summary refs log tree commit diff
path: root/src/freetype/afscript.h
diff options
context:
space:
mode:
authorWlodekM <[email protected]>2024-06-16 10:35:45 +0300
committerWlodekM <[email protected]>2024-06-16 10:35:45 +0300
commitabef6da56913f1c55528103e60a50451a39628b1 (patch)
treeb3c8092471ecbb73e568cd0d336efa0e7871ee8d /src/freetype/afscript.h
initial commit
Diffstat (limited to 'src/freetype/afscript.h')
-rw-r--r--src/freetype/afscript.h351
1 files changed, 351 insertions, 0 deletions
diff --git a/src/freetype/afscript.h b/src/freetype/afscript.h
new file mode 100644
index 0000000..1bf5790
--- /dev/null
+++ b/src/freetype/afscript.h
@@ -0,0 +1,351 @@
+/***************************************************************************/
+/*                                                                         */
+/*  afscript.h                                                             */
+/*                                                                         */
+/*    Auto-fitter scripts (specification only).                            */
+/*                                                                         */
+/*  Copyright 2013-2018 by                                                 */
+/*  David Turner, Robert Wilhelm, and Werner Lemberg.                      */
+/*                                                                         */
+/*  This file is part of the FreeType project, and may only be used,       */
+/*  modified, and distributed under the terms of the FreeType project      */
+/*  license, LICENSE.TXT.  By continuing to use, modify, or distribute     */
+/*  this file you indicate that you have read the license and              */
+/*  understand and accept it fully.                                        */
+/*                                                                         */
+/***************************************************************************/
+
+
+  /* The following part can be included multiple times. */
+  /* Define `SCRIPT' as needed.                         */
+
+
+  /* Add new scripts here.  The first and second arguments are the    */
+  /* script name in lowercase and uppercase, respectively, followed   */
+  /* by a description string.  Then comes the corresponding HarfBuzz  */
+  /* script name tag, followed by a string of standard characters (to */
+  /* derive the standard width and height of stems).                  */
+  /*                                                                  */
+  /* Note that fallback scripts only have a default style, thus we    */
+  /* use `HB_SCRIPT_INVALID' as the HarfBuzz script name tag for      */
+  /* them.                                                            */
+
+  SCRIPT( adlm, ADLM,
+          "Adlam",
+          HB_SCRIPT_ADLAM,
+          HINTING_BOTTOM_TO_TOP,
+          "\xF0\x9E\xA4\x8C \xF0\x9E\xA4\xAE" ) /* 𞤌 𞤮 */
+
+  SCRIPT( arab, ARAB,
+          "Arabic",
