22 * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS. |
22 * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS. |
23 * |
23 * |
24 * Google Author(s): Behdad Esfahbod |
24 * Google Author(s): Behdad Esfahbod |
25 */ |
25 */ |
26 |
26 |
27 #include "hb-ot-shape-complex-khmer-private.hh" |
27 #include "hb-ot-shape-complex-khmer.hh" |
28 #include "hb-ot-layout-private.hh" |
28 #include "hb-ot-layout.hh" |
29 |
29 |
30 |
30 |
31 /* |
31 /* |
32 * Khmer shaper. |
32 * Khmer shaper. |
33 */ |
33 */ |
34 |
34 |
35 struct feature_list_t { |
35 static const hb_ot_map_feature_t |
36 hb_tag_t tag; |
|
37 hb_ot_map_feature_flags_t flags; |
|
38 }; |
|
39 |
|
40 static const feature_list_t |
|
41 khmer_features[] = |
36 khmer_features[] = |
42 { |
37 { |
43 /* |
38 /* |
44 * Basic features. |
39 * Basic features. |
45 * These features are applied in order, one at a time, after initial_reordering. |
40 * These features are applied in order, one at a time, after reordering. |
46 */ |
41 */ |
47 {HB_TAG('p','r','e','f'), F_NONE}, |
42 {HB_TAG('p','r','e','f'), F_MANUAL_JOINERS}, |
48 {HB_TAG('b','l','w','f'), F_NONE}, |
43 {HB_TAG('b','l','w','f'), F_MANUAL_JOINERS}, |
49 {HB_TAG('a','b','v','f'), F_NONE}, |
44 {HB_TAG('a','b','v','f'), F_MANUAL_JOINERS}, |
50 {HB_TAG('p','s','t','f'), F_NONE}, |
45 {HB_TAG('p','s','t','f'), F_MANUAL_JOINERS}, |
51 {HB_TAG('c','f','a','r'), F_NONE}, |
46 {HB_TAG('c','f','a','r'), F_MANUAL_JOINERS}, |
52 /* |
47 /* |
53 * Other features. |
48 * Other features. |
54 * These features are applied all at once, after final_reordering. |
49 * These features are applied all at once after clearing syllables. |
55 * Default Bengali font in Windows for example has intermixed |
|
56 * lookups for init,pres,abvs,blws features. |
|
57 */ |
50 */ |
58 {HB_TAG('p','r','e','s'), F_GLOBAL}, |
51 {HB_TAG('p','r','e','s'), F_GLOBAL_MANUAL_JOINERS}, |
59 {HB_TAG('a','b','v','s'), F_GLOBAL}, |
52 {HB_TAG('a','b','v','s'), F_GLOBAL_MANUAL_JOINERS}, |
60 {HB_TAG('b','l','w','s'), F_GLOBAL}, |
53 {HB_TAG('b','l','w','s'), F_GLOBAL_MANUAL_JOINERS}, |
61 {HB_TAG('p','s','t','s'), F_GLOBAL}, |
54 {HB_TAG('p','s','t','s'), F_GLOBAL_MANUAL_JOINERS}, |
62 /* Positioning features, though we don't care about the types. */ |
55 /* |
|
56 * Positioning features. |
|
57 * We don't care about the types. |
|
58 */ |
63 {HB_TAG('d','i','s','t'), F_GLOBAL}, |
59 {HB_TAG('d','i','s','t'), F_GLOBAL}, |
64 {HB_TAG('a','b','v','m'), F_GLOBAL}, |
60 {HB_TAG('a','b','v','m'), F_GLOBAL}, |
65 {HB_TAG('b','l','w','m'), F_GLOBAL}, |
61 {HB_TAG('b','l','w','m'), F_GLOBAL}, |
66 }; |
62 }; |
67 |
63 |
109 { |
102 { |
110 hb_ot_map_builder_t *map = &plan->map; |
103 hb_ot_map_builder_t *map = &plan->map; |
111 |
104 |
112 /* Do this before any lookups have been applied. */ |
105 /* Do this before any lookups have been applied. */ |
113 map->add_gsub_pause (setup_syllables); |
106 map->add_gsub_pause (setup_syllables); |
114 |
107 map->add_gsub_pause (reorder); |
115 map->add_global_bool_feature (HB_TAG('l','o','c','l')); |
108 |
116 /* The Indic specs do not require ccmp, but we apply it here since if |
109 /* Testing suggests that Uniscribe does NOT pause between basic |
117 * there is a use of it, it's typically at the beginning. */ |
