rockbox/apps/onplay.c

1965 lines
65 KiB
C
Raw Normal View History

/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
* $Id$
*
* Copyright (C) 2002 Björn Stenberg
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
* KIND, either express or implied.
*
****************************************************************************/
#include <errno.h>
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <stdbool.h>
#include "debug.h"
#include "lcd.h"
#include "file.h"
#include "audio.h"
#include "menu.h"
#include "lang.h"
#include "playlist.h"
#include "button.h"
#include "kernel.h"
#include "keyboard.h"
#include "mp3data.h"
#include "metadata.h"
#include "screens.h"
#include "tree.h"
#include "settings.h"
#include "playlist_viewer.h"
#include "talk.h"
#include "onplay.h"
#include "filetypes.h"
#include "open_plugin.h"
#include "plugin.h"
#include "bookmark.h"
#include "action.h"
#include "splash.h"
#include "yesno.h"
#include "menus/exported_menus.h"
#include "icons.h"
#include "sound_menu.h"
#include "playlist_menu.h"
#include "playlist_catalog.h"
#ifdef HAVE_TAGCACHE
#include "tagtree.h"
#endif
#include "cuesheet.h"
#include "statusbar-skinned.h"
#include "pitchscreen.h"
#include "viewport.h"
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
#include "pathfuncs.h"
#include "shortcuts.h"
static int context;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
static const char *selected_file = NULL;
static char selected_file_path[MAX_PATH];
static int selected_file_attr = 0;
static int onplay_result = ONPLAY_OK;
extern struct menu_item_ex file_menu; /* settings_menu.c */
/* redefine MAKE_MENU so the MENU_EXITAFTERTHISMENU flag can be added easily */
#define MAKE_ONPLAYMENU( name, str, callback, icon, ... ) \
static const struct menu_item_ex *name##_[] = {__VA_ARGS__}; \
static const struct menu_callback_with_desc name##__ = {callback,str,icon};\
static const struct menu_item_ex name = \
{MT_MENU|MENU_HAS_DESC|MENU_EXITAFTERTHISMENU| \
MENU_ITEM_COUNT(sizeof( name##_)/sizeof(*name##_)), \
{ (void*)name##_},{.callback_and_desc = & name##__}};
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* Used for directory move, copy and delete */
struct dirrecurse_params
{
char path[MAX_PATH]; /* Buffer for full path */
size_t append; /* Append position in 'path' for stack push */
};
enum clipboard_op_flags
{
PASTE_CUT = 0x00, /* Is a move (cut) operation (default) */
PASTE_COPY = 0x01, /* Is a copy operation */
PASTE_OVERWRITE = 0x02, /* Overwrite destination */
PASTE_EXDEV = 0x04, /* Actually copy/move across volumes */
};
/* result codec of various onplay operations */
enum onplay_result_code
{
/* Anything < 0 is failure */
OPRC_SUCCESS = 0, /* All operations completed successfully */
OPRC_NOOP = 1, /* Operation didn't need to do anything */
OPRC_CANCELLED = 2, /* Operation was cancelled by user */
OPRC_NOOVERWRT = 3,
};
static struct clipboard
{
char path[MAX_PATH]; /* Clipped file's path */
unsigned int attr; /* Clipped file's attributes */
unsigned int flags; /* Operation type flags */
} clipboard;
/* Empty the clipboard */
static void clipboard_clear_selection(struct clipboard *clip)
{
clip->path[0] = '\0';
clip->attr = 0;
clip->flags = 0;
}
/* Store the selection in the clipboard */
static bool clipboard_clip(struct clipboard *clip, const char *path,
unsigned int attr, unsigned int flags)
{
/* if it fits it clips */
if (strlcpy(clip->path, path, sizeof (clip->path))
< sizeof (clip->path)) {
clip->attr = attr;
clip->flags = flags;
return true;
}
else {
clipboard_clear_selection(clip);
return false;
}
}
/* ----------------------------------------------------------------------- */
/* Displays the bookmark menu options for the user to decide. This is an */
/* interface function. */
/* ----------------------------------------------------------------------- */
static int bookmark_menu_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list);
MENUITEM_FUNCTION(bookmark_create_menu_item, 0,
ID2P(LANG_BOOKMARK_MENU_CREATE),
bookmark_create_menu, NULL,
bookmark_menu_callback, Icon_Bookmark);
MENUITEM_FUNCTION(bookmark_load_menu_item, 0,
ID2P(LANG_BOOKMARK_MENU_LIST),
bookmark_load_menu, NULL,
bookmark_menu_callback, Icon_Bookmark);
MAKE_ONPLAYMENU(bookmark_menu, ID2P(LANG_BOOKMARK_MENU),
bookmark_menu_callback, Icon_Bookmark,
&bookmark_create_menu_item, &bookmark_load_menu_item);
static int bookmark_menu_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list)
{
(void) this_list;
switch (action)
{
case ACTION_REQUEST_MENUITEM:
/* hide create bookmark option if bookmarking isn't currently possible (no track playing, queued tracks...) */
if (this_item == &bookmark_create_menu_item)
{
if (!bookmark_is_bookmarkable_state())
return ACTION_EXIT_MENUITEM;
}
/* hide loading bookmarks menu if no bookmarks exist */
else if (this_item == &bookmark_load_menu_item)
{
if (!bookmark_exists())
return ACTION_EXIT_MENUITEM;
}
/* hide the bookmark menu if bookmarks can't be loaded or created */
else if (!bookmark_is_bookmarkable_state() && !bookmark_exists())
return ACTION_EXIT_MENUITEM;
break;
case ACTION_EXIT_MENUITEM:
settings_save();
break;
}
return action;
}
/* playing_time screen context */
struct playing_time_info {
int curr_playing; /* index of currently playing track in playlist */
int nb_tracks; /* how many tracks in playlist */
/* seconds before and after current position, and total. Datatype
allows for values up to 68years. If I had kept it in ms
though, it would have overflowed at 24days, which takes
something like 8.5GB at 32kbps, and so we could conceivably
have playlists lasting longer than that. */
long secs_bef, secs_aft, secs_ttl;
long trk_secs_bef, trk_secs_aft, trk_secs_ttl;
/* kilobytes played before and after current pos, and total.
Kilobytes because bytes would overflow. Data type range is up
to 2TB. */
long kbs_bef, kbs_aft, kbs_ttl;
};
/* list callback for playing_time screen */
static const char * playing_time_get_or_speak_info(int selected_item, void * data,
char *buf, size_t buffer_len,
bool say_it)
{
struct playing_time_info *pti = (struct playing_time_info *)data;
switch(selected_item) {
case 0: { /* elapsed and total time */
char timestr1[25], timestr2[25];
format_time_auto(timestr1, sizeof(timestr1), pti->secs_bef,
UNIT_SEC, false);
format_time_auto(timestr2, sizeof(timestr2), pti->secs_ttl,
UNIT_SEC, false);
long elapsed_perc; /* percentage of duration elapsed */
if (pti->secs_ttl == 0)
elapsed_perc = 0;
else if (pti->secs_ttl <= 0xFFFFFF)
elapsed_perc = pti->secs_bef *100 / pti->secs_ttl;
else /* sacrifice some precision to avoid overflow */
elapsed_perc = (pti->secs_bef>>7) *100 /(pti->secs_ttl>>7);
snprintf(buf, buffer_len, str(LANG_PLAYTIME_ELAPSED),
timestr1, timestr2, elapsed_perc);
if (say_it)
talk_ids(false, LANG_PLAYTIME_ELAPSED,
TALK_ID(pti->secs_bef, UNIT_TIME),
VOICE_OF,
TALK_ID(pti->secs_ttl, UNIT_TIME),
VOICE_PAUSE,
TALK_ID(elapsed_perc, UNIT_PERCENT));
break;
}
case 1: { /* playlist remaining time */
char timestr[25];
format_time_auto(timestr, sizeof(timestr), pti->secs_aft,
UNIT_SEC, false);
snprintf(buf, buffer_len, str(LANG_PLAYTIME_REMAINING),
timestr);
if (say_it)
talk_ids(false, LANG_PLAYTIME_REMAINING,
TALK_ID(pti->secs_aft, UNIT_TIME));
break;
}
case 2: { /* track elapsed and duration */
char timestr1[25], timestr2[25];
format_time_auto(timestr1, sizeof(timestr1), pti->trk_secs_bef,
UNIT_SEC, false);
format_time_auto(timestr2, sizeof(timestr2), pti->trk_secs_ttl,
UNIT_SEC, false);
long elapsed_perc; /* percentage of duration elapsed */
if (pti->trk_secs_ttl == 0)
elapsed_perc = 0;
else if (pti->trk_secs_ttl <= 0xFFFFFF)
elapsed_perc = pti->trk_secs_bef *100 / pti->trk_secs_ttl;
else /* sacrifice some precision to avoid overflow */
elapsed_perc = (pti->trk_secs_bef>>7) *100 /(pti->trk_secs_ttl>>7);
snprintf(buf, buffer_len, str(LANG_PLAYTIME_TRK_ELAPSED),
timestr1, timestr2, elapsed_perc);
if (say_it)
talk_ids(false, LANG_PLAYTIME_TRK_ELAPSED,
TALK_ID(pti->trk_secs_bef, UNIT_TIME),
VOICE_OF,
TALK_ID(pti->trk_secs_ttl, UNIT_TIME),
VOICE_PAUSE,
TALK_ID(elapsed_perc, UNIT_PERCENT));
break;
}
case 3: { /* track remaining time */
char timestr[25];
format_time_auto(timestr, sizeof(timestr), pti->trk_secs_aft,
UNIT_SEC, false);
snprintf(buf, buffer_len, str(LANG_PLAYTIME_TRK_REMAINING),
timestr);
if (say_it)
talk_ids(false, LANG_PLAYTIME_TRK_REMAINING,
TALK_ID(pti->trk_secs_aft, UNIT_TIME));
break;
}
case 4: { /* track index */
int track_perc = (pti->curr_playing+1) *100 / pti->nb_tracks;
snprintf(buf, buffer_len, str(LANG_PLAYTIME_TRACK),
pti->curr_playing + 1, pti->nb_tracks, track_perc);
if (say_it)
talk_ids(false, LANG_PLAYTIME_TRACK,
TALK_ID(pti->curr_playing+1, UNIT_INT),
VOICE_OF,
TALK_ID(pti->nb_tracks, UNIT_INT),
VOICE_PAUSE,
TALK_ID(track_perc, UNIT_PERCENT));
break;
}
case 5: { /* storage size */
char str1[10], str2[10], str3[10];
