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-indic.hh" |
28 | #include "hb-ot-layout.hh" |
29 | |
30 | |
31 | /* |
32 | * Indic shaper. |
33 | */ |
34 | |
35 | |
36 | /* |
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. |
42 | */ |
43 | |
44 | enum base_position_t { |
45 | BASE_POS_LAST_SINHALA, |
46 | BASE_POS_LAST |
47 | }; |
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 |
54 | }; |
55 | enum reph_mode_t { |
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. */ |
59 | }; |
60 | enum blwf_mode_t { |
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. */ |
63 | }; |
64 | struct indic_config_t |
65 | { |
66 | hb_script_t script; |
67 | bool has_old_spec; |
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; |
73 | }; |
74 | |
75 | static const indic_config_t indic_configs[] = |
76 | { |
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}, |
90 | }; |
91 | |
92 | |
93 | |
94 | /* |
95 | * Indic shaper. |
96 | */ |
97 | |
98 | struct feature_list_t { |
99 | hb_tag_t tag; |
100 | hb_ot_map_feature_flags_t flags; |
101 | }; |
102 | |
103 | static const feature_list_t |
104 | indic_features[] = |
105 | { |
106 | /* |
107 | * Basic features. |
108 | * These features are applied in order, one at a time, after initial_reordering. |
109 | */ |
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}, |
121 | /* |
122 | * Other features. |
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. |
126 | */ |
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}, |
137 | }; |
138 | |
139 | /* |
140 | * Must be in the same order as the indic_features array. |
141 | */ |
142 | enum { |
143 | _NUKT, |
144 | _AKHN, |
145 | RPHF, |
146 | _RKRF, |
147 | PREF, |
148 | BLWF, |
149 | ABVF, |
150 | HALF, |
151 | PSTF, |
152 | _VATU, |
153 | _CJCT, |
154 | |
155 | INIT, |
156 | _PRES, |
157 | _ABVS, |
158 | _BLWS, |
159 | _PSTS, |
160 | _HALN, |
161 | _DIST, |
162 | _ABVM, |
163 | _BLWM, |
164 | |
165 | INDIC_NUM_FEATURES, |
166 | INDIC_BASIC_FEATURES = INIT /* Don't forget to update this! */ |
167 | }; |
168 | |
169 | static void |
170 | setup_syllables (const hb_ot_shape_plan_t *plan, |
171 | hb_font_t *font, |
172 | hb_buffer_t *buffer); |
173 | static void |
174 | initial_reordering (const hb_ot_shape_plan_t *plan, |
175 | hb_font_t *font, |
176 | hb_buffer_t *buffer); |
177 | static void |
178 | final_reordering (const hb_ot_shape_plan_t *plan, |
179 | hb_font_t *font, |
180 | hb_buffer_t *buffer); |
181 | static void |
182 | clear_syllables (const hb_ot_shape_plan_t *plan, |
183 | hb_font_t *font, |
184 | hb_buffer_t *buffer); |
185 | |
186 | static void |
187 | collect_features_indic (hb_ot_shape_planner_t *plan) |
188 | { |
189 | hb_ot_map_builder_t *map = &plan->map; |
190 | |
191 | /* Do this before any lookups have been applied. */ |
192 | map->add_gsub_pause (setup_syllables); |
193 | |
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')); |
198 | |
199 | |
200 | unsigned int i = 0; |
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); |
205 | } |
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); |
209 | } |
210 | |
211 | map->add_global_bool_feature (HB_TAG('c','a','l','t')); |
212 | map->add_global_bool_feature (HB_TAG('c','l','i','g')); |
213 | |
214 | map->add_gsub_pause (clear_syllables); |
215 | } |
216 | |
217 | static void |
218 | override_features_indic (hb_ot_shape_planner_t *plan) |
219 | { |
220 | plan->map.add_feature (HB_TAG('l','i','g','a'), 0, F_GLOBAL); |
221 | } |
222 | |
223 | |
224 | struct would_substitute_feature_t |
225 | { |
226 | inline void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_) |
227 | { |
228 | zero_context = zero_context_; |
229 | map->get_stage_lookups (0/*GSUB*/, |
230 | map->get_feature_stage (0/*GSUB*/, feature_tag), |
231 | &lookups, &count); |
232 | } |
233 | |
234 | inline bool would_substitute (const hb_codepoint_t *glyphs, |
235 | unsigned int glyphs_count, |
236 | hb_face_t *face) const |
237 | { |
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)) |
240 | return true; |
241 | return false; |
242 | } |
243 | |
244 | private: |
245 | const hb_ot_map_t::lookup_map_t *lookups; |
246 | unsigned int count; |
247 | bool zero_context; |
248 | }; |
249 | |
250 | struct indic_shape_plan_t |
251 | { |
252 | ASSERT_POD (); |
253 | |
254 | inline bool load_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const |
255 | { |
256 | hb_codepoint_t glyph = virama_glyph.get_relaxed (); |
257 | if (unlikely (glyph == (hb_codepoint_t) -1)) |
258 | { |
259 | if (!config->virama || !font->get_nominal_glyph (config->virama, &glyph)) |
260 | glyph = 0; |
261 | /* Technically speaking, the spec says we should apply 'locl' to virama too. |
262 | * Maybe one day... */ |
263 | |
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. */ |
266 | virama_glyph.set_relaxed ((int) glyph); |
267 | } |
268 | |
269 | *pglyph = glyph; |
270 | return glyph != 0; |
271 | } |
272 | |
273 | const indic_config_t *config; |
274 | |
275 | bool is_old_spec; |
276 | mutable hb_atomic_int_t virama_glyph; |
277 | |
278 | would_substitute_feature_t rphf; |
279 | would_substitute_feature_t pref; |
280 | would_substitute_feature_t blwf; |
281 | would_substitute_feature_t pstf; |
282 | |
283 | hb_mask_t mask_array[INDIC_NUM_FEATURES]; |
284 | }; |
285 | |
286 | static void * |
287 | data_create_indic (const hb_ot_shape_plan_t *plan) |
288 | { |
289 | indic_shape_plan_t *indic_plan = (indic_shape_plan_t *) calloc (1, sizeof (indic_shape_plan_t)); |
290 | if (unlikely (!indic_plan)) |
291 | return nullptr; |
292 | |
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]; |
297 | break; |
298 | } |
299 | |
300 | indic_plan->is_old_spec = indic_plan->config->has_old_spec && ((plan->map.chosen_script[0] & 0x000000FFu) != '2'); |
301 | indic_plan->virama_glyph.set_relaxed (-1); |
302 | |
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. |
306 | * |
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. |
311 | */ |
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); |
317 | |
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); |
321 | |
322 | return indic_plan; |
323 | } |
324 | |
325 | static void |
