diff options
author | Reinhard Tartler <siretart@tauware.de> | 2011-10-10 17:43:39 +0200 |
---|---|---|
committer | Reinhard Tartler <siretart@tauware.de> | 2011-10-10 17:43:39 +0200 |
commit | f4092abdf94af6a99aff944d6264bc1284e8bdd4 (patch) | |
tree | 2ac1c9cc16ceb93edb2c4382c088dac5aeafdf0f /nx-X11/extras/ttf2pt1/ttf2pt1.c | |
parent | a840692edc9c6d19cd7c057f68e39c7d95eb767d (diff) | |
download | nx-libs-f4092abdf94af6a99aff944d6264bc1284e8bdd4.tar.gz nx-libs-f4092abdf94af6a99aff944d6264bc1284e8bdd4.tar.bz2 nx-libs-f4092abdf94af6a99aff944d6264bc1284e8bdd4.zip |
Imported nx-X11-3.1.0-1.tar.gznx-X11/3.1.0-1
Summary: Imported nx-X11-3.1.0-1.tar.gz
Keywords:
Imported nx-X11-3.1.0-1.tar.gz
into Git repository
Diffstat (limited to 'nx-X11/extras/ttf2pt1/ttf2pt1.c')
-rw-r--r-- | nx-X11/extras/ttf2pt1/ttf2pt1.c | 2722 |
1 files changed, 2722 insertions, 0 deletions
diff --git a/nx-X11/extras/ttf2pt1/ttf2pt1.c b/nx-X11/extras/ttf2pt1/ttf2pt1.c new file mode 100644 index 000000000..a85a41658 --- /dev/null +++ b/nx-X11/extras/ttf2pt1/ttf2pt1.c @@ -0,0 +1,2722 @@ +/* + * True Type Font to Adobe Type 1 font converter + * By Mark Heath <mheath@netspace.net.au> + * Based on ttf2pfa by Andrew Weeks <ccsaw@bath.ac.uk> + * With help from Frank M. Siegert <fms@this.net> + * + * see COPYRIGHT for full copyright notice + * +*********************************************************************** + * + * Sergey Babkin <babkin@users.sourceforge.net>, <sab123@hotmail.com> + * + * Added post-processing of resulting outline to correct the errors + * both introduced during conversion and present in the original font, + * autogeneration of hints (has yet to be improved though) and BlueValues, + * scaling to 1000x1000 matrix, option to print the result on STDOUT, + * support of Unicode to CP1251 conversion, optimization of the + * resulting font code by space (that improves the speed too). Excluded + * the glyphs that are unaccessible through the encoding table from + * the output file. Added the built-in Type1 assembler (taken from + * the `t1utils' package). + * +*********************************************************************** + * + * Thomas Henlich <thenlich@rcs.urz.tu-dresden.de> + * + * Added generation of .afm file (font metrics) + * Read encoding information from encoding description file + * Fixed bug in error message about unknown language ('-l' option) + * Added `:' after %%!PS-AdobeFont-1.0 + * changed unused entries in ISOLatin1Encoding[] from .notdef to c127,c128... + * +*********************************************************************** + * + * Thomas Henlich <thenlich@rcs.urz.tu-dresden.de> + * + * Added generation of .afm file (font metrics) + * +*********************************************************************** + * + * Bug Fixes: +************************************************************************ + * + * Sun, 21 Jun 1998 Thomas Henlich <thenlich@Rcs1.urz.tu-dresden.de> + * 1. "width" should be "short int" because otherwise: + * characters with negative widths (e.g. -4) become *very* wide (65532) + * 2. the number of /CharStrings is numglyphs and not numglyphs+1 + * +*********************************************************************** + * + * + * + * The resultant font file produced by this program still needs to be ran + * through t1asm (from the t1utils archive) to produce a completely valid + * font. + * + */ +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <sys/types.h> +#include <sys/stat.h> +#include <fcntl.h> +#include <time.h> +#include <ctype.h> +#include <math.h> + +#ifdef _GNU_SOURCE +#include <getopt.h> +#endif + +#ifndef WINDOWS +# include <unistd.h> +# include <netinet/in.h> +# define BITBUCKET "/dev/null" +# include <sys/wait.h> +#else +# define WINDOWS_FUNCTIONS /* ask to define functions - in one file only */ +# include "windows.h" +# define BITBUCKET "NUL" +# define snprintf _snprintf +#endif + +#ifdef XP_PSTEXT +#include <ft2build.h> +#include FT_FREETYPE_H +#include FT_TRUETYPE_TABLES_H +#include FT_BBOX_H +#include FT_GLYPH_H + +#include FT_CONFIG_CONFIG_H +#include FT_CONFIG_OPTIONS_H +#include FT_ERRORS_H +#include FT_SYSTEM_H +#include FT_IMAGE_H +#include FT_TYPES_H +#include FT_OUTLINE_H +#include FT_MODULE_H +#include FT_RENDER_H +#include FT_TYPE1_TABLES_H +#include FT_TRUETYPE_IDS_H +#include FT_TRUETYPE_TAGS_H +#include FT_MULTIPLE_MASTERS_H +#include FT_SFNT_NAMES_H + +#include "os.h" +#include "Xproto.h" +#include "font.h" +#include "fontstruct.h" +#include "fntfilst.h" +#include "fontutil.h" +#include "fontenc.h" +#include "ft.h" +#define NOT_IN_FTFUNCS +#include "ftfuncs.h" +#endif /* XP_PSTEXT */ + +#include "pt1.h" +#include "global.h" +#include "version.h" + +/* globals */ + +/* table of front-ends */ + +#ifdef USE_TTF +extern struct frontsw ttf_sw; +#endif /* USE_TTF */ +#ifdef USE_BDF +extern struct frontsw bdf_sw; +#endif /* USE_BDF */ +#if defined(USE_FREETYPE) + extern struct frontsw freetype_sw; +#endif + +struct frontsw *frontswtab[] = { +#ifdef USE_BDF + &bdf_sw, +#endif /* USE_BDF */ +#if defined(USE_FREETYPE) && defined(PREFER_FREETYPE) + &freetype_sw, +#endif +#ifdef USE_TTF + &ttf_sw, +#endif /* USE_TTF */ +#if defined(USE_FREETYPE) && !defined(PREFER_FREETYPE) + &freetype_sw, +#endif + NULL /* end of table */ +}; + +struct frontsw *cursw=0; /* the active front end */ +char *front_arg=""; /* optional argument */ + +/* options */ +int encode = 0; /* encode the resulting file */ +int pfbflag = 0; /* produce compressed file */ +int wantafm=0; /* want to see .afm instead of .t1a on stdout */ +int correctvsize=0; /* try to correct the vertical size of characters */ +int wantuid = 0; /* user wants UniqueID entry in the font */ +int allglyphs = 0; /* convert all glyphs, not only 256 of them */ +int warnlevel = 3; /* the level of permitted warnings */ +int forcemap = 0; /* do mapping even on non-Unicode fonts */ +/* options - maximal limits */ +int max_stemdepth = 128; /* maximal depth of stem stack in interpreter (128 - limit from X11) */ +/* options - debugging */ +int absolute = 0; /* print out in absolute values */ +int reverse = 1; /* reverse font to Type1 path directions */ +/* options - suboptions of Outline Processing, defaults are set in table */ +int optimize; /* enables space optimization */ +int smooth; /* enable smoothing of outlines */ +int transform; /* enables transformation to 1000x1000 matrix */ +int hints; /* enables autogeneration of hints */ +int subhints; /* enables autogeneration of substituted hints */ +int trybold; /* try to guess whether the font is bold */ +int correctwidth; /* try to correct the character width */ +int vectorize; /* vectorize the bitmaps */ +int use_autotrace; /* use the autotrace library on bitmap */ +/* options - suboptions of File Generation, defaults are set in table */ +int gen_pfa; /* generate the font file */ +int gen_afm; /* generate the metrics file */ +int gen_dvienc; /* generate the dvips encoding file */ + +/* not quite options to select a particular source encoding */ +int force_pid = -1; /* specific platform id */ +int force_eid = -1; /* specific encoding id */ + +/* structure to define the sub-option lists controlled by the + * case: uppercase enables them, lowercase disables + */ +struct subo_case { + char disbl; /* character to disable - enforced lowercase */ + char enbl; /* character to enable - auto-set as toupper(disbl) */ + int *valp; /* pointer to the actual variable containing value */ + int dflt; /* default value */ + char *descr; /* description */ +}; + +#ifdef DEBUG +int debug = DEBUG; /* debugging flag */ +#else +int debug = 0; +#endif /* DEBUG */ + +FILE *null_file, *pfa_file, *afm_file, *dvienc_file; +int numglyphs; +struct font_metrics fontm; + +/* non-globals */ +static char *strUID = 0; /* user-supplied UniqueID */ +static unsigned long numUID; /* auto-generated UniqueID */ + +static int ps_fmt_3 = 0; +static double scale_factor, original_scale_factor; + +static char *glyph_rename[ENCTABSZ]; + +/* the names assigned if the original font + * does not specify any + */ + +static char *Fmt3Encoding[256] = { + "c0", "c1", "c2", "c3", + "c4", "c5", "c6", "c7", + "c8", "c9", "c10", "c11", + "c12", "CR", "c14", "c15", + "c16", "c17", "c18", "c19", + "c20", "c21", "c22", "c23", + "c24", "c25", "c26", "c27", + "c28", "c29", "c30", "c31", + "space", "exclam", "quotedbl", "numbersign", + "dollar", "percent", "ampersand", "quotesingle", + "parenleft", "parenright", "asterisk", "plus", + "comma", "hyphen", "period", "slash", + "zero", "one", "two", "three", + "four", "five", "six", "seven", + "eight", "nine", "colon", "semicolon", + "less", "equal", "greater", "question", + "at", "A", "B", "C", + "D", "E", "F", "G", + "H", "I", "J", "K", + "L", "M", "N", "O", + "P", "Q", "R", "S", + "T", "U", "V", "W", + "X", "Y", "Z", "bracketleft", + "backslash", "bracketright", "asciicircum", "underscore", + "grave", "a", "b", "c", + "d", "e", "f", "g", + "h", "i", "j", "k", + "l", "m", "n", "o", + "p", "q", "r", "s", + "t", "u", "v", "w", + "x", "y", "z", "braceleft", + "bar", "braceright", "asciitilde", "c127", + "c128", "c129", "quotesinglbase", "florin", + "quotedblbase", "ellipsis", "dagger", "daggerdbl", + "circumflex", "perthousand", "Scaron", "guilsinglleft", + "OE", "c141", "c142", "c143", + "c144", "quoteleft", "quoteright", "quotedblleft", + "quotedblright", "bullet", "endash", "emdash", + "tilde", "trademark", "scaron", "guilsinglright", + "oe", "c157", "c158", "Ydieresis", + "nbspace", "exclamdown", "cent", "sterling", + "currency", "yen", "brokenbar", "section", + "dieresis", "copyright", "ordfeminine", "guillemotleft", + "logicalnot", "sfthyphen", "registered", "macron", + "degree", "plusminus", "twosuperior", "threesuperior", + "acute", "mu", "paragraph", "periodcentered", + "cedilla", "onesuperior", "ordmasculine", "guillemotright", + "onequarter", "onehalf", "threequarters", "questiondown", + "Agrave", "Aacute", "Acircumflex", "Atilde", + "Adieresis", "Aring", "AE", "Ccedilla", + "Egrave", "Eacute", "Ecircumflex", "Edieresis", + "Igrave", "Iacute", "Icircumflex", "Idieresis", + "Eth", "Ntilde", "Ograve", "Oacute", + "Ocircumflex", "Otilde", "Odieresis", "multiply", + "Oslash", "Ugrave", "Uacute", "Ucircumflex", + "Udieresis", "Yacute", "Thorn", "germandbls", + "agrave", "aacute", "acircumflex", "atilde", + "adieresis", "aring", "ae", "ccedilla", + "egrave", "eacute", "ecircumflex", "edieresis", + "igrave", "iacute", "icircumflex", "idieresis", + "eth", "ntilde", "ograve", "oacute", + "ocircumflex", "otilde", "odieresis", "divide", + "oslash", "ugrave", "uacute", "ucircumflex", + "udieresis", "yacute", "thorn", "ydieresis" +}; + +#ifdef