output_dyn_value(str1, sizeof(str1), pti->kbs_ttl, kibyte_units, 3, true);
output_dyn_value(str2, sizeof(str2), pti->kbs_bef, kibyte_units, 3, true);
output_dyn_value(str3, sizeof(str3), pti->kbs_aft, kibyte_units, 3, true);
snprintf(buf, buffer_len, str(LANG_PLAYTIME_STORAGE),
str1,str2,str3);
if (say_it) {
talk_id(LANG_PLAYTIME_STORAGE, false);
output_dyn_value(NULL, 0, pti->kbs_ttl, kibyte_units, 3, true);
talk_ids(true, VOICE_PAUSE, VOICE_PLAYTIME_DONE);
output_dyn_value(NULL, 0, pti->kbs_bef, kibyte_units, 3, true);
talk_id(LANG_PLAYTIME_REMAINING, true);
output_dyn_value(NULL, 0, pti->kbs_aft, kibyte_units, 3, true);
}
break;
}
case 6: { /* Average track file size */
char str[10];
long avg_track_size = pti->kbs_ttl /pti->nb_tracks;
output_dyn_value(str, sizeof(str), avg_track_size, kibyte_units, 3, true);
snprintf(buf, buffer_len, str(LANG_PLAYTIME_AVG_TRACK_SIZE),
str);
if (say_it) {
talk_id(LANG_PLAYTIME_AVG_TRACK_SIZE, false);
output_dyn_value(NULL, 0, avg_track_size, kibyte_units, 3, true);
}
break;
}
case 7: { /* Average bitrate */
/* Convert power of 2 kilobytes to power of 10 kilobits */
long avg_bitrate = pti->kbs_ttl / pti->secs_ttl *1024 *8 /1000;
snprintf(buf, buffer_len, str(LANG_PLAYTIME_AVG_BITRATE),
avg_bitrate);
if (say_it)
talk_ids(false, LANG_PLAYTIME_AVG_BITRATE,
TALK_ID(avg_bitrate, UNIT_KBIT));
break;
}
}
return buf;
}
static const char * playing_time_get_info(int selected_item, void * data,
char *buffer, size_t buffer_len)
{
return playing_time_get_or_speak_info(selected_item, data,
buffer, buffer_len, false);
}
static int playing_time_speak_info(int selected_item, void * data)
{
static char buffer[MAX_PATH];
playing_time_get_or_speak_info(selected_item, data,
buffer, MAX_PATH, true);
return 0;
}
/* playing time screen: shows total and elapsed playlist duration and
other stats */
static bool playing_time(void)
{
unsigned long talked_tick = current_tick;
struct playing_time_info pti;
struct playlist_track_info pltrack;
struct mp3entry id3;
int i, fd, ret;
pti.nb_tracks = playlist_amount();
playlist_get_resume_info(&pti.curr_playing);
struct mp3entry *curr_id3 = audio_current_track();
if (pti.curr_playing == -1 || !curr_id3)
return false;
pti.secs_bef = pti.trk_secs_bef = curr_id3->elapsed/1000;
pti.secs_aft = pti.trk_secs_aft
= (curr_id3->length -curr_id3->elapsed)/1000;
pti.kbs_bef = curr_id3->offset/1024;
pti.kbs_aft = (curr_id3->filesize -curr_id3->offset)/1024;
splash(0, ID2P(LANG_WAIT));
/* Go through each file in the playlist and get its stats. For
huge playlists this can take a while... The reference position
is the position at the moment this function was invoked,
although playback continues forward. */
for (i = 0; i < pti.nb_tracks; i++) {
/* Show a splash while we are loading. */
splashf(0, str(LANG_LOADING_PERCENT),
i*100/pti.nb_tracks, str(LANG_OFF_ABORT));
/* Voice equivalent */
if (TIME_AFTER(current_tick, talked_tick+5*HZ)) {
talked_tick = current_tick;
talk_ids(false, LANG_LOADING_PERCENT,
TALK_ID(i*100/pti.nb_tracks, UNIT_PERCENT));
}
if (action_userabort(TIMEOUT_NOBLOCK))
goto exit;
if (i == pti.curr_playing)
continue;
if (playlist_get_track_info(NULL, i, &pltrack) < 0)
goto error;
if ((fd = open(pltrack.filename, O_RDONLY)) < 0)
goto error;
ret = get_metadata(&id3, fd, pltrack.filename);
close(fd);
if (!ret)
goto error;
if (i < pti.curr_playing) {
pti.secs_bef += id3.length/1000;
pti.kbs_bef += id3.filesize/1024;
} else {
pti.secs_aft += id3.length/1000;
pti.kbs_aft += id3.filesize/1024;
}
}
pti.secs_ttl = pti.secs_bef +pti.secs_aft;
pti.trk_secs_ttl = pti.trk_secs_bef +pti.trk_secs_aft;
pti.kbs_ttl = pti.kbs_bef +pti.kbs_aft;
struct gui_synclist pt_lists;
int key;
gui_synclist_init(&pt_lists, &playing_time_get_info, &pti, true, 1, NULL);
if (global_settings.talk_menu)
gui_synclist_set_voice_callback(&pt_lists, playing_time_speak_info);
gui_synclist_set_nb_items(&pt_lists, 8);
gui_synclist_draw(&pt_lists);
/* gui_syncstatusbar_draw(&statusbars, true); */
gui_synclist_speak_item(&pt_lists);
while (true) {
/* gui_syncstatusbar_draw(&statusbars, false); */
if (list_do_action(CONTEXT_LIST, HZ/2,
&pt_lists, &key, LIST_WRAP_UNLESS_HELD) == 0
&& key!=ACTION_NONE && key!=ACTION_UNKNOWN)
{
talk_force_shutup();
return(default_event_handler(key) == SYS_USB_CONNECTED);
}
}
error:
splash(HZ, ID2P(LANG_PLAYTIME_ERROR));
exit:
return false;
}
/* CONTEXT_WPS playlist options */
static bool shuffle_playlist(void)
{
if (!warn_on_pl_erase())
return false;
playlist_sort(NULL, true);
playlist_randomise(NULL, current_tick, true);
return false;
}
static bool save_playlist(void)
{
/* save_playlist_screen should load the newly saved playlist and resume */
save_playlist_screen(NULL);
return false;
}
extern struct menu_item_ex view_cur_playlist; /* from playlist_menu.c */
MENUITEM_FUNCTION(search_playlist_item, 0, ID2P(LANG_SEARCH_IN_PLAYLIST),
search_playlist, NULL, NULL, Icon_Playlist);
MENUITEM_FUNCTION(playlist_save_item, 0, ID2P(LANG_SAVE_DYNAMIC_PLAYLIST),
save_playlist, NULL, NULL, Icon_Playlist);
MENUITEM_FUNCTION(reshuffle_item, 0, ID2P(LANG_SHUFFLE_PLAYLIST),
shuffle_playlist, NULL, NULL, Icon_Playlist);
MENUITEM_FUNCTION(playing_time_item, 0, ID2P(LANG_PLAYING_TIME),
playing_time, NULL, NULL, Icon_Playlist);
MAKE_ONPLAYMENU( wps_playlist_menu, ID2P(LANG_PLAYLIST),
NULL, Icon_Playlist,
&view_cur_playlist, &search_playlist_item,
&playlist_save_item, &reshuffle_item, &playing_time_item
);
/* CONTEXT_[TREE|ID3DB] playlist options */
static bool add_to_playlist(int position, bool queue)
{
bool new_playlist = false;
if (!(audio_status() & AUDIO_STATUS_PLAY))
{
new_playlist = true;
if (position == PLAYLIST_REPLACE)
position = PLAYLIST_INSERT;
}
const char *lines[] = {
ID2P(LANG_RECURSE_DIRECTORY_QUESTION),
selected_file
};
const struct text_message message={lines, 2};
splash(0, ID2P(LANG_WAIT));
if (new_playlist)
playlist_create(NULL, NULL);
/* always set seed before inserting shuffled */
if (position == PLAYLIST_INSERT_SHUFFLED ||
position == PLAYLIST_INSERT_LAST_SHUFFLED)
{
srand(current_tick);
if (position == PLAYLIST_INSERT_LAST_SHUFFLED)
playlist_set_last_shuffled_start();
}
#ifdef HAVE_TAGCACHE
if (context == CONTEXT_ID3DB)
{
tagtree_insert_selection_playlist(position, queue);
}
else
#endif
{
if ((selected_file_attr & FILE_ATTR_MASK) == FILE_ATTR_AUDIO)
playlist_insert_track(NULL, selected_file, position, queue, true);
else if (selected_file_attr & ATTR_DIRECTORY)
{
bool recurse = false;
if (global_settings.recursive_dir_insert != RECURSE_ASK)
recurse = (bool)global_settings.recursive_dir_insert;
else
{
/* Ask if user wants to recurse directory */
recurse = (gui_syncyesno_run(&message, NULL, NULL)==YESNO_YES);
}
playlist_insert_directory(NULL, selected_file, position, queue,
recurse);
}
else if ((selected_file_attr & FILE_ATTR_MASK) == FILE_ATTR_M3U)
playlist_insert_playlist(NULL, selected_file, position, queue);
}
if (new_playlist && (playlist_amount() > 0))
{
/* nothing is currently playing so begin playing what we just
inserted */
if (global_settings.playlist_shuffle)
playlist_shuffle(current_tick, -1);
playlist_start(0, 0, 0);
onplay_result = ONPLAY_START_PLAY;
}
return false;
}
static bool view_playlist(void)
{
bool was_playing = audio_status() & AUDIO_STATUS_PLAY;
bool result;
result = playlist_viewer_ex(selected_file);
if (!was_playing && (audio_status() & AUDIO_STATUS_PLAY) &&
onplay_result == ONPLAY_OK)
/* playlist was started from viewer */
onplay_result = ONPLAY_START_PLAY;
return result;
}
static int playlist_insert_func(void *param)
{
if (((intptr_t)param == PLAYLIST_REPLACE ||
((intptr_t)param == PLAYLIST_INSERT_SHUFFLED && !(audio_status() & AUDIO_STATUS_PLAY))) &&
!warn_on_pl_erase())
return 0;
add_to_playlist((intptr_t)param, false);
return 0;
}
static int playlist_queue_func(void *param)
{
add_to_playlist((intptr_t)param, true);
return 0;
}
static int treeplaylist_wplayback_callback(int action,
const struct menu_item_ex* this_item,
struct gui_synclist *this_list);
static int treeplaylist_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list);
/* insert items */
MENUITEM_FUNCTION(i_pl_item, MENU_FUNC_USEPARAM, ID2P(LANG_INSERT),
playlist_insert_func, (intptr_t*)PLAYLIST_INSERT,
treeplaylist_wplayback_callback, Icon_Playlist);
MENUITEM_FUNCTION(i_first_pl_item, MENU_FUNC_USEPARAM, ID2P(LANG_INSERT_FIRST),
playlist_insert_func, (intptr_t*)PLAYLIST_INSERT_FIRST,
treeplaylist_wplayback_callback, Icon_Playlist);
MENUITEM_FUNCTION(i_last_pl_item, MENU_FUNC_USEPARAM, ID2P(LANG_INSERT_LAST),
playlist_insert_func, (intptr_t*)PLAYLIST_INSERT_LAST,
treeplaylist_wplayback_callback, Icon_Playlist);
MENUITEM_FUNCTION(i_shuf_pl_item, MENU_FUNC_USEPARAM,
ID2P(LANG_INSERT_SHUFFLED), playlist_insert_func,
(intptr_t*)PLAYLIST_INSERT_SHUFFLED,
treeplaylist_callback, Icon_Playlist);
MENUITEM_FUNCTION(i_last_shuf_pl_item, MENU_FUNC_USEPARAM,
ID2P(LANG_INSERT_LAST_SHUFFLED), playlist_insert_func,