326 | data_destroy_indic (void *data) |
327 | { |
328 | free (data); |
329 | } |
330 | |
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, |
335 | hb_face_t *face) |
336 | { |
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 |
346 | * to work. */ |
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)) |
350 | return POS_BELOW_C; |
351 | if (indic_plan->pstf.would_substitute (glyphs , 2, face) || |
352 | indic_plan->pstf.would_substitute (glyphs+1, 2, face)) |
353 | return POS_POST_C; |
354 | if (indic_plan->pref.would_substitute (glyphs , 2, face) || |
355 | indic_plan->pref.would_substitute (glyphs+1, 2, face)) |
356 | return POS_POST_C; |
357 | return POS_BASE_C; |
358 | } |
359 | |
360 | |
361 | enum syllable_type_t { |
362 | consonant_syllable, |
363 | vowel_syllable, |
364 | standalone_cluster, |
365 | symbol_cluster, |
366 | broken_cluster, |
367 | non_indic_cluster, |
368 | }; |
369 | |
370 | #include "hb-ot-shape-complex-indic-machine.hh" |
371 | |
372 | |
373 | static void |
374 | setup_masks_indic (const hb_ot_shape_plan_t *plan HB_UNUSED, |
375 | hb_buffer_t *buffer, |
376 | hb_font_t *font HB_UNUSED) |
377 | { |
378 | HB_BUFFER_ALLOCATE_VAR (buffer, indic_category); |
379 | HB_BUFFER_ALLOCATE_VAR (buffer, indic_position); |
380 | |
381 | /* We cannot setup masks here. We save information about characters |
382 | * and setup masks later on in a pause-callback. */ |
383 | |
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]); |
388 | } |
389 | |
390 | static void |
391 | setup_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED, |
392 | hb_font_t *font HB_UNUSED, |
393 | hb_buffer_t *buffer) |
394 | { |
395 | find_syllables (buffer); |
396 | foreach_syllable (buffer, start, end) |
397 | buffer->unsafe_to_break (start, end); |
398 | } |
399 | |
400 | static int |
401 | compare_indic_order (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb) |
402 | { |
403 | int a = pa->indic_position(); |
404 | int b = pb->indic_position(); |
405 | |
406 | return a < b ? -1 : a == b ? 0 : +1; |
407 | } |
408 | |
409 | |
410 | |
411 | static void |
412 | update_consonant_positions (const hb_ot_shape_plan_t *plan, |
413 | hb_font_t *font, |
414 | hb_buffer_t *buffer) |
415 | { |
416 | const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; |
417 | |
418 | if (indic_plan->config->base_pos != BASE_POS_LAST) |
419 | return; |
420 | |
421 | hb_codepoint_t virama; |
422 | if (indic_plan->load_virama_glyph (font, &virama)) |
423 | { |
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) |
429 | { |
430 | hb_codepoint_t consonant = info[i].codepoint; |
431 | info[i].indic_position() = consonant_position_from_face (indic_plan, consonant, virama, face); |
432 | } |
433 | } |
434 | } |
435 | |
436 | |
437 | /* Rules from: |
438 | * https://docs.microsqoft.com/en-us/typography/script-development/devanagari */ |
439 | |
440 | static void |
441 | initial_reordering_consonant_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 | const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; |
447 | hb_glyph_info_t *info = buffer->info; |
448 | |
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... |
452 | */ |
453 | if (buffer->props.script == HB_SCRIPT_KANNADA && |
454 | start + 3 <= end && |
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))) |
458 | { |
459 | buffer->merge_clusters (start+1, start+3); |
460 | hb_glyph_info_t tmp = info[start+1]; |
461 | info[start+1] = info[start+2]; |
462 | info[start+2] = tmp; |
463 | } |
464 | |
465 | /* 1. Find base consonant: |
466 | * |
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. |
473 | * |
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 |
476 | * base consonants. |
477 | */ |
478 | |
479 | unsigned int base = end; |
480 | bool has_reph = false; |
481 | |
482 | { |
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] && |
488 | start + 3 <= end && |
489 | ( |
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) |
492 | )) |
493 | { |
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))) |
502 | { |
503 | limit += 2; |
504 | while (limit < end && is_joiner (info[limit])) |
505 | limit++; |
506 | base = start; |
507 | has_reph = true; |
508 | } |
509 | } else if (indic_plan->config->reph_mode == REPH_MODE_LOG_REPHA && info[start].indic_category() == OT_Repha) |
510 | { |
511 | limit += 1; |
512 | while (limit < end && is_joiner (info[limit])) |
513 | limit++; |
514 | base = start; |
515 | has_reph = true; |
516 | } |
517 | |
518 | switch (indic_plan->config->base_pos) |
519 | { |
520 | case BASE_POS_LAST: |
521 | { |
522 | /* -> starting from the end of the syllable, move backwards */ |
523 | unsigned int i = end; |
524 | bool seen_below = false; |
525 | do { |
526 | i--; |
527 | /* -> until a consonant is found */ |
528 | if (is_consonant (info[i])) |
529 | { |
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)) |
534 | { |
535 | base = i; |
536 | break; |
537 | } |
538 | if (info[i].indic_position() == POS_BELOW_C) |
539 | seen_below = true; |
540 | |
541 | /* -> or that is not a pre-base-reordering Ra, |
542 | * |
543 | * IMPLEMENTATION NOTES: |
544 | * |
545 | * Our pre-base-reordering Ra's are marked POS_POST_C, so will be skipped |
546 | * by the logic above already. |
547 | */ |
548 | |
549 | /* -> or arrive at the first consonant. The consonant stopped at will |
550 | * be the base. */ |
551 | base = i; |
552 | } |
553 | else |
554 | { |
555 | /* A ZWJ after a Halant stops the base search, and requests an explicit |
556 | * half form. |
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. */ |
560 | if (start < i && |
561 | info[i].indic_category() == OT_ZWJ && |
562 | info[i - 1].indic_category() == OT_H) |
563 | break; |
564 | } |
565 | } while (i > limit); |
566 | } |
567 | break; |
568 | |
569 | case BASE_POS_LAST_SINHALA: |
570 | { |
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. |
574 | */ |
575 | |
576 | if (!has_reph) |
577 | base = limit; |
578 | |
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])) |
583 | { |
584 | if (limit < i && info[i - 1].indic_category() == OT_ZWJ) |
585 | break; |
586 | else |
587 | base = i; |
588 | } |
589 | |
