diff options
Diffstat (limited to 'nx-X11/lib/font/fontfile/encparse.c.NX.original')
-rw-r--r-- | nx-X11/lib/font/fontfile/encparse.c.NX.original | 966 |
1 files changed, 0 insertions, 966 deletions
diff --git a/nx-X11/lib/font/fontfile/encparse.c.NX.original b/nx-X11/lib/font/fontfile/encparse.c.NX.original deleted file mode 100644 index 6262aff07..000000000 --- a/nx-X11/lib/font/fontfile/encparse.c.NX.original +++ /dev/null @@ -1,966 +0,0 @@ -/* -Copyright (c) 1998-2001 by Juliusz Chroboczek - -Permission is hereby granted, free of charge, to any person obtaining a copy -of this software and associated documentation files (the "Software"), to deal -in the Software without restriction, including without limitation the rights -to use, copy, modify, merge, publish, distribute, sublicense, and/or sell -copies of the Software, and to permit persons to whom the Software is -furnished to do so, subject to the following conditions: - -The above copyright notice and this permission notice shall be included in -all copies or substantial portions of the Software. - -THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR -IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, -FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE -AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER -LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, -OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN -THE SOFTWARE. -*/ - -/* $XFree86: xc/lib/font/fontfile/encparse.c,v 1.20tsi Exp $ */ - -/* Parser for encoding files */ - -/* This code assumes that we are using ASCII. We don't use the ctype - functions, as they depend on the current locale. On the other - hand, we do use strcasecmp, but only on strings that we've checked - to be pure ASCII. Bloody ``Code Set Independence''. */ - -#include <string.h> -#include <stdio.h> - -#if defined(__SCO__) || defined(__UNIXWARE__) -#include <strings.h> -#endif - -#ifndef FONTENC_NO_LIBFONT - -#include <X11/fonts/fntfilio.h> -#include <X11/fonts/fntfilst.h> - -#else - -#include <stdlib.h> -#define xalloc(n) malloc(n) -#define xrealloc(p, n) realloc(p, n) -#define xfree(p) free(p) - -#include "zlib.h" -typedef gzFile FontFilePtr; -#define FontFileGetc(f) gzgetc(f) -#define FontFileOpen(filename) gzopen(filename, "rb") -#define FontFileClose(f) gzclose(f) - -#define MAXFONTFILENAMELEN 1024 -#define MAXFONTNAMELEN 1024 - -#endif - -#include <X11/fonts/fontenc.h> -#include "fontencI.h" - -#define MAXALIASES 20 - -#define EOF_TOKEN -1 -#define ERROR_TOKEN -2 -#define EOL_TOKEN 0 -#define NUMBER_TOKEN 1 -#define KEYWORD_TOKEN 2 - -#define EOF_LINE -1 -#define ERROR_LINE -2 -#define STARTENCODING_LINE 1 -#define STARTMAPPING_LINE 2 -#define ENDMAPPING_LINE 3 -#define CODE_LINE 4 -#define CODE_RANGE_LINE 5 -#define CODE_UNDEFINE_LINE 6 -#define NAME_LINE 7 -#define SIZE_LINE 8 -#define ALIAS_LINE 9 -#define FIRSTINDEX_LINE 10 - -/* Return from lexer */ -#define MAXKEYWORDLEN 100 - -static long number_value; -static char keyword_value[MAXKEYWORDLEN+1]; - -static long value1, value2, value3; - -/* Lexer code */ - -/* Skip to the beginning of new line */ -static void -skipEndOfLine(FontFilePtr f, int c) -{ - if(c == 0) - c = FontFileGetc(f); - - for(;;) - if(c <= 0 || c == '\n') - return; - else - c = FontFileGetc(f); -} - -/* Get a number; we're at the first digit. */ -static unsigned -getnum(FontFilePtr f, int c, int *cp) -{ - unsigned n = 0; - int base = 10; - - /* look for `0' or `0x' prefix */ - if(c == '0') { - c = FontFileGetc(f); - base = 8; - if(c == 'x' || c == 'X') { - base = 16; - c = FontFileGetc(f); - } - } - - /* accumulate digits */ - for(;;) { - if ('0' <= c && c <= '9') { - n *= base; n += c - '0'; - } else if('a' <= c && c <= 'f') { - n *= base; n += c - 'a' + 10; - } else if('A' <=c && c <= 'F') { - n *= base; n += c - 'A' + 10; - } else - break; - c = FontFileGetc(f); - } - - *cp = c; return n; -} - -/* Skip to beginning of new line; return 1 if only whitespace was found. */ -static int -endOfLine(FontFilePtr f, int c) -{ - if(c == 0) - c = FontFileGetc(f); - - for(;;) { - if(c <= 0 || c == '\n') - return 1; - else if(c == '#') { - skipEndOfLine(f,c); - return 1; - } - else if(c == ' ' || c == '\t') { - skipEndOfLine(f,c); - return 0; - } - c = FontFileGetc(f); - } -} - -/* Get a token; we're at first char */ -static int -gettoken(FontFilePtr f, int c, int *cp) -{ - char *p; - - if(c <= 0) - c = FontFileGetc(f); - - if(c <= 0) { - return EOF_TOKEN; - } - - while(c == ' ' || c == '\t') - c = FontFileGetc(f); - - if(c=='\n') { - return EOL_TOKEN; - } else if(c == '#') { - skipEndOfLine(f,c); - return EOL_TOKEN; - } else if(c >= '0' && c <= '9') { - number_value = getnum(f,c,cp); - return NUMBER_TOKEN; - } else if((c >= 'A' && c <= 'Z') || - (c >= 'a' && c <= 'z') || - c == '/' || c == '_' || c == '-' || c == '.') { - p = keyword_value; - *p++ = c; - while(p-keyword_value < MAXKEYWORDLEN) { - c = FontFileGetc(f); - if(c <= ' ' || c > '~' || c == '#') - break; - *p++ = c; - } - *cp = c; - *p = '\0'; - return KEYWORD_TOKEN; - } else { - *cp = c; - return ERROR_TOKEN; - } -} - -/* Parse a line. - * Always skips to the beginning of a new line, even if an error occurs */ -static int -getnextline(FontFilePtr f) -{ - int c, token; - c = FontFileGetc(f); - if(c <= 0) - return EOF_LINE; - - again: - token=gettoken(f,c,&c); - - switch(token) { - case EOF_TOKEN: - return EOF_LINE; - case EOL_TOKEN: - /* empty line */ - c = FontFileGetc(f); - goto again; - case NUMBER_TOKEN: - value1 = number_value; - token = gettoken(f,c,&c); - switch(token) { - case NUMBER_TOKEN: - value2 = number_value; - token = gettoken(f,c,&c); - switch(token) { - case NUMBER_TOKEN: - value3 = number_value; - return CODE_RANGE_LINE; - case EOL_TOKEN: - return CODE_LINE; - default: - skipEndOfLine(f,c); - return ERROR_LINE; - } - case KEYWORD_TOKEN: - if(!endOfLine(f,c)) - return ERROR_LINE; - else - return NAME_LINE; - default: - skipEndOfLine(f,c); - return ERROR_LINE; - } - case KEYWORD_TOKEN: - if(!strcasecmp(keyword_value, "STARTENCODING")) { - token = gettoken(f,c,&c); - if(token == KEYWORD_TOKEN) { - if(endOfLine(f,c)) - return STARTENCODING_LINE; - else - return ERROR_LINE; - } else { - skipEndOfLine(f,c); - return ERROR_LINE; - } - } else if(!strcasecmp(keyword_value, "ALIAS")) { - token = gettoken(f,c,&c); - if(token == KEYWORD_TOKEN) { - if(endOfLine(f,c)) - return ALIAS_LINE; - else - return ERROR_LINE; - } else { - skipEndOfLine(f,c); - return ERROR_LINE; - } - } else if(!strcasecmp(keyword_value, "SIZE")) { - token = gettoken(f,c,&c); - if(token == NUMBER_TOKEN) { - value1 = number_value; - token = gettoken(f,c,&c); - switch(token) { - case NUMBER_TOKEN: - value2 = number_value; - return SIZE_LINE; - case EOL_TOKEN: - value2=0; - return SIZE_LINE; - default: - skipEndOfLine(f,c); - return ERROR_LINE; - } - } else { - skipEndOfLine(f,c); - return ERROR_LINE; - } - } else if(!strcasecmp(keyword_value, "FIRSTINDEX")) { - token = gettoken(f,c,&c); - if(token == NUMBER_TOKEN) { - value1 = number_value; - token = gettoken(f,c,&c); - switch(token) { - case NUMBER_TOKEN: - value2 = number_value; - return FIRSTINDEX_LINE; - case EOL_TOKEN: - value2 = 0; - return FIRSTINDEX_LINE; - default: - skipEndOfLine(f,c); - return ERROR_LINE; - } - } else { - skipEndOfLine(f,c); - return ERROR_LINE; - } - } else