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.hh" |
28 | |
29 | #ifndef HB_NO_OT_SHAPE |
30 | |
31 | #include "hb-ot-shaper-indic.hh" |
32 | #include "hb-ot-shaper-indic-machine.hh" |
33 | #include "hb-ot-shaper-vowel-constraints.hh" |
34 | #include "hb-ot-layout.hh" |
35 | |
36 | |
37 | /* |
38 | * Indic shaper. |
39 | */ |
40 | |
41 | |
42 | static inline void |
43 | set_indic_properties (hb_glyph_info_t &info) |
44 | { |
45 | hb_codepoint_t u = info.codepoint; |
46 | unsigned int type = hb_indic_get_categories (u); |
47 | |
48 | info.indic_category() = (indic_category_t) (type & 0xFFu); |
49 | info.indic_position() = (indic_position_t) (type >> 8); |
50 | } |
51 | |
52 | |
53 | static inline bool |
54 | is_one_of (const hb_glyph_info_t &info, unsigned int flags) |
55 | { |
56 | /* If it ligated, all bets are off. */ |
57 | if (_hb_glyph_info_ligated (&info)) return false; |
58 | return !!(FLAG_UNSAFE (info.indic_category()) & flags); |
59 | } |
60 | |
61 | /* Note: |
62 | * |
63 | * We treat Vowels and placeholders as if they were consonants. This is safe because Vowels |
64 | * cannot happen in a consonant syllable. The plus side however is, we can call the |
65 | * consonant syllable logic from the vowel syllable function and get it all right! |
66 | * |
67 | * Keep in sync with consonant_categories in the generator. */ |
68 | #define CONSONANT_FLAGS_INDIC (FLAG (I_Cat(C)) | FLAG (I_Cat(CS)) | FLAG (I_Cat(Ra)) | FLAG (I_Cat(CM)) | FLAG (I_Cat(V)) | FLAG (I_Cat(PLACEHOLDER)) | FLAG (I_Cat(DOTTEDCIRCLE))) |
69 | |
70 | static inline bool |
71 | is_consonant (const hb_glyph_info_t &info) |
72 | { |
73 | return is_one_of (info, CONSONANT_FLAGS_INDIC); |
74 | } |
75 | |
76 | #define JOINER_FLAGS (FLAG (I_Cat(ZWJ)) | FLAG (I_Cat(ZWNJ))) |
77 | |
78 | static inline bool |
79 | is_joiner (const hb_glyph_info_t &info) |
80 | { |
81 | return is_one_of (info, JOINER_FLAGS); |
82 | } |
83 | |
84 | static inline bool |
85 | is_halant (const hb_glyph_info_t &info) |
86 | { |
87 | return is_one_of (info, FLAG (I_Cat(H))); |
88 | } |
89 | |
90 | struct hb_indic_would_substitute_feature_t |
91 | { |
92 | void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_) |
93 | { |
94 | zero_context = zero_context_; |
95 | lookups = map->get_stage_lookups (0/*GSUB*/, |
96 | map->get_feature_stage (0/*GSUB*/, feature_tag)); |
97 | } |
98 | |
99 | bool would_substitute (const hb_codepoint_t *glyphs, |
100 | unsigned int glyphs_count, |
101 | hb_face_t *face) const |
102 | { |
103 | for (const auto &lookup : lookups) |
104 | if (hb_ot_layout_lookup_would_substitute (face, lookup.index, glyphs, glyphs_count, zero_context)) |
105 | return true; |
106 | return false; |
107 | } |
108 | |
109 | private: |
110 | hb_array_t<const hb_ot_map_t::lookup_map_t> lookups; |
111 | bool zero_context; |
112 | }; |
113 | |
114 | |
115 | /* |
116 | * Indic configurations. Note that we do not want to keep every single script-specific |
117 | * behavior in these tables necessarily. This should mainly be used for per-script |
118 | * properties that are cheaper keeping here, than in the code. Ie. if, say, one and |
119 | * only one script has an exception, that one script can be if'ed directly in the code, |
120 | * instead of adding a new flag in these structs. |
121 | */ |
122 | |
123 | enum reph_position_t { |
124 | REPH_POS_AFTER_MAIN = POS_AFTER_MAIN, |
125 | REPH_POS_BEFORE_SUB = POS_BEFORE_SUB, |
126 | REPH_POS_AFTER_SUB = POS_AFTER_SUB, |
127 | REPH_POS_BEFORE_POST = POS_BEFORE_POST, |
128 | REPH_POS_AFTER_POST = POS_AFTER_POST |
129 | }; |
130 | enum reph_mode_t { |
131 | REPH_MODE_IMPLICIT, /* Reph formed out of initial Ra,H sequence. */ |
132 | REPH_MODE_EXPLICIT, /* Reph formed out of initial Ra,H,ZWJ sequence. */ |
133 | REPH_MODE_LOG_REPHA /* Encoded Repha character, needs reordering. */ |
134 | }; |
135 | enum blwf_mode_t { |
136 | BLWF_MODE_PRE_AND_POST, /* Below-forms feature applied to pre-base and post-base. */ |
137 | BLWF_MODE_POST_ONLY /* Below-forms feature applied to post-base only. */ |
138 | }; |
139 | struct indic_config_t |
140 | { |
141 | hb_script_t script; |
142 | bool has_old_spec; |
143 | hb_codepoint_t virama; |
144 | reph_position_t reph_pos; |
145 | reph_mode_t reph_mode; |
146 | blwf_mode_t blwf_mode; |
147 | }; |
148 | |
149 | static const indic_config_t indic_configs[] = |
150 | { |
151 | /* Default. Should be first. */ |
152 | {HB_SCRIPT_INVALID, false, 0,REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST}, |
153 | {HB_SCRIPT_DEVANAGARI,true, 0x094Du,REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST}, |
154 | {HB_SCRIPT_BENGALI, true, 0x09CDu,REPH_POS_AFTER_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST}, |
155 | {HB_SCRIPT_GURMUKHI, true, 0x0A4Du,REPH_POS_BEFORE_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST}, |
156 | {HB_SCRIPT_GUJARATI, true, 0x0ACDu,REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST}, |
157 | {HB_SCRIPT_ORIYA, true, 0x0B4Du,REPH_POS_AFTER_MAIN, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST}, |
158 | {HB_SCRIPT_TAMIL, true, 0x0BCDu,REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST}, |
159 | {HB_SCRIPT_TELUGU, true, 0x0C4Du,REPH_POS_AFTER_POST, REPH_MODE_EXPLICIT, BLWF_MODE_POST_ONLY}, |
160 | {HB_SCRIPT_KANNADA, true, 0x0CCDu,REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_POST_ONLY}, |
161 | {HB_SCRIPT_MALAYALAM, true, 0x0D4Du,REPH_POS_AFTER_MAIN, REPH_MODE_LOG_REPHA,BLWF_MODE_PRE_AND_POST}, |
162 | }; |
163 | |
164 | |
165 | static const hb_ot_map_feature_t |
166 | indic_features[] = |
167 | { |
168 | /* |
169 | * Basic features. |
170 | * These features are applied in order, one at a time, after initial_reordering, |
171 | * constrained to the syllable. |
172 | */ |
173 | {HB_TAG('n','u','k','t'), F_GLOBAL_MANUAL_JOINERS | F_PER_SYLLABLE}, |
174 | {HB_TAG('a','k','h','n'), F_GLOBAL_MANUAL_JOINERS | F_PER_SYLLABLE}, |
175 | {HB_TAG('r','p','h','f'), F_MANUAL_JOINERS | F_PER_SYLLABLE}, |
176 | {HB_TAG('r','k','r','f'), F_GLOBAL_MANUAL_JOINERS | F_PER_SYLLABLE}, |
177 | {HB_TAG('p','r','e','f'), F_MANUAL_JOINERS | F_PER_SYLLABLE}, |
178 | {HB_TAG('b','l','w','f'), F_MANUAL_JOINERS | F_PER_SYLLABLE}, |
179 | {HB_TAG('a','b','v','f'), F_MANUAL_JOINERS | F_PER_SYLLABLE}, |
