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
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 */
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)
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);
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 */
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 */
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])
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_SAFE (info.indic_category()) & flags);
149 is_joiner (const hb_glyph_info_t &info)
151 return is_one_of (info, JOINER_FLAGS);
155 is_consonant (const hb_glyph_info_t &info)
157 return is_one_of (info, CONSONANT_FLAGS);
161 is_halant_or_coeng (const hb_glyph_info_t &info)
163 return is_one_of (info, HALANT_OR_COENG_FLAGS);
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);
179 /* The following act more like the Bindus. */
180 if (unlikely (hb_in_range (u, 0x0953u, 0x0954u)))
182 /* The following act like consonants. */
183 else if (unlikely (hb_in_ranges (u, 0x0A72u, 0x0A73u,
186 /* TODO: The following should only be allowed after a Visarga.
187 * For now, just treat them like regular tone marks. */
188 else if (unlikely (hb_in_range (u, 0x1CE2u, 0x1CE8u)))
190 /* TODO: The following should only be allowed after some of
191 * the nasalization marks, maybe only for U+1CE9..U+1CF1.
192 * For now, just treat them like tone marks. */
193 else if (unlikely (u == 0x1CEDu))
195 /* The following take marks in standalone clusters, similar to Avagraha. */
196 else if (unlikely (hb_in_ranges (u, 0xA8F2u, 0xA8F7u,
201 ASSERT_STATIC ((int) INDIC_SYLLABIC_CATEGORY_AVAGRAHA == OT_Symbol);
203 else if (unlikely (u == 0x17DDu)) /* https://github.com/roozbehp/unicode-data/issues/2 */
208 else if (unlikely (u == 0x17C6u)) cat = OT_N; /* Khmer Bindu doesn't like to be repositioned. */
209 else if (unlikely (hb_in_range (u, 0x2010u, 0x2011u)))
210 cat = OT_PLACEHOLDER;
211 else if (unlikely (u == 0x25CCu)) cat = OT_DOTTEDCIRCLE;
212 else if (unlikely (u == 0xA982u)) cat = OT_SM; /* Javanese repha. */
213 else if (unlikely (u == 0xA9BEu)) cat = OT_CM2; /* Javanese medial ya. */
214 else if (unlikely (u == 0xA9BDu)) { cat = OT_M; pos = POS_POST_C; } /* Javanese vocalic r. */
218 * Re-assign position.
221 if ((FLAG_SAFE (cat) & CONSONANT_FLAGS))
227 else if (cat == OT_M)
229 pos = matra_position (u, pos);
231 else if ((FLAG_SAFE (cat) & (FLAG (OT_SM) | FLAG (OT_VD) | FLAG (OT_A) | FLAG (OT_Symbol))))
236 if (unlikely (u == 0x0B01u)) pos = POS_BEFORE_SUB; /* Oriya Bindu is BeforeSub in the spec. */
240 info.indic_category() = cat;
241 info.indic_position() = pos;
245 * Things above this line should ideally be moved to the Indic table itself.
250 * Indic configurations. Note that we do not want to keep every single script-specific
251 * behavior in these tables necessarily. This should mainly be used for per-script
252 * properties that are cheaper keeping here, than in the code. Ie. if, say, one and
253 * only one script has an exception, that one script can be if'ed directly in the code,
254 * instead of adding a new flag in these structs.
257 enum base_position_t {
259 BASE_POS_LAST_SINHALA,
262 enum reph_position_t {
263 REPH_POS_AFTER_MAIN = POS_AFTER_MAIN,
264 REPH_POS_BEFORE_SUB = POS_BEFORE_SUB,
265 REPH_POS_AFTER_SUB = POS_AFTER_SUB,
266 REPH_POS_BEFORE_POST = POS_BEFORE_POST,
267 REPH_POS_AFTER_POST = POS_AFTER_POST,
268 REPH_POS_DONT_CARE = POS_RA_TO_BECOME_REPH
271 REPH_MODE_IMPLICIT, /* Reph formed out of initial Ra,H sequence. */
272 REPH_MODE_EXPLICIT, /* Reph formed out of initial Ra,H,ZWJ sequence. */
273 REPH_MODE_VIS_REPHA, /* Encoded Repha character, no reordering needed. */
274 REPH_MODE_LOG_REPHA /* Encoded Repha character, needs reordering. */
277 BLWF_MODE_PRE_AND_POST, /* Below-forms feature applied to pre-base and post-base. */
278 BLWF_MODE_POST_ONLY /* Below-forms feature applied to post-base only. */
283 PREF_LEN_DONT_CARE = PREF_LEN_2
285 struct indic_config_t
289 hb_codepoint_t virama;
290 base_position_t base_pos;
291 reph_position_t reph_pos;
292 reph_mode_t reph_mode;
293 blwf_mode_t blwf_mode;
297 static const indic_config_t indic_configs[] =
299 /* Default. Should be first. */
300 {HB_SCRIPT_INVALID, false, 0,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_1},
301 {HB_SCRIPT_DEVANAGARI,true, 0x094Du,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
302 {HB_SCRIPT_BENGALI, true, 0x09CDu,BASE_POS_LAST, REPH_POS_AFTER_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
303 {HB_SCRIPT_GURMUKHI, true, 0x0A4Du,BASE_POS_LAST, REPH_POS_BEFORE_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
304 {HB_SCRIPT_GUJARATI, true, 0x0ACDu,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
305 {HB_SCRIPT_ORIYA, true, 0x0B4Du,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
306 {HB_SCRIPT_TAMIL, true, 0x0BCDu,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_2},
307 {HB_SCRIPT_TELUGU, true, 0x0C4Du,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_EXPLICIT, BLWF_MODE_POST_ONLY, PREF_LEN_2},
308 {HB_SCRIPT_KANNADA, true, 0x0CCDu,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_POST_ONLY, PREF_LEN_2},
309 {HB_SCRIPT_MALAYALAM, true, 0x0D4Du,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_LOG_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_2},
310 {HB_SCRIPT_SINHALA, false,0x0DCAu,BASE_POS_LAST_SINHALA,
311 REPH_POS_AFTER_MAIN, REPH_MODE_EXPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
312 {HB_SCRIPT_KHMER, false,0x17D2u,BASE_POS_FIRST,REPH_POS_DONT_CARE, REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_2},
313 {HB_SCRIPT_JAVANESE, false,0xA9C0u,BASE_POS_FIRST,REPH_POS_DONT_CARE, REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_1},
322 struct feature_list_t {
324 hb_ot_map_feature_flags_t flags;
327 static const feature_list_t
332 * These features are applied in order, one at a time, after initial_reordering.