if(!strcasecmp(keyword_value, "STARTMAPPING")) { - keyword_value[0] = 0; - value1 = 0; value1 = 0; - /* first a keyword */ - token = gettoken(f,c,&c); - if(token != KEYWORD_TOKEN) { - skipEndOfLine(f, c); - return ERROR_LINE; - } - - /* optional first integer */ - token = gettoken(f,c,&c); - if(token == NUMBER_TOKEN) { - value1 = number_value; - } else if(token == EOL_TOKEN) { - return STARTMAPPING_LINE; - } else { - skipEndOfLine(f, c); - return ERROR_LINE; - } - - /* optional second integer */ - token = gettoken(f,c,&c); - if(token == NUMBER_TOKEN) { - value2 = number_value; - } else if(token == EOL_TOKEN) { - return STARTMAPPING_LINE; - } else { - skipEndOfLine(f, c); - return ERROR_LINE; - } - - if(!endOfLine(f,c)) - return ERROR_LINE; - else { - return STARTMAPPING_LINE; - } - } else if(!strcasecmp(keyword_value, "UNDEFINE")) { - /* first integer */ - token = gettoken(f,c,&c); - if(token != NUMBER_TOKEN) { - skipEndOfLine(f,c); - return ERROR_LINE; - } - value1 = number_value; - /* optional second integer */ - token = gettoken(f,c,&c); - if(token == EOL_TOKEN) { - value2 = value1; - return CODE_UNDEFINE_LINE; - } else if(token == NUMBER_TOKEN) { - value2 = number_value; - if(endOfLine(f,c)) { - return CODE_UNDEFINE_LINE; - } else - return ERROR_LINE; - } else { - skipEndOfLine(f,c); - return ERROR_LINE; - } - } else if(!strcasecmp(keyword_value, "ENDENCODING")) { - if(endOfLine(f,c)) - return EOF_LINE; - else - return ERROR_LINE; - } else if(!strcasecmp(keyword_value, "ENDMAPPING")) { - if(endOfLine(f,c)) - return ENDMAPPING_LINE; - else - return ERROR_LINE; - } else { - skipEndOfLine(f,c); - return ERROR_LINE; - } - default: - return ERROR_LINE; - } -} - -static void -install_mapping(FontEncPtr encoding, FontMapPtr mapping) -{ - FontMapPtr m; - - if(encoding->mappings == NULL) - encoding->mappings = mapping; - else { - m = encoding->mappings; - while(m->next != NULL) - m = m->next; - m->next = mapping; - } - mapping->next = NULL; - mapping->encoding = encoding; -} - -static int -setCode(unsigned from, unsigned to, unsigned row_size, - unsigned *first, unsigned *last, - unsigned *encsize, unsigned short **enc) -{ - unsigned index, i; - unsigned short *newenc; - - if(from>0xFFFF) - return 0; /* success */ - - if(row_size==0) - index=from; - else { - if((value1 & 0xFF) >= row_size) - return 0; /* ignore out of range mappings */ - index = (from>>8) * row_size + (from&0xFF); - } - - /* Optimize away useless identity mappings. This is only expected - to be useful with linear encodings. */ - if(index == to && (index < *first || index > *last)) - return 0; - if(*encsize == 0) { - *encsize = (index < 256) ? 256 : 0x10000; - *enc = (unsigned short*)xalloc((*encsize) * sizeof(unsigned short)); - if(*enc == NULL) { - *encsize = 0; - return 1; - } - } else if(*encsize <= index) { - *encsize = 0x10000; - if((newenc = (unsigned short*)xrealloc(enc, *encsize))==NULL) - return 1; - *enc = newenc; - } - if(*first > *last) { - *first = *last = index; - } - if(index < *first) { - for(i = index; i < *first; i++) - (*enc)[i] = i; - *first = index; - } - if(index > *last) { - for(i = *last + 1; i <= index; i++) - (*enc)[i] = i; - *last = index; - } - (*enc)[index] = to; - return 0; -} - -/* Parser. If headerOnly is true, we're only interested in the - data contained in the encoding file's header. */ - -/* As font encodings are currently never freed, the allocations done - by this function are mostly its private business. Note, however, - that FontEncIdentify needs to free the header fields -- so if you - change this function, you may need to change FontEncIdentify. */ - -/* I want a garbage collector. */ - -static FontEncPtr -parseEncodingFile(FontFilePtr f, int