rockbox/utils/imxtools/sbtools/dbparser.c
Amaury Pouly 8785a499f4 imxtools/efltosb: fix extern() in db files
Change-Id: I5f1acd619b2f280e709d2bd06a28a3e88cc05e27
2013-09-10 23:19:34 +02:00

999 lines
30 KiB
C

/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
* $Id$
*
* Copyright (C) 2011 Amaury Pouly
*
* 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.
*
****************************************************************************/
#define _POSIX_C_SOURCE 200809L /* for strdup */
#include <stdio.h>
#include <ctype.h>
#include <stdint.h>
#include <string.h>
#include "dbparser.h"
#include "misc.h"
enum lexem_type_t
{
LEX_IDENTIFIER,
LEX_LPAREN,
LEX_RPAREN,
LEX_NUMBER,
LEX_STRING, /* double-quoted string */
LEX_EQUAL,
LEX_SEMICOLON,
LEX_LBRACE,
LEX_RBRACE,
LEX_RANGLE,
LEX_OR,
LEX_LSHIFT,
LEX_COLON,
LEX_LE,
LEX_EOF
};
struct lexem_t
{
enum lexem_type_t type;
/* if str is not NULL, it must be a malloc'd pointer */
char *str;
uint32_t num;
int line;
const char *file;
};
struct context_t
{
const char *file;
char *begin;
char *end;
char *ptr;
int line;
};
#define parse_error(ctx, ...) \
do { fprintf(stderr, "%s:%d: ", ctx->file, ctx->line); \
fprintf(stderr, __VA_ARGS__); exit(2); } while(0)
static void advance(struct context_t *ctx, int nr_chars)
{
while(nr_chars--)
{
if(*(ctx->ptr++) == '\n')
ctx->line++;
}
}
static inline bool eof(struct context_t *ctx)
{
return ctx->ptr == ctx->end;
}
static inline bool next_valid(struct context_t *ctx, int nr)
{
return ctx->ptr + nr < ctx->end;
}
static inline char cur_char(struct context_t *ctx)
{
return *ctx->ptr;
}
static inline char next_char(struct context_t *ctx, int nr)
{
return ctx->ptr[nr];
}
static inline void locate_lexem(struct lexem_t *lex, struct context_t *ctx)
{
lex->file = ctx->file;
lex->line = ctx->line;
}
static void __parse_string(struct context_t *ctx, void *user, void (*emit_fn)(void *user, char c))
{
while(!eof(ctx))
{
if(cur_char(ctx) == '"')
break;
else if(cur_char(ctx) == '\\')
{
advance(ctx, 1);
if(eof(ctx))
parse_error(ctx, "Unfinished string\n");
if(cur_char(ctx) == '\\') emit_fn(user, '\\');
else if(cur_char(ctx) == '\'') emit_fn(user, '\'');
else if(cur_char(ctx) == '\"') emit_fn(user, '\"');
else parse_error(ctx, "Unknown escape sequence \\%c\n", cur_char(ctx));
advance(ctx, 1);
}
else
{
emit_fn(user, cur_char(ctx));
advance(ctx, 1);
}
}
if(eof(ctx) || cur_char(ctx) != '"')
parse_error(ctx, "Unfinished string\n");
advance(ctx, 1);
}
static void __parse_string_emit(void *user, char c)
{
char **pstr = (char **)user;
*(*pstr)++ = c;
}
static void __parse_string_count(void *user, char c)
{
(void) c;
(*(int *)user)++;
}
static void parse_string(struct context_t *ctx, struct lexem_t *lexem)
{
locate_lexem(lexem, ctx);
/* skip " */
advance(ctx, 1);
/* compute length */
struct context_t cpy_ctx = *ctx;