180 | {HB_TAG('h','a','l','f'), F_MANUAL_JOINERS | F_PER_SYLLABLE}, |
181 | {HB_TAG('p','s','t','f'), F_MANUAL_JOINERS | F_PER_SYLLABLE}, |
182 | {HB_TAG('v','a','t','u'), F_GLOBAL_MANUAL_JOINERS | F_PER_SYLLABLE}, |
183 | {HB_TAG('c','j','c','t'), F_GLOBAL_MANUAL_JOINERS | F_PER_SYLLABLE}, |
184 | /* |
185 | * Other features. |
186 | * These features are applied all at once, after final_reordering, constrained |
187 | * to the syllable. |
188 | * Default Bengali font in Windows for example has intermixed |
189 | * lookups for init,pres,abvs,blws features. |
190 | */ |
191 | {HB_TAG('i','n','i','t'), F_MANUAL_JOINERS | F_PER_SYLLABLE}, |
192 | {HB_TAG('p','r','e','s'), F_GLOBAL_MANUAL_JOINERS | F_PER_SYLLABLE}, |
193 | {HB_TAG('a','b','v','s'), F_GLOBAL_MANUAL_JOINERS | F_PER_SYLLABLE}, |
194 | {HB_TAG('b','l','w','s'), F_GLOBAL_MANUAL_JOINERS | F_PER_SYLLABLE}, |
195 | {HB_TAG('p','s','t','s'), F_GLOBAL_MANUAL_JOINERS | F_PER_SYLLABLE}, |
196 | {HB_TAG('h','a','l','n'), F_GLOBAL_MANUAL_JOINERS | F_PER_SYLLABLE}, |
197 | }; |
198 | |
199 | /* |
200 | * Must be in the same order as the indic_features array. |
201 | */ |
202 | enum { |
203 | _INDIC_NUKT, |
204 | _INDIC_AKHN, |
205 | INDIC_RPHF, |
206 | _INDIC_RKRF, |
207 | INDIC_PREF, |
208 | INDIC_BLWF, |
209 | INDIC_ABVF, |
210 | INDIC_HALF, |
211 | INDIC_PSTF, |
212 | _INDIC_VATU, |
213 | _INDIC_CJCT, |
214 | |
215 | INDIC_INIT, |
216 | _INDIC_PRES, |
217 | _INDIC_ABVS, |
218 | _INDIC_BLWS, |
219 | _INDIC_PSTS, |
220 | _INDIC_HALN, |
221 | |
222 | INDIC_NUM_FEATURES, |
223 | INDIC_BASIC_FEATURES = INDIC_INIT, /* Don't forget to update this! */ |
224 | }; |
225 | |
226 | static bool |
227 | setup_syllables_indic (const hb_ot_shape_plan_t *plan, |
228 | hb_font_t *font, |
229 | hb_buffer_t *buffer); |
230 | static bool |
231 | initial_reordering_indic (const hb_ot_shape_plan_t *plan, |
232 | hb_font_t *font, |
233 | hb_buffer_t *buffer); |
234 | static bool |
235 | final_reordering_indic (const hb_ot_shape_plan_t *plan, |
236 | hb_font_t *font, |
237 | hb_buffer_t *buffer); |
238 | |
239 | static void |
240 | collect_features_indic (hb_ot_shape_planner_t *plan) |
241 | { |
242 | hb_ot_map_builder_t *map = &plan->map; |
243 | |
244 | /* Do this before any lookups have been applied. */ |
245 | map->add_gsub_pause (setup_syllables_indic); |
246 | |
247 | map->enable_feature (HB_TAG('l','o','c','l'), F_PER_SYLLABLE); |
248 | /* The Indic specs do not require ccmp, but we apply it here since if |
249 | * there is a use of it, it's typically at the beginning. */ |
250 | map->enable_feature (HB_TAG('c','c','m','p'), F_PER_SYLLABLE); |
251 | |
252 | |
253 | unsigned int i = 0; |
254 | map->add_gsub_pause (initial_reordering_indic); |
255 | |
256 | for (; i < INDIC_BASIC_FEATURES; i++) { |
257 | map->add_feature (indic_features[i]); |
258 | map->add_gsub_pause (nullptr); |
259 | } |
260 | |
261 | map->add_gsub_pause (final_reordering_indic); |
262 | |
263 | for (; i < INDIC_NUM_FEATURES; i++) |
264 | map->add_feature (indic_features[i]); |
265 | } |
266 | |
267 | static void |
268 | override_features_indic (hb_ot_shape_planner_t *plan) |
269 | { |
270 | plan->map.disable_feature (HB_TAG('l','i','g','a')); |
271 | plan->map.add_gsub_pause (hb_syllabic_clear_var); // Don't need syllables anymore, use stop to free buffer var |
272 | } |
273 | |
274 | |
275 | struct indic_shape_plan_t |
276 | { |
277 | bool load_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const |
278 | { |
279 | hb_codepoint_t glyph = virama_glyph; |
280 | if (unlikely (glyph == (hb_codepoint_t) -1)) |
281 | { |
282 | if (!config->virama || !font->get_nominal_glyph (config->virama, &glyph)) |
283 | glyph = 0; |
284 | /* Technically speaking, the spec says we should apply 'locl' to virama too. |
285 | * Maybe one day... */ |
286 | |
287 | /* Our get_nominal_glyph() function needs a font, so we can't get the virama glyph |
288 | * during shape planning... Instead, overwrite it here. */ |
289 | virama_glyph = (int) glyph; |
290 | } |
291 | |
292 | *pglyph = glyph; |
293 | return glyph != 0; |
294 | } |
295 | |
296 | const indic_config_t *config; |
297 | |
298 | bool is_old_spec; |
299 | #ifndef HB_NO_UNISCRIBE_BUG_COMPATIBLE |
300 | bool uniscribe_bug_compatible; |
301 | #else |
302 | static constexpr bool uniscribe_bug_compatible = false; |
303 | #endif |
304 | mutable hb_atomic_int_t virama_glyph; |
305 | |
306 | hb_indic_would_substitute_feature_t rphf; |
307 | hb_indic_would_substitute_feature_t pref; |
308 | hb_indic_would_substitute_feature_t blwf; |
309 | hb_indic_would_substitute_feature_t pstf; |
310 | hb_indic_would_substitute_feature_t vatu; |
311 | |
312 | hb_mask_t mask_array[INDIC_NUM_FEATURES]; |
313 | }; |
314 | |
315 | static void * |
316 | data_create_indic (const hb_ot_shape_plan_t *plan) |
317 | { |
318 | indic_shape_plan_t *indic_plan = (indic_shape_plan_t *) hb_calloc (1, sizeof (indic_shape_plan_t)); |
319 | if (unlikely (!indic_plan)) |
320 | return nullptr; |
321 | |
322 | indic_plan->config = &indic_configs[0]; |
323 | for (unsigned int i = 1; i < ARRAY_LENGTH (indic_configs); i++) |
324 | if (plan->props.script == indic_configs[i].script) { |
325 | indic_plan->config = &indic_configs[i]; |
326 | break; |
327 | } |
328 | |
329 | indic_plan->is_old_spec = indic_plan->config->has_old_spec && ((plan->map.chosen_script[0] & 0x000000FFu) != '2'); |
330 | #ifndef HB_NO_UNISCRIBE_BUG_COMPATIBLE |
331 | indic_plan->uniscribe_bug_compatible = hb_options ().uniscribe_bug_compatible; |
332 | #endif |
333 | indic_plan->virama_glyph = -1; |
334 | |
335 | /* Use zero-context would_substitute() matching for new-spec of the main |
336 | * Indic scripts, and scripts with one spec only, but not for old-specs. |
337 | * The new-spec for all dual-spec scripts says zero-context matching happens. |
338 | * |
339 | * However, testing with Malayalam shows that old and new spec both allow |
340 | * context. Testing with Bengali new-spec however shows that it doesn't. |
341 | * So, the heuristic here is the way it is. It should *only* be changed, |
342 | * as we discover more cases of what Windows does. DON'T TOUCH OTHERWISE. |
343 | */ |
344 | bool zero_context = !indic_plan->is_old_spec && plan->props.script != HB_SCRIPT_MALAYALAM; |
345 | indic_plan->rphf.init (&plan->map, HB_TAG('r','p','h','f'), zero_context); |