headerOnly) -{ - int line; - - unsigned short *enc=NULL; - char **nam = NULL, **newnam; - unsigned i, first = 0xFFFF, last=0, encsize=0, namsize=0; - FontEncPtr encoding = NULL; - FontMapPtr mapping = NULL; - FontEncSimpleMapPtr sm; - FontEncSimpleNamePtr sn; - char *aliases[MAXALIASES]; - int numaliases=0; - -#if 0 - /* GCC complains about unused labels. Please fix GCC rather than - obfuscating my code. */ - no_encoding: -#endif - line = getnextline(f); - switch(line) { - case EOF_LINE: - goto error; - case STARTENCODING_LINE: - encoding = (FontEncPtr)xalloc(sizeof(FontEncRec)); - if(encoding == NULL) - goto error; - encoding->name = (char*)xalloc(strlen(keyword_value)+1); - if(encoding->name == NULL) - goto error; - strcpy(encoding->name, keyword_value); - encoding->size = 256; - encoding->row_size = 0; - encoding->mappings = NULL; - encoding->next = NULL; - encoding->first = encoding->first_col=0; - goto no_mapping; - default: - goto error; - } - - no_mapping: - line = getnextline(f); - switch(line) { - case EOF_LINE: goto done; - case ALIAS_LINE: - if(numaliases < MAXALIASES) { - aliases[numaliases] = (char*)xalloc(strlen(keyword_value)+1); - if(aliases[numaliases] == NULL) - goto error; - strcpy(aliases[numaliases], keyword_value); - numaliases++; - } - goto no_mapping; - case SIZE_LINE: - encoding->size = value1; - encoding->row_size = value2; - goto no_mapping; - case FIRSTINDEX_LINE: - encoding->first = value1; - encoding->first_col = value2; - goto no_mapping; - case STARTMAPPING_LINE: - if(headerOnly) - goto done; - if(!strcasecmp(keyword_value, "unicode")) { - mapping = (FontMapPtr)xalloc(sizeof(FontMapRec)); - if(mapping == NULL) - goto error; - mapping->type = FONT_ENCODING_UNICODE; - mapping->pid = 0; - mapping->eid = 0; - mapping->recode = 0; - mapping->name = 0; - mapping->client_data = 0; - mapping->next = 0; - goto mapping; - } else if(!strcasecmp(keyword_value, "cmap")) { - mapping = (FontMapPtr)xalloc(sizeof(FontMapRec)); - if(mapping == NULL) - goto error; - mapping->type = FONT_ENCODING_TRUETYPE; - mapping->pid = value1; - mapping->eid = value2; - mapping->recode = 0; - mapping->name = 0; - mapping->client_data = 0; - mapping->next = 0; - goto mapping; - } else if(!strcasecmp(keyword_value, "postscript")) { - mapping = (FontMapPtr)xalloc(sizeof(FontMapRec)); - if(mapping == NULL) - goto error; - mapping->type = FONT_ENCODING_POSTSCRIPT; - mapping->pid = 0; - mapping->eid = 0; - mapping->recode = 0; - mapping->name = 0; - mapping->client_data = 0; - mapping->next = 0; - goto string_mapping; - } else { /* unknown mapping type -- ignore */ - goto skipmapping; - } - /* NOTREACHED */ - goto error; - default: goto no_mapping; /* ignore unknown lines */ - } - - skipmapping: - line = getnextline(f); - switch(line) { - case ENDMAPPING_LINE: - goto no_mapping; - case EOF_LINE: - goto error; - default: - goto skipmapping; - } - - mapping: - line = getnextline(f); - switch(line) { - case EOF_LINE: goto error; - case ENDMAPPING_LINE: - mapping->recode = FontEncSimpleRecode; - mapping->name = FontEncUndefinedName; - mapping->client_data = sm = - (FontEncSimpleMapPtr)xalloc(sizeof(FontEncSimpleMapRec)); - if(sm == NULL) - goto error; - sm->row_size = encoding->row_size; - if(first <= last) { - sm->first = first; - sm->len=last-first+1; - sm->map = - (unsigned short*)xalloc(sm->len * sizeof(unsigned short)); - if(sm->map == NULL) { - xfree(sm); - mapping->client_data = sm = NULL; - goto error; - } - } else { - sm->first = 0; - sm->len = 0; - sm->map = 0; - } - for(i=0; i < sm->len; i++) - sm -> map[i] = enc[first+i]; - install_mapping(encoding, mapping); - mapping = 0; - first = 0xFFFF; last=0; - goto