1 /*
2 * Copyright © 2011,2012 Google, Inc.
3 *
4 * This is part of HarfBuzz, a text shaping library.
5 *
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.
11 *
12 * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
13 * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
14 * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
15 * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
16 * DAMAGE.
17 *
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.
23 *
24 * Google Author(s): Behdad Esfahbod
25 */
26
27 #include "hb-ot-shape-complex-khmer-private.hh"
28 #include "hb-ot-layout-private.hh"
29
30
31 /*
32 * Khmer shaper.
33 */
34
35 struct feature_list_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[] =
42 {
43 /*
44 * Basic features.
45 * These features are applied in order, one at a time, after initial_reordering.
46 */
47 {HB_TAG('p','r','e','f'), F_NONE},
48 {HB_TAG('b','l','w','f'), F_NONE},
49 {HB_TAG('a','b','v','f'), F_NONE},
50 {HB_TAG('p','s','t','f'), F_NONE},
51 {HB_TAG('c','f','a','r'), F_NONE},
52 /*
53 * Other features.
54 * These features are applied all at once, after final_reordering.
55 * Default Bengali font in Windows for example has intermixed
56 * lookups for init,pres,abvs,blws features.
57 */
58 {HB_TAG('p','r','e','s'), F_GLOBAL},
59 {HB_TAG('a','b','v','s'), F_GLOBAL},
60 {HB_TAG('b','l','w','s'), F_GLOBAL},
61 {HB_TAG('p','s','t','s'), F_GLOBAL},
62 /* Positioning features, though we don't care about the types. */
63 {HB_TAG('d','i','s','t'), F_GLOBAL},
64 {HB_TAG('a','b','v','m'), F_GLOBAL},
65 {HB_TAG('b','l','w','m'), F_GLOBAL},
66 };
67
68 /*
69 * Must be in the same order as the khmer_features array.
70 */
71 enum {
72 PREF,
73 BLWF,
74 ABVF,
75 PSTF,
76 CFAR,
77
78 _PRES,
79 _ABVS,
80 _BLWS,
81 _PSTS,
82 _DIST,
83 _ABVM,
84 _BLWM,
85
86 KHMER_NUM_FEATURES,
87 KHMER_BASIC_FEATURES = _PRES /* Don't forget to update this! */
88 };
89
90 static void
91 setup_syllables (const hb_ot_shape_plan_t *plan,
92 hb_font_t *font,
93 hb_buffer_t *buffer);
94 static void
95 initial_reordering (const hb_ot_shape_plan_t *plan,
96 hb_font_t *font,
97 hb_buffer_t *buffer);
98 static void
99 final_reordering (const hb_ot_shape_plan_t *plan,
100 hb_font_t *font,
101 hb_buffer_t *buffer);
102 static void
103 clear_syllables (const hb_ot_shape_plan_t *plan,
104 hb_font_t *font,
105 hb_buffer_t *buffer);
106
107 static void
108 collect_features_khmer (hb_ot_shape_planner_t *plan)
109 {
110 hb_ot_map_builder_t *map = &plan->map;
111
112 /* Do this before any lookups have been applied. */
113 map->add_gsub_pause (setup_syllables);
114
115 map->add_global_bool_feature (HB_TAG('l','o','c','l'));
116 /* The Indic specs do not require ccmp, but we apply it here since if
117 * there is a use of it, it's typically at the beginning. */
118 map->add_global_bool_feature (HB_TAG('c','c','m','p'));
119
120
121 unsigned int i = 0;
122 map->add_gsub_pause (initial_reordering);
123 for (; i < KHMER_BASIC_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
135 map->add_gsub_pause (clear_syllables);
136 }
137
138 static void
139 override_features_khmer (hb_ot_shape_planner_t *plan)
140 {
141 /* Uniscribe does not apply 'kern' in Khmer. */
142 if (hb_options ().uniscribe_bug_compatible)
143 {
144 plan->map.add_feature (HB_TAG('k','e','r','n'), 0, F_GLOBAL);
145 }
146
147 plan->map.add_feature (HB_TAG('l','i','g','a'), 0, F_GLOBAL);
148 }
149
150
151 struct would_substitute_feature_t
152 {
153 inline void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_)
154 {
155 zero_context = zero_context_;
156 map->get_stage_lookups (0/*GSUB*/,
157 map->get_feature_stage (0/*GSUB*/, feature_tag),
158 &lookups, &count);
159 }
160
161 inline bool would_substitute (const hb_codepoint_t *glyphs,
162 unsigned int glyphs_count,
163 hb_face_t *face) const
164 {
165 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 return true;
168 return false;
169 }
170
171 private:
172 const hb_ot_map_t::lookup_map_t *lookups;
173 unsigned int count;
174 bool zero_context;
175 };
176
177 struct khmer_shape_plan_t
178 {
179 ASSERT_POD ();
180
181 inline bool get_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const
182 {
183 hb_codepoint_t glyph = virama_glyph;
184 if (unlikely (virama_glyph == (hb_codepoint_t) -1))
185 {
186 if (!font->get_nominal_glyph (0x17D2u, &glyph))
187 glyph = 0;
188 /* Technically speaking, the spec says we should apply 'locl' to virama too.