(intptr_t*)PLAYLIST_INSERT_LAST_SHUFFLED,
treeplaylist_callback, Icon_Playlist);
/* queue items */
MENUITEM_FUNCTION(q_pl_item, MENU_FUNC_USEPARAM, ID2P(LANG_QUEUE),
playlist_queue_func, (intptr_t*)PLAYLIST_INSERT,
treeplaylist_callback, Icon_Playlist);
MENUITEM_FUNCTION(q_first_pl_item, MENU_FUNC_USEPARAM, ID2P(LANG_QUEUE_FIRST),
playlist_queue_func, (intptr_t*)PLAYLIST_INSERT_FIRST,
treeplaylist_callback, Icon_Playlist);
MENUITEM_FUNCTION(q_last_pl_item, MENU_FUNC_USEPARAM, ID2P(LANG_QUEUE_LAST),
playlist_queue_func, (intptr_t*)PLAYLIST_INSERT_LAST,
treeplaylist_callback, Icon_Playlist);
MENUITEM_FUNCTION(q_shuf_pl_item, MENU_FUNC_USEPARAM,
ID2P(LANG_QUEUE_SHUFFLED), playlist_queue_func,
(intptr_t*)PLAYLIST_INSERT_SHUFFLED,
treeplaylist_callback, Icon_Playlist);
MENUITEM_FUNCTION(q_last_shuf_pl_item, MENU_FUNC_USEPARAM,
ID2P(LANG_QUEUE_LAST_SHUFFLED), playlist_queue_func,
(intptr_t*)PLAYLIST_INSERT_LAST_SHUFFLED,
treeplaylist_callback, Icon_Playlist);
/* queue items in submenu */
MENUITEM_FUNCTION(q_pl_submenu_item, MENU_FUNC_USEPARAM, ID2P(LANG_QUEUE),
playlist_queue_func, (intptr_t*)PLAYLIST_INSERT,
NULL, Icon_Playlist);
MENUITEM_FUNCTION(q_first_pl_submenu_item, MENU_FUNC_USEPARAM, ID2P(LANG_QUEUE_FIRST),
playlist_queue_func, (intptr_t*)PLAYLIST_INSERT_FIRST,
NULL, Icon_Playlist);
MENUITEM_FUNCTION(q_last_pl_submenu_item, MENU_FUNC_USEPARAM, ID2P(LANG_QUEUE_LAST),
playlist_queue_func, (intptr_t*)PLAYLIST_INSERT_LAST,
NULL, Icon_Playlist);
MENUITEM_FUNCTION(q_shuf_pl_submenu_item, MENU_FUNC_USEPARAM,
ID2P(LANG_QUEUE_SHUFFLED), playlist_queue_func,
(intptr_t*)PLAYLIST_INSERT_SHUFFLED,
treeplaylist_callback, Icon_Playlist);
MENUITEM_FUNCTION(q_last_shuf_pl_submenu_item, MENU_FUNC_USEPARAM,
ID2P(LANG_QUEUE_LAST_SHUFFLED), playlist_queue_func,
(intptr_t*)PLAYLIST_INSERT_LAST_SHUFFLED,
treeplaylist_callback, Icon_Playlist);
MAKE_ONPLAYMENU(queue_menu, ID2P(LANG_QUEUE_MENU),
treeplaylist_wplayback_callback, Icon_Playlist,
&q_pl_submenu_item,
&q_first_pl_submenu_item,
&q_last_pl_submenu_item,
&q_shuf_pl_submenu_item,
&q_last_shuf_pl_submenu_item);
static int treeplaylist_wplayback_callback(int action,
const struct menu_item_ex* this_item,
struct gui_synclist *this_list)
{
(void)this_list;
switch (action)
{
case ACTION_REQUEST_MENUITEM:
if ((audio_status() & AUDIO_STATUS_PLAY) &&
(this_item != &queue_menu ||
global_settings.show_queue_options == QUEUE_SHOW_IN_SUBMENU))
return action;
else
return ACTION_EXIT_MENUITEM;
break;
}
return action;
}
/* replace playlist */
MENUITEM_FUNCTION(replace_pl_item, MENU_FUNC_USEPARAM, ID2P(LANG_CLEAR_LIST_AND_PLAY_NEXT),
playlist_insert_func, (intptr_t*)PLAYLIST_REPLACE,
NULL, Icon_Playlist);
MENUITEM_FUNCTION(replace_shuf_pl_item, MENU_FUNC_USEPARAM,
ID2P(LANG_CLEAR_LIST_AND_PLAY_SHUFFLED), playlist_insert_func,
(intptr_t*)PLAYLIST_INSERT_SHUFFLED,
treeplaylist_callback, Icon_Playlist);
MAKE_ONPLAYMENU(tree_playlist_menu, ID2P(LANG_CURRENT_PLAYLIST),
treeplaylist_callback, Icon_Playlist,
/* insert */
&i_pl_item,
&i_first_pl_item,
&i_last_pl_item,
&i_shuf_pl_item,
&i_last_shuf_pl_item,
/* queue */
&q_pl_item,
&q_first_pl_item,
&q_last_pl_item,
&q_shuf_pl_item,
&q_last_shuf_pl_item,
/* Queue submenu */
&queue_menu,
/* replace */
&replace_pl_item,
&replace_shuf_pl_item
);
static int treeplaylist_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list)
{
(void)this_list;
switch (action)
{
case ACTION_REQUEST_MENUITEM:
if (this_item == &tree_playlist_menu)
{
if ((selected_file_attr & FILE_ATTR_MASK) == FILE_ATTR_AUDIO ||
(selected_file_attr & FILE_ATTR_MASK) == FILE_ATTR_M3U ||
(selected_file_attr & ATTR_DIRECTORY))
return action;
}
else if ((this_item == &q_pl_item ||
this_item == &q_first_pl_item ||
this_item == &q_last_pl_item) &&
global_settings.show_queue_options == QUEUE_SHOW_AT_TOPLEVEL &&
(audio_status() & AUDIO_STATUS_PLAY))
{
return action;
}
else if (this_item == &i_shuf_pl_item)
{
if (global_settings.show_shuffled_adding_options &&
(audio_status() & AUDIO_STATUS_PLAY))
{
return action;
}
}
else if (this_item == &replace_shuf_pl_item)
{
if (global_settings.show_shuffled_adding_options &&
!(audio_status() & AUDIO_STATUS_PLAY) &&
((selected_file_attr & ATTR_DIRECTORY) ||
((selected_file_attr & FILE_ATTR_MASK) == FILE_ATTR_M3U)))
{
return action;
}
}
else if (this_item == &q_shuf_pl_submenu_item ||
(this_item == &q_shuf_pl_item &&
global_settings.show_queue_options == QUEUE_SHOW_AT_TOPLEVEL))
{
if (global_settings.show_shuffled_adding_options &&
(audio_status() & AUDIO_STATUS_PLAY))
{
return action;
}
}
else if (this_item == &i_last_shuf_pl_item ||
this_item == &q_last_shuf_pl_submenu_item ||
(this_item == &q_last_shuf_pl_item &&
global_settings.show_queue_options == QUEUE_SHOW_AT_TOPLEVEL))
{
if (global_settings.show_shuffled_adding_options &&
(playlist_amount() > 0) &&
(audio_status() & AUDIO_STATUS_PLAY) &&
((selected_file_attr & ATTR_DIRECTORY) ||
((selected_file_attr & FILE_ATTR_MASK) == FILE_ATTR_M3U)))
{
return action;
}
}
return ACTION_EXIT_MENUITEM;
break;
}
return action;
}
void onplay_show_playlist_menu(char* path)
{
selected_file = path;
if (dir_exists(path))
selected_file_attr = ATTR_DIRECTORY;
else
selected_file_attr = filetype_get_attr(path);
do_menu(&tree_playlist_menu, NULL, NULL, false);
}
/* playlist catalog options */
static bool cat_add_to_a_playlist(void)
{
return catalog_add_to_a_playlist(selected_file, selected_file_attr,
false, NULL);
}
static bool cat_add_to_a_new_playlist(void)
{
return catalog_add_to_a_playlist(selected_file, selected_file_attr,
true, NULL);
}
static int clipboard_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list);
static bool set_catalogdir(void)
{
catalog_set_directory(selected_file);
settings_save();
return false;
}
MENUITEM_FUNCTION(set_catalogdir_item, 0, ID2P(LANG_SET_AS_PLAYLISTCAT_DIR),
set_catalogdir, NULL, clipboard_callback, Icon_Playlist);
static int cat_playlist_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list);
MENUITEM_FUNCTION(cat_add_to_list, 0, ID2P(LANG_CATALOG_ADD_TO),
cat_add_to_a_playlist, 0, NULL, Icon_Playlist);
MENUITEM_FUNCTION(cat_add_to_new, 0, ID2P(LANG_CATALOG_ADD_TO_NEW),
cat_add_to_a_new_playlist, 0, NULL, Icon_Playlist);
MAKE_ONPLAYMENU(cat_playlist_menu, ID2P(LANG_CATALOG),
cat_playlist_callback, Icon_Playlist,
&cat_add_to_list, &cat_add_to_new, &set_catalogdir_item);
void onplay_show_playlist_cat_menu(char* track_name)
{
selected_file = track_name;
selected_file_attr = FILE_ATTR_AUDIO;
do_menu(&cat_playlist_menu, NULL, NULL, false);
}
static int cat_playlist_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list)
{
(void)this_item;
(void)this_list;
if (!selected_file ||
(((selected_file_attr & FILE_ATTR_MASK) != FILE_ATTR_AUDIO) &&
((selected_file_attr & FILE_ATTR_MASK) != FILE_ATTR_M3U) &&
((selected_file_attr & ATTR_DIRECTORY) == 0)))
{
return ACTION_EXIT_MENUITEM;
}
#ifdef HAVE_TAGCACHE
if (context == CONTEXT_ID3DB &&
((selected_file_attr & FILE_ATTR_MASK) != FILE_ATTR_AUDIO))
{
return ACTION_EXIT_MENUITEM;
}
#endif
switch (action)
{
case ACTION_REQUEST_MENUITEM:
if ((audio_status() & AUDIO_STATUS_PLAY) || context != CONTEXT_WPS)
{
return action;
}
else
return ACTION_EXIT_MENUITEM;
break;
}
return action;
}
static void draw_slider(void)
{
LCD core move buf ptr and address look up function viewport struct I'm currently running up against the limitations of the lcd_draw functions I want these functions to be able to be used on any size buffer not just buffers with a stride matching the underlying device [DONE] allow the framebuffer to be decoupled from the device framebuffer [DONE need examples] allow for some simple blit like transformations [DONE] remove the device framebuffer from the plugin api [DONE}ditto remote framebuffer [DONE] remove _viewport_get_framebuffer you can call struct *vp = lcd_set_viewport(NULL) and vp->buffer->fb_ptr while remote lcds may compile (and work in the sim) its not been tested on targets [FIXED] backdrops need work to be screen agnostic [FIXED] screen statusbar is not being combined into the main viewport correctly yet [FIXED] screen elements are displayed incorrectly after switch to void* [FIXED] core didn't restore proper viewport on splash etc. [NEEDS TESTING] remote lcd garbled data [FIXED] osd lib garbled screen on bmp_part [FIXED] grey_set_vp needs to return old viewport like lcd_set_viewport [FIXED] Viewport update now handles viewports with differing buffers/strides by copying to the main buffer [FIXED] splash on top of WPS leaves old framebuffer data (doesn't redraw) [UPDATE] refined this a bit more to have clear_viewport set the clean bit and have skin_render do its own screen clear scrolling viewports no longer trigger wps refresh also fixed a bug where guisyncyesno was displaying and then disappearing [ADDED!] New LCD macros that allow you to create properly size frame buffers in you desired size without wasting bytes (LCD_ and LCD_REMOTE_) LCD_STRIDE(w, h) same as STRIDE_MAIN LCD_FBSTRIDE(w, h) returns target specific stride for a buffer W x H LCD_NBELEMS(w, h) returns the number of fb_data sized elemenst needed for a buffer W x H LCD_NATIVE_STRIDE(s) conversion between rockbox native vertical and lcd native stride (2bitH) test_viewports.c has an example of usage [FIXED!!] 2bit targets don't respect non-native strides [FIXED] Few define snags Change-Id: I0d04c3834e464eca84a5a715743a297a0cefd0af