int length = 0;
__parse_string(&cpy_ctx, (void *)&length, __parse_string_count);
/* parse again */
lexem->type = LEX_STRING;
lexem->str = xmalloc(length + 1);
lexem->str[length] = 0;
char *pstr = lexem->str;
__parse_string(ctx, (void *)&pstr, __parse_string_emit);
}
static void parse_ascii_number(struct context_t *ctx, struct lexem_t *lexem)
{
locate_lexem(lexem, ctx);
/* skip ' */
advance(ctx, 1);
/* we expect n<=4 character and then ' */
int len = 0;
uint32_t value = 0;
while(!eof(ctx))
{
if(cur_char(ctx) != '\'')
{
value = value << 8 | cur_char(ctx);
len++;
advance(ctx, 1);
}
else
break;
}
if(eof(ctx) || cur_char(ctx) != '\'')
parse_error(ctx, "Unterminated ascii number literal\n");
if(len == 0 || len > 4)
parse_error(ctx, "Invalid ascii number literal length: only 1 to 4 characters allowed\n");
/* skip ' */
advance(ctx, 1);
lexem->type = LEX_NUMBER;
lexem->num = value;
}
static void parse_number(struct context_t *ctx, struct lexem_t *lexem)
{
locate_lexem(lexem, ctx);
/* check base */
int base = 10;
if(cur_char(ctx) == '0' && next_valid(ctx, 1) && next_char(ctx, 1) == 'x')
{
advance(ctx, 2);
base = 16;
}
lexem->type = LEX_NUMBER;
lexem->num = 0;
while(!eof(ctx) && isxdigit(cur_char(ctx)))
{
if(base == 10 && !isdigit(cur_char(ctx)))
break;
byte v;
if(convxdigit(cur_char(ctx), &v))
break;
lexem->num = base * lexem->num + v;
advance(ctx, 1);
}
}
static void parse_identifier(struct context_t *ctx, struct lexem_t *lexem)
{
locate_lexem(lexem, ctx);
/* remember position */
char *old = ctx->ptr;
while(!eof(ctx) && (isalnum(cur_char(ctx)) || cur_char(ctx) == '_'))
advance(ctx, 1);
lexem->type = LEX_IDENTIFIER;
int len = ctx->ptr - old;
lexem->str = xmalloc(len + 1);
lexem->str[len] = 0;
memcpy(lexem->str, old, len);
}
static void next_lexem(struct context_t *ctx, struct lexem_t *lexem)
{
#define ret_simple(t, adv) \
do {locate_lexem(lexem, ctx); \
lexem->type = t; \
advance(ctx, adv); \
return;} while(0)
while(!eof(ctx))
{
char c = cur_char(ctx);
/* skip whitespace */
if(c == ' ' || c == '\t' || c == '\n' || c == '\r')
{
advance(ctx, 1);
continue;
}
/* skip C++ style comments */
if(c == '/' && next_valid(ctx, 1) && next_char(ctx, 1) == '/')
{
while(!eof(ctx) && cur_char(ctx) != '\n')
advance(ctx, 1);
continue;
}
/* skip C-style comments */
if(c == '/' && next_valid(ctx, 1) && next_char(ctx, 1) == '*')
{
advance(ctx, 2);
while(true)
{
if(!next_valid(ctx, 1))
parse_error(ctx, "Unterminated comment");
if(cur_char(ctx) == '*' && next_char(ctx, 1) == '/')
{
advance(ctx, 2);
break;
}
advance(ctx, 1);
}
continue;
}
break;
}
if(eof(ctx)) ret_simple(LEX_EOF, 0);
char c = cur_char(ctx);
bool nv = next_valid(ctx, 1);
char nc = nv ? next_char(ctx, 1) : 0;
if(c == '(') ret_simple(LEX_LPAREN, 1);
if(c == ')') ret_simple(LEX_RPAREN, 1);
if(c == '{') ret_simple(LEX_LBRACE, 1);