+          HB_SCRIPT_ARABIC,
+          HINTING_BOTTOM_TO_TOP,
+          "\xD9\x84 \xD8\xAD \xD9\x80" ) /* ل ح ـ */
+
+  SCRIPT( armn, ARMN,
+          "Armenian",
+          HB_SCRIPT_ARMENIAN,
+          HINTING_BOTTOM_TO_TOP,
+          "\xD5\xBD \xD5\x8D" ) /* ս Ս */
+
+  SCRIPT( avst, AVST,
+          "Avestan",
+          HB_SCRIPT_AVESTAN,
+          HINTING_BOTTOM_TO_TOP,
+          "\xF0\x90\xAC\x9A" ) /* 𐬚 */
+
+  SCRIPT( bamu, BAMU,
+          "Bamum",
+          HB_SCRIPT_BAMUM,
+          HINTING_BOTTOM_TO_TOP,
+          "\xEA\x9B\x81 \xEA\x9B\xAF" ) /* ꛁ ꛯ */
+
+  /* there are no simple forms for letters; we thus use two digit shapes */
+  SCRIPT( beng, BENG,
+          "Bengali",
+          HB_SCRIPT_BENGALI,
+          HINTING_TOP_TO_BOTTOM,
+          "\xE0\xA7\xA6 \xE0\xA7\xAA" ) /* ০ ৪ */
+
+  SCRIPT( buhd, BUHD,
+          "Buhid",
+          HB_SCRIPT_BUHID,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE1\x9D\x8B \xE1\x9D\x8F" ) /* ᝋ ᝏ */
+
+  SCRIPT( cakm, CAKM,
+          "Chakma",
+          HB_SCRIPT_CHAKMA,
+          HINTING_BOTTOM_TO_TOP,
+          "\xF0\x91\x84\xA4 \xF0\x91\x84\x89 \xF0\x91\x84\x9B" ) /* 𑄤 𑄉 𑄛 */
+
+  SCRIPT( cans, CANS,
+          "Canadian Syllabics",
+          HB_SCRIPT_CANADIAN_SYLLABICS,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE1\x91\x8C \xE1\x93\x9A" ) /* ᑌ ᓚ */
+
+  SCRIPT( cari, CARI,
+          "Carian",
+          HB_SCRIPT_CARIAN,
+          HINTING_BOTTOM_TO_TOP,
+          "\xF0\x90\x8A\xAB \xF0\x90\x8B\x89" ) /* 𐊫 𐋉 */
+
+  SCRIPT( cher, CHER,
+          "Cherokee",
+          HB_SCRIPT_CHEROKEE,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE1\x8E\xA4 \xE1\x8F\x85 \xEA\xAE\x95" ) /* Ꭴ Ꮕ ꮕ */
+
+  SCRIPT( copt, COPT,
+          "Coptic",
+          HB_SCRIPT_COPTIC,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE2\xB2\x9E \xE2\xB2\x9F" ) /* Ⲟ ⲟ */
+
+  SCRIPT( cprt, CPRT,
+          "Cypriot",
+          HB_SCRIPT_CYPRIOT,
+          HINTING_BOTTOM_TO_TOP,
+          "\xF0\x90\xA0\x85 \xF0\x90\xA0\xA3" ) /* 𐠅 𐠣 */
+
+  SCRIPT( cyrl, CYRL,
+          "Cyrillic",
+          HB_SCRIPT_CYRILLIC,
+          HINTING_BOTTOM_TO_TOP,
+          "\xD0\xBE \xD0\x9E" ) /* о О */
+
+  SCRIPT( deva, DEVA,
+          "Devanagari",
+          HB_SCRIPT_DEVANAGARI,
+          HINTING_TOP_TO_BOTTOM,
+          "\xE0\xA4\xA0 \xE0\xA4\xB5 \xE0\xA4\x9F" ) /* ठ व ट */
+
+  SCRIPT( dsrt, DSRT,
+          "Deseret",
+          HB_SCRIPT_DESERET,
+          HINTING_BOTTOM_TO_TOP,
+          "\xF0\x90\x90\x84 \xF0\x90\x90\xAC" ) /* 𐐄 𐐬 */
+
+  SCRIPT( ethi, ETHI,
+          "Ethiopic",
+          HB_SCRIPT_ETHIOPIC,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE1\x8B\x90" ) /* ዐ */
+
+  SCRIPT( geor, GEOR,
+          "Georgian (Mkhedruli)",
+          HB_SCRIPT_GEORGIAN,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE1\x83\x98 \xE1\x83\x94 \xE1\x83\x90 \xE1\xB2\xBF" ) /* ი ე ა Ი */
+
