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.hh" |
28 | #include "hb-ot-layout.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 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. |
55 | */ |
56 | {HB_TAG('p','r','e','s'), F_GLOBAL}, |
57 | {HB_TAG('a','b','v','s'), F_GLOBAL}, |
58 | {HB_TAG('b','l','w','s'), F_GLOBAL}, |
59 | {HB_TAG('p','s','t','s'), F_GLOBAL}, |
60 | /* Positioning features, though we don't care about the types. */ |
61 | {HB_TAG('d','i','s','t'), F_GLOBAL}, |
62 | {HB_TAG('a','b','v','m'), F_GLOBAL}, |
63 | {HB_TAG('b','l','w','m'), F_GLOBAL}, |
64 | }; |
65 | |
66 | /* |
67 | * Must be in the same order as the khmer_features array. |
68 | */ |
69 | enum { |
70 | PREF, |
71 | BLWF, |
72 | ABVF, |
73 | PSTF, |
74 | CFAR, |
75 | |
76 | _PRES, |
77 | _ABVS, |
78 | _BLWS, |
79 | _PSTS, |
80 | _DIST, |
81 | _ABVM, |
82 | _BLWM, |
83 | |
84 | KHMER_NUM_FEATURES, |
85 | KHMER_BASIC_FEATURES = _PRES /* Don't forget to update this! */ |
86 | }; |
87 | |
88 | static void |
89 | setup_syllables (const hb_ot_shape_plan_t *plan, |
90 | hb_font_t *font, |
91 | hb_buffer_t *buffer); |
92 | static void |
93 | reorder (const hb_ot_shape_plan_t *plan, |
94 | hb_font_t *font, |
95 | hb_buffer_t *buffer); |
96 | static void |
97 | clear_syllables (const hb_ot_shape_plan_t *plan, |
98 | hb_font_t *font, |
99 | hb_buffer_t *buffer); |
100 | |
101 | static void |
102 | collect_features_khmer (hb_ot_shape_planner_t *plan) |
103 | { |
104 | hb_ot_map_builder_t *map = &plan->map; |
105 | |
106 | /* Do this before any lookups have been applied. */ |
107 | map->add_gsub_pause (setup_syllables); |
108 | map->add_gsub_pause (reorder); |
109 | |
110 | /* Testing suggests that Uniscribe does NOT pause between basic |
111 | * features. Test with KhmerUI.ttf and the following three |
112 | * sequences: |
113 | * |
114 | * U+1789,U+17BC |
115 | * U+1789,U+17D2,U+1789 |
116 | * U+1789,U+17D2,U+1789,U+17BC |
117 | * |
118 | * https://github.com/harfbuzz/harfbuzz/issues/974 |
119 | */ |
120 | map->add_global_bool_feature (HB_TAG('l','o','c','l')); |
121 | map->add_global_bool_feature (HB_TAG('c','c','m','p')); |
122 | |
123 | unsigned int i = 0; |
124 | for (; i < KHMER_BASIC_FEATURES; i++) { |
125 | map->add_feature (khmer_features[i].tag, 1, khmer_features[i].flags | F_MANUAL_ZWJ | F_MANUAL_ZWNJ); |
126 | } |
127 | |
128 | map->add_gsub_pause (clear_syllables); |
129 | |
130 | for (; i < KHMER_NUM_FEATURES; i++) { |
131 | map->add_feature (khmer_features[i].tag, 1, khmer_features[i].flags | F_MANUAL_ZWJ | F_MANUAL_ZWNJ); |
132 | } |
133 | |
134 | map->add_global_bool_feature (HB_TAG('c','a','l','t')); |
135 | map->add_global_bool_feature (HB_TAG('c','l','i','g')); |
136 | |
137 | } |
138 | |
139 | static void |
140 | override_features_khmer (hb_ot_shape_planner_t *plan) |
141 | { |
142 | /* Uniscribe does not apply 'kern' in Khmer. */ |
143 | if (hb_options ().uniscribe_bug_compatible) |
144 | { |
145 | plan->map.add_feature (HB_TAG('k','e','r','n'), 0, F_GLOBAL); |
146 | } |
147 | |
148 | plan->map.add_feature (HB_TAG('l','i','g','a'), 0, F_GLOBAL); |
149 | } |
150 | |
151 | |
152 | struct would_substitute_feature_t |
153 | { |
154 | inline void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_) |
155 | { |
156 | zero_context = zero_context_; |
157 | map->get_stage_lookups (0/*GSUB*/, |
158 | map->get_feature_stage (0/*GSUB*/, feature_tag), |
159 | &lookups, &count); |
160 | } |
161 | |
162 | inline bool would_substitute (const hb_codepoint_t *glyphs, |
163 | unsigned int glyphs_count, |
164 | hb_face_t *face) const |
165 | { |
166 | for (unsigned int i = 0; i < count; i++) |
167 | if (hb_ot_layout_lookup_would_substitute_fast (face, lookups[i].index, glyphs, glyphs_count, zero_context)) |