110 * features. Test with KhmerUI.ttf and the following three |
118 map->add_global_bool_feature (HB_TAG('c','c','m','p')); |
111 * sequences: |
119 |
112 * |
|
113 * U+1789,U+17BC |
|
114 * U+1789,U+17D2,U+1789 |
|
115 * U+1789,U+17D2,U+1789,U+17BC |
|
116 * |
|
117 * https://github.com/harfbuzz/harfbuzz/issues/974 |
|
118 */ |
|
119 map->enable_feature (HB_TAG('l','o','c','l')); |
|
120 map->enable_feature (HB_TAG('c','c','m','p')); |
120 |
121 |
121 unsigned int i = 0; |
122 unsigned int i = 0; |
122 map->add_gsub_pause (initial_reordering); |
123 for (; i < KHMER_BASIC_FEATURES; i++) |
123 for (; i < KHMER_BASIC_FEATURES; i++) { |
124 map->add_feature (khmer_features[i]); |
124 map->add_feature (khmer_features[i].tag, 1, khmer_features[i].flags | F_MANUAL_ZWJ | F_MANUAL_ZWNJ); |
|
125 map->add_gsub_pause (nullptr); |
|
126 } |
|
127 map->add_gsub_pause (final_reordering); |
|
128 for (; i < KHMER_NUM_FEATURES; i++) { |
|
129 map->add_feature (khmer_features[i].tag, 1, khmer_features[i].flags | F_MANUAL_ZWJ | F_MANUAL_ZWNJ); |
|
130 } |
|
131 |
|
132 map->add_global_bool_feature (HB_TAG('c','a','l','t')); |
|
133 map->add_global_bool_feature (HB_TAG('c','l','i','g')); |
|
134 |
125 |
135 map->add_gsub_pause (clear_syllables); |
126 map->add_gsub_pause (clear_syllables); |
|
127 |
|
128 for (; i < KHMER_NUM_FEATURES; i++) |
|
129 map->add_feature (khmer_features[i]); |
136 } |
130 } |
137 |
131 |
138 static void |
132 static void |
139 override_features_khmer (hb_ot_shape_planner_t *plan) |
133 override_features_khmer (hb_ot_shape_planner_t *plan) |
140 { |
134 { |
|
135 hb_ot_map_builder_t *map = &plan->map; |
|
136 |
|
137 /* Khmer spec has 'clig' as part of required shaping features: |
|
138 * "Apply feature 'clig' to form ligatures that are desired for |
|
139 * typographical correctness.", hence in overrides... */ |
|
140 map->enable_feature (HB_TAG('c','l','i','g')); |
|
141 |
141 /* Uniscribe does not apply 'kern' in Khmer. */ |
142 /* Uniscribe does not apply 'kern' in Khmer. */ |
142 if (hb_options ().uniscribe_bug_compatible) |
143 if (hb_options ().uniscribe_bug_compatible) |
143 { |
144 { |
144 plan->map.add_feature (HB_TAG('k','e','r','n'), 0, F_GLOBAL); |
145 map->disable_feature (HB_TAG('k','e','r','n')); |
145 } |
146 } |
146 |
147 |
147 plan->map.add_feature (HB_TAG('l','i','g','a'), 0, F_GLOBAL); |
148 map->disable_feature (HB_TAG('l','i','g','a')); |
148 } |
149 } |
149 |
150 |
150 |
151 |
151 struct would_substitute_feature_t |
152 struct would_substitute_feature_t |
152 { |
153 { |
153 inline void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_) |
154 void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_) |
154 { |
155 { |
155 zero_context = zero_context_; |
156 zero_context = zero_context_; |
156 map->get_stage_lookups (0/*GSUB*/, |
157 map->get_stage_lookups (0/*GSUB*/, |
157 map->get_feature_stage (0/*GSUB*/, feature_tag), |
158 map->get_feature_stage (0/*GSUB*/, feature_tag), |
158 &lookups, &count); |
159 &lookups, &count); |
159 } |
160 } |
160 |
161 |
161 inline bool would_substitute (const hb_codepoint_t *glyphs, |
162 bool would_substitute (const hb_codepoint_t *glyphs, |
162 unsigned int glyphs_count, |