189 * Maybe one day... */
190
191 /* Our get_nominal_glyph() function needs a font, so we can't get the virama glyph
192 * during shape planning... Instead, overwrite it here. It's safe. Don't worry! */
193 virama_glyph = glyph;
194 }
195
196 *pglyph = glyph;
197 return glyph != 0;
198 }
199
200 mutable hb_codepoint_t virama_glyph;
201
202 would_substitute_feature_t pref;
203
204 hb_mask_t mask_array[KHMER_NUM_FEATURES];
205 };
206
207 static void *
208 data_create_khmer (const hb_ot_shape_plan_t *plan)
209 {
210 khmer_shape_plan_t *khmer_plan = (khmer_shape_plan_t *) calloc (1, sizeof (khmer_shape_plan_t));
211 if (unlikely (!khmer_plan))
212 return nullptr;
213
214 khmer_plan->virama_glyph = (hb_codepoint_t) -1;
215
216 khmer_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), true);
217
218 for (unsigned int i = 0; i < ARRAY_LENGTH (khmer_plan->mask_array); i++)
219 khmer_plan->mask_array[i] = (khmer_features[i].flags & F_GLOBAL) ?
220 0 : plan->map.get_1_mask (khmer_features[i].tag);
221
222 return khmer_plan;
223 }
224
225 static void
226 data_destroy_khmer (void *data)
227 {
228 free (data);
229 }
230
231
232 enum syllable_type_t {
233 consonant_syllable,
234 broken_cluster,
235 non_khmer_cluster,
236 };
237
238 #include "hb-ot-shape-complex-khmer-machine.hh"
239
240 static void
241 setup_masks_khmer (const hb_ot_shape_plan_t *plan HB_UNUSED,
242 hb_buffer_t *buffer,
243 hb_font_t *font HB_UNUSED)
244 {
245 HB_BUFFER_ALLOCATE_VAR (buffer, khmer_category);
246 HB_BUFFER_ALLOCATE_VAR (buffer, khmer_position);
247
248 /* We cannot setup masks here. We save information about characters
249 * and setup masks later on in a pause-callback. */
250
251 unsigned int count = buffer->len;
252 hb_glyph_info_t *info = buffer->info;
253 for (unsigned int i = 0; i < count; i++)
254 set_khmer_properties (info[i]);
255 }
256
257 static void
258 setup_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
259 hb_font_t *font HB_UNUSED,
260 hb_buffer_t *buffer)
261 {
262 find_syllables (buffer);
263 foreach_syllable (buffer, start, end)
264 buffer->unsafe_to_break (start, end);
265 }
266
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
277 /* Rules from:
278 * https://docs.microsoft.com/en-us/typography/script-development/devanagari */
279
280 static void
281 initial_reordering_consonant_syllable (const hb_ot_shape_plan_t *plan,
282 hb_face_t *face,
283 hb_buffer_t *buffer,
284 unsigned int start, unsigned int end)
285 {
286 const khmer_shape_plan_t *khmer_plan = (const khmer_shape_plan_t *) plan->data;
287 hb_glyph_info_t *info = buffer->info;
288
289 /* 1. Khmer shaping assumes that a syllable will begin with a Cons, IndV, or Number. */
290
291 /* The first consonant is always the base. */
292 unsigned int base = start;
293 info[base].khmer_position() = POS_BASE_C;
294
295 /* Mark all subsequent consonants as below. */
296 for (unsigned int i = base + 1; i < end; i++)
297 if (is_consonant_or_vowel (info[i]))
298 info[i].khmer_position() = POS_BELOW_C;
299
300 /* Mark final consonants. A final consonant is one appearing after a matra,