346 | indic_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), zero_context); |
347 | indic_plan->blwf.init (&plan->map, HB_TAG('b','l','w','f'), zero_context); |
348 | indic_plan->pstf.init (&plan->map, HB_TAG('p','s','t','f'), zero_context); |
349 | indic_plan->vatu.init (&plan->map, HB_TAG('v','a','t','u'), zero_context); |
350 | |
351 | for (unsigned int i = 0; i < ARRAY_LENGTH (indic_plan->mask_array); i++) |
352 | indic_plan->mask_array[i] = (indic_features[i].flags & F_GLOBAL) ? |
353 | 0 : plan->map.get_1_mask (indic_features[i].tag); |
354 | |
355 | return indic_plan; |
356 | } |
357 | |
358 | static void |
359 | data_destroy_indic (void *data) |
360 | { |
361 | hb_free (data); |
362 | } |
363 | |
364 | static indic_position_t |
365 | consonant_position_from_face (const indic_shape_plan_t *indic_plan, |
366 | const hb_codepoint_t consonant, |
367 | const hb_codepoint_t virama, |
368 | hb_face_t *face) |
369 | { |
370 | /* For old-spec, the order of glyphs is Consonant,Virama, |
371 | * whereas for new-spec, it's Virama,Consonant. However, |
372 | * some broken fonts (like Free Sans) simply copied lookups |
373 | * from old-spec to new-spec without modification. |
374 | * And oddly enough, Uniscribe seems to respect those lookups. |
375 | * Eg. in the sequence U+0924,U+094D,U+0930, Uniscribe finds |
376 | * base at 0. The font however, only has lookups matching |
377 | * 930,94D in 'blwf', not the expected 94D,930 (with new-spec |
378 | * table). As such, we simply match both sequences. Seems |
379 | * to work. |
380 | * |
381 | * Vatu is done as well, for: |
382 | * https://github.com/harfbuzz/harfbuzz/issues/1587 |
383 | */ |
384 | hb_codepoint_t glyphs[3] = {virama, consonant, virama}; |
385 | if (indic_plan->blwf.would_substitute (glyphs , 2, face) || |
386 | indic_plan->blwf.would_substitute (glyphs+1, 2, face) || |
387 | indic_plan->vatu.would_substitute (glyphs , 2, face) || |
388 | indic_plan->vatu.would_substitute (glyphs+1, 2, face)) |
389 | return POS_BELOW_C; |
390 | if (indic_plan->pstf.would_substitute (glyphs , 2, face) || |
391 | indic_plan->pstf.would_substitute (glyphs+1, 2, face)) |
392 | return POS_POST_C; |
393 | if (indic_plan->pref.would_substitute (glyphs , 2, face) || |
394 | indic_plan->pref.would_substitute (glyphs+1, 2, face)) |
395 | return POS_POST_C; |
396 | return POS_BASE_C; |
397 | } |
398 | |
399 | static void |
400 | setup_masks_indic (const hb_ot_shape_plan_t *plan HB_UNUSED, |
401 | hb_buffer_t *buffer, |
402 | hb_font_t *font HB_UNUSED) |
403 | { |
404 | HB_BUFFER_ALLOCATE_VAR (buffer, indic_category); |
405 | HB_BUFFER_ALLOCATE_VAR (buffer, indic_position); |
406 | |
407 | /* We cannot setup masks here. We save information about characters |
408 | * and setup masks later on in a pause-callback. */ |
409 | |
410 | unsigned int count = buffer->len; |
411 | hb_glyph_info_t *info = buffer->info; |
412 | for (unsigned int i = 0; i < count; i++) |
413 | set_indic_properties (info[i]); |
414 | } |
415 | |
416 | static bool |
417 | setup_syllables_indic (const hb_ot_shape_plan_t *plan HB_UNUSED, |
418 | hb_font_t *font HB_UNUSED, |
419 | hb_buffer_t *buffer) |
420 | { |
421 | HB_BUFFER_ALLOCATE_VAR (buffer, syllable); |
422 | find_syllables_indic (buffer); |
423 | foreach_syllable (buffer, start, end) |
424 | buffer->unsafe_to_break (start, end); |
425 | return false; |
426 | } |
427 | |
428 | static int |
429 | compare_indic_order (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb) |
430 | { |
431 | int a = pa->indic_position(); |
432 | int b = pb->indic_position(); |
433 | |
434 | return (int) a - (int) b; |
435 | } |
436 | |
437 | |
438 | |
439 | static void |
440 | update_consonant_positions_indic (const hb_ot_shape_plan_t *plan, |
441 | hb_font_t *font, |
442 | hb_buffer_t *buffer) |
443 | { |
444 | const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; |
445 | |
446 | hb_codepoint_t virama; |
447 | if (indic_plan->load_virama_glyph (font, &virama)) |
448 | { |
449 | hb_face_t *face = font->face; |
450 | unsigned int count = buffer->len; |
451 | hb_glyph_info_t *info = buffer->info; |
452 | for (unsigned int i = 0; i < count; i++) |
453 | if (info[i].indic_position() == POS_BASE_C) |
454 | { |
455 | hb_codepoint_t consonant = info[i].codepoint; |
456 | info[i].indic_position() = consonant_position_from_face (indic_plan, consonant, virama, face); |
457 | } |
458 | } |
459 | } |
460 | |
461 | |
462 | /* Rules from: |
463 | * https://docs.microsqoft.com/en-us/typography/script-development/devanagari */ |
464 | |
465 | static void |
466 | initial_reordering_consonant_syllable (const hb_ot_shape_plan_t *plan, |
467 | hb_face_t *face, |
468 | hb_buffer_t *buffer, |
469 | unsigned int start, unsigned int end) |
470 | { |
471 | const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; |
472 | hb_glyph_info_t *info = buffer->info; |
473 | |
474 | /* https://github.com/harfbuzz/harfbuzz/issues/435#issuecomment-335560167 |
475 | * // For compatibility with legacy usage in Kannada, |
476 | * // Ra+h+ZWJ must behave like Ra+ZWJ+h... |
477 | */ |
478 | if (buffer->props.script == HB_SCRIPT_KANNADA && |
479 | start + 3 <= end && |
480 | is_one_of (info[start ], FLAG (I_Cat(Ra))) && |
481 | is_one_of (info[start+1], FLAG (I_Cat(H))) && |
482 | is_one_of (info[start+2], FLAG (I_Cat(ZWJ)))) |
483 | { |
484 | buffer->merge_clusters (start+1, start+3); |
485 | hb_swap (info[start+1], info[start+2]); |
486 | } |
487 | |
488 | /* 1. Find base consonant: |
489 | * |
490 | * The shaping engine finds the base consonant of the syllable, using the |
491 | * following algorithm: starting from the end of the syllable, move backwards |
492 | * until a consonant is found that does not have a below-base or post-base |
493 | * form (post-base forms have to follow below-base forms), or that is not a |
494 | * pre-base-reordering Ra, or arrive at the first consonant. The consonant |
495 | * stopped at will be the base. |
496 | * |
497 | * o If the syllable starts with Ra + Halant (in a script that has Reph) |
498 | * and has more than one consonant, Ra is excluded from candidates for |
499 | * base consonants. |
500 | */ |
501 | |
502 | unsigned int base = end; |
503 | bool has_reph = false; |
504 | |
505 | { |