if(c == '}') ret_simple(LEX_RBRACE, 1);
if(c == '>') ret_simple(LEX_RANGLE, 1);
if(c == '=') ret_simple(LEX_EQUAL, 1);
if(c == ';') ret_simple(LEX_SEMICOLON, 1);
if(c == ',') ret_simple(LEX_COLON, 1);
if(c == '|') ret_simple(LEX_OR, 1);
if(c == '<' && nv && nc == '<') ret_simple(LEX_LSHIFT, 2);
if(c == '<' && nv && nc == '=') ret_simple(LEX_LE, 2);
if(c == '"') return parse_string(ctx, lexem);
if(c == '\'') return parse_ascii_number(ctx, lexem);
if(isdigit(c)) return parse_number(ctx, lexem);
if(isalpha(c) || c == '_') return parse_identifier(ctx, lexem);
parse_error(ctx, "Unexpected character '%c'\n", c);
#undef ret_simple
}
#if 0
static void log_lexem(struct lexem_t *lexem)
{
switch(lexem->type)
{
case LEX_EOF: printf("<eof>"); break;
case LEX_EQUAL: printf("="); break;
case LEX_IDENTIFIER: printf("id(%s)", lexem->str); break;
case LEX_LPAREN: printf("("); break;
case LEX_RPAREN: printf(")"); break;
case LEX_LBRACE: printf("{"); break;
case LEX_RBRACE: printf("}"); break;
case LEX_SEMICOLON: printf(";"); break;
case LEX_NUMBER: printf("num(%d)", lexem->num); break;
case LEX_STRING: printf("str(%s)", lexem->str); break;
case LEX_OR: printf("|"); break;
case LEX_LSHIFT: printf("<<"); break;
default: printf("<unk>");
}
}
#endif
struct cmd_option_t *db_add_opt(struct cmd_option_t **opt, const char *identifier, bool is_str)
{
while(*opt)
opt = &(*opt)->next;
*opt = xmalloc(sizeof(struct cmd_option_t));
memset(*opt, 0, sizeof(struct cmd_option_t));
(*opt)->name = strdup(identifier);
(*opt)->is_string = is_str;
return *opt;
}
void db_add_str_opt(struct cmd_option_t **opt, const char *name, const char *value)
{
db_add_opt(opt, name, true)->str = strdup(value);
}
void db_add_int_opt(struct cmd_option_t **opt, const char *name, uint32_t value)
{
db_add_opt(opt, name, false)->val = value;
}
static struct cmd_source_t *db_add_src(struct cmd_source_t **src, const char *identifier, bool is_extern)
{
while(*src)
src = &(*src)->next;
*src = xmalloc(sizeof(struct cmd_source_t));
memset(*src, 0, sizeof(struct cmd_source_t));
(*src)->identifier = strdup(identifier);
(*src)->is_extern = is_extern;
return *src;
}
void db_add_source(struct cmd_file_t *cmd_file, const char *identifier, const char *filename)
{
db_add_src(&cmd_file->source_list, identifier, false)->filename = strdup(filename);
}
void db_add_extern_source(struct cmd_file_t *cmd_file, const char *identifier, int extern_nr)
{
db_add_src(&cmd_file->source_list, identifier, true)->extern_nr = extern_nr;
}
static struct cmd_inst_t *db_add_inst(struct cmd_inst_t **list, enum cmd_inst_type_t type,
uint32_t argument)
{
while(*list)
list = &(*list)->next;
*list = xmalloc(sizeof(struct cmd_inst_t));
memset(*list, 0, sizeof(struct cmd_inst_t));
(*list)->type = type;
(*list)->argument = argument;
return *list;
}