168 | return true; |
169 | return false; |
170 | } |
171 | |
172 | private: |
173 | const hb_ot_map_t::lookup_map_t *lookups; |
174 | unsigned int count; |
175 | bool zero_context; |
176 | }; |
177 | |
178 | struct khmer_shape_plan_t |
179 | { |
180 | ASSERT_POD (); |
181 | |
182 | inline bool get_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const |
183 | { |
184 | hb_codepoint_t glyph = virama_glyph; |
185 | if (unlikely (virama_glyph == (hb_codepoint_t) -1)) |
186 | { |
187 | if (!font->get_nominal_glyph (0x17D2u, &glyph)) |
188 | glyph = 0; |
189 | /* Technically speaking, the spec says we should apply 'locl' to virama too. |
190 | * Maybe one day... */ |
191 | |
192 | /* Our get_nominal_glyph() function needs a font, so we can't get the virama glyph |
193 | * during shape planning... Instead, overwrite it here. It's safe. Don't worry! */ |
194 | virama_glyph = glyph; |
195 | } |
196 | |
197 | *pglyph = glyph; |
198 | return glyph != 0; |
199 | } |
200 | |
201 | mutable hb_codepoint_t virama_glyph; |
202 | |
203 | would_substitute_feature_t pref; |
204 | |
205 | hb_mask_t mask_array[KHMER_NUM_FEATURES]; |
206 | }; |
207 | |
208 | static void * |
209 | data_create_khmer (const hb_ot_shape_plan_t *plan) |
210 | { |
211 | khmer_shape_plan_t *khmer_plan = (khmer_shape_plan_t *) calloc (1, sizeof (khmer_shape_plan_t)); |
212 | if (unlikely (!khmer_plan)) |
213 | return nullptr; |
214 | |
215 | khmer_plan->virama_glyph = (hb_codepoint_t) -1; |
216 | |
217 | khmer_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), true); |
218 | |
219 | for (unsigned int i = 0; i < ARRAY_LENGTH (khmer_plan->mask_array); i++) |
220 | khmer_plan->mask_array[i] = (khmer_features[i].flags & F_GLOBAL) ? |
221 | 0 : plan->map.get_1_mask (khmer_features[i].tag); |
222 | |
223 | return khmer_plan; |
224 | } |
225 | |
226 | static void |
227 | data_destroy_khmer (void *data) |
228 | { |
229 | free (data); |
230 | } |
231 | |
232 | |
233 | enum syllable_type_t { |
234 | consonant_syllable, |
235 | broken_cluster, |
236 | non_khmer_cluster, |
237 | }; |
238 | |
239 | #include "hb-ot-shape-complex-khmer-machine.hh" |
240 | |
241 | static void |
242 | setup_masks_khmer (const hb_ot_shape_plan_t *plan HB_UNUSED, |
243 | hb_buffer_t *buffer, |
244 | hb_font_t *font HB_UNUSED) |
245 | { |
246 | HB_BUFFER_ALLOCATE_VAR (buffer, khmer_category); |
247 | HB_BUFFER_ALLOCATE_VAR (buffer, khmer_position); |
248 | |
249 | /* We cannot setup masks here. We save information about characters |
250 | * and setup masks later on in a pause-callback. */ |
251 | |
252 | unsigned int count = buffer->len; |
253 | hb_glyph_info_t *info = buffer->info; |
254 | for (unsigned int i = 0; i < count; i++) |
255 | set_khmer_properties (info[i]); |
256 | } |
257 | |
258 | static void |
259 | setup_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED, |
260 | hb_font_t *font HB_UNUSED, |
261 | hb_buffer_t *buffer) |
262 | { |
263 | find_syllables (buffer); |
264 | foreach_syllable (buffer, start, end) |
265 | buffer->unsafe_to_break (start, end); |
266 | } |
267 | |
268 | |
269 | /* Rules from: |
270 | * https://docs.microsoft.com/en-us/typography/script-development/devanagari */ |
271 | |
272 | static void |
273 | reorder_consonant_syllable (const hb_ot_shape_plan_t *plan, |
274 | hb_face_t *face, |
275 | hb_buffer_t *buffer, |
276 | unsigned int start, unsigned int end) |
277 | { |
278 | const khmer_shape_plan_t *khmer_plan = (const khmer_shape_plan_t *) plan->data; |
279 | hb_glyph_info_t *info = buffer->info; |
280 | |
281 | /* Setup masks. */ |
282 | { |
283 | /* Post-base */ |
284 | hb_mask_t mask = khmer_plan->mask_array[BLWF] | khmer_plan->mask_array[ABVF] | khmer_plan->mask_array[PSTF]; |
285 | for (unsigned int i = start + 1; i < end; i++) |