334 {HB_TAG('n','u','k','t'), F_GLOBAL},
335 {HB_TAG('a','k','h','n'), F_GLOBAL},
336 {HB_TAG('r','p','h','f'), F_NONE},
337 {HB_TAG('r','k','r','f'), F_GLOBAL},
338 {HB_TAG('p','r','e','f'), F_NONE},
339 {HB_TAG('b','l','w','f'), F_NONE},
340 {HB_TAG('a','b','v','f'), F_NONE},
341 {HB_TAG('h','a','l','f'), F_NONE},
342 {HB_TAG('p','s','t','f'), F_NONE},
343 {HB_TAG('v','a','t','u'), F_GLOBAL},
344 {HB_TAG('c','j','c','t'), F_GLOBAL},
345 {HB_TAG('c','f','a','r'), F_NONE},
348 * These features are applied all at once, after final_reordering.
349 * Default Bengali font in Windows for example has intermixed
350 * lookups for init,pres,abvs,blws features.
352 {HB_TAG('i','n','i','t'), F_NONE},
353 {HB_TAG('p','r','e','s'), F_GLOBAL},
354 {HB_TAG('a','b','v','s'), F_GLOBAL},
355 {HB_TAG('b','l','w','s'), F_GLOBAL},
356 {HB_TAG('p','s','t','s'), F_GLOBAL},
357 {HB_TAG('h','a','l','n'), F_GLOBAL},
358 /* Positioning features, though we don't care about the types. */
359 {HB_TAG('d','i','s','t'), F_GLOBAL},
360 {HB_TAG('a','b','v','m'), F_GLOBAL},
361 {HB_TAG('b','l','w','m'), F_GLOBAL},
365 * Must be in the same order as the indic_features array.
392 INDIC_BASIC_FEATURES = INIT /* Don't forget to update this! */
396 setup_syllables (const hb_ot_shape_plan_t *plan,
398 hb_buffer_t *buffer);
400 initial_reordering (const hb_ot_shape_plan_t *plan,
402 hb_buffer_t *buffer);
404 final_reordering (const hb_ot_shape_plan_t *plan,
406 hb_buffer_t *buffer);
408 clear_syllables (const hb_ot_shape_plan_t *plan,
410 hb_buffer_t *buffer);
413 collect_features_indic (hb_ot_shape_planner_t *plan)
415 hb_ot_map_builder_t *map = &plan->map;
417 /* Do this before any lookups have been applied. */
418 map->add_gsub_pause (setup_syllables);
420 map->add_global_bool_feature (HB_TAG('l','o','c','l'));
421 /* The Indic specs do not require ccmp, but we apply it here since if
422 * there is a use of it, it's typically at the beginning. */
423 map->add_global_bool_feature (HB_TAG('c','c','m','p'));
427 map->add_gsub_pause (initial_reordering);
428 for (; i < INDIC_BASIC_FEATURES; i++) {
429 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ);
430 map->add_gsub_pause (NULL);
432 map->add_gsub_pause (final_reordering);
433 for (; i < INDIC_NUM_FEATURES; i++) {
434 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ);
437 map->add_global_bool_feature (HB_TAG('c','a','l','t'));
438 map->add_global_bool_feature (HB_TAG('c','l','i','g'));
440 map->add_gsub_pause (clear_syllables);
444 override_features_indic (hb_ot_shape_planner_t *plan)
446 /* Uniscribe does not apply 'kern' in Khmer. */
447 if (hb_options ().uniscribe_bug_compatible)
449 switch ((hb_tag_t) plan->props.script)
451 case HB_SCRIPT_KHMER:
452 plan->map.add_feature (HB_TAG('k','e','r','n'), 0, F_GLOBAL);
457 plan->map.add_feature (HB_TAG('l','i','g','a'), 0, F_GLOBAL);
461 struct would_substitute_feature_t
463 inline void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_)
465 zero_context = zero_context_;
466 map->get_stage_lookups (0/*GSUB*/,
467 map->get_feature_stage (0/*GSUB*/, feature_tag),
471 inline bool would_substitute (const hb_codepoint_t *glyphs,
472 unsigned int glyphs_count,
473 hb_face_t *face) const
475 for (unsigned int i = 0; i < count; i++)
476 if (hb_ot_layout_lookup_would_substitute_fast (face, lookups[i].index, glyphs, glyphs_count, zero_context))
482 const hb_ot_map_t::lookup_map_t *lookups;
487 struct indic_shape_plan_t
491 inline bool get_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const
493 hb_codepoint_t glyph = virama_glyph;
494 if (unlikely (virama_glyph == (hb_codepoint_t) -1))
496 if (!config->virama || !font->get_nominal_glyph (config->virama, &glyph))
498 /* Technically speaking, the spec says we should apply 'locl' to virama too.
499 * Maybe one day... */
501 /* Our get_nominal_glyph() function needs a font, so we can't get the virama glyph
502 * during shape planning... Instead, overwrite it here. It's safe. Don't worry! */
503 (const_cast<indic_shape_plan_t *> (this))->virama_glyph = glyph;
510 const indic_config_t *config;
513 hb_codepoint_t virama_glyph;
515 would_substitute_feature_t rphf;
516 would_substitute_feature_t pref;
517 would_substitute_feature_t blwf;
518 would_substitute_feature_t pstf;
520 hb_mask_t mask_array[INDIC_NUM_FEATURES];
524 data_create_indic (const hb_ot_shape_plan_t *plan)
526 indic_shape_plan_t *indic_plan = (indic_shape_plan_t *) calloc (1, sizeof (indic_shape_plan_t));
527 if (unlikely (!indic_plan))
530 indic_plan->config = &indic_configs[0];
531 for (unsigned int i = 1; i < ARRAY_LENGTH (indic_configs); i++)
532 if (plan->props.script == indic_configs[i].script) {
533 indic_plan->config = &indic_configs[i];
537 indic_plan->is_old_spec = indic_plan->config->has_old_spec && ((plan->map.chosen_script[0] & 0x000000FFu) != '2');
538 indic_plan->virama_glyph = (hb_codepoint_t) -1;
540 /* Use zero-context would_substitute() matching for new-spec of the main
541 * Indic scripts, and scripts with one spec only, but not for old-specs.
542 * The new-spec for all dual-spec scripts says zero-context matching happens.
544 * However, testing with Malayalam shows that old and new spec both allow
545 * context. Testing with Bengali new-spec however shows that it doesn't.
546 * So, the heuristic here is the way it is. It should *only* be changed,
547 * as we discover more cases of what Windows does. DON'T TOUCH OTHERWISE.
549 bool zero_context = !indic_plan->is_old_spec && plan->props.script != HB_SCRIPT_MALAYALAM;
550 indic_plan->rphf.init (&plan->map, HB_TAG('r','p','h','f'), zero_context);
551 indic_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), zero_context);
552 indic_plan->blwf.init (&plan->map, HB_TAG('b','l','w','f'), zero_context);
553 indic_plan->pstf.init (&plan->map, HB_TAG('p','s','t','f'), zero_context);
555 for (unsigned int i = 0; i < ARRAY_LENGTH (indic_plan->mask_array); i++)
556 indic_plan->mask_array[i] = (indic_features[i].flags & F_GLOBAL) ?