void db_add_inst_id(struct cmd_section_t *cmd_section, enum cmd_inst_type_t type,
const char *identifier, uint32_t argument)
{
db_add_inst(&cmd_section->inst_list, type, argument)->identifier = strdup(identifier);
}
void db_add_inst_addr(struct cmd_section_t *cmd_section, enum cmd_inst_type_t type,
uint32_t addr, uint32_t argument)
{
db_add_inst(&cmd_section->inst_list, type, argument)->addr = addr;
}
struct cmd_section_t *db_add_section(struct cmd_file_t *cmd_file, uint32_t identifier, bool data)
{
struct cmd_section_t **prev = &cmd_file->section_list;
while(*prev)
prev = &(*prev)->next;
*prev = xmalloc(sizeof(struct cmd_section_t));
memset(*prev, 0, sizeof(struct cmd_section_t));
(*prev)->identifier = identifier;
(*prev)->is_data = data;
return *prev;
}
struct cmd_source_t *db_find_source_by_id(struct cmd_file_t *cmd_file, const char *id)
{
struct cmd_source_t *src = cmd_file->source_list;
while(src)
{
if(strcmp(src->identifier, id) == 0)
return src;
src = src->next;
}
return NULL;
}
struct cmd_option_t *db_find_option_by_id(struct cmd_option_t *opt, const char *name)
{
while(opt)
{
if(strcmp(opt->name, name) == 0)
return opt;
opt = opt->next;
}
return NULL;
}
#define INVALID_SB_SUBVERSION 0xffff
static const char *parse_sb_subversion(const char *str, uint16_t *v)
{
int len = 0;
*v = 0;
while(isdigit(str[len]) && len < 3)
*v = (*v) << 4 | (str[len++] - '0');
if(len == 0)
*v = INVALID_SB_SUBVERSION;
return str + len;
}
bool db_parse_sb_version(struct sb_version_t *ver, const char *str)
{
str = parse_sb_subversion(str, &ver->major);
if(ver->major == INVALID_SB_SUBVERSION || *str != '.')
return false;
str = parse_sb_subversion(str + 1, &ver->minor);
if(ver->minor == INVALID_SB_SUBVERSION || *str != '.')
return false;
str = parse_sb_subversion(str + 1, &ver->revision);
if(ver->revision == INVALID_SB_SUBVERSION || *str != 0)
return false;
return true;
}
static bool db_generate_sb_subversion(uint16_t subver, char *str)
{
str[0] = '0' + ((subver >> 8) & 0xf);
str[1] = '0' + ((subver >> 4) & 0xf);
str[2] = '0' + (subver & 0xf);
return true;
}
bool db_generate_sb_version(struct sb_version_t *ver, char *str, int size)
{
if(size < 12)
return false;
str[3] = '.';
str[7] = '.';
str[11] = 0;
return db_generate_sb_subversion(ver->major, str) &&
db_generate_sb_subversion(ver->minor, str + 4) &&
db_generate_sb_subversion(ver->revision, str + 8);
}
#undef parse_error
#define parse_error(lexem, ...) \
do { fprintf(stderr, "%s:%d: ", lexem.file, lexem.line); \
fprintf(stderr, __VA_ARGS__); exit(2); } while(0)
struct lex_ctx_t
{
struct context_t ctx;
struct lexem_t lexem;
};
/* When lexems hold strings (like identifier), it might be useful to steal
* the pointer and don't clean the lexem but in other case, one don't want
* to keep the pointer to the string and just want to release the memory.
