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"
37 * Indic configurations. Note that we do not want to keep every single script-specific
38 * behavior in these tables necessarily. This should mainly be used for per-script
39 * properties that are cheaper keeping here, than in the code. Ie. if, say, one and
40 * only one script has an exception, that one script can be if'ed directly in the code,
41 * instead of adding a new flag in these structs.
44 enum base_position_t {
45 BASE_POS_LAST_SINHALA,
48 enum reph_position_t {
49 REPH_POS_AFTER_MAIN = POS_AFTER_MAIN,
50 REPH_POS_BEFORE_SUB = POS_BEFORE_SUB,
51 REPH_POS_AFTER_SUB = POS_AFTER_SUB,
52 REPH_POS_BEFORE_POST = POS_BEFORE_POST,
53 REPH_POS_AFTER_POST = POS_AFTER_POST
56 REPH_MODE_IMPLICIT, /* Reph formed out of initial Ra,H sequence. */
57 REPH_MODE_EXPLICIT, /* Reph formed out of initial Ra,H,ZWJ sequence. */
58 REPH_MODE_LOG_REPHA /* Encoded Repha character, needs reordering. */
61 BLWF_MODE_PRE_AND_POST, /* Below-forms feature applied to pre-base and post-base. */
62 BLWF_MODE_POST_ONLY /* Below-forms feature applied to post-base only. */
68 hb_codepoint_t virama;
69 base_position_t base_pos;
70 reph_position_t reph_pos;
71 reph_mode_t reph_mode;
72 blwf_mode_t blwf_mode;
75 static const indic_config_t indic_configs[] =
77 /* Default. Should be first. */
78 {HB_SCRIPT_INVALID, false, 0,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
79 {HB_SCRIPT_DEVANAGARI,true, 0x094Du,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
80 {HB_SCRIPT_BENGALI, true, 0x09CDu,BASE_POS_LAST, REPH_POS_AFTER_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
81 {HB_SCRIPT_GURMUKHI, true, 0x0A4Du,BASE_POS_LAST, REPH_POS_BEFORE_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
82 {HB_SCRIPT_GUJARATI, true, 0x0ACDu,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
83 {HB_SCRIPT_ORIYA, true, 0x0B4Du,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
84 {HB_SCRIPT_TAMIL, true, 0x0BCDu,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST},
85 {HB_SCRIPT_TELUGU, true, 0x0C4Du,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_EXPLICIT, BLWF_MODE_POST_ONLY},
86 {HB_SCRIPT_KANNADA, true, 0x0CCDu,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_POST_ONLY},
87 {HB_SCRIPT_MALAYALAM, true, 0x0D4Du,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_LOG_REPHA,BLWF_MODE_PRE_AND_POST},
88 {HB_SCRIPT_SINHALA, false,0x0DCAu,BASE_POS_LAST_SINHALA,
89 REPH_POS_AFTER_POST, REPH_MODE_EXPLICIT, BLWF_MODE_PRE_AND_POST},
98 struct feature_list_t {
100 hb_ot_map_feature_flags_t flags;
103 static const feature_list_t
108 * These features are applied in order, one at a time, after initial_reordering.
110 {HB_TAG('n','u','k','t'), F_GLOBAL},
111 {HB_TAG('a','k','h','n'), F_GLOBAL},
112 {HB_TAG('r','p','h','f'), F_NONE},
113 {HB_TAG('r','k','r','f'), F_GLOBAL},
114 {HB_TAG('p','r','e','f'), F_NONE},
115 {HB_TAG('b','l','w','f'), F_NONE},
116 {HB_TAG('a','b','v','f'), F_NONE},
117 {HB_TAG('h','a','l','f'), F_NONE},
118 {HB_TAG('p','s','t','f'), F_NONE},
119 {HB_TAG('v','a','t','u'), F_GLOBAL},
120 {HB_TAG('c','j','c','t'), F_GLOBAL},
123 * These features are applied all at once, after final_reordering.
124 * Default Bengali font in Windows for example has intermixed
125 * lookups for init,pres,abvs,blws features.
127 {HB_TAG('i','n','i','t'), F_NONE},
128 {HB_TAG('p','r','e','s'), F_GLOBAL},
129 {HB_TAG('a','b','v','s'), F_GLOBAL},
130 {HB_TAG('b','l','w','s'), F_GLOBAL},
131 {HB_TAG('p','s','t','s'), F_GLOBAL},
132 {HB_TAG('h','a','l','n'), F_GLOBAL},
133 /* Positioning features, though we don't care about the types. */
134 {HB_TAG('d','i','s','t'), F_GLOBAL},
135 {HB_TAG('a','b','v','m'), F_GLOBAL},
136 {HB_TAG('b','l','w','m'), F_GLOBAL},
140 * Must be in the same order as the indic_features array.