557 0 : plan->map.get_1_mask (indic_features[i].tag);
563 data_destroy_indic (void *data)
568 static indic_position_t
569 consonant_position_from_face (const indic_shape_plan_t *indic_plan,
570 const hb_codepoint_t consonant,
571 const hb_codepoint_t virama,
574 /* For old-spec, the order of glyphs is Consonant,Virama,
575 * whereas for new-spec, it's Virama,Consonant. However,
576 * some broken fonts (like Free Sans) simply copied lookups
577 * from old-spec to new-spec without modification.
578 * And oddly enough, Uniscribe seems to respect those lookups.
579 * Eg. in the sequence U+0924,U+094D,U+0930, Uniscribe finds
580 * base at 0. The font however, only has lookups matching
581 * 930,94D in 'blwf', not the expected 94D,930 (with new-spec
582 * table). As such, we simply match both sequences. Seems
584 hb_codepoint_t glyphs[3] = {virama, consonant, virama};
585 if (indic_plan->blwf.would_substitute (glyphs , 2, face) ||
586 indic_plan->blwf.would_substitute (glyphs+1, 2, face))
588 if (indic_plan->pstf.would_substitute (glyphs , 2, face) ||
589 indic_plan->pstf.would_substitute (glyphs+1, 2, face))
591 unsigned int pref_len = indic_plan->config->pref_len;
592 if ((pref_len == PREF_LEN_2 &&
593 (indic_plan->pref.would_substitute (glyphs , 2, face) ||
594 indic_plan->pref.would_substitute (glyphs+1, 2, face)))
595 || (pref_len == PREF_LEN_1 &&
596 indic_plan->pref.would_substitute (glyphs+1, 1, face)))
602 enum syllable_type_t {
611 #include "hb-ot-shape-complex-indic-machine.hh"
615 setup_masks_indic (const hb_ot_shape_plan_t *plan HB_UNUSED,
617 hb_font_t *font HB_UNUSED)
619 HB_BUFFER_ALLOCATE_VAR (buffer, indic_category);
620 HB_BUFFER_ALLOCATE_VAR (buffer, indic_position);
622 /* We cannot setup masks here. We save information about characters
623 * and setup masks later on in a pause-callback. */
625 unsigned int count = buffer->len;
626 hb_glyph_info_t *info = buffer->info;
627 for (unsigned int i = 0; i < count; i++)
628 set_indic_properties (info[i]);
632 setup_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
633 hb_font_t *font HB_UNUSED,
636 find_syllables (buffer);
640 compare_indic_order (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb)
642 int a = pa->indic_position();
643 int b = pb->indic_position();
645 return a < b ? -1 : a == b ? 0 : +1;
651 update_consonant_positions (const hb_ot_shape_plan_t *plan,
655 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
657 if (indic_plan->config->base_pos != BASE_POS_LAST)
660 hb_codepoint_t virama;
661 if (indic_plan->get_virama_glyph (font, &virama))
663 hb_face_t *face = font->face;
664 unsigned int count = buffer->len;
665 hb_glyph_info_t *info = buffer->info;
666 for (unsigned int i = 0; i < count; i++)
667 if (info[i].indic_position() == POS_BASE_C)
669 hb_codepoint_t consonant = info[i].codepoint;
670 info[i].indic_position() = consonant_position_from_face (indic_plan, consonant, virama, face);
677 * https://www.microsoft.com/typography/otfntdev/devanot/shaping.aspx */
680 initial_reordering_consonant_syllable (const hb_ot_shape_plan_t *plan,
683 unsigned int start, unsigned int end)
685 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
686 hb_glyph_info_t *info = buffer->info;
689 /* 1. Find base consonant:
691 * The shaping engine finds the base consonant of the syllable, using the
692 * following algorithm: starting from the end of the syllable, move backwards
693 * until a consonant is found that does not have a below-base or post-base
694 * form (post-base forms have to follow below-base forms), or that is not a
695 * pre-base reordering Ra, or arrive at the first consonant. The consonant
696 * stopped at will be the base.
698 * o If the syllable starts with Ra + Halant (in a script that has Reph)
699 * and has more than one consonant, Ra is excluded from candidates for
703 unsigned int base = end;
704 bool has_reph = false;
707 /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
708 * and has more than one consonant, Ra is excluded from candidates for
709 * base consonants. */
710 unsigned int limit = start;
711 if (indic_plan->config->reph_pos != REPH_POS_DONT_CARE &&
712 indic_plan->mask_array[RPHF] &&
715 (indic_plan->config->reph_mode == REPH_MODE_IMPLICIT && !is_joiner (info[start + 2])) ||
716 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT && info[start + 2].indic_category() == OT_ZWJ)
719 /* See if it matches the 'rphf' feature. */
720 hb_codepoint_t glyphs[3] = {info[start].codepoint,
721 info[start + 1].codepoint,
722 indic_plan->config->reph_mode == REPH_MODE_EXPLICIT ?
723 info[start + 2].codepoint : 0};
724 if (indic_plan->rphf.would_substitute (glyphs, 2, face) ||
725 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT &&
726 indic_plan->rphf.would_substitute (glyphs, 3, face)))
729 while (limit < end && is_joiner (info[limit]))
734 } else if (indic_plan->config->reph_mode == REPH_MODE_LOG_REPHA && info[start].indic_category() == OT_Repha)
737 while (limit < end && is_joiner (info[limit]))
743 switch (indic_plan->config->base_pos)
747 /* -> starting from the end of the syllable, move backwards */
748 unsigned int i = end;
749 bool seen_below = false;
752 /* -> until a consonant is found */
753 if (is_consonant (info[i]))
755 /* -> that does not have a below-base or post-base form
756 * (post-base forms have to follow below-base forms), */
757 if (info[i].indic_position() != POS_BELOW_C &&
758 (info[i].indic_position() != POS_POST_C || seen_below))
763 if (info[i].indic_position() == POS_BELOW_C)
766 /* -> or that is not a pre-base reordering Ra,
768 * IMPLEMENTATION NOTES:
770 * Our pre-base reordering Ra's are marked POS_POST_C, so will be skipped
771 * by the logic above already.
774 /* -> or arrive at the first consonant. The consonant stopped at will
780 /* A ZWJ after a Halant stops the base search, and requests an explicit
782 * A ZWJ before a Halant, requests a subjoined form instead, and hence
783 * search continues. This is particularly important for Bengali
784 * sequence Ra,H,Ya that should form Ya-Phalaa by subjoining Ya. */
786 info[i].indic_category() == OT_ZWJ &&
787 info[i - 1].indic_category() == OT_H)
794 case BASE_POS_LAST_SINHALA:
796 /* Sinhala base positioning is slightly different from main Indic, in that:
797 * 1. Its ZWJ behavior is different,
798 * 2. We don't need to look into the font for consonant positions.