286 | info[i].mask |= mask; |
287 | } |
288 | |
289 | unsigned int num_coengs = 0; |
290 | for (unsigned int i = start + 1; i < end; i++) |
291 | { |
292 | /* """ |
293 | * When a COENG + (Cons | IndV) combination are found (and subscript count |
294 | * is less than two) the character combination is handled according to the |
295 | * subscript type of the character following the COENG. |
296 | * |
297 | * ... |
298 | * |
299 | * Subscript Type 2 - The COENG + RO characters are reordered to immediately |
300 | * before the base glyph. Then the COENG + RO characters are assigned to have |
301 | * the 'pref' OpenType feature applied to them. |
302 | * """ |
303 | */ |
304 | if (info[i].khmer_category() == OT_Coeng && num_coengs <= 2 && i + 1 < end) |
305 | { |
306 | num_coengs++; |
307 | |
308 | if (info[i + 1].khmer_category() == OT_Ra) |
309 | { |
310 | for (unsigned int j = 0; j < 2; j++) |
311 | info[i + j].mask |= khmer_plan->mask_array[PREF]; |
312 | |
313 | /* Move the Coeng,Ro sequence to the start. */ |
314 | buffer->merge_clusters (start, i + 2); |
315 | hb_glyph_info_t t0 = info[i]; |
316 | hb_glyph_info_t t1 = info[i + 1]; |
317 | memmove (&info[start + 2], &info[start], (i - start) * sizeof (info[0])); |
318 | info[start] = t0; |
319 | info[start + 1] = t1; |
320 | |
321 | /* Mark the subsequent stuff with 'cfar'. Used in Khmer. |
322 | * Read the feature spec. |
323 | * This allows distinguishing the following cases with MS Khmer fonts: |
324 | * U+1784,U+17D2,U+179A,U+17D2,U+1782 |
325 | * U+1784,U+17D2,U+1782,U+17D2,U+179A |
326 | */ |
327 | if (khmer_plan->mask_array[CFAR]) |
328 | for (unsigned int j = i + 2; j < end; j++) |
329 | info[j].mask |= khmer_plan->mask_array[CFAR]; |
330 | |
331 | num_coengs = 2; /* Done. */ |
332 | } |
333 | } |
334 | |
335 | /* Reorder left matra piece. */ |
336 | else if (info[i].khmer_position() == POS_PRE_M) |
337 | { |
338 | /* Move to the start. */ |
339 | buffer->merge_clusters (start, i + 1); |
340 | hb_glyph_info_t t = info[i]; |
341 | memmove (&info[start + 1], &info[start], (i - start) * sizeof (info[0])); |
342 | info[start] = t; |
343 | } |
344 | } |
345 | } |
346 | |
347 | static void |
348 | initial_reordering_syllable (const hb_ot_shape_plan_t *plan, |
349 | hb_face_t *face, |
350 | hb_buffer_t *buffer, |
351 | unsigned int start, unsigned int end) |
352 | { |
353 | syllable_type_t syllable_type = (syllable_type_t) (buffer->info[start].syllable() & 0x0F); |
354 | switch (syllable_type) |
355 | { |
356 | case broken_cluster: /* We already inserted dotted-circles, so just call the consonant_syllable. */ |
357 | case consonant_syllable: |
358 | reorder_consonant_syllable (plan, face, buffer, start, end); |
359 | break; |
360 | |
361 | case non_khmer_cluster: |
362 | break; |
363 | } |
364 | } |
365 | |
366 | static inline void |
367 | insert_dotted_circles (const hb_ot_shape_plan_t *plan HB_UNUSED, |
368 | hb_font_t *font, |
369 | hb_buffer_t *buffer) |
370 | { |
371 | /* Note: This loop is extra overhead, but should not be measurable. */ |
372 | bool has_broken_syllables = false; |
373 | unsigned int count = buffer->len; |
374 | hb_glyph_info_t *info = buffer->info; |
375 | for (unsigned int i = 0; i < count; i++) |
376 | if ((info[i].syllable() & 0x0F) == broken_cluster) |
377 | { |
378 | has_broken_syllables = true; |
379 | break; |
380 | } |
381 | if (likely (!has_broken_syllables)) |
382 | return; |
383 | |
384 | |
385 | hb_codepoint_t dottedcircle_glyph; |
386 | if (!font->get_nominal_glyph (0x25CCu, &dottedcircle_glyph)) |
387 | return; |
388 | |
389 | hb_glyph_info_t dottedcircle = {0}; |
390 | dottedcircle.codepoint = 0x25CCu; |
391 | set_khmer_properties (dottedcircle); |
392 | dottedcircle.codepoint = dottedcircle_glyph; |
393 | |