166 INDIC_BASIC_FEATURES = INIT /* Don't forget to update this! */
170 setup_syllables (const hb_ot_shape_plan_t *plan,
172 hb_buffer_t *buffer);
174 initial_reordering (const hb_ot_shape_plan_t *plan,
176 hb_buffer_t *buffer);
178 final_reordering (const hb_ot_shape_plan_t *plan,
180 hb_buffer_t *buffer);
182 clear_syllables (const hb_ot_shape_plan_t *plan,
184 hb_buffer_t *buffer);
187 collect_features_indic (hb_ot_shape_planner_t *plan)
189 hb_ot_map_builder_t *map = &plan->map;
191 /* Do this before any lookups have been applied. */
192 map->add_gsub_pause (setup_syllables);
194 map->add_global_bool_feature (HB_TAG('l','o','c','l'));
195 /* The Indic specs do not require ccmp, but we apply it here since if
196 * there is a use of it, it's typically at the beginning. */
197 map->add_global_bool_feature (HB_TAG('c','c','m','p'));
201 map->add_gsub_pause (initial_reordering);
202 for (; i < INDIC_BASIC_FEATURES; i++) {
203 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ | F_MANUAL_ZWNJ);
204 map->add_gsub_pause (nullptr);
206 map->add_gsub_pause (final_reordering);
207 for (; i < INDIC_NUM_FEATURES; i++) {
208 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ | F_MANUAL_ZWNJ);
211 map->add_global_bool_feature (HB_TAG('c','a','l','t'));
212 map->add_global_bool_feature (HB_TAG('c','l','i','g'));
214 map->add_gsub_pause (clear_syllables);
218 override_features_indic (hb_ot_shape_planner_t *plan)
220 plan->map.add_feature (HB_TAG('l','i','g','a'), 0, F_GLOBAL);
224 struct would_substitute_feature_t
226 inline void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_)
228 zero_context = zero_context_;
229 map->get_stage_lookups (0/*GSUB*/,
230 map->get_feature_stage (0/*GSUB*/, feature_tag),
234 inline bool would_substitute (const hb_codepoint_t *glyphs,
235 unsigned int glyphs_count,
236 hb_face_t *face) const
238 for (unsigned int i = 0; i < count; i++)
239 if (hb_ot_layout_lookup_would_substitute_fast (face, lookups[i].index, glyphs, glyphs_count, zero_context))
245 const hb_ot_map_t::lookup_map_t *lookups;
250 struct indic_shape_plan_t
254 inline bool get_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const
256 hb_codepoint_t glyph = virama_glyph;
257 if (unlikely (virama_glyph == (hb_codepoint_t) -1))
259 if (!config->virama || !font->get_nominal_glyph (config->virama, &glyph))
261 /* Technically speaking, the spec says we should apply 'locl' to virama too.
262 * Maybe one day... */
264 /* Our get_nominal_glyph() function needs a font, so we can't get the virama glyph
265 * during shape planning... Instead, overwrite it here. It's safe. Don't worry! */
266 virama_glyph = glyph;
273 const indic_config_t *config;
276 mutable hb_codepoint_t virama_glyph;
278 would_substitute_feature_t rphf;
279 would_substitute_feature_t pref;
280 would_substitute_feature_t blwf;
281 would_substitute_feature_t pstf;
283 hb_mask_t mask_array[INDIC_NUM_FEATURES];
287 data_create_indic (const hb_ot_shape_plan_t *plan)
289 indic_shape_plan_t *indic_plan = (indic_shape_plan_t *) calloc (1, sizeof (indic_shape_plan_t));
290 if (unlikely (!indic_plan))
293 indic_plan->config = &indic_configs[0];
294 for (unsigned int i = 1; i < ARRAY_LENGTH (indic_configs); i++)
295 if (plan->props.script == indic_configs[i].script) {
296 indic_plan->config = &indic_configs[i];
300 indic_plan->is_old_spec = indic_plan->config->has_old_spec && ((plan->map.chosen_script[0] & 0x000000FFu) != '2');
301 indic_plan->virama_glyph = (hb_codepoint_t) -1;
303 /* Use zero-context would_substitute() matching for new-spec of the main
304 * Indic scripts, and scripts with one spec only, but not for old-specs.
305 * The new-spec for all dual-spec scripts says zero-context matching happens.
307 * However, testing with Malayalam shows that old and new spec both allow
308 * context. Testing with Bengali new-spec however shows that it doesn't.
309 * So, the heuristic here is the way it is. It should *only* be changed,
310 * as we discover more cases of what Windows does. DON'T TOUCH OTHERWISE.
312 bool zero_context = !indic_plan->is_old_spec && plan->props.script != HB_SCRIPT_MALAYALAM;
313 indic_plan->rphf.init (&plan->map, HB_TAG('r','p','h','f'), zero_context);
314 indic_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), zero_context);
315 indic_plan->blwf.init (&plan->map, HB_TAG('b','l','w','f'), zero_context);
316 indic_plan->pstf.init (&plan->map, HB_TAG('p','s','t','f'), zero_context);
318 for (unsigned int i = 0; i < ARRAY_LENGTH (indic_plan->mask_array); i++)
319 indic_plan->mask_array[i] = (indic_features[i].flags & F_GLOBAL) ?
320 0 : plan->map.get_1_mask (indic_features[i].tag);
326 data_destroy_indic (void *data)
331 static indic_position_t
332 consonant_position_from_face (const indic_shape_plan_t *indic_plan,
333 const hb_codepoint_t consonant,
334 const hb_codepoint_t virama,
337 /* For old-spec, the order of glyphs is Consonant,Virama,
338 * whereas for new-spec, it's Virama,Consonant. However,
339 * some broken fonts (like Free Sans) simply copied lookups
340 * from old-spec to new-spec without modification.
341 * And oddly enough, Uniscribe seems to respect those lookups.
