Fixes for Android GN build input/outputs
[chromium-blink-merge.git] / third_party / harfbuzz-ng / src / hb-ot-shape-complex-indic.cc
blob7723600410a9be780dcb304936351d1b8cc1798a
1 /*
2 * Copyright © 2011,2012 Google, Inc.
4 * This is part of HarfBuzz, a text shaping library.
6 * Permission is hereby granted, without written agreement and without
7 * license or royalty fees, to use, copy, modify, and distribute this
8 * software and its documentation for any purpose, provided that the
9 * above copyright notice and the following two paragraphs appear in
10 * all copies of this software.
12 * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
13 * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
14 * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
15 * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
16 * DAMAGE.
18 * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
19 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
20 * FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS
21 * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
22 * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
24 * Google Author(s): Behdad Esfahbod
27 #include "hb-ot-shape-complex-indic-private.hh"
28 #include "hb-ot-layout-private.hh"
30 /* buffer var allocations */
31 #define indic_category() complex_var_u8_0() /* indic_category_t */
32 #define indic_position() complex_var_u8_1() /* indic_position_t */
36 * Indic shaper.
40 #define IN_HALF_BLOCK(u, Base) (((u) & ~0x7Fu) == (Base))
42 #define IS_DEVA(u) (IN_HALF_BLOCK (u, 0x0900u))
43 #define IS_BENG(u) (IN_HALF_BLOCK (u, 0x0980u))
44 #define IS_GURU(u) (IN_HALF_BLOCK (u, 0x0A00u))
45 #define IS_GUJR(u) (IN_HALF_BLOCK (u, 0x0A80u))
46 #define IS_ORYA(u) (IN_HALF_BLOCK (u, 0x0B00u))
47 #define IS_TAML(u) (IN_HALF_BLOCK (u, 0x0B80u))
48 #define IS_TELU(u) (IN_HALF_BLOCK (u, 0x0C00u))
49 #define IS_KNDA(u) (IN_HALF_BLOCK (u, 0x0C80u))
50 #define IS_MLYM(u) (IN_HALF_BLOCK (u, 0x0D00u))
51 #define IS_SINH(u) (IN_HALF_BLOCK (u, 0x0D80u))
52 #define IS_KHMR(u) (IN_HALF_BLOCK (u, 0x1780u))
55 #define MATRA_POS_LEFT(u) POS_PRE_M
56 #define MATRA_POS_RIGHT(u) ( \
57 IS_DEVA(u) ? POS_AFTER_SUB : \
58 IS_BENG(u) ? POS_AFTER_POST : \
59 IS_GURU(u) ? POS_AFTER_POST : \
60 IS_GUJR(u) ? POS_AFTER_POST : \
61 IS_ORYA(u) ? POS_AFTER_POST : \
62 IS_TAML(u) ? POS_AFTER_POST : \
63 IS_TELU(u) ? (u <= 0x0C42u ? POS_BEFORE_SUB : POS_AFTER_SUB) : \
64 IS_KNDA(u) ? (u < 0x0CC3u || u > 0xCD6u ? POS_BEFORE_SUB : POS_AFTER_SUB) : \
65 IS_MLYM(u) ? POS_AFTER_POST : \
66 IS_SINH(u) ? POS_AFTER_SUB : \
67 IS_KHMR(u) ? POS_AFTER_POST : \
68 /*default*/ POS_AFTER_SUB \
70 #define MATRA_POS_TOP(u) ( /* BENG and MLYM don't have top matras. */ \
71 IS_DEVA(u) ? POS_AFTER_SUB : \
72 IS_GURU(u) ? POS_AFTER_POST : /* Deviate from spec */ \
73 IS_GUJR(u) ? POS_AFTER_SUB : \
74 IS_ORYA(u) ? POS_AFTER_MAIN : \
75 IS_TAML(u) ? POS_AFTER_SUB : \
76 IS_TELU(u) ? POS_BEFORE_SUB : \
77 IS_KNDA(u) ? POS_BEFORE_SUB : \
78 IS_SINH(u) ? POS_AFTER_SUB : \
79 IS_KHMR(u) ? POS_AFTER_POST : \
80 /*default*/ POS_AFTER_SUB \
82 #define MATRA_POS_BOTTOM(u) ( \
83 IS_DEVA(u) ? POS_AFTER_SUB : \
84 IS_BENG(u) ? POS_AFTER_SUB : \
85 IS_GURU(u) ? POS_AFTER_POST : \
86 IS_GUJR(u) ? POS_AFTER_POST : \
87 IS_ORYA(u) ? POS_AFTER_SUB : \
88 IS_TAML(u) ? POS_AFTER_POST : \
89 IS_TELU(u) ? POS_BEFORE_SUB : \
90 IS_KNDA(u) ? POS_BEFORE_SUB : \
91 IS_MLYM(u) ? POS_AFTER_POST : \
92 IS_SINH(u) ? POS_AFTER_SUB : \
93 IS_KHMR(u) ? POS_AFTER_POST : \
94 /*default*/ POS_AFTER_SUB \
97 static inline indic_position_t
98 matra_position (hb_codepoint_t u, indic_position_t side)
100 switch ((int) side)
102 case POS_PRE_C: return MATRA_POS_LEFT (u);
103 case POS_POST_C: return MATRA_POS_RIGHT (u);
104 case POS_ABOVE_C: return MATRA_POS_TOP (u);
105 case POS_BELOW_C: return MATRA_POS_BOTTOM (u);
107 return side;
110 /* XXX
111 * This is a hack for now. We should move this data into the main Indic table.
112 * Or completely remove it and just check in the tables.
114 static const hb_codepoint_t ra_chars[] = {
115 0x0930u, /* Devanagari */
116 0x09B0u, /* Bengali */
117 0x09F0u, /* Bengali */
118 0x0A30u, /* Gurmukhi */ /* No Reph */
119 0x0AB0u, /* Gujarati */
120 0x0B30u, /* Oriya */
121 0x0BB0u, /* Tamil */ /* No Reph */
122 0x0C30u, /* Telugu */ /* Reph formed only with ZWJ */
123 0x0CB0u, /* Kannada */
124 0x0D30u, /* Malayalam */ /* No Reph, Logical Repha */
126 0x0DBBu, /* Sinhala */ /* Reph formed only with ZWJ */
128 0x179Au, /* Khmer */ /* No Reph, Visual Repha */
131 static inline bool
132 is_ra (hb_codepoint_t u)
134 for (unsigned int i = 0; i < ARRAY_LENGTH (ra_chars); i++)
135 if (u == ra_chars[i])
136 return true;
137 return false;
140 static inline bool
141 is_one_of (const hb_glyph_info_t &info, unsigned int flags)
143 /* If it ligated, all bets are off. */
144 if (_hb_glyph_info_ligated (&info)) return false;
145 return !!(FLAG (info.indic_category()) & flags);
148 static inline bool
149 is_joiner (const hb_glyph_info_t &info)
151 return is_one_of (info, JOINER_FLAGS);
154 static inline bool
155 is_consonant (const hb_glyph_info_t &info)
157 return is_one_of (info, CONSONANT_FLAGS);
160 static inline bool
161 is_halant_or_coeng (const hb_glyph_info_t &info)
163 return is_one_of (info, HALANT_OR_COENG_FLAGS);
166 static inline void
167 set_indic_properties (hb_glyph_info_t &info)
169 hb_codepoint_t u = info.codepoint;
170 unsigned int type = hb_indic_get_categories (u);
171 indic_category_t cat = (indic_category_t) (type & 0x7Fu);
172 indic_position_t pos = (indic_position_t) (type >> 8);
176 * Re-assign category
180 /* The spec says U+0952 is OT_A. However, testing shows that Uniscribe
181 * treats a whole bunch of characters similarly.
182 * TESTS: For example, for U+0951:
183 * U+092E,U+0947,U+0952
184 * U+092E,U+0952,U+0947
185 * U+092E,U+0947,U+0951
186 * U+092E,U+0951,U+0947
187 * U+092E,U+0951,U+0952
188 * U+092E,U+0952,U+0951
190 if (unlikely (hb_in_ranges (u, 0x0951u, 0x0952u,
191 0x1CD0u, 0x1CD2u,
192 0x1CD4u, 0x1CE1u) ||
193 u == 0x1CF4u))
194 cat = OT_A;
195 /* The following act more like the Bindus. */
196 else if (unlikely (hb_in_range (u, 0x0953u, 0x0954u)))
197 cat = OT_SM;
198 /* The following act like consonants. */
199 else if (unlikely (hb_in_ranges (u, 0x0A72u, 0x0A73u,
200 0x1CF5u, 0x1CF6u)))
201 cat = OT_C;
202 /* TODO: The following should only be allowed after a Visarga.
