457 lines
14 KiB
C
457 lines
14 KiB
C
/*
|
|
* Utility routines
|
|
*
|
|
* Copyright 1998 Bertho A. Stultiens
|
|
*
|
|
* This library is free software; you can redistribute it and/or
|
|
* modify it under the terms of the GNU Lesser General Public
|
|
* License as published by the Free Software Foundation; either
|
|
* version 2.1 of the License, or (at your option) any later version.
|
|
*
|
|
* This library is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
* Lesser General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU Lesser General Public
|
|
* License along with this library; if not, write to the Free Software
|
|
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
|
|
*/
|
|
|
|
#include "config.h"
|
|
#include "wine/port.h"
|
|
|
|
#include <assert.h>
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <stdarg.h>
|
|
#include <string.h>
|
|
#include <ctype.h>
|
|
|
|
#include "wine/unicode.h"
|
|
#include "wrc.h"
|
|
#include "utils.h"
|
|
#include "parser.h"
|
|
|
|
/* #define WANT_NEAR_INDICATION */
|
|
|
|
#ifdef WANT_NEAR_INDICATION
|
|
void make_print(char *str)
|
|
{
|
|
while(*str)
|
|
{
|
|
if(!isprint(*str))
|
|
*str = ' ';
|
|
str++;
|
|
}
|
|
}
|
|
#endif
|
|
|
|
static void generic_msg(const char *s, const char *t, const char *n, va_list ap)
|
|
{
|
|
fprintf(stderr, "%s:%d:%d: %s: ", input_name ? input_name : "stdin", line_number, char_number, t);
|
|
vfprintf(stderr, s, ap);
|
|
#ifdef WANT_NEAR_INDICATION
|
|
{
|
|
char *cpy;
|
|
if(n)
|
|
{
|
|
cpy = xstrdup(n);
|
|
make_print(cpy);
|
|
fprintf(stderr, " near '%s'", cpy);
|
|
free(cpy);
|
|
}
|
|
}
|
|
#endif
|
|
}
|
|
|
|
|
|
int parser_error(const char *s, ...)
|
|
{
|
|
va_list ap;
|
|
va_start(ap, s);
|
|
generic_msg(s, "Error", parser_text, ap);
|
|
fputc( '\n', stderr );
|
|
va_end(ap);
|
|
exit(1);
|
|
return 1;
|
|
}
|
|
|
|
int parser_warning(const char *s, ...)
|
|
{
|
|
va_list ap;
|
|
va_start(ap, s);
|
|
generic_msg(s, "Warning", parser_text, ap);
|
|
va_end(ap);
|
|
return 0;
|
|
}
|
|
|
|
void internal_error(const char *file, int line, const char *s, ...)
|
|
{
|
|
va_list ap;
|
|
va_start(ap, s);
|
|
fprintf(stderr, "Internal error (please report) %s %d: ", file, line);
|
|
vfprintf(stderr, s, ap);
|
|
va_end(ap);
|
|
exit(3);
|
|
}
|
|
|
|
void error(const char *s, ...)
|
|
{
|
|
va_list ap;
|
|
va_start(ap, s);
|
|
fprintf(stderr, "Error: ");
|
|
vfprintf(stderr, s, ap);
|
|
va_end(ap);
|
|
exit(2);
|
|
}
|
|
|
|
void warning(const char *s, ...)
|
|
{
|
|
va_list ap;
|
|
va_start(ap, s);
|
|
fprintf(stderr, "Warning: ");
|
|
vfprintf(stderr, s, ap);
|
|
va_end(ap);
|
|
}
|
|
|
|
void chat(const char *s, ...)