342 * Eg. in the sequence U+0924,U+094D,U+0930, Uniscribe finds
343 * base at 0. The font however, only has lookups matching
344 * 930,94D in 'blwf', not the expected 94D,930 (with new-spec
345 * table). As such, we simply match both sequences. Seems
347 hb_codepoint_t glyphs[3] = {virama, consonant, virama};
348 if (indic_plan->blwf.would_substitute (glyphs , 2, face) ||
349 indic_plan->blwf.would_substitute (glyphs+1, 2, face))
351 if (indic_plan->pstf.would_substitute (glyphs , 2, face) ||
352 indic_plan->pstf.would_substitute (glyphs+1, 2, face))
354 if (indic_plan->pref.would_substitute (glyphs , 2, face) ||
355 indic_plan->pref.would_substitute (glyphs+1, 2, face))
361 enum syllable_type_t {
370 #include "hb-ot-shape-complex-indic-machine.hh"
374 setup_masks_indic (const hb_ot_shape_plan_t *plan HB_UNUSED,
376 hb_font_t *font HB_UNUSED)
378 HB_BUFFER_ALLOCATE_VAR (buffer, indic_category);
379 HB_BUFFER_ALLOCATE_VAR (buffer, indic_position);
381 /* We cannot setup masks here. We save information about characters
382 * and setup masks later on in a pause-callback. */
384 unsigned int count = buffer->len;
385 hb_glyph_info_t *info = buffer->info;
386 for (unsigned int i = 0; i < count; i++)
387 set_indic_properties (info[i]);
391 setup_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
392 hb_font_t *font HB_UNUSED,
395 find_syllables (buffer);
396 foreach_syllable (buffer, start, end)
397 buffer->unsafe_to_break (start, end);
401 compare_indic_order (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb)
403 int a = pa->indic_position();
404 int b = pb->indic_position();
406 return a < b ? -1 : a == b ? 0 : +1;
412 update_consonant_positions (const hb_ot_shape_plan_t *plan,
416 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
418 if (indic_plan->config->base_pos != BASE_POS_LAST)
421 hb_codepoint_t virama;
422 if (indic_plan->get_virama_glyph (font, &virama))
424 hb_face_t *face = font->face;
425 unsigned int count = buffer->len;
426 hb_glyph_info_t *info = buffer->info;
427 for (unsigned int i = 0; i < count; i++)
428 if (info[i].indic_position() == POS_BASE_C)
430 hb_codepoint_t consonant = info[i].codepoint;
431 info[i].indic_position() = consonant_position_from_face (indic_plan, consonant, virama, face);
438 * https://docs.microsqoft.com/en-us/typography/script-development/devanagari */
441 initial_reordering_consonant_syllable (const hb_ot_shape_plan_t *plan,
444 unsigned int start, unsigned int end)
446 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
447 hb_glyph_info_t *info = buffer->info;
449 /* https://github.com/harfbuzz/harfbuzz/issues/435#issuecomment-335560167
450 * // For compatibility with legacy usage in Kannada,
451 * // Ra+h+ZWJ must behave like Ra+ZWJ+h...
453 if (buffer->props.script == HB_SCRIPT_KANNADA &&
455 is_one_of (info[start ], FLAG (OT_Ra)) &&
456 is_one_of (info[start+1], FLAG (OT_H)) &&
457 is_one_of (info[start+2], FLAG (OT_ZWJ)))
459 buffer->merge_clusters (start+1, start+3);
460 hb_glyph_info_t tmp = info[start+1];
461 info[start+1] = info[start+2];
465 /* 1. Find base consonant:
467 * The shaping engine finds the base consonant of the syllable, using the
468 * following algorithm: starting from the end of the syllable, move backwards
469 * until a consonant is found that does not have a below-base or post-base
470 * form (post-base forms have to follow below-base forms), or that is not a
471 * pre-base-reordering Ra, or arrive at the first consonant. The consonant
472 * stopped at will be the base.
474 * o If the syllable starts with Ra + Halant (in a script that has Reph)
475 * and has more than one consonant, Ra is excluded from candidates for
479 unsigned int base = end;
480 bool has_reph = false;
483 /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
484 * and has more than one consonant, Ra is excluded from candidates for
485 * base consonants. */
486 unsigned int limit = start;
487 if (indic_plan->mask_array[RPHF] &&
490 (indic_plan->config->reph_mode == REPH_MODE_IMPLICIT && !is_joiner (info[start + 2])) ||
491 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT && info[start + 2].indic_category() == OT_ZWJ)
494 /* See if it matches the 'rphf' feature. */
495 hb_codepoint_t glyphs[3] = {info[start].codepoint,
496 info[start + 1].codepoint,
497 indic_plan->config->reph_mode == REPH_MODE_EXPLICIT ?
498 info[start + 2].codepoint : 0};
499 if (indic_plan->rphf.would_substitute (glyphs, 2, face) ||
500 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT &&
501 indic_plan->rphf.would_substitute (glyphs, 3, face)))
504 while (limit < end && is_joiner (info[limit]))
509 } else if (indic_plan->config->reph_mode == REPH_MODE_LOG_REPHA && info[start].indic_category() == OT_Repha)
512 while (limit < end && is_joiner (info[limit]))
518 switch (indic_plan->config->base_pos)
522 /* -> starting from the end of the syllable, move backwards */
523 unsigned int i = end;
524 bool seen_below = false;
527 /* -> until a consonant is found */
528 if (is_consonant (info[i]))
530 /* -> that does not have a below-base or post-base form
531 * (post-base forms have to follow below-base forms), */
532 if (info[i].indic_position() != POS_BELOW_C &&
533 (info[i].indic_position() != POS_POST_C || seen_below))
538 if (info[i].indic_position() == POS_BELOW_C)
541 /* -> or that is not a pre-base-reordering Ra,
543 * IMPLEMENTATION NOTES:
545 * Our pre-base-reordering Ra's are marked POS_POST_C, so will be skipped
546 * by the logic above already.
549 /* -> or arrive at the first consonant. The consonant stopped at will
555 /* A ZWJ after a Halant stops the base search, and requests an explicit
557 * A ZWJ before a Halant, requests a subjoined form instead, and hence
558 * search continues. This is particularly important for Bengali
559 * sequence Ra,H,Ya that should form Ya-Phalaa by subjoining Ya. */
561 info[i].indic_category() == OT_ZWJ &&
562 info[i - 1].indic_category() == OT_H)
569 case BASE_POS_LAST_SINHALA:
571 /* Sinhala base positioning is slightly different from main Indic, in that:
572 * 1. Its ZWJ behavior is different,
573 * 2. We don't need to look into the font for consonant positions.
579 /* Find the last base consonant that is not blocked by ZWJ. If there is
580 * a ZWJ right before a base consonant, that would request a subjoined form. */
581 for (unsigned int i = limit; i < end; i++)
582 if (is_consonant (info[i]))
584 if (limit < i && info[i - 1].indic_category() == OT_ZWJ)
590 /* Mark all subsequent consonants as below. */
591 for (unsigned int i = base + 1; i < end; i++)
592 if (is_consonant (info[i]))
593 info[i].indic_position() = POS_BELOW_C;
598 /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
599 * and has more than one consonant, Ra is excluded from candidates for
602 * Only do this for unforced Reph. (ie. not for Ra,H,ZWJ. */
603 if (has_reph && base == start && limit - base <= 2) {
604 /* Have no other consonant, so Reph is not formed and Ra becomes base. */
610 /* 2. Decompose and reorder Matras:
612 * Each matra and any syllable modifier sign in the syllable are moved to the
613 * appropriate position relative to the consonant(s) in the syllable. The
614 * shaping engine decomposes two- or three-part matras into their constituent
615 * parts before any repositioning. Matra characters are classified by which
616 * consonant in a conjunct they have affinity for and are reordered to the
617 * following positions:
619 * o Before first half form in the syllable
620 * o After subjoined consonants
621 * o After post-form consonant
622 * o After main consonant (for above marks)
624 * IMPLEMENTATION NOTES:
626 * The normalize() routine has already decomposed matras for us, so we don't
627 * need to worry about that.