* Thus clean_lexem should be true except when one keeps a pointer */
static inline void next(struct lex_ctx_t *ctx, bool clean_lexem)
{
if(clean_lexem)
free(ctx->lexem.str);
memset(&ctx->lexem, 0, sizeof(struct lexem_t));
next_lexem(&ctx->ctx, &ctx->lexem);
}
static uint32_t parse_term_expr(struct lex_ctx_t *ctx, struct cmd_option_t *const_list)
{
uint32_t ret = 0;
if(ctx->lexem.type == LEX_NUMBER)
ret = ctx->lexem.num;
else if(ctx->lexem.type == LEX_IDENTIFIER)
{
struct cmd_option_t *c = db_find_option_by_id(const_list, ctx->lexem.str);
if(c == NULL)
parse_error(ctx->lexem, "Undefined reference to constant '%s'\n", ctx->lexem.str);
if(c->is_string)
parse_error(ctx->lexem, "Internal error: constant '%s' is not an integer\n", ctx->lexem.str);
ret = c->val;
}
else
parse_error(ctx->lexem, "Number or constant identifier expected\n");
next(ctx, true);
return ret;
}
static uint32_t parse_shift_expr(struct lex_ctx_t *ctx, struct cmd_option_t *const_list)
{
uint32_t v = parse_term_expr(ctx, const_list);
while(ctx->lexem.type == LEX_LSHIFT)
{
next(ctx, true);
v <<= parse_term_expr(ctx, const_list);
}
return v;
}
static uint32_t parse_or_expr(struct lex_ctx_t *ctx, struct cmd_option_t *const_list)
{
uint32_t v = parse_shift_expr(ctx, const_list);
while(ctx->lexem.type == LEX_OR)
{
next(ctx, true);
v |= parse_shift_expr(ctx, const_list);
}
return v;
}
static uint32_t parse_intexpr(struct lex_ctx_t *ctx, struct cmd_option_t *const_list)
{
return parse_or_expr(ctx, const_list);
}
#define NR_INITIAL_CONSTANTS 4
static char *init_const_name[NR_INITIAL_CONSTANTS] = {"true", "false", "yes", "no"};
static uint32_t init_const_value[NR_INITIAL_CONSTANTS] = {1, 0, 1, 0};
struct cmd_file_t *db_parse_file(const char *file)
{
size_t size;
FILE *f = fopen(file, "r");
if(f == NULL)
{
if(g_debug)
perror("Cannot open db file");
return NULL;
}
fseek(f, 0, SEEK_END);
size = ftell(f);
fseek(f, 0, SEEK_SET);
char *buf = xmalloc(size);
if(fread(buf, size, 1, f) != 1)
{
if(g_debug)
perror("Cannot read db file");
return NULL;
}
fclose(f);
if(g_debug)
printf("Parsing db file '%s'\n", file);
struct cmd_file_t *cmd_file = xmalloc(sizeof(struct cmd_file_t));
memset(cmd_file, 0, sizeof(struct cmd_file_t));
/* add initial constants */
for(int i = 0; i < NR_INITIAL_CONSTANTS; i++)
db_add_int_opt(&cmd_file->constant_list, init_const_name[i], init_const_value[i]);
struct lex_ctx_t lctx;
lctx.ctx.file = file;
lctx.ctx.line = 1;
lctx.ctx.begin = buf;
lctx.ctx.ptr = buf;
lctx.ctx.end = buf + size;
#define next(clean_lexem) next(&lctx, clean_lexem)
#define lexem lctx.lexem
/* init lexer */
next(false); /* don't clean init lexem because it doesn't exist */
/* constants ? */
if(lexem.type == LEX_IDENTIFIER && !strcmp(lexem.str, "constants"))
{
next(true);
if(lexem.type != LEX_LBRACE)
parse_error(lexem, "'{' expected after 'constants'\n");
while(true)
{
next(true);
if(lexem.type == LEX_RBRACE)
break;
if(lexem.type != LEX_IDENTIFIER)
parse_error(lexem, "Identifier expected in constants\n");
const char *name = lexem.str;
next(false); /* lexem string is kept as option name */
if(lexem.type != LEX_EQUAL)
parse_error(lexem, "'=' expected after identifier\n");
next(true);
db_add_int_opt(&cmd_file->constant_list, name, parse_intexpr(&lctx, cmd_file->constant_list));