804 /* Find the last base consonant that is not blocked by ZWJ. If there is
805 * a ZWJ right before a base consonant, that would request a subjoined form. */
806 for (unsigned int i = limit; i < end; i++)
807 if (is_consonant (info[i]))
809 if (limit < i && info[i - 1].indic_category() == OT_ZWJ)
815 /* Mark all subsequent consonants as below. */
816 for (unsigned int i = base + 1; i < end; i++)
817 if (is_consonant (info[i]))
818 info[i].indic_position() = POS_BELOW_C;
824 /* The first consonant is always the base. */
826 assert (indic_plan->config->reph_mode == REPH_MODE_VIS_REPHA);
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;
839 /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
840 * and has more than one consonant, Ra is excluded from candidates for
843 * Only do this for unforced Reph. (ie. not for Ra,H,ZWJ. */
844 if (has_reph && base == start && limit - base <= 2) {
845 /* Have no other consonant, so Reph is not formed and Ra becomes base. */
851 /* 2. Decompose and reorder Matras:
853 * Each matra and any syllable modifier sign in the cluster are moved to the
854 * appropriate position relative to the consonant(s) in the cluster. The
855 * shaping engine decomposes two- or three-part matras into their constituent
856 * parts before any repositioning. Matra characters are classified by which
857 * consonant in a conjunct they have affinity for and are reordered to the
858 * following positions:
860 * o Before first half form in the syllable
861 * o After subjoined consonants
862 * o After post-form consonant
863 * o After main consonant (for above marks)
865 * IMPLEMENTATION NOTES:
867 * The normalize() routine has already decomposed matras for us, so we don't
868 * need to worry about that.
872 /* 3. Reorder marks to canonical order:
874 * Adjacent nukta and halant or nukta and vedic sign are always repositioned
875 * if necessary, so that the nukta is first.
877 * IMPLEMENTATION NOTES:
879 * We don't need to do this: the normalize() routine already did this for us.
883 /* Reorder characters */
885 for (unsigned int i = start; i < base; i++)
886 info[i].indic_position() = MIN (POS_PRE_C, (indic_position_t) info[i].indic_position());
889 info[base].indic_position() = POS_BASE_C;
891 /* Mark final consonants. A final consonant is one appearing after a matra,
893 for (unsigned int i = base + 1; i < end; i++)
894 if (info[i].indic_category() == OT_M) {
895 for (unsigned int j = i + 1; j < end; j++)
896 if (is_consonant (info[j])) {
897 info[j].indic_position() = POS_FINAL_C;
903 /* Handle beginning Ra */
905 info[start].indic_position() = POS_RA_TO_BECOME_REPH;
907 /* For old-style Indic script tags, move the first post-base Halant after
910 * Reports suggest that in some scripts Uniscribe does this only if there
911 * is *not* a Halant after last consonant already (eg. Kannada), while it
912 * does it unconditionally in other scripts (eg. Malayalam). We don't
913 * currently know about other scripts, so we single out Malayalam for now.
916 * U+0C9A,U+0CCD,U+0C9A,U+0CCD
917 * With some versions of Lohit Kannada.
918 * https://bugs.freedesktop.org/show_bug.cgi?id=59118
920 * Malayalam test case:
921 * U+0D38,U+0D4D,U+0D31,U+0D4D,U+0D31,U+0D4D
922 * With lohit-ttf-20121122/Lohit-Malayalam.ttf
924 if (indic_plan->is_old_spec)
926 bool disallow_double_halants = buffer->props.script != HB_SCRIPT_MALAYALAM;
927 for (unsigned int i = base + 1; i < end; i++)
928 if (info[i].indic_category() == OT_H)
931 for (j = end - 1; j > i; j--)
932 if (is_consonant (info[j]) ||
933 (disallow_double_halants && info[j].indic_category() == OT_H))
935 if (info[j].indic_category() != OT_H && j > i) {
936 /* Move Halant to after last consonant. */
937 hb_glyph_info_t t = info[i];
938 memmove (&info[i], &info[i + 1], (j - i) * sizeof (info[0]));
945 /* Attach misc marks to previous char to move with them. */
947 indic_position_t last_pos = POS_START;
948 for (unsigned int i = start; i < end; i++)
950 if ((FLAG_SAFE (info[i].indic_category()) & (JOINER_FLAGS | FLAG (OT_N) | FLAG (OT_RS) | MEDIAL_FLAGS | HALANT_OR_COENG_FLAGS)))
952 info[i].indic_position() = last_pos;
953 if (unlikely (info[i].indic_category() == OT_H &&
954 info[i].indic_position() == POS_PRE_M))
957 * Uniscribe doesn't move the Halant with Left Matra.
958 * TEST: U+092B,U+093F,U+094DE
959 * We follow. This is important for the Sinhala
960 * U+0DDA split matra since it decomposes to U+0DD9,U+0DCA
961 * where U+0DD9 is a left matra and U+0DCA is the virama.
962 * We don't want to move the virama with the left matra.
963 * TEST: U+0D9A,U+0DDA
965 for (unsigned int j = i; j > start; j--)
966 if (info[j - 1].indic_position() != POS_PRE_M) {
967 info[i].indic_position() = info[j - 1].indic_position();
971 } else if (info[i].indic_position() != POS_SMVD) {
972 last_pos = (indic_position_t) info[i].indic_position();
976 /* For post-base consonants let them own anything before them
977 * since the last consonant or matra. */
979 unsigned int last = base;
980 for (unsigned int i = base + 1; i < end; i++)
981 if (is_consonant (info[i]))
983 for (unsigned int j = last + 1; j < i; j++)
984 if (info[j].indic_position() < POS_SMVD)
985 info[j].indic_position() = info[i].indic_position();
987 } else if (info[i].indic_category() == OT_M)
993 /* Use syllable() for sort accounting temporarily. */
994 unsigned int syllable = info[start].syllable();
995 for (unsigned int i = start; i < end; i++)
996 info[i].syllable() = i - start;
998 /* Sit tight, rock 'n roll! */
999 hb_stable_sort (info + start, end - start, compare_indic_order);
1000 /* Find base again */
1002 for (unsigned int i = start; i < end; i++)
1003 if (info[i].indic_position() == POS_BASE_C)
1008 /* Things are out-of-control for post base positions, they may shuffle
1009 * around like crazy. In old-spec mode, we move halants around, so in
1010 * that case merge all clusters after base. Otherwise, check the sort
1011 * order and merge as needed.
1012 * For pre-base stuff, we handle cluster issues in final reordering.