631 /* 3. Reorder marks to canonical order:
633 * Adjacent nukta and halant or nukta and vedic sign are always repositioned
634 * if necessary, so that the nukta is first.
636 * IMPLEMENTATION NOTES:
638 * We don't need to do this: the normalize() routine already did this for us.
642 /* Reorder characters */
644 for (unsigned int i = start; i < base; i++)
645 info[i].indic_position() = MIN (POS_PRE_C, (indic_position_t) info[i].indic_position());
648 info[base].indic_position() = POS_BASE_C;
650 /* Mark final consonants. A final consonant is one appearing after a matra.
651 * Happens in Sinhala. */
652 for (unsigned int i = base + 1; i < end; i++)
653 if (info[i].indic_category() == OT_M) {
654 for (unsigned int j = i + 1; j < end; j++)
655 if (is_consonant (info[j])) {
656 info[j].indic_position() = POS_FINAL_C;
662 /* Handle beginning Ra */
664 info[start].indic_position() = POS_RA_TO_BECOME_REPH;
666 /* For old-style Indic script tags, move the first post-base Halant after
669 * Reports suggest that in some scripts Uniscribe does this only if there
670 * is *not* a Halant after last consonant already (eg. Kannada), while it
671 * does it unconditionally in other scripts (eg. Malayalam). We don't
672 * currently know about other scripts, so we single out Malayalam for now.
675 * U+0C9A,U+0CCD,U+0C9A,U+0CCD
676 * With some versions of Lohit Kannada.
677 * https://bugs.freedesktop.org/show_bug.cgi?id=59118
679 * Malayalam test case:
680 * U+0D38,U+0D4D,U+0D31,U+0D4D,U+0D31,U+0D4D
681 * With lohit-ttf-20121122/Lohit-Malayalam.ttf
683 if (indic_plan->is_old_spec)
685 bool disallow_double_halants = buffer->props.script != HB_SCRIPT_MALAYALAM;
686 for (unsigned int i = base + 1; i < end; i++)
687 if (info[i].indic_category() == OT_H)
690 for (j = end - 1; j > i; j--)
691 if (is_consonant (info[j]) ||
692 (disallow_double_halants && info[j].indic_category() == OT_H))
694 if (info[j].indic_category() != OT_H && j > i) {
695 /* Move Halant to after last consonant. */
696 hb_glyph_info_t t = info[i];
697 memmove (&info[i], &info[i + 1], (j - i) * sizeof (info[0]));
704 /* Attach misc marks to previous char to move with them. */
706 indic_position_t last_pos = POS_START;
707 for (unsigned int i = start; i < end; i++)
709 if ((FLAG_UNSAFE (info[i].indic_category()) & (JOINER_FLAGS | FLAG (OT_N) | FLAG (OT_RS) | MEDIAL_FLAGS | FLAG (OT_H))))
711 info[i].indic_position() = last_pos;
712 if (unlikely (info[i].indic_category() == OT_H &&
713 info[i].indic_position() == POS_PRE_M))
716 * Uniscribe doesn't move the Halant with Left Matra.
717 * TEST: U+092B,U+093F,U+094DE
718 * We follow. This is important for the Sinhala
719 * U+0DDA split matra since it decomposes to U+0DD9,U+0DCA
720 * where U+0DD9 is a left matra and U+0DCA is the virama.
721 * We don't want to move the virama with the left matra.
722 * TEST: U+0D9A,U+0DDA
724 for (unsigned int j = i; j > start; j--)
725 if (info[j - 1].indic_position() != POS_PRE_M) {
726 info[i].indic_position() = info[j - 1].indic_position();
730 } else if (info[i].indic_position() != POS_SMVD) {
731 last_pos = (indic_position_t) info[i].indic_position();
735 /* For post-base consonants let them own anything before them
736 * since the last consonant or matra. */
738 unsigned int last = base;
739 for (unsigned int i = base + 1; i < end; i++)
740 if (is_consonant (info[i]))
742 for (unsigned int j = last + 1; j < i; j++)
743 if (info[j].indic_position() < POS_SMVD)
744 info[j].indic_position() = info[i].indic_position();
746 } else if (info[i].indic_category() == OT_M)
752 /* Use syllable() for sort accounting temporarily. */
753 unsigned int syllable = info[start].syllable();
754 for (unsigned int i = start; i < end; i++)
755 info[i].syllable() = i - start;
757 /* Sit tight, rock 'n roll! */
758 hb_stable_sort (info + start, end - start, compare_indic_order);
759 /* Find base again */
761 for (unsigned int i = start; i < end; i++)
762 if (info[i].indic_position() == POS_BASE_C)
767 /* Things are out-of-control for post base positions, they may shuffle
768 * around like crazy. In old-spec mode, we move halants around, so in
769 * that case merge all clusters after base. Otherwise, check the sort
770 * order and merge as needed.
771 * For pre-base stuff, we handle cluster issues in final reordering.
773 * We could use buffer->sort() for this, if there was no special
774 * reordering of pre-base stuff happening later...