506 | /* -> If the syllable starts with Ra + Halant (in a script that has Reph) |
507 | * and has more than one consonant, Ra is excluded from candidates for |
508 | * base consonants. */ |
509 | unsigned int limit = start; |
510 | if (indic_plan->mask_array[INDIC_RPHF] && |
511 | start + 3 <= end && |
512 | ( |
513 | (indic_plan->config->reph_mode == REPH_MODE_IMPLICIT && !is_joiner (info[start + 2])) || |
514 | (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT && info[start + 2].indic_category() == I_Cat(ZWJ)) |
515 | )) |
516 | { |
517 | /* See if it matches the 'rphf' feature. */ |
518 | hb_codepoint_t glyphs[3] = {info[start].codepoint, |
519 | info[start + 1].codepoint, |
520 | indic_plan->config->reph_mode == REPH_MODE_EXPLICIT ? |
521 | info[start + 2].codepoint : 0}; |
522 | if (indic_plan->rphf.would_substitute (glyphs, 2, face) || |
523 | (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT && |
524 | indic_plan->rphf.would_substitute (glyphs, 3, face))) |
525 | { |
526 | limit += 2; |
527 | while (limit < end && is_joiner (info[limit])) |
528 | limit++; |
529 | base = start; |
530 | has_reph = true; |
531 | } |
532 | } else if (indic_plan->config->reph_mode == REPH_MODE_LOG_REPHA && info[start].indic_category() == I_Cat(Repha)) |
533 | { |
534 | limit += 1; |
535 | while (limit < end && is_joiner (info[limit])) |
536 | limit++; |
537 | base = start; |
538 | has_reph = true; |
539 | } |
540 | |
541 | { |
542 | /* -> starting from the end of the syllable, move backwards */ |
543 | unsigned int i = end; |
544 | bool seen_below = false; |
545 | do { |
546 | i--; |
547 | /* -> until a consonant is found */ |
548 | if (is_consonant (info[i])) |
549 | { |
550 | /* -> that does not have a below-base or post-base form |
551 | * (post-base forms have to follow below-base forms), */ |
552 | if (info[i].indic_position() != POS_BELOW_C && |
553 | (info[i].indic_position() != POS_POST_C || seen_below)) |
554 | { |
555 | base = i; |
556 | break; |
557 | } |
558 | if (info[i].indic_position() == POS_BELOW_C) |
559 | seen_below = true; |
560 | |
561 | /* -> or that is not a pre-base-reordering Ra, |
562 | * |
563 | * IMPLEMENTATION NOTES: |
564 | * |
565 | * Our pre-base-reordering Ra's are marked POS_POST_C, so will be skipped |
566 | * by the logic above already. |
567 | */ |
568 | |
569 | /* -> or arrive at the first consonant. The consonant stopped at will |
570 | * be the base. */ |
571 | base = i; |
572 | } |
573 | else |
574 | { |
575 | /* A ZWJ after a Halant stops the base search, and requests an explicit |
576 | * half form. |
577 | * A ZWJ before a Halant, requests a subjoined form instead, and hence |
578 | * search continues. This is particularly important for Bengali |
579 | * sequence Ra,H,Ya that should form Ya-Phalaa by subjoining Ya. */ |
580 | if (start < i && |
581 | info[i].indic_category() == I_Cat(ZWJ) && |
582 | info[i - 1].indic_category() == I_Cat(H)) |
583 | break; |
584 | } |
585 | } while (i > limit); |
586 | } |
587 | |
588 | /* -> If the syllable starts with Ra + Halant (in a script that has Reph) |
589 | * and has more than one consonant, Ra is excluded from candidates for |
590 | * base consonants. |
591 | * |
592 | * Only do this for unforced Reph. (ie. not for Ra,H,ZWJ. */ |
593 | if (has_reph && base == start && limit - base <= 2) { |
594 | /* Have no other consonant, so Reph is not formed and Ra becomes base. */ |
595 | has_reph = false; |
596 | } |
597 | } |
598 | |
599 | |
600 | /* 2. Decompose and reorder Matras: |
601 | * |
602 | * Each matra and any syllable modifier sign in the syllable are moved to the |
603 | * appropriate position relative to the consonant(s) in the syllable. The |
604 | * shaping engine decomposes two- or three-part matras into their constituent |
605 | * parts before any repositioning. Matra characters are classified by which |
606 | * consonant in a conjunct they have affinity for and are reordered to the |
607 | * following positions: |
608 | * |
609 | * o Before first half form in the syllable |
610 | * o After subjoined consonants |
611 | * o After post-form consonant |
612 | * o After main consonant (for above marks) |
613 | * |
614 | * IMPLEMENTATION NOTES: |
615 | * |
616 | * The normalize() routine has already decomposed matras for us, so we don't |
617 | * need to worry about that. |
618 | */ |
619 | |
620 | |
621 | /* 3. Reorder marks to canonical order: |
622 | * |
623 | * Adjacent nukta and halant or nukta and vedic sign are always repositioned |
624 | * if necessary, so that the nukta is first. |
625 | * |
626 | * IMPLEMENTATION NOTES: |
627 | * |
628 | * We don't need to do this: the normalize() routine already did this for us. |
629 | */ |
630 | |
631 | |
632 | /* Reorder characters */ |
633 | |
634 | for (unsigned int i = start; i < base; i++) |
635 | info[i].indic_position() = hb_min (POS_PRE_C, (indic_position_t) info[i].indic_position()); |
636 | |
637 | if (base < end) |
638 | info[base].indic_position() = POS_BASE_C; |
639 | |
640 | /* Handle beginning Ra */ |
641 | if (has_reph) |
642 | info[start].indic_position() = POS_RA_TO_BECOME_REPH; |
643 | |
644 | /* For old-style Indic script tags, move the first post-base Halant after |
645 | * last consonant. |
646 | * |
647 | * Reports suggest that in some scripts Uniscribe does this only if there |
648 | * is *not* a Halant after last consonant already. We know that is the |
649 | * case for Kannada, while it reorders unconditionally in other scripts, |
650 | * eg. Malayalam, Bengali, and Devanagari. We don't currently know about |
651 | * other scripts, so we block Kannada. |
652 | * |
653 | * Kannada test case: |
654 | * U+0C9A,U+0CCD,U+0C9A,U+0CCD |
655 | * With some versions of Lohit Kannada. |
656 | * https://bugs.freedesktop.org/show_bug.cgi?id=59118 |
657 | * |
658 | * Malayalam test case: |
659 | * U+0D38,U+0D4D,U+0D31,U+0D4D,U+0D31,U+0D4D |
660 | * With lohit-ttf-20121122/Lohit-Malayalam.ttf |
661 | * |
662 | * Bengali test case: |
663 | * U+0998,U+09CD,U+09AF,U+09CD |
664 | * With Windows XP vrinda.ttf |
665 | * https://github.com/harfbuzz/harfbuzz/issues/1073 |
666 | * |
667 | * Devanagari test case: |
668 | * U+091F,U+094D,U+0930,U+094D |
669 | * With chandas.ttf |
670 | * https://github.com/harfbuzz/harfbuzz/issues/1071 |
671 | */ |
672 | if (indic_plan->is_old_spec) |