|
|
{
|
|
if(debuglevel & DEBUGLEVEL_CHAT)
|
|
{
|
|
va_list ap;
|
|
va_start(ap, s);
|
|
fprintf(stderr, "FYI: ");
|
|
vfprintf(stderr, s, ap);
|
|
va_end(ap);
|
|
}
|
|
}
|
|
|
|
char *dup_basename(const char *name, const char *ext)
|
|
{
|
|
int namelen;
|
|
int extlen = strlen(ext);
|
|
char *base;
|
|
char *slash;
|
|
|
|
if(!name)
|
|
name = "wrc.tab";
|
|
|
|
slash = strrchr(name, '/');
|
|
if (slash)
|
|
name = slash + 1;
|
|
|
|
namelen = strlen(name);
|
|
|
|
/* +4 for later extension and +1 for '\0' */
|
|
base = xmalloc(namelen +4 +1);
|
|
strcpy(base, name);
|
|
if(!strcasecmp(name + namelen-extlen, ext))
|
|
{
|
|
base[namelen - extlen] = '\0';
|
|
}
|
|
return base;
|
|
}
|
|
|
|
void *xmalloc(size_t size)
|
|
{
|
|
void *res;
|
|
|
|
assert(size > 0);
|
|
res = malloc(size);
|
|
if(res == NULL)
|
|
{
|
|
error("Virtual memory exhausted.\n");
|
|
}
|
|
memset(res, 0x55, size);
|
|
return res;
|
|
}
|
|
|
|
|
|
void *xrealloc(void *p, size_t size)
|
|
{
|
|
void *res;
|
|
|
|
assert(size > 0);
|
|
res = realloc(p, size);
|
|
if(res == NULL)
|
|
{
|
|
error("Virtual memory exhausted.\n");
|
|
}
|
|
return res;
|
|
}
|
|
|
|
char *xstrdup(const char *str)
|
|
{
|
|
char *s;
|
|
|
|
assert(str != NULL);
|
|
s = xmalloc(strlen(str)+1);
|
|
return strcpy(s, str);
|
|
}
|
|
|
|
|
|
/*
|
|
*****************************************************************************
|
|
* Function : compare_name_id
|
|
* Syntax : int compare_name_id(const name_id_t *n1, const name_id_t *n2)
|
|
* Input :
|
|
* Output :
|
|
* Description :
|
|
* Remarks :
|
|
*****************************************************************************
|
|
*/
|
|
int compare_name_id(const name_id_t *n1, const name_id_t *n2)
|
|
{
|
|
if(n1->type == name_ord && n2->type == name_ord)
|
|
{
|
|
return n1->name.i_name - n2->name.i_name;
|
|
}
|
|
else if(n1->type == name_str && n2->type == name_str)
|
|
{
|
|
if(n1->name.s_name->type == str_char
|
|
&& n2->name.s_name->type == str_char)
|
|
{
|
|
return strcasecmp(n1->name.s_name->str.cstr, n2->name.s_name->str.cstr);
|
|
}
|
|
else if(n1->name.s_name->type == str_unicode
|
|
&& n2->name.s_name->type == str_unicode)
|
|
{
|
|
return strcmpiW(n1->name.s_name->str.wstr, n2->name.s_name->str.wstr);
|
|
}
|
|
else
|
|
{
|
|
internal_error(__FILE__, __LINE__, "Can't yet compare strings of mixed type\n");
|
|
}
|
|
}
|
|
else if(n1->type == name_ord && n2->type == name_str)
|
|
return 1;
|
|
else if(n1->type == name_str && n2->type == name_ord)
|
|
return -1;
|
|
else
|
|
internal_error(__FILE__, __LINE__, "Comparing name-ids with unknown types (%d, %d)\n",
|
|
n1->type, n2->type);
|
|
|
|
return 0; /* Keep the compiler happy */
|
|
}
|
|
|
|
string_t *convert_string(const string_t *str, enum str_e type, int codepage)
|
|
{
|
|