notdef /* { */ +/* This table is not used anywhere in the code + * so it's ifdef-ed out by default but left in + * the source code for reference purposes (and + * possibly for future use) + */ + +static char *ISOLatin1Encoding[256] = { + ".null", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", "CR", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + "space", "exclam", "quotedbl", "numbersign", + "dollar", "percent", "ampersand", "quoteright", + "parenleft", "parenright", "asterisk", "plus", + "comma", "hyphen", "period", "slash", + "zero", "one", "two", "three", + "four", "five", "six", "seven", + "eight", "nine", "colon", "semicolon", + "less", "equal", "greater", "question", + "at", "A", "B", "C", + "D", "E", "F", "G", + "H", "I", "J", "K", + "L", "M", "N", "O", + "P", "Q", "R", "S", + "T", "U", "V", "W", + "X", "Y", "Z", "bracketleft", + "backslash", "bracketright", "asciicircum", "underscore", + "grave", "a", "b", "c", + "d", "e", "f", "g", + "h", "i", "j", "k", + "l", "m", "n", "o", + "p", "q", "r", "s", + "t", "u", "v", "w", + "x", "y", "z", "braceleft", + "bar", "braceright", "asciitilde", "c127", + "c128", "c129", "quotesinglbase", "florin", + "quotedblbase", "ellipsis", "dagger", "daggerdbl", + "circumflex", "perthousand", "Scaron", "guilsinglleft", + "OE", "c141", "c142", "c143", + "c144", "quoteleft", "quoteright", "quotedblleft", + "quotedblright", "bullet", "endash", "emdash", + "tilde", "trademark", "scaron", "guilsinglright", + "oe", "c157", "c158", "Ydieresis", + "nbspace", "exclamdown", "cent", "sterling", + "currency", "yen", "brokenbar", "section", + "dieresis", "copyright", "ordfeminine", "guillemotleft", + "logicalnot", "sfthyphen", "registered", "macron", + "degree", "plusminus", "twosuperior", "threesuperior", + "acute", "mu", "paragraph", "periodcentered", + "cedilla", "onesuperior", "ordmasculine", "guillemotright", + "onequarter", "onehalf", "threequarters", "questiondown", + "Agrave", "Aacute", "Acircumflex", "Atilde", + "Adieresis", "Aring", "AE", "Ccedilla", + "Egrave", "Eacute", "Ecircumflex", "Edieresis", + "Igrave", "Iacute", "Icircumflex", "Idieresis", + "Eth", "Ntilde", "Ograve", "Oacute", + "Ocircumflex", "Otilde", "Odieresis", "multiply", + "Oslash", "Ugrave", "Uacute", "Ucircumflex", + "Udieresis", "Yacute", "Thorn", "germandbls", + "agrave", "aacute", "acircumflex", "atilde", + "adieresis", "aring", "ae", "ccedilla", + "egrave", "eacute", "ecircumflex", "edieresis", + "igrave", "iacute", "icircumflex", "idieresis", + "eth", "ntilde", "ograve", "oacute", + "ocircumflex", "otilde", "odieresis", "divide", + "oslash", "ugrave", "uacute", "ucircumflex", + "udieresis", "yacute", "thorn", "ydieresis" +}; + +#endif /* } notdef */ + +static char *adobe_StandardEncoding[256] = { + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + "space", "exclam", "quotedbl", "numbersign", + "dollar", "percent", "ampersand", "quoteright", + "parenleft", "parenright", "asterisk", "plus", + "comma", "hyphen", "period", "slash", + "zero", "one", "two", "three", + "four", "five", "six", "seven", + "eight", "nine", "colon", "semicolon", + "less", "equal", "greater", "question", + "at", "A", "B", "C", "D", "E", "F", "G", + "H", "I", "J", "K", "L", "M", "N", "O", + "P", "Q", "R", "S", "T", "U", "V", "W", + "X", "Y", "Z", "bracketleft", + "backslash", "bracketright", "asciicircum", "underscore", + "quoteleft", "a", "b", "c", "d", "e", "f", "g", + "h", "i", "j", "k", "l", "m", "n", "o", + "p", "q", "r", "s", "t", "u", "v", "w", + "x", "y", "z", "braceleft", + "bar", "braceright", "asciitilde", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", "exclamdown", "cent", "sterling", + "fraction", "yen", "florin", "section", + "currency", "quotesingle", "quotedblleft", "guillemotleft", + "guilsinglleft", "guilsinglright", "fi", "fl", + ".notdef", "endash", "dagger", "daggerdbl", + "periodcentered", ".notdef", "paragraph", "bullet", + "quotesinglbase", "quotedblbase", "quotedblright", "guillemotright", + "ellipsis", "perthousand", ".notdef", "questiondown", + ".notdef", "grave", "acute", "circumflex", + "tilde", "macron", "breve", "dotaccent", + "dieresis", ".notdef", "ring", "cedilla", + ".notdef", "hungarumlaut", "ogonek", "caron", + "emdash", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", "AE", ".notdef", "ordfeminine", + ".notdef", ".notdef", ".notdef", ".notdef", + "Lslash", "Oslash", "OE", "ordmasculine", + ".notdef", ".notdef", ".notdef", ".notdef", + ".notdef", "ae", ".notdef", ".notdef", + ".notdef", "dotlessi", ".notdef", ".notdef", + "lslash", "oslash", "oe", "germandbls", + ".notdef", ".notdef", ".notdef", ".notdef" +}; + +/* + * Decription of the supported conversions from Unicode + * + * SB + * Yes, I know that the compiled-in conversion is stupid but + * it is simple to implement and allows not to worry about the + * filesystem context. After all, the source is always available + * and adding another language to it is easy. + * + * The language name is expected to be the same as the subdirectory name + * in the `encodings' directory (for possible future extensions). + * The primary use of the aliases is for guessing based on the current + * locale. + */ + +#define MAXUNIALIAS 10 +#define MAXUNITABLES 3 + +/* the character used as the language argument separator */ +#define LANG_ARG_SEP '+' + + +/* + * Types of language-related routines. Arguments are: + * name is the glyph name + * arg is the user-specified language-dependent argument + * which can for example select the subfont plane for Eastern fonts. + * If none is supplied by user then an empty string ("") is passed. + * If no language is specified by user and auto-guessing happens + * then NULL is passed. + * when shows if the conversion by name was called before conversion by + * map or after (it's called twice) + */ + +/* type of the Unicode map initialization routine */ +typedef void uni_init_t(char *arg); + +/* type of Unicode converter-by-name function + * it's called for each glyph twice: one time for each glyph + * before doing conversion by map and one time after + */ +typedef int uni_conv_t(char *name, char *arg, int when); +#define UNICONV_BYNAME_BEFORE 0 +#define UNICONV_BYNAME_AFTER 1 + +struct uni_language { + uni_init_t *init[MAXUNITABLES]; /* map initialization routines */ + uni_conv_t *convbyname; /* the name-based conversion function */ + char *name; /* the language name */ + char *descr; /* description */ + char *alias[MAXUNIALIAS]; /* aliases of the language name */ + int sample_upper; /* code of some uppercase character for correctvsize() */ +}; + +/* the converter routines have an option of adding this suffix to the font name */ +static char *uni_font_name_suffix = ""; /* empty by default */ +/* this buffer may be used to store the suffix */ +#define UNI_MAX_SUFFIX_LEN 100 +static char uni_suffix_buf[UNI_MAX_SUFFIX_LEN+1]; + +/* + * Prototypes of the conversion routines + */ + +static uni_init_t unicode_latin1; +static uni_init_t unicode_latin2; +static uni_init_t unicode_latin4; +static uni_init_t unicode_latin5; +static uni_init_t unicode_cyrillic; +static uni_init_t unicode_adobestd; +static uni_init_t unicode_plane; +static uni_conv_t unicode_adobestd_byname; + +static uni_init_t unicode_init_user; + +/* + * The order of descriptions is important: if we can't guess the + * language we just call all the conversion routines in order until + * we find one that understands this glyph. + */ +static struct uni_language uni_lang[]= { + /* pseudo-language for all the languages using Latin1 */ + { + { unicode_latin1 }, + 0, /* no name-based mapping */ + "latin1", + "works for most of the Western languages", + { "en_", "de_", "fr_", "nl_", "no_", "da_", "it_" }, + 'A' + }, + { /* by Szalay Tamas <tomek@elender.hu> */ + { unicode_latin2 }, + 0, /* no name-based mapping */ + "latin2", + "works for Central European languages", + { "hu_","pl_","cz_","si_","sk_" }, + 'A' + }, + { /* by Ričardas Čepas <rch@WriteMe.Com> */ + { unicode_latin4 }, + 0, /* no name-based mapping */ + "latin4", + "works for Baltic languages", + { "lt_", "lv_" }, /* doubt about ee_ */ + 'A' + }, + { /* by Turgut Uyar <uyar@cs.itu.edu.tr> */ + { unicode_latin5 }, + 0, /* no name-based mapping */ + "latin5", + "for Turkish", + { "tr_" }, + 'A' + }, + { /* by Zvezdan Petkovic <z.petkovic@computer.org> */ + { unicode_cyrillic, unicode_latin1 }, + 0, /* no name-based mapping */ + "cyrillic", + "in Windows encoding", + { "bg_", "be_", "mk_", "ru_", "sr_", "su_", "uk_" }, + 'A' + }, + { + { unicode_cyrillic, unicode_latin1 }, + 0, /* no name-based mapping */ + "russian", + "obsolete, use cyrillic instead", + { 0 }, + 'A' + }, + { + { unicode_cyrillic, unicode_latin1 }, + 0, /* no name-based mapping */ + "bulgarian", + "obsolete, use cyrillic instead", + { 0 }, + 'A' + }, + { + { unicode_adobestd }, + unicode_adobestd_byname, + "adobestd", + "Adobe Standard, expected by TeX", + { NULL }, + 'A' + }, + { + { unicode_plane }, + 0, /* no name-based mapping */ + "plane", + "one plane of Unicode or other multi-byte encoding as is", + { NULL }, + 0 /* no easy way to predict the capital letters */ + }, +}; + +static struct uni_language uni_lang_user = { + { unicode_init_user }, + 0, /* no name-based mapping */ + 0, /* no name */ + 0, /* no description */ + { 0 }, + 0 /* no sample */ +}; + +static struct uni_language *uni_lang_selected=0; /* 0 means "unknown, try all" */ +static int uni_sample='A'; /* sample of an uppercase character */ +static char *uni_lang_arg=""; /* user-supplied language-dependent argument */ + +extern int runt1asm(int); + +/* + * user-defined loadable maps + */ + + +/* The idea begind buckets is to avoid comparing every code with all ENCTABSZ codes in table. + * All the 16-bit unicode space is divided between a number of equal-sized buckets. + * Initially all the buckets are marked with 0. Then if any code in the bucket is + * used it's marked with 1. Later during translation we check the code's bucket first + * and it it's 0 then return failure right