163 unsigned int glyphs_count, |
163 hb_face_t *face) const |
164 hb_face_t *face) const |
164 { |
165 { |
165 for (unsigned int i = 0; i < count; i++) |
166 for (unsigned int i = 0; i < count; i++) |
166 if (hb_ot_layout_lookup_would_substitute_fast (face, lookups[i].index, glyphs, glyphs_count, zero_context)) |
167 if (hb_ot_layout_lookup_would_substitute_fast (face, lookups[i].index, glyphs, glyphs_count, zero_context)) |
167 return true; |
168 return true; |
168 return false; |
169 return false; |
262 find_syllables (buffer); |
260 find_syllables (buffer); |
263 foreach_syllable (buffer, start, end) |
261 foreach_syllable (buffer, start, end) |
264 buffer->unsafe_to_break (start, end); |
262 buffer->unsafe_to_break (start, end); |
265 } |
263 } |
266 |
264 |
267 static int |
|
268 compare_khmer_order (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb) |
|
269 { |
|
270 int a = pa->khmer_position(); |
|
271 int b = pb->khmer_position(); |
|
272 |
|
273 return a < b ? -1 : a == b ? 0 : +1; |
|
274 } |
|
275 |
|
276 |
265 |
277 /* Rules from: |
266 /* Rules from: |
278 * https://docs.microsoft.com/en-us/typography/script-development/devanagari */ |
267 * https://docs.microsoft.com/en-us/typography/script-development/devanagari */ |
279 |
268 |
280 static void |
269 static void |
281 initial_reordering_consonant_syllable (const hb_ot_shape_plan_t *plan, |
270 reorder_consonant_syllable (const hb_ot_shape_plan_t *plan, |
282 hb_face_t *face, |
271 hb_face_t *face HB_UNUSED, |
283 hb_buffer_t *buffer, |
272 hb_buffer_t *buffer, |
284 unsigned int start, unsigned int end) |
273 unsigned int start, unsigned int end) |
285 { |
274 { |
286 const khmer_shape_plan_t *khmer_plan = (const khmer_shape_plan_t *) plan->data; |
275 const khmer_shape_plan_t *khmer_plan = (const khmer_shape_plan_t *) plan->data; |
287 hb_glyph_info_t *info = buffer->info; |
276 hb_glyph_info_t *info = buffer->info; |
288 |
277 |
289 /* 1. Khmer shaping assumes that a syllable will begin with a Cons, IndV, or Number. */ |
278 /* Setup masks. */ |
290 |
279 { |
291 /* The first consonant is always the base. */ |
280 /* Post-base */ |
292 unsigned int base = start; |
281 hb_mask_t mask = khmer_plan->mask_array[BLWF] | khmer_plan->mask_array[ABVF] | khmer_plan->mask_array[PSTF]; |
293 info[base].khmer_position() = POS_BASE_C; |
282 for (unsigned int i = start + 1; i < end; i++) |
294 |
283 info[i].mask |= mask; |
295 /* Mark all subsequent consonants as below. */ |
284 } |
296 for (unsigned int i = base + 1; i < end; i++) |
285 |
297 if (is_consonant_or_vowel (info[i])) |
286 unsigned int num_coengs = 0; |
298 info[i].khmer_position() = POS_BELOW_C; |
287 for (unsigned int i = start + 1; i < end; i++) |
299 |
288 { |
300 /* Mark final consonants. A final consonant is one appearing after a matra, |
289 /* """ |
301 * like in Khmer. */ |
290 * When a COENG + (Cons | IndV) combination are found (and subscript count |
302 for (unsigned int i = base + 1; i < end; i++) |
291 * is less than two) the character combination is handled according to the |
303 if (info[i].khmer_category() == OT_M) { |
292 * subscript type of the character following the COENG. |
304 for (unsigned int j = i + 1; j < end; j++) |
293 * |
305 if (is_consonant_or_vowel (info[j])) { |