203 * For now, just treat them like regular tone marks. */
204 else if (unlikely (hb_in_range (u, 0x1CE2u, 0x1CE8u)))
205 cat = OT_A;
206 /* TODO: The following should only be allowed after some of
207 * the nasalization marks, maybe only for U+1CE9..U+1CF1.
208 * For now, just treat them like tone marks. */
209 else if (unlikely (u == 0x1CEDu))
210 cat = OT_A;
211 /* The following take marks in standalone clusters, similar to Avagraha. */
212 else if (unlikely (hb_in_ranges (u, 0xA8F2u, 0xA8F7u,
213 0x1CE9u, 0x1CECu,
214 0x1CEEu, 0x1CF1u)))
216 cat = OT_Symbol;
217 ASSERT_STATIC ((int) INDIC_SYLLABIC_CATEGORY_AVAGRAHA == OT_Symbol);
219 else if (unlikely (hb_in_range (u, 0x17CDu, 0x17D1u) ||
220 u == 0x17CBu || u == 0x17D3u || u == 0x17DDu)) /* Khmer Various signs */
222 /* These are like Top Matras. */
223 cat = OT_M;
224 pos = POS_ABOVE_C;
226 else if (unlikely (u == 0x17C6u)) cat = OT_N; /* Khmer Bindu doesn't like to be repositioned. */
227 else if (unlikely (u == 0x17D2u)) cat = OT_Coeng; /* Khmer coeng */
228 else if (unlikely (hb_in_range (u, 0x2010u, 0x2011u)))
229 cat = OT_PLACEHOLDER;
230 else if (unlikely (u == 0x25CCu)) cat = OT_DOTTEDCIRCLE;
231 else if (unlikely (u == 0xA982u)) cat = OT_SM; /* Javanese repha. */
232 else if (unlikely (u == 0xA9BEu)) cat = OT_CM2; /* Javanese medial ya. */
233 else if (unlikely (u == 0xA9BDu)) { cat = OT_M; pos = POS_POST_C; } /* Javanese vocalic r. */
237 * Re-assign position.
240 if ((FLAG (cat) & CONSONANT_FLAGS))
242 pos = POS_BASE_C;
243 if (is_ra (u))
244 cat = OT_Ra;
246 else if (cat == OT_M)
248 pos = matra_position (u, pos);
250 else if ((FLAG (cat) & (FLAG (OT_SM) | FLAG (OT_VD) | FLAG (OT_A) | FLAG (OT_Symbol))))
252 pos = POS_SMVD;
255 if (unlikely (u == 0x0B01u)) pos = POS_BEFORE_SUB; /* Oriya Bindu is BeforeSub in the spec. */
259 info.indic_category() = cat;
260 info.indic_position() = pos;
264 * Things above this line should ideally be moved to the Indic table itself.
269 * Indic configurations. Note that we do not want to keep every single script-specific
270 * behavior in these tables necessarily. This should mainly be used for per-script
271 * properties that are cheaper keeping here, than in the code. Ie. if, say, one and
272 * only one script has an exception, that one script can be if'ed directly in the code,
273 * instead of adding a new flag in these structs.
276 enum base_position_t {
277 BASE_POS_FIRST,
278 BASE_POS_LAST_SINHALA,
279 BASE_POS_LAST
281 enum reph_position_t {
282 REPH_POS_AFTER_MAIN = POS_AFTER_MAIN,
283 REPH_POS_BEFORE_SUB = POS_BEFORE_SUB,
284 REPH_POS_AFTER_SUB = POS_AFTER_SUB,
285 REPH_POS_BEFORE_POST = POS_BEFORE_POST,
286 REPH_POS_AFTER_POST = POS_AFTER_POST,
287 REPH_POS_DONT_CARE = POS_RA_TO_BECOME_REPH
289 enum reph_mode_t {
290 REPH_MODE_IMPLICIT, /* Reph formed out of initial Ra,H sequence. */
291 REPH_MODE_EXPLICIT, /* Reph formed out of initial Ra,H,ZWJ sequence. */
292 REPH_MODE_VIS_REPHA, /* Encoded Repha character, no reordering needed. */
293 REPH_MODE_LOG_REPHA /* Encoded Repha character, needs reordering. */
295 enum blwf_mode_t {
296 BLWF_MODE_PRE_AND_POST, /* Below-forms feature applied to pre-base and post-base. */
297 BLWF_MODE_POST_ONLY /* Below-forms feature applied to post-base only. */
299 enum pref_len_t {
300 PREF_LEN_1 = 1,
301 PREF_LEN_2 = 2,
302 PREF_LEN_DONT_CARE = PREF_LEN_2
304 struct indic_config_t
306 hb_script_t script;
307 bool has_old_spec;
308 hb_codepoint_t virama;
309 base_position_t base_pos;
310 reph_position_t reph_pos;
311 reph_mode_t reph_mode;
312 blwf_mode_t blwf_mode;
313 pref_len_t pref_len;
316 static const indic_config_t indic_configs[] =
318 /* Default. Should be first. */
319 {HB_SCRIPT_INVALID, false, 0,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_1},
320 {HB_SCRIPT_DEVANAGARI,true, 0x094Du,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
321 {HB_SCRIPT_BENGALI, true, 0x09CDu,BASE_POS_LAST, REPH_POS_AFTER_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
322 {HB_SCRIPT_GURMUKHI, true, 0x0A4Du,BASE_POS_LAST, REPH_POS_BEFORE_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
323 {HB_SCRIPT_GUJARATI, true, 0x0ACDu,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
324 {HB_SCRIPT_ORIYA, true, 0x0B4Du,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
325 {HB_SCRIPT_TAMIL, true, 0x0BCDu,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_2},
326 {HB_SCRIPT_TELUGU, true, 0x0C4Du,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_EXPLICIT, BLWF_MODE_POST_ONLY, PREF_LEN_2},
327 {HB_SCRIPT_KANNADA, true, 0x0CCDu,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_POST_ONLY, PREF_LEN_2},
328 {HB_SCRIPT_MALAYALAM, true, 0x0D4Du,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_LOG_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_2},
329 {HB_SCRIPT_SINHALA, false,0x0DCAu,BASE_POS_LAST_SINHALA,
330 REPH_POS_AFTER_MAIN, REPH_MODE_EXPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
331 {HB_SCRIPT_KHMER, false,0x17D2u,BASE_POS_FIRST,REPH_POS_DONT_CARE, REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_2},
332 {HB_SCRIPT_JAVANESE, false,0xA9C0u,BASE_POS_FIRST,REPH_POS_DONT_CARE, REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_1},
338 * Indic shaper.
341 struct feature_list_t {
342 hb_tag_t tag;
343 hb_ot_map_feature_flags_t flags;
346 static const feature_list_t
347 indic_features[] =
350 * Basic features.
351 * These features are applied in order, one at a time, after initial_reordering.
353 {HB_TAG('n','u','k','t'), F_GLOBAL},
354 {HB_TAG('a','k','h','n'), F_GLOBAL},
355 {HB_TAG('r','p','h','f'), F_NONE},
356 {HB_TAG('r','k','r','f'), F_GLOBAL},
357 {HB_TAG('p','r','e','f'), F_NONE},
358 {HB_TAG('b','l','w','f'), F_NONE},
359 {HB_TAG('a','b','v','f'), F_NONE},
360 {HB_TAG('h','a','l','f'), F_NONE},
361 {HB_TAG('p','s','t','f'), F_NONE},
362 {HB_TAG('v','a','t','u'), F_GLOBAL},
363 {HB_TAG('c','j','c','t'), F_GLOBAL},
364 {HB_TAG('c','f','a','r'), F_NONE},
366 * Other features.
367 * These features are applied all at once, after final_reordering.
368 * Default Bengali font in Windows for example has intermixed
369 * lookups for init,pres,abvs,blws features.
371 {HB_TAG('i','n','i','t'), F_NONE},
372 {HB_TAG('p','r','e','s'), F_GLOBAL},
373 {HB_TAG('a','b','v','s'), F_GLOBAL},
374 {HB_TAG('b','l','w','s'), F_GLOBAL},
375 {HB_TAG('p','s','t','s'), F_GLOBAL},
376 {HB_TAG('h','a','l','n'), F_GLOBAL},
377 /* Positioning features, though we don't care about the types. */
378 {HB_TAG('d','i','s','t'), F_GLOBAL},
379 {HB_TAG('a','b','v','m'), F_GLOBAL},
380 {HB_TAG('b','l','w','m'), F_GLOBAL},
384 * Must be in the same order as the indic_features array.