776 if (indic_plan->is_old_spec || end - base > 127)
777 buffer->merge_clusters (base, end);
780 /* Note! syllable() is a one-byte field. */
781 for (unsigned int i = base; i < end; i++)
782 if (info[i].syllable() != 255)
784 unsigned int max = i;
785 unsigned int j = start + info[i].syllable();
789 unsigned int next = start + info[j].syllable();
790 info[j].syllable() = 255; /* So we don't process j later again. */
794 buffer->merge_clusters (i, max + 1);
798 /* Put syllable back in. */
799 for (unsigned int i = start; i < end; i++)
800 info[i].syllable() = syllable;
803 /* Setup masks now */
809 for (unsigned int i = start; i < end && info[i].indic_position() == POS_RA_TO_BECOME_REPH; i++)
810 info[i].mask |= indic_plan->mask_array[RPHF];
813 mask = indic_plan->mask_array[HALF];
814 if (!indic_plan->is_old_spec &&
815 indic_plan->config->blwf_mode == BLWF_MODE_PRE_AND_POST)
816 mask |= indic_plan->mask_array[BLWF];
817 for (unsigned int i = start; i < base; i++)
818 info[i].mask |= mask;
822 info[base].mask |= mask;
824 mask = indic_plan->mask_array[BLWF] | indic_plan->mask_array[ABVF] | indic_plan->mask_array[PSTF];
825 for (unsigned int i = base + 1; i < end; i++)
826 info[i].mask |= mask;
829 if (indic_plan->is_old_spec &&
830 buffer->props.script == HB_SCRIPT_DEVANAGARI)
832 /* Old-spec eye-lash Ra needs special handling. From the
835 * "The feature 'below-base form' is applied to consonants
836 * having below-base forms and following the base consonant.
837 * The exception is vattu, which may appear below half forms
838 * as well as below the base glyph. The feature 'below-base
839 * form' will be applied to all such occurrences of Ra as well."
841 * Test case: U+0924,U+094D,U+0930,U+094d,U+0915
842 * with Sanskrit 2003 font.
844 * However, note that Ra,Halant,ZWJ is the correct way to
845 * request eyelash form of Ra, so we wouldbn't inhibit it
848 * Test case: U+0924,U+094D,U+0930,U+094d,U+200D,U+0915
850 for (unsigned int i = start; i + 1 < base; i++)
851 if (info[i ].indic_category() == OT_Ra &&
852 info[i+1].indic_category() == OT_H &&
854 info[i+2].indic_category() != OT_ZWJ))
856 info[i ].mask |= indic_plan->mask_array[BLWF];
857 info[i+1].mask |= indic_plan->mask_array[BLWF];
861 unsigned int pref_len = 2;
862 if (indic_plan->mask_array[PREF] && base + pref_len < end)
864 /* Find a Halant,Ra sequence and mark it for pre-base-reordering processing. */
865 for (unsigned int i = base + 1; i + pref_len - 1 < end; i++) {
866 hb_codepoint_t glyphs[2];
867 for (unsigned int j = 0; j < pref_len; j++)
868 glyphs[j] = info[i + j].codepoint;
869 if (indic_plan->pref.would_substitute (glyphs, pref_len, face))
871 for (unsigned int j = 0; j < pref_len; j++)
872 info[i++].mask |= indic_plan->mask_array[PREF];
878 /* Apply ZWJ/ZWNJ effects */
879 for (unsigned int i = start + 1; i < end; i++)
880 if (is_joiner (info[i])) {
881 bool non_joiner = info[i].indic_category() == OT_ZWNJ;
887 /* ZWJ/ZWNJ should disable CJCT. They do that by simply
888 * being there, since we don't skip them for the CJCT
889 * feature (ie. F_MANUAL_ZWJ) */
891 /* A ZWNJ disables HALF. */
893 info[j].mask &= ~indic_plan->mask_array[HALF];
895 } while (j > start && !is_consonant (info[j]));
900 initial_reordering_standalone_cluster (const hb_ot_shape_plan_t *plan,
903 unsigned int start, unsigned int end)
905 /* We treat placeholder/dotted-circle as if they are consonants, so we
906 * should just chain. Only if not in compatibility mode that is... */
908 if (hb_options ().uniscribe_bug_compatible)
910 /* For dotted-circle, this is what Uniscribe does:
911 * If dotted-circle is the last glyph, it just does nothing.
912 * Ie. It doesn't form Reph. */
913 if (buffer->info[end - 1].indic_category() == OT_DOTTEDCIRCLE)
917 initial_reordering_consonant_syllable (plan, face, buffer, start, end);
921 initial_reordering_syllable (const hb_ot_shape_plan_t *plan,
924 unsigned int start, unsigned int end)
926 syllable_type_t syllable_type = (syllable_type_t) (buffer->info[start].syllable() & 0x0F);
927 switch (syllable_type)
929 case vowel_syllable: /* We made the vowels look like consonants. So let's call the consonant logic! */
930 case consonant_syllable:
931 initial_reordering_consonant_syllable (plan, face, buffer, start, end);
934 case broken_cluster: /* We already inserted dotted-circles, so just call the standalone_cluster. */
935 case standalone_cluster:
936 initial_reordering_standalone_cluster (plan, face, buffer, start, end);
940 case non_indic_cluster:
946 insert_dotted_circles (const hb_ot_shape_plan_t *plan HB_UNUSED,
950 /* Note: This loop is extra overhead, but should not be measurable. */
951 bool has_broken_syllables = false;
952 unsigned int count = buffer->len;
953 hb_glyph_info_t *info = buffer->info;
954 for (unsigned int i = 0; i < count; i++)
955 if ((info[i].syllable() & 0x0F) == broken_cluster)
957 has_broken_syllables = true;
960 if (likely (!has_broken_syllables))
964 hb_codepoint_t dottedcircle_glyph;
965 if (!font->get_nominal_glyph (0x25CCu, &dottedcircle_glyph))
968 hb_glyph_info_t dottedcircle = {0};
969 dottedcircle.codepoint = 0x25CCu;
970 set_indic_properties (dottedcircle);
971 dottedcircle.codepoint = dottedcircle_glyph;
973 buffer->clear_output ();
976 unsigned int last_syllable = 0;
977 while (buffer->idx < buffer->len && buffer->successful)
979 unsigned int syllable = buffer->cur().syllable();
980 syllable_type_t syllable_type = (syllable_type_t) (syllable & 0x0F);
981 if (unlikely (last_syllable != syllable && syllable_type == broken_cluster))