away. This may be useful for + * Chinese fonts with many thousands of glyphs. + */ + +#define BUCKET_ID_BITS 11 +#define MARK_UNI_BUCKET(unicode) SET_BITMAP(uni_user_buckets, (unicode)>>(16-BUCKET_ID_BITS)) +#define IS_UNI_BUCKET(unicode) IS_BITMAP(uni_user_buckets, (unicode)>>(16-BUCKET_ID_BITS)) + +static DEF_BITMAP(uni_user_buckets, 1<<BUCKET_ID_BITS); + +static unsigned int unicode_map[ENCTABSZ]; /* font-encoding to unicode map */ +static int enctabsz = 256; /* actual number of codes used */ + +static void +unicode_init_user( + char *path +) +{ + FILE *unicode_map_file; +#define UNIBFSZ 256 + char buffer[UNIBFSZ]; + unsigned code, unicode, curpos, unicode2; + char *arg, *p; + int enabled, found, sawplane; + int lineno, cnt, n, nchars; + char next; + int pid, eid, overid=0; + + /* check if we have an argument (plane name) */ + arg = strrchr(path, LANG_ARG_SEP); + if(arg != 0) { + *arg++ = 0; + if( sscanf(arg, "pid=%d,eid=%d%n", &pid, &eid, &nchars) == 2 ) { + force_pid = pid; force_eid = eid; overid = 1; + WARNING_1 fprintf(stderr, "User override of the source encoding: pid=%d eid=%d\n", pid, eid); + forcemap = 1; + arg += nchars; + if(*arg == ',') + arg++; + } + if( *arg == 0 || strlen(arg) > UNI_MAX_SUFFIX_LEN-1) + arg = NULL; + else { + sprintf(uni_suffix_buf, "-%s", arg); + uni_font_name_suffix = uni_suffix_buf; + } + } + + /* now read in the encoding description file, if requested */ + if ((unicode_map_file = fopen(path, "r")) == NULL) { + fprintf(stderr, "**** Cannot access map file '%s' ****\n", path); + exit(1); + } + + sawplane = 0; + if(arg==NULL) + enabled = found = 1; + else + enabled = found = 0; + + lineno=0; curpos=0; + while (fgets (buffer, UNIBFSZ, unicode_map_file) != NULL) { + char name[UNIBFSZ]; + + lineno++; + + if(sscanf(buffer, "plane %s", name)==1) { + sawplane = 1; + if(arg == 0) { + fprintf(stderr, "**** map file '%s' requires plane name\n", path); + fprintf(stderr, "for example:\n"); + fprintf(stderr, " ttf2pt1 -L %s%c[pid=N,eid=N,]%s ...\n", + path, LANG_ARG_SEP, name); + fprintf(stderr, "to select plane '%s'\n", name); + exit(1); + } + if( !strcmp(arg, name) ) { + enabled = found = 1; + curpos = 0; + } else { + enabled = 0; + if(found) /* no need to read further */ + break; + } + continue; + } + + if(sscanf(buffer, "id %d %d", &pid, &eid)==2) { + if( !overid /* only if the user has not overriden */ + && (enabled || !sawplane) ) { + force_pid = pid; force_eid = eid; + forcemap = 1; + } + continue; + } + + if( !enabled ) + continue; /* skip to the next plane */ + + if( sscanf(buffer, "at %i", &curpos) == 1 ) { + if(curpos > 255) { + fprintf(stderr, "**** map file '%s' line %d: code over 255\n", path, lineno); + exit(1); + } + if(ISDBG(EXTMAP)) fprintf(stderr, "=== at 0x%x\n", curpos); + continue; + } + + /* try the format of Roman Czyborra's files */ + if ( sscanf (buffer, " =%x U+%4x", &code, &unicode) == 2 + /* try the format of Linux locale charmap file */ + || sscanf (buffer, " <%*s /x%x <U%4x>", &code, &unicode) == 2 ) { + if (code < ENCTABSZ) { + if(code >= enctabsz) enctabsz=code+1; + unicode_map[code] = unicode; + glyph_rename[code] = NULL; + } + } + /* try the format with glyph renaming */ + else if (sscanf (buffer, " !%x U+%4x %128s", &code, + &unicode, name) == 3) { + if (code < ENCTABSZ) { + if(code >= enctabsz) enctabsz=code+1; + unicode_map[code] = unicode; + glyph_rename[code] = strdup(name); + } + } + /* try the compact sequence format */ + else if( (n=sscanf(buffer, " %i%n", &unicode, &cnt)) == 1 ) { + p = buffer; + do { + if(curpos > 255) { + fprintf(stderr, "**** map file '%s' line %d: code over 255 for unicode 0x%x\n", + path, lineno, unicode); + exit(1); + } + if(ISDBG(EXTMAP)) fprintf(stderr, "=== 0x%d -> 0x%x\n", curpos, unicode); + unicode_map[curpos++] = unicode; + p += cnt; + if( sscanf(p, " %[,-]%n", &next,&cnt) == 1 ) { + if(ISDBG(EXTMAP)) fprintf(stderr, "=== next: '%c'\n", next); + p += cnt; + if( next == '-' ) { /* range */ + if ( sscanf(p, " %i%n", &unicode2, &cnt) != 1 ) { + fprintf(stderr, "**** map file '%s' line %d: missing end of range\n", path, lineno); + exit(1); + } + p += cnt; + if(ISDBG(EXTMAP)) fprintf(stderr, "=== range 0x%x to 0x%x\n", unicode, unicode2); + for(unicode++; unicode <= unicode2; unicode++) { + if(curpos > 255) { + fprintf(stderr, "**** map file '%s' line %d: code over 255 in unicode range ...-0x%x\n", + path, lineno, unicode2); + exit(1); + } + if(ISDBG(EXTMAP)) fprintf(stderr, "=== 0x%x -> 0x%x\n", curpos, unicode); + unicode_map[curpos++] = unicode; + } + } + } + } while ( sscanf(p, " %i%n", &unicode, &cnt) == 1 ); + } + + } + + fclose (unicode_map_file); + + if( !found ) { + fprintf(stderr, "**** map file '%s' has no plane '%s'\n", path, arg); + exit(1); + } + + if(unicode_map['A'] == 'A') + uni_sample = 'A'; /* seems to be compatible with Latin */ + else + uni_sample = 0; /* don't make any assumptions */ +} + +/* + * by Zvezdan Petkovic <z.petkovic@computer.org> + */ +static void +unicode_cyrillic( + char *arg +) +{ + int i; + static unsigned int cyrillic_unicode_map[] = { + 0x0402, 0x0403, 0x201a, 0x0453, 0x201e, 0x2026, 0x2020, 0x2021, /* 80 */ + 0x20ac, 0x2030, 0x0409, 0x2039, 0x040a, 0x040c, 0x040b, 0x040f, /* 88 */ + 0x0452, 0x2018, 0x2019, 0x201c, 0x201d, 0x2022, 0x2013, 0x2014, /* 90 */ + 0x02dc, 0x2122, 0x0459, 0x203a, 0x045a, 0x045c, 0x045b, 0x045f, /* 98 */ + 0x00a0, 0x040e, 0x045e, 0x0408, 0x00a4, 0x0490, 0x00a6, 0x00a7, /* A0 */ + 0x0401, 0x00a9, 0x0404, 0x00ab, 0x00ac, 0x00ad, 0x00ae, 0x0407, /* A8 */ + 0x00b0, 0x00b1, 0x0406, 0x0456, 0x0491, 0x00b5, 0x00b6, 0x00b7, /* B0 */ + 0x0451, 0x2116, 0x0454, 0x00bb, 0x0458, 0x0405, 0x0455, 0x0457, /* B8 */ + }; + + for(i=0; i<=0x7F; i++) + unicode_map[i] = i; + + for(i=0x80; i<=0xBF; i++) + unicode_map[i] = cyrillic_unicode_map[i-0x80]; + + for(i=0xC0; i<=0xFF; i++) + unicode_map[i] = i+0x350; + +} + +static void +unicode_latin1( + char *arg +) +{ + int i; + static unsigned int latin1_unicode_map[] = { + 0x20ac, -1, 0x201a, 0x0192, 0x201e, 0x2026, 0x2020, 0x2021, /* 80 */ + 0x02c6, 0x2030, 0x0160, 0x2039, 0x0152, 0x008d, 0x017d, 0x008f, /* 88 */ + 0x0090, 0x2018, 0x2019, 0x201c, 0x201d, 0x2022, 0x2013, 0x2014, /* 90 */ + 0x02dc, 0x2122, 0x0161, 0x203a, 0x0153, 0x009d, 0x017e, 0x0178, /* 98 */ + }; + + for(i=0; i<=0x7F; i++) + unicode_map[i] = i; + + for(i=0x80; i<=0x9F; i++) + unicode_map[i] = latin1_unicode_map[i-0x80]; + + for(i=0xA0; i<=0xFF; i++) + unicode_map[i] = i; +} + +static void +unicode_adobestd( + char *arg +) +{ + int i; + static unsigned int adobestd_unicode_map[] = { + -1, 0x00a1, 0x00a2, 0x00a3, 0x2215, 0x00a5, 0x0192, 0x00a7, /* A0 */ + 0x00a4, 0x0027, 0x201c, 0x00ab, 0x2039, 0x203a, 0xfb01, 0xfb02, /* A8 */ + -1, 0x2013, 0x2020, 0x2021, 0x2219, -1, 0x00b6, 0x2022, /* B0 */ + 0x201a, 0x201e, 0x201d, 0x00bb, 0x2026, 0x2030, -1, 0x00bf, /* B8 */ + -1, 0x0060, 0x00b4, 0x02c6, 0x02dc, 0x02c9, 0x02d8, 0x02d9, /* C0 */ + 0x00a8, -1, 0x02da, 0x00b8, -1, 0x02dd, 0x02db, 0x02c7, /* C8 */ + 0x2014, -1, -1, -1, -1, -1, -1, -1, /* D0 */ + -1, -1, -1, -1, -1, -1, -1, -1, /* D8 */ + -1, 0x00c6, -1, 0x00aa, -1, -1, -1, -1, /* E0 */ + 0x0141, 0x00d8, 0x0152, 0x00ba, -1, -1, -1, -1, /* E8 */ + -1, 0x00e6, -1, -1, -1, 0x0131, -1, -1, /* F0 */ + 0x0142, 0x00f8, 0x0153, 0x00df, -1, -1, -1, -1, /* F8 */ + }; + + for(i=0; i<=0x7F; i++) + unicode_map[i] = i; + + unicode_map[0x27] = 0x2019; + unicode_map[0x60] = -1; + + /* 0x80 to 0x9F is a hole */ + + for(i=0xA0; i<=0xFF; i++) + unicode_map[i] = adobestd_unicode_map[i-0xA0]; +} + +/* + * Not all of the Adobe glyphs are in the Unicode + * standard maps, so the font creators have + * different ideas about their codes. Because + * of this we try to map based on the glyph + * names instead of Unicode codes. If there are + * no glyph names (ps_fmt_3!=0) we fall back + * to the code-based scheme. + */ + +static int +unicode_adobestd_byname( + char *name, + char *arg, + int where +) +{ + int i; + + /* names always take precedence over codes */ + if(where == UNICONV_BYNAME_AFTER) + return -1; + + for(i=32; i<256; i++) { + if(!strcmp(name, adobe_StandardEncoding[i])) + return i; + } + return -1; + +} + +static void +unicode_latin2( + char *arg +) +{ + int i; + static unsigned int latin2_unicode_map[] = { + 0x00a0, 0x0104, 0x02d8, 0x0141, 0x00a4, 0x013d, 0x015a, 0x00a7, /* A0 */ + 0x00a8, 0x0160, 0x015e, 0x0164, 0x0179, 0x00ad, 0x017d, 0x017b, /* A8 */ + 0x00b0, 0x0105, 0x02db, 0x0142, 0x00b4, 0x013e, 0x015b, 0x02c7, /* B0 */ + 0x00b8, 0x0161, 0x015f, 0x0165, 0x017a, 0x02dd, 0x017e, 0x017c, /* B8 */ + 0x0154, 0x00c1, 0x00c2, 0x0102, 0x00c4, 0x0139, 0x0106, 0x00c7, /* C0 */ + 0x010c, 0x00c9, 0x0118, 0x00cb, 0x011a, 0x00cd, 0x00ce, 0x010e, /* C8 */ + 0x0110, 0x0143, 0x0147, 0x00d3, 0x00d4, 0x0150, 0x00d6, 0x00d7, /* D0 */ + 0x0158, 0x016e, 0x00da, 0x0170, 0x00dc, 0x00dd, 0x0162, 0x00df, /* D8 */ + 0x0155, 0x00e1, 0x00e2, 0x0103, 0x00e4, 0x013a, 0x0107, 0x00e7, /* E0 */ + 0x010d, 0x00e9, 0x0119, 0x00eb, 0x011b, 0x00ed, 0x00ee, 0x010f, /* E8 */ + 0x0111, 0x0144, 0x0148, 0x00f3, 0x00f4, 0x0151, 0x00f6, 0x00f7, /* F0 */ + 0x0159, 0x016f, 0x00fa, 0x0171, 0x00fc, 0x00fd, 0x0163, 0x02d9, /* F8 */ + }; + + for(i=0; i<=0x7E; i++) + unicode_map[i] = i; + + /* 7F-9F are unused */ + + for(i=0xA0; i<=0xFF; i++) + unicode_map[i] = latin2_unicode_map[i-0xA0]; +} + +static void +unicode_latin4( + char *arg +) +{ + int i; + static unsigned int latin4_unicode_map[] = { + 0x0080, 0x0081, 0x201a, 0x0192, -1, 0x2026, 0x2020, 0x2021, /* 80 */ + 0x02c6, 0x2030, -1, 0x2039, 0x0152, 0x008d, 0x008e, 0x008f, /* 88 */ + 0x201e, 0x201c, 0x2019, -1, 0x201d, 0x2022, 0x2013, 0x2014, /* 90 */ + 0x02dc, 0x2122, -1, 0x203a, 0x0153, 0x009d, 0x009e, 0x0178, /* 98 */ + 0x00a0, 0x0104, 0x0138, 0x0156, 0x00a4, 0x0128, 0x013b, 0x00a7, /* A0 */ + 0x00a8, 0x0160, 0x0112, 0x0122, 0x0166, 0x00ad, 0x017d, 0x00af, /* A8 */ + 0x00b0, 0x0105, 0x02db, 0x0157, 0x00b4, 0x0129, 0x013c, 0x02c7, /* B0 */ + 0x00b8, 0x0161, 0x0113, 0x0123, 0x0167, 0x014a, 0x017e, 0x014b, /* B8 */ + 0x0100, 0x00c1, 0x00c2, 0x00c3, 0x00c4, 0x00c5, 0x00c6, 0x012e, /* C0 */ + 0x010c, 0x00c9, 0x0118, 0x00cb, 0x0116, 0x00cd, 0x00ce, 0x012a, /* C8 */ + 0x0110, 0x0145, 0x014c, 0x0136, 0x00d4, 0x00d5, 0x00d6, 0x00d7, /* D0 */ + 0x00d8, 0x0172, 0x00da, 0x00db, 0x00dc, 0x0168, 0x016a, 0x00df, /* D8 */ + 0x0101, 0x00e1, 0x00e2, 0x00e3, 0x00e4, 0x00e5, 