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; |
594 | } |
595 | break; |
596 | } |
597 | |
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 |
600 | * base consonants. |
601 | * |
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. */ |
605 | has_reph = false; |
606 | } |
607 | } |
608 | |
609 | |
610 | /* 2. Decompose and reorder Matras: |
611 | * |
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: |
618 | * |
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) |
623 | * |
624 | * IMPLEMENTATION NOTES: |
625 | * |
626 | * The normalize() routine has already decomposed matras for us, so we don't |
627 | * need to worry about that. |
628 | */ |
629 | |
630 | |
631 | /* 3. Reorder marks to canonical order: |
632 | * |
633 | * Adjacent nukta and halant or nukta and vedic sign are always repositioned |
634 | * if necessary, so that the nukta is first. |
635 | * |
636 | * IMPLEMENTATION NOTES: |
637 | * |
638 | * We don't need to do this: the normalize() routine already did this for us. |
639 | */ |
640 | |
641 | |
642 | /* Reorder characters */ |
643 | |
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()); |
646 | |
647 | if (base < end) |
648 | info[base].indic_position() = POS_BASE_C; |
649 | |
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; |
657 | break; |
658 | } |
659 | break; |
660 | } |
661 | |
662 | /* Handle beginning Ra */ |
663 | if (has_reph) |
664 | info[start].indic_position() = POS_RA_TO_BECOME_REPH; |
665 | |
666 | /* For old-style Indic script tags, move the first post-base Halant after |
667 | * last consonant. |
668 | * |
669 | * Reports suggest that in some scripts Uniscribe does this only if there |
670 | * is *not* a Halant after last consonant already. We know that is the |
671 | * case for Kannada, while it reorders unconditionally in other scripts, |
672 | * eg. Malayalam, Bengali, and Devanagari. We don't currently know about |
673 | * other scripts, so we blacklist Kannada. |
674 | * |
675 | * Kannada test case: |
676 | * U+0C9A,U+0CCD,U+0C9A,U+0CCD |
677 | * With some versions of Lohit Kannada. |
678 | * https://bugs.freedesktop.org/show_bug.cgi?id=59118 |
679 | * |
680 | * Malayalam test case: |
681 | * U+0D38,U+0D4D,U+0D31,U+0D4D,U+0D31,U+0D4D |
682 | * With lohit-ttf-20121122/Lohit-Malayalam.ttf |
683 | * |
684 | * Bengali test case: |
685 | * U+0998,U+09CD,U+09AF,U+09CD |
686 | * With Windows XP vrinda.ttf |
687 | * https://github.com/harfbuzz/harfbuzz/issues/1073 |
688 | * |
689 | * Devanagari test case: |
690 | * U+091F,U+094D,U+0930,U+094D |
691 | * With chandas.ttf |
692 | * https://github.com/harfbuzz/harfbuzz/issues/1071 |
693 | */ |
694 | if (indic_plan->is_old_spec) |
695 | { |
696 | bool disallow_double_halants = buffer->props.script == HB_SCRIPT_KANNADA; |
697 | for (unsigned int i = base + 1; i < end; i++) |
698 | if (info[i].indic_category() == OT_H) |
699 | { |
700 | unsigned int j; |
701 | for (j = end - 1; j > i; j--) |
702 | if (is_consonant (info[j]) || |
703 | (disallow_double_halants && info[j].indic_category() == OT_H)) |
704 | break; |
705 | if (info[j].indic_category() != OT_H && j > i) { |
706 | /* Move Halant to after last consonant. */ |
707 | hb_glyph_info_t t = info[i]; |
708 | memmove (&info[i], &info[i + 1], (j - i) * sizeof (info[0])); |
709 | info[j] = t; |
710 | } |
711 | break; |
712 | } |
713 | } |
714 | |
715 | /* Attach misc marks to previous char to move with them. */ |
716 | { |
717 | indic_position_t last_pos = POS_START; |
718 | for (unsigned int i = start; i < end; i++) |
719 | { |
720 | if ((FLAG_UNSAFE (info[i].indic_category()) & (JOINER_FLAGS | FLAG (OT_N) | FLAG (OT_RS) | MEDIAL_FLAGS | FLAG (OT_H)))) |
721 | { |
722 | info[i].indic_position() = last_pos; |
723 | if (unlikely (info[i].indic_category() == OT_H && |
724 | info[i].indic_position() == POS_PRE_M)) |
725 | { |
726 | /* |
727 | * Uniscribe doesn't move the Halant with Left Matra. |
728 | * TEST: U+092B,U+093F,U+094DE |
729 | * We follow. This is important for the Sinhala |
730 | * U+0DDA split matra since it decomposes to U+0DD9,U+0DCA |
731 | * where U+0DD9 is a left matra and U+0DCA is the virama. |
732 | * We don't want to move the virama with the left matra. |
733 | * TEST: U+0D9A,U+0DDA |
734 | */ |
735 | for (unsigned int j = i; j > start; j--) |
736 | if (info[j - 1].indic_position() != POS_PRE_M) { |
737 | info[i].indic_position() = info[j - 1].indic_position(); |
738 | break; |
739 | } |
740 | } |
741 | } else if (info[i].indic_position() != POS_SMVD) { |
742 | last_pos = (indic_position_t) info[i].indic_position(); |
743 | } |
744 | } |
745 | } |
746 | /* For post-base consonants let them own anything before them |
747 | * since the last consonant or matra. */ |
748 | { |
749 | unsigned int last = base; |
750 | for (unsigned int i = base + 1; i < end; i++) |
751 | if (is_consonant (info[i])) |
752 | { |
753 | for (unsigned int j = last + 1; j < i; j++) |
754 | if (info[j].indic_position() < POS_SMVD) |
755 | info[j].indic_position() = info[i].indic_position(); |
756 | last = i; |
757 | } else if (info[i].indic_category() == OT_M) |
758 | last = i; |
759 | } |
760 | |
761 | |
762 | { |
763 | /* Use syllable() for sort accounting temporarily. */ |
764 | unsigned int syllable = info[start].syllable(); |
765 | for (unsigned int i = start; i < end; i++) |
766 | info[i].syllable() = i - start; |
767 | |
768 | /* Sit tight, rock 'n roll! */ |
769 | hb_stable_sort (info + start, end - start, compare_indic_order); |
770 | /* Find base again */ |
771 | base = end; |
772 | for (unsigned int i = start; i < end; i++) |
773 | if (info[i].indic_position() == POS_BASE_C) |
774 | { |
775 | base = i; |
776 | break; |
777 | } |
778 | /* Things are out-of-control for post base positions, they may shuffle |
779 | * around like crazy. In old-spec mode, we move halants around, so in |
780 | * that case merge all clusters after base. Otherwise, check the sort |
781 | * order and merge as needed. |
782 | * For pre-base stuff, we handle cluster issues in final reordering. |
783 | * |
784 | * We could use buffer->sort() for this, if there was no special |
785 | * reordering of pre-base stuff happening later... |