294 * ... |
306 info[j].khmer_position() = POS_FINAL_C; |
295 * |
307 break; |
296 * Subscript Type 2 - The COENG + RO characters are reordered to immediately |
308 } |
297 * before the base glyph. Then the COENG + RO characters are assigned to have |
309 break; |
298 * the 'pref' OpenType feature applied to them. |
310 } |
299 * """ |
311 |
300 */ |
312 /* Attach misc marks to previous char to move with them. */ |
301 if (info[i].khmer_category() == OT_Coeng && num_coengs <= 2 && i + 1 < end) |
313 { |
|
314 khmer_position_t last_pos = POS_START; |
|
315 for (unsigned int i = start; i < end; i++) |
|
316 { |
302 { |
317 if ((FLAG_UNSAFE (info[i].khmer_category()) & (JOINER_FLAGS | FLAG (OT_N) | FLAG (OT_RS) | MEDIAL_FLAGS | FLAG (OT_Coeng)))) |
303 num_coengs++; |
|
304 |
|
305 if (info[i + 1].khmer_category() == OT_Ra) |
318 { |
306 { |
319 info[i].khmer_position() = last_pos; |
307 for (unsigned int j = 0; j < 2; j++) |
320 if (unlikely (info[i].khmer_category() == OT_Coeng && |
308 info[i + j].mask |= khmer_plan->mask_array[PREF]; |
321 info[i].khmer_position() == POS_PRE_M)) |
309 |
322 { |
310 /* Move the Coeng,Ro sequence to the start. */ |
323 /* |
311 buffer->merge_clusters (start, i + 2); |
324 * Uniscribe doesn't move the Halant with Left Matra. |
312 hb_glyph_info_t t0 = info[i]; |
325 * TEST: U+092B,U+093F,U+094DE |
313 hb_glyph_info_t t1 = info[i + 1]; |
326 * We follow. This is important for the Sinhala |
314 memmove (&info[start + 2], &info[start], (i - start) * sizeof (info[0])); |
327 * U+0DDA split matra since it decomposes to U+0DD9,U+0DCA |
315 info[start] = t0; |
328 * where U+0DD9 is a left matra and U+0DCA is the virama. |
316 info[start + 1] = t1; |
329 * We don't want to move the virama with the left matra. |
|
330 * TEST: U+0D9A,U+0DDA |
|
331 */ |
|
332 for (unsigned int j = i; j > start; j--) |
|
333 if (info[j - 1].khmer_position() != POS_PRE_M) { |
|
334 info[i].khmer_position() = info[j - 1].khmer_position(); |
|
335 break; |
|
336 } |
|
337 } |
|
338 } else if (info[i].khmer_position() != POS_SMVD) { |
|
339 last_pos = (khmer_position_t) info[i].khmer_position(); |
|
340 } |
|
341 } |
|
342 } |
|
343 /* For post-base consonants let them own anything before them |
|
344 * since the last consonant or matra. */ |
|
345 { |
|
346 unsigned int last = base; |
|
347 for (unsigned int i = base + 1; i < end; i++) |
|
348 if (is_consonant_or_vowel (info[i])) |
|
349 { |
|
350 for (unsigned int j = last + 1; j < i; j++) |
|
351 if (info[j].khmer_position() < POS_SMVD) |
|
352 info[j].khmer_position() = info[i].khmer_position(); |
|
353 last = i; |
|
354 } else if (info[i].khmer_category() == OT_M) |
|
355 last = i; |
|
356 } |
|
357 |
|
358 { |
|
359 /* Use syllable() for sort accounting temporarily. */ |
|
360 unsigned int syllable = info[start].syllable(); |
|
361 for (unsigned int i = start; i < end; i++) |
|
362 info[i].syllable() = i - start; |
|
363 |
|
364 /* Sit tight, rock 'n roll! */ |
|
365 hb_stable_sort (info + start, end - start, compare_khmer_order); |
|
366 /* Find base again */ |
|
367 base = end; |
|
368 for (unsigned int i = start; i < end; i++) |
|
369 if (info[i].khmer_position() == POS_BASE_C) |
|
370 { |
|
371 base = i; |
|