2020-10-07 06:01:35 +00:00
struct viewport *last_vp;
FOR_NB_SCREENS(i)
{
struct viewport vp;
int slider_height = 2*screens[i].getcharheight();
viewport_set_defaults(&vp, i);
LCD core move buf ptr and address look up function viewport struct I'm currently running up against the limitations of the lcd_draw functions I want these functions to be able to be used on any size buffer not just buffers with a stride matching the underlying device [DONE] allow the framebuffer to be decoupled from the device framebuffer [DONE need examples] allow for some simple blit like transformations [DONE] remove the device framebuffer from the plugin api [DONE}ditto remote framebuffer [DONE] remove _viewport_get_framebuffer you can call struct *vp = lcd_set_viewport(NULL) and vp->buffer->fb_ptr while remote lcds may compile (and work in the sim) its not been tested on targets [FIXED] backdrops need work to be screen agnostic [FIXED] screen statusbar is not being combined into the main viewport correctly yet [FIXED] screen elements are displayed incorrectly after switch to void* [FIXED] core didn't restore proper viewport on splash etc. [NEEDS TESTING] remote lcd garbled data [FIXED] osd lib garbled screen on bmp_part [FIXED] grey_set_vp needs to return old viewport like lcd_set_viewport [FIXED] Viewport update now handles viewports with differing buffers/strides by copying to the main buffer [FIXED] splash on top of WPS leaves old framebuffer data (doesn't redraw) [UPDATE] refined this a bit more to have clear_viewport set the clean bit and have skin_render do its own screen clear scrolling viewports no longer trigger wps refresh also fixed a bug where guisyncyesno was displaying and then disappearing [ADDED!] New LCD macros that allow you to create properly size frame buffers in you desired size without wasting bytes (LCD_ and LCD_REMOTE_) LCD_STRIDE(w, h) same as STRIDE_MAIN LCD_FBSTRIDE(w, h) returns target specific stride for a buffer W x H LCD_NBELEMS(w, h) returns the number of fb_data sized elemenst needed for a buffer W x H LCD_NATIVE_STRIDE(s) conversion between rockbox native vertical and lcd native stride (2bitH) test_viewports.c has an example of usage [FIXED!!] 2bit targets don't respect non-native strides [FIXED] Few define snags Change-Id: I0d04c3834e464eca84a5a715743a297a0cefd0af
2020-10-07 06:01:35 +00:00
last_vp = screens[i].set_viewport(&vp);
show_busy_slider(&screens[i], 1, vp.height - slider_height,
vp.width-2, slider_height-1);
screens[i].update_viewport();
LCD core move buf ptr and address look up function viewport struct I'm currently running up against the limitations of the lcd_draw functions I want these functions to be able to be used on any size buffer not just buffers with a stride matching the underlying device [DONE] allow the framebuffer to be decoupled from the device framebuffer [DONE need examples] allow for some simple blit like transformations [DONE] remove the device framebuffer from the plugin api [DONE}ditto remote framebuffer [DONE] remove _viewport_get_framebuffer you can call struct *vp = lcd_set_viewport(NULL) and vp->buffer->fb_ptr while remote lcds may compile (and work in the sim) its not been tested on targets [FIXED] backdrops need work to be screen agnostic [FIXED] screen statusbar is not being combined into the main viewport correctly yet [FIXED] screen elements are displayed incorrectly after switch to void* [FIXED] core didn't restore proper viewport on splash etc. [NEEDS TESTING] remote lcd garbled data [FIXED] osd lib garbled screen on bmp_part [FIXED] grey_set_vp needs to return old viewport like lcd_set_viewport [FIXED] Viewport update now handles viewports with differing buffers/strides by copying to the main buffer [FIXED] splash on top of WPS leaves old framebuffer data (doesn't redraw) [UPDATE] refined this a bit more to have clear_viewport set the clean bit and have skin_render do its own screen clear scrolling viewports no longer trigger wps refresh also fixed a bug where guisyncyesno was displaying and then disappearing [ADDED!] New LCD macros that allow you to create properly size frame buffers in you desired size without wasting bytes (LCD_ and LCD_REMOTE_) LCD_STRIDE(w, h) same as STRIDE_MAIN LCD_FBSTRIDE(w, h) returns target specific stride for a buffer W x H LCD_NBELEMS(w, h) returns the number of fb_data sized elemenst needed for a buffer W x H LCD_NATIVE_STRIDE(s) conversion between rockbox native vertical and lcd native stride (2bitH) test_viewports.c has an example of usage [FIXED!!] 2bit targets don't respect non-native strides [FIXED] Few define snags Change-Id: I0d04c3834e464eca84a5a715743a297a0cefd0af
2020-10-07 06:01:35 +00:00
screens[i].set_viewport(last_vp);
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
static void clear_display(bool update)
{
struct viewport vp;
LCD core move buf ptr and address look up function viewport struct I'm currently running up against the limitations of the lcd_draw functions I want these functions to be able to be used on any size buffer not just buffers with a stride matching the underlying device [DONE] allow the framebuffer to be decoupled from the device framebuffer [DONE need examples] allow for some simple blit like transformations [DONE] remove the device framebuffer from the plugin api [DONE}ditto remote framebuffer [DONE] remove _viewport_get_framebuffer you can call struct *vp = lcd_set_viewport(NULL) and vp->buffer->fb_ptr while remote lcds may compile (and work in the sim) its not been tested on targets [FIXED] backdrops need work to be screen agnostic [FIXED] screen statusbar is not being combined into the main viewport correctly yet [FIXED] screen elements are displayed incorrectly after switch to void* [FIXED] core didn't restore proper viewport on splash etc. [NEEDS TESTING] remote lcd garbled data [FIXED] osd lib garbled screen on bmp_part [FIXED] grey_set_vp needs to return old viewport like lcd_set_viewport [FIXED] Viewport update now handles viewports with differing buffers/strides by copying to the main buffer [FIXED] splash on top of WPS leaves old framebuffer data (doesn't redraw) [UPDATE] refined this a bit more to have clear_viewport set the clean bit and have skin_render do its own screen clear scrolling viewports no longer trigger wps refresh also fixed a bug where guisyncyesno was displaying and then disappearing [ADDED!] New LCD macros that allow you to create properly size frame buffers in you desired size without wasting bytes (LCD_ and LCD_REMOTE_) LCD_STRIDE(w, h) same as STRIDE_MAIN LCD_FBSTRIDE(w, h) returns target specific stride for a buffer W x H LCD_NBELEMS(w, h) returns the number of fb_data sized elemenst needed for a buffer W x H LCD_NATIVE_STRIDE(s) conversion between rockbox native vertical and lcd native stride (2bitH) test_viewports.c has an example of usage [FIXED!!] 2bit targets don't respect non-native strides [FIXED] Few define snags Change-Id: I0d04c3834e464eca84a5a715743a297a0cefd0af
2020-10-07 06:01:35 +00:00
struct viewport *last_vp;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
FOR_NB_SCREENS(i)
{
struct screen * screen = &screens[i];
viewport_set_defaults(&vp, screen->screen_type);
LCD core move buf ptr and address look up function viewport struct I'm currently running up against the limitations of the lcd_draw functions I want these functions to be able to be used on any size buffer not just buffers with a stride matching the underlying device [DONE] allow the framebuffer to be decoupled from the device framebuffer [DONE need examples] allow for some simple blit like transformations [DONE] remove the device framebuffer from the plugin api [DONE}ditto remote framebuffer [DONE] remove _viewport_get_framebuffer you can call struct *vp = lcd_set_viewport(NULL) and vp->buffer->fb_ptr while remote lcds may compile (and work in the sim) its not been tested on targets [FIXED] backdrops need work to be screen agnostic [FIXED] screen statusbar is not being combined into the main viewport correctly yet [FIXED] screen elements are displayed incorrectly after switch to void* [FIXED] core didn't restore proper viewport on splash etc. [NEEDS TESTING] remote lcd garbled data [FIXED] osd lib garbled screen on bmp_part [FIXED] grey_set_vp needs to return old viewport like lcd_set_viewport [FIXED] Viewport update now handles viewports with differing buffers/strides by copying to the main buffer [FIXED] splash on top of WPS leaves old framebuffer data (doesn't redraw) [UPDATE] refined this a bit more to have clear_viewport set the clean bit and have skin_render do its own screen clear scrolling viewports no longer trigger wps refresh also fixed a bug where guisyncyesno was displaying and then disappearing [ADDED!] New LCD macros that allow you to create properly size frame buffers in you desired size without wasting bytes (LCD_ and LCD_REMOTE_) LCD_STRIDE(w, h) same as STRIDE_MAIN LCD_FBSTRIDE(w, h) returns target specific stride for a buffer W x H LCD_NBELEMS(w, h) returns the number of fb_data sized elemenst needed for a buffer W x H LCD_NATIVE_STRIDE(s) conversion between rockbox native vertical and lcd native stride (2bitH) test_viewports.c has an example of usage [FIXED!!] 2bit targets don't respect non-native strides [FIXED] Few define snags Change-Id: I0d04c3834e464eca84a5a715743a297a0cefd0af