const union cptable *cptable = codepage ? wine_cp_get_table( codepage ) : NULL;
|
|
string_t *ret = xmalloc(sizeof(*ret));
|
|
int res;
|
|
|
|
if (!codepage && str->type != type)
|
|
parser_error( "Current language is Unicode only, cannot convert string" );
|
|
|
|
if((str->type == str_char) && (type == str_unicode))
|
|
{
|
|
ret->type = str_unicode;
|
|
ret->size = cptable ? wine_cp_mbstowcs( cptable, 0, str->str.cstr, str->size, NULL, 0 )
|
|
: wine_utf8_mbstowcs( 0, str->str.cstr, str->size, NULL, 0 );
|
|
ret->str.wstr = xmalloc( (ret->size+1) * sizeof(WCHAR) );
|
|
if (cptable)
|
|
res = wine_cp_mbstowcs( cptable, MB_ERR_INVALID_CHARS, str->str.cstr, str->size,
|
|
ret->str.wstr, ret->size );
|
|
else
|
|
res = wine_utf8_mbstowcs( MB_ERR_INVALID_CHARS, str->str.cstr, str->size,
|
|
ret->str.wstr, ret->size );
|
|
if (res == -2)
|
|
parser_error( "Invalid character in string '%.*s' for codepage %u",
|
|
str->size, str->str.cstr, codepage );
|
|
ret->str.wstr[ret->size] = 0;
|
|
}
|
|
else if((str->type == str_unicode) && (type == str_char))
|
|
{
|
|
ret->type = str_char;
|
|
ret->size = cptable ? wine_cp_wcstombs( cptable, 0, str->str.wstr, str->size, NULL, 0, NULL, NULL )
|
|
: wine_utf8_wcstombs( 0, str->str.wstr, str->size, NULL, 0 );
|
|
ret->str.cstr = xmalloc( ret->size + 1 );
|
|
if (cptable)
|
|
wine_cp_wcstombs( cptable, 0, str->str.wstr, str->size, ret->str.cstr, ret->size, NULL, NULL );
|
|
else
|
|
wine_utf8_wcstombs( 0, str->str.wstr, str->size, ret->str.cstr, ret->size );
|
|
ret->str.cstr[ret->size] = 0;
|
|
}
|
|
else if(str->type == str_unicode)
|
|
{
|
|
ret->type = str_unicode;
|
|
ret->size = str->size;
|
|
ret->str.wstr = xmalloc(sizeof(WCHAR)*(ret->size+1));
|
|
memcpy( ret->str.wstr, str->str.wstr, ret->size * sizeof(WCHAR) );
|
|
ret->str.wstr[ret->size] = 0;
|
|
}
|
|
else /* str->type == str_char */
|
|
{
|
|
ret->type = str_char;
|
|
ret->size = str->size;
|
|
ret->str.cstr = xmalloc( ret->size + 1 );
|
|
memcpy( ret->str.cstr, str->str.cstr, ret->size );
|
|
ret->str.cstr[ret->size] = 0;
|
|
}
|
|
return ret;
|
|
}
|
|
|
|
|
|
void free_string(string_t *str)
|
|
{
|
|
if (str->type == str_unicode) free( str->str.wstr );
|
|
else free( str->str.cstr );
|
|
free( str );
|
|
}
|
|
|
|
|
|
int check_unicode_conversion( const string_t *str_a, const string_t *str_w, int codepage )
|
|
{
|
|
int ok;
|
|
string_t *teststr = convert_string( str_w, str_char, codepage );
|
|
|
|
ok = (teststr->size == str_a->size && !memcmp( teststr->str.cstr, str_a->str.cstr, str_a->size ));
|
|
|
|
if (!ok)
|
|
{
|
|
int i;
|
|
|
|
fprintf( stderr, "Source: %s", str_a->str.cstr );
|
|
for (i = 0; i < str_a->size; i++)
|
|
fprintf( stderr, " %02x", (unsigned char)str_a->str.cstr[i] );
|
|
fprintf( stderr, "\nUnicode: " );
|
|
for (i = 0; i < str_w->size; i++)
|
|