0x00e6, 0x012f, /* E0 */ + 0x010d, 0x00e9, 0x0119, 0x00eb, 0x0117, 0x00ed, 0x00ee, 0x012b, /* E8 */ + 0x0111, 0x0146, 0x014d, 0x0137, 0x00f4, 0x00f5, 0x00f6, 0x00f7, /* F0 */ + 0x00f8, 0x0173, 0x00fa, 0x00fb, 0x00fc, 0x0169, 0x016b, 0x02d9, /* F8 */ + }; + + for(i=0; i<=0x7F; i++) + unicode_map[i] = i; + + for(i=0x80; i<=0xFF; i++) + unicode_map[i] = latin4_unicode_map[i-0x80]; + +#if 0 /* for documentation purposes only */ + case 0x201e: return 0x90; /* these two quotes are a hack only */ + case 0x201c: return 0x91; /* these two quotes are a hack only */ + case 0x00A0: return 0xA0; /* NO-BREAK SPACE */ + case 0x0104: return 0xA1; /* LATIN CAPITAL LETTER A WITH OGONEK */ + case 0x0138: return 0xA2; /* LATIN SMALL LETTER KRA */ + case 0x0156: return 0xA3; /* LATIN CAPITAL LETTER R WITH CEDILLA */ + case 0x00A4: return 0xA4; /* CURRENCY SIGN */ + case 0x0128: return 0xA5; /* LATIN CAPITAL LETTER I WITH TILDE */ + case 0x013B: return 0xA6; /* LATIN CAPITAL LETTER L WITH CEDILLA */ + case 0x00A7: return 0xA7; /* SECTION SIGN */ + case 0x00A8: return 0xA8; /* DIAERESIS */ + case 0x0160: return 0xA9; /* LATIN CAPITAL LETTER S WITH CARON */ + case 0x0112: return 0xAA; /* LATIN CAPITAL LETTER E WITH MACRON */ + case 0x0122: return 0xAB; /* LATIN CAPITAL LETTER G WITH CEDILLA */ + case 0x0166: return 0xAC; /* LATIN CAPITAL LETTER T WITH STROKE */ + case 0x00AD: return 0xAD; /* SOFT HYPHEN */ + case 0x017D: return 0xAE; /* LATIN CAPITAL LETTER Z WITH CARON */ + case 0x00AF: return 0xAF; /* MACRON */ + case 0x00B0: return 0xB0; /* DEGREE SIGN */ + case 0x0105: return 0xB1; /* LATIN SMALL LETTER A WITH OGONEK */ + case 0x02DB: return 0xB2; /* OGONEK */ + case 0x0157: return 0xB3; /* LATIN SMALL LETTER R WITH CEDILLA */ + case 0x00B4: return 0xB4; /* ACUTE ACCENT */ + case 0x0129: return 0xB5; /* LATIN SMALL LETTER I WITH TILDE */ + case 0x013C: return 0xB6; /* LATIN SMALL LETTER L WITH CEDILLA */ + case 0x02C7: return 0xB7; /* CARON */ + case 0x00B8: return 0xB8; /* CEDILLA */ + case 0x0161: return 0xB9; /* LATIN SMALL LETTER S WITH CARON */ + case 0x0113: return 0xBA; /* LATIN SMALL LETTER E WITH MACRON */ + case 0x0123: return 0xBB; /* LATIN SMALL LETTER G WITH CEDILLA */ + case 0x0167: return 0xBC; /* LATIN SMALL LETTER T WITH STROKE */ + case 0x014A: return 0xBD; /* LATIN CAPITAL LETTER ENG */ + case 0x017E: return 0xBE; /* LATIN SMALL LETTER Z WITH CARON */ + case 0x014B: return 0xBF; /* LATIN SMALL LETTER ENG */ + case 0x0100: return 0xC0; /* LATIN CAPITAL LETTER A WITH MACRON */ + case 0x00C1: return 0xC1; /* LATIN CAPITAL LETTER A WITH ACUTE */ + case 0x00C2: return 0xC2; /* LATIN CAPITAL LETTER A WITH CIRCUMFLEX */ + case 0x00C3: return 0xC3; /* LATIN CAPITAL LETTER A WITH TILDE */ + case 0x00C4: return 0xC4; /* LATIN CAPITAL LETTER A WITH DIAERESIS */ + case 0x00C5: return 0xC5; /* LATIN CAPITAL LETTER A WITH RING ABOVE */ + case 0x00C6: return 0xC6; /* LATIN CAPITAL LIGATURE AE */ + case 0x012E: return 0xC7; /* LATIN CAPITAL LETTER I WITH OGONEK */ + case 0x010C: return 0xC8; /* LATIN CAPITAL LETTER C WITH CARON */ + case 0x00C9: return 0xC9; /* LATIN CAPITAL LETTER E WITH ACUTE */ + case 0x0118: return 0xCA; /* LATIN CAPITAL LETTER E WITH OGONEK */ + case 0x00CB: return 0xCB; /* LATIN CAPITAL LETTER E WITH DIAERESIS */ + case 0x0116: return 0xCC; /* LATIN CAPITAL LETTER E WITH DOT ABOVE */ + case 0x00CD: return 0xCD; /* LATIN CAPITAL LETTER I WITH ACUTE */ + case 0x00CE: return 0xCE; /* LATIN CAPITAL LETTER I WITH CIRCUMFLEX */ + case 0x012A: return 0xCF; /* LATIN CAPITAL LETTER I WITH MACRON */ + case 0x0110: return 0xD0; /* LATIN CAPITAL LETTER D WITH STROKE */ + case 0x0145: return 0xD1; /* LATIN CAPITAL LETTER N WITH CEDILLA */ + case 0x014C: return 0xD2; /* LATIN CAPITAL LETTER O WITH MACRON */ + case 0x0136: return 0xD3; /* LATIN CAPITAL LETTER K WITH CEDILLA */ + case 0x00D4: return 0xD4; /* LATIN CAPITAL LETTER O WITH CIRCUMFLEX */ + case 0x00D5: return 0xD5; /* LATIN CAPITAL LETTER O WITH TILDE */ + case 0x00D6: return 0xD6; /* LATIN CAPITAL LETTER O WITH DIAERESIS */ + case 0x00D7: return 0xD7; /* MULTIPLICATION SIGN */ + case 0x00D8: return 0xD8; /* LATIN CAPITAL LETTER O WITH STROKE */ + case 0x0172: return 0xD9; /* LATIN CAPITAL LETTER U WITH OGONEK */ + case 0x00DA: return 0xDA; /* LATIN CAPITAL LETTER U WITH ACUTE */ + case 0x00DB: return 0xDB; /* LATIN CAPITAL LETTER U WITH CIRCUMFLEX */ + case 0x00DC: return 0xDC; /* LATIN CAPITAL LETTER U WITH DIAERESIS */ + case 0x0168: return 0xDD; /* LATIN CAPITAL LETTER U WITH TILDE */ + case 0x016A: return 0xDE; /* LATIN CAPITAL LETTER U WITH MACRON */ + case 0x00DF: return 0xDF; /* LATIN SMALL LETTER SHARP S */ + case 0x0101: return 0xE0; /* LATIN SMALL LETTER A WITH MACRON */ + case 0x00E1: return 0xE1; /* LATIN SMALL LETTER A WITH ACUTE */ + case 0x00E2: return 0xE2; /* LATIN SMALL LETTER A WITH CIRCUMFLEX */ + case 0x00E3: return 0xE3; /* LATIN SMALL LETTER A WITH TILDE */ + case 0x00E4: return 0xE4; /* LATIN SMALL LETTER A WITH DIAERESIS */ + case 0x00E5: return 0xE5; /* LATIN SMALL LETTER A WITH RING ABOVE */ + case 0x00E6: return 0xE6; /* LATIN SMALL LIGATURE AE */ + case 0x012F: return 0xE7; /* LATIN SMALL LETTER I WITH OGONEK */ + case 0x010D: return 0xE8; /* LATIN SMALL LETTER C WITH CARON */ + case 0x00E9: return 0xE9; /* LATIN SMALL LETTER E WITH ACUTE */ + case 0x0119: return 0xEA; /* LATIN SMALL LETTER E WITH OGONEK */ + case 0x00EB: return 0xEB; /* LATIN SMALL LETTER E WITH DIAERESIS */ + case 0x0117: return 0xEC; /* LATIN SMALL LETTER E WITH DOT ABOVE */ + case 0x00ED: return 0xED; /* LATIN SMALL LETTER I WITH ACUTE */ + case 0x00EE: return 0xEE; /* LATIN SMALL LETTER I WITH CIRCUMFLEX */ + case 0x012B: return 0xEF; /* LATIN SMALL LETTER I WITH MACRON */ + case 0x0111: return 0xF0; /* LATIN SMALL LETTER D WITH STROKE */ + case 0x0146: return 0xF1; /* LATIN SMALL LETTER N WITH CEDILLA */ + case 0x014D: return 0xF2; /* LATIN SMALL LETTER O WITH MACRON */ + case 0x0137: return 0xF3; /* LATIN SMALL LETTER K WITH CEDILLA */ + case 0x00F4: return 0xF4; /* LATIN SMALL LETTER O WITH CIRCUMFLEX */ + case 0x00F5: return 0xF5; /* LATIN SMALL LETTER O WITH TILDE */ + case 0x00F6: return 0xF6; /* LATIN SMALL LETTER O WITH DIAERESIS */ + case 0x00F7: return 0xF7; /* DIVISION SIGN */ + case 0x00F8: return 0xF8; /* LATIN SMALL LETTER O WITH STROKE */ + case 0x0173: return 0xF9; /* LATIN SMALL LETTER U WITH OGONEK */ + case 0x00FA: return 0xFA; /* LATIN SMALL LETTER U WITH ACUTE */ + case 0x00FB: return 0xFB; /* LATIN SMALL LETTER U WITH CIRCUMFLEX */ + case 0x00FC: return 0xFC; /* LATIN SMALL LETTER U WITH DIAERESIS */ + case 0x0169: return 0xFD; /* LATIN SMALL LETTER U WITH TILDE */ + case 0x016B: return 0xFE; /* LATIN SMALL LETTER U WITH MACRON */ + case 0x02D9: return 0xFF; /* DOT ABOVE */ +#endif +} + +static void +unicode_latin5( + char *arg +) +{ + int i; + static unsigned int latin5_unicode_map1[] = { + 0x0080, 0x0081, 0x201a, 0x0192, 0x201e, 0x2026, 0x2020, 0x2021, /* 80 */ + 0x02c6, 0x2030, 0x0160, 0x2039, 0x0152, 0x008d, 0x008e, 0x008f, /* 88 */ + 0x0090, 0x2018, 0x2019, 0x201c, 0x201d, 0x2022, 0x2013, 0x2014, /* 90 */ + 0x02dc, 0x2122, 0x0161, 0x203a, 0x0153, 0x009d, 0x009e, 0x0178, /* 98 */ + }; + static unsigned int latin5_unicode_map2[] = { + 0x011e, 0x00d1, 0x00d2, 0x00d3, 0x00d4, 0x00d5, 0x00d6, 0x00d7, /* D0 */ + 0x00d8, 0x00d9, 0x00da, 0x00db, 0x00dc, 0x0130, 0x015e, 0x00df, /* D8 */ + 0x00e0, 0x00e1, 0x00e2, 0x00e3, 0x00e4, 0x00e5, 0x00e6, 0x00e7, /* E0 direct */ + 0x00e8, 0x00e9, 0x00ea, 0x00eb, 0x00ec, 0x00ed, 0x00ee, 0x00ef, /* E8 direct */ + 0x011f, 0x00f1, 0x00f2, 0x00f3, 0x00f4, 0x00f5, 0x00f6, 0x00f7, /* F0 */ + 0x00f8, 0x00f9, 0x00fa, 0x00fb, 0x00fc, 0x0131, 0x015f, 0x00ff, /* F8 */ + }; + + for(i=0; i<=0x7F; i++) + unicode_map[i] = i; + + for(i=0x80; i<=0x9F; i++) + unicode_map[i] = latin5_unicode_map1[i-0x80]; + + for(i=0xA0; i<=0xCF; i++) + unicode_map[i] = i; + + for(i=0xD0; i<=0xFF; i++) + unicode_map[i] = latin5_unicode_map2[i-0xD0]; +} + +/* a way to select one 256-character plane from Unicode + * or other multi-byte encoding + */ + +static void +unicode_plane( + char *arg +) +{ + static unsigned plane; + int nchars; + int c1, c2, i; + + if(uni_lang_selected == 0) + return; /* don't participate in auto-guessing */ + + plane = 0; force_pid = force_eid = -1; + + c1 = sscanf(arg, "pid=%d,eid=%d%n", &force_pid, &force_eid, &nchars); + if(c1 == 2) { + arg += nchars; + if(*arg == ',') + arg++; + } + if(arg[0] == '0' && (arg[1]=='x' || arg[1]=='X') ) { + arg += 2; + c2 = sscanf(arg, "%x", &plane); + } else { + c2 = sscanf(arg, "%d", &plane); + } + + if( (c1!=2 && c1!=0) || (c1==0 && c2==0) ) { + fprintf(stderr, "**** option -l plane expects one of the following formats:\n"); + fprintf(stderr, " -l plane+0xNN - select hexadecimal number of plane of Unicode\n"); + fprintf(stderr, " -l plane+NN - select decimal number of plane of Unicode\n"); + fprintf(stderr, " -l plane+pid=N,eid=N - select plane 0 of specified encoding\n"); + fprintf(stderr, " -l plane+pid=N,eid=N,0xNN - select hex plane of TTF encoding with this PID/EID\n"); + fprintf(stderr, " -l plane+pid=N,eid=N,NN - select decimal plane of TTF encoding with this PID/EID\n"); + exit(1); + } + + if(c2!=0) { + if(strlen(arg) > sizeof(uni_suffix_buf)-2) { + fprintf(stderr, "**** plane number is too large\n"); + } + + sprintf(uni_suffix_buf, "-%s", arg); + uni_font_name_suffix = uni_suffix_buf; + } else { + uni_font_name_suffix = ""; + } + + plane <<= 8; + for(i=0; i<=0xFF; i++) + unicode_map[i] = plane | i; +} + +/* look up the 8-bit code by unicode */ + +int +unicode_rev_lookup( + int unival +) +{ + int res; + + if( ! IS_UNI_BUCKET(unival) ) + return -1; + + for (res = 0; res < enctabsz; res++) + if (unicode_map[res] == unival) + return res; + return -1; +} + +/* mark the buckets for quick lookup */ + +static void +unicode_prepare_buckets( + void +) +{ + int i; + + memset(uni_user_buckets, 0, sizeof uni_user_buckets); + for(i=0; i<enctabsz; i++) { + if(unicode_map[i] != (unsigned) -1) + MARK_UNI_BUCKET(unicode_map[i]); + } +} + +/* + * When we print errors about bad names we want to print these names in + * some decent-looking form + */ + +static char * +nametoprint( + unsigned char *s +) +{ + static char res[50]; + int c, i; + + for(i=0; ( c =* s )!