786 | */ |
787 | if (indic_plan->is_old_spec || end - base > 127) |
788 | buffer->merge_clusters (base, end); |
789 | else |
790 | { |
791 | /* Note! syllable() is a one-byte field. */ |
792 | for (unsigned int i = base; i < end; i++) |
793 | if (info[i].syllable() != 255) |
794 | { |
795 | unsigned int max = i; |
796 | unsigned int j = start + info[i].syllable(); |
797 | while (j != i) |
798 | { |
799 | max = MAX (max, j); |
800 | unsigned int next = start + info[j].syllable(); |
801 | info[j].syllable() = 255; /* So we don't process j later again. */ |
802 | j = next; |
803 | } |
804 | if (i != max) |
805 | buffer->merge_clusters (i, max + 1); |
806 | } |
807 | } |
808 | |
809 | /* Put syllable back in. */ |
810 | for (unsigned int i = start; i < end; i++) |
811 | info[i].syllable() = syllable; |
812 | } |
813 | |
814 | /* Setup masks now */ |
815 | |
816 | { |
817 | hb_mask_t mask; |
818 | |
819 | /* Reph */ |
820 | for (unsigned int i = start; i < end && info[i].indic_position() == POS_RA_TO_BECOME_REPH; i++) |
821 | info[i].mask |= indic_plan->mask_array[RPHF]; |
822 | |
823 | /* Pre-base */ |
824 | mask = indic_plan->mask_array[HALF]; |
825 | if (!indic_plan->is_old_spec && |
826 | indic_plan->config->blwf_mode == BLWF_MODE_PRE_AND_POST) |
827 | mask |= indic_plan->mask_array[BLWF]; |
828 | for (unsigned int i = start; i < base; i++) |
829 | info[i].mask |= mask; |
830 | /* Base */ |
831 | mask = 0; |
832 | if (base < end) |
833 | info[base].mask |= mask; |
834 | /* Post-base */ |
835 | mask = indic_plan->mask_array[BLWF] | indic_plan->mask_array[ABVF] | indic_plan->mask_array[PSTF]; |
836 | for (unsigned int i = base + 1; i < end; i++) |
837 | info[i].mask |= mask; |
838 | } |
839 | |
840 | if (indic_plan->is_old_spec && |
841 | buffer->props.script == HB_SCRIPT_DEVANAGARI) |
842 | { |
843 | /* Old-spec eye-lash Ra needs special handling. From the |
844 | * spec: |
845 | * |
846 | * "The feature 'below-base form' is applied to consonants |
847 | * having below-base forms and following the base consonant. |
848 | * The exception is vattu, which may appear below half forms |
849 | * as well as below the base glyph. The feature 'below-base |
850 | * form' will be applied to all such occurrences of Ra as well." |
851 | * |
852 | * Test case: U+0924,U+094D,U+0930,U+094d,U+0915 |
853 | * with Sanskrit 2003 font. |
854 | * |
855 | * However, note that Ra,Halant,ZWJ is the correct way to |
856 | * request eyelash form of Ra, so we wouldbn't inhibit it |
857 | * in that sequence. |
858 | * |
859 | * Test case: U+0924,U+094D,U+0930,U+094d,U+200D,U+0915 |
860 | */ |
861 | for (unsigned int i = start; i + 1 < base; i++) |
862 | if (info[i ].indic_category() == OT_Ra && |
863 | info[i+1].indic_category() == OT_H && |
864 | (i + 2 == base || |
865 | info[i+2].indic_category() != OT_ZWJ)) |
866 | { |
867 | info[i ].mask |= indic_plan->mask_array[BLWF]; |
868 | info[i+1].mask |= indic_plan->mask_array[BLWF]; |
869 | } |
870 | } |
871 | |
872 | unsigned int pref_len = 2; |
873 | if (indic_plan->mask_array[PREF] && base + pref_len < end) |
874 | { |
875 | /* Find a Halant,Ra sequence and mark it for pre-base-reordering processing. */ |
876 | for (unsigned int i = base + 1; i + pref_len - 1 < end; i++) { |
877 | hb_codepoint_t glyphs[2]; |
878 | for (unsigned int j = 0; j < pref_len; j++) |
879 | glyphs[j] = info[i + j].codepoint; |
880 | if (indic_plan->pref.would_substitute (glyphs, pref_len, face)) |
881 | { |
882 | for (unsigned int j = 0; j < pref_len; j++) |
883 | info[i++].mask |= indic_plan->mask_array[PREF]; |
884 | break; |
885 | } |
886 | } |
887 | } |
888 | |
889 | /* Apply ZWJ/ZWNJ effects */ |
890 | for (unsigned int i = start + 1; i < end; i++) |
891 | if (is_joiner (info[i])) { |
892 | bool non_joiner = info[i].indic_category() == OT_ZWNJ; |
893 | unsigned int j = i; |
894 | |
895 | do { |
896 | j--; |
897 | |
898 | /* ZWJ/ZWNJ should disable CJCT. They do that by simply |
899 | * being there, since we don't skip them for the CJCT |
900 | * feature (ie. F_MANUAL_ZWJ) */ |
901 | |
902 | /* A ZWNJ disables HALF. */ |
903 | if (non_joiner) |
904 | info[j].mask &= ~indic_plan->mask_array[HALF]; |
905 | |
906 | } while (j > start && !is_consonant (info[j])); |
907 | } |
908 | } |
909 | |
910 | static void |
911 | initial_reordering_standalone_cluster (const hb_ot_shape_plan_t *plan, |
912 | hb_face_t *face, |
913 | hb_buffer_t *buffer, |
914 | unsigned int start, unsigned int end) |
915 | { |
916 | /* We treat placeholder/dotted-circle as if they are consonants, so we |
917 | * should just chain. Only if not in compatibility mode that is... */ |
918 | |
919 | if (hb_options ().uniscribe_bug_compatible) |
920 | { |
921 | /* For dotted-circle, this is what Uniscribe does: |
922 | * If dotted-circle is the last glyph, it just does nothing. |
923 | * Ie. It doesn't form Reph. */ |
924 | if (buffer->info[end - 1].indic_category() == OT_DOTTEDCIRCLE) |
925 | return; |
926 | } |
927 | |
928 | initial_reordering_consonant_syllable (plan, face, buffer, start, end); |
929 | } |
930 | |
931 | static void |
932 | initial_reordering_syllable (const hb_ot_shape_plan_t *plan, |
933 | hb_face_t *face, |
934 | hb_buffer_t *buffer, |
935 | unsigned int start, unsigned int end) |
936 | { |
937 | syllable_type_t syllable_type = (syllable_type_t) (buffer->info[start].syllable() & 0x0F); |
938 | switch (syllable_type) |
939 | { |
940 | case vowel_syllable: /* We made the vowels look like consonants. So let's call the consonant logic! */ |
941 | case consonant_syllable: |
942 | initial_reordering_consonant_syllable (plan, face, buffer, start, end); |
943 | break; |
944 | |
945 | case broken_cluster: /* We already inserted dotted-circles, so just call the standalone_cluster. */ |
946 | case standalone_cluster: |
947 | initial_reordering_standalone_cluster (plan, face, buffer, start, end); |
948 | break; |
949 | |
950 | case symbol_cluster: |
951 | case non_indic_cluster: |
952 | break; |
953 | } |
954 | } |
955 | |
956 | static inline void |
957 | insert_dotted_circles (const hb_ot_shape_plan_t *plan HB_UNUSED, |
958 | hb_font_t *font, |
959 | hb_buffer_t *buffer) |