673 | { |
674 | bool disallow_double_halants = buffer->props.script == HB_SCRIPT_KANNADA; |
675 | for (unsigned int i = base + 1; i < end; i++) |
676 | if (info[i].indic_category() == I_Cat(H)) |
677 | { |
678 | unsigned int j; |
679 | for (j = end - 1; j > i; j--) |
680 | if (is_consonant (info[j]) || |
681 | (disallow_double_halants && info[j].indic_category() == I_Cat(H))) |
682 | break; |
683 | if (info[j].indic_category() != I_Cat(H) && j > i) { |
684 | /* Move Halant to after last consonant. */ |
685 | hb_glyph_info_t t = info[i]; |
686 | memmove (&info[i], &info[i + 1], (j - i) * sizeof (info[0])); |
687 | info[j] = t; |
688 | } |
689 | break; |
690 | } |
691 | } |
692 | |
693 | /* Attach misc marks to previous char to move with them. */ |
694 | { |
695 | indic_position_t last_pos = POS_START; |
696 | for (unsigned int i = start; i < end; i++) |
697 | { |
698 | if ((FLAG_UNSAFE (info[i].indic_category()) & (JOINER_FLAGS | FLAG (I_Cat(N)) | FLAG (I_Cat(RS)) | FLAG (I_Cat(CM)) | FLAG (I_Cat(H))))) |
699 | { |
700 | info[i].indic_position() = last_pos; |
701 | if (unlikely (info[i].indic_category() == I_Cat(H) && |
702 | info[i].indic_position() == POS_PRE_M)) |
703 | { |
704 | /* |
705 | * Uniscribe doesn't move the Halant with Left Matra. |
706 | * TEST: U+092B,U+093F,U+094D |
707 | * We follow. |
708 | */ |
709 | for (unsigned int j = i; j > start; j--) |
710 | if (info[j - 1].indic_position() != POS_PRE_M) { |
711 | info[i].indic_position() = info[j - 1].indic_position(); |
712 | break; |
713 | } |
714 | } |
715 | } else if (info[i].indic_position() != POS_SMVD) { |
716 | if (info[i].indic_category() == I_Cat(MPst) && |
717 | i > start && info[i - 1].indic_category() == I_Cat(SM)) |
718 | info[i - 1].indic_position() = info[i].indic_position(); |
719 | last_pos = (indic_position_t) info[i].indic_position(); |
720 | } |
721 | } |
722 | } |
723 | /* For post-base consonants let them own anything before them |
724 | * since the last consonant or matra. */ |
725 | { |
726 | unsigned int last = base; |
727 | for (unsigned int i = base + 1; i < end; i++) |
728 | if (is_consonant (info[i])) |
729 | { |
730 | for (unsigned int j = last + 1; j < i; j++) |
731 | if (info[j].indic_position() < POS_SMVD) |
732 | info[j].indic_position() = info[i].indic_position(); |
733 | last = i; |
734 | } else if (FLAG_UNSAFE (info[i].indic_category()) & (FLAG (I_Cat(M)) | FLAG (I_Cat(MPst)))) |
735 | last = i; |
736 | } |
737 | |
738 | |
739 | { |
740 | /* Use syllable() for sort accounting temporarily. */ |
741 | unsigned int syllable = info[start].syllable(); |
742 | for (unsigned int i = start; i < end; i++) |
743 | info[i].syllable() = i - start; |
744 | |
745 | /* Sit tight, rock 'n roll! */ |
746 | hb_stable_sort (info + start, end - start, compare_indic_order); |
747 | |
748 | /* Find base again; also flip left-matra sequence. */ |
749 | unsigned first_left_matra = end; |
750 | unsigned last_left_matra = end; |
751 | base = end; |
752 | for (unsigned int i = start; i < end; i++) |
753 | { |
754 | if (info[i].indic_position() == POS_BASE_C) |
755 | { |
756 | base = i; |
757 | break; |
758 | } |
759 | else if (info[i].indic_position() == POS_PRE_M) |
760 | { |
761 | if (first_left_matra == end) |
762 | first_left_matra = i; |
763 | last_left_matra = i; |
764 | } |
765 | } |
766 | /* https://github.com/harfbuzz/harfbuzz/issues/3863 */ |
767 | if (first_left_matra < last_left_matra) |
768 | { |
769 | /* No need to merge clusters, handled later. */ |
770 | buffer->reverse_range (first_left_matra, last_left_matra + 1); |
771 | /* Reverse back nuktas, etc. */ |
772 | unsigned i = first_left_matra; |
773 | for (unsigned j = i; j <= last_left_matra; j++) |
774 | if (FLAG_UNSAFE (info[j].indic_category()) & (FLAG (I_Cat(M)) | FLAG (I_Cat(MPst)))) |
775 | { |
776 | buffer->reverse_range (i, j + 1); |
777 | i = j + 1; |
778 | } |
779 | } |
780 | |
781 | /* Things are out-of-control for post base positions, they may shuffle |
782 | * around like crazy. In old-spec mode, we move halants around, so in |
783 | * that case merge all clusters after base. Otherwise, check the sort |
784 | * order and merge as needed. |
785 | * For pre-base stuff, we handle cluster issues in final reordering. |
786 | * |
787 | * We could use buffer->sort() for this, if there was no special |
788 | * reordering of pre-base stuff happening later... |
789 | * We don't want to merge_clusters all of that, which buffer->sort() |
790 | * would. Here's a concrete example: |
791 | * |
792 | * Assume there's a pre-base consonant and explicit Halant before base, |
793 | * followed by a prebase-reordering (left) Matra: |
794 | * |
795 | * C,H,ZWNJ,B,M |
796 | * |
797 | * At this point in reordering we would have: |
798 | * |
799 | * M,C,H,ZWNJ,B |
800 | * |
801 | * whereas in final reordering we will bring the Matra closer to Base: |
802 | * |
803 | * C,H,ZWNJ,M,B |
804 | * |
805 | * That's why we don't want to merge-clusters anything before the Base |
806 | * at this point. But if something moved from after Base to before it, |
807 | * we should merge clusters from base to them. In final-reordering, we |
808 | * only move things around before base, and merge-clusters up to base. |
809 | * These two merge-clusters from the two sides of base will interlock |
810 | * to merge things correctly. See: |
811 | * https://github.com/harfbuzz/harfbuzz/issues/2272 |
812 | */ |
813 | if (indic_plan->is_old_spec || end - start > 127) |
814 | buffer->merge_clusters (base, end); |
815 | else |
816 | { |
817 | /* Note! syllable() is a one-byte field. */ |
818 | for (unsigned int i = base; i < end; i++) |
819 | if (info[i].syllable() != 255) |
820 | { |
821 | unsigned int min = i; |
822 | unsigned int max = i; |
823 | unsigned int j = start + info[i].syllable(); |
824 | while (j != i) |
825 | { |
826 | min = hb_min (min, j); |
827 | max = hb_max (max, j); |
828 | unsigned int next = start + info[j].syllable(); |
829 | info[j].syllable() = 255; /* So we don't process j later again. */ |
830 | j = next; |
831 | } |
832 | buffer->merge_clusters (hb_max (base, min), max + 1); |
833 | } |
834 | } |
835 | |
836 | /* Put syllable back in. */ |
837 | for (unsigned int i = start; i < end; i++) |
838 | info[i].syllable() = syllable; |
839 | } |
840 | |