386 enum {
387 _NUKT,
388 _AKHN,
389 RPHF,
390 _RKRF,
391 PREF,
392 BLWF,
393 ABVF,
394 HALF,
395 PSTF,
396 _VATU,
397 _CJCT,
398 CFAR,
400 INIT,
401 _PRES,
402 _ABVS,
403 _BLWS,
404 _PSTS,
405 _HALN,
406 _DIST,
407 _ABVM,
408 _BLWM,
410 INDIC_NUM_FEATURES,
411 INDIC_BASIC_FEATURES = INIT /* Don't forget to update this! */
414 static void
415 setup_syllables (const hb_ot_shape_plan_t *plan,
416 hb_font_t *font,
417 hb_buffer_t *buffer);
418 static void
419 initial_reordering (const hb_ot_shape_plan_t *plan,
420 hb_font_t *font,
421 hb_buffer_t *buffer);
422 static void
423 final_reordering (const hb_ot_shape_plan_t *plan,
424 hb_font_t *font,
425 hb_buffer_t *buffer);
426 static void
427 clear_syllables (const hb_ot_shape_plan_t *plan,
428 hb_font_t *font,
429 hb_buffer_t *buffer);
431 static void
432 collect_features_indic (hb_ot_shape_planner_t *plan)
434 hb_ot_map_builder_t *map = &plan->map;
436 /* Do this before any lookups have been applied. */
437 map->add_gsub_pause (setup_syllables);
439 map->add_global_bool_feature (HB_TAG('l','o','c','l'));
440 /* The Indic specs do not require ccmp, but we apply it here since if
441 * there is a use of it, it's typically at the beginning. */
442 map->add_global_bool_feature (HB_TAG('c','c','m','p'));
445 unsigned int i = 0;
446 map->add_gsub_pause (initial_reordering);
447 for (; i < INDIC_BASIC_FEATURES; i++) {
448 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ);
449 map->add_gsub_pause (NULL);
451 map->add_gsub_pause (final_reordering);
452 for (; i < INDIC_NUM_FEATURES; i++) {
453 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ);
456 map->add_global_bool_feature (HB_TAG('c','a','l','t'));
457 map->add_global_bool_feature (HB_TAG('c','l','i','g'));
459 map->add_gsub_pause (clear_syllables);
462 static void
463 override_features_indic (hb_ot_shape_planner_t *plan)
465 /* Uniscribe does not apply 'kern' in Khmer. */
466 if (hb_options ().uniscribe_bug_compatible)
468 switch ((hb_tag_t) plan->props.script)
470 case HB_SCRIPT_KHMER:
471 plan->map.add_feature (HB_TAG('k','e','r','n'), 0, F_GLOBAL);
472 break;
476 plan->map.add_feature (HB_TAG('l','i','g','a'), 0, F_GLOBAL);
480 struct would_substitute_feature_t
482 inline void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_)
484 zero_context = zero_context_;
485 map->get_stage_lookups (0/*GSUB*/,
486 map->get_feature_stage (0/*GSUB*/, feature_tag),
487 &lookups, &count);
490 inline bool would_substitute (const hb_codepoint_t *glyphs,
491 unsigned int glyphs_count,
492 hb_face_t *face) const
494 for (unsigned int i = 0; i < count; i++)
495 if (hb_ot_layout_lookup_would_substitute_fast (face, lookups[i].index, glyphs, glyphs_count, zero_context))
496 return true;
497 return false;
500 private:
501 const hb_ot_map_t::lookup_map_t *lookups;
502 unsigned int count;
503 bool zero_context;
506 struct indic_shape_plan_t
508 ASSERT_POD ();
510 inline bool get_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const
512 hb_codepoint_t glyph = virama_glyph;
513 if (unlikely (virama_glyph == (hb_codepoint_t) -1))
515 if (!config->virama || !font->get_glyph (config->virama, 0, &glyph))
516 glyph = 0;
517 /* Technically speaking, the spec says we should apply 'locl' to virama too.
518 * Maybe one day... */
520 /* Our get_glyph() function needs a font, so we can't get the virama glyph
521 * during shape planning... Instead, overwrite it here. It's safe. Don't worry! */
522 (const_cast<indic_shape_plan_t *> (this))->virama_glyph = glyph;
525 *pglyph = glyph;
526 return glyph != 0;
529 const indic_config_t *config;
531 bool is_old_spec;
532 hb_codepoint_t virama_glyph;
534 would_substitute_feature_t rphf;
535 would_substitute_feature_t pref;
536 would_substitute_feature_t blwf;
537 would_substitute_feature_t pstf;
539 hb_mask_t mask_array[INDIC_NUM_FEATURES];
542 static void *
543 data_create_indic (const hb_ot_shape_plan_t *plan)
545 indic_shape_plan_t *indic_plan = (indic_shape_plan_t *) calloc (1, sizeof (indic_shape_plan_t));
546 if (unlikely (!indic_plan))
547 return NULL;
549 indic_plan->config = &indic_configs[0];
550 for (unsigned int i = 1; i < ARRAY_LENGTH (indic_configs); i++)
551 if (plan->props.script == indic_configs[i].script) {
552 indic_plan->config = &indic_configs[i];
553 break;
556 indic_plan->is_old_spec = indic_plan->config->has_old_spec && ((plan->map.chosen_script[0] & 0x000000FFu) != '2');
557 indic_plan->virama_glyph = (hb_codepoint_t) -1;
559 /* Use zero-context would_substitute() matching for new-spec of the main
560 * Indic scripts, and scripts with one spec only, but not for old-specs. */
561 bool zero_context = !indic_plan->is_old_spec;
562 indic_plan->rphf.init (&plan->map, HB_TAG('r','p','h','f'), zero_context);
563 indic_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), zero_context);
564 indic_plan->blwf.init (&plan->map, HB_TAG('b','l','w','f'), zero_context);
565 indic_plan->pstf.init (&plan->map, HB_TAG('p','s','t','f'), zero_context);
567 for (unsigned int i = 0; i < ARRAY_LENGTH (indic_plan->mask_array); i++)
568 indic_plan->mask_array[i] = (indic_features[i].flags & F_GLOBAL) ?
569 0 : plan->map.get_1_mask (indic_features[i].tag);
571 return indic_plan;
574 static void
575 data_destroy_indic (void *data)
577 free (data);
580 static indic_position_t
581 consonant_position_from_face (const indic_shape_plan_t *indic_plan,
582 const hb_codepoint_t consonant,
583 const hb_codepoint_t virama,
584 hb_face_t *face)
586 /* For old-spec, the order of glyphs is Consonant,Virama,
587 * whereas for new-spec, it's Virama,Consonant. However,
588 * some broken fonts (like Free Sans) simply copied lookups
589 * from old-spec to new-spec without modification.
590 * And oddly enough, Uniscribe seems to respect those lookups.
591 * Eg. in the sequence U+0924,U+094D,U+0930, Uniscribe finds
592 * base at 0. The font however, only has lookups matching
593 * 930,94D in 'blwf', not the expected 94D,930 (with new-spec
594 * table). As such, we simply match both sequences. Seems
595 * to work. */
596 hb_codepoint_t glyphs[3] = {virama, consonant, virama};
597 if (indic_plan->blwf.would_substitute (glyphs , 2, face) ||
598 indic_plan->blwf.would_substitute (glyphs+1, 2, face))
599 return POS_BELOW_C;
600 if (indic_plan->pstf.would_substitute (glyphs , 2, face) ||
601 indic_plan->pstf.would_substitute (glyphs+1, 2, face))
602 return POS_POST_C;
603 unsigned int pref_len = indic_plan->config->pref_len;
604 if ((pref_len == PREF_LEN_2 &&
605 (indic_plan->pref.would_substitute (glyphs , 2, face) ||
606 indic_plan->pref.would_substitute (glyphs+1, 2, face)))
607 || (pref_len == PREF_LEN_1 &&
608 indic_plan->pref.would_substitute (glyphs+1, 1, face)))
609 return POS_POST_C;
610 return POS_BASE_C;
614 enum syllable_type_t {
615 consonant_syllable,
616 vowel_syllable,
617 standalone_cluster,
618 symbol_cluster,
619 broken_cluster,
620 non_indic_cluster,
623 #include "hb-ot-shape-complex-indic-machine.hh"
626 static void
627 setup_masks_indic (const hb_ot_shape_plan_t *plan HB_UNUSED,
628 hb_buffer_t *buffer,
629 hb_font_t *font HB_UNUSED)
631 HB_BUFFER_ALLOCATE_VAR (buffer, indic_category);
632 HB_BUFFER_ALLOCATE_VAR (buffer, indic_position);
634 /* We cannot setup masks here. We save information about characters
635 * and setup masks later on in a pause-callback. */
637 unsigned int count = buffer->len;
638 hb_glyph_info_t *info = buffer->info;
639 for (unsigned int i = 0; i < count; i++)
640 set_indic_properties (info[i]);
643 static void
644 setup_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
645 hb_font_t *font HB_UNUSED,
646 hb_buffer_t *buffer)
648 find_syllables (buffer);
651 static int
652 compare_indic_order (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb)
654 int a = pa->indic_position();
655 int b = pb->indic_position();
657 return a < b ? -1 : a == b ? 0 : +1;
662 static void
663 update_consonant_positions (const hb_ot_shape_plan_t *plan,
664 hb_font_t *font,
665 hb_buffer_t *buffer)
667 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
669 if (indic_plan->config->base_pos != BASE_POS_LAST)
670 return;
672 hb_codepoint_t virama;
673 if (indic_plan->get_virama_glyph (font, &virama))
675 hb_face_t *face = font->face;
676 unsigned int count = buffer->len;
677 hb_glyph_info_t *info = buffer->info;
678 for (unsigned int i = 0; i < count; i++)
679 if (info[i].indic_position() == POS_BASE_C)
681 hb_codepoint_t consonant = info[i].codepoint;
682 info[i].indic_position() = consonant_position_from_face (indic_plan, consonant, virama, face);
688 /* Rules from:
689 * https://www.microsoft.com/typography/otfntdev/devanot/shaping.aspx */
691 static void
692 initial_reordering_consonant_syllable (const hb_ot_shape_plan_t *plan,
693 hb_face_t *face,
694 hb_buffer_t *buffer,
695 unsigned int start, unsigned int end)
697 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
698 hb_glyph_info_t *info = buffer->info;
701 /* 1. Find base consonant:
703 * The shaping engine finds the base consonant of the syllable, using the
704 * following algorithm: starting from the end of the syllable, move backwards
705 * until a consonant is found that does not have a below-base or post-base
706 * form (post-base forms have to follow below-base forms), or that is not a
707 * pre-base reordering Ra, or arrive at the first consonant. The consonant
708 * stopped at will be the base.