960 | { |
961 | /* Note: This loop is extra overhead, but should not be measurable. */ |
962 | bool has_broken_syllables = false; |
963 | unsigned int count = buffer->len; |
964 | hb_glyph_info_t *info = buffer->info; |
965 | for (unsigned int i = 0; i < count; i++) |
966 | if ((info[i].syllable() & 0x0F) == broken_cluster) |
967 | { |
968 | has_broken_syllables = true; |
969 | break; |
970 | } |
971 | if (likely (!has_broken_syllables)) |
972 | return; |
973 | |
974 | |
975 | hb_codepoint_t dottedcircle_glyph; |
976 | if (!font->get_nominal_glyph (0x25CCu, &dottedcircle_glyph)) |
977 | return; |
978 | |
979 | hb_glyph_info_t dottedcircle = {0}; |
980 | dottedcircle.codepoint = 0x25CCu; |
981 | set_indic_properties (dottedcircle); |
982 | dottedcircle.codepoint = dottedcircle_glyph; |
983 | |
984 | buffer->clear_output (); |
985 | |
986 | buffer->idx = 0; |
987 | unsigned int last_syllable = 0; |
988 | while (buffer->idx < buffer->len && buffer->successful) |
989 | { |
990 | unsigned int syllable = buffer->cur().syllable(); |
991 | syllable_type_t syllable_type = (syllable_type_t) (syllable & 0x0F); |
992 | if (unlikely (last_syllable != syllable && syllable_type == broken_cluster)) |
993 | { |
994 | last_syllable = syllable; |
995 | |
996 | hb_glyph_info_t ginfo = dottedcircle; |
997 | ginfo.cluster = buffer->cur().cluster; |
998 | ginfo.mask = buffer->cur().mask; |
999 | ginfo.syllable() = buffer->cur().syllable(); |
1000 | /* TODO Set glyph_props? */ |
1001 | |
1002 | /* Insert dottedcircle after possible Repha. */ |
1003 | while (buffer->idx < buffer->len && buffer->successful && |
1004 | last_syllable == buffer->cur().syllable() && |
1005 | buffer->cur().indic_category() == OT_Repha) |
1006 | buffer->next_glyph (); |
1007 | |
1008 | buffer->output_info (ginfo); |
1009 | } |
1010 | else |
1011 | buffer->next_glyph (); |
1012 | } |
1013 | |
1014 | buffer->swap_buffers (); |
1015 | } |
1016 | |
1017 | static void |
1018 | initial_reordering (const hb_ot_shape_plan_t *plan, |
1019 | hb_font_t *font, |
1020 | hb_buffer_t *buffer) |
1021 | { |
1022 | update_consonant_positions (plan, font, buffer); |
1023 | insert_dotted_circles (plan, font, buffer); |
1024 | |
1025 | foreach_syllable (buffer, start, end) |
1026 | initial_reordering_syllable (plan, font->face, buffer, start, end); |
1027 | } |
1028 | |
1029 | static void |
1030 | final_reordering_syllable (const hb_ot_shape_plan_t *plan, |
1031 | hb_buffer_t *buffer, |
1032 | unsigned int start, unsigned int end) |
1033 | { |
1034 | const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; |
1035 | hb_glyph_info_t *info = buffer->info; |
1036 | |
1037 | |
1038 | /* This function relies heavily on halant glyphs. Lots of ligation |
1039 | * and possibly multiple substitutions happened prior to this |
1040 | * phase, and that might have messed up our properties. Recover |
1041 | * from a particular case of that where we're fairly sure that a |
1042 | * class of OT_H is desired but has been lost. */ |
1043 | /* We don't call load_virama_glyph(), since we know it's already |
1044 | * loaded. */ |
1045 | hb_codepoint_t virama_glyph = indic_plan->virama_glyph.get_relaxed (); |
1046 | if (virama_glyph) |
1047 | { |
1048 | for (unsigned int i = start; i < end; i++) |
1049 | if (info[i].codepoint == virama_glyph && |
1050 | _hb_glyph_info_ligated (&info[i]) && |
1051 | _hb_glyph_info_multiplied (&info[i])) |
1052 | { |
1053 | /* This will make sure that this glyph passes is_halant() test. */ |
1054 | info[i].indic_category() = OT_H; |
1055 | _hb_glyph_info_clear_ligated_and_multiplied (&info[i]); |
1056 | } |
1057 | } |
1058 | |
1059 | |
1060 | /* 4. Final reordering: |
1061 | * |
1062 | * After the localized forms and basic shaping forms GSUB features have been |
1063 | * applied (see below), the shaping engine performs some final glyph |
1064 | * reordering before applying all the remaining font features to the entire |
1065 | * syllable. |
1066 | */ |
1067 | |
1068 | bool try_pref = !!indic_plan->mask_array[PREF]; |
1069 | |
1070 | /* Find base again */ |
1071 | unsigned int base; |
1072 | for (base = start; base < end; base++) |
1073 | if (info[base].indic_position() >= POS_BASE_C) |
1074 | { |
1075 | if (try_pref && base + 1 < end) |
1076 | { |
1077 | for (unsigned int i = base + 1; i < end; i++) |
1078 | if ((info[i].mask & indic_plan->mask_array[PREF]) != 0) |
1079 | { |
1080 | if (!(_hb_glyph_info_substituted (&info[i]) && |
1081 | _hb_glyph_info_ligated_and_didnt_multiply (&info[i]))) |
1082 | { |
1083 | /* Ok, this was a 'pref' candidate but didn't form any. |
1084 | * Base is around here... */ |
1085 | base = i; |
1086 | while (base < end && is_halant (info[base])) |
1087 | base++; |
1088 | info[base].indic_position() = POS_BASE_C; |
1089 | |
1090 | try_pref = false; |
1091 | } |
1092 | break; |
1093 | } |
1094 | } |
1095 | /* For Malayalam, skip over unformed below- (but NOT post-) forms. */ |
1096 | if (buffer->props.script == HB_SCRIPT_MALAYALAM) |
1097 | { |
1098 | for (unsigned int i = base + 1; i < end; i++) |
1099 | { |
1100 | while (i < end && is_joiner (info[i])) |
1101 | i++; |
1102 | if (i == end || !is_halant (info[i])) |
1103 | break; |
1104 | i++; /* Skip halant. */ |
1105 | while (i < end && is_joiner (info[i])) |
1106 | i++; |
1107 | if (i < end && is_consonant (info[i]) && info[i].indic_position() == POS_BELOW_C) |
1108 | { |
1109 | base = i; |
1110 | info[base].indic_position() = POS_BASE_C; |
1111 | } |
1112 | } |
1113 | } |
1114 | |
1115 | if (start < base && info[base].indic_position() > POS_BASE_C) |
1116 | base--; |
1117 | break; |
1118 | } |
1119 | if (base == end && start < base && |
1120 | is_one_of (info[base - 1], FLAG (OT_ZWJ))) |
1121 | base--; |
1122 | if (base < end) |
1123 | while (start < base && |
1124 | is_one_of (info[base], (FLAG (OT_N) | FLAG (OT_H)))) |
1125 | base--; |
1126 | |
1127 | |
1128 | /* o Reorder matras: |
1129 | * |
1130 | * If a pre-base matra character had been reordered before applying basic |
1131 | * features, the glyph can be moved closer to the main consonant based on |
1132 | * whether half-forms had been formed. Actual position for the matra is |