841 | /* Setup masks now */ |
842 | |
843 | { |
844 | hb_mask_t mask; |
845 | |
846 | /* Reph */ |
847 | for (unsigned int i = start; i < end && info[i].indic_position() == POS_RA_TO_BECOME_REPH; i++) |
848 | info[i].mask |= indic_plan->mask_array[INDIC_RPHF]; |
849 | |
850 | /* Pre-base */ |
851 | mask = indic_plan->mask_array[INDIC_HALF]; |
852 | if (!indic_plan->is_old_spec && |
853 | indic_plan->config->blwf_mode == BLWF_MODE_PRE_AND_POST) |
854 | mask |= indic_plan->mask_array[INDIC_BLWF]; |
855 | for (unsigned int i = start; i < base; i++) |
856 | info[i].mask |= mask; |
857 | /* Base */ |
858 | mask = 0; |
859 | if (base < end) |
860 | info[base].mask |= mask; |
861 | /* Post-base */ |
862 | mask = indic_plan->mask_array[INDIC_BLWF] | |
863 | indic_plan->mask_array[INDIC_ABVF] | |
864 | indic_plan->mask_array[INDIC_PSTF]; |
865 | for (unsigned int i = base + 1; i < end; i++) |
866 | info[i].mask |= mask; |
867 | } |
868 | |
869 | if (indic_plan->is_old_spec && |
870 | buffer->props.script == HB_SCRIPT_DEVANAGARI) |
871 | { |
872 | /* Old-spec eye-lash Ra needs special handling. From the |
873 | * spec: |
874 | * |
875 | * "The feature 'below-base form' is applied to consonants |
876 | * having below-base forms and following the base consonant. |
877 | * The exception is vattu, which may appear below half forms |
878 | * as well as below the base glyph. The feature 'below-base |
879 | * form' will be applied to all such occurrences of Ra as well." |
880 | * |
881 | * Test case: U+0924,U+094D,U+0930,U+094d,U+0915 |
882 | * with Sanskrit 2003 font. |
883 | * |
884 | * However, note that Ra,Halant,ZWJ is the correct way to |
885 | * request eyelash form of Ra, so we wouldbn't inhibit it |
886 | * in that sequence. |
887 | * |
888 | * Test case: U+0924,U+094D,U+0930,U+094d,U+200D,U+0915 |
889 | */ |
890 | for (unsigned int i = start; i + 1 < base; i++) |
891 | if (info[i ].indic_category() == I_Cat(Ra) && |
892 | info[i+1].indic_category() == I_Cat(H) && |
893 | (i + 2 == base || |
894 | info[i+2].indic_category() != I_Cat(ZWJ))) |
895 | { |
896 | info[i ].mask |= indic_plan->mask_array[INDIC_BLWF]; |
897 | info[i+1].mask |= indic_plan->mask_array[INDIC_BLWF]; |
898 | } |
899 | } |
900 | |
901 | unsigned int pref_len = 2; |
902 | if (indic_plan->mask_array[INDIC_PREF] && base + pref_len < end) |
903 | { |
904 | /* Find a Halant,Ra sequence and mark it for pre-base-reordering processing. */ |
905 | for (unsigned int i = base + 1; i + pref_len - 1 < end; i++) { |
906 | hb_codepoint_t glyphs[2]; |
907 | for (unsigned int j = 0; j < pref_len; j++) |
908 | glyphs[j] = info[i + j].codepoint; |
909 | if (indic_plan->pref.would_substitute (glyphs, pref_len, face)) |
910 | { |
911 | for (unsigned int j = 0; j < pref_len; j++) |
912 | info[i++].mask |= indic_plan->mask_array[INDIC_PREF]; |
913 | break; |
914 | } |
915 | } |
916 | } |
917 | |
918 | /* Apply ZWJ/ZWNJ effects */ |
919 | for (unsigned int i = start + 1; i < end; i++) |
920 | if (is_joiner (info[i])) { |
921 | bool non_joiner = info[i].indic_category() == I_Cat(ZWNJ); |
922 | unsigned int j = i; |
923 | |
924 | do { |
925 | j--; |
926 | |
927 | /* ZWJ/ZWNJ should disable CJCT. They do that by simply |
928 | * being there, since we don't skip them for the CJCT |
929 | * feature (ie. F_MANUAL_ZWJ) */ |
930 | |
931 | /* A ZWNJ disables HALF. */ |
932 | if (non_joiner) |
933 | info[j].mask &= ~indic_plan->mask_array[INDIC_HALF]; |
934 | |
935 | } while (j > start && !is_consonant (info[j])); |
936 | } |
937 | } |
938 | |
939 | static void |
940 | initial_reordering_standalone_cluster (const hb_ot_shape_plan_t *plan, |
941 | hb_face_t *face, |
942 | hb_buffer_t *buffer, |
943 | unsigned int start, unsigned int end) |
944 | { |
945 | /* We treat placeholder/dotted-circle as if they are consonants, so we |
946 | * should just chain. Only if not in compatibility mode that is... */ |
947 | |
948 | const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; |
949 | if (indic_plan->uniscribe_bug_compatible) |
950 | { |
951 | /* For dotted-circle, this is what Uniscribe does: |
952 | * If dotted-circle is the last glyph, it just does nothing. |
953 | * Ie. It doesn't form Reph. */ |
954 | if (buffer->info[end - 1].indic_category() == I_Cat(DOTTEDCIRCLE)) |
955 | return; |
956 | } |
957 | |
958 | initial_reordering_consonant_syllable (plan, face, buffer, start, end); |
959 | } |
960 | |
961 | static void |
962 | initial_reordering_syllable_indic (const hb_ot_shape_plan_t *plan, |
963 | hb_face_t *face, |
964 | hb_buffer_t *buffer, |
965 | unsigned int start, unsigned int end) |
966 | { |
967 | indic_syllable_type_t syllable_type = (indic_syllable_type_t) (buffer->info[start].syllable() & 0x0F); |
968 | switch (syllable_type) |
969 | { |
970 | case indic_vowel_syllable: /* We made the vowels look like consonants. So let's call the consonant logic! */ |
971 | case indic_consonant_syllable: |
972 | initial_reordering_consonant_syllable (plan, face, buffer, start, end); |
973 | break; |
974 | |
975 | case indic_broken_cluster: /* We already inserted dotted-circles, so just call the standalone_cluster. */ |
976 | case indic_standalone_cluster: |
977 | initial_reordering_standalone_cluster (plan, face, buffer, start, end); |
978 | break; |
979 | |
980 | case indic_symbol_cluster: |
981 | case indic_non_indic_cluster: |
982 | break; |
983 | } |
984 | } |
985 | |
986 | static bool |
987 | initial_reordering_indic (const hb_ot_shape_plan_t *plan, |
988 | hb_font_t *font, |
989 | hb_buffer_t *buffer) |
990 | { |
991 | bool ret = false; |
992 | if (!buffer->message (font, "start reordering indic initial" )) |
993 | return ret; |
994 | |
995 | update_consonant_positions_indic (plan, font, buffer); |
996 | if (hb_syllabic_insert_dotted_circles (font, buffer, |
997 | indic_broken_cluster, |
998 | I_Cat(DOTTEDCIRCLE), |
999 | I_Cat(Repha), |
1000 | POS_END)) |
1001 | ret = true; |
1002 | |
1003 | foreach_syllable (buffer, start, end) |
1004 | initial_reordering_syllable_indic (plan, font->face, buffer, start, end); |
1005 | |
1006 | (void) buffer->message (font, "end reordering indic initial" ); |
1007 | |
1008 | return ret; |
1009 | } |
1010 | |
1011 | static void |
1012 | final_reordering_syllable_indic (const hb_ot_shape_plan_t *plan, |
1013 | hb_buffer_t *buffer, |