372 break; |
|
373 } |
|
374 |
|
375 if (unlikely (end - start >= 127)) |
|
376 buffer->merge_clusters (start, end); |
|
377 else |
|
378 /* Note! syllable() is a one-byte field. */ |
|
379 for (unsigned int i = base; i < end; i++) |
|
380 if (info[i].syllable() != 255) |
|
381 { |
|
382 unsigned int max = i; |
|
383 unsigned int j = start + info[i].syllable(); |
|
384 while (j != i) |
|
385 { |
|
386 max = MAX (max, j); |
|
387 unsigned int next = start + info[j].syllable(); |
|
388 info[j].syllable() = 255; /* So we don't process j later again. */ |
|
389 j = next; |
|
390 } |
|
391 if (i != max) |
|
392 buffer->merge_clusters (i, max + 1); |
|
393 } |
|
394 |
|
395 /* Put syllable back in. */ |
|
396 for (unsigned int i = start; i < end; i++) |
|
397 info[i].syllable() = syllable; |
|
398 } |
|
399 |
|
400 /* Setup masks now */ |
|
401 |
|
402 { |
|
403 hb_mask_t mask; |
|
404 |
|
405 /* Post-base */ |
|
406 mask = khmer_plan->mask_array[BLWF] | khmer_plan->mask_array[ABVF] | khmer_plan->mask_array[PSTF]; |
|
407 for (unsigned int i = base + 1; i < end; i++) |
|
408 info[i].mask |= mask; |
|
409 } |
|
410 |
|
411 unsigned int pref_len = 2; |
|
412 if (khmer_plan->mask_array[PREF] && base + pref_len < end) |
|
413 { |
|
414 /* Find a Halant,Ra sequence and mark it for pre-base-reordering processing. */ |
|
415 for (unsigned int i = base + 1; i + pref_len - 1 < end; i++) { |
|
416 hb_codepoint_t glyphs[2]; |
|
417 for (unsigned int j = 0; j < pref_len; j++) |
|
418 glyphs[j] = info[i + j].codepoint; |
|
419 if (khmer_plan->pref.would_substitute (glyphs, pref_len, face)) |
|
420 { |
|
421 for (unsigned int j = 0; j < pref_len; j++) |
|
422 info[i++].mask |= khmer_plan->mask_array[PREF]; |
|
423 |
317 |
424 /* Mark the subsequent stuff with 'cfar'. Used in Khmer. |
318 /* Mark the subsequent stuff with 'cfar'. Used in Khmer. |
425 * Read the feature spec. |
319 * Read the feature spec. |
426 * This allows distinguishing the following cases with MS Khmer fonts: |
320 * This allows distinguishing the following cases with MS Khmer fonts: |
427 * U+1784,U+17D2,U+179A,U+17D2,U+1782 |
321 * U+1784,U+17D2,U+179A,U+17D2,U+1782 |
428 * U+1784,U+17D2,U+1782,U+17D2,U+179A |
322 * U+1784,U+17D2,U+1782,U+17D2,U+179A |
429 */ |
323 */ |
430 if (khmer_plan->mask_array[CFAR]) |
324 if (khmer_plan->mask_array[CFAR]) |
431 for (; i < end; i++) |
325 for (unsigned int j = i + 2; j < end; j++) |
432 info[i].mask |= khmer_plan->mask_array[CFAR]; |
326 info[j].mask |= khmer_plan->mask_array[CFAR]; |
433 |
327 |
434 break; |
328 num_coengs = 2; /* Done. */ |
435 } |
329 } |
|
330 } |
|
331 |
|
332 /* Reorder left matra piece. */ |
|
333 else if (info[i].khmer_category() == OT_VPre) |
|
334 { |
|
335 /* Move to the start. */ |
|
336 buffer->merge_clusters (start, i + 1); |
|
337 hb_glyph_info_t t = info[i]; |
|
338 memmove (&info[start + 1], &info[start], (i - start) * sizeof (info[0])); |
|
339 info[start] = t; |
436 } |
340 } |
437 } |
341 } |
438 } |
342 } |
439 |
343 |
440 static void |
344 static void |
511 buffer->output_info (ginfo); |
415 buffer->output_info (ginfo); |
512 } |
416 } |
513 else |
417 else |
514 buffer->next_glyph (); |
418 buffer->next_glyph (); |
515 } |
419 } |
516 |
|
517 buffer->swap_buffers (); |