983 last_syllable = syllable;
985 hb_glyph_info_t ginfo = dottedcircle;
986 ginfo.cluster = buffer->cur().cluster;
987 ginfo.mask = buffer->cur().mask;
988 ginfo.syllable() = buffer->cur().syllable();
989 /* TODO Set glyph_props? */
991 /* Insert dottedcircle after possible Repha. */
992 while (buffer->idx < buffer->len && buffer->successful &&
993 last_syllable == buffer->cur().syllable() &&
994 buffer->cur().indic_category() == OT_Repha)
995 buffer->next_glyph ();
997 buffer->output_info (ginfo);
1000 buffer->next_glyph ();
1003 buffer->swap_buffers ();
1007 initial_reordering (const hb_ot_shape_plan_t *plan,
1009 hb_buffer_t *buffer)
1011 update_consonant_positions (plan, font, buffer);
1012 insert_dotted_circles (plan, font, buffer);
1014 foreach_syllable (buffer, start, end)
1015 initial_reordering_syllable (plan, font->face, buffer, start, end);
1019 final_reordering_syllable (const hb_ot_shape_plan_t *plan,
1020 hb_buffer_t *buffer,
1021 unsigned int start, unsigned int end)
1023 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
1024 hb_glyph_info_t *info = buffer->info;
1027 /* This function relies heavily on halant glyphs. Lots of ligation
1028 * and possibly multiple substitutions happened prior to this
1029 * phase, and that might have messed up our properties. Recover
1030 * from a particular case of that where we're fairly sure that a
1031 * class of OT_H is desired but has been lost. */
1032 if (indic_plan->virama_glyph)
1034 unsigned int virama_glyph = indic_plan->virama_glyph;
1035 for (unsigned int i = start; i < end; i++)
1036 if (info[i].codepoint == virama_glyph &&
1037 _hb_glyph_info_ligated (&info[i]) &&
1038 _hb_glyph_info_multiplied (&info[i]))
1040 /* This will make sure that this glyph passes is_halant() test. */
1041 info[i].indic_category() = OT_H;
1042 _hb_glyph_info_clear_ligated_and_multiplied (&info[i]);
1047 /* 4. Final reordering:
1049 * After the localized forms and basic shaping forms GSUB features have been
1050 * applied (see below), the shaping engine performs some final glyph
1051 * reordering before applying all the remaining font features to the entire
1055 bool try_pref = !!indic_plan->mask_array[PREF];
1057 /* Find base again */
1059 for (base = start; base < end; base++)
1060 if (info[base].indic_position() >= POS_BASE_C)
1062 if (try_pref && base + 1 < end)
1064 for (unsigned int i = base + 1; i < end; i++)
1065 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0)
1067 if (!(_hb_glyph_info_substituted (&info[i]) &&
1068 _hb_glyph_info_ligated_and_didnt_multiply (&info[i])))
1070 /* Ok, this was a 'pref' candidate but didn't form any.
1071 * Base is around here... */
1073 while (base < end && is_halant (info[base]))
1075 info[base].indic_position() = POS_BASE_C;
1082 /* For Malayalam, skip over unformed below- (but NOT post-) forms. */
1083 if (buffer->props.script == HB_SCRIPT_MALAYALAM)
1085 for (unsigned int i = base + 1; i < end; i++)
1087 while (i < end && is_joiner (info[i]))
1089 if (i == end || !is_halant (info[i]))
1091 i++; /* Skip halant. */
1092 while (i < end && is_joiner (info[i]))
1094 if (i < end && is_consonant (info[i]) && info[i].indic_position() == POS_BELOW_C)
1097 info[base].indic_position() = POS_BASE_C;
1102 if (start < base && info[base].indic_position() > POS_BASE_C)
1106 if (base == end && start < base &&
1107 is_one_of (info[base - 1], FLAG (OT_ZWJ)))
1110 while (start < base &&
1111 is_one_of (info[base], (FLAG (OT_N) | FLAG (OT_H))))
1115 /* o Reorder matras:
1117 * If a pre-base matra character had been reordered before applying basic
1118 * features, the glyph can be moved closer to the main consonant based on
1119 * whether half-forms had been formed. Actual position for the matra is
1120 * defined as “after last standalone halant glyph, after initial matra
1121 * position and before the main consonant”. If ZWJ or ZWNJ follow this
1122 * halant, position is moved after it.
1125 if (start + 1 < end && start < base) /* Otherwise there can't be any pre-base matra characters. */
1127 /* If we lost track of base, alas, position before last thingy. */
1128 unsigned int new_pos = base == end ? base - 2 : base - 1;
1130 /* Malayalam / Tamil do not have "half" forms or explicit virama forms.
1131 * The glyphs formed by 'half' are Chillus or ligated explicit viramas.
1132 * We want to position matra after them.
1134 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL)
1136 while (new_pos > start &&
1137 !(is_one_of (info[new_pos], (FLAG (OT_M) | FLAG (OT_H)))))
1140 /* If we found no Halant we are done.
1141 * Otherwise only proceed if the Halant does
1142 * not belong to the Matra itself! */
1143 if (is_halant (info[new_pos]) &&
1144 info[new_pos].indic_position() != POS_PRE_M)
1146 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */
1147 if (new_pos + 1 < end && is_joiner (info[new_pos + 1]))
1151 new_pos = start; /* No move. */
1154 if (start < new_pos && info[new_pos].indic_position () != POS_PRE_M)
1156 /* Now go see if there's actually any matras... */
1157 for (unsigned int i = new_pos; i > start; i--)
1158 if (info[i - 1].indic_position () == POS_PRE_M)
1160 unsigned int old_pos = i - 1;
1161 if (old_pos < base && base <= new_pos) /* Shouldn't actually happen. */
1164 hb_glyph_info_t tmp = info[old_pos];
1165 memmove (&info[old_pos], &info[old_pos + 1], (new_pos - old_pos) * sizeof (info[0]));
1166 info[new_pos] = tmp;
1168 /* Note: this merge_clusters() is intentionally *after* the reordering.