fprintf( stderr, " %04x", str_w->str.wstr[i] );
|
|
fprintf( stderr, "\nBack: %s", teststr->str.cstr );
|
|
for (i = 0; i < teststr->size; i++)
|
|
fprintf( stderr, " %02x", (unsigned char)teststr->str.cstr[i] );
|
|
fprintf( stderr, "\n" );
|
|
}
|
|
free_string( teststr );
|
|
return ok;
|
|
}
|
|
|
|
|
|
struct lang2cp
|
|
{
|
|
unsigned short lang;
|
|
unsigned short sublang;
|
|
unsigned int cp;
|
|
} lang2cp_t;
|
|
|
|
/* language to codepage conversion table */
|
|
/* specific sublanguages need only be specified if their codepage */
|
|
/* differs from the default (SUBLANG_NEUTRAL) */
|
|
static const struct lang2cp lang2cps[] =
|
|
{
|
|
{ LANG_AFRIKAANS, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_ALBANIAN, SUBLANG_NEUTRAL, 1250 },
|
|
{ LANG_ARABIC, SUBLANG_NEUTRAL, 1256 },
|
|
{ LANG_ARMENIAN, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_AZERI, SUBLANG_NEUTRAL, 1254 },
|
|
{ LANG_AZERI, SUBLANG_AZERI_CYRILLIC, 1251 },
|
|
{ LANG_BASQUE, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_BELARUSIAN, SUBLANG_NEUTRAL, 1251 },
|
|
#ifdef LANG_BRETON
|
|
{ LANG_BRETON, SUBLANG_NEUTRAL, 1252 },
|
|
#endif /* LANG_BRETON */
|
|
{ LANG_BULGARIAN, SUBLANG_NEUTRAL, 1251 },
|
|
{ LANG_CATALAN, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_CHINESE, SUBLANG_NEUTRAL, 950 },
|
|
{ LANG_CHINESE, SUBLANG_CHINESE_SINGAPORE, 936 },
|
|
{ LANG_CHINESE, SUBLANG_CHINESE_SIMPLIFIED, 936 },
|
|
#ifdef LANG_CORNISH
|
|
{ LANG_CORNISH, SUBLANG_NEUTRAL, 1252 },
|
|
#endif /* LANG_CORNISH */
|
|
{ LANG_CROATIAN, SUBLANG_NEUTRAL, 1250 },
|
|
{ LANG_CZECH, SUBLANG_NEUTRAL, 1250 },
|
|
{ LANG_DANISH, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_DIVEHI, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_DUTCH, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_ENGLISH, SUBLANG_NEUTRAL, 1252 },
|
|
#ifdef LANG_ESPERANTO
|
|
{ LANG_ESPERANTO, SUBLANG_NEUTRAL, 1252 },
|
|
#endif /* LANG_ESPERANTO */
|
|
{ LANG_ESTONIAN, SUBLANG_NEUTRAL, 1257 },
|
|
{ LANG_FAEROESE, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_FARSI, SUBLANG_NEUTRAL, 1256 },
|
|
{ LANG_FINNISH, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_FRENCH, SUBLANG_NEUTRAL, 1252 },
|
|
#ifdef LANG_GAELIC
|
|
{ LANG_GAELIC, SUBLANG_NEUTRAL, 1252 },
|
|
#endif /* LANG_GAELIC */
|
|
{ LANG_GALICIAN, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_GEORGIAN, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_GERMAN, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_GREEK, SUBLANG_NEUTRAL, 1253 },
|
|
{ LANG_GUJARATI, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_HEBREW, SUBLANG_NEUTRAL, 1255 },
|
|
{ LANG_HINDI, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_HUNGARIAN, SUBLANG_NEUTRAL, 1250 },
|
|