+  SCRIPT( geok, GEOK,
+          "Georgian (Khutsuri)",
+          HB_SCRIPT_INVALID,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE1\x82\xB6 \xE1\x82\xB1 \xE2\xB4\x99" ) /* Ⴖ Ⴑ ⴙ */
+
+  SCRIPT( glag, GLAG,
+          "Glagolitic",
+          HB_SCRIPT_GLAGOLITIC,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE2\xB0\x95 \xE2\xB1\x85" ) /* Ⱅ ⱅ */
+
+  SCRIPT( goth, GOTH,
+          "Gothic",
+          HB_SCRIPT_GOTHIC,
+          HINTING_TOP_TO_BOTTOM,
+          "\xF0\x90\x8C\xB4 \xF0\x90\x8C\xBE \xF0\x90\x8D\x83" ) /* 𐌴 𐌾 𐍃 */
+
+  SCRIPT( grek, GREK,
+          "Greek",
+          HB_SCRIPT_GREEK,
+          HINTING_BOTTOM_TO_TOP,
+          "\xCE\xBF \xCE\x9F" ) /* ο Ο */
+
+  SCRIPT( gujr, GUJR,
+          "Gujarati",
+          HB_SCRIPT_GUJARATI,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE0\xAA\x9F \xE0\xAB\xA6" ) /* ટ ૦ */
+
+  SCRIPT( guru, GURU,
+          "Gurmukhi",
+          HB_SCRIPT_GURMUKHI,
+          HINTING_TOP_TO_BOTTOM,
+          "\xE0\xA8\xA0 \xE0\xA8\xB0 \xE0\xA9\xA6" ) /* ਠ ਰ ੦ */
+
+  SCRIPT( hebr, HEBR,
+          "Hebrew",
+          HB_SCRIPT_HEBREW,
+          HINTING_BOTTOM_TO_TOP,
+          "\xD7\x9D" ) /* ם */
+
+  SCRIPT( kali, KALI,
+          "Kayah Li",
+          HB_SCRIPT_KAYAH_LI,
+          HINTING_BOTTOM_TO_TOP,
+          "\xEA\xA4\x8D \xEA\xA4\x80" ) /* ꤍ ꤀ */
+
+  /* only digit zero has a simple shape in the Khmer script */
+  SCRIPT( khmr, KHMR,
+          "Khmer",
+          HB_SCRIPT_KHMER,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE1\x9F\xA0" ) /* ០ */
+
+  SCRIPT( khms, KHMS,
+          "Khmer Symbols",
+          HB_SCRIPT_INVALID,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE1\xA7\xA1 \xE1\xA7\xAA" ) /* ᧡ ᧪ */
+
+  SCRIPT( knda, KNDA,
+          "Kannada",
+          HB_SCRIPT_KANNADA,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE0\xB3\xA6 \xE0\xB2\xAC" ) /* ೦ ಬ */
+
+  /* only digit zero has a simple shape in the Lao script */
+  SCRIPT( lao, LAO,
+          "Lao",
+          HB_SCRIPT_LAO,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE0\xBB\x90" ) /* ໐ */
+
+  SCRIPT( latn, LATN,
+          "Latin",
+          HB_SCRIPT_LATIN,
+          HINTING_BOTTOM_TO_TOP,
+          "o O 0" )
+
+  SCRIPT( latb, LATB,
+          "Latin Subscript Fallback",
+          HB_SCRIPT_INVALID,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE2\x82\x92 \xE2\x82\x80" ) /* ₒ ₀ */
+
+  SCRIPT( latp, LATP,
+          "Latin Superscript Fallback",
+          HB_SCRIPT_INVALID,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE1\xB5\x92 \xE1\xB4\xBC \xE2\x81\xB0" ) /* ᵒ ᴼ ⁰ */
+
+  SCRIPT( lisu, LISU,
+          "Lisu",
+          HB_SCRIPT_LISU,
+          HINTING_BOTTOM_TO_TOP,
+          "\xEA\x93\xB3" ) /* ꓳ */
+
+  SCRIPT( mlym, MLYM,
+          "Malayalam",
+          HB_SCRIPT_MALAYALAM,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE0\xB4\xA0 \xE0\xB4\xB1" ) /* ഠ റ */
+