1133 | * defined as “after last standalone halant glyph, after initial matra |
1134 | * position and before the main consonant”. If ZWJ or ZWNJ follow this |
1135 | * halant, position is moved after it. |
1136 | * |
1137 | * IMPLEMENTATION NOTES: |
1138 | * |
1139 | * It looks like the last sentence is wrong. Testing, with Windows 7 Uniscribe |
1140 | * and Devanagari shows that the behavior is best described as: |
1141 | * |
1142 | * "If ZWJ follows this halant, matra is NOT repositioned after this halant. |
1143 | * If ZWNJ follows this halant, position is moved after it." |
1144 | * |
1145 | * Test case, with Adobe Devanagari or Nirmala UI: |
1146 | * |
1147 | * U+091F,U+094D,U+200C,U+092F,U+093F |
1148 | * (Matra moves to the middle, after ZWNJ.) |
1149 | * |
1150 | * U+091F,U+094D,U+200D,U+092F,U+093F |
1151 | * (Matra does NOT move, stays to the left.) |
1152 | * |
1153 | * https://github.com/harfbuzz/harfbuzz/issues/1070 |
1154 | */ |
1155 | |
1156 | if (start + 1 < end && start < base) /* Otherwise there can't be any pre-base matra characters. */ |
1157 | { |
1158 | /* If we lost track of base, alas, position before last thingy. */ |
1159 | unsigned int new_pos = base == end ? base - 2 : base - 1; |
1160 | |
1161 | /* Malayalam / Tamil do not have "half" forms or explicit virama forms. |
1162 | * The glyphs formed by 'half' are Chillus or ligated explicit viramas. |
1163 | * We want to position matra after them. |
1164 | */ |
1165 | if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL) |
1166 | { |
1167 | search: |
1168 | while (new_pos > start && |
1169 | !(is_one_of (info[new_pos], (FLAG (OT_M) | FLAG (OT_H))))) |
1170 | new_pos--; |
1171 | |
1172 | /* If we found no Halant we are done. |
1173 | * Otherwise only proceed if the Halant does |
1174 | * not belong to the Matra itself! */ |
1175 | if (is_halant (info[new_pos]) && |
1176 | info[new_pos].indic_position() != POS_PRE_M) |
1177 | { |
1178 | #if 0 // See comment above |
1179 | /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */ |
1180 | if (new_pos + 1 < end && is_joiner (info[new_pos + 1])) |
1181 | new_pos++; |
1182 | #endif |
1183 | if (new_pos + 1 < end) |
1184 | { |
1185 | /* -> If ZWJ follows this halant, matra is NOT repositioned after this halant. */ |
1186 | if (info[new_pos + 1].indic_category() == OT_ZWJ) |
1187 | { |
1188 | /* Keep searching. */ |
1189 | if (new_pos > start) |
1190 | { |
1191 | new_pos--; |
1192 | goto search; |
1193 | } |
1194 | } |
1195 | /* -> If ZWNJ follows this halant, position is moved after it. */ |
1196 | if (info[new_pos + 1].indic_category() == OT_ZWNJ) |
1197 | new_pos++; |
1198 | } |
1199 | } |
1200 | else |
1201 | new_pos = start; /* No move. */ |
1202 | } |
1203 | |
1204 | if (start < new_pos && info[new_pos].indic_position () != POS_PRE_M) |
1205 | { |
1206 | /* Now go see if there's actually any matras... */ |
1207 | for (unsigned int i = new_pos; i > start; i--) |
1208 | if (info[i - 1].indic_position () == POS_PRE_M) |
1209 | { |
1210 | unsigned int old_pos = i - 1; |
1211 | if (old_pos < base && base <= new_pos) /* Shouldn't actually happen. */ |
1212 | base--; |
1213 | |
1214 | hb_glyph_info_t tmp = info[old_pos]; |
1215 | memmove (&info[old_pos], &info[old_pos + 1], (new_pos - old_pos) * sizeof (info[0])); |
1216 | info[new_pos] = tmp; |
1217 | |
1218 | /* Note: this merge_clusters() is intentionally *after* the reordering. |
1219 | * Indic matra reordering is special and tricky... */ |
1220 | buffer->merge_clusters (new_pos, MIN (end, base + 1)); |
1221 | |
1222 | new_pos--; |
1223 | } |
1224 | } else { |
1225 | for (unsigned int i = start; i < base; i++) |
1226 | if (info[i].indic_position () == POS_PRE_M) { |
1227 | buffer->merge_clusters (i, MIN (end, base + 1)); |
1228 | break; |
1229 | } |
1230 | } |
1231 | } |
1232 | |
1233 | |
1234 | /* o Reorder reph: |
1235 | * |
1236 | * Reph’s original position is always at the beginning of the syllable, |
1237 | * (i.e. it is not reordered at the character reordering stage). However, |
1238 | * it will be reordered according to the basic-forms shaping results. |
1239 | * Possible positions for reph, depending on the script, are; after main, |
1240 | * before post-base consonant forms, and after post-base consonant forms. |
1241 | */ |
1242 | |
1243 | /* Two cases: |
1244 | * |
1245 | * - If repha is encoded as a sequence of characters (Ra,H or Ra,H,ZWJ), then |
1246 | * we should only move it if the sequence ligated to the repha form. |
1247 | * |
1248 | * - If repha is encoded separately and in the logical position, we should only |
1249 | * move it if it did NOT ligate. If it ligated, it's probably the font trying |
1250 | * to make it work without the reordering. |
1251 | */ |
1252 | if (start + 1 < end && |
1253 | info[start].indic_position() == POS_RA_TO_BECOME_REPH && |
1254 | ((info[start].indic_category() == OT_Repha) ^ |
1255 | _hb_glyph_info_ligated_and_didnt_multiply (&info[start]))) |
1256 | { |
1257 | unsigned int new_reph_pos; |
1258 | reph_position_t reph_pos = indic_plan->config->reph_pos; |
1259 | |
1260 | /* 1. If reph should be positioned after post-base consonant forms, |
1261 | * proceed to step 5. |
1262 | */ |
1263 | if (reph_pos == REPH_POS_AFTER_POST) |
1264 | { |
1265 | goto reph_step_5; |
1266 | } |
1267 | |
1268 | /* 2. If the reph repositioning class is not after post-base: target |
1269 | * position is after the first explicit halant glyph between the |
1270 | * first post-reph consonant and last main consonant. If ZWJ or ZWNJ |
1271 | * are following this halant, position is moved after it. If such |
1272 | * position is found, this is the target position. Otherwise, |
1273 | * proceed to the next step. |
1274 | * |
1275 | * Note: in old-implementation fonts, where classifications were |
1276 | * fixed in shaping engine, there was no case where reph position |
1277 | * will be found on this step. |
1278 | */ |
1279 | { |
1280 | new_reph_pos = start + 1; |
1281 | while (new_reph_pos < base && !is_halant (info[new_reph_pos])) |
1282 | new_reph_pos++; |
1283 | |
1284 | if (new_reph_pos < base && is_halant (info[new_reph_pos])) |
1285 | { |