710 * o If the syllable starts with Ra + Halant (in a script that has Reph)
711 * and has more than one consonant, Ra is excluded from candidates for
712 * base consonants.
715 unsigned int base = end;
716 bool has_reph = false;
719 /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
720 * and has more than one consonant, Ra is excluded from candidates for
721 * base consonants. */
722 unsigned int limit = start;
723 if (indic_plan->config->reph_pos != REPH_POS_DONT_CARE &&
724 indic_plan->mask_array[RPHF] &&
725 start + 3 <= end &&
727 (indic_plan->config->reph_mode == REPH_MODE_IMPLICIT && !is_joiner (info[start + 2])) ||
728 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT && info[start + 2].indic_category() == OT_ZWJ)
731 /* See if it matches the 'rphf' feature. */
732 hb_codepoint_t glyphs[3] = {info[start].codepoint,
733 info[start + 1].codepoint,
734 indic_plan->config->reph_mode == REPH_MODE_EXPLICIT ?
735 info[start + 2].codepoint : 0};
736 if (indic_plan->rphf.would_substitute (glyphs, 2, face) ||
737 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT &&
738 indic_plan->rphf.would_substitute (glyphs, 3, face)))
740 limit += 2;
741 while (limit < end && is_joiner (info[limit]))
742 limit++;
743 base = start;
744 has_reph = true;
746 } else if (indic_plan->config->reph_mode == REPH_MODE_LOG_REPHA && info[start].indic_category() == OT_Repha)
748 limit += 1;
749 while (limit < end && is_joiner (info[limit]))
750 limit++;
751 base = start;
752 has_reph = true;
755 switch (indic_plan->config->base_pos)
757 default:
758 assert (false);
759 /* fallthrough */
761 case BASE_POS_LAST:
763 /* -> starting from the end of the syllable, move backwards */
764 unsigned int i = end;
765 bool seen_below = false;
766 do {
767 i--;
768 /* -> until a consonant is found */
769 if (is_consonant (info[i]))
771 /* -> that does not have a below-base or post-base form
772 * (post-base forms have to follow below-base forms), */
773 if (info[i].indic_position() != POS_BELOW_C &&
774 (info[i].indic_position() != POS_POST_C || seen_below))
776 base = i;
777 break;
779 if (info[i].indic_position() == POS_BELOW_C)
780 seen_below = true;
782 /* -> or that is not a pre-base reordering Ra,
784 * IMPLEMENTATION NOTES:
786 * Our pre-base reordering Ra's are marked POS_POST_C, so will be skipped
787 * by the logic above already.
790 /* -> or arrive at the first consonant. The consonant stopped at will
791 * be the base. */
792 base = i;
794 else
796 /* A ZWJ after a Halant stops the base search, and requests an explicit
797 * half form.
798 * A ZWJ before a Halant, requests a subjoined form instead, and hence
799 * search continues. This is particularly important for Bengali
800 * sequence Ra,H,Ya that should form Ya-Phalaa by subjoining Ya. */
801 if (start < i &&
802 info[i].indic_category() == OT_ZWJ &&
803 info[i - 1].indic_category() == OT_H)
804 break;
806 } while (i > limit);
808 break;
810 case BASE_POS_LAST_SINHALA:
812 /* Sinhala base positioning is slightly different from main Indic, in that:
813 * 1. Its ZWJ behavior is different,
814 * 2. We don't need to look into the font for consonant positions.
817 if (!has_reph)
818 base = limit;
820 /* Find the last base consonant that is not blocked by ZWJ. If there is
821 * a ZWJ right before a base consonant, that would request a subjoined form. */
822 for (unsigned int i = limit; i < end; i++)
823 if (is_consonant (info[i]))
825 if (limit < i && info[i - 1].indic_category() == OT_ZWJ)
826 break;
827 else
828 base = i;
831 /* Mark all subsequent consonants as below. */
832 for (unsigned int i = base + 1; i < end; i++)
833 if (is_consonant (info[i]))
834 info[i].indic_position() = POS_BELOW_C;
836 break;
838 case BASE_POS_FIRST:
840 /* The first consonant is always the base. */
842 assert (indic_plan->config->reph_mode == REPH_MODE_VIS_REPHA);
843 assert (!has_reph);
845 base = start;
847 /* Mark all subsequent consonants as below. */
848 for (unsigned int i = base + 1; i < end; i++)
849 if (is_consonant (info[i]))
850 info[i].indic_position() = POS_BELOW_C;
852 break;
855 /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
856 * and has more than one consonant, Ra is excluded from candidates for
857 * base consonants.
859 * Only do this for unforced Reph. (ie. not for Ra,H,ZWJ. */
860 if (has_reph && base == start && limit - base <= 2) {
861 /* Have no other consonant, so Reph is not formed and Ra becomes base. */
862 has_reph = false;
867 /* 2. Decompose and reorder Matras:
869 * Each matra and any syllable modifier sign in the cluster are moved to the
870 * appropriate position relative to the consonant(s) in the cluster. The
871 * shaping engine decomposes two- or three-part matras into their constituent
872 * parts before any repositioning. Matra characters are classified by which
873 * consonant in a conjunct they have affinity for and are reordered to the
874 * following positions:
876 * o Before first half form in the syllable
877 * o After subjoined consonants
878 * o After post-form consonant
879 * o After main consonant (for above marks)
881 * IMPLEMENTATION NOTES:
883 * The normalize() routine has already decomposed matras for us, so we don't
884 * need to worry about that.
888 /* 3. Reorder marks to canonical order:
890 * Adjacent nukta and halant or nukta and vedic sign are always repositioned
891 * if necessary, so that the nukta is first.
893 * IMPLEMENTATION NOTES:
895 * We don't need to do this: the normalize() routine already did this for us.
899 /* Reorder characters */
901 for (unsigned int i = start; i < base; i++)
902 info[i].indic_position() = MIN (POS_PRE_C, (indic_position_t) info[i].indic_position());
904 if (base < end)
905 info[base].indic_position() = POS_BASE_C;
907 /* Mark final consonants. A final consonant is one appearing after a matra,
908 * like in Khmer. */
909 for (unsigned int i = base + 1; i < end; i++)
910 if (info[i].indic_category() == OT_M) {
911 for (unsigned int j = i + 1; j < end; j++)
912 if (is_consonant (info[j])) {
913 info[j].indic_position() = POS_FINAL_C;
914 break;
916 break;
919 /* Handle beginning Ra */
920 if (has_reph)
921 info[start].indic_position() = POS_RA_TO_BECOME_REPH;
923 /* For old-style Indic script tags, move the first post-base Halant after
924 * last consonant.
926 * Reports suggest that in some scripts Uniscribe does this only if there
927 * is *not* a Halant after last consonant already (eg. Kannada), while it
928 * does it unconditionally in other scripts (eg. Malayalam). We don't
929 * currently know about other scripts, so we single out Malayalam for now.