=0 && i<sizeof(res)-8; s++) { + if(c < ' ' || c > 126) { + sprintf(res+i, "\\x%02X", c); + i+=4; + } else { + res[i++] = c; + } + } + if(*s != 0) { + res[i++] = '.'; + res[i++] = '.'; + res[i++] = '.'; + } + res[i++] = 0; + return res; +} + +/* + * Scale the values according to the scale_factor + */ + +double +fscale( + double val +) +{ + return scale_factor * val; +} + +int +iscale( + int val +) +{ + return (int) (val > 0 ? scale_factor * val + 0.5 + : scale_factor * val - 0.5); +} + +/* + * Try to force fixed width of characters + */ + +static void +alignwidths(void) +{ + int i; + int n = 0, avg, max = 0, min = 3000, sum = 0, x; + + for (i = 0; i < numglyphs; i++) { + if (glyph_list[i].flags & GF_USED) { + x = glyph_list[i].width; + + if (x != 0) { + if (x < min) + min = x; + if (x > max) + max = x; + + sum += x; + n++; + } + } + } + + if (n == 0) + return; + + avg = sum / n; + + WARNING_3 fprintf(stderr, "widths: max=%d avg=%d min=%d\n", max, avg, min); + + /* if less than 5% variation from average */ + /* force fixed width */ + if (20 * (avg - min) < avg && 20 * (max - avg) < avg) { + for (i = 0; i < numglyphs; i++) { + if (glyph_list[i].flags & GF_USED) + glyph_list[i].width = avg; + } + fontm.is_fixed_pitch = 1; + } +} + +static void +convert_glyf( + int glyphno +) +{ + GLYPH *g; + int ncurves; + + g = &glyph_list[glyphno]; + + + g->scaledwidth = iscale(g->width); + + g->entries = 0; + g->lastentry = 0; + g->path = 0; + if (g->ttf_pathlen != 0) { + cursw->glpath(glyphno, glyph_list); + g->lastentry = 0; + + if(ISDBG(BUILDG)) + dumppaths(g, NULL, NULL); + + assertpath(g->entries, __FILE__, __LINE__, g->name); + + fclosepaths(g); + assertpath(g->entries, __FILE__, __LINE__, g->name); + + /* float processing */ + if(smooth) { + ffixquadrants(g); + assertpath(g->entries, __FILE__, __LINE__, g->name); + + fsplitzigzags(g); + assertpath(g->entries, __FILE__, __LINE__, g->name); + + fforceconcise(g); + assertpath(g->entries, __FILE__, __LINE__, g->name); + + fstraighten(g); + assertpath(g->entries, __FILE__, __LINE__, g->name); + } + + pathtoint(g); + /* all processing past this point expects integer path */ + assertpath(g->entries, __FILE__, __LINE__, g->name); + +#if 0 + fixcontours(g); + testfixcvdir(g); +#endif + + /* int processing */ + if (smooth) { + smoothjoints(g); + assertpath(g->entries, __FILE__, __LINE__, g->name); + } + + ncurves = 0; + { + GENTRY *ge; + for(ge = g->entries; ge; ge = ge->next) + ncurves++; + } + if (ncurves > 200) { + WARNING_3 fprintf(stderr, + "** Glyph %s is too long, may display incorrectly\n", + g->name); + } + } else { + /* for buildstems */ + g->flags &= ~GF_FLOAT; + } +} + +static void +handle_gnames(void) +{ + int i, n, found, c, type; + + /* get the names from the font file */ + ps_fmt_3 = cursw->glnames(glyph_list); + +/* These checks are not required by Xprt's PS DDX... */ +#ifndef XP_PSTEXT + /* check for names with wrong characters */ + for (n = 0; n < numglyphs; n++) { + int c; + for (i = 0; (c = glyph_list[n].name[i]) != 0; i++) { + if (!(isalnum(c) || c == '.' || c == '_' || c == '-') + || i==0 && isdigit(c)) { /* must not start with a digit */ + WARNING_3 fprintf(stderr, "Glyph %d %s (%s), ", + n, isdigit(c) ? "name starts with a digit" : + "has bad characters in name", + nametoprint(glyph_list[n].name)); + glyph_list[n].name = malloc(16); + sprintf(glyph_list[n].name, "_b_%d", n); + WARNING_3 fprintf(stderr, "changing to %s\n", glyph_list[n].name); + break; + } + } + } + + if( !ps_fmt_3 ) { + /* check for duplicate names */ + for (n = 0; n < numglyphs; n++) { + found = 0; + for (i = 0; i < n && !found; i++) { + if (strcmp(glyph_list[i].name, glyph_list[n].name) == 0) { + if (( glyph_list[n].name = malloc(16) )==0) { + fprintf (stderr, "****malloc failed %s line %d\n", __FILE__, __LINE__); + exit(255); + } + sprintf(glyph_list[n].name, "_d_%d", n); + + /* if the font has no names in it (what the native parser + * recognises as ps_fmt_3), FreeType returns all the + * names as .notdef, so don't complain in this case + */ + if(strcmp(glyph_list[i].name, ".notdef")) { + WARNING_3 fprintf(stderr, + "Glyph %d has the same name as %d: (%s), changing to %s\n", + n, i, + glyph_list[i].name, + glyph_list[n].name); + } + found = 1; + } + } + } + + } +#endif /* !XP_PSTEXT */ + + /* start the encoding stuff */ + for (i = 0; i < ENCTABSZ; i++) { + encoding[i] = -1; + } + + /* do the 1st round of encoding by name */ + if(!ps_fmt_3 && uni_lang_selected && uni_lang_selected->convbyname) { + for (n = 0; n < numglyphs; n++) { + c = uni_lang_selected->convbyname(glyph_list[n].name, + uni_lang_arg, UNICONV_BYNAME_BEFORE); + if(c>=0 && c<ENCTABSZ && encoding[c] == -1) + encoding[c] = n; + } + } + + /* now do the encoding by table */ + if(uni_lang_selected) { + for(i=0; i < MAXUNITABLES && uni_lang_selected->init[i]; i++) { + for (n = 0; n < ENCTABSZ; n++) + unicode_map[n] = -1; + uni_lang_selected->init[i](uni_lang_arg); + unicode_prepare_buckets(); + type = cursw->glenc(glyph_list, encoding, unicode_map); + if( type == 0 ) + /* if we have an 8-bit encoding we don't need more tries */ + break; + } + } else { + /* language is unknown, try the first table of each */ + for(i=0; i < sizeof uni_lang/(sizeof uni_lang[0]); i++) { + if(uni_lang[i].init[0] == NULL) + continue; + for (n = 0; n < ENCTABSZ; n++) + unicode_map[n] = -1; + uni_lang[i].init[0](uni_lang_arg); + unicode_prepare_buckets(); + type = cursw->glenc(glyph_list, encoding, unicode_map); + if( type == 0 ) + /* if we have an 8-bit encoding we don't need more tries */ + break; + } + } + + if (ps_fmt_3) { + /* get rid of the old names, they are all "UNKNOWN" anyawy */ + for (i = 0; i < numglyphs; i++) { + glyph_list[i].name = 0; + } + if(type == 0) { + /* 8-bit - give 8859/1 names to the first 256 glyphs */ + for (i = 0; i < 256; i++) { /* here 256, not ENCTABSZ */ + if (encoding[i] > 0) { + glyph_list[encoding[i]].name = Fmt3Encoding[i]; + } + } + } else if(type == 1) { + /* Unicode - give 8859/1 names to the first 256 glyphs of Unicode */ + for (n = 0; n < 256; n++) { /* here 256, not ENCTABSZ */ + i = unicode_rev_lookup(n); + if (i>=0 && encoding[i] > 0) { + glyph_list[encoding[i]].name = Fmt3Encoding[i]; + } + } + } /* for other types of encodings just give generated names */ + /* assign unique names to the rest of the glyphs */ + for (i = 0; i < numglyphs; i++) { + if (glyph_list[i].name == 0) { + if (( glyph_list[i].name = malloc(16) )==0) { + fprintf (stderr, "****malloc failed %s line %d\n", __FILE__, __LINE__); + exit(255); + } + sprintf(glyph_list[i].name, "_d_%d", i); + } + } + } + + /* do the 2nd round of encoding by name */ + if(uni_lang_selected && uni_lang_selected->convbyname) { + for (n = 0; n < numglyphs; n++) { + c = uni_lang_selected->convbyname(glyph_list[n].name, + uni_lang_arg, UNICONV_BYNAME_AFTER); + if(c>=0 && c<ENCTABSZ && encoding[c] == -1) + encoding[c] = n; + } + } + /* all the encoding things are done */ + + for (i = 0; i < ENCTABSZ; i++) + if(encoding[i] == -1) { + /* check whether this character might be a duplicate + * (in which case it would be missed by unicode_rev_lookup()) + */ + c = unicode_map[i]; + if((type != 0 || forcemap) && c != -1) { + for(n = 0; n < i; n++) { + if(unicode_map[n] == c) { + encoding[i] = encoding[n]; + } + } + } + if(encoding[i] == -1) /* still not found, defaults to .notdef */ + encoding[i] = 0; + } + + for (i = 0; i < 256; i++) /* here 256, not ENCTABSZ */ + glyph_list[encoding[i]].char_no = i; + +#ifndef XP_PSTEXT + /* enforce two special cases defined in TTF manual */ + if(numglyphs > 0) + glyph_list[0].name = ".notdef"; + if(numglyphs > 1) + glyph_list[1].name = ".null"; +#endif /* !XP_PSTEXT */ + + for (i = 0; i < ENCTABSZ; i++) { + if ((encoding[i] != 0) && glyph_rename[i]) { + glyph_list[encoding[i]].name = glyph_rename[i]; + } + } +} + +static void +usage(void) +{ + +#ifdef XP_PSTEXT + fputs("ft2pt1: Internal startup error\n", stderr); +#else + +#ifdef _GNU_SOURCE +# define fplop(txt) fputs(txt, stderr); +#else +# define fplop(txt) +#endif + + fputs("Use:\n", stderr); + fputs("ttf2pt1 [-<opts>] [-l language | -L file] <ttf-file> [<fontname>]\n", stderr); + fputs(" or\n", stderr); + fputs("ttf2pt1 [-<opts>] [-l language | -L file] <ttf-file> -\n", stderr); + fputs(" or\n", stderr); + fputs("ttf2pt1 [-<opts>] [-l language | -L file] <ttf-file> - | t1asm > <pfa-file>\n", stderr); + + fplop("\n"); + fplop("This build supports both short and long option names,\n"); + fplop("the long options are listed before corresponding short ones\n"); + + fplop(" --all-glyphs\n"); + fputs(" -a - include all glyphs, even those not in the encoding table\n", stderr); + fplop(" --pfb\n"); + fputs(" -b - produce a compressed .pfb file\n", stderr); + fplop(" --debug dbg_suboptions\n"); + fputs(" -d dbg_suboptions - debugging options, run ttf2pt1 -d? for help\n", stderr); + fplop(" --encode\n"); + fputs(" -e - produce a fully encoded .pfa file\n", stderr); + fplop(" --force-unicode\n"); + fputs(" -F - force use of Unicode encoding even if other MS encoding detected\n", stderr); + fplop(" --generate suboptions\n"); + fputs(" -G suboptions - control the file generation, run ttf2pt1 -G? for help\n", stderr); + fplop(" --language language\n"); + fputs(" -l language - convert Unicode to specified language, run ttf2pt1 -l? for list\n", stderr); + fplop(" --language-map file\n"); + fputs(" -L file - convert Unicode according to encoding description file\n", stderr); + fplop(" --limit <type>=<value>\n"); + fputs(" -m <type>=<value> - set maximal limit of given type to value, types:\n", stderr); + fputs(" h - maximal hint stack depth in the PostScript interpreter\n", stderr); + fplop(" --processing suboptions\n"); + fputs(" -O suboptions - control outline processing, run ttf2pt1 -O? for help\n", stderr); + fplop(" --parser name\n"); + fputs(" -p name - use specific front-end parser, run ttf2pt1 -p? for list\n", stderr); + fplop(" --uid id\n"); + fputs(" -u id - use this UniqueID, -u A means autogeneration\n", stderr); + fplop(" --vertical-autoscale size\n"); + fputs(" -v size - scale the font to make uppercase letters >size/1000 high\n", stderr); + fplop(" --version\n"); + fputs(" -V - print ttf2pt1 version number\n", stderr); + fplop(" --warning number\n"); + fputs(" -W number - set the level of permitted warnings (0 - disable)\n", stderr); + fputs("Obsolete options (will be removed in future releases):\n", stderr); + fplop(" --afm\n"); + fputs(" -A - write the .afm file to STDOUT instead of the font, now -GA\n", stderr); + fputs(" -f - don't try to guess the value of the ForceBold hint, now -Ob\n", stderr); + fputs(" -h - disable autogeneration of hints, now -Oh\n", stderr); + fputs(" -H - disable hint substitution, now -Ou\n", stderr); + fputs(" -o - disable outline optimization, now -Oo\n", stderr); + fputs(" -s - disable outline smoothing, now -Os\n", stderr); + fputs(" -t - disable auto-scaling to 1000x1000 standard matrix, now -Ot\n", stderr); + fputs(" -w - correct the glyph widths (use only for buggy fonts), now -OW\n", stderr); + fputs("With no <fontname>, write to <ttf-file> with suffix replaced.