1014 | unsigned int start, unsigned int end) |
1015 | { |
1016 | const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; |
1017 | hb_glyph_info_t *info = buffer->info; |
1018 | |
1019 | |
1020 | /* This function relies heavily on halant glyphs. Lots of ligation |
1021 | * and possibly multiple substitutions happened prior to this |
1022 | * phase, and that might have messed up our properties. Recover |
1023 | * from a particular case of that where we're fairly sure that a |
1024 | * class of I_Cat(H) is desired but has been lost. */ |
1025 | /* We don't call load_virama_glyph(), since we know it's already |
1026 | * loaded. */ |
1027 | hb_codepoint_t virama_glyph = indic_plan->virama_glyph; |
1028 | if (virama_glyph) |
1029 | { |
1030 | for (unsigned int i = start; i < end; i++) |
1031 | if (info[i].codepoint == virama_glyph && |
1032 | _hb_glyph_info_ligated (&info[i]) && |
1033 | _hb_glyph_info_multiplied (&info[i])) |
1034 | { |
1035 | /* This will make sure that this glyph passes is_halant() test. */ |
1036 | info[i].indic_category() = I_Cat(H); |
1037 | _hb_glyph_info_clear_ligated_and_multiplied (&info[i]); |
1038 | } |
1039 | } |
1040 | |
1041 | |
1042 | /* 4. Final reordering: |
1043 | * |
1044 | * After the localized forms and basic shaping forms GSUB features have been |
1045 | * applied (see below), the shaping engine performs some final glyph |
1046 | * reordering before applying all the remaining font features to the entire |
1047 | * syllable. |
1048 | */ |
1049 | |
1050 | bool try_pref = !!indic_plan->mask_array[INDIC_PREF]; |
1051 | |
1052 | /* Find base again */ |
1053 | unsigned int base; |
1054 | for (base = start; base < end; base++) |
1055 | if (info[base].indic_position() >= POS_BASE_C) |
1056 | { |
1057 | if (try_pref && base + 1 < end) |
1058 | { |
1059 | for (unsigned int i = base + 1; i < end; i++) |
1060 | if ((info[i].mask & indic_plan->mask_array[INDIC_PREF]) != 0) |
1061 | { |
1062 | if (!(_hb_glyph_info_substituted (&info[i]) && |
1063 | _hb_glyph_info_ligated_and_didnt_multiply (&info[i]))) |
1064 | { |
1065 | /* Ok, this was a 'pref' candidate but didn't form any. |
1066 | * Base is around here... */ |
1067 | base = i; |
1068 | while (base < end && is_halant (info[base])) |
1069 | base++; |
1070 | if (base < end) |
1071 | info[base].indic_position() = POS_BASE_C; |
1072 | |
1073 | try_pref = false; |
1074 | } |
1075 | break; |
1076 | } |
1077 | if (base == end) |
1078 | break; |
1079 | } |
1080 | /* For Malayalam, skip over unformed below- (but NOT post-) forms. */ |
1081 | if (buffer->props.script == HB_SCRIPT_MALAYALAM) |
1082 | { |
1083 | for (unsigned int i = base + 1; i < end; i++) |
1084 | { |
1085 | while (i < end && is_joiner (info[i])) |
1086 | i++; |
1087 | if (i == end || !is_halant (info[i])) |
1088 | break; |
1089 | i++; /* Skip halant. */ |
1090 | while (i < end && is_joiner (info[i])) |
1091 | i++; |
1092 | if (i < end && is_consonant (info[i]) && info[i].indic_position() == POS_BELOW_C) |
1093 | { |
1094 | base = i; |
1095 | info[base].indic_position() = POS_BASE_C; |
1096 | } |
1097 | } |
1098 | } |
1099 | |
1100 | if (start < base && info[base].indic_position() > POS_BASE_C) |
1101 | base--; |
1102 | break; |
1103 | } |
1104 | if (base == end && start < base && |
1105 | is_one_of (info[base - 1], FLAG (I_Cat(ZWJ)))) |
1106 | base--; |
1107 | if (base < end) |
1108 | while (start < base && |
1109 | is_one_of (info[base], (FLAG (I_Cat(N)) | FLAG (I_Cat(H))))) |
1110 | base--; |
1111 | |
1112 | |
1113 | /* o Reorder matras: |
1114 | * |
1115 | * If a pre-base matra character had been reordered before applying basic |
1116 | * features, the glyph can be moved closer to the main consonant based on |
1117 | * whether half-forms had been formed. Actual position for the matra is |
1118 | * defined as “after last standalone halant glyph, after initial matra |
1119 | * position and before the main consonant”. If ZWJ or ZWNJ follow this |
1120 | * halant, position is moved after it. |
1121 | * |
1122 | * IMPLEMENTATION NOTES: |
1123 | * |
1124 | * It looks like the last sentence is wrong. Testing, with Windows 7 Uniscribe |
1125 | * and Devanagari shows that the behavior is best described as: |
1126 | * |
1127 | * "If ZWJ follows this halant, matra is NOT repositioned after this halant. |
1128 | * If ZWNJ follows this halant, position is moved after it." |
1129 | * |
1130 | * Test case, with Adobe Devanagari or Nirmala UI: |
1131 | * |
1132 | * U+091F,U+094D,U+200C,U+092F,U+093F |
1133 | * (Matra moves to the middle, after ZWNJ.) |
1134 | * |
1135 | * U+091F,U+094D,U+200D,U+092F,U+093F |
1136 | * (Matra does NOT move, stays to the left.) |
1137 | * |
1138 | * https://github.com/harfbuzz/harfbuzz/issues/1070 |
1139 | */ |
1140 | |
1141 | if (start + 1 < end && start < base) /* Otherwise there can't be any pre-base matra characters. */ |
1142 | { |
1143 | /* If we lost track of base, alas, position before last thingy. */ |
1144 | unsigned int new_pos = base == end ? base - 2 : base - 1; |
1145 | |
1146 | /* Malayalam / Tamil do not have "half" forms or explicit virama forms. |
1147 | * The glyphs formed by 'half' are Chillus or ligated explicit viramas. |
1148 | * We want to position matra after them. |
1149 | */ |
1150 | if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL) |
1151 | { |
1152 | search: |
1153 | while (new_pos > start && |
1154 | !(is_one_of (info[new_pos], (FLAG (I_Cat(M)) | FLAG (I_Cat(MPst)) | FLAG (I_Cat(H)))))) |
1155 | new_pos--; |
1156 | |
1157 | /* If we found no Halant we are done. |
1158 | * Otherwise only proceed if the Halant does |
1159 | * not belong to the Matra itself! */ |
1160 | if (is_halant (info[new_pos]) && |
1161 | info[new_pos].indic_position() != POS_PRE_M) |
1162 | { |
1163 | #if 0 // See comment above |
1164 | /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */ |
1165 | if (new_pos + 1 < end && is_joiner (info[new_pos + 1])) |
1166 | new_pos++; |
1167 | #endif |
1168 | if (new_pos + 1 < end) |
1169 | { |
1170 | /* -> If ZWJ follows this halant, matra is NOT repositioned after this halant. */ |
1171 | if (info[new_pos + 1].indic_category() == I_Cat(ZWJ)) |
1172 | { |
1173 | /* Keep searching. */ |
1174 | if (new_pos > start) |
1175 | { |
1176 | new_pos--; |
1177 | goto search; |
1178 | } |