1286 | /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */ |
1287 | if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1])) |
1288 | new_reph_pos++; |
1289 | goto reph_move; |
1290 | } |
1291 | } |
1292 | |
1293 | /* 3. If reph should be repositioned after the main consonant: find the |
1294 | * first consonant not ligated with main, or find the first |
1295 | * consonant that is not a potential pre-base-reordering Ra. |
1296 | */ |
1297 | if (reph_pos == REPH_POS_AFTER_MAIN) |
1298 | { |
1299 | new_reph_pos = base; |
1300 | while (new_reph_pos + 1 < end && info[new_reph_pos + 1].indic_position() <= POS_AFTER_MAIN) |
1301 | new_reph_pos++; |
1302 | if (new_reph_pos < end) |
1303 | goto reph_move; |
1304 | } |
1305 | |
1306 | /* 4. If reph should be positioned before post-base consonant, find |
1307 | * first post-base classified consonant not ligated with main. If no |
1308 | * consonant is found, the target position should be before the |
1309 | * first matra, syllable modifier sign or vedic sign. |
1310 | */ |
1311 | /* This is our take on what step 4 is trying to say (and failing, BADLY). */ |
1312 | if (reph_pos == REPH_POS_AFTER_SUB) |
1313 | { |
1314 | new_reph_pos = base; |
1315 | while (new_reph_pos + 1 < end && |
1316 | !( FLAG_UNSAFE (info[new_reph_pos + 1].indic_position()) & (FLAG (POS_POST_C) | FLAG (POS_AFTER_POST) | FLAG (POS_SMVD)))) |
1317 | new_reph_pos++; |
1318 | if (new_reph_pos < end) |
1319 | goto reph_move; |
1320 | } |
1321 | |
1322 | /* 5. If no consonant is found in steps 3 or 4, move reph to a position |
1323 | * immediately before the first post-base matra, syllable modifier |
1324 | * sign or vedic sign that has a reordering class after the intended |
1325 | * reph position. For example, if the reordering position for reph |
1326 | * is post-main, it will skip above-base matras that also have a |
1327 | * post-main position. |
1328 | */ |
1329 | reph_step_5: |
1330 | { |
1331 | /* Copied from step 2. */ |
1332 | new_reph_pos = start + 1; |
1333 | while (new_reph_pos < base && !is_halant (info[new_reph_pos])) |
1334 | new_reph_pos++; |
1335 | |
1336 | if (new_reph_pos < base && is_halant (info[new_reph_pos])) |
1337 | { |
1338 | /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */ |
1339 | if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1])) |
1340 | new_reph_pos++; |
1341 | goto reph_move; |
1342 | } |
1343 | } |
1344 | |
1345 | /* 6. Otherwise, reorder reph to the end of the syllable. |
1346 | */ |
1347 | { |
1348 | new_reph_pos = end - 1; |
1349 | while (new_reph_pos > start && info[new_reph_pos].indic_position() == POS_SMVD) |
1350 | new_reph_pos--; |
1351 | |
1352 | /* |
1353 | * If the Reph is to be ending up after a Matra,Halant sequence, |
1354 | * position it before that Halant so it can interact with the Matra. |
1355 | * However, if it's a plain Consonant,Halant we shouldn't do that. |
1356 | * Uniscribe doesn't do this. |
1357 | * TEST: U+0930,U+094D,U+0915,U+094B,U+094D |
1358 | */ |
1359 | if (!hb_options ().uniscribe_bug_compatible && |
1360 | unlikely (is_halant (info[new_reph_pos]))) { |
1361 | for (unsigned int i = base + 1; i < new_reph_pos; i++) |
1362 | if (info[i].indic_category() == OT_M) { |
1363 | /* Ok, got it. */ |
1364 | new_reph_pos--; |
1365 | } |
1366 | } |
1367 | goto reph_move; |
1368 | } |
1369 | |
1370 | reph_move: |
1371 | { |
1372 | /* Move */ |
1373 | buffer->merge_clusters (start, new_reph_pos + 1); |
1374 | hb_glyph_info_t reph = info[start]; |
1375 | memmove (&info[start], &info[start + 1], (new_reph_pos - start) * sizeof (info[0])); |
1376 | info[new_reph_pos] = reph; |
1377 | |
1378 | if (start < base && base <= new_reph_pos) |
1379 | base--; |
1380 | } |
1381 | } |
1382 | |
1383 | |
1384 | /* o Reorder pre-base-reordering consonants: |
1385 | * |
1386 | * If a pre-base-reordering consonant is found, reorder it according to |
1387 | * the following rules: |
1388 | */ |
1389 | |
1390 | if (try_pref && base + 1 < end) /* Otherwise there can't be any pre-base-reordering Ra. */ |
1391 | { |
1392 | for (unsigned int i = base + 1; i < end; i++) |
1393 | if ((info[i].mask & indic_plan->mask_array[PREF]) != 0) |
1394 | { |
1395 | /* 1. Only reorder a glyph produced by substitution during application |
1396 | * of the <pref> feature. (Note that a font may shape a Ra consonant with |
1397 | * the feature generally but block it in certain contexts.) |
1398 | */ |
1399 | /* Note: We just check that something got substituted. We don't check that |
1400 | * the <pref> feature actually did it... |
1401 | * |
1402 | * Reorder pref only if it ligated. */ |
1403 | if (_hb_glyph_info_ligated_and_didnt_multiply (&info[i])) |
1404 | { |
1405 | /* |
1406 | * 2. Try to find a target position the same way as for pre-base matra. |
1407 | * If it is found, reorder pre-base consonant glyph. |
1408 | * |
1409 | * 3. If position is not found, reorder immediately before main |
1410 | * consonant. |
1411 | */ |
1412 | |
1413 | unsigned int new_pos = base; |
1414 | /* Malayalam / Tamil do not have "half" forms or explicit virama forms. |
1415 | * The glyphs formed by 'half' are Chillus or ligated explicit viramas. |
1416 | * We want to position matra after them. |
1417 | */ |
1418 | if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL) |
1419 | { |
1420 | while (new_pos > start && |
1421 | !(is_one_of (info[new_pos - 1], FLAG(OT_M) | FLAG (OT_H)))) |
1422 | new_pos--; |
1423 | } |
1424 | |
1425 | if (new_pos > start && is_halant (info[new_pos - 1])) |
1426 | { |
1427 | /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */ |
1428 | if (new_pos < end && is_joiner (info[new_pos])) |
1429 | new_pos++; |
1430 | } |
1431 | |
1432 | { |
1433 | unsigned int old_pos = i; |
1434 | |
1435 | buffer->merge_clusters (new_pos, old_pos + 1); |
1436 | hb_glyph_info_t tmp = info[old_pos]; |
1437 | memmove (&info[new_pos + 1], &info[new_pos], (old_pos - new_pos) * sizeof (info[0])); |
1438 | info[new_pos] = tmp; |
1439 | |
1440 | if (new_pos <= base && base < old_pos) |
1441 | base++; |
1442 | } |
1443 | } |
1444 | |
1445 | break; |
1446 | } |
1447 | } |
1448 | |
1449 | |
1450 | /* Apply 'init' to the Left Matra if it's a word start. */ |