420 buffer->swap_buffers (); |
518 } |
421 } |
519 |
422 |
520 static void |
423 static void |
521 initial_reordering (const hb_ot_shape_plan_t *plan, |
424 reorder (const hb_ot_shape_plan_t *plan, |
522 hb_font_t *font, |
425 hb_font_t *font, |
523 hb_buffer_t *buffer) |
426 hb_buffer_t *buffer) |
524 { |
427 { |
525 insert_dotted_circles (plan, font, buffer); |
428 insert_dotted_circles (plan, font, buffer); |
526 |
429 |
527 foreach_syllable (buffer, start, end) |
430 foreach_syllable (buffer, start, end) |
528 initial_reordering_syllable (plan, font->face, buffer, start, end); |
431 initial_reordering_syllable (plan, font->face, buffer, start, end); |
529 } |
|
530 |
|
531 static void |
|
532 final_reordering_syllable (const hb_ot_shape_plan_t *plan, |
|
533 hb_buffer_t *buffer, |
|
534 unsigned int start, unsigned int end) |
|
535 { |
|
536 const khmer_shape_plan_t *khmer_plan = (const khmer_shape_plan_t *) plan->data; |
|
537 hb_glyph_info_t *info = buffer->info; |
|
538 |
|
539 |
|
540 /* This function relies heavily on halant glyphs. Lots of ligation |
|
541 * and possibly multiple substitutions happened prior to this |
|
542 * phase, and that might have messed up our properties. Recover |
|
543 * from a particular case of that where we're fairly sure that a |
|
544 * class of OT_Coeng is desired but has been lost. */ |
|
545 if (khmer_plan->virama_glyph) |
|
546 { |
|
547 unsigned int virama_glyph = khmer_plan->virama_glyph; |
|
548 for (unsigned int i = start; i < end; i++) |
|
549 if (info[i].codepoint == virama_glyph && |
|
550 _hb_glyph_info_ligated (&info[i]) && |
|
551 _hb_glyph_info_multiplied (&info[i])) |
|
552 { |
|
553 /* This will make sure that this glyph passes is_coeng() test. */ |
|
554 info[i].khmer_category() = OT_Coeng; |
|
555 _hb_glyph_info_clear_ligated_and_multiplied (&info[i]); |
|
556 } |
|
557 } |
|
558 |
|
559 |
|
560 /* 4. Final reordering: |
|
561 * |
|
562 * After the localized forms and basic shaping forms GSUB features have been |
|
563 * applied (see below), the shaping engine performs some final glyph |
|
564 * reordering before applying all the remaining font features to the entire |
|
565 * syllable. |
|
566 */ |
|
567 |
|
568 bool try_pref = !!khmer_plan->mask_array[PREF]; |
|
569 |
|
570 /* Find base again */ |
|
571 unsigned int base; |
|
572 for (base = start; base < end; base++) |
|
573 if (info[base].khmer_position() >= POS_BASE_C) |
|
574 { |
|
575 if (try_pref && base + 1 < end) |
|
576 { |
|
577 for (unsigned int i = base + 1; i < end; i++) |
|
578 if ((info[i].mask & khmer_plan->mask_array[PREF]) != 0) |
|
579 { |
|
580 if (!(_hb_glyph_info_substituted (&info[i]) && |
|
581 _hb_glyph_info_ligated_and_didnt_multiply (&info[i]))) |
|
582 { |
|
583 /* Ok, this was a 'pref' candidate but didn't form any. |
|
584 * Base is around here... */ |
|
585 base = i; |
|
586 while (base < end && is_coeng (info[base])) |
|
587 base++; |
|
588 info[base].khmer_position() = POS_BASE_C; |
|
589 |
|
590 try_pref = false; |
|
591 } |
|
592 break; |
|
593 } |
|
594 } |
|
595 |
|
596 if (start < base && info[base].khmer_position() > POS_BASE_C) |
|
597 base--; |
|
598 break; |
|
599 } |
|
600 if (base == end && start < base && |
|