1014 * We could use buffer->sort() for this, if there was no special
1015 * reordering of pre-base stuff happening later...
1017 if (indic_plan->is_old_spec || end - base > 127)
1018 buffer->merge_clusters (base, end);
1021 /* Note! syllable() is a one-byte field. */
1022 for (unsigned int i = base; i < end; i++)
1023 if (info[i].syllable() != 255)
1025 unsigned int max = i;
1026 unsigned int j = start + info[i].syllable();
1030 unsigned int next = start + info[j].syllable();
1031 info[j].syllable() = 255; /* So we don't process j later again. */
1035 buffer->merge_clusters (i, max + 1);
1039 /* Put syllable back in. */
1040 for (unsigned int i = start; i < end; i++)
1041 info[i].syllable() = syllable;
1044 /* Setup masks now */
1050 for (unsigned int i = start; i < end && info[i].indic_position() == POS_RA_TO_BECOME_REPH; i++)
1051 info[i].mask |= indic_plan->mask_array[RPHF];
1054 mask = indic_plan->mask_array[HALF];
1055 if (!indic_plan->is_old_spec &&
1056 indic_plan->config->blwf_mode == BLWF_MODE_PRE_AND_POST)
1057 mask |= indic_plan->mask_array[BLWF];
1058 for (unsigned int i = start; i < base; i++)
1059 info[i].mask |= mask;
1063 info[base].mask |= mask;
1065 mask = indic_plan->mask_array[BLWF] | indic_plan->mask_array[ABVF] | indic_plan->mask_array[PSTF];
1066 for (unsigned int i = base + 1; i < end; i++)
1067 info[i].mask |= mask;
1070 if (indic_plan->is_old_spec &&
1071 buffer->props.script == HB_SCRIPT_DEVANAGARI)
1073 /* Old-spec eye-lash Ra needs special handling. From the
1076 * "The feature 'below-base form' is applied to consonants
1077 * having below-base forms and following the base consonant.
1078 * The exception is vattu, which may appear below half forms
1079 * as well as below the base glyph. The feature 'below-base
1080 * form' will be applied to all such occurrences of Ra as well."
1082 * Test case: U+0924,U+094D,U+0930,U+094d,U+0915
1083 * with Sanskrit 2003 font.
1085 * However, note that Ra,Halant,ZWJ is the correct way to
1086 * request eyelash form of Ra, so we wouldbn't inhibit it
1089 * Test case: U+0924,U+094D,U+0930,U+094d,U+200D,U+0915
1091 for (unsigned int i = start; i + 1 < base; i++)
1092 if (info[i ].indic_category() == OT_Ra &&
1093 info[i+1].indic_category() == OT_H &&
1095 info[i+2].indic_category() != OT_ZWJ))
1097 info[i ].mask |= indic_plan->mask_array[BLWF];
1098 info[i+1].mask |= indic_plan->mask_array[BLWF];
1102 unsigned int pref_len = indic_plan->config->pref_len;
1103 if (indic_plan->mask_array[PREF] && base + pref_len < end)
1105 assert (1 <= pref_len && pref_len <= 2);
1106 /* Find a Halant,Ra sequence and mark it for pre-base reordering processing. */
1107 for (unsigned int i = base + 1; i + pref_len - 1 < end; i++) {
1108 hb_codepoint_t glyphs[2];
1109 for (unsigned int j = 0; j < pref_len; j++)
1110 glyphs[j] = info[i + j].codepoint;
1111 if (indic_plan->pref.would_substitute (glyphs, pref_len, face))
1113 for (unsigned int j = 0; j < pref_len; j++)
1114 info[i++].mask |= indic_plan->mask_array[PREF];
1116 /* Mark the subsequent stuff with 'cfar'. Used in Khmer.
1117 * Read the feature spec.
1118 * This allows distinguishing the following cases with MS Khmer fonts:
1119 * U+1784,U+17D2,U+179A,U+17D2,U+1782
1120 * U+1784,U+17D2,U+1782,U+17D2,U+179A
1122 if (indic_plan->mask_array[CFAR])
1123 for (; i < end; i++)
1124 info[i].mask |= indic_plan->mask_array[CFAR];
1131 /* Apply ZWJ/ZWNJ effects */
1132 for (unsigned int i = start + 1; i < end; i++)
1133 if (is_joiner (info[i])) {
1134 bool non_joiner = info[i].indic_category() == OT_ZWNJ;
1140 /* ZWJ/ZWNJ should disable CJCT. They do that by simply
1141 * being there, since we don't skip them for the CJCT
1142 * feature (ie. F_MANUAL_ZWJ) */
1144 /* A ZWNJ disables HALF. */
1146 info[j].mask &= ~indic_plan->mask_array[HALF];
1148 } while (j > start && !is_consonant (info[j]));
1153 initial_reordering_standalone_cluster (const hb_ot_shape_plan_t *plan,
1155 hb_buffer_t *buffer,
1156 unsigned int start, unsigned int end)
1158 /* We treat placeholder/dotted-circle as if they are consonants, so we
1159 * should just chain. Only if not in compatibility mode that is... */
1161 if (hb_options ().uniscribe_bug_compatible)
1163 /* For dotted-circle, this is what Uniscribe does:
1164 * If dotted-circle is the last glyph, it just does nothing.