{ LANG_ICELANDIC, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_INDONESIAN, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_ITALIAN, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_JAPANESE, SUBLANG_NEUTRAL, 932 },
|
|
{ LANG_KANNADA, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_KAZAK, SUBLANG_NEUTRAL, 1251 },
|
|
{ LANG_KONKANI, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_KOREAN, SUBLANG_NEUTRAL, 949 },
|
|
{ LANG_KYRGYZ, SUBLANG_NEUTRAL, 1251 },
|
|
{ LANG_LATVIAN, SUBLANG_NEUTRAL, 1257 },
|
|
{ LANG_LITHUANIAN, SUBLANG_NEUTRAL, 1257 },
|
|
{ LANG_MACEDONIAN, SUBLANG_NEUTRAL, 1251 },
|
|
{ LANG_MALAY, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_MARATHI, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_MONGOLIAN, SUBLANG_NEUTRAL, 1251 },
|
|
{ LANG_NEUTRAL, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_NORWEGIAN, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_POLISH, SUBLANG_NEUTRAL, 1250 },
|
|
{ LANG_PORTUGUESE, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_PUNJABI, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_ROMANIAN, SUBLANG_NEUTRAL, 1250 },
|
|
{ LANG_ROMANSH, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_RUSSIAN, SUBLANG_NEUTRAL, 1251 },
|
|
{ LANG_SANSKRIT, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_SERBIAN, SUBLANG_NEUTRAL, 1250 },
|
|
{ LANG_SERBIAN, SUBLANG_SERBIAN_CYRILLIC, 1251 },
|
|
{ LANG_SLOVAK, SUBLANG_NEUTRAL, 1250 },
|
|
{ LANG_SLOVENIAN, SUBLANG_NEUTRAL, 1250 },
|
|
{ LANG_SPANISH, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_SWAHILI, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_SWEDISH, SUBLANG_NEUTRAL, 1252 },
|
|
{ LANG_SYRIAC, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_TAMIL, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_TATAR, SUBLANG_NEUTRAL, 1251 },
|
|
{ LANG_TELUGU, SUBLANG_NEUTRAL, 0 },
|
|
{ LANG_THAI, SUBLANG_NEUTRAL, 874 },
|
|
{ LANG_TURKISH, SUBLANG_NEUTRAL, 1254 },
|
|
{ LANG_UKRAINIAN, SUBLANG_NEUTRAL, 1251 },
|
|
{ LANG_URDU, SUBLANG_NEUTRAL, 1256 },
|
|
{ LANG_UZBEK, SUBLANG_NEUTRAL, 1254 },
|
|
{ LANG_UZBEK, SUBLANG_UZBEK_CYRILLIC, 1251 },
|
|
{ LANG_VIETNAMESE, SUBLANG_NEUTRAL, 1258 }
|
|
#ifdef LANG_WALON
|
|
, { LANG_WALON, SUBLANG_NEUTRAL, 1252 }
|
|
#endif /* LANG_WALON */
|
|
#ifdef LANG_WELSH
|
|
, { LANG_WELSH, SUBLANG_NEUTRAL, 1252 }
|
|
#endif /* LANG_WELSH */
|
|
};
|
|
|
|
int get_language_codepage( unsigned short lang, unsigned short sublang )
|
|
{
|
|
unsigned int i;
|
|
int cp = -1, defcp = -1;
|
|
|
|
for (i = 0; i < sizeof(lang2cps)/sizeof(lang2cps[0]); i++)
|
|
{
|
|
if (lang2cps[i].lang != lang) continue;
|
|
if (lang2cps[i].sublang == sublang)
|
|
{
|
|
cp = lang2cps[i].cp;
|
|
break;
|
|
}
|
|
if (lang2cps[i].sublang == SUBLANG_NEUTRAL) defcp = lang2cps[i].cp;
|
|
}
|
|
|
|
if (cp == -1) cp = defcp;
|
|
assert( cp <= 0 || wine_cp_get_table(cp) );
|
|
return cp;
|
|
}
|