From 16d779115982012db0c93d8c4c735a9fb7a57dfb Mon Sep 17 00:00:00 2001 From: Behdad Esfahbod Date: Fri, 4 Aug 2017 15:13:34 +0100 Subject: [PATCH] Remove use of psnames for charset construction This is ancient. No font we care baout uses them. Kill. This also makes fc-glyphname machinery obsolete. Should be removed. --- src/fcfreetype.c | 179 ----------------------------------------------- 1 file changed, 179 deletions(-) diff --git a/src/fcfreetype.c b/src/fcfreetype.c index 6224ca6..1115463 100644 --- a/src/fcfreetype.c +++ b/src/fcfreetype.c @@ -2007,122 +2007,6 @@ static const FT_Encoding fcFontEncodings[] = { #define NUM_DECODE (int) (sizeof (fcFontEncodings) / sizeof (fcFontEncodings[0])) -#include "../fc-glyphname/fcglyphname.h" - -static FcChar32 -FcHashGlyphName (const FcChar8 *name) -{ - FcChar32 h = 0; - FcChar8 c; - - while ((c = *name++)) - { - h = ((h << 1) | (h >> 31)) ^ c; - } - return h; -} - -#if HAVE_FT_HAS_PS_GLYPH_NAMES -/* - * Use Type1 glyph names for fonts which have reliable names - * and which export an Adobe Custom mapping - */ -static FcBool -FcFreeTypeUseNames (FT_Face face) -{ - FT_Int map; - - if (!FT_Has_PS_Glyph_Names (face)) - return FcFalse; - for (map = 0; map < face->num_charmaps; map++) - if (face->charmaps[map]->encoding == ft_encoding_adobe_custom) - return FcTrue; - return FcFalse; -} - -static const FcChar8 * -FcUcs4ToGlyphName (FcChar32 ucs4) -{ - int i = (int) (ucs4 % FC_GLYPHNAME_HASH); - int r = 0; - FcGlyphId gn; - - while ((gn = _fc_ucs_to_name[i]) != -1) - { - if (_fc_glyph_names[gn].ucs == ucs4) - return _fc_glyph_names[gn].name; - if (!r) - { - r = (int) (ucs4 % FC_GLYPHNAME_REHASH); - if (!r) - r = 1; - } - i += r; - if (i >= FC_GLYPHNAME_HASH) - i -= FC_GLYPHNAME_HASH; - } - return 0; -} - -static FcChar32 -FcGlyphNameToUcs4 (FcChar8 *name) -{ - FcChar32 h = FcHashGlyphName (name); - int i = (int) (h % FC_GLYPHNAME_HASH); - int r = 0; - FcGlyphId gn; - - while ((gn = _fc_name_to_ucs[i]) != -1) - { - if (!strcmp ((char *) name, (char *) _fc_glyph_names[gn].name)) - return _fc_glyph_names[gn].ucs; - if (!r) - { - r = (int) (h % FC_GLYPHNAME_REHASH); - if (!r) - r = 1; - } - i += r; - if (i >= FC_GLYPHNAME_HASH) - i -= FC_GLYPHNAME_HASH; - } - return 0xffff; -} - -/* - * Work around a bug in some FreeType versions which fail - * to correctly bounds check glyph name buffers and overwrite - * the stack. As Postscript names have a limit of 127 characters, - * this should be sufficient. - */ - -#if FC_GLYPHNAME_MAXLEN < 127 -# define FC_GLYPHNAME_BUFLEN 127 -#else -# define FC_GLYPHNAME_BUFLEN FC_GLYPHNAME_MAXLEN -#endif - -/* - * Search through a font for a glyph by name. This is - * currently a linear search as there doesn't appear to be - * any defined order within the font - */ -static FT_UInt -FcFreeTypeGlyphNameIndex (FT_Face face, const FcChar8 *name) -{ - FT_UInt gindex; - FcChar8 name_buf[FC_GLYPHNAME_BUFLEN + 2]; - - for (gindex = 0; gindex < (FT_UInt) face->num_glyphs; gindex++) - { - if (FT_Get_Glyph_Name (face, gindex, name_buf, FC_GLYPHNAME_BUFLEN+1) == 0) - if (!strcmp ((char *) name, (char *) name_buf)) - return gindex; - } - return 0; -} -#endif - /* * Map a UCS4 glyph to a glyph index. Use all available encoding * tables to try and find one that works. This information is expected @@ -2180,21 +2064,6 @@ FcFreeTypeCharIndex (FT_Face face, FcChar32 ucs4) return glyphindex; } } -#if HAVE_FT_HAS_PS_GLYPH_NAMES - /* - * Check postscript name table if present - */ - if (FcFreeTypeUseNames (face)) - { - const FcChar8 *name = FcUcs4ToGlyphName (ucs4); - if (name) - { - glyphindex = FcFreeTypeGlyphNameIndex (face, name); - if (glyphindex) - return glyphindex; - } - } -#endif return 0; } @@ -2362,54 +2231,6 @@ FcFreeTypeCharSetAndSpacing (FT_Face face, FcBlanks *blanks, int *spacing) break; } -#if HAVE_FT_HAS_PS_GLYPH_NAMES - /* - * Add mapping from PS glyph names if available - */ - if (FcFreeTypeUseNames (face)) - { - FcChar8 name_buf[FC_GLYPHNAME_BUFLEN + 2]; - - for (glyph = 0; glyph < (FT_UInt) face->num_glyphs; glyph++) - { - if (FT_Get_Glyph_Name (face, glyph, name_buf, FC_GLYPHNAME_BUFLEN+1) == 0) - { - ucs4 = FcGlyphNameToUcs4 (name_buf); - if (ucs4 != 0xffff && - FcFreeTypeCheckGlyph (face, ucs4, glyph, &advance)) - { - if (advance) - { - if (!has_advance) - { - has_advance = FcTrue; - advance_one = advance; - } - else if (!APPROXIMATELY_EQUAL (advance, advance_one)) - { - if (fixed_advance) - { - dual_advance = FcTrue; - fixed_advance = FcFalse; - advance_two = advance; - } - else if (!APPROXIMATELY_EQUAL (advance, advance_two)) - dual_advance = FcFalse; - } - } - leaf = FcCharSetFindLeafCreate (fcs, ucs4); - if (!leaf) - goto bail1; - leaf->map[(ucs4 & 0xff) >> 5] |= (1 << (ucs4 & 0x1f)); -#ifdef CHECK - if (ucs4 > font_max) - font_max = ucs4; -#endif - } - } - } - } -#endif if (fixed_advance) *spacing = FC_MONO; else if (dual_advance && APPROXIMATELY_EQUAL (2 * FC_MIN (advance_one, advance_two), FC_MAX (advance_one, advance_two)))