601 is_one_of (info[base - 1], FLAG (OT_ZWJ))) |
|
602 base--; |
|
603 if (base < end) |
|
604 while (start < base && |
|
605 is_one_of (info[base], (FLAG (OT_N) | FLAG (OT_Coeng)))) |
|
606 base--; |
|
607 |
|
608 |
|
609 /* o Reorder matras: |
|
610 * |
|
611 * If a pre-base matra character had been reordered before applying basic |
|
612 * features, the glyph can be moved closer to the main consonant based on |
|
613 * whether half-forms had been formed. Actual position for the matra is |
|
614 * defined as “after last standalone halant glyph, after initial matra |
|
615 * position and before the main consonant”. If ZWJ or ZWNJ follow this |
|
616 * halant, position is moved after it. |
|
617 */ |
|
618 |
|
619 if (start + 1 < end && start < base) /* Otherwise there can't be any pre-base matra characters. */ |
|
620 { |
|
621 /* If we lost track of base, alas, position before last thingy. */ |
|
622 unsigned int new_pos = base == end ? base - 2 : base - 1; |
|
623 |
|
624 while (new_pos > start && |
|
625 !(is_one_of (info[new_pos], (FLAG (OT_M) | FLAG (OT_Coeng))))) |
|
626 new_pos--; |
|
627 |
|
628 /* If we found no Halant we are done. |
|
629 * Otherwise only proceed if the Halant does |
|
630 * not belong to the Matra itself! */ |
|
631 if (is_coeng (info[new_pos]) && |
|
632 info[new_pos].khmer_position() != POS_PRE_M) |
|
633 { |
|
634 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */ |
|
635 if (new_pos + 1 < end && is_joiner (info[new_pos + 1])) |
|
636 new_pos++; |
|
637 } |
|
638 else |
|
639 new_pos = start; /* No move. */ |
|
640 |
|
641 if (start < new_pos && info[new_pos].khmer_position () != POS_PRE_M) |
|
642 { |
|
643 /* Now go see if there's actually any matras... */ |
|
644 for (unsigned int i = new_pos; i > start; i--) |
|
645 if (info[i - 1].khmer_position () == POS_PRE_M) |
|
646 { |
|
647 unsigned int old_pos = i - 1; |
|
648 if (old_pos < base && base <= new_pos) /* Shouldn't actually happen. */ |
|
649 base--; |
|
650 |
|
651 hb_glyph_info_t tmp = info[old_pos]; |
|
652 memmove (&info[old_pos], &info[old_pos + 1], (new_pos - old_pos) * sizeof (info[0])); |
|
653 info[new_pos] = tmp; |
|
654 |
|
655 /* Note: this merge_clusters() is intentionally *after* the reordering. |
|
656 * Indic matra reordering is special and tricky... */ |
|
657 buffer->merge_clusters (new_pos, MIN (end, base + 1)); |
|
658 |
|
659 new_pos--; |
|
660 } |
|
661 } else { |
|
662 for (unsigned int i = start; i < base; i++) |
|
663 if (info[i].khmer_position () == POS_PRE_M) { |
|
664 buffer->merge_clusters (i, MIN (end, base + 1)); |
|
665 break; |
|
666 } |
|
667 } |
|
668 } |
|
669 |
|
670 |
|
671 /* o Reorder pre-base-reordering consonants: |
|
672 * |
|
673 * If a pre-base-reordering consonant is found, reorder it according to |
|
674 * the following rules: |
|
675 */ |
|
676 |
|
677 if (try_pref && base + 1 < end) /* Otherwise there can't be any pre-base-reordering Ra. */ |
|
678 { |
|
679 for (unsigned int i = base + 1; i < end; i++) |
|
680 if ((info[i].mask & khmer_plan->mask_array[PREF]) != 0) |
|
681 { |
|
682 /* 1. Only reorder a glyph produced by substitution during application |
|
683 * of the <pref> feature. (Note that a font may shape a Ra consonant with |
|