\n", stderr); + fputs("The last '-' means 'use STDOUT'.\n", stderr); + +#undef fplop + +#endif /* XP_PSTEXT */ +} + +static void +printversion(void) +{ + fprintf(stderr, "ttf2pt1 %s\n", TTF2PT1_VERSION); +} + +/* initialize a table of suboptions */ +static void +init_subo_tbl( + struct subo_case *tbl +) +{ + int i; + + for(i=0; tbl[i].disbl != 0; i++) { + tbl[i].disbl = tolower(tbl[i].disbl); + tbl[i].enbl = toupper(tbl[i].disbl); + *(tbl[i].valp) = tbl[i].dflt; + } +} + +/* print the default value of the suboptions */ +static void +print_subo_dflt( + FILE *f, + struct subo_case *tbl +) +{ + int i; + + for(i=0; tbl[i].disbl != 0; i++) { + if(tbl[i].dflt) + putc(tbl[i].enbl, f); + else + putc(tbl[i].disbl, f); + } +} + +/* print the usage message for the suboptions */ +static void +print_subo_usage( + FILE *f, + struct subo_case *tbl +) +{ +#ifdef XP_PSTEXT + fputs("ft2pt1: Internal startup error\n", stderr); +#else + int i; + + fprintf(f,"The lowercase suboptions disable features, corresponding\n"); + fprintf(f,"uppercase suboptions enable them. The supported suboptions,\n"); + fprintf(f,"their default states and the features they control are:\n"); + for(i=0; tbl[i].disbl != 0; i++) { + fprintf(f," %c/%c - [%s] %s\n", tbl[i].disbl, tbl[i].enbl, + tbl[i].dflt ? "enabled" : "disabled", tbl[i].descr); + } +#endif /* XP_PSTEXT */ +} + +/* find and set the entry according to suboption, + * return the found entry (or if not found return NULL) + */ +struct subo_case * +set_subo( + struct subo_case *tbl, + int subopt +) +{ + int i; + + for(i=0; tbl[i].disbl != 0; i++) { + if(subopt == tbl[i].disbl) { + *(tbl[i].valp) = 0; + return &tbl[i]; + } else if(subopt == tbl[i].enbl) { + *(tbl[i].valp) = 1; + return &tbl[i]; + } + } + return NULL; +} + + +#ifdef XP_PSTEXT +FT_Face xp_pstext_ft_face = NULL; /* used by ft.c */ +FTFontPtr xp_xtf = NULL; +const char *xp_psfontname = NULL; +unsigned long xp_font_block_offset = 0UL; + +int +ft2pt1_main( + int argc, + char **argv, + FTFontPtr tf, + const char *download_psfontname, + unsigned long download_font_block_offset +) +#else +int +main( + int argc, + char **argv +) +#endif /* XP_PSTEXT */ +{ + long i, j; + time_t now; + char filename[4096]; + int c,nchars,nmetrics; + int ws; + int forcebold= -1; /* -1 means "don't know" */ + char *lang; + int oc; + int subid; + char *cmdline; +#ifdef _GNU_SOURCE +# define ttf2pt1_getopt(a, b, c, d, e) getopt_long(a, b, c, d, e) + static struct option longopts[] = { + { "afm", 0, NULL, 'A' }, + { "all-glyphs", 0, NULL, 'a' }, + { "pfb", 0, NULL, 'b' }, + { "debug", 1, NULL, 'd' }, + { "encode", 0, NULL, 'e' }, + { "force-unicode", 0, NULL, 'F' }, + { "generate", 1, NULL, 'G' }, + { "language", 1, NULL, 'l' }, + { "language-map", 1, NULL, 'L' }, + { "limit", 1, NULL, 'm' }, + { "processing", 1, NULL, 'O' }, + { "parser", 1, NULL, 'p' }, + { "uid", 1, NULL, 'u' }, + { "vertical-autoscale", 1, NULL, 'v' }, + { "version", 0, NULL, 'V' }, + { "warning", 1, NULL, 'W' }, + { NULL, 0, NULL, 0 } + }; +#else +# define ttf2pt1_getopt(a, b, c, d, e) getopt(a, b, c) +#endif + /* table of Outline Processing (may think also as Optimization) options */ + static struct subo_case opotbl[] = { + { 'b', 0/*auto-set*/, &trybold, 1, "guessing of the ForceBold hint" }, + { 'h', 0/*auto-set*/, &hints, 1, "autogeneration of hints" }, + { 'u', 0/*auto-set*/, &subhints, 1, "hint substitution technique" }, + { 'o', 0/*auto-set*/, &optimize, 1, "space optimization of font files" }, + { 's', 0/*auto-set*/, &smooth, 1, "smoothing and repair of outlines" }, + { 't', 0/*auto-set*/, &transform, 1, "auto-scaling to the standard matrix 1000x1000" }, + { 'w', 0/*auto-set*/, &correctwidth, 0, "correct the glyph widths (use only for buggy fonts)" }, + { 'v', 0/*auto-set*/, &vectorize, 0, "vectorize (trace) the bitmaps" }, +#ifdef USE_AUTOTRACE + { 'z', 0/*auto-set*/, &use_autotrace, 0, "use the autotrace library on bitmaps (works badly)" }, +#endif /*USE_AUTOTRACE*/ + { 0, 0, 0, 0, 0} /* terminator */ + }; + /* table of the File Generation options */ + static struct subo_case fgotbl[] = { + { 'f', 0/*auto-set*/, &gen_pfa, 1, "generate the font file (.t1a, .pfa or .pfb)" }, + { 'a', 0/*auto-set*/, &gen_afm, 1, "generate the Adobe metrics file (.afm)" }, + { 'e', 0/*auto-set*/, &gen_dvienc, 0, "generate the dvips encoding file (.enc)" }, + { 0, 0, 0, 0, 0} /* terminator */ + }; + int *genlast = NULL; + +#ifdef XP_PSTEXT + xp_pstext_ft_face = tf->instance->face->face; + xp_xtf = tf; + xp_psfontname = download_psfontname; + xp_font_block_offset = download_font_block_offset; +#endif /* XP_PSTEXT */ + + init_subo_tbl(opotbl); /* initialize sub-options of -O */ + init_subo_tbl(fgotbl); /* initialize sub-options of -G */ + + /* save the command line for the record + * (we don't bother about escaping the shell special characters) + */ + + j = 0; + for(i=1; i<argc; i++) { + j += strlen(argv[i])+1; + } + if ((cmdline = malloc(j+1)) == NULL) { + fprintf (stderr, "****malloc failed %s line %d\n", __FILE__, __LINE__); + exit(255); + } + cmdline[0] = 0; + for(i=1; i<argc; i++) { + strcat(cmdline, argv[i]); + strcat(cmdline, " "); + } + for(i=0; (j=cmdline[i])!=0; i++) + if(j == '\n') + cmdline[i] = ' '; + + + while(( oc=ttf2pt1_getopt(argc, argv, "FaoebAsthHfwVv:p:l:d:u:L:m:W:O:G:", + longopts, NULL) )!= -1) { + switch(oc) { + case 'W': + if(sscanf(optarg, "%d", &warnlevel) < 1 || warnlevel < 0) { + fprintf(stderr, "**** warning level must be a positive number\n"); + exit(1); + } + break; + case 'F': + forcemap = 1; + break; + case 'o': + fputs("Warning: option -o is obsolete, use -Oo instead\n", stderr); + optimize = 0; + break; + case 'e': + encode = 1; + break; + case 'b': + encode = pfbflag = 1; + break; + case 'A': + fputs("Warning: option -A is obsolete, use -GA instead\n", stderr); + wantafm = 1; + break; + case 'a': + allglyphs = 1; + break; + case 's': + fputs("Warning: option -s is obsolete, use -Os instead\n", stderr); + smooth = 0; + break; + case 't': + fputs("Warning: option -t is obsolete, use -Ot instead\n", stderr); + transform = 0; + break; + case 'd': + for(i=0; optarg[i]!=0; i++) + switch(optarg[i]) { + case 'a': + absolute = 1; + break; + case 'r': + reverse = 0; + break; + default: + if (optarg[i] != '?') + fprintf(stderr, "**** Unknown debugging option '%c' ****\n", optarg[i]); + fputs("The recognized debugging options are:\n", stderr); + fputs(" a - enable absolute coordinates\n", stderr); + fputs(" r - do not reverse font outlines directions\n", stderr); + exit(1); + break; + }; + break; + case 'm': + { + char subopt; + int val; + + if(sscanf(optarg, "%c=%d", &subopt, &val) !=2) { + fprintf(stderr, "**** Misformatted maximal limit ****\n"); + fprintf(stderr, "spaces around the equal sign are not allowed\n"); + fprintf(stderr, "good examples: -mh=100 -m h=100\n"); + fprintf(stderr, "bad examples: -mh = 100 -mh= 100\n"); + exit(1); + break; + } + switch(subopt) { + case 'h': + max_stemdepth = val; + break; + default: + if (subopt != '?') + fprintf(stderr, "**** Unknown limit type '%c' ****\n", subopt); + fputs("The recognized limit types are:\n", stderr); + fputs(" h - maximal hint stack depth in the PostScript interpreter\n", stderr); + exit(1); + break; + } + break; + } + case 'O': + { + char *p; + for(p=optarg; *p != 0; p++) { + if(set_subo(opotbl, *p) == NULL) { /* found no match */ + if (*p != '?') + fprintf(stderr, "**** Unknown outline processing suboption '%c' ****\n", *p); + fprintf(stderr,"The general form of the outline processing option is:\n"); + fprintf(stderr," -O suboptions\n"); + fprintf(stderr,"(To remember easily -O may be also thought of as \"optimization\").\n"); + print_subo_usage(stderr, opotbl); + fprintf(stderr, "The default state corresponds to the option -O "); + print_subo_dflt(stderr, opotbl); + fprintf(stderr, "\n"); + exit(1); + } + } + break; + } + case 'G': + { + char *p; + struct subo_case *s; + + for(p=optarg; *p != 0; p++) { + if(( s = set_subo(fgotbl, *p) )==NULL) { /* found no match */ + if (*p != '?') + fprintf(stderr, "**** Unknown outline processing suboption '%c' ****\n", *p); + fprintf(stderr,"The general form of the file generation option is:\n"); + fprintf(stderr," -G suboptions\n"); + print_subo_usage(stderr, fgotbl); + fprintf(stderr, "The default state corresponds to the option -G "); + print_subo_dflt(stderr, fgotbl); + fprintf(stderr, "\n"); + fprintf(stderr, "If the result is written to STDOUT, the last specified enabling suboption of -G\n"); + fprintf(stderr, "selects the file to be written to STDOUT (the font file by default).\n"); + exit(1); + } + if( *(s->valp) ) + genlast = s->valp; + } + break; + } + case 'h': + fputs("Warning: option -h is obsolete, use -Oh instead\n", stderr); + hints = 0; + break; + case 'H': + fputs("Warning: meaning of option -H has been changed to its opposite\n", stderr); + fputs("Warning: option -H is obsolete, use -Ou instead\n", stderr); + subhints = 0; + break; + case 'f': + fputs("Warning: option -f is obsolete, use -Ob instead\n", stderr); + trybold = 0; + break; + case 'w': + fputs("Warning: option -w is obsolete, use -OW instead\n", stderr); + correctwidth = 1; + break; + case 'u': + if(wantuid) { + fprintf(stderr, "**** UniqueID may be specified only once ****\n"); + exit(1); + } + wantuid = 1; + if(optarg[0]=='A' && optarg[1]==0) + strUID=0; /* will be generated automatically */ + else { + strUID=optarg; + for(i=0; optarg[i]!=0; i++) + if( !isdigit(optarg[i]) ) { + fprintf(stderr, "**** UniqueID must be numeric or A for automatic ****\n"); + exit(1); + } + } + break; + case 'v': + correctvsize = atoi(optarg); + if(correctvsize <= 0 && correctvsize > 1000) { + fprintf(stderr, "**** wrong vsize '%d', ignored ****\n", correctvsize); + correctvsize=0; + } + break; + case 'p': + if(cursw!