2020-10-07 06:01:35 +00:00
last_vp = screen->set_viewport(&vp);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
screen->clear_viewport();
if (update) {
screen->update_viewport();
}
LCD core move buf ptr and address look up function viewport struct I'm currently running up against the limitations of the lcd_draw functions I want these functions to be able to be used on any size buffer not just buffers with a stride matching the underlying device [DONE] allow the framebuffer to be decoupled from the device framebuffer [DONE need examples] allow for some simple blit like transformations [DONE] remove the device framebuffer from the plugin api [DONE}ditto remote framebuffer [DONE] remove _viewport_get_framebuffer you can call struct *vp = lcd_set_viewport(NULL) and vp->buffer->fb_ptr while remote lcds may compile (and work in the sim) its not been tested on targets [FIXED] backdrops need work to be screen agnostic [FIXED] screen statusbar is not being combined into the main viewport correctly yet [FIXED] screen elements are displayed incorrectly after switch to void* [FIXED] core didn't restore proper viewport on splash etc. [NEEDS TESTING] remote lcd garbled data [FIXED] osd lib garbled screen on bmp_part [FIXED] grey_set_vp needs to return old viewport like lcd_set_viewport [FIXED] Viewport update now handles viewports with differing buffers/strides by copying to the main buffer [FIXED] splash on top of WPS leaves old framebuffer data (doesn't redraw) [UPDATE] refined this a bit more to have clear_viewport set the clean bit and have skin_render do its own screen clear scrolling viewports no longer trigger wps refresh also fixed a bug where guisyncyesno was displaying and then disappearing [ADDED!] New LCD macros that allow you to create properly size frame buffers in you desired size without wasting bytes (LCD_ and LCD_REMOTE_) LCD_STRIDE(w, h) same as STRIDE_MAIN LCD_FBSTRIDE(w, h) returns target specific stride for a buffer W x H LCD_NBELEMS(w, h) returns the number of fb_data sized elemenst needed for a buffer W x H LCD_NATIVE_STRIDE(s) conversion between rockbox native vertical and lcd native stride (2bitH) test_viewports.c has an example of usage [FIXED!!] 2bit targets don't respect non-native strides [FIXED] Few define snags Change-Id: I0d04c3834e464eca84a5a715743a297a0cefd0af
2020-10-07 06:01:35 +00:00
screen->set_viewport(last_vp);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
}
}
static void splash_path(const char *path)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
clear_display(false);
path_basename(path, &path);
splash(0, path);
draw_slider();
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* Splashes the path and checks the keys */
static bool poll_cancel_action(const char *path)
{
splash_path(path);
return ACTION_STD_CANCEL == get_action(CONTEXT_STD, TIMEOUT_NOBLOCK);
}
static int confirm_overwrite(void)
{
static const char *lines[] = { ID2P(LANG_REALLY_OVERWRITE) };
static const struct text_message message = { lines, 1 };
return gui_syncyesno_run(&message, NULL, NULL);
}
static int confirm_delete(const char *file)
{
const char *lines[] = { ID2P(LANG_REALLY_DELETE), file };
const char *yes_lines[] = { ID2P(LANG_DELETING), file };
const struct text_message message = { lines, 2 };
const struct text_message yes_message = { yes_lines, 2 };
return gui_syncyesno_run(&message, &yes_message, NULL);
}
static bool check_new_name(const char *basename)
{
/* at least prevent escapes out of the base directory from keyboard-
entered filenames; the file code should reject other invalidities */
return *basename != '\0' && !strchr(basename, PATH_SEPCH) &&
!is_dotdir_name(basename);
}
static void splash_cancelled(void)
{
clear_display(true);
splash(HZ, ID2P(LANG_CANCEL));
}
static void splash_failed(int lang_what)
{
cond_talk_ids_fq(lang_what, LANG_FAILED);
clear_display(true);
splashf(HZ*2, "%s %s", str(lang_what), str(LANG_FAILED));
}
/* helper function to remove a non-empty directory */
static int remove_dir(struct dirrecurse_params *parm)
{
DIR *dir = opendir(parm->path);
if (!dir) {
return -1; /* open error */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
size_t append = parm->append;
int rc = OPRC_SUCCESS;
/* walk through the directory content */
while (rc == OPRC_SUCCESS) {
errno = 0; /* distinguish failure from eod */
struct dirent *entry = readdir(dir);
if (!entry) {
if (errno) {
rc = -1;
}
break;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
}
struct dirinfo info = dir_get_info(dir, entry);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if ((info.attribute & ATTR_DIRECTORY) &&
is_dotdir_name(entry->d_name)) {
continue; /* skip these */
}
/* append name to current directory */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
parm->append = append + path_append(&parm->path[append],
PA_SEP_HARD, entry->d_name,
sizeof (parm->path) - append);
if (parm->append >= sizeof (parm->path)) {
rc = -1;
break; /* no space left in buffer */
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (info.attribute & ATTR_DIRECTORY) {
/* remove a subdirectory */
rc = remove_dir(parm);
} else {
/* remove a file */
if (poll_cancel_action(parm->path)) {
rc = OPRC_CANCELLED;
break;
}
rc = remove(parm->path);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* Remove basename we added above */
parm->path[append] = '\0';
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
closedir(dir);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (rc == 0) {
/* remove the now empty directory */
if (poll_cancel_action(parm->path)) {
rc = OPRC_CANCELLED;
} else {
rc = rmdir(parm->path);
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return rc;
}
/* share code for file and directory deletion, saves space */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
static int delete_file_dir(void)
{
const char *to_delete=selected_file;
const int to_delete_attr=selected_file_attr;
if (confirm_delete(to_delete) != YESNO_YES) {
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return 1;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
clear_display(true);
splash(HZ/2, str(LANG_DELETING));
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int rc = -1;
if (to_delete_attr & ATTR_DIRECTORY) { /* true if directory */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
struct dirrecurse_params parm;
parm.append = strlcpy(parm.path, to_delete, sizeof (parm.path));
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (parm.append < sizeof (parm.path)) {
cpu_boost(true);
rc = remove_dir(&parm);
cpu_boost(false);
}
} else {
rc = remove(to_delete);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (rc < OPRC_SUCCESS) {
splash_failed(LANG_DELETE);
} else if (rc == OPRC_CANCELLED) {
splash_cancelled();
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (rc != OPRC_NOOP) {
/* Could have failed after some but not all needed changes; reload */
onplay_result = ONPLAY_RELOAD_DIR;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return 1;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
static int rename_file(void)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int rc = -1;
char newname[MAX_PATH];
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
const char *oldbase, *selection = selected_file;
path_basename(selection, &oldbase);
size_t pathlen = oldbase - selection;
char *newbase = newname + pathlen;
if (strlcpy(newname, selection, sizeof (newname)) >= sizeof (newname)) {
/* Too long */
} else if (kbd_input(newbase, sizeof (newname) - pathlen, NULL) < 0) {
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = OPRC_CANCELLED;
} else if (!strcmp(oldbase, newbase)) {
rc = OPRC_NOOP; /* No change at all */
} else if (check_new_name(newbase)) {
switch (relate(selection, newname))
{
case RELATE_DIFFERENT:
if (file_exists(newname)) {
break; /* don't overwrite */
}
/* Fall-through */
case RELATE_SAME:
rc = rename(selection, newname);
break;
case RELATE_PREFIX:
default:
break;
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (rc < OPRC_SUCCESS) {
splash_failed(LANG_RENAME);
} else if (rc == OPRC_CANCELLED) {
/* splash_cancelled(); kbd_input() splashes it */
} else if (rc == OPRC_SUCCESS) {
onplay_result = ONPLAY_RELOAD_DIR;
}
return 1;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
static int create_dir(void)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int rc = -1;
char dirname[MAX_PATH];
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
size_t pathlen = path_append(dirname, getcwd(NULL, 0), PA_SEP_HARD,
sizeof (dirname));
char *basename = dirname + pathlen;
if (pathlen >= sizeof (dirname)) {
/* Too long */
} else if (kbd_input(basename, sizeof (dirname) - pathlen, NULL) < 0) {
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = OPRC_CANCELLED;
} else if (check_new_name(basename)) {
rc = mkdir(dirname);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (rc < OPRC_SUCCESS) {
splash_failed(LANG_CREATE_DIR);
} else if (rc == OPRC_CANCELLED) {
/* splash_cancelled(); kbd_input() splashes it */
} else if (rc == OPRC_SUCCESS) {
onplay_result = ONPLAY_RELOAD_DIR;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return 1;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* Paste a file */
static int clipboard_pastefile(const char *src, const char *target,
unsigned int flags)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int rc = -1;
while (!(flags & (PASTE_COPY | PASTE_EXDEV))) {
if ((flags & PASTE_OVERWRITE) || !file_exists(target)) {
/* Rename and possibly overwrite the file */
if (poll_cancel_action(src)) {
rc = OPRC_CANCELLED;
} else {
rc = rename(src, target);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
#ifdef HAVE_MULTIVOLUME
if (rc < 0 && errno == EXDEV) {
/* Failed because cross volume rename doesn't work; force
a move instead */
flags |= PASTE_EXDEV;
break;
}
#endif /* HAVE_MULTIVOLUME */
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* See if we can get the plugin buffer for the file copy buffer */
size_t buffersize;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
char *buffer = (char *) plugin_get_buffer(&buffersize);
if (buffer == NULL || buffersize < 512) {
/* Not large enough, try for a disk sector worth of stack
instead */
buffersize = 512;
buffer = (char *)alloca(buffersize);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (buffer == NULL) {
return -1;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
buffersize &= ~0x1ff; /* Round buffer size to multiple of sector
size */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int src_fd = open(src, O_RDONLY);
if (src_fd >= 0) {
int oflag = O_WRONLY|O_CREAT;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (!(flags & PASTE_OVERWRITE)) {
oflag |= O_EXCL;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int target_fd = open(target, oflag, 0666);
if (target_fd >= 0) {
off_t total_size = 0;
off_t next_cancel_test = 0; /* No excessive button polling */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
rc = OPRC_SUCCESS;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
while (rc == OPRC_SUCCESS) {
if (total_size >= next_cancel_test) {
next_cancel_test = total_size + 0x10000;
if (poll_cancel_action(src)) {
rc = OPRC_CANCELLED;
break;
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
ssize_t bytesread = read(src_fd, buffer, buffersize);
if (bytesread <= 0) {
if (bytesread < 0) {
rc = -1;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* else eof on buffer boundary; nothing to write */
break;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
ssize_t byteswritten = write(target_fd, buffer, bytesread);
if (byteswritten < bytesread) {
/* Some I/O error */
rc = -1;
break;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
total_size += byteswritten;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (bytesread < (ssize_t)buffersize) {
/* EOF with trailing bytes */
break;
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (rc == OPRC_SUCCESS) {
/* If overwriting, set the correct length if original was
longer */
rc = ftruncate(target_fd, total_size);
}
close(target_fd);
if (rc != OPRC_SUCCESS) {
/* Copy failed. Cleanup. */
remove(target);
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
close(src_fd);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (rc == OPRC_SUCCESS && !(flags & PASTE_COPY)) {
/* Remove the source file */
rc = remove(src);
}
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* Paste a directory */
static int clipboard_pastedirectory(struct dirrecurse_params *src,
struct dirrecurse_params *target,
unsigned int flags)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int rc = -1;
while (!(flags & (PASTE_COPY | PASTE_EXDEV))) {
if ((flags & PASTE_OVERWRITE) || !file_exists(target->path)) {
/* Just try to move the directory */
if (poll_cancel_action(src->path)) {
rc = OPRC_CANCELLED;
} else {
rc = rename(src->path, target->path);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (rc < 0) {
int errnum = errno;
if (errnum == ENOTEMPTY && (flags & PASTE_OVERWRITE)) {
/* Directory is not empty thus rename() will not do a quick
overwrite */
break;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
#ifdef HAVE_MULTIVOLUME
else if (errnum == EXDEV) {
/* Failed because cross volume rename doesn't work; force
a move instead */
flags |= PASTE_EXDEV;
break;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
#endif /* HAVE_MULTIVOLUME */
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
DIR *srcdir = opendir(src->path);
if (srcdir) {
/* Make a directory to copy things to */
rc = mkdir(target->path);
if (rc < 0 && errno == EEXIST && (flags & PASTE_OVERWRITE)) {
/* Exists and overwrite was approved */
rc = OPRC_SUCCESS;
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
size_t srcap = src->append, targetap = target->append;
/* Walk through the directory content; this loop will exit as soon as
there's a problem */
while (rc == OPRC_SUCCESS) {
errno = 0; /* Distinguish failure from eod */
struct dirent *entry = readdir(srcdir);
if (!entry) {
if (errno) {
rc = -1;
}
break;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
}
struct dirinfo info = dir_get_info(srcdir, entry);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if ((info.attribute & ATTR_DIRECTORY) &&
is_dotdir_name(entry->d_name)) {
continue; /* Skip these */
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* Append names to current directories */
src->append = srcap +
path_append(&src->path[srcap], PA_SEP_HARD, entry->d_name,
sizeof(src->path) - srcap);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
target->append = targetap +
path_append(&target->path[targetap], PA_SEP_HARD, entry->d_name,
sizeof (target->path) - targetap);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (src->append >= sizeof (src->path) ||
target->append >= sizeof (target->path)) {
rc = -1; /* No space left in buffer */
break;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (poll_cancel_action(src->path)) {
rc = OPRC_CANCELLED;
break;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
DEBUGF("Copy %s to %s\n", src->path, target->path);
if (info.attribute & ATTR_DIRECTORY) {
/* Copy/move a subdirectory */
rc = clipboard_pastedirectory(src, target, flags); /* recursion */
} else {
/* Copy/move a file */
rc = clipboard_pastefile(src->path, target->path, flags);
}
/* Remove basenames we added above */
src->path[srcap] = target->path[targetap] = '\0';
}
if (rc == OPRC_SUCCESS && !(flags & PASTE_COPY)) {
/* Remove the now empty directory */
rc = rmdir(src->path);
}
closedir(srcdir);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return rc;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
static bool clipboard_cut(void)
{
return clipboard_clip(&clipboard, selected_file, selected_file_attr,
PASTE_CUT);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
static bool clipboard_copy(void)
{
return clipboard_clip(&clipboard, selected_file, selected_file_attr,
PASTE_COPY);
}
/* Paste the clipboard to the current directory */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
static int clipboard_paste(void)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (!clipboard.path[0])
return 1;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
int rc = -1;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
struct dirrecurse_params src, target;
unsigned int flags = clipboard.flags;
/* Figure out the name of the selection */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
const char *nameptr;
path_basename(clipboard.path, &nameptr);
/* Final target is current directory plus name of selection */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
target.append = path_append(target.path, getcwd(NULL, 0),
nameptr, sizeof (target.path));
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
switch (target.append < sizeof (target.path) ?