684 * the feature generally but block it in certain contexts.) |
|
685 */ |
|
686 /* Note: We just check that something got substituted. We don't check that |
|
687 * the <pref> feature actually did it... |
|
688 * |
|
689 * Reorder pref only if it ligated. */ |
|
690 if (_hb_glyph_info_ligated_and_didnt_multiply (&info[i])) |
|
691 { |
|
692 /* |
|
693 * 2. Try to find a target position the same way as for pre-base matra. |
|
694 * If it is found, reorder pre-base consonant glyph. |
|
695 * |
|
696 * 3. If position is not found, reorder immediately before main |
|
697 * consonant. |
|
698 */ |
|
699 |
|
700 unsigned int new_pos = base; |
|
701 while (new_pos > start && |
|
702 !(is_one_of (info[new_pos - 1], FLAG(OT_M) | FLAG (OT_Coeng)))) |
|
703 new_pos--; |
|
704 |
|
705 /* In Khmer coeng model, a H,Ra can go *after* matras. If it goes after a |
|
706 * split matra, it should be reordered to *before* the left part of such matra. */ |
|
707 if (new_pos > start && info[new_pos - 1].khmer_category() == OT_M) |
|
708 { |
|
709 unsigned int old_pos = i; |
|
710 for (unsigned int j = base + 1; j < old_pos; j++) |
|
711 if (info[j].khmer_category() == OT_M) |
|
712 { |
|
713 new_pos--; |
|
714 break; |
|
715 } |
|
716 } |
|
717 |
|
718 if (new_pos > start && is_coeng (info[new_pos - 1])) |
|
719 { |
|
720 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */ |
|
721 if (new_pos < end && is_joiner (info[new_pos])) |
|
722 new_pos++; |
|
723 } |
|
724 |
|
725 { |
|
726 unsigned int old_pos = i; |
|
727 |
|
728 buffer->merge_clusters (new_pos, old_pos + 1); |
|
729 hb_glyph_info_t tmp = info[old_pos]; |
|
730 memmove (&info[new_pos + 1], &info[new_pos], (old_pos - new_pos) * sizeof (info[0])); |
|
731 info[new_pos] = tmp; |
|
732 |
|
733 if (new_pos <= base && base < old_pos) |
|
734 base++; |
|
735 } |
|
736 } |
|
737 |
|
738 break; |
|
739 } |
|
740 } |
|
741 |
|
742 |
|
743 /* |
|
744 * Finish off the clusters and go home! |
|
745 */ |
|
746 if (hb_options ().uniscribe_bug_compatible) |
|
747 { |
|
748 /* Uniscribe merges the entire syllable into a single cluster... Except for Tamil & Sinhala. |
|
749 * This means, half forms are submerged into the main consonant's cluster. |
|
750 * This is unnecessary, and makes cursor positioning harder, but that's what |
|
751 * Uniscribe does. */ |
|
752 buffer->merge_clusters (start, end); |
|
753 } |
|
754 } |
|
755 |
|
756 |
|
757 static void |
|
758 final_reordering (const hb_ot_shape_plan_t *plan, |
|
759 hb_font_t *font HB_UNUSED, |
|
760 hb_buffer_t *buffer) |
|
761 { |
|
762 unsigned int count = buffer->len; |
|
763 if (unlikely (!count)) return; |
|
764 |
|
765 foreach_syllable (buffer, start, end) |
|
766 final_reordering_syllable (plan, buffer, start, end); |
|
767 |
432 |
768 HB_BUFFER_DEALLOCATE_VAR (buffer, khmer_category); |
433 HB_BUFFER_DEALLOCATE_VAR (buffer, khmer_category); |
769 HB_BUFFER_DEALLOCATE_VAR (buffer, khmer_position); |
434 } |
770 } |
|
771 |
|
772 |
435 |
773 static void |
436 static void |
774 clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED, |
437 clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED, |
775 hb_font_t *font HB_UNUSED, |
438 hb_font_t *font HB_UNUSED, |
776 hb_buffer_t *buffer) |
439 hb_buffer_t *buffer) |