931 * Kannada test case:
932 * U+0C9A,U+0CCD,U+0C9A,U+0CCD
933 * With some versions of Lohit Kannada.
934 * https://bugs.freedesktop.org/show_bug.cgi?id=59118
936 * Malayalam test case:
937 * U+0D38,U+0D4D,U+0D31,U+0D4D,U+0D31,U+0D4D
938 * With lohit-ttf-20121122/Lohit-Malayalam.ttf
940 if (indic_plan->is_old_spec)
942 bool disallow_double_halants = buffer->props.script != HB_SCRIPT_MALAYALAM;
943 for (unsigned int i = base + 1; i < end; i++)
944 if (info[i].indic_category() == OT_H)
946 unsigned int j;
947 for (j = end - 1; j > i; j--)
948 if (is_consonant (info[j]) ||
949 (disallow_double_halants && info[j].indic_category() == OT_H))
950 break;
951 if (info[j].indic_category() != OT_H && j > i) {
952 /* Move Halant to after last consonant. */
953 hb_glyph_info_t t = info[i];
954 memmove (&info[i], &info[i + 1], (j - i) * sizeof (info[0]));
955 info[j] = t;
957 break;
961 /* Attach misc marks to previous char to move with them. */
963 indic_position_t last_pos = POS_START;
964 for (unsigned int i = start; i < end; i++)
966 if ((FLAG (info[i].indic_category()) & (JOINER_FLAGS | FLAG (OT_N) | FLAG (OT_RS) | MEDIAL_FLAGS | HALANT_OR_COENG_FLAGS)))
968 info[i].indic_position() = last_pos;
969 if (unlikely (info[i].indic_category() == OT_H &&
970 info[i].indic_position() == POS_PRE_M))
973 * Uniscribe doesn't move the Halant with Left Matra.
974 * TEST: U+092B,U+093F,U+094DE
975 * We follow. This is important for the Sinhala
976 * U+0DDA split matra since it decomposes to U+0DD9,U+0DCA
977 * where U+0DD9 is a left matra and U+0DCA is the virama.
978 * We don't want to move the virama with the left matra.
979 * TEST: U+0D9A,U+0DDA
981 for (unsigned int j = i; j > start; j--)
982 if (info[j - 1].indic_position() != POS_PRE_M) {
983 info[i].indic_position() = info[j - 1].indic_position();
984 break;
987 } else if (info[i].indic_position() != POS_SMVD) {
988 last_pos = (indic_position_t) info[i].indic_position();
992 /* For post-base consonants let them own anything before them
993 * since the last consonant or matra. */
995 unsigned int last = base;
996 for (unsigned int i = base + 1; i < end; i++)
997 if (is_consonant (info[i]))
999 for (unsigned int j = last + 1; j < i; j++)
1000 if (info[j].indic_position() < POS_SMVD)
1001 info[j].indic_position() = info[i].indic_position();
1002 last = i;
1003 } else if (info[i].indic_category() == OT_M)
1004 last = i;
1009 /* Use syllable() for sort accounting temporarily. */
1010 unsigned int syllable = info[start].syllable();
1011 for (unsigned int i = start; i < end; i++)
1012 info[i].syllable() = i - start;
1014 /* Sit tight, rock 'n roll! */
1015 hb_bubble_sort (info + start, end - start, compare_indic_order);
1016 /* Find base again */
1017 base = end;
1018 for (unsigned int i = start; i < end; i++)
1019 if (info[i].indic_position() == POS_BASE_C)
1021 base = i;
1022 break;
1024 /* Things are out-of-control for post base positions, they may shuffle
1025 * around like crazy. In old-spec mode, we move halants around, so in
1026 * that case merge all clusters after base. Otherwise, check the sort
1027 * order and merge as needed.
1028 * For pre-base stuff, we handle cluster issues in final reordering. */
1029 if (indic_plan->is_old_spec || end - base > 127)
1030 buffer->merge_clusters (base, end);
1031 else
1033 /* Note! syllable() is a one-byte field. */
1034 for (unsigned int i = base; i < end; i++)
1035 if (info[i].syllable() != 255)
1037 unsigned int max = i;
1038 unsigned int j = start + info[i].syllable();
1039 while (j != i)
1041 max = MAX (max, j);
1042 unsigned int next = start + info[j].syllable();
1043 info[j].syllable() = 255; /* So we don't process j later again. */
1044 j = next;
1046 if (i != max)
1047 buffer->merge_clusters (i, max + 1);
1051 /* Put syllable back in. */
1052 for (unsigned int i = start; i < end; i++)
1053 info[i].syllable() = syllable;
1056 /* Setup masks now */
1059 hb_mask_t mask;
1061 /* Reph */
1062 for (unsigned int i = start; i < end && info[i].indic_position() == POS_RA_TO_BECOME_REPH; i++)
1063 info[i].mask |= indic_plan->mask_array[RPHF];
1065 /* Pre-base */
1066 mask = indic_plan->mask_array[HALF];
1067 if (!indic_plan->is_old_spec &&
1068 indic_plan->config->blwf_mode == BLWF_MODE_PRE_AND_POST)
1069 mask |= indic_plan->mask_array[BLWF];
1070 for (unsigned int i = start; i < base; i++)
1071 info[i].mask |= mask;
1072 /* Base */
1073 mask = 0;
1074 if (base < end)
1075 info[base].mask |= mask;
1076 /* Post-base */
1077 mask = indic_plan->mask_array[BLWF] | indic_plan->mask_array[ABVF] | indic_plan->mask_array[PSTF];
1078 for (unsigned int i = base + 1; i < end; i++)
1079 info[i].mask |= mask;
1082 if (indic_plan->is_old_spec &&
1083 buffer->props.script == HB_SCRIPT_DEVANAGARI)
1085 /* Old-spec eye-lash Ra needs special handling. From the
1086 * spec:
1088 * "The feature 'below-base form' is applied to consonants
1089 * having below-base forms and following the base consonant.
1090 * The exception is vattu, which may appear below half forms
1091 * as well as below the base glyph. The feature 'below-base
1092 * form' will be applied to all such occurrences of Ra as well."
1094 * Test case: U+0924,U+094D,U+0930,U+094d,U+0915
1095 * with Sanskrit 2003 font.
1097 * However, note that Ra,Halant,ZWJ is the correct way to
1098 * request eyelash form of Ra, so we wouldbn't inhibit it
1099 * in that sequence.
1101 * Test case: U+0924,U+094D,U+0930,U+094d,U+200D,U+0915
1103 for (unsigned int i = start; i + 1 < base; i++)
1104 if (info[i ].indic_category() == OT_Ra &&
1105 info[i+1].indic_category() == OT_H &&
1106 (i + 2 == base ||
1107 info[i+2].indic_category() != OT_ZWJ))
1109 info[i ].mask |= indic_plan->mask_array[BLWF];
1110 info[i+1].mask |= indic_plan->mask_array[BLWF];
1114 unsigned int pref_len = indic_plan->config->pref_len;
1115 if (indic_plan->mask_array[PREF] && base + pref_len < end)
1117 assert (1 <= pref_len && pref_len <= 2);
1118 /* Find a Halant,Ra sequence and mark it for pre-base reordering processing. */
1119 for (unsigned int i = base + 1; i + pref_len - 1 < end; i++) {
1120 hb_codepoint_t glyphs[2];
1121 for (unsigned int j = 0; j < pref_len; j++)
1122 glyphs[j] = info[i + j].codepoint;
1123 if (indic_plan->pref.would_substitute (glyphs, pref_len, face))
1125 for (unsigned int j = 0; j < pref_len; j++)
1126 info[i++].mask |= indic_plan->mask_array[PREF];
1128 /* Mark the subsequent stuff with 'cfar'. Used in Khmer.
1129 * Read the feature spec.
1130 * This allows distinguishing the following cases with MS Khmer fonts:
1131 * U+1784,U+17D2,U+179A,U+17D2,U+1782
1132 * U+1784,U+17D2,U+1782,U+17D2,U+179A
1134 if (indic_plan->mask_array[CFAR])
1135 for (; i < end; i++)
1136 info[i].mask |= indic_plan->mask_array[CFAR];
1138 break;
1143 /* Apply ZWJ/ZWNJ effects */
1144 for (unsigned int i = start + 1; i < end; i++)
1145 if (is_joiner (info[i])) {
1146 bool non_joiner = info[i].indic_category() == OT_ZWNJ;
1147 unsigned int j = i;
1149 do {
1150 j--;
1152 /* ZWJ/ZWNJ should disable CJCT. They do that by simply
1153 * being there, since we don't skip them for the CJCT
1154 * feature (ie. F_MANUAL_ZWJ) */
1156 /* A ZWNJ disables HALF. */
1157 if (non_joiner)
1158 info[j].mask &= ~indic_plan->mask_array[HALF];
1160 } while (j > start && !is_consonant (info[j]));
1165 static void
1166 initial_reordering_vowel_syllable (const hb_ot_shape_plan_t *plan,
1167 hb_face_t *face,
1168 hb_buffer_t *buffer,
1169 unsigned int start, unsigned int end)
1171 /* We made the vowels look like consonants. So let's call the consonant logic! */
1172 initial_reordering_consonant_syllable (plan, face, buffer, start, end);
1175 static void
1176 initial_reordering_standalone_cluster (const hb_ot_shape_plan_t *plan,
1177 hb_face_t *face,
1178 hb_buffer_t *buffer,
1179 unsigned int start, unsigned int end)
1181 /* We treat placeholder/dotted-circle as if they are consonants, so we
1182 * should just chain. Only if not in compatibility mode that is... */
1184 if (hb_options ().uniscribe_bug_compatible)
1186 /* For dotted-circle, this is what Uniscribe does:
1187 * If dotted-circle is the last glyph, it just does nothing.