if(lexem.type != LEX_SEMICOLON)
parse_error(lexem, "';' expected after string\n");
}
next(true);
}
/* options ? */
if(lexem.type == LEX_IDENTIFIER && !strcmp(lexem.str, "options"))
{
next(true);
if(lexem.type != LEX_LBRACE)
parse_error(lexem, "'{' expected after 'options'\n");
while(true)
{
next(true);
if(lexem.type == LEX_RBRACE)
break;
if(lexem.type != LEX_IDENTIFIER)
parse_error(lexem, "Identifier expected in options\n");
const char *name = lexem.str;
next(false); /* lexem string is kept as option name */
if(lexem.type != LEX_EQUAL)
parse_error(lexem, "'=' expected after identifier\n");
next(true);
if(lexem.type == LEX_STRING)
{
db_add_str_opt(&cmd_file->opt_list, name, lexem.str);
next(true);
}
else
db_add_int_opt(&cmd_file->opt_list, name, parse_intexpr(&lctx, cmd_file->constant_list));
if(lexem.type != LEX_SEMICOLON)
parse_error(lexem, "';' expected after string\n");
}
next(true);
}
/* sources */
if(lexem.type != LEX_IDENTIFIER || strcmp(lexem.str, "sources"))
parse_error(lexem, "'sources' expected\n");
next(true);
if(lexem.type != LEX_LBRACE)
parse_error(lexem, "'{' expected after 'sources'\n");
while(true)
{
next(true);
if(lexem.type == LEX_RBRACE)
break;
if(lexem.type != LEX_IDENTIFIER)
parse_error(lexem, "identifier expected in sources\n");
const char *srcid = lexem.str;
if(db_find_source_by_id(cmd_file, srcid) != NULL)
parse_error(lexem, "Duplicate source identifier\n");
next(false); /* lexem string is kept as source name */
if(lexem.type != LEX_EQUAL)
parse_error(lexem, "'=' expected after identifier\n");
next(true);
if(lexem.type == LEX_STRING)
{
db_add_source(cmd_file, srcid, lexem.str);
next(true);
}
else if(lexem.type == LEX_IDENTIFIER && !strcmp(lexem.str, "extern"))
{
next(true);
if(lexem.type != LEX_LPAREN)
parse_error(lexem, "'(' expected after 'extern'\n");
next(true);
db_add_extern_source(cmd_file, srcid, parse_intexpr(&lctx, cmd_file->constant_list));
if(lexem.type != LEX_RPAREN)
parse_error(lexem, "')' expected\n");
next(true);
}
else
parse_error(lexem, "String or 'extern' expected after '='\n");
if(lexem.type != LEX_SEMICOLON)
parse_error(lexem, "';' expected\n");
}
/* sections */
while(true)
{
next(true);
if(lexem.type == LEX_EOF)
break;
if(lexem.type != LEX_IDENTIFIER || strcmp(lexem.str, "section") != 0)
parse_error(lexem, "'section' expected\n");
next(true);
if(lexem.type != LEX_LPAREN)
parse_error(lexem, "'(' expected after 'section'\n");
next(true);
/* can be any number */
struct cmd_section_t *sec = db_add_section(cmd_file, parse_intexpr(&lctx, cmd_file->constant_list), false);
/* options ? */
if(lexem.type == LEX_SEMICOLON)
{
do
{
next(true);
if(lexem.type != LEX_IDENTIFIER)
parse_error(lexem, "Identifier expected for section option\n");
const char *name = lexem.str;
next(false); /* lexem string is kept as option name */
if(lexem.type != LEX_EQUAL)
parse_error(lexem, "'=' expected after option identifier\n");
next(true);
if(lexem.type == LEX_STRING)
{
db_add_str_opt(&sec->opt_list, name, lexem.str);
next(true);
}
else
db_add_int_opt(&sec->opt_list, name, parse_intexpr(&lctx, cmd_file->constant_list));
}while(lexem.type == LEX_COLON);
}
if(lexem.type != LEX_RPAREN)