no_mapping; - - case CODE_LINE: - if(setCode(value1, value2, encoding->row_size, - &first, &last, &encsize, &enc)) - goto error; - goto mapping; - - case CODE_RANGE_LINE: - if(value1 > 0x10000) - value1 = 0x10000; - if(value2 > 0x10000) - value2 = 0x10000; - if(value2 < value1) - goto mapping; - /* Do the last value first to avoid having to realloc() */ - if(setCode(value2, value3+(value2-value1), encoding->row_size, - &first, &last, &encsize, &enc)) - goto error; - for(i=value1; i<value2; i++) { - if(setCode(i, value3+(i-value1), encoding->row_size, - &first, &last, &encsize, &enc)) - goto error; - } - goto mapping; - - case CODE_UNDEFINE_LINE: - if(value1 > 0x10000) - value1 = 0x10000; - if(value2 > 0x10000) - value2 = 0x10000; - if(value2 < value1) - goto mapping; - /* Do the last value first to avoid having to realloc() */ - if(setCode(value2, 0, encoding->row_size, - &first, &last, &encsize, &enc)) - goto error; - for(i = value1; i < value2; i++) { - if(setCode(i, 0, encoding->row_size, - &first, &last, &encsize, &enc)) - goto error; - } - goto mapping; - - default: goto mapping; /* ignore unknown lines */ - } - - string_mapping: - line = getnextline(f); - switch(line) { - case EOF_LINE: goto error; - case ENDMAPPING_LINE: - mapping->recode = FontEncUndefinedRecode; - mapping->name = FontEncSimpleName; - mapping->client_data = sn = - (FontEncSimpleNamePtr)xalloc(sizeof(FontEncSimpleNameRec)); - if(sn == NULL) - goto error; - if(first > last) { - xfree(sn); - mapping->client_data = sn = NULL; - goto error; - } - sn->first = first; - sn->len = last - first + 1; - sn->map = (char**)xalloc(sn->len*sizeof(char*)); - if(sn->map == NULL) { - xfree(sn); - mapping->client_data = sn = NULL; - goto error; - } - for(i = 0; i < sn->len; i++) - sn->map[i] = nam[first+i]; - install_mapping(encoding,mapping); - mapping = 0; - first = 0xFFFF; last=0; - goto no_mapping; - case NAME_LINE: - if(value1 >= 0x10000) goto string_mapping; - if(namsize == 0) { - namsize = (value1) < 256 ? 256 : 0x10000; - nam = (char**)xalloc(namsize * sizeof(char*)); - if(nam == NULL) { - namsize=0; - goto error; - } - } else if(namsize <= value1) { - namsize = 0x10000; - if((newnam = (char**)xrealloc(nam, namsize)) == NULL) - goto error; - nam = newnam; - } - if(first > last) { - first = last = value1; - } - if(value1 < first) { - for(i = value1; i < first; i++) - nam[i] = NULL; - first = value1; - } - if(value1 > last) { - for(i=last+1; i <= value1; i++) - nam[i]=NULL; - last = value1; - } - nam[value1] = (char*)xalloc(strlen(keyword_value)+1); - if(nam[value1] == NULL) { - goto error; - } - strcpy(nam[value1], keyword_value); - goto string_mapping; - - default: goto string_mapping; /* ignore unknown lines */ - } - - done: - if(encsize) xfree(enc); encsize=0; enc = NULL; - if(namsize) xfree(nam); namsize=0; nam = NULL; /* don't free entries! */ - - encoding->aliases=NULL; - if(numaliases) { - encoding->aliases = (char**)xalloc((numaliases+1)*sizeof(char*)); - if(encoding->aliases == NULL) - goto error; - for(i=0; i<numaliases; i++) - encoding->aliases[i] = aliases[i]; - encoding->aliases[numaliases]=NULL; - } - - return encoding; - -error: - if(encsize) xfree(enc); encsize=0; - if(namsize) { - for(i = first; i <= last; i++) - if(nam[i]) - xfree(nam[i]); - xfree(nam); - namsize = 0; - } - if(mapping) { - if(mapping->client_data) xfree(mapping->client_data); - xfree(mapping); - } - if(encoding) { - if(encoding->name) xfree(encoding->name); - for(mapping = encoding->mappings; mapping; mapping = mapping->next) { - if(mapping->client_data) xfree(mapping->client_data); - xfree(mapping); - } - xfree(encoding); - } - for(i = 0; i < numaliases; i++) - xfree(aliases[i]); - /* We don't need to free sn and sm as they handled locally in the body.