1188 * Ie. It doesn't form Reph. */
1189 if (buffer->info[end - 1].indic_category() == OT_DOTTEDCIRCLE)
1190 return;
1193 initial_reordering_consonant_syllable (plan, face, buffer, start, end);
1196 static void
1197 initial_reordering_broken_cluster (const hb_ot_shape_plan_t *plan,
1198 hb_face_t *face,
1199 hb_buffer_t *buffer,
1200 unsigned int start, unsigned int end)
1202 /* We already inserted dotted-circles, so just call the standalone_cluster. */
1203 initial_reordering_standalone_cluster (plan, face, buffer, start, end);
1206 static void
1207 initial_reordering_symbol_cluster (const hb_ot_shape_plan_t *plan HB_UNUSED,
1208 hb_face_t *face HB_UNUSED,
1209 hb_buffer_t *buffer HB_UNUSED,
1210 unsigned int start HB_UNUSED, unsigned int end HB_UNUSED)
1212 /* Nothing to do right now. If we ever switch to using the output
1213 * buffer in the reordering process, we'd need to next_glyph() here. */
1216 static void
1217 initial_reordering_non_indic_cluster (const hb_ot_shape_plan_t *plan HB_UNUSED,
1218 hb_face_t *face HB_UNUSED,
1219 hb_buffer_t *buffer HB_UNUSED,
1220 unsigned int start HB_UNUSED, unsigned int end HB_UNUSED)
1222 /* Nothing to do right now. If we ever switch to using the output
1223 * buffer in the reordering process, we'd need to next_glyph() here. */
1227 static void
1228 initial_reordering_syllable (const hb_ot_shape_plan_t *plan,
1229 hb_face_t *face,
1230 hb_buffer_t *buffer,
1231 unsigned int start, unsigned int end)
1233 syllable_type_t syllable_type = (syllable_type_t) (buffer->info[start].syllable() & 0x0F);
1234 switch (syllable_type) {
1235 case consonant_syllable: initial_reordering_consonant_syllable (plan, face, buffer, start, end); return;
1236 case vowel_syllable: initial_reordering_vowel_syllable (plan, face, buffer, start, end); return;
1237 case standalone_cluster: initial_reordering_standalone_cluster (plan, face, buffer, start, end); return;
1238 case symbol_cluster: initial_reordering_symbol_cluster (plan, face, buffer, start, end); return;
1239 case broken_cluster: initial_reordering_broken_cluster (plan, face, buffer, start, end); return;
1240 case non_indic_cluster: initial_reordering_non_indic_cluster (plan, face, buffer, start, end); return;
1244 static inline void
1245 insert_dotted_circles (const hb_ot_shape_plan_t *plan HB_UNUSED,
1246 hb_font_t *font,
1247 hb_buffer_t *buffer)
1249 /* Note: This loop is extra overhead, but should not be measurable. */
1250 bool has_broken_syllables = false;
1251 unsigned int count = buffer->len;
1252 hb_glyph_info_t *info = buffer->info;
1253 for (unsigned int i = 0; i < count; i++)
1254 if ((info[i].syllable() & 0x0F) == broken_cluster)
1256 has_broken_syllables = true;
1257 break;
1259 if (likely (!has_broken_syllables))
1260 return;
1263 hb_codepoint_t dottedcircle_glyph;
1264 if (!font->get_glyph (0x25CCu, 0, &dottedcircle_glyph))
1265 return;
1267 hb_glyph_info_t dottedcircle = {0};
1268 dottedcircle.codepoint = 0x25CCu;
1269 set_indic_properties (dottedcircle);
1270 dottedcircle.codepoint = dottedcircle_glyph;
1272 buffer->clear_output ();
1274 buffer->idx = 0;
1275 unsigned int last_syllable = 0;
1276 while (buffer->idx < buffer->len)
1278 unsigned int syllable = buffer->cur().syllable();
1279 syllable_type_t syllable_type = (syllable_type_t) (syllable & 0x0F);
1280 if (unlikely (last_syllable != syllable && syllable_type == broken_cluster))
1282 last_syllable = syllable;
1284 hb_glyph_info_t info = dottedcircle;
1285 info.cluster = buffer->cur().cluster;
1286 info.mask = buffer->cur().mask;
1287 info.syllable() = buffer->cur().syllable();
1288 /* TODO Set glyph_props? */
1290 /* Insert dottedcircle after possible Repha. */
1291 while (buffer->idx < buffer->len &&
1292 last_syllable == buffer->cur().syllable() &&
1293 buffer->cur().indic_category() == OT_Repha)
1294 buffer->next_glyph ();
1296 buffer->output_info (info);
1298 else
1299 buffer->next_glyph ();
1302 buffer->swap_buffers ();
1305 static void
1306 initial_reordering (const hb_ot_shape_plan_t *plan,
1307 hb_font_t *font,
1308 hb_buffer_t *buffer)
1310 update_consonant_positions (plan, font, buffer);
1311 insert_dotted_circles (plan, font, buffer);
1313 hb_glyph_info_t *info = buffer->info;
1314 unsigned int count = buffer->len;
1315 if (unlikely (!count)) return;
1316 unsigned int last = 0;
1317 unsigned int last_syllable = info[0].syllable();
1318 for (unsigned int i = 1; i < count; i++)
1319 if (last_syllable != info[i].syllable()) {
1320 initial_reordering_syllable (plan, font->face, buffer, last, i);
1321 last = i;
1322 last_syllable = info[last].syllable();
1324 initial_reordering_syllable (plan, font->face, buffer, last, count);
1327 static void
1328 final_reordering_syllable (const hb_ot_shape_plan_t *plan,
1329 hb_buffer_t *buffer,
1330 unsigned int start, unsigned int end)
1332 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
1333 hb_glyph_info_t *info = buffer->info;
1336 /* This function relies heavily on halant glyphs. Lots of ligation
1337 * and possibly multiplication substitutions happened prior to this
1338 * phase, and that might have messed up our properties. Recover
1339 * from a particular case of that where we're fairly sure that a
1340 * class of OT_H is desired but has been lost. */
1341 if (indic_plan->virama_glyph)
1343 unsigned int virama_glyph = indic_plan->virama_glyph;
1344 for (unsigned int i = start; i < end; i++)
1345 if (info[i].codepoint == virama_glyph &&
1346 _hb_glyph_info_ligated (&info[i]) &&
1347 _hb_glyph_info_multiplied (&info[i]))
1349 /* This will make sure that this glyph passes is_halant_or_coeng() test. */
1350 info[i].indic_category() = OT_H;
1351 _hb_glyph_info_clear_ligated_and_multiplied (&info[i]);
1356 /* 4. Final reordering:
1358 * After the localized forms and basic shaping forms GSUB features have been
1359 * applied (see below), the shaping engine performs some final glyph
1360 * reordering before applying all the remaining font features to the entire
1361 * cluster.
1364 bool try_pref = !!indic_plan->mask_array[PREF];
1366 /* Find base again */
1367 unsigned int base;
1368 for (base = start; base < end; base++)
1369 if (info[base].indic_position() >= POS_BASE_C)
1371 if (try_pref && base + 1 < end && indic_plan->config->pref_len == 2)
1373 for (unsigned int i = base + 1; i < end; i++)
1374 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0)
1376 if (!(_hb_glyph_info_substituted (&info[i]) &&
1377 _hb_glyph_info_ligated_and_didnt_multiply (&info[i])))
1379 /* Ok, this was a 'pref' candidate but didn't form any.