relate(clipboard.path, target.path) : -1)
{
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
case RELATE_SAME:
rc = OPRC_NOOP;
break;
case RELATE_DIFFERENT:
if (file_exists(target.path)) {
/* If user chooses not to overwrite, cancel */
if (confirm_overwrite() == YESNO_NO) {
rc = OPRC_NOOVERWRT;
break;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
flags |= PASTE_OVERWRITE;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
clear_display(true);
splash(HZ/2, (flags & PASTE_COPY) ? ID2P(LANG_COPYING) :
ID2P(LANG_MOVING));
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
/* Now figure out what we're doing */
cpu_boost(true);
if (clipboard.attr & ATTR_DIRECTORY) {
/* Copy or move a subdirectory */
src.append = strlcpy(src.path, clipboard.path,
sizeof (src.path));
if (src.append < sizeof (src.path)) {
rc = clipboard_pastedirectory(&src, &target, flags);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
} else {
/* Copy or move a file */
rc = clipboard_pastefile(clipboard.path, target.path, flags);
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
cpu_boost(false);
break;
case RELATE_PREFIX:
default: /* Some other relation / failure */
break;
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
clear_display(true);
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
switch (rc)
{
case OPRC_CANCELLED:
splash_cancelled();
case OPRC_SUCCESS:
onplay_result = ONPLAY_RELOAD_DIR;
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
case OPRC_NOOP:
clipboard_clear_selection(&clipboard);
case OPRC_NOOVERWRT:
break;
default:
if (rc < OPRC_SUCCESS) {
splash_failed(LANG_PASTE);
onplay_result = ONPLAY_RELOAD_DIR;
}
}
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return 1;
}
#ifdef HAVE_TAGCACHE
static int set_rating_inline(void)
{
struct mp3entry* id3 = audio_current_track();
if (id3 && id3->tagcache_idx && global_settings.runtimedb)
{
set_int_ex(str(LANG_MENU_SET_RATING), "", UNIT_INT, (void*)(&id3->rating),
NULL, 1, 0, 10, NULL, NULL);
tagcache_update_numeric(id3->tagcache_idx-1, tag_rating, id3->rating);
}
else
splash(HZ*2, ID2P(LANG_ID3_NO_INFO));
return 0;
}
static int ratingitem_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list)
{
(void)this_item;
(void)this_list;
switch (action)
{
case ACTION_REQUEST_MENUITEM:
if (!selected_file || !global_settings.runtimedb ||
!tagcache_is_usable())
return ACTION_EXIT_MENUITEM;
break;
}
return action;
}
MENUITEM_FUNCTION(rating_item, 0, ID2P(LANG_MENU_SET_RATING),
set_rating_inline, NULL,
ratingitem_callback, Icon_Questionmark);
#endif
MENUITEM_RETURNVALUE(plugin_item, ID2P(LANG_OPEN_PLUGIN),
GO_TO_PLUGIN, NULL, Icon_Plugin);
static bool view_cue(void)
{
struct mp3entry* id3 = audio_current_track();
if (id3 && id3->cuesheet)
{
browse_cuesheet(id3->cuesheet);
}
return false;
}
static int view_cue_item_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list)
{
(void)this_item;
(void)this_list;
struct mp3entry* id3 = audio_current_track();
switch (action)
{
case ACTION_REQUEST_MENUITEM:
if (!selected_file
|| !id3 || !id3->cuesheet)
return ACTION_EXIT_MENUITEM;
break;
}
return action;
}
MENUITEM_FUNCTION(view_cue_item, 0, ID2P(LANG_BROWSE_CUESHEET),
view_cue, NULL, view_cue_item_callback, Icon_NOICON);
static int browse_id3_wrapper(void)
{
if (browse_id3())
return GO_TO_ROOT;
return GO_TO_PREVIOUS;
}
/* CONTEXT_WPS items */
MENUITEM_FUNCTION(browse_id3_item, MENU_FUNC_CHECK_RETVAL, ID2P(LANG_MENU_SHOW_ID3_INFO),
browse_id3_wrapper, NULL, NULL, Icon_NOICON);
#ifdef HAVE_PITCHCONTROL
MENUITEM_FUNCTION(pitch_screen_item, 0, ID2P(LANG_PITCH),
gui_syncpitchscreen_run, NULL, NULL, Icon_Audio);
#endif
/* CONTEXT_[TREE|ID3DB] items */
static int clipboard_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list);
MENUITEM_FUNCTION(rename_file_item, 0, ID2P(LANG_RENAME),
rename_file, NULL, clipboard_callback, Icon_NOICON);
MENUITEM_FUNCTION(clipboard_cut_item, 0, ID2P(LANG_CUT),
clipboard_cut, NULL, clipboard_callback, Icon_NOICON);
MENUITEM_FUNCTION(clipboard_copy_item, 0, ID2P(LANG_COPY),
clipboard_copy, NULL, clipboard_callback, Icon_NOICON);
MENUITEM_FUNCTION(clipboard_paste_item, 0, ID2P(LANG_PASTE),
clipboard_paste, NULL, clipboard_callback, Icon_NOICON);
MENUITEM_FUNCTION(delete_file_item, 0, ID2P(LANG_DELETE),
delete_file_dir, NULL, clipboard_callback, Icon_NOICON);
MENUITEM_FUNCTION(delete_dir_item, 0, ID2P(LANG_DELETE_DIR),
delete_file_dir, NULL, clipboard_callback, Icon_NOICON);
MENUITEM_FUNCTION(create_dir_item, 0, ID2P(LANG_CREATE_DIR),
create_dir, NULL, clipboard_callback, Icon_NOICON);
/* other items */
static bool list_viewers(void)
{
int ret = filetype_list_viewers(selected_file);
if (ret == PLUGIN_USB_CONNECTED)
onplay_result = ONPLAY_RELOAD_DIR;
return false;
}
static bool onplay_load_plugin(void *param)
{
int ret = filetype_load_plugin((const char*)param, selected_file);
if (ret == PLUGIN_USB_CONNECTED)
onplay_result = ONPLAY_RELOAD_DIR;
return false;
}
MENUITEM_FUNCTION(list_viewers_item, 0, ID2P(LANG_ONPLAY_OPEN_WITH),
list_viewers, NULL, clipboard_callback, Icon_NOICON);
MENUITEM_FUNCTION(properties_item, MENU_FUNC_USEPARAM, ID2P(LANG_PROPERTIES),
onplay_load_plugin, (void *)"properties",
clipboard_callback, Icon_NOICON);
static bool onplay_add_to_shortcuts(void)
{
shortcuts_add(SHORTCUT_BROWSER, selected_file);
return false;
}
MENUITEM_FUNCTION(add_to_faves_item, 0, ID2P(LANG_ADD_TO_FAVES),
onplay_add_to_shortcuts, NULL,
clipboard_callback, Icon_NOICON);
#if LCD_DEPTH > 1
static bool set_backdrop(void)
{
strlcpy(global_settings.backdrop_file, selected_file,
sizeof(global_settings.backdrop_file));
settings_save();
skin_backdrop_load_setting();
skin_backdrop_show(sb_get_backdrop(SCREEN_MAIN));
return true;
}
MENUITEM_FUNCTION(set_backdrop_item, 0, ID2P(LANG_SET_AS_BACKDROP),
set_backdrop, NULL, clipboard_callback, Icon_NOICON);
#endif
#ifdef HAVE_RECORDING
static bool set_recdir(void)
{
strlcpy(global_settings.rec_directory, selected_file,
sizeof(global_settings.rec_directory));
settings_save();
return false;
}
MENUITEM_FUNCTION(set_recdir_item, 0, ID2P(LANG_SET_AS_REC_DIR),
set_recdir, NULL, clipboard_callback, Icon_Recording);
#endif
static bool set_startdir(void)
{
snprintf(global_settings.start_directory,
sizeof(global_settings.start_directory),
"%s/", selected_file);
settings_save();
return false;
}
MENUITEM_FUNCTION(set_startdir_item, 0, ID2P(LANG_SET_AS_START_DIR),
set_startdir, NULL, clipboard_callback, Icon_file_view_menu);
static int clipboard_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list)
{
(void)this_list;
switch (action)
{
case ACTION_REQUEST_MENUITEM:
#ifdef HAVE_MULTIVOLUME
/* no rename+delete for volumes */
if ((selected_file_attr & ATTR_VOLUME) &&
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
(this_item == &rename_file_item ||
this_item == &delete_dir_item ||
this_item == &clipboard_cut_item ||
this_item == &list_viewers_item))
return ACTION_EXIT_MENUITEM;
#endif
#ifdef HAVE_TAGCACHE
if (context == CONTEXT_ID3DB)
{
if (((selected_file_attr & FILE_ATTR_MASK) ==
FILE_ATTR_AUDIO) &&
this_item == &properties_item)
return action;
return ACTION_EXIT_MENUITEM;
}
#endif
if (this_item == &clipboard_paste_item)
{ /* visible if there is something to paste */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
return (clipboard.path[0] != 0) ?
action : ACTION_EXIT_MENUITEM;
}
else if (this_item == &create_dir_item)
{
/* always visible */
return action;
}
else if (selected_file)
{
/* requires an actual file */
if (this_item == &rename_file_item ||
this_item == &clipboard_cut_item ||
this_item == &clipboard_copy_item ||
this_item == &properties_item ||
this_item == &add_to_faves_item)
{
return action;
}
else if ((selected_file_attr & ATTR_DIRECTORY))
{
/* only for directories */
if (this_item == &delete_dir_item ||
this_item == &set_startdir_item ||
this_item == &set_catalogdir_item
#ifdef HAVE_RECORDING
|| this_item == &set_recdir_item
#endif
)
return action;
}
else if (this_item == &delete_file_item ||
this_item == &list_viewers_item)
{
/* only for files */
return action;
}
#if LCD_DEPTH > 1
else if (this_item == &set_backdrop_item)
{
char *suffix = strrchr(selected_file, '.');
if (suffix)
{
if (strcasecmp(suffix, ".bmp") == 0)
{
return action;
}
}
}
#endif
}
return ACTION_EXIT_MENUITEM;
break;
}
return action;
}
static int onplaymenu_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list);
/* used when onplay() is called in the CONTEXT_WPS context */
MAKE_ONPLAYMENU( wps_onplay_menu, ID2P(LANG_ONPLAY_MENU_TITLE),
onplaymenu_callback, Icon_Audio,
&wps_playlist_menu, &cat_playlist_menu,
&sound_settings, &playback_settings,
#ifdef HAVE_TAGCACHE
&rating_item,
#endif
&bookmark_menu,
&plugin_item,
&browse_id3_item, &list_viewers_item,
&delete_file_item, &view_cue_item,
#ifdef HAVE_PITCHCONTROL
&pitch_screen_item,
#endif
);
MENUITEM_FUNCTION(view_playlist_item, 0, ID2P(LANG_VIEW),
view_playlist, NULL,
onplaymenu_callback, Icon_Playlist);
/* used when onplay() is not called in the CONTEXT_WPS context */
MAKE_ONPLAYMENU( tree_onplay_menu, ID2P(LANG_ONPLAY_MENU_TITLE),
onplaymenu_callback, Icon_file_view_menu,
&view_playlist_item, &tree_playlist_menu, &cat_playlist_menu,
&rename_file_item, &clipboard_cut_item, &clipboard_copy_item,
&clipboard_paste_item, &delete_file_item, &delete_dir_item,
#if LCD_DEPTH > 1
&set_backdrop_item,
#endif
&list_viewers_item, &create_dir_item, &properties_item,
#ifdef HAVE_RECORDING
&set_recdir_item,
#endif
&set_startdir_item, &add_to_faves_item, &file_menu,
);
static int onplaymenu_callback(int action,
const struct menu_item_ex *this_item,
struct gui_synclist *this_list)
{
(void)this_list;
switch (action)
{
case ACTION_TREE_STOP:
if (this_item == &wps_onplay_menu)
{
list_stop_handler();
return ACTION_STD_CANCEL;
}
break;
case ACTION_REQUEST_MENUITEM:
if (this_item == &view_playlist_item)
{
if ((selected_file_attr & FILE_ATTR_MASK) == FILE_ATTR_M3U &&
context == CONTEXT_TREE)
return action;
}
return ACTION_EXIT_MENUITEM;
break;
case ACTION_EXIT_MENUITEM:
return ACTION_EXIT_AFTER_THIS_MENUITEM;
break;
}
return action;
}
#ifdef HAVE_HOTKEY
/* direct function calls, no need for menu callbacks */
static bool delete_item(void)
{
#ifdef HAVE_MULTIVOLUME
/* no delete for volumes */