1169 * Indic matra reordering is special and tricky... */
1170 buffer->merge_clusters (new_pos, MIN (end, base + 1));
1175 for (unsigned int i = start; i < base; i++)
1176 if (info[i].indic_position () == POS_PRE_M) {
1177 buffer->merge_clusters (i, MIN (end, base + 1));
1186 * Reph’s original position is always at the beginning of the syllable,
1187 * (i.e. it is not reordered at the character reordering stage). However,
1188 * it will be reordered according to the basic-forms shaping results.
1189 * Possible positions for reph, depending on the script, are; after main,
1190 * before post-base consonant forms, and after post-base consonant forms.
1195 * - If repha is encoded as a sequence of characters (Ra,H or Ra,H,ZWJ), then
1196 * we should only move it if the sequence ligated to the repha form.
1198 * - If repha is encoded separately and in the logical position, we should only
1199 * move it if it did NOT ligate. If it ligated, it's probably the font trying
1200 * to make it work without the reordering.
1202 if (start + 1 < end &&
1203 info[start].indic_position() == POS_RA_TO_BECOME_REPH &&
1204 ((info[start].indic_category() == OT_Repha) ^
1205 _hb_glyph_info_ligated_and_didnt_multiply (&info[start])))
1207 unsigned int new_reph_pos;
1208 reph_position_t reph_pos = indic_plan->config->reph_pos;
1210 /* 1. If reph should be positioned after post-base consonant forms,
1211 * proceed to step 5.
1213 if (reph_pos == REPH_POS_AFTER_POST)
1218 /* 2. If the reph repositioning class is not after post-base: target
1219 * position is after the first explicit halant glyph between the
1220 * first post-reph consonant and last main consonant. If ZWJ or ZWNJ
1221 * are following this halant, position is moved after it. If such
1222 * position is found, this is the target position. Otherwise,
1223 * proceed to the next step.
1225 * Note: in old-implementation fonts, where classifications were
1226 * fixed in shaping engine, there was no case where reph position
1227 * will be found on this step.
1230 new_reph_pos = start + 1;
1231 while (new_reph_pos < base && !is_halant (info[new_reph_pos]))
1234 if (new_reph_pos < base && is_halant (info[new_reph_pos]))
1236 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */
1237 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1]))
1243 /* 3. If reph should be repositioned after the main consonant: find the
1244 * first consonant not ligated with main, or find the first
1245 * consonant that is not a potential pre-base-reordering Ra.
1247 if (reph_pos == REPH_POS_AFTER_MAIN)
1249 new_reph_pos = base;
1250 while (new_reph_pos + 1 < end && info[new_reph_pos + 1].indic_position() <= POS_AFTER_MAIN)
1252 if (new_reph_pos < end)
1256 /* 4. If reph should be positioned before post-base consonant, find
1257 * first post-base classified consonant not ligated with main. If no
1258 * consonant is found, the target position should be before the
1259 * first matra, syllable modifier sign or vedic sign.
1261 /* This is our take on what step 4 is trying to say (and failing, BADLY). */
1262 if (reph_pos == REPH_POS_AFTER_SUB)
1264 new_reph_pos = base;
1265 while (new_reph_pos + 1 < end &&
1266 !( FLAG_UNSAFE (info[new_reph_pos + 1].indic_position()) & (FLAG (POS_POST_C) | FLAG (POS_AFTER_POST) | FLAG (POS_SMVD))))
1268 if (new_reph_pos < end)
1272 /* 5. If no consonant is found in steps 3 or 4, move reph to a position
1273 * immediately before the first post-base matra, syllable modifier
1274 * sign or vedic sign that has a reordering class after the intended
1275 * reph position. For example, if the reordering position for reph
1276 * is post-main, it will skip above-base matras that also have a
1277 * post-main position.
1281 /* Copied from step 2. */
1282 new_reph_pos = start + 1;
1283 while (new_reph_pos < base && !is_halant (info[new_reph_pos]))
1286 if (new_reph_pos < base && is_halant (info[new_reph_pos]))
1288 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */
1289 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1]))
1295 /* 6. Otherwise, reorder reph to the end of the syllable.
1298 new_reph_pos = end - 1;
1299 while (new_reph_pos > start && info[new_reph_pos].indic_position() == POS_SMVD)
1303 * If the Reph is to be ending up after a Matra,Halant sequence,
1304 * position it before that Halant so it can interact with the Matra.
1305 * However, if it's a plain Consonant,Halant we shouldn't do that.
1306 * Uniscribe doesn't do this.
1307 * TEST: U+0930,U+094D,U+0915,U+094B,U+094D
1309 if (!hb_options ().uniscribe_bug_compatible &&
1310 unlikely (is_halant (info[new_reph_pos]))) {
1311 for (unsigned int i = base + 1; i < new_reph_pos; i++)
1312 if (info[i].indic_category() == OT_M) {
1323 buffer->merge_clusters (start, new_reph_pos + 1);
1324 hb_glyph_info_t reph = info[start];
1325 memmove (&info[start], &info[start + 1], (new_reph_pos - start) * sizeof (info[0]));
1326 info[new_reph_pos] = reph;
1328 if (start < base && base <= new_reph_pos)
1334 /* o Reorder pre-base-reordering consonants:
1336 * If a pre-base-reordering consonant is found, reorder it according to
1337 * the following rules:
1340 if (try_pref && base + 1 < end) /* Otherwise there can't be any pre-base-reordering Ra. */
1342 for (unsigned int i = base + 1; i < end; i++)
1343 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0)
1345 /* 1. Only reorder a glyph produced by substitution during application
1346 * of the <pref> feature. (Note that a font may shape a Ra consonant with
1347 * the feature generally but block it in certain contexts.)
1349 /* Note: We just check that something got substituted. We don't check that
1350 * the <pref> feature actually did it...
1352 * Reorder pref only if it ligated. */
1353 if (_hb_glyph_info_ligated_and_didnt_multiply (&info[i]))
1356 * 2. Try to find a target position the same way as for pre-base matra.