=0) { + fprintf(stderr, "**** only one front-end parser be used ****\n"); + exit(1); + } + + { /* separate parser from parser-specific argument */ + char *p = strchr(optarg, LANG_ARG_SEP); + if(p != 0) { + *p = 0; + front_arg = p+1; + } else + front_arg = ""; + } + for(i=0; frontswtab[i] != NULL; i++) + if( !strcmp(frontswtab[i]->name, optarg) ) { + cursw = frontswtab[i]; + break; + } + + if(cursw==0) { + if (strcmp(optarg, "?")) + fprintf(stderr, "**** unknown front-end parser '%s' ****\n", optarg); + fputs("the following front-ends are supported now:\n", stderr); + for(i=0; frontswtab[i] != NULL; i++) { + fprintf(stderr," %s (%s)\n file suffixes: ", + frontswtab[i]->name, + frontswtab[i]->descr ? frontswtab[i]->descr : "no description" + ); + for(j=0; j<MAXSUFFIX; j++) + if(frontswtab[i]->suffix[j]) + fprintf(stderr, "%s ", frontswtab[i]->suffix[j]); + fprintf(stderr, "\n"); + } + exit(1); + } + break; + case 'l': + if(uni_lang_selected!=0) { + fprintf(stderr, "**** only one language option may be used ****\n"); + exit(1); + } + + { /* separate language from language-specific argument */ + char *p = strchr(optarg, LANG_ARG_SEP); + if(p != 0) { + *p = 0; + uni_lang_arg = p+1; + } else + uni_lang_arg = ""; + } + for(i=0; i < sizeof uni_lang/(sizeof uni_lang[0]); i++) + if( !strcmp(uni_lang[i].name, optarg) ) { + uni_lang_selected = &uni_lang[i]; + uni_sample = uni_lang[i].sample_upper; + break; + } + + if(uni_lang_selected==0) { + if (strcmp(optarg, "?")) + fprintf(stderr, "**** unknown language '%s' ****\n", optarg); + fputs(" the following languages are supported now:\n", stderr); + for(i=0; i < sizeof uni_lang/(sizeof uni_lang[0]); i++) + fprintf(stderr," %s (%s)\n", + uni_lang[i].name, + uni_lang[i].descr ? uni_lang[i].descr : "no description" + ); + exit(1); + } + break; + case 'L': + if(uni_lang_selected!=0) { + fprintf(stderr, "**** only one language option may be used ****\n"); + exit(1); + } + uni_lang_selected = &uni_lang_user; + uni_lang_arg = optarg; + break; + case 'V': + printversion(); + exit(0); + break; + default: + usage(); + exit(1); + break; + } + } + argc-=optind-1; /* the rest of code counts from argv[0] */ + argv+=optind-1; + + if (absolute && encode) { + fprintf(stderr, "**** options -a and -e are incompatible ****\n"); + exit(1); + } + if ((argc != 2) && (argc != 3)) { + usage(); + exit(1); + } + + /* try to guess the language by the locale used */ + if(uni_lang_selected==0 && (lang=getenv("LANG"))!=0 ) { + for(i=0; i < sizeof uni_lang/sizeof(struct uni_language); i++) { + if( !strncmp(uni_lang[i].name, lang, strlen(uni_lang[i].name)) ) { + uni_lang_selected = &uni_lang[i]; + goto got_a_language; + } + } + /* no full name ? try aliases */ + for(i=0; i < sizeof uni_lang/sizeof(struct uni_language); i++) { + for(c=0; c<MAXUNIALIAS; c++) + if( uni_lang[i].alias[c]!=0 + && !strncmp(uni_lang[i].alias[c], lang, strlen(uni_lang[i].alias[c])) ) { + uni_lang_selected = &uni_lang[i]; + goto got_a_language; + } + } + got_a_language: + if(uni_lang_selected!=0) { + WARNING_1 fprintf(stderr, "Using language '%s' for Unicode fonts\n", uni_lang[i].name); + uni_sample = uni_lang[i].sample_upper; + } + } + + /* try to guess the front-end parser by the file name suffix */ + if(cursw==0) { + char *p = strrchr(argv[1], '.'); + char *s; + + if(p!=0 && (s = strdup(p+1))!=0) { + for(p=s; *p; p++) + *p = tolower(*p); + p = s; + + for(i=0; frontswtab[i] != 0 && cursw == 0; i++) { + for(j=0; j<MAXSUFFIX; j++) + if(frontswtab[i]->suffix[j] + && !strcmp(p, frontswtab[i]->suffix[j]) ) { + cursw = frontswtab[i]; + WARNING_1 fprintf(stderr, "Auto-detected front-end parser '%s'\n", + cursw->name); + WARNING_1 fprintf(stderr, " (use ttf2pt1 -p? to get the full list of available front-ends)\n"); + break; + } + } + free(s); + } + + if(cursw==0) { + cursw = frontswtab[0]; + WARNING_1 fprintf(stderr, "Can't detect front-end parser, using '%s' by default\n", + cursw->name); + WARNING_1 fprintf(stderr, " (use ttf2pt1 -p? to get the full list of available front-ends)\n"); + } + } + + /* open the input file */ + cursw->open(argv[1], front_arg); + + /* Get base name of output file (if not specified) + * by removing (known) suffixes + */ + if (argc == 2) { + char *p; + argv[2] = strdup (argv[1]); + p = strrchr(argv[2], '.'); + if (p != NULL) + for (j = 0; (j < MAXSUFFIX) && (cursw->suffix[j]); j++) + if (!strcmp(p+1, cursw->suffix[j])) { + *p = '\0'; + break; + } + } + + if ((null_file = fopen(BITBUCKET, "w")) == NULL) { + fprintf(stderr, "**** Cannot open %s ****\n", + BITBUCKET); + exit(1); + } + + if (argv[2][0] == '-' && argv[2][1] == 0) { +#ifdef WINDOWS + if(encode) { + fprintf(stderr, "**** can't write encoded file to stdout ***\n"); + exit(1); + } +#endif /* WINDOWS */ + pfa_file = afm_file = dvienc_file = null_file; + + if(wantafm || genlast == &gen_afm) { /* print .afm instead of .pfa */ + afm_file=stdout; + } else if(genlast == &gen_dvienc) { /* print .enc instead of .pfa */ + dvienc_file=stdout; + } else { + pfa_file=stdout; + } + } else { +#ifndef WINDOWS + snprintf(filename, sizeof filename, "%s.%s", argv[2], encode ? (pfbflag ? "pfb" : "pfa") : "t1a" ); +#else /* WINDOWS */ + snprintf(filename, sizeof filename, "%s.t1a", argv[2]); +#endif /* WINDOWS */ + if(gen_pfa) { + if ((pfa_file = fopen(filename, "w+b")) == NULL) { + fprintf(stderr, "**** Cannot create %s ****\n", filename); + exit(1); + } else { + WARNING_2 fprintf(stderr, "Creating file %s\n", filename); + } + } else + pfa_file = null_file; + + if(gen_afm) { + snprintf(filename, sizeof filename, "%s.afm", argv[2]) ; + if ((afm_file = fopen(filename, "w+")) == NULL) { + fprintf(stderr, "**** Cannot create %s ****\n", filename); + exit(1); + } + } else + afm_file = null_file; + + if(gen_dvienc) { + snprintf(filename, sizeof filename, "%s.enc", argv[2]) ; + if ((dvienc_file = fopen(filename, "w+")) == NULL) { + fprintf(stderr, "**** Cannot create %s ****\n", filename); + exit(1); + } + } else + dvienc_file = null_file; + } + + /* + * Now check whether we want a fully encoded .pfa file + */ +#ifndef WINDOWS + if (encode && pfa_file != null_file) { + int p[2]; + extern FILE *ifp, *ofp; /* from t1asm.c */ + + ifp=stdin; + ofp=stdout; + + if (pipe(p) < 0) { + perror("**** Cannot create pipe ****\n"); + exit(1); + } + ofp = pfa_file; + ifp = fdopen(p[0], "r"); + if (ifp == NULL) { + perror("**** Cannot use pipe for reading ****\n"); + exit(1); + } + pfa_file = fdopen(p[1], "w"); + if (pfa_file == NULL) { + perror("**** Cannot use pipe for writing ****\n"); + exit(1); + } + switch (fork()) { + case -1: + perror("**** Cannot fork the assembler process ****\n"); + exit(1); + case 0: /* child */ + fclose(pfa_file); + exit(runt1asm(pfbflag)); + default: /* parent */ + fclose(ifp); fclose(ofp); + } + } +#endif /* WINDOWS */ + + numglyphs = cursw->nglyphs(); + + WARNING_3 fprintf(stderr, "numglyphs = %d\n", numglyphs); + + glyph_list = (GLYPH *) calloc(numglyphs, sizeof(GLYPH)); + + /* initialize non-0 fields */ + for (i = 0; i < numglyphs; i++) { + GLYPH *g; + + g = &glyph_list[i]; + g->char_no = -1; + g->orig_code = -1; + g->name = "UNKNOWN"; + g->flags = GF_FLOAT; /* we start with float representation */ + } + + handle_gnames(); + + cursw->glmetrics(glyph_list); + cursw->fnmetrics(&fontm); + + original_scale_factor = 1000.0 / (double) fontm.units_per_em; + + if(transform == 0) + scale_factor = 1.0; /* don't transform */ + else + scale_factor = original_scale_factor; + + if(correctvsize && uni_sample!=0) { /* only for known languages */ + /* try to adjust the scale factor to make a typical + * uppercase character of hight at least (correctvsize), this + * may improve the appearance of the font but also + * make it weird, use with caution + */ + int ysz; + + ysz = iscale(glyph_list[encoding[uni_sample]].yMax); + if( ysz<correctvsize ) { + scale_factor *= (double)correctvsize / ysz; + } + } + +#ifdef XP_ONLY_BLOCKS + /* All glyphs from |xp_font_block_offset| to |xp_font_block_offset+255|*/ + for( i = xp_font_block_offset ; i < (xp_font_block_offset+256) ; i++ ) { + unsigned long ftindex; + + /* Remap X11 font index to FreeType font index */ + ftindex = FTRemap(xp_pstext_ft_face, &xp_xtf->mapping, i); + + if( ftindex < numglyphs ) { + glyph_list[ftindex].flags |= GF_USED; + } + } + + /* also always include .notdef */ + { + int notdef_found = FALSE; + + for (i = 0; i < numglyphs; i++) { + if(!strcmp(glyph_list[i].name, ".notdef")) { + glyph_list[i].flags |= GF_USED; + notdef_found = TRUE; + break; + } + } + + if( !notdef_found ) + { + /* No ".notdef" found ? + * Then copy outlines of char 0 to the first "free" slot and make + * it our ".notdef" char. + */ + for (i = 0; i < numglyphs; i++) { + if((glyph_list[i].flags & GF_USED) == 0) { + glyph_list[i] = glyph_list[0]; + glyph_list[i].flags |= GF_USED; + glyph_list[i].name = ".notdef"; + notdef_found = TRUE; + break; + } + } + } + + if( !notdef_found ) + { + /* This shoudl never happen... */ + fprintf(stderr, "ft2pt1: '.notdef' missing in generated font.