301 * like in Khmer. */
302 for (unsigned int i = base + 1; i < end; i++)
303 if (info[i].khmer_category() == OT_M) {
304 for (unsigned int j = i + 1; j < end; j++)
305 if (is_consonant_or_vowel (info[j])) {
306 info[j].khmer_position() = POS_FINAL_C;
307 break;
308 }
309 break;
310 }
311
312 /* Attach misc marks to previous char to move with them. */
313 {
314 khmer_position_t last_pos = POS_START;
315 for (unsigned int i = start; i < end; i++)
316 {
317 if ((FLAG_UNSAFE (info[i].khmer_category()) & (JOINER_FLAGS | FLAG (OT_N) | FLAG (OT_RS) | MEDIAL_FLAGS | FLAG (OT_Coeng))))
318 {
319 info[i].khmer_position() = last_pos;
320 if (unlikely (info[i].khmer_category() == OT_Coeng &&
321 info[i].khmer_position() == POS_PRE_M))
322 {
323 /*
324 * Uniscribe doesn't move the Halant with Left Matra.
325 * TEST: U+092B,U+093F,U+094DE
326 * We follow. This is important for the Sinhala
327 * U+0DDA split matra since it decomposes to U+0DD9,U+0DCA
328 * where U+0DD9 is a left matra and U+0DCA is the virama.
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
424 /* Mark the subsequent stuff with 'cfar'. Used in Khmer.
425 * Read the feature spec.
426 * This allows distinguishing the following cases with MS Khmer fonts:
427 * U+1784,U+17D2,U+179A,U+17D2,U+1782
428 * U+1784,U+17D2,U+1782,U+17D2,U+179A
429 */
430 if (khmer_plan->mask_array[CFAR])
431 for (; i < end; i++)
432 info[i].mask |= khmer_plan->mask_array[CFAR];
433
434 break;
435 }
436 }
437 }
438 }
439
440 static void
441 initial_reordering_syllable (const hb_ot_shape_plan_t *plan,
442 hb_face_t *face,
443 hb_buffer_t *buffer,
444 unsigned int start, unsigned int end)
445 {
446 syllable_type_t syllable_type = (syllable_type_t) (buffer->info[start].syllable() & 0x0F);
447 switch (syllable_type)
448 {
449 case broken_cluster: /* We already inserted dotted-circles, so just call the consonant_syllable. */
450 case consonant_syllable:
451 initial_reordering_consonant_syllable (plan, face, buffer, start, end);
452 break;
453
454 case non_khmer_cluster:
455 break;
456 }
457 }
458
459 static inline void
460 insert_dotted_circles (const hb_ot_shape_plan_t *plan HB_UNUSED,
461 hb_font_t *font,
462 hb_buffer_t *buffer)
463 {
464 /* Note: This loop is extra overhead, but should not be measurable. */
465 bool has_broken_syllables = false;
466 unsigned int count = buffer->len;
467 hb_glyph_info_t *info = buffer->info;
468 for (unsigned int i = 0; i < count; i++)
469 if ((info[i].syllable() & 0x0F) == broken_cluster)
470 {
471 has_broken_syllables = true;
472 break;
473 }
474 if (likely (!has_broken_syllables))
475 return;
476
477
478 hb_codepoint_t dottedcircle_glyph;
479 if (!font->get_nominal_glyph (0x25CCu, &dottedcircle_glyph))
480 return;
481
482 hb_glyph_info_t dottedcircle = {0};
483 dottedcircle.codepoint = 0x25CCu;
484 set_khmer_properties (dottedcircle);
485 dottedcircle.codepoint = dottedcircle_glyph;
486
487 buffer->clear_output ();
488
489 buffer->idx = 0;
490 unsigned int last_syllable = 0;
491 while (buffer->idx < buffer->len && buffer->successful)