1165 * Ie. It doesn't form Reph. */
1166 if (buffer->info[end - 1].indic_category() == OT_DOTTEDCIRCLE)
1170 initial_reordering_consonant_syllable (plan, face, buffer, start, end);
1174 initial_reordering_syllable (const hb_ot_shape_plan_t *plan,
1176 hb_buffer_t *buffer,
1177 unsigned int start, unsigned int end)
1179 syllable_type_t syllable_type = (syllable_type_t) (buffer->info[start].syllable() & 0x0F);
1180 switch (syllable_type)
1182 case vowel_syllable: /* We made the vowels look like consonants. So let's call the consonant logic! */
1183 case consonant_syllable:
1184 initial_reordering_consonant_syllable (plan, face, buffer, start, end);
1187 case broken_cluster: /* We already inserted dotted-circles, so just call the standalone_cluster. */
1188 case standalone_cluster:
1189 initial_reordering_standalone_cluster (plan, face, buffer, start, end);
1192 case symbol_cluster:
1193 case non_indic_cluster:
1199 insert_dotted_circles (const hb_ot_shape_plan_t *plan HB_UNUSED,
1201 hb_buffer_t *buffer)
1203 /* Note: This loop is extra overhead, but should not be measurable. */
1204 bool has_broken_syllables = false;
1205 unsigned int count = buffer->len;
1206 hb_glyph_info_t *info = buffer->info;
1207 for (unsigned int i = 0; i < count; i++)
1208 if ((info[i].syllable() & 0x0F) == broken_cluster)
1210 has_broken_syllables = true;
1213 if (likely (!has_broken_syllables))
1217 hb_codepoint_t dottedcircle_glyph;
1218 if (!font->get_nominal_glyph (0x25CCu, &dottedcircle_glyph))
1221 hb_glyph_info_t dottedcircle = {0};
1222 dottedcircle.codepoint = 0x25CCu;
1223 set_indic_properties (dottedcircle);
1224 dottedcircle.codepoint = dottedcircle_glyph;
1226 buffer->clear_output ();
1229 unsigned int last_syllable = 0;
1230 while (buffer->idx < buffer->len && !buffer->in_error)
1232 unsigned int syllable = buffer->cur().syllable();
1233 syllable_type_t syllable_type = (syllable_type_t) (syllable & 0x0F);
1234 if (unlikely (last_syllable != syllable && syllable_type == broken_cluster))
1236 last_syllable = syllable;
1238 hb_glyph_info_t ginfo = dottedcircle;
1239 ginfo.cluster = buffer->cur().cluster;
1240 ginfo.mask = buffer->cur().mask;
1241 ginfo.syllable() = buffer->cur().syllable();
1242 /* TODO Set glyph_props? */
1244 /* Insert dottedcircle after possible Repha. */
1245 while (buffer->idx < buffer->len && !buffer->in_error &&
1246 last_syllable == buffer->cur().syllable() &&
1247 buffer->cur().indic_category() == OT_Repha)
1248 buffer->next_glyph ();
1250 buffer->output_info (ginfo);
1253 buffer->next_glyph ();
1256 buffer->swap_buffers ();
1260 initial_reordering (const hb_ot_shape_plan_t *plan,
1262 hb_buffer_t *buffer)
1264 update_consonant_positions (plan, font, buffer);
1265 insert_dotted_circles (plan, font, buffer);
1267 foreach_syllable (buffer, start, end)
1268 initial_reordering_syllable (plan, font->face, buffer, start, end);
1272 final_reordering_syllable (const hb_ot_shape_plan_t *plan,
1273 hb_buffer_t *buffer,
1274 unsigned int start, unsigned int end)
1276 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
1277 hb_glyph_info_t *info = buffer->info;
1280 /* This function relies heavily on halant glyphs. Lots of ligation
1281 * and possibly multiplication substitutions happened prior to this
1282 * phase, and that might have messed up our properties. Recover
1283 * from a particular case of that where we're fairly sure that a
1284 * class of OT_H is desired but has been lost. */
1285 if (indic_plan->virama_glyph)
1287 unsigned int virama_glyph = indic_plan->virama_glyph;
1288 for (unsigned int i = start; i < end; i++)
1289 if (info[i].codepoint == virama_glyph &&
1290 _hb_glyph_info_ligated (&info[i]) &&
1291 _hb_glyph_info_multiplied (&info[i]))
1293 /* This will make sure that this glyph passes is_halant_or_coeng() test. */
1294 info[i].indic_category() = OT_H;
1295 _hb_glyph_info_clear_ligated_and_multiplied (&info[i]);
1300 /* 4. Final reordering:
1302 * After the localized forms and basic shaping forms GSUB features have been
1303 * applied (see below), the shaping engine performs some final glyph
1304 * reordering before applying all the remaining font features to the entire
1308 bool try_pref = !!indic_plan->mask_array[PREF];
1310 /* Find base again */
1312 for (base = start; base < end; base++)
1313 if (info[base].indic_position() >= POS_BASE_C)
1315 if (try_pref && base + 1 < end && indic_plan->config->pref_len == 2)
1317 for (unsigned int i = base + 1; i < end; i++)
1318 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0)
1320 if (!(_hb_glyph_info_substituted (&info[i]) &&
1321 _hb_glyph_info_ligated_and_didnt_multiply (&info[i])))
1323 /* Ok, this was a 'pref' candidate but didn't form any.
1324 * Base is around here... */
1326 while (base < end && is_halant_or_coeng (info[base]))
1328 info[base].indic_position() = POS_BASE_C;
1335 /* For Malayalam, skip over unformed below- (but NOT post-) forms. */
1336 if (buffer->props.script == HB_SCRIPT_MALAYALAM)
1338 for (unsigned int i = base + 1; i < end; i++)
1340 while (i < end && is_joiner (info[i]))
1342 if (i == end || !is_halant_or_coeng (info[i]))
1344 i++; /* Skip halant. */
1345 while (i < end && is_joiner (info[i]))
1347 if (i < end && is_consonant (info[i]) && info[i].indic_position() == POS_BELOW_C)
1350 info[base].indic_position() = POS_BASE_C;
1355 if (start < base && info[base].indic_position() > POS_BASE_C)
1359 if (base == end && start < base &&
1360 is_one_of (info[base - 1], FLAG (OT_ZWJ)))
1363 while (start < base &&
1364 is_one_of (info[base], (FLAG (OT_N) | HALANT_OR_COENG_FLAGS)))
1368 /* o Reorder matras:
1370 * If a pre-base matra character had been reordered before applying basic
1371 * features, the glyph can be moved closer to the main consonant based on
1372 * whether half-forms had been formed. Actual position for the matra is
1373 * defined as “after last standalone halant glyph, after initial matra
1374 * position and before the main consonant”. If ZWJ or ZWNJ follow this
1375 * halant, position is moved after it.
1378 if (start + 1 < end && start < base) /* Otherwise there can't be any pre-base matra characters. */
1380 /* If we lost track of base, alas, position before last thingy. */
1381 unsigned int new_pos = base == end ? base - 2 : base - 1;
1383 /* Malayalam / Tamil do not have "half" forms or explicit virama forms.
1384 * The glyphs formed by 'half' are Chillus or ligated explicit viramas.
1385 * We want to position matra after them.
1387 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL)
1389 while (new_pos > start &&
1390 !(is_one_of (info[new_pos], (FLAG (OT_M) | HALANT_OR_COENG_FLAGS))))
1393 /* If we found no Halant we are done.
1394 * Otherwise only proceed if the Halant does
1395 * not belong to the Matra itself! */
1396 if (is_halant_or_coeng (info[new_pos]) &&
1397 info[new_pos].indic_position() != POS_PRE_M)
1399 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */
1400 if (new_pos + 1 < end && is_joiner (info[new_pos + 1]))
1404 new_pos = start; /* No move. */
1407 if (start < new_pos && info[new_pos].indic_position () != POS_PRE_M)
1409 /* Now go see if there's actually any matras... */
1410 for (unsigned int i = new_pos; i > start; i--)
1411 if (info[i - 1].indic_position () == POS_PRE_M)
1413 unsigned int old_pos = i - 1;
1414 if (old_pos < base && base <= new_pos) /* Shouldn't actually happen. */
1417 hb_glyph_info_t tmp = info[old_pos];
1418 memmove (&info[old_pos], &info[old_pos + 1], (new_pos - old_pos) * sizeof (info[0]));
1419 info[new_pos] = tmp;
1421 /* Note: this merge_clusters() is intentionally *after* the reordering.