\n"); + } + } +#else + if(allglyphs) { + for (i = 0; i < numglyphs; i++) { + glyph_list[i].flags |= GF_USED; + } + } else { + for (i = 0; i < ENCTABSZ; i++) { + glyph_list[encoding[i]].flags |= GF_USED; + } + + /* also always include .notdef */ + for (i = 0; i < numglyphs; i++) + if(!strcmp(glyph_list[i].name, ".notdef")) { + glyph_list[i].flags |= GF_USED; + break; + } + } +#endif /* XP_ONLY_BLOCKS */ + + for (i = 0; i < numglyphs; i++) { + if (glyph_list[i].flags & GF_USED) { + DBG_TO_GLYPH(&glyph_list[i]); + convert_glyf(i); + DBG_FROM_GLYPH(&glyph_list[i]); + } + } + + italic_angle = fontm.italic_angle; + + if (italic_angle > 45.0 || italic_angle < -45.0) + italic_angle = 0.0; /* consider buggy */ + + if (hints) { + findblues(); + for (i = 0; i < numglyphs; i++) { + if (glyph_list[i].flags & GF_USED) { + DBG_TO_GLYPH(&glyph_list[i]); + buildstems(&glyph_list[i]); + assertpath(glyph_list[i].entries, __FILE__, __LINE__, glyph_list[i].name); + DBG_FROM_GLYPH(&glyph_list[i]); + } + } + stemstatistics(); + } else { + for(i=0; i<4; i++) + bbox[i] = iscale(fontm.bbox[i]); + } + /* don't touch the width of fixed width fonts */ + if( fontm.is_fixed_pitch ) + correctwidth=0; + docorrectwidth(); /* checks correctwidth inside */ + if (reverse) + for (i = 0; i < numglyphs; i++) { + if (glyph_list[i].flags & GF_USED) { + DBG_TO_GLYPH(&glyph_list[i]); + reversepaths(&glyph_list[i]); + assertpath(glyph_list[i].entries, __FILE__, __LINE__, glyph_list[i].name); + DBG_FROM_GLYPH(&glyph_list[i]); + } + } + + +#if 0 + /* + ** It seems to bring troubles. The problem is that some + ** styles of the font may be recognized as fixed-width + ** while other styles of the same font as proportional. + ** So it's better to be commented out yet. + */ + if (tryfixed) + alignwidths(); +#endif + + if(trybold) { + forcebold = fontm.force_bold; + } + +#ifdef XP_PSTEXT + fprintf(pfa_file, "%%!PS-AdobeFont-1.0: %s\n", fontm.name_ps); +#else + fprintf(pfa_file, "%%!PS-AdobeFont-1.0: %s %s\n", fontm.name_ps, fontm.name_copyright); + time(&now); + fprintf(pfa_file, "%%%%CreationDate: %s", ctime(&now)); + fprintf(pfa_file, "%% Converted by ttf2pt1 %s/%s\n", TTF2PT1_VERSION, cursw->name); + fprintf(pfa_file, "%% Args: %s\n", cmdline); + fprintf(pfa_file, "%%%%EndComments\n"); +#endif /* XP_PSTEXT */ + fprintf(pfa_file, "12 dict begin\n/FontInfo 9 dict dup begin\n"); + + WARNING_3 fprintf(stderr, "FontName %s%s\n", fontm.name_ps, uni_font_name_suffix); + + + fprintf(pfa_file, " /version (%s) readonly def\n", fontm.name_version); + + fprintf(pfa_file, " /Notice (%s) readonly def\n", fontm.name_copyright); + + fprintf(pfa_file, " /FullName (%s) readonly def\n", fontm.name_full); + fprintf(pfa_file, " /FamilyName (%s) readonly def\n", fontm.name_family); + + if(wantuid) { + if(strUID) + fprintf(pfa_file, " /UniqueID %s def\n", strUID); + else { + numUID=0; + for(i=0; fontm.name_full[i]!=0; i++) { + numUID *= 37; /* magic number, good for hash */ + numUID += fontm.name_full[i]-' '; + /* if the name is long the first chars + * may be lost forever, so re-insert + * them thus making kind of CRC + */ + numUID += (numUID>>24) & 0xFF; + } + /* the range for private UIDs is 4 000 000 - 4 999 999 */ + fprintf(pfa_file, " /UniqueID %lu def\n", numUID%1000000+4000000); + } + } + + fprintf(pfa_file, " /Weight (%s) readonly def\n", fontm.name_style); + + fprintf(pfa_file, " /ItalicAngle %f def\n", italic_angle); + fprintf(pfa_file, " /isFixedPitch %s def\n", + fontm.is_fixed_pitch ? "true" : "false"); + + /* we don't print out the unused glyphs */ + nchars = 0; + for (i = 0; i < numglyphs; i++) { + if (glyph_list[i].flags & GF_USED) { + nchars++; + } + } + + fprintf(afm_file, "StartFontMetrics 4.1\n"); + fprintf(afm_file, "FontName %s%s\n", fontm.name_ps, uni_font_name_suffix); + fprintf(afm_file, "FullName %s\n", fontm.name_full); + fprintf(afm_file, "Notice %s\n", fontm.name_copyright); + fprintf(afm_file, "EncodingScheme FontSpecific\n"); + fprintf(afm_file, "FamilyName %s\n", fontm.name_family); + fprintf(afm_file, "Weight %s\n", fontm.name_style); + fprintf(afm_file, "Version %s\n", fontm.name_version); + fprintf(afm_file, "Characters %d\n", nchars); + fprintf(afm_file, "ItalicAngle %.1f\n", italic_angle); + + fprintf(afm_file, "Ascender %d\n", iscale(fontm.ascender)); + fprintf(afm_file, "Descender %d\n", iscale(fontm.descender)); + + fprintf(pfa_file, " /UnderlinePosition %d def\n", + iscale(fontm.underline_position)); + + fprintf(pfa_file, " /UnderlineThickness %hd def\n", + iscale(fontm.underline_thickness)); + + fprintf(pfa_file, "end readonly def\n"); + + fprintf(afm_file, "UnderlineThickness %d\n", + iscale(fontm.underline_thickness)); + + fprintf(afm_file, "UnderlinePosition %d\n", + iscale(fontm.underline_position)); + + fprintf(afm_file, "IsFixedPitch %s\n", + fontm.is_fixed_pitch ? "true" : "false"); + fprintf(afm_file, "FontBBox %d %d %d %d\n", + bbox[0], bbox[1], bbox[2], bbox[3]); + + fprintf(pfa_file, "/FontName /%s%s def\n", fontm.name_ps, uni_font_name_suffix); + fprintf(pfa_file, "/PaintType 0 def\n/StrokeWidth 0 def\n"); + /* I'm not sure if these are fixed */ + fprintf(pfa_file, "/FontType 1 def\n"); + + if (transform) { + fprintf(pfa_file, "/FontMatrix [0.001 0 0 0.001 0 0] def\n"); + } else { + fprintf(pfa_file, "/FontMatrix [%9.7f 0 0 %9.7f 0 0] def\n", + original_scale_factor / 1000.0, original_scale_factor / 1000.0); + } + + fprintf(pfa_file, "/FontBBox {%d %d %d %d} readonly def\n", + bbox[0], bbox[1], bbox[2], bbox[3]); + + fprintf(pfa_file, "/Encoding 256 array\n"); + /* determine number of elements for metrics table */ + nmetrics = 256; + for (i = 0; i < numglyphs; i++) { + if( glyph_list[i].flags & GF_USED + && glyph_list[i].char_no == -1 ) { + nmetrics++; + } + } + fprintf(afm_file, "StartCharMetrics %d\n", nmetrics); + + fprintf(dvienc_file, "/%s%sEncoding [\n", + fontm.name_ps, uni_font_name_suffix); + +#ifdef XP_PSTEXT + { + int linewidth = 0; + for (i = 0; i < 256; i++) { /* here 256, not ENCTABSZ */ + linewidth += strlen(glyph_list[encoding[i]].name) + 14 + 8; + fprintf(pfa_file, "dup %d /%s put%s", + i, + glyph_list[encoding[i]].name, + (linewidth > 70 || i == 255)?(linewidth = 0, "\n"):("\t")); + if( glyph_list[encoding[i]].flags & GF_USED ) { + print_glyph_metrics(i, encoding[i]); + } + if (encoding[i]) + fprintf (dvienc_file, "/index0x%04X\n", encoding[i]); + else + fprintf (dvienc_file, "/.notdef\n"); + } + } +#else + for (i = 0; i < 256; i++) { /* here 256, not ENCTABSZ */ + fprintf(pfa_file, + "dup %d /%s put\n", i, glyph_list[encoding[i]].name); + if( glyph_list[encoding[i]].flags & GF_USED ) { + print_glyph_metrics(i, encoding[i]); + } + if (encoding[i]) + fprintf (dvienc_file, "/index0x%04X\n", encoding[i]); + else + fprintf (dvienc_file, "/.notdef\n"); + } +#endif /* XP_PSTEXT */ + /* print the metrics for glyphs not in encoding table */ + for(i=0; i<numglyphs; i++) { + if( (glyph_list[i].flags & GF_USED) + && glyph_list[i].char_no == -1 ) { + print_glyph_metrics(-1, i); + } + } + + fprintf(pfa_file, "readonly def\ncurrentdict end\ncurrentfile eexec\n"); + fprintf(pfa_file, "dup /Private 16 dict dup begin\n"); + + fprintf(pfa_file, "/RD{string currentfile exch readstring pop}executeonly def\n"); + fprintf(pfa_file, "/ND{noaccess def}executeonly def\n"); + fprintf(pfa_file, "/NP{noaccess put}executeonly def\n"); + + /* UniqueID must be shown twice, in both font and Private dictionary */ + if(wantuid) { + if(strUID) + fprintf(pfa_file, "/UniqueID %s def\n", strUID); + else + /* the range for private UIDs is 4 000 000 - 4 999 999 */ + fprintf(pfa_file, "/UniqueID %lu def\n", numUID%1000000+4000000); + } + + if(forcebold==0) + fprintf(pfa_file, "/ForceBold false def\n"); + else if(forcebold==1) + fprintf(pfa_file, "/ForceBold true def\n"); + + fprintf(pfa_file, "/BlueValues [ "); + for (i = 0; i < nblues; i++) + fprintf(pfa_file, "%d ", bluevalues[i]); + fprintf(pfa_file, "] def\n"); + + fprintf(pfa_file, "/OtherBlues [ "); + for (i = 0; i < notherb; i++) + fprintf(pfa_file, "%d ", otherblues[i]); + fprintf(pfa_file, "] def\n"); + + if (stdhw != 0) + fprintf(pfa_file, "/StdHW [ %d ] def\n", stdhw); + if (stdvw != 0) + fprintf(pfa_file, "/StdVW [ %d ] def\n", stdvw); + fprintf(pfa_file, "/StemSnapH [ "); + for (i = 0; i < 12 && stemsnaph[i] != 0; i++) + fprintf(pfa_file, "%d ", stemsnaph[i]); + fprintf(pfa_file, "] def\n"); + fprintf(pfa_file, "/StemSnapV [ "); + for (i = 0; i < 12 && stemsnapv[i] != 0; i++) + fprintf(pfa_file, "%d ", stemsnapv[i]); + fprintf(pfa_file, "] def\n"); + + fprintf(pfa_file, "/MinFeature {16 16} def\n"); + /* Are these fixed also ? */ + fprintf(pfa_file, "/password 5839 def\n"); + + /* calculate the number of subroutines */ + + subid=5; + for (i = 0; i < numglyphs; i++) { + if (glyph_list[i].flags & GF_USED) { + subid+=glyph_list[i].nsg; + } + } + + fprintf(pfa_file, "/Subrs %d array\n", subid); + /* standard subroutines */ + fprintf(pfa_file, "dup 0 {\n\t3 0 callothersubr pop pop setcurrentpoint return\n\t} NP\n"); + fprintf(pfa_file, "dup 1 {\n\t0 1 callothersubr return\n\t} NP\n"); + fprintf(pfa_file, "dup 2 {\n\t0 2 callothersubr return\n\t} NP\n"); + fprintf(pfa_file, "dup 3 {\n\treturn\n\t} NP\n"); + /* our sub to make the hint substitution code shorter */ + fprintf(pfa_file, "dup 4 {\n\t1 3 callothersubr pop callsubr return\n\t} NP\n"); + + if(pfa_file != null_file) { /* save time if the output would be wasted */ + /* print the hinting subroutines */ + subid=5; + for (i = 0; i < numglyphs; i++) { + if (glyph_list[i].flags & GF_USED) { + subid+=print_glyph_subs(i, subid); + } + } + + fprintf(pfa_file, "ND\n"); + + fprintf(pfa_file, "2 index /CharStrings %d dict dup begin\n", nchars); + + for (i = 0; i < numglyphs; i++) { + if (glyph_list[i].flags & GF_USED) { + print_glyph(i); + } + } + } + + + fprintf(pfa_file, "end\nend\nreadonly put\n"); + fprintf(pfa_file, "noaccess put\n"); + fprintf(pfa_file, "dup/FontName get exch definefont pop\n"); + fprintf(pfa_file, "mark currentfile closefile\n"); + fprintf(pfa_file, "cleartomark\n"); + if(pfa_file != null_file) + fclose(pfa_file); + + fprintf(afm_file, "EndCharMetrics\n"); + + if(afm_file != null_file) { /* save time if the output would be wasted */ + /* print the kerning data if present */ + cursw->kerning(glyph_list); + print_kerning(afm_file); + } + + fprintf(afm_file, "EndFontMetrics\n"); + if(afm_file != null_file) + fclose(afm_file); + + fprintf(dvienc_file, "] def\n"); + if(dvienc_file != null_file) + fclose(dvienc_file); + + WARNING_1 fprintf(stderr, "Finished - font files created\n"); + + cursw->close(); + +#ifndef WINDOWS + while (wait(&ws) > 0) { + } +#else + if (encode && pfa_file != null_file) { + extern FILE *ifp, *ofp; /* from t1asm.c */ + + snprintf(filename, sizeof filename, "%s.%s", argv[2], pfbflag ? "pfb" : "pfa" ); + + if ((ofp = fopen(filename, "w+b")) == NULL) { + fprintf(stderr, "**** Cannot create %s ****\n", filename); + exit(1); + } else { + WARNING_2 fprintf(stderr, "Creating file %s\n", filename); + } + + snprintf(filename, sizeof filename, "%s.t1a", argv[2]); + + if ((ifp = fopen(filename, "rb")) == NULL) { + fprintf(stderr, "**** Cannot read %s ****\n", filename); + exit(1); + } else { + WARNING_2 fprintf(stderr, "Converting file %s\n", filename); + } + + runt1asm(pfbflag); + + WARNING_2 fprintf(stderr, "Removing file %s\n", filename); + if(unlink(filename) < 0) + WARNING_1 fprintf(stderr, "Unable to remove file %s\n", filename); + } +#endif /* WINDOWS */ + + fclose(null_file); + return 0; +} |