+  SCRIPT( mymr, MYMR,
+          "Myanmar",
+          HB_SCRIPT_MYANMAR,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE1\x80\x9D \xE1\x80\x84 \xE1\x80\x82" ) /* ဝ င ဂ */
+
+  SCRIPT( nkoo, NKOO,
+          "N'Ko",
+          HB_SCRIPT_NKO,
+          HINTING_BOTTOM_TO_TOP,
+          "\xDF\x8B \xDF\x80" ) /* ߋ ߀ */
+
+  SCRIPT( none, NONE,
+          "no script",
+          HB_SCRIPT_INVALID,
+          HINTING_BOTTOM_TO_TOP,
+          "" )
+
+  SCRIPT( olck, OLCK,
+          "Ol Chiki",
+          HB_SCRIPT_OL_CHIKI,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE1\xB1\x9B" ) /* ᱛ */
+
+  SCRIPT( orkh, ORKH,
+          "Old Turkic",
+          HB_SCRIPT_OLD_TURKIC,
+          HINTING_BOTTOM_TO_TOP,
+          "\xF0\x90\xB0\x97" ) /* 𐰗 */
+
+  SCRIPT( osge, OSGE,
+          "Osage",
+          HB_SCRIPT_OSAGE,
+          HINTING_BOTTOM_TO_TOP,
+          "\xF0\x90\x93\x82 \xF0\x90\x93\xAA" ) /* 𐓂 𐓪 */
+
+  SCRIPT( osma, OSMA,
+          "Osmanya",
+          HB_SCRIPT_OSMANYA,
+          HINTING_BOTTOM_TO_TOP,
+          "\xF0\x90\x92\x86 \xF0\x90\x92\xA0" ) /* 𐒆 𐒠 */
+
+  SCRIPT( saur, SAUR,
+          "Saurashtra",
+          HB_SCRIPT_SAURASHTRA,
+          HINTING_BOTTOM_TO_TOP,
+          "\xEA\xA2\x9D \xEA\xA3\x90" ) /* ꢝ ꣐ */
+
+  SCRIPT( shaw, SHAW,
+          "Shavian",
+          HB_SCRIPT_SHAVIAN,
+          HINTING_BOTTOM_TO_TOP,
+          "\xF0\x90\x91\xB4" ) /* 𐑴 */
+
+  SCRIPT( sinh, SINH,
+          "Sinhala",
+          HB_SCRIPT_SINHALA,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE0\xB6\xA7" ) /* ට */
+
+  /* only digit zero has a simple (round) shape in the Sundanese script */
+  SCRIPT( sund, SUND,
+          "Sundanese",
+          HB_SCRIPT_SUNDANESE,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE1\xAE\xB0" ) /* ᮰ */
+
+  /* only digit zero has a simple (round) shape in the Tamil script */
+  SCRIPT( taml, TAML,
+          "Tamil",
+          HB_SCRIPT_TAMIL,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE0\xAF\xA6" ) /* ௦ */
+
+  SCRIPT( tavt, TAVT,
+          "Tai Viet",
+          HB_SCRIPT_TAI_VIET,
+          HINTING_BOTTOM_TO_TOP,
+          "\xEA\xAA\x92 \xEA\xAA\xAB" ) /* ꪒ ꪫ */
+
+  /* there are no simple forms for letters; we thus use two digit shapes */
+  SCRIPT( telu, TELU,
+          "Telugu",
+          HB_SCRIPT_TELUGU,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE0\xB1\xA6 \xE0\xB1\xA7" ) /* ౦ ౧ */
+
+  SCRIPT( tfng, TFNG,
+          "Tifinagh",
+          HB_SCRIPT_TIFINAGH,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE2\xB5\x94" ) /* ⵔ */
+
+  SCRIPT( thai, THAI,
+          "Thai",
+          HB_SCRIPT_THAI,
+          HINTING_BOTTOM_TO_TOP,
+          "\xE0\xB8\xB2 \xE0\xB9\x85 \xE0\xB9\x90" ) /* า ๅ ๐ */
+
+  SCRIPT( vaii, VAII,
+          "Vai",
+          HB_SCRIPT_VAI,
+          HINTING_BOTTOM_TO_TOP,
+          "\xEA\x98\x93 \xEA\x96\x9C \xEA\x96\xB4" ) /* ꘓ ꖜ ꖴ */
+
+/* END */