1380 * Base is around here... */
1381 base = i;
1382 while (base < end && is_halant_or_coeng (info[base]))
1383 base++;
1384 info[base].indic_position() = POS_BASE_C;
1386 try_pref = false;
1388 break;
1392 if (start < base && info[base].indic_position() > POS_BASE_C)
1393 base--;
1394 break;
1396 if (base == end && start < base &&
1397 is_one_of (info[base - 1], FLAG (OT_ZWJ)))
1398 base--;
1399 if (base < end)
1400 while (start < base &&
1401 is_one_of (info[base], (FLAG (OT_N) | HALANT_OR_COENG_FLAGS)))
1402 base--;
1405 /* o Reorder matras:
1407 * If a pre-base matra character had been reordered before applying basic
1408 * features, the glyph can be moved closer to the main consonant based on
1409 * whether half-forms had been formed. Actual position for the matra is
1410 * defined as “after last standalone halant glyph, after initial matra
1411 * position and before the main consonant”. If ZWJ or ZWNJ follow this
1412 * halant, position is moved after it.
1415 if (start + 1 < end && start < base) /* Otherwise there can't be any pre-base matra characters. */
1417 /* If we lost track of base, alas, position before last thingy. */
1418 unsigned int new_pos = base == end ? base - 2 : base - 1;
1420 /* Malayalam / Tamil do not have "half" forms or explicit virama forms.
1421 * The glyphs formed by 'half' are Chillus or ligated explicit viramas.
1422 * We want to position matra after them.
1424 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL)
1426 while (new_pos > start &&
1427 !(is_one_of (info[new_pos], (FLAG (OT_M) | HALANT_OR_COENG_FLAGS))))
1428 new_pos--;
1430 /* If we found no Halant we are done.
1431 * Otherwise only proceed if the Halant does
1432 * not belong to the Matra itself! */
1433 if (is_halant_or_coeng (info[new_pos]) &&
1434 info[new_pos].indic_position() != POS_PRE_M)
1436 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */
1437 if (new_pos + 1 < end && is_joiner (info[new_pos + 1]))
1438 new_pos++;
1440 else
1441 new_pos = start; /* No move. */
1444 if (start < new_pos && info[new_pos].indic_position () != POS_PRE_M)
1446 /* Now go see if there's actually any matras... */
1447 for (unsigned int i = new_pos; i > start; i--)
1448 if (info[i - 1].indic_position () == POS_PRE_M)
1450 unsigned int old_pos = i - 1;
1451 hb_glyph_info_t tmp = info[old_pos];
1452 memmove (&info[old_pos], &info[old_pos + 1], (new_pos - old_pos) * sizeof (info[0]));
1453 info[new_pos] = tmp;
1454 if (old_pos < base && base <= new_pos) /* Shouldn't actually happen. */
1455 base--;
1456 buffer->merge_clusters (new_pos, MIN (end, base + 1));
1457 new_pos--;
1459 } else {
1460 for (unsigned int i = start; i < base; i++)
1461 if (info[i].indic_position () == POS_PRE_M) {
1462 buffer->merge_clusters (i, MIN (end, base + 1));
1463 break;
1469 /* o Reorder reph:
1471 * Reph’s original position is always at the beginning of the syllable,
1472 * (i.e. it is not reordered at the character reordering stage). However,
1473 * it will be reordered according to the basic-forms shaping results.
1474 * Possible positions for reph, depending on the script, are; after main,
1475 * before post-base consonant forms, and after post-base consonant forms.
1478 /* Two cases:
1480 * - If repha is encoded as a sequence of characters (Ra,H or Ra,H,ZWJ), then
1481 * we should only move it if the sequence ligated to the repha form.
1483 * - If repha is encoded separately and in the logical position, we should only
1484 * move it if it did NOT ligate. If it ligated, it's probably the font trying
1485 * to make it work without the reordering.
1487 if (start + 1 < end &&
1488 info[start].indic_position() == POS_RA_TO_BECOME_REPH &&
1489 ((info[start].indic_category() == OT_Repha) ^
1490 _hb_glyph_info_ligated_and_didnt_multiply (&info[start])))
1492 unsigned int new_reph_pos;
1493 reph_position_t reph_pos = indic_plan->config->reph_pos;
1495 assert (reph_pos != REPH_POS_DONT_CARE);
1497 /* 1. If reph should be positioned after post-base consonant forms,
1498 * proceed to step 5.
1500 if (reph_pos == REPH_POS_AFTER_POST)
1502 goto reph_step_5;
1505 /* 2. If the reph repositioning class is not after post-base: target
1506 * position is after the first explicit halant glyph between the
1507 * first post-reph consonant and last main consonant. If ZWJ or ZWNJ
1508 * are following this halant, position is moved after it. If such
1509 * position is found, this is the target position. Otherwise,
1510 * proceed to the next step.
1512 * Note: in old-implementation fonts, where classifications were
1513 * fixed in shaping engine, there was no case where reph position
1514 * will be found on this step.
1517 new_reph_pos = start + 1;
1518 while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos]))
1519 new_reph_pos++;
1521 if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos]))
1523 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */
1524 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1]))
1525 new_reph_pos++;
1526 goto reph_move;
1530 /* 3. If reph should be repositioned after the main consonant: find the
1531 * first consonant not ligated with main, or find the first
1532 * consonant that is not a potential pre-base reordering Ra.
1534 if (reph_pos == REPH_POS_AFTER_MAIN)
1536 new_reph_pos = base;
1537 while (new_reph_pos + 1 < end && info[new_reph_pos + 1].indic_position() <= POS_AFTER_MAIN)
1538 new_reph_pos++;
1539 if (new_reph_pos < end)
1540 goto reph_move;
1543 /* 4. If reph should be positioned before post-base consonant, find
1544 * first post-base classified consonant not ligated with main. If no
1545 * consonant is found, the target position should be before the
1546 * first matra, syllable modifier sign or vedic sign.
1548 /* This is our take on what step 4 is trying to say (and failing, BADLY). */
1549 if (reph_pos == REPH_POS_AFTER_SUB)
1551 new_reph_pos = base;
1552 while (new_reph_pos < end &&
1553 !( FLAG (info[new_reph_pos + 1].indic_position()) & (FLAG (POS_POST_C) | FLAG (POS_AFTER_POST) | FLAG (POS_SMVD))))
1554 new_reph_pos++;
1555 if (new_reph_pos < end)
1556 goto reph_move;
1559 /* 5. If no consonant is found in steps 3 or 4, move reph to a position
1560 * immediately before the first post-base matra, syllable modifier
1561 * sign or vedic sign that has a reordering class after the intended
1562 * reph position. For example, if the reordering position for reph
1563 * is post-main, it will skip above-base matras that also have a
1564 * post-main position.
1566 reph_step_5:
1568 /* Copied from step 2. */
1569 new_reph_pos = start + 1;
1570 while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos]))
1571 new_reph_pos++;
1573 if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos]))
1575 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */
1576 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1]))
1577 new_reph_pos++;
1578 goto reph_move;
1582 /* 6. Otherwise, reorder reph to the end of the syllable.
1585 new_reph_pos = end - 1;
1586 while (new_reph_pos > start && info[new_reph_pos].indic_position() == POS_SMVD)
1587 new_reph_pos--;
1590 * If the Reph is to be ending up after a Matra,Halant sequence,
1591 * position it before that Halant so it can interact with the Matra.
1592 * However, if it's a plain Consonant,Halant we shouldn't do that.
1593 * Uniscribe doesn't do this.
1594 * TEST: U+0930,U+094D,U+0915,U+094B,U+094D
1596 if (!hb_options ().uniscribe_bug_compatible &&
1597 unlikely (is_halant_or_coeng (info[new_reph_pos]))) {
1598 for (unsigned int i = base + 1; i < new_reph_pos; i++)
1599 if (info[i].indic_category() == OT_M) {
1600 /* Ok, got it. */
1601 new_reph_pos--;
1604 goto reph_move;
1607 reph_move:
1609 buffer->merge_clusters (start, new_reph_pos + 1);
1611 /* Move */
1612 hb_glyph_info_t reph = info[start];
1613 memmove (&info[start], &info[start + 1], (new_reph_pos - start) * sizeof (info[0]));
1614 info[new_reph_pos] = reph;
1615 if (start < base && base <= new_reph_pos)
1616 base--;
1621 /* o Reorder pre-base reordering consonants:
1623 * If a pre-base reordering consonant is found, reorder it according to
1624 * the following rules:
1627 if (try_pref && base + 1 < end) /* Otherwise there can't be any pre-base reordering Ra. */
1629 unsigned int pref_len = indic_plan->config->pref_len;
1630 for (unsigned int i = base + 1; i < end; i++)
1631 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0)
1633 /* 1. Only reorder a glyph produced by substitution during application
1634 * of the <pref> feature. (Note that a font may shape a Ra consonant with
1635 * the feature generally but block it in certain contexts.)
1637 /* Note: We just check that something got substituted. We don't check that
1638 * the <pref> feature actually did it...