*/ - return 0; -} - -char* -FontEncDirectory() -{ - static char* dir = NULL; - - if(dir == NULL) { - char *c = getenv("FONT_ENCODINGS_DIRECTORY"); - if(c) { - dir = malloc(strlen(c) + 1); - if(!dir) - return NULL; - strcpy(dir, c); - } else { - dir = FONT_ENCODINGS_DIRECTORY; - } - } - return dir; -} - -static void -parseFontFileName(const char *fontFileName, char *buf, char *dir) -{ - const char *p; - char *q, *lastslash; - - for(p = fontFileName, q = dir, lastslash = NULL; *p; p++, q++) { - *q = *p; - if(*p == '/') - lastslash = q+1; - } - - if(!lastslash) - lastslash = dir; - - *lastslash = '\0'; - - if(buf && strlen(dir) + 14 < MAXFONTFILENAMELEN) { - strcpy(buf, dir); - strcat(buf, "encodings.dir"); - } -} - -static FontEncPtr -FontEncReallyReallyLoad(const char *charset, - const char *dirname, const char *dir) -{ - FontFilePtr f; - FILE *file; - FontEncPtr encoding; - char file_name[MAXFONTFILENAMELEN], encoding_name[MAXFONTNAMELEN], - buf[MAXFONTFILENAMELEN]; - int count, n; - static char format[24] = ""; - - /* As we don't really expect to open encodings that often, we don't - take the trouble of caching encodings directories. */ - - if((file = fopen(dirname, "r")) == NULL) { - return NULL; - } - - count = fscanf(file, "%d\n", &n); - if(count == EOF || count != 1) { - fclose(file); - return NULL; - } - - encoding = NULL; - if (!format[0]) { - sprintf(format, "%%%ds %%%d[^\n]\n", (int)sizeof(encoding_name) - 1, - (int)sizeof(file_name) - 1); - } - for(;;) { - count = fscanf(file, format, encoding_name, file_name); - if(count == EOF) - break; - if(count != 2) - break; - - if(!strcasecmp(encoding_name, charset)) { - /* Found it */ - if(file_name[0] != '/') { - if(strlen(dir) + strlen(file_name) >= MAXFONTFILENAMELEN) { - fclose(file); - return NULL; - } - strcpy(buf, dir); - strcat(buf, file_name); - } else { - strcpy(buf , file_name); - } - - f = FontFileOpen(buf); - if(f == NULL) { - fclose(file); - return NULL; - } - encoding = parseEncodingFile(f, 0); - FontFileClose(f); - break; - } - } - - fclose(file); - - return encoding; -} - -/* Parser ntrypoint -- used by FontEncLoad */ -FontEncPtr -FontEncReallyLoad(const char *charset, const char *fontFileName) -{ - FontEncPtr encoding; - char dir[MAXFONTFILENAMELEN], dirname[MAXFONTFILENAMELEN]; - char *d; - - if(fontFileName) { - parseFontFileName(fontFileName, dirname, dir); - encoding = FontEncReallyReallyLoad(charset, dirname, dir); - if(encoding) - return(encoding); - } - - d = FontEncDirectory(); - if(d) { - parseFontFileName(d, NULL, dir); - encoding = FontEncReallyReallyLoad(charset, d, dir); - return encoding; - } - - return NULL; -} - -/* Return a NULL-terminated array of encoding names. Note that this - * function has incestuous knowledge of the allocations done by - * parseEncodingFile. */ - -char ** -FontEncIdentify(const char *fileName) -{ - FontFilePtr f; - FontEncPtr encoding; - char **names, **name, **alias; - int numaliases; - - if((f = FontFileOpen(fileName))==NULL) { - return NULL; - } - encoding = parseEncodingFile(f, 1); - FontFileClose(f); - - if(!encoding) - return NULL; - - numaliases = 0; - if(encoding->aliases) - for(alias = encoding->aliases; *alias; alias++) - numaliases++; - - names = (char**)xalloc((numaliases+2)*sizeof(char*)); - if(names == NULL) { - if(encoding->aliases) - xfree(encoding->aliases); - xfree(encoding); - return NULL; - } - - name = names; - *(name++) = encoding->name; - if(numaliases > 0) - for(alias = encoding->aliases; *alias; alias++, name++) - *name = *alias; - - *name = 0; - xfree(encoding->aliases); - xfree(encoding); - - return names; -} |