I can only assume in the course of the original conversion to fixed point
math in RbLua the fact that division by zero and NaN handling was to be
caught as a graceful exception by the floating point handler was overlooked.
As a result lua doesn't handle these exceptions and instead results in a
panic on the device.
This patch fixes this handling in the lexer for compile time Inf and Nan
results and in the luavm for runtime division by zero (Inf)
I missed the runtime exception of n%0 added checks for that as well..
Change-Id: I7746c087ea93678e5875f15ec6fe3f29f26bdb31
Fixes action fall through of select|release on fuze during FMS screen
allowing detection of unmapped button
Change-Id: I53945766263d8710f31621a4c9ad50cec7be2b0d
When long pressing PLAY/PAUSE and FM_MODE on devices that use the clip keymap
the volume changes in addition to toggling PLAY/PAUSE and SCAN/SEEK,this patch
removes the volume functionality as these devices provide dedicated volume
buttons
Change-Id: I3ab5af6da07900417c31c7a1036343b32b950e4e
Allows changing file sort options directly through context menu in the file browser
rather than having to exit the browser and go through General Settings>File_View
Change-Id: Iec124d512dcb934bd261b2144a699c23f536c165
lua would not return or set arbitrary file positions
file:seek("set", 0) worked file:seek("cur") worked
but setting an offset or file:seek("end") failed
I tracked this down to a bug checking the return of rb->lseek
on error lseek returns a negative number and returns the file
position otherwise, the function was checking for if(N) instead of
if(N < 0)
Fixed - limited size of lseek to size of signed LuaNumber
Fixed - io:lines() stopped after first line containing only a newline
instead of returning a blank line and continuing till EOF
this fixes file:read("*l") as well
Fixed - ssize_t for read() with error checking
Change-Id: Ie859b288fb8f6814f1b3ae30992ecf78f2669de7
The upstream code changed a little bit with regard to the request_keys()
API. Also, we save some bytes (especially on the c200v2) by compiling with
-ffunction-sections and -fdata-sections, which allows Net to fit once again.
Change-Id: I3ab30127169c73e4cd8996f0c12e1223ee18d79f
This includes an upstream change to the Galaxies help text. `genhelp.sh'
no longer leaves temporary files sitting around, and the self-test feature
of lz4tiny.c works again.
Change-Id: I787f4cb3c258baade31638d6be18f95b7aa0705e
This brings the upstream version to b3da238 (though some of my own
changes are included on top of that).
Change-Id: Ida73e8cd86765413147ce891af3cc2b7aeda2b2a
This should cut just enough size off the binary to let Net still compile
for the c200v2. This is probably close to the last time I'll be able to do
this; I'm already resorting to dirty hacks like writing super terse error
messages.
Change-Id: I43344b9a601696d7ca56fc02af4a611fd9d1a150
They all work now :). I merged in part of Chris Boyle's Android port of
Puzzles to give the front end a way to know what keys the back end needs.
This also re-syncs to the latest upstream sources.
Change-Id: Ie0409bbb32f617ae5abf4f81be3b45d1552db9bb
The help text is now processed to generate a style array to pass to the
display_text library in addition to the text itself. The help text is still
compressed using LZ4, and still fits on the c200v2.
Change-Id: I7a3a664f90f67a1a018956c72d2b62d92b8ffd17
This prevents a (highly improbable) race condition when exiting, and uses
pcm_set_frequency() instead of mixer_set_frequency(), since that seems like
the Right Thing to Do (TM).
Change-Id: Icb6a4597843215f08b3835490ac63f67c9a04736
Rather than holding intermediate results as fixed-point, this converts them
directly to normal integers (in the range of the PCM sample) while mixing,
instead of waiting till the end to perform a separate shifting step. Also,
this precalculates some constants in the reverb code.
Change-Id: Ie04e444d145bc28ce67eef9ae0ead6d328acf28a
This allows CPU load to drop below 100%, which is good for anything
that scales frequency and voltage based on CPU load. Also conserves
some energy by letting the core go idle if there aren't any available
buffers.
Change-Id: I9385ac9e030f97010b12eb825875a900463ab0ac
Using the coprocessor was a good idea in theory, but didn't actually work.
This moves all SDL threads to the main core, which isn't ideal, but at
least works. Additionally, this also adds some good-practice stuff such as
setting the default sample rate on exit.
Change-Id: If1636b33d439000ec7c4e50f46e8d002708d3121
This is attempt to simplify porting programs to rockbox (as plugins).
Currently this compat layer implements:
fopen(), fclose(), fflush(), fread(), fwrite(), fseek(), fseeko(),
ftell(), ftello(), fgetc(), ungetc(), fputc(), fgets(), clearerr(),
ferror(), feof(), fprintf()
In order to use it you need to include in ported sources
"lib/stdio_compat.h"
Change-Id: I5add615dd19c5af9c767ccbfb1bd5a4e466741cb
I removed the mechanism to fall through to internally mapped contexts
after plugin contexts were exhausted
Change-Id: Id287248184fd67cb2a2242475296247ac86af807
Playback checked the files' presence before attempting to buffer
the track. Just get rid of that and save an extra open/close call.
It will find out if the path is bad when the metadata fails.
Fix some size_t/off_t conflation. No need to update plugin version
because no plugin actually uses bufopen().
Change-Id: I3db112449dc0b2eeb91c546f308880ac82494fc7
It must be set to something else valid (unless it's the only one
left) when closing it, IN ALL CASES, not just if it's first or last.
Don't know what was in my head. Hopefully takes care of a reported
issue. Even if it's not causing any issues, it was still incorrect.
Change-Id: I594af8b35d774ec222dadce80dfa8b95138f037e