1357 * If it is found, reorder pre-base consonant glyph.
1359 * 3. If position is not found, reorder immediately before main
1363 unsigned int new_pos = base;
1364 /* Malayalam / Tamil do not have "half" forms or explicit virama forms.
1365 * The glyphs formed by 'half' are Chillus or ligated explicit viramas.
1366 * We want to position matra after them.
1368 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL)
1370 while (new_pos > start &&
1371 !(is_one_of (info[new_pos - 1], FLAG(OT_M) | FLAG (OT_H))))
1375 if (new_pos > start && is_halant (info[new_pos - 1]))
1377 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */
1378 if (new_pos < end && is_joiner (info[new_pos]))
1383 unsigned int old_pos = i;
1385 buffer->merge_clusters (new_pos, old_pos + 1);
1386 hb_glyph_info_t tmp = info[old_pos];
1387 memmove (&info[new_pos + 1], &info[new_pos], (old_pos - new_pos) * sizeof (info[0]));
1388 info[new_pos] = tmp;
1390 if (new_pos <= base && base < old_pos)
1400 /* Apply 'init' to the Left Matra if it's a word start. */
1401 if (info[start].indic_position () == POS_PRE_M)
1404 !(FLAG_UNSAFE (_hb_glyph_info_get_general_category (&info[start - 1])) &
1405 FLAG_RANGE (HB_UNICODE_GENERAL_CATEGORY_FORMAT, HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK)))
1406 info[start].mask |= indic_plan->mask_array[INIT];
1408 buffer->unsafe_to_break (start - 1, start + 1);
1413 * Finish off the clusters and go home!
1415 if (hb_options ().uniscribe_bug_compatible)
1417 switch ((hb_tag_t) plan->props.script)
1419 case HB_SCRIPT_TAMIL:
1420 case HB_SCRIPT_SINHALA:
1424 /* Uniscribe merges the entire syllable into a single cluster... Except for Tamil & Sinhala.
1425 * This means, half forms are submerged into the main consonant's cluster.
1426 * This is unnecessary, and makes cursor positioning harder, but that's what
1427 * Uniscribe does. */
1428 buffer->merge_clusters (start, end);
1436 final_reordering (const hb_ot_shape_plan_t *plan,
1437 hb_font_t *font HB_UNUSED,
1438 hb_buffer_t *buffer)
1440 unsigned int count = buffer->len;
1441 if (unlikely (!count)) return;
1443 foreach_syllable (buffer, start, end)
1444 final_reordering_syllable (plan, buffer, start, end);
1446 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_category);
1447 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_position);
1452 clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
1453 hb_font_t *font HB_UNUSED,
1454 hb_buffer_t *buffer)
1456 hb_glyph_info_t *info = buffer->info;
1457 unsigned int count = buffer->len;
1458 for (unsigned int i = 0; i < count; i++)
1459 info[i].syllable() = 0;
1464 decompose_indic (const hb_ot_shape_normalize_context_t *c,
1471 /* Don't decompose these. */
1472 case 0x0931u : return false; /* DEVANAGARI LETTER RRA */
1473 // https://github.com/harfbuzz/harfbuzz/issues/779
1474 case 0x09DCu : return false; /* BENGALI LETTER RRA */
1475 case 0x09DDu : return false; /* BENGALI LETTER RHA */
1476 case 0x0B94u : return false; /* TAMIL LETTER AU */
1480 * Decompose split matras that don't have Unicode decompositions.
1485 /* This one has no decomposition in Unicode, but needs no decomposition either. */
1486 /* case 0x0AC9u : return false; */
1489 case 0x0B57u : *a = no decomp, -> RIGHT; return true;
1493 if ((ab == 0x0DDAu || hb_in_range<hb_codepoint_t> (ab, 0x0DDCu, 0x0DDEu)))
1496 * Sinhala split matras... Let the fun begin.
1498 * These four characters have Unicode decompositions. However, Uniscribe
1499 * decomposes them "Khmer-style", that is, it uses the character itself to
1500 * get the second half. The first half of all four decompositions is always
1503 * Now, there are buggy fonts, namely, the widely used lklug.ttf, that are
1504 * broken with Uniscribe. But we need to support them. As such, we only
1505 * do the Uniscribe-style decomposition if the character is transformed into
1506 * its "sec.half" form by the 'pstf' feature. Otherwise, we fall back to
1507 * Unicode decomposition.
1509 * Note that we can't unconditionally use Unicode decomposition. That would
1510 * break some other fonts, that are designed to work with Uniscribe, and
1511 * don't have positioning features for the Unicode-style decomposition.
1515 * The Uniscribe behavior is now documented in the newly published Sinhala
1518 * https://docs.microsoft.com/en-us/typography/script-development/sinhala#shaping
1521 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) c->plan->data;
1523 hb_codepoint_t glyph;
1525 if (hb_options ().uniscribe_bug_compatible ||
1526 (c->font->get_nominal_glyph (ab, &glyph) &&
1527 indic_plan->pstf.would_substitute (&glyph, 1, c->font->face)))
1529 /* Ok, safe to use Uniscribe-style decomposition. */
1536 return (bool) c->unicode->decompose (ab, a, b);
1540 compose_indic (const hb_ot_shape_normalize_context_t *c,
1545 /* Avoid recomposing split matras. */
1546 if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a)))
1549 /* Composition-exclusion exceptions that we want to recompose. */
1550 if (a == 0x09AFu && b == 0x09BCu) { *ab = 0x09DFu; return true; }
1552 return (bool) c->unicode->compose (a, b, ab);
1556 const hb_ot_complex_shaper_t _hb_ot_complex_shaper_indic =
1558 collect_features_indic,
1559 override_features_indic,
1562 nullptr, /* preprocess_text */
1563 nullptr, /* postprocess_glyphs */
1564 HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT,
1568 nullptr, /* disable_otl */
1569 nullptr, /* reorder_marks */
1570 HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE,
1571 false, /* fallback_position */