Rewrite filesystem code (WIP) This patch redoes the filesystem code from the FAT driver up to the clipboard code in onplay.c. Not every aspect of this is finished therefore it is still "WIP". I don't wish to do too much at once (haha!). What is left to do is get dircache back in the sim and find an implementation for the dircache indicies in the tagcache and playlist code or do something else that has the same benefit. Leaving these out for now does not make anything unusable. All the basics are done. Phone app code should probably get vetted (and app path handling just plain rewritten as environment expansions); the SDL app and Android run well. Main things addressed: 1) Thread safety: There is none right now in the trunk code. Most of what currently works is luck when multiple threads are involved or multiple descriptors to the same file are open. 2) POSIX compliance: Many of the functions behave nothing like their counterparts on a host system. This leads to inconsistent code or very different behavior from native to hosted. One huge offender was rename(). Going point by point would fill a book. 3) Actual running RAM usage: Many targets will use less RAM and less stack space (some more RAM because I upped the number of cache buffers for large memory). There's very little memory lying fallow in rarely-used areas (see 'Key core changes' below). Also, all targets may open the same number of directory streams whereas before those with less than 8MB RAM were limited to 8, not 12 implying those targets will save slightly less. 4) Performance: The test_disk plugin shows markedly improved performance, particularly in the area of (uncached) directory scanning, due partly to more optimal directory reading and to a better sector cache algorithm. Uncached times tend to be better while there is a bit of a slowdown in dircache due to it being a bit heavier of an implementation. It's not noticeable by a human as far as I can say. Key core changes: 1) Files and directories share core code and data structures. 2) The filesystem code knows which descriptors refer to same file. This ensures that changes from one stream are appropriately reflected in every open descriptor for that file (fileobj_mgr.c). 3) File and directory cache buffers are borrowed from the main sector cache. This means that when they are not in use by a file, they are not wasted, but used for the cache. Most of the time, only a few of them are needed. It also means that adding more file and directory handles is less expensive. All one must do in ensure a large enough cache to borrow from. 4) Relative path components are supported and the namespace is unified. It does not support full relative paths to an implied current directory; what is does support is use of "." and "..". Adding the former would not be very difficult. The namespace is unified in the sense that volumes may be specified several times along with relative parts, e.g.: "/<0>/foo/../../<1>/bar" :<=> "/<1>/bar". 5) Stack usage is down due to sharing of data, static allocation and less duplication of strings on the stack. This requires more serialization than I would like but since the number of threads is limited to a low number, the tradoff in favor of the stack seems reasonable. 6) Separates and heirarchicalizes (sic) the SIM and APP filesystem code. SIM path and volume handling is just like the target. Some aspects of the APP file code get more straightforward (e.g. no path hashing is needed). Dircache: Deserves its own section. Dircache is new but pays homage to the old. The old one was not compatible and so it, since it got redone, does all the stuff it always should have done such as: 1) It may be update and used at any time during the build process. No longer has one to wait for it to finish building to do basic file management (create, remove, rename, etc.). 2) It does not need to be either fully scanned or completely disabled; it can be incomplete (i.e. overfilled, missing paths), still be of benefit and be correct. 3) Handles mounting and dismounting of individual volumes which means a full rebuild is not needed just because you pop a new SD card in the slot. Now, because it reuses its freed entry data, may rebuild only that volume. 4) Much more fundamental to the file code. When it is built, it is the keeper of the master file list whether enabled or not ("disabled" is just a state of the cache). Its must always to ready to be started and bind all streams opened prior to being enabled. 5) Maintains any short filenames in OEM format which means that it does not need to be rebuilt when changing the default codepage. Miscellaneous Compatibility: 1) Update any other code that would otherwise not work such as the hotswap mounting code in various card drivers. 2) File management: Clipboard needed updating because of the behavioral changes. Still needs a little more work on some finer points. 3) Remove now-obsolete functionality such as the mutex's "no preempt" flag (which was only for the prior FAT driver). 4) struct dirinfo uses time_t rather than raw FAT directory entry time fields. I plan to follow up on genericizing everything there (i.e. no FAT attributes). 5) unicode.c needed some redoing so that the file code does not try try to load codepages during a scan, which is actually a problem with the current code. The default codepage, if any is required, is now kept in RAM separarately (bufalloced) from codepages specified to iso_decode() (which must not be bufalloced because the conversion may be done by playback threads). Brings with it some additional reusable core code: 1) Revised file functions: Reusable code that does things such as safe path concatenation and parsing without buffer limitations or data duplication. Variants that copy or alter the input path may be based off these. To do: 1) Put dircache functionality back in the sim. Treating it internally as a different kind of file system seems the best approach at this time. 2) Restore use of dircache indexes in the playlist and database or something effectively the same. Since the cache doesn't have to be complete in order to be used, not getting a hit on the cache doesn't unambiguously say if the path exists or not. Change-Id: Ia30f3082a136253e3a0eae0784e3091d138915c8 Reviewed-on: http://gerrit.rockbox.org/566 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested: Michael Sevakis <jethead71@rockbox.org>
2013-08-06 02:02:45 +00:00
if (selected_file_attr & ATTR_VOLUME)
return false;
#endif
return delete_file_dir();
}
static bool open_with(void)
{
/* only open files */
if (selected_file_attr & ATTR_DIRECTORY)
return false;
#ifdef HAVE_MULTIVOLUME
if (selected_file_attr & ATTR_VOLUME)
return false;
#endif
return list_viewers();
}
static int playlist_insert_shuffled(void)
{
if ((audio_status() & AUDIO_STATUS_PLAY) ||
(selected_file_attr & ATTR_DIRECTORY) ||
((selected_file_attr & FILE_ATTR_MASK) == FILE_ATTR_M3U))
{
playlist_insert_func((intptr_t*)PLAYLIST_INSERT_SHUFFLED);
return ONPLAY_START_PLAY;
}
return ONPLAY_RELOAD_DIR;
}
static void hotkey_run_plugin(void)
{
open_plugin_run(ID2P(LANG_HOTKEY_WPS));
}
struct hotkey_assignment {
int action; /* hotkey_action */
int lang_id; /* Language ID */
struct menu_func func; /* Function to run if this entry is selected */
int return_code; /* What to return after the function is run */
};
#define HOTKEY_FUNC(func, param) {{(void *)func}, param}
/* Any desired hotkey functions go here, in the enum in onplay.h,
and in the settings menu in settings_list.c. The order here
is not important. */
static struct hotkey_assignment hotkey_items[] = {
{ HOTKEY_VIEW_PLAYLIST, LANG_VIEW_DYNAMIC_PLAYLIST,
HOTKEY_FUNC(NULL, NULL),
ONPLAY_PLAYLIST },
{ HOTKEY_SHOW_TRACK_INFO, LANG_MENU_SHOW_ID3_INFO,
HOTKEY_FUNC(browse_id3, NULL),
ONPLAY_RELOAD_DIR },
#ifdef HAVE_PITCHCONTROL
{ HOTKEY_PITCHSCREEN, LANG_PITCH,
HOTKEY_FUNC(gui_syncpitchscreen_run, NULL),
ONPLAY_RELOAD_DIR },
#endif
{ HOTKEY_OPEN_WITH, LANG_ONPLAY_OPEN_WITH,
HOTKEY_FUNC(open_with, NULL),
ONPLAY_RELOAD_DIR },
{ HOTKEY_DELETE, LANG_DELETE,
HOTKEY_FUNC(delete_item, NULL),
ONPLAY_RELOAD_DIR },
{ HOTKEY_INSERT, LANG_INSERT,
HOTKEY_FUNC(playlist_insert_func, (intptr_t*)PLAYLIST_INSERT),
ONPLAY_RELOAD_DIR },
{ HOTKEY_INSERT_SHUFFLED, LANG_INSERT_SHUFFLED,
HOTKEY_FUNC(playlist_insert_shuffled, NULL),
ONPLAY_RELOAD_DIR },
{ HOTKEY_PLUGIN, LANG_OPEN_PLUGIN,
HOTKEY_FUNC(hotkey_run_plugin, NULL),
ONPLAY_OK },
{ HOTKEY_BOOKMARK, LANG_BOOKMARK_MENU_CREATE,
HOTKEY_FUNC(bookmark_create_menu, NULL),
ONPLAY_OK },
};
/* Return the language ID for this action */
int get_hotkey_lang_id(int action)
{
int i = ARRAYLEN(hotkey_items);
while (i--)
{
if (hotkey_items[i].action == action)
return hotkey_items[i].lang_id;
}
return LANG_OFF;
}
/* Execute the hotkey function, if listed */
static int execute_hotkey(bool is_wps)
{
int i = ARRAYLEN(hotkey_items);
struct hotkey_assignment *this_item;
const int action = (is_wps ? global_settings.hotkey_wps :
global_settings.hotkey_tree);
/* search assignment struct for a match for the hotkey setting */
while (i--)
{
this_item = &hotkey_items[i];
if (this_item->action == action)
{
/* run the associated function (with optional param), if any */
const struct menu_func func = this_item->func;
int func_return = ONPLAY_RELOAD_DIR;
if (func.function != NULL)
{
if (func.param != NULL)
func_return = (*func.function_w_param)(func.param);
else
func_return = (*func.function)();
}
/* return with the associated code */
const int return_code = this_item->return_code;
/* ONPLAY_OK here means to use the function return code */
if (return_code == ONPLAY_OK)
return func_return;
return return_code;
}
}
/* no valid hotkey set, ignore hotkey */
return ONPLAY_RELOAD_DIR;
}
#endif /* HOTKEY */
int onplay(char* file, int attr, int from, bool hotkey)
{
const struct menu_item_ex *menu;
onplay_result = ONPLAY_OK;
context = from;
if (file == NULL)
selected_file = NULL;
else
{
strlcpy(selected_file_path, file, MAX_PATH);
selected_file = selected_file_path;
}
selected_file_attr = attr;
int menu_selection;
#ifdef HAVE_HOTKEY
if (hotkey)
return execute_hotkey(context == CONTEXT_WPS);
#else
(void)hotkey;
#endif
push_current_activity(ACTIVITY_CONTEXTMENU);
if (context == CONTEXT_WPS)
menu = &wps_onplay_menu;
else
menu = &tree_onplay_menu;
menu_selection = do_menu(menu, NULL, NULL, false);
pop_current_activity();
switch (menu_selection)
{
case GO_TO_WPS:
return ONPLAY_START_PLAY;
case GO_TO_ROOT:
case GO_TO_MAINMENU:
return ONPLAY_MAINMENU;
case GO_TO_PLAYLIST_VIEWER:
return ONPLAY_PLAYLIST;
case GO_TO_PLUGIN:
return ONPLAY_PLUGIN;
default:
return onplay_result;
}
}