492 {
493 unsigned int syllable = buffer->cur().syllable();
494 syllable_type_t syllable_type = (syllable_type_t) (syllable & 0x0F);
495 if (unlikely (last_syllable != syllable && syllable_type == broken_cluster))
496 {
497 last_syllable = syllable;
498
499 hb_glyph_info_t ginfo = dottedcircle;
500 ginfo.cluster = buffer->cur().cluster;
501 ginfo.mask = buffer->cur().mask;
502 ginfo.syllable() = buffer->cur().syllable();
503 /* TODO Set glyph_props? */
504
505 /* Insert dottedcircle after possible Repha. */
506 while (buffer->idx < buffer->len && buffer->successful &&
507 last_syllable == buffer->cur().syllable() &&
508 buffer->cur().khmer_category() == OT_Repha)
509 buffer->next_glyph ();
510
511 buffer->output_info (ginfo);
512 }
513 else
514 buffer->next_glyph ();
515 }
516
517 buffer->swap_buffers ();
518 }
519
520 static void
521 initial_reordering (const hb_ot_shape_plan_t *plan,
522 hb_font_t *font,
523 hb_buffer_t *buffer)
524 {
525 insert_dotted_circles (plan, font, buffer);
526
527 foreach_syllable (buffer, start, end)
528 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
768 HB_BUFFER_DEALLOCATE_VAR (buffer, khmer_category);
769 HB_BUFFER_DEALLOCATE_VAR (buffer, khmer_position);
770 }
771
772
773 static void
774 clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
775 hb_font_t *font HB_UNUSED,
776 hb_buffer_t *buffer)
777 {
778 hb_glyph_info_t *info = buffer->info;
779 unsigned int count = buffer->len;
780 for (unsigned int i = 0; i < count; i++)
781 info[i].syllable() = 0;
782 }
783
784
785 static bool
786 decompose_khmer (const hb_ot_shape_normalize_context_t *c,
787 hb_codepoint_t ab,
788 hb_codepoint_t *a,
789 hb_codepoint_t *b)
790 {
791 switch (ab)
792 {
793 /*
794 * Decompose split matras that don't have Unicode decompositions.
795 */
796
797 /* Khmer */
798 case 0x17BEu : *a = 0x17C1u; *b= 0x17BEu; return true;
799 case 0x17BFu : *a = 0x17C1u; *b= 0x17BFu; return true;
800 case 0x17C0u : *a = 0x17C1u; *b= 0x17C0u; return true;
801 case 0x17C4u : *a = 0x17C1u; *b= 0x17C4u; return true;
802 case 0x17C5u : *a = 0x17C1u; *b= 0x17C5u; return true;
803 }
804
805 return (bool) c->unicode->decompose (ab, a, b);
806 }
807
808 static bool
809 compose_khmer (const hb_ot_shape_normalize_context_t *c,
810 hb_codepoint_t a,
811 hb_codepoint_t b,
812 hb_codepoint_t *ab)
813 {
814 /* Avoid recomposing split matras. */
815 if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a)))
816 return false;
817
818 return (bool) c->unicode->compose (a, b, ab);
819 }
820
821
822 const hb_ot_complex_shaper_t _hb_ot_complex_shaper_khmer =
823 {
824 collect_features_khmer,
825 override_features_khmer,
826 data_create_khmer,
827 data_destroy_khmer,
828 nullptr, /* preprocess_text */
829 nullptr, /* postprocess_glyphs */
830 HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT,
831 decompose_khmer,
832 compose_khmer,
833 setup_masks_khmer,
834 nullptr, /* disable_otl */
835 nullptr, /* reorder_marks */
836 HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE,
837 false, /* fallback_position */
838 };