1422 * Indic matra reordering is special and tricky... */
1423 buffer->merge_clusters (new_pos, MIN (end, base + 1));
1428 for (unsigned int i = start; i < base; i++)
1429 if (info[i].indic_position () == POS_PRE_M) {
1430 buffer->merge_clusters (i, MIN (end, base + 1));
1439 * Reph’s original position is always at the beginning of the syllable,
1440 * (i.e. it is not reordered at the character reordering stage). However,
1441 * it will be reordered according to the basic-forms shaping results.
1442 * Possible positions for reph, depending on the script, are; after main,
1443 * before post-base consonant forms, and after post-base consonant forms.
1448 * - If repha is encoded as a sequence of characters (Ra,H or Ra,H,ZWJ), then
1449 * we should only move it if the sequence ligated to the repha form.
1451 * - If repha is encoded separately and in the logical position, we should only
1452 * move it if it did NOT ligate. If it ligated, it's probably the font trying
1453 * to make it work without the reordering.
1455 if (start + 1 < end &&
1456 info[start].indic_position() == POS_RA_TO_BECOME_REPH &&
1457 ((info[start].indic_category() == OT_Repha) ^
1458 _hb_glyph_info_ligated_and_didnt_multiply (&info[start])))
1460 unsigned int new_reph_pos;
1461 reph_position_t reph_pos = indic_plan->config->reph_pos;
1463 assert (reph_pos != REPH_POS_DONT_CARE);
1465 /* 1. If reph should be positioned after post-base consonant forms,
1466 * proceed to step 5.
1468 if (reph_pos == REPH_POS_AFTER_POST)
1473 /* 2. If the reph repositioning class is not after post-base: target
1474 * position is after the first explicit halant glyph between the
1475 * first post-reph consonant and last main consonant. If ZWJ or ZWNJ
1476 * are following this halant, position is moved after it. If such
1477 * position is found, this is the target position. Otherwise,
1478 * proceed to the next step.
1480 * Note: in old-implementation fonts, where classifications were
1481 * fixed in shaping engine, there was no case where reph position
1482 * will be found on this step.
1485 new_reph_pos = start + 1;
1486 while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos]))
1489 if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos]))
1491 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */
1492 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1]))
1498 /* 3. If reph should be repositioned after the main consonant: find the
1499 * first consonant not ligated with main, or find the first
1500 * consonant that is not a potential pre-base reordering Ra.
1502 if (reph_pos == REPH_POS_AFTER_MAIN)
1504 new_reph_pos = base;
1505 while (new_reph_pos + 1 < end && info[new_reph_pos + 1].indic_position() <= POS_AFTER_MAIN)
1507 if (new_reph_pos < end)
1511 /* 4. If reph should be positioned before post-base consonant, find
1512 * first post-base classified consonant not ligated with main. If no
1513 * consonant is found, the target position should be before the
1514 * first matra, syllable modifier sign or vedic sign.
1516 /* This is our take on what step 4 is trying to say (and failing, BADLY). */
1517 if (reph_pos == REPH_POS_AFTER_SUB)
1519 new_reph_pos = base;
1520 while (new_reph_pos < end &&
1521 !( FLAG_SAFE (info[new_reph_pos + 1].indic_position()) & (FLAG (POS_POST_C) | FLAG (POS_AFTER_POST) | FLAG (POS_SMVD))))
1523 if (new_reph_pos < end)
1527 /* 5. If no consonant is found in steps 3 or 4, move reph to a position
1528 * immediately before the first post-base matra, syllable modifier
1529 * sign or vedic sign that has a reordering class after the intended
1530 * reph position. For example, if the reordering position for reph
1531 * is post-main, it will skip above-base matras that also have a
1532 * post-main position.
1536 /* Copied from step 2. */
1537 new_reph_pos = start + 1;
1538 while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos]))
1541 if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos]))
1543 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */
1544 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1]))
1550 /* 6. Otherwise, reorder reph to the end of the syllable.
1553 new_reph_pos = end - 1;
1554 while (new_reph_pos > start && info[new_reph_pos].indic_position() == POS_SMVD)
1558 * If the Reph is to be ending up after a Matra,Halant sequence,
1559 * position it before that Halant so it can interact with the Matra.
1560 * However, if it's a plain Consonant,Halant we shouldn't do that.
1561 * Uniscribe doesn't do this.
1562 * TEST: U+0930,U+094D,U+0915,U+094B,U+094D
1564 if (!hb_options ().uniscribe_bug_compatible &&
1565 unlikely (is_halant_or_coeng (info[new_reph_pos]))) {
1566 for (unsigned int i = base + 1; i < new_reph_pos; i++)
1567 if (info[i].indic_category() == OT_M) {
1578 buffer->merge_clusters (start, new_reph_pos + 1);
1579 hb_glyph_info_t reph = info[start];
1580 memmove (&info[start], &info[start + 1], (new_reph_pos - start) * sizeof (info[0]));
1581 info[new_reph_pos] = reph;
1583 if (start < base && base <= new_reph_pos)
1589 /* o Reorder pre-base reordering consonants:
1591 * If a pre-base reordering consonant is found, reorder it according to
1592 * the following rules:
1595 if (try_pref && base + 1 < end) /* Otherwise there can't be any pre-base reordering Ra. */
1597 unsigned int pref_len = indic_plan->config->pref_len;
1598 for (unsigned int i = base + 1; i < end; i++)
1599 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0)
1601 /* 1. Only reorder a glyph produced by substitution during application
1602 * of the <pref> feature. (Note that a font may shape a Ra consonant with
1603 * the feature generally but block it in certain contexts.)
1605 /* Note: We just check that something got substituted. We don't check that
1606 * the <pref> feature actually did it...
1608 * If pref len is longer than one, then only reorder if it ligated. If
1609 * pref len is one, only reorder if it didn't ligate with other things. */
1610 if (_hb_glyph_info_substituted (&info[i]) &&
1611 ((pref_len == 1) ^ _hb_glyph_info_ligated_and_didnt_multiply (&info[i])))
1614 * 2. Try to find a target position the same way as for pre-base matra.
1615 * If it is found, reorder pre-base consonant glyph.
1617 * 3. If position is not found, reorder immediately before main
1621 unsigned int new_pos = base;
1622 /* Malayalam / Tamil do not have "half" forms or explicit virama forms.
1623 * The glyphs formed by 'half' are Chillus or ligated explicit viramas.
1624 * We want to position matra after them.