394 | buffer->clear_output (); |
395 | |
396 | buffer->idx = 0; |
397 | unsigned int last_syllable = 0; |
398 | while (buffer->idx < buffer->len && buffer->successful) |
399 | { |
400 | unsigned int syllable = buffer->cur().syllable(); |
401 | syllable_type_t syllable_type = (syllable_type_t) (syllable & 0x0F); |
402 | if (unlikely (last_syllable != syllable && syllable_type == broken_cluster)) |
403 | { |
404 | last_syllable = syllable; |
405 | |
406 | hb_glyph_info_t ginfo = dottedcircle; |
407 | ginfo.cluster = buffer->cur().cluster; |
408 | ginfo.mask = buffer->cur().mask; |
409 | ginfo.syllable() = buffer->cur().syllable(); |
410 | /* TODO Set glyph_props? */ |
411 | |
412 | /* Insert dottedcircle after possible Repha. */ |
413 | while (buffer->idx < buffer->len && buffer->successful && |
414 | last_syllable == buffer->cur().syllable() && |
415 | buffer->cur().khmer_category() == OT_Repha) |
416 | buffer->next_glyph (); |
417 | |
418 | buffer->output_info (ginfo); |
419 | } |
420 | else |
421 | buffer->next_glyph (); |
422 | } |
423 | |
424 | buffer->swap_buffers (); |
425 | } |
426 | |
427 | static void |
428 | reorder (const hb_ot_shape_plan_t *plan, |
429 | hb_font_t *font, |
430 | hb_buffer_t *buffer) |
431 | { |
432 | insert_dotted_circles (plan, font, buffer); |
433 | |
434 | foreach_syllable (buffer, start, end) |
435 | initial_reordering_syllable (plan, font->face, buffer, start, end); |
436 | |
437 | HB_BUFFER_DEALLOCATE_VAR (buffer, khmer_category); |
438 | HB_BUFFER_DEALLOCATE_VAR (buffer, khmer_position); |
439 | } |
440 | |
441 | static void |
442 | clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED, |
443 | hb_font_t *font HB_UNUSED, |
444 | hb_buffer_t *buffer) |
445 | { |
446 | /* TODO: In USE, we clear syllables right after reorder. Figure out |
447 | * what Uniscribe does. */ |
448 | hb_glyph_info_t *info = buffer->info; |
449 | unsigned int count = buffer->len; |
450 | for (unsigned int i = 0; i < count; i++) |
451 | info[i].syllable() = 0; |
452 | } |
453 | |
454 | |
455 | static bool |
456 | decompose_khmer (const hb_ot_shape_normalize_context_t *c, |
457 | hb_codepoint_t ab, |
458 | hb_codepoint_t *a, |
459 | hb_codepoint_t *b) |
460 | { |
461 | switch (ab) |
462 | { |
463 | /* |
464 | * Decompose split matras that don't have Unicode decompositions. |
465 | */ |
466 | |
467 | /* Khmer */ |
468 | case 0x17BEu : *a = 0x17C1u; *b= 0x17BEu; return true; |
469 | case 0x17BFu : *a = 0x17C1u; *b= 0x17BFu; return true; |
470 | case 0x17C0u : *a = 0x17C1u; *b= 0x17C0u; return true; |
471 | case 0x17C4u : *a = 0x17C1u; *b= 0x17C4u; return true; |
472 | case 0x17C5u : *a = 0x17C1u; *b= 0x17C5u; return true; |
473 | } |
474 | |
475 | return (bool) c->unicode->decompose (ab, a, b); |
476 | } |
477 | |
478 | static bool |
479 | compose_khmer (const hb_ot_shape_normalize_context_t *c, |
480 | hb_codepoint_t a, |
481 | hb_codepoint_t b, |
482 | hb_codepoint_t *ab) |
483 | { |
484 | /* Avoid recomposing split matras. */ |
485 | if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a))) |
486 | return false; |
487 | |
488 | return (bool) c->unicode->compose (a, b, ab); |
489 | } |
490 | |
491 | |
492 | const hb_ot_complex_shaper_t _hb_ot_complex_shaper_khmer = |
493 | { |
494 | collect_features_khmer, |
495 | override_features_khmer, |
496 | data_create_khmer, |
497 | data_destroy_khmer, |
498 | nullptr, /* preprocess_text */ |
499 | nullptr, /* postprocess_glyphs */ |
500 | HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT, |
501 | decompose_khmer, |
502 | compose_khmer, |
503 | setup_masks_khmer, |
504 | nullptr, /* disable_otl */ |
505 | nullptr, /* reorder_marks */ |
506 | HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE, |
507 | false, /* fallback_position */ |
508 | }; |
509 | |