1640 * If pref len is longer than one, then only reorder if it ligated. If
1641 * pref len is one, only reorder if it didn't ligate with other things. */
1642 if (_hb_glyph_info_substituted (&info[i]) &&
1643 ((pref_len == 1) ^ _hb_glyph_info_ligated_and_didnt_multiply (&info[i])))
1646 * 2. Try to find a target position the same way as for pre-base matra.
1647 * If it is found, reorder pre-base consonant glyph.
1649 * 3. If position is not found, reorder immediately before main
1650 * consonant.
1653 unsigned int new_pos = base;
1654 /* Malayalam / Tamil do not have "half" forms or explicit virama forms.
1655 * The glyphs formed by 'half' are Chillus or ligated explicit viramas.
1656 * We want to position matra after them.
1658 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL)
1660 while (new_pos > start &&
1661 !(is_one_of (info[new_pos - 1], FLAG(OT_M) | HALANT_OR_COENG_FLAGS)))
1662 new_pos--;
1664 /* In Khmer coeng model, a H,Ra can go *after* matras. If it goes after a
1665 * split matra, it should be reordered to *before* the left part of such matra. */
1666 if (new_pos > start && info[new_pos - 1].indic_category() == OT_M)
1668 unsigned int old_pos = i;
1669 for (unsigned int i = base + 1; i < old_pos; i++)
1670 if (info[i].indic_category() == OT_M)
1672 new_pos--;
1673 break;
1678 if (new_pos > start && is_halant_or_coeng (info[new_pos - 1]))
1680 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */
1681 if (new_pos < end && is_joiner (info[new_pos]))
1682 new_pos++;
1686 unsigned int old_pos = i;
1687 buffer->merge_clusters (new_pos, old_pos + 1);
1688 hb_glyph_info_t tmp = info[old_pos];
1689 memmove (&info[new_pos + 1], &info[new_pos], (old_pos - new_pos) * sizeof (info[0]));
1690 info[new_pos] = tmp;
1691 if (new_pos <= base && base < old_pos)
1692 base++;
1696 break;
1701 /* Apply 'init' to the Left Matra if it's a word start. */
1702 if (info[start].indic_position () == POS_PRE_M &&
1703 (!start ||
1704 !(FLAG (_hb_glyph_info_get_general_category (&info[start - 1])) &
1705 FLAG_RANGE (HB_UNICODE_GENERAL_CATEGORY_FORMAT, HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK))))
1706 info[start].mask |= indic_plan->mask_array[INIT];
1710 * Finish off the clusters and go home!
1712 if (hb_options ().uniscribe_bug_compatible)
1714 switch ((hb_tag_t) plan->props.script)
1716 case HB_SCRIPT_TAMIL:
1717 case HB_SCRIPT_SINHALA:
1718 break;
1720 default:
1721 /* Uniscribe merges the entire cluster... Except for Tamil & Sinhala.
1722 * This means, half forms are submerged into the main consonants cluster.
1723 * This is unnecessary, and makes cursor positioning harder, but that's what
1724 * Uniscribe does. */
1725 buffer->merge_clusters (start, end);
1726 break;
1732 static void
1733 final_reordering (const hb_ot_shape_plan_t *plan,
1734 hb_font_t *font HB_UNUSED,
1735 hb_buffer_t *buffer)
1737 unsigned int count = buffer->len;
1738 if (unlikely (!count)) return;
1740 hb_glyph_info_t *info = buffer->info;
1741 unsigned int last = 0;
1742 unsigned int last_syllable = info[0].syllable();
1743 for (unsigned int i = 1; i < count; i++)
1744 if (last_syllable != info[i].syllable()) {
1745 final_reordering_syllable (plan, buffer, last, i);
1746 last = i;
1747 last_syllable = info[last].syllable();
1749 final_reordering_syllable (plan, buffer, last, count);
1751 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_category);
1752 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_position);
1756 static void
1757 clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
1758 hb_font_t *font HB_UNUSED,
1759 hb_buffer_t *buffer)
1761 hb_glyph_info_t *info = buffer->info;
1762 unsigned int count = buffer->len;
1763 for (unsigned int i = 0; i < count; i++)
1764 info[i].syllable() = 0;
1768 static bool
1769 decompose_indic (const hb_ot_shape_normalize_context_t *c,
1770 hb_codepoint_t ab,
1771 hb_codepoint_t *a,
1772 hb_codepoint_t *b)
1774 switch (ab)
1776 /* Don't decompose these. */
1777 case 0x0931u : return false;
1778 case 0x0B94u : return false;
1782 * Decompose split matras that don't have Unicode decompositions.
1785 case 0x0F77u : *a = 0x0FB2u; *b= 0x0F81u; return true;
1786 case 0x0F79u : *a = 0x0FB3u; *b= 0x0F81u; return true;
1787 case 0x17BEu : *a = 0x17C1u; *b= 0x17BEu; return true;
1788 case 0x17BFu : *a = 0x17C1u; *b= 0x17BFu; return true;
1789 case 0x17C0u : *a = 0x17C1u; *b= 0x17C0u; return true;
1790 case 0x17C4u : *a = 0x17C1u; *b= 0x17C4u; return true;
1791 case 0x17C5u : *a = 0x17C1u; *b= 0x17C5u; return true;
1792 case 0x1925u : *a = 0x1920u; *b= 0x1923u; return true;
1793 case 0x1926u : *a = 0x1920u; *b= 0x1924u; return true;
1794 case 0x1B3Cu : *a = 0x1B42u; *b= 0x1B3Cu; return true;
1795 case 0x1112Eu : *a = 0x11127u; *b= 0x11131u; return true;
1796 case 0x1112Fu : *a = 0x11127u; *b= 0x11132u; return true;
1797 #if 0
1798 /* This one has no decomposition in Unicode, but needs no decomposition either. */
1799 /* case 0x0AC9u : return false; */
1800 case 0x0B57u : *a = no decomp, -> RIGHT; return true;
1801 case 0x1C29u : *a = no decomp, -> LEFT; return true;
1802 case 0xA9C0u : *a = no decomp, -> RIGHT; return true;
1803 case 0x111BuF : *a = no decomp, -> ABOVE; return true;
1804 #endif
1807 if ((ab == 0x0DDAu || hb_in_range (ab, 0x0DDCu, 0x0DDEu)))
1810 * Sinhala split matras... Let the fun begin.
1812 * These four characters have Unicode decompositions. However, Uniscribe
1813 * decomposes them "Khmer-style", that is, it uses the character itself to
1814 * get the second half. The first half of all four decompositions is always
1815 * U+0DD9.
1817 * Now, there are buggy fonts, namely, the widely used lklug.ttf, that are
1818 * broken with Uniscribe. But we need to support them. As such, we only
1819 * do the Uniscribe-style decomposition if the character is transformed into
1820 * its "sec.half" form by the 'pstf' feature. Otherwise, we fall back to
1821 * Unicode decomposition.
1823 * Note that we can't unconditionally use Unicode decomposition. That would
1824 * break some other fonts, that are designed to work with Uniscribe, and
1825 * don't have positioning features for the Unicode-style decomposition.
1827 * Argh...
1829 * The Uniscribe behavior is now documented in the newly published Sinhala
1830 * spec in 2012:
1832 * http://www.microsoft.com/typography/OpenTypeDev/sinhala/intro.htm#shaping
1835 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) c->plan->data;
1837 hb_codepoint_t glyph;
1839 if (hb_options ().uniscribe_bug_compatible ||
1840 (c->font->get_glyph (ab, 0, &glyph) &&
1841 indic_plan->pstf.would_substitute (&glyph, 1, c->font->face)))
1843 /* Ok, safe to use Uniscribe-style decomposition. */
1844 *a = 0x0DD9u;
1845 *b = ab;
1846 return true;
1850 return c->unicode->decompose (ab, a, b);
1853 static bool
1854 compose_indic (const hb_ot_shape_normalize_context_t *c,
1855 hb_codepoint_t a,
1856 hb_codepoint_t b,
1857 hb_codepoint_t *ab)
1859 /* Avoid recomposing split matras. */
1860 if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a)))
1861 return false;
1863 /* Composition-exclusion exceptions that we want to recompose. */
1864 if (a == 0x09AFu && b == 0x09BCu) { *ab = 0x09DFu; return true; }
1866 return c->unicode->compose (a, b, ab);
1870 const hb_ot_complex_shaper_t _hb_ot_complex_shaper_indic =
1872 "indic",
1873 collect_features_indic,
1874 override_features_indic,
1875 data_create_indic,
1876 data_destroy_indic,
1877 NULL, /* preprocess_text */
1878 HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT,
1879 decompose_indic,
1880 compose_indic,
1881 setup_masks_indic,
1882 HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE,
1883 false, /* fallback_position */