1451 | if (info[start].indic_position () == POS_PRE_M) |
1452 | { |
1453 | if (!start || |
1454 | !(FLAG_UNSAFE (_hb_glyph_info_get_general_category (&info[start - 1])) & |
1455 | FLAG_RANGE (HB_UNICODE_GENERAL_CATEGORY_FORMAT, HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK))) |
1456 | info[start].mask |= indic_plan->mask_array[INIT]; |
1457 | else |
1458 | buffer->unsafe_to_break (start - 1, start + 1); |
1459 | } |
1460 | |
1461 | |
1462 | /* |
1463 | * Finish off the clusters and go home! |
1464 | */ |
1465 | if (hb_options ().uniscribe_bug_compatible) |
1466 | { |
1467 | switch ((hb_tag_t) plan->props.script) |
1468 | { |
1469 | case HB_SCRIPT_TAMIL: |
1470 | case HB_SCRIPT_SINHALA: |
1471 | break; |
1472 | |
1473 | default: |
1474 | /* Uniscribe merges the entire syllable into a single cluster... Except for Tamil & Sinhala. |
1475 | * This means, half forms are submerged into the main consonant's cluster. |
1476 | * This is unnecessary, and makes cursor positioning harder, but that's what |
1477 | * Uniscribe does. */ |
1478 | buffer->merge_clusters (start, end); |
1479 | break; |
1480 | } |
1481 | } |
1482 | } |
1483 | |
1484 | |
1485 | static void |
1486 | final_reordering (const hb_ot_shape_plan_t *plan, |
1487 | hb_font_t *font HB_UNUSED, |
1488 | hb_buffer_t *buffer) |
1489 | { |
1490 | unsigned int count = buffer->len; |
1491 | if (unlikely (!count)) return; |
1492 | |
1493 | foreach_syllable (buffer, start, end) |
1494 | final_reordering_syllable (plan, buffer, start, end); |
1495 | |
1496 | HB_BUFFER_DEALLOCATE_VAR (buffer, indic_category); |
1497 | HB_BUFFER_DEALLOCATE_VAR (buffer, indic_position); |
1498 | } |
1499 | |
1500 | |
1501 | static void |
1502 | clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED, |
1503 | hb_font_t *font HB_UNUSED, |
1504 | hb_buffer_t *buffer) |
1505 | { |
1506 | hb_glyph_info_t *info = buffer->info; |
1507 | unsigned int count = buffer->len; |
1508 | for (unsigned int i = 0; i < count; i++) |
1509 | info[i].syllable() = 0; |
1510 | } |
1511 | |
1512 | |
1513 | static bool |
1514 | decompose_indic (const hb_ot_shape_normalize_context_t *c, |
1515 | hb_codepoint_t ab, |
1516 | hb_codepoint_t *a, |
1517 | hb_codepoint_t *b) |
1518 | { |
1519 | switch (ab) |
1520 | { |
1521 | /* Don't decompose these. */ |
1522 | case 0x0931u : return false; /* DEVANAGARI LETTER RRA */ |
1523 | // https://github.com/harfbuzz/harfbuzz/issues/779 |
1524 | case 0x09DCu : return false; /* BENGALI LETTER RRA */ |
1525 | case 0x09DDu : return false; /* BENGALI LETTER RHA */ |
1526 | case 0x0B94u : return false; /* TAMIL LETTER AU */ |
1527 | |
1528 | |
1529 | /* |
1530 | * Decompose split matras that don't have Unicode decompositions. |
1531 | */ |
1532 | |
1533 | #if 0 |
1534 | /* Gujarati */ |
1535 | /* This one has no decomposition in Unicode, but needs no decomposition either. */ |
1536 | /* case 0x0AC9u : return false; */ |
1537 | |
1538 | /* Oriya */ |
1539 | case 0x0B57u : *a = no decomp, -> RIGHT; return true; |
1540 | #endif |
1541 | } |
1542 | |
1543 | if ((ab == 0x0DDAu || hb_in_range<hb_codepoint_t> (ab, 0x0DDCu, 0x0DDEu))) |
1544 | { |
1545 | /* |
1546 | * Sinhala split matras... Let the fun begin. |
1547 | * |
1548 | * These four characters have Unicode decompositions. However, Uniscribe |
1549 | * decomposes them "Khmer-style", that is, it uses the character itself to |
1550 | * get the second half. The first half of all four decompositions is always |
1551 | * U+0DD9. |
1552 | * |
1553 | * Now, there are buggy fonts, namely, the widely used lklug.ttf, that are |
1554 | * broken with Uniscribe. But we need to support them. As such, we only |
1555 | * do the Uniscribe-style decomposition if the character is transformed into |
1556 | * its "sec.half" form by the 'pstf' feature. Otherwise, we fall back to |
1557 | * Unicode decomposition. |
1558 | * |
1559 | * Note that we can't unconditionally use Unicode decomposition. That would |
1560 | * break some other fonts, that are designed to work with Uniscribe, and |
1561 | * don't have positioning features for the Unicode-style decomposition. |
1562 | * |
1563 | * Argh... |
1564 | * |
1565 | * The Uniscribe behavior is now documented in the newly published Sinhala |
1566 | * spec in 2012: |
1567 | * |
1568 | * https://docs.microsoft.com/en-us/typography/script-development/sinhala#shaping |
1569 | */ |
1570 | |
1571 | const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) c->plan->data; |
1572 | |
1573 | hb_codepoint_t glyph; |
1574 | |
1575 | if (hb_options ().uniscribe_bug_compatible || |
1576 | (c->font->get_nominal_glyph (ab, &glyph) && |
1577 | indic_plan->pstf.would_substitute (&glyph, 1, c->font->face))) |
1578 | { |
1579 | /* Ok, safe to use Uniscribe-style decomposition. */ |
1580 | *a = 0x0DD9u; |
1581 | *b = ab; |
1582 | return true; |
1583 | } |
1584 | } |
1585 | |
1586 | return (bool) c->unicode->decompose (ab, a, b); |
1587 | } |
1588 | |
1589 | static bool |
1590 | compose_indic (const hb_ot_shape_normalize_context_t *c, |
1591 | hb_codepoint_t a, |
1592 | hb_codepoint_t b, |
1593 | hb_codepoint_t *ab) |
1594 | { |
1595 | /* Avoid recomposing split matras. */ |
1596 | if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a))) |
1597 | return false; |
1598 | |
1599 | /* Composition-exclusion exceptions that we want to recompose. */ |
1600 | if (a == 0x09AFu && b == 0x09BCu) { *ab = 0x09DFu; return true; } |
1601 | |
1602 | return (bool) c->unicode->compose (a, b, ab); |
1603 | } |
1604 | |
1605 | |
1606 | const hb_ot_complex_shaper_t _hb_ot_complex_shaper_indic = |
1607 | { |
1608 | collect_features_indic, |
1609 | override_features_indic, |
1610 | data_create_indic, |
1611 | data_destroy_indic, |
1612 | nullptr, /* preprocess_text */ |
1613 | nullptr, /* postprocess_glyphs */ |
1614 | HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT, |
1615 | decompose_indic, |
1616 | compose_indic, |
1617 | setup_masks_indic, |
1618 | nullptr, /* disable_otl */ |
1619 | nullptr, /* reorder_marks */ |
1620 | HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE, |
1621 | false, /* fallback_position */ |
1622 | }; |
1623 | |