parse_error(lexem, "')' expected after section identifier\n");
next(true);
if(lexem.type == LEX_LBRACE)
{
sec->is_data = false;
/* commands */
while(true)
{
next(true);
if(lexem.type == LEX_RBRACE)
break;
struct cmd_inst_t *inst = db_add_inst(&sec->inst_list, CMD_LOAD, 0);
if(lexem.type != LEX_IDENTIFIER)
parse_error(lexem, "Instruction expected in section\n");
if(strcmp(lexem.str, "load") == 0)
inst->type = CMD_LOAD;
else if(strcmp(lexem.str, "call") == 0)
inst->type = CMD_CALL;
else if(strcmp(lexem.str, "jump") == 0)
inst->type = CMD_JUMP;
else if(strcmp(lexem.str, "mode") == 0)
inst->type = CMD_MODE;
else
parse_error(lexem, "Instruction expected in section\n");
next(true);
if(inst->type == CMD_LOAD)
{
if(lexem.type != LEX_IDENTIFIER)
parse_error(lexem, "Identifier expected after instruction\n");
inst->identifier = lexem.str;
if(db_find_source_by_id(cmd_file, inst->identifier) == NULL)
parse_error(lexem, "Undefined reference to source '%s'\n", inst->identifier);
next(false); /* lexem string kept as identifier */
if(lexem.type == LEX_RANGLE)
{
// load at
inst->type = CMD_LOAD_AT;
next(true);
inst->addr = parse_intexpr(&lctx, cmd_file->constant_list);
}
if(lexem.type != LEX_SEMICOLON)
parse_error(lexem, "';' expected after command\n");
}
else if(inst->type == CMD_CALL || inst->type == CMD_JUMP)
{
if(lexem.type == LEX_IDENTIFIER)
{
inst->identifier = lexem.str;
if(db_find_source_by_id(cmd_file, inst->identifier) == NULL)
parse_error(lexem, "Undefined reference to source '%s'\n", inst->identifier);
next(false); /* lexem string kept as identifier */
}
else
{
inst->type = (inst->type == CMD_CALL) ? CMD_CALL_AT : CMD_JUMP_AT;
inst->addr = parse_intexpr(&lctx, cmd_file->constant_list);
}
if(lexem.type == LEX_LPAREN)
{
next(true);
inst->argument = parse_intexpr(&lctx, cmd_file->constant_list);
if(lexem.type != LEX_RPAREN)
parse_error(lexem, "Expected closing brace\n");
next(true);
}
if(lexem.type != LEX_SEMICOLON)
parse_error(lexem, "';' expected after command\n");
}
else if(inst->type == CMD_MODE)
{
inst->argument = parse_intexpr(&lctx, cmd_file->constant_list);
if(lexem.type != LEX_SEMICOLON)
parse_error(lexem, "Expected ';' after command\n");
}
else
parse_error(lexem, "Internal error");
}
}
else if(lexem.type == LEX_LE)
{
sec->is_data = true;
next(true);
if(lexem.type != LEX_IDENTIFIER)
parse_error(lexem, "Identifier expected after '<='\n");
sec->source_id = lexem.str;
next(false); /* lexem string is kept as source id */
if(lexem.type != LEX_SEMICOLON)
parse_error(lexem, "';' expected after identifier\n");
}
else
parse_error(lexem, "'{' or '<=' expected after section directive\n");
}
#undef lexem
#undef next
free(buf);
return cmd_file;
}
void db_free_option_list(struct cmd_option_t *opt_list)
{
while(opt_list)
{
struct cmd_option_t *next = opt_list->next;
fflush(stdout);
free(opt_list->name);
free(opt_list->str);
free(opt_list);
opt_list = next;
}
}
static bool db_generate_options(FILE *f, const char *secname, struct cmd_option_t *list)
{
fprintf(f, "%s\n", secname);
fprintf(f, "{\n");
while(list)
{
fprintf(f, " %s = ", list->name);
if(list->is_string)
fprintf(f, "\"%s\";\n", list->str); // FIXME handle escape
else
fprintf(f, "0x%x;\n", list->val);
list = list->next;
}
fprintf(f, "}\n");