1626 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL)
1628 while (new_pos > start &&
1629 !(is_one_of (info[new_pos - 1], FLAG(OT_M) | HALANT_OR_COENG_FLAGS)))
1632 /* In Khmer coeng model, a H,Ra can go *after* matras. If it goes after a
1633 * split matra, it should be reordered to *before* the left part of such matra. */
1634 if (new_pos > start && info[new_pos - 1].indic_category() == OT_M)
1636 unsigned int old_pos = i;
1637 for (unsigned int j = base + 1; j < old_pos; j++)
1638 if (info[j].indic_category() == OT_M)
1646 if (new_pos > start && is_halant_or_coeng (info[new_pos - 1]))
1648 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */
1649 if (new_pos < end && is_joiner (info[new_pos]))
1654 unsigned int old_pos = i;
1656 buffer->merge_clusters (new_pos, old_pos + 1);
1657 hb_glyph_info_t tmp = info[old_pos];
1658 memmove (&info[new_pos + 1], &info[new_pos], (old_pos - new_pos) * sizeof (info[0]));
1659 info[new_pos] = tmp;
1661 if (new_pos <= base && base < old_pos)
1671 /* Apply 'init' to the Left Matra if it's a word start. */
1672 if (info[start].indic_position () == POS_PRE_M &&
1674 !(FLAG_SAFE (_hb_glyph_info_get_general_category (&info[start - 1])) &
1675 FLAG_RANGE (HB_UNICODE_GENERAL_CATEGORY_FORMAT, HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK))))
1676 info[start].mask |= indic_plan->mask_array[INIT];
1680 * Finish off the clusters and go home!
1682 if (hb_options ().uniscribe_bug_compatible)
1684 switch ((hb_tag_t) plan->props.script)
1686 case HB_SCRIPT_TAMIL:
1687 case HB_SCRIPT_SINHALA:
1691 /* Uniscribe merges the entire cluster... Except for Tamil & Sinhala.
1692 * This means, half forms are submerged into the main consonants cluster.
1693 * This is unnecessary, and makes cursor positioning harder, but that's what
1694 * Uniscribe does. */
1695 buffer->merge_clusters (start, end);
1703 final_reordering (const hb_ot_shape_plan_t *plan,
1704 hb_font_t *font HB_UNUSED,
1705 hb_buffer_t *buffer)
1707 unsigned int count = buffer->len;
1708 if (unlikely (!count)) return;
1710 foreach_syllable (buffer, start, end)
1711 final_reordering_syllable (plan, buffer, start, end);
1713 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_category);
1714 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_position);
1719 clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
1720 hb_font_t *font HB_UNUSED,
1721 hb_buffer_t *buffer)
1723 hb_glyph_info_t *info = buffer->info;
1724 unsigned int count = buffer->len;
1725 for (unsigned int i = 0; i < count; i++)
1726 info[i].syllable() = 0;
1731 decompose_indic (const hb_ot_shape_normalize_context_t *c,
1738 /* Don't decompose these. */
1739 case 0x0931u : return false;
1740 case 0x0B94u : return false;
1744 * Decompose split matras that don't have Unicode decompositions.
1747 case 0x0F77u : *a = 0x0FB2u; *b= 0x0F81u; return true;
1748 case 0x0F79u : *a = 0x0FB3u; *b= 0x0F81u; return true;
1749 case 0x17BEu : *a = 0x17C1u; *b= 0x17BEu; return true;
1750 case 0x17BFu : *a = 0x17C1u; *b= 0x17BFu; return true;
1751 case 0x17C0u : *a = 0x17C1u; *b= 0x17C0u; return true;
1752 case 0x17C4u : *a = 0x17C1u; *b= 0x17C4u; return true;
1753 case 0x17C5u : *a = 0x17C1u; *b= 0x17C5u; return true;
1754 case 0x1925u : *a = 0x1920u; *b= 0x1923u; return true;
1755 case 0x1926u : *a = 0x1920u; *b= 0x1924u; return true;
1756 case 0x1B3Cu : *a = 0x1B42u; *b= 0x1B3Cu; return true;
1757 case 0x1112Eu : *a = 0x11127u; *b= 0x11131u; return true;
1758 case 0x1112Fu : *a = 0x11127u; *b= 0x11132u; return true;
1760 /* This one has no decomposition in Unicode, but needs no decomposition either. */
1761 /* case 0x0AC9u : return false; */
1762 case 0x0B57u : *a = no decomp, -> RIGHT; return true;
1763 case 0x1C29u : *a = no decomp, -> LEFT; return true;
1764 case 0xA9C0u : *a = no decomp, -> RIGHT; return true;
1765 case 0x111BuF : *a = no decomp, -> ABOVE; return true;
1769 if ((ab == 0x0DDAu || hb_in_range (ab, 0x0DDCu, 0x0DDEu)))
1772 * Sinhala split matras... Let the fun begin.
1774 * These four characters have Unicode decompositions. However, Uniscribe
1775 * decomposes them "Khmer-style", that is, it uses the character itself to
1776 * get the second half. The first half of all four decompositions is always
1779 * Now, there are buggy fonts, namely, the widely used lklug.ttf, that are
1780 * broken with Uniscribe. But we need to support them. As such, we only
1781 * do the Uniscribe-style decomposition if the character is transformed into
1782 * its "sec.half" form by the 'pstf' feature. Otherwise, we fall back to
1783 * Unicode decomposition.
1785 * Note that we can't unconditionally use Unicode decomposition. That would
1786 * break some other fonts, that are designed to work with Uniscribe, and
1787 * don't have positioning features for the Unicode-style decomposition.
1791 * The Uniscribe behavior is now documented in the newly published Sinhala
1794 * http://www.microsoft.com/typography/OpenTypeDev/sinhala/intro.htm#shaping
1797 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) c->plan->data;
1799 hb_codepoint_t glyph;
1801 if (hb_options ().uniscribe_bug_compatible ||
1802 (c->font->get_nominal_glyph (ab, &glyph) &&
1803 indic_plan->pstf.would_substitute (&glyph, 1, c->font->face)))
1805 /* Ok, safe to use Uniscribe-style decomposition. */
1812 return (bool) c->unicode->decompose (ab, a, b);
1816 compose_indic (const hb_ot_shape_normalize_context_t *c,
1821 /* Avoid recomposing split matras. */
1822 if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a)))
1825 /* Composition-exclusion exceptions that we want to recompose. */
1826 if (a == 0x09AFu && b == 0x09BCu) { *ab = 0x09DFu; return true; }
1828 return (bool) c->unicode->compose (a, b, ab);
1832 const hb_ot_complex_shaper_t _hb_ot_complex_shaper_indic =
1835 collect_features_indic,
1836 override_features_indic,
1839 NULL, /* preprocess_text */
1840 NULL, /* postprocess_glyphs */
1841 HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT,
1845 HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE,
1846 false, /* fallback_position */