return true;
}
static bool db_generate_section_options(FILE *f, struct cmd_option_t *list)
{
bool first = true;
while(list)
{
fprintf(f, "%c %s = ", first ? ';' : ',', list->name);
if(list->is_string)
fprintf(f, "\"%s\"", list->str); // FIXME handle escape
else
fprintf(f, "0x%x", list->val);
first = false;
list = list->next;
}
return true;
}
static bool db_generate_sources(FILE *f, struct cmd_source_t *list)
{
fprintf(f, "sources\n"),
fprintf(f, "{\n");
while(list)
{
fprintf(f, " %s = ", list->identifier);
if(list->is_extern)
fprintf(f, "extern(%d);\n", list->extern_nr);
else
fprintf(f, "\"%s\";\n", list->filename); // FIXME handle escape
list = list->next;
}
fprintf(f, "}\n");
return true;
}
static bool db_generate_section(FILE *f, struct cmd_section_t *section)
{
fprintf(f, "section(%#x", section->identifier);
db_generate_section_options(f, section->opt_list);
if(section->is_data)
{
fprintf(f, ") <= %s;\n", section->source_id);
return true;
}
fprintf(f, ")\n{\n");
struct cmd_inst_t *inst = section->inst_list;
while(inst)
{
fprintf(f, " ");
switch(inst->type)
{
case CMD_LOAD:
fprintf(f, "load %s;\n", inst->identifier);
break;
case CMD_LOAD_AT:
fprintf(f, "load %s > %#x;\n", inst->identifier, inst->addr);
break;
case CMD_CALL:
fprintf(f, "call %s(%#x);\n", inst->identifier, inst->argument);
break;
case CMD_CALL_AT:
fprintf(f, "call %#x(%#x);\n", inst->addr, inst->argument);
break;
case CMD_JUMP:
fprintf(f, "jump %s(%#x);\n", inst->identifier, inst->argument);
break;
case CMD_JUMP_AT:
fprintf(f, "jump %#x(%#x);\n", inst->addr, inst->argument);
break;
case CMD_MODE:
fprintf(f, "mode %#x;\n", inst->argument);
break;
default:
bug("die");
}
inst = inst->next;
}
fprintf(f, "}\n");
return true;
}
static bool db_generate_sections(FILE *f, struct cmd_section_t *section)
{
while(section)
if(!db_generate_section(f, section))
return false;
else
section = section->next;
return true;
}
bool db_generate_file(struct cmd_file_t *file, const char *filename, void *user, db_color_printf printf)
{
FILE *f = fopen(filename, "w");
if(f == NULL)
return printf(user, true, GREY, "Cannot open '%s' for writing: %m\n", filename), false;
if(!db_generate_options(f, "constants", file->constant_list))
goto Lerr;
if(!db_generate_options(f, "options", file->opt_list))
goto Lerr;
if(!db_generate_sources(f, file->source_list))
goto Lerr;
if(!db_generate_sections(f, file->section_list))
goto Lerr;
fclose(f);
return true;
Lerr:
fclose(f);
return false;
}
void db_free(struct cmd_file_t *file)
{
db_free_option_list(file->opt_list);
db_free_option_list(file->constant_list);
struct cmd_source_t *src = file->source_list;
while(src)
{
struct cmd_source_t *next = src->next;
free(src->identifier);
fflush(stdout);
free(src->filename);
if(src->loaded)
{
if(src->type == CMD_SRC_BIN)
free(src->bin.data);
if(src->type == CMD_SRC_ELF)
elf_release(&src->elf);
}
free(src);
src = next;
}
struct cmd_section_t *sec = file->section_list;
while(sec)
{
struct cmd_section_t *next = sec->next;
db_free_option_list(sec->opt_list);
free(sec->source_id);
struct cmd_inst_t *inst = sec->inst_list;
while(inst)
{
struct cmd_inst_t *next = inst->next;
free(inst->identifier);
free(inst);
inst = next;
}
free(sec);
sec = next;
}
free(file);
}