1179 | } |
1180 | /* -> If ZWNJ follows this halant, position is moved after it. |
1181 | * |
1182 | * IMPLEMENTATION NOTES: |
1183 | * |
1184 | * This is taken care of by the state-machine. A Halant,ZWNJ is a terminating |
1185 | * sequence for a consonant syllable; any pre-base matras occurring after it |
1186 | * will belong to the subsequent syllable. |
1187 | */ |
1188 | } |
1189 | } |
1190 | else |
1191 | new_pos = start; /* No move. */ |
1192 | } |
1193 | |
1194 | if (start < new_pos && info[new_pos].indic_position () != POS_PRE_M) |
1195 | { |
1196 | /* Now go see if there's actually any matras... */ |
1197 | for (unsigned int i = new_pos; i > start; i--) |
1198 | if (info[i - 1].indic_position () == POS_PRE_M) |
1199 | { |
1200 | unsigned int old_pos = i - 1; |
1201 | if (old_pos < base && base <= new_pos) /* Shouldn't actually happen. */ |
1202 | base--; |
1203 | |
1204 | hb_glyph_info_t tmp = info[old_pos]; |
1205 | memmove (&info[old_pos], &info[old_pos + 1], (new_pos - old_pos) * sizeof (info[0])); |
1206 | info[new_pos] = tmp; |
1207 | |
1208 | /* Note: this merge_clusters() is intentionally *after* the reordering. |
1209 | * Indic matra reordering is special and tricky... */ |
1210 | buffer->merge_clusters (new_pos, hb_min (end, base + 1)); |
1211 | |
1212 | new_pos--; |
1213 | } |
1214 | } else { |
1215 | for (unsigned int i = start; i < base; i++) |
1216 | if (info[i].indic_position () == POS_PRE_M) { |
1217 | buffer->merge_clusters (i, hb_min (end, base + 1)); |
1218 | break; |
1219 | } |
1220 | } |
1221 | } |
1222 | |
1223 | |
1224 | /* o Reorder reph: |
1225 | * |
1226 | * Reph’s original position is always at the beginning of the syllable, |
1227 | * (i.e. it is not reordered at the character reordering stage). However, |
1228 | * it will be reordered according to the basic-forms shaping results. |
1229 | * Possible positions for reph, depending on the script, are; after main, |
1230 | * before post-base consonant forms, and after post-base consonant forms. |
1231 | */ |
1232 | |
1233 | /* Two cases: |
1234 | * |
1235 | * - If repha is encoded as a sequence of characters (Ra,H or Ra,H,ZWJ), then |
1236 | * we should only move it if the sequence ligated to the repha form. |
1237 | * |
1238 | * - If repha is encoded separately and in the logical position, we should only |
1239 | * move it if it did NOT ligate. If it ligated, it's probably the font trying |
1240 | * to make it work without the reordering. |
1241 | */ |
1242 | if (start + 1 < end && |
1243 | info[start].indic_position() == POS_RA_TO_BECOME_REPH && |
1244 | ((info[start].indic_category() == I_Cat(Repha)) ^ |
1245 | _hb_glyph_info_ligated_and_didnt_multiply (&info[start]))) |
1246 | { |
1247 | unsigned int new_reph_pos; |
1248 | reph_position_t reph_pos = indic_plan->config->reph_pos; |
1249 | |
1250 | /* 1. If reph should be positioned after post-base consonant forms, |
1251 | * proceed to step 5. |
1252 | */ |
1253 | if (reph_pos == REPH_POS_AFTER_POST) |
1254 | { |
1255 | goto reph_step_5; |
1256 | } |
1257 | |
1258 | /* 2. If the reph repositioning class is not after post-base: target |
1259 | * position is after the first explicit halant glyph between the |
1260 | * first post-reph consonant and last main consonant. If ZWJ or ZWNJ |
1261 | * are following this halant, position is moved after it. If such |
1262 | * position is found, this is the target position. Otherwise, |
1263 | * proceed to the next step. |
1264 | * |
1265 | * Note: in old-implementation fonts, where classifications were |
1266 | * fixed in shaping engine, there was no case where reph position |
1267 | * will be found on this step. |
1268 | */ |
1269 | { |
1270 | new_reph_pos = start + 1; |
1271 | while (new_reph_pos < base && !is_halant (info[new_reph_pos])) |
1272 | new_reph_pos++; |
1273 | |
1274 | if (new_reph_pos < base && is_halant (info[new_reph_pos])) |
1275 | { |
1276 | /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */ |
1277 | if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1])) |
1278 | new_reph_pos++; |
1279 | goto reph_move; |
1280 | } |
1281 | } |
1282 | |
1283 | /* 3. If reph should be repositioned after the main consonant: find the |
1284 | * first consonant not ligated with main, or find the first |
1285 | * consonant that is not a potential pre-base-reordering Ra. |
1286 | */ |
1287 | if (reph_pos == REPH_POS_AFTER_MAIN) |
1288 | { |
1289 | new_reph_pos = base; |
1290 | while (new_reph_pos + 1 < end && info[new_reph_pos + 1].indic_position() <= POS_AFTER_MAIN) |
1291 | new_reph_pos++; |
1292 | if (new_reph_pos < end) |
1293 | goto reph_move; |
1294 | } |
1295 | |
1296 | /* 4. If reph should be positioned before post-base consonant, find |
1297 | * first post-base classified consonant not ligated with main. If no |
1298 | * consonant is found, the target position should be before the |
1299 | * first matra, syllable modifier sign or vedic sign. |
1300 | */ |
1301 | /* This is our take on what step 4 is trying to say (and failing, BADLY). */ |
1302 | if (reph_pos == REPH_POS_AFTER_SUB) |
1303 | { |
1304 | new_reph_pos = base; |
1305 | while (new_reph_pos + 1 < end && |
1306 | !( FLAG_UNSAFE (info[new_reph_pos + 1].indic_position()) & (FLAG (POS_POST_C) | FLAG (POS_AFTER_POST) | FLAG (POS_SMVD)))) |
1307 | new_reph_pos++; |
1308 | if (new_reph_pos < end) |
1309 | goto reph_move; |
1310 | } |
1311 | |
1312 | /* 5. If no consonant is found in steps 3 or 4, move reph to a position |
1313 | * immediately before the first post-base matra, syllable modifier |
1314 | * sign or vedic sign that has a reordering class after the intended |
1315 | * reph position. For example, if the reordering position for reph |
1316 | * is post-main, it will skip above-base matras that also have a |
1317 | * post-main position. |
1318 | */ |
1319 | reph_step_5: |
1320 | { |
1321 | /* Copied from step 2. */ |
1322 | new_reph_pos = start + 1; |
1323 | while (new_reph_pos < base && !is_halant (info[new_reph_pos])) |
1324 | new_reph_pos++; |
1325 | |
1326 | if (new_reph_pos < base && is_halant (info[new_reph_pos])) |
1327 | { |
1328 | /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */ |
1329 | if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1])) |
1330 | new_reph_pos++; |
1331 | goto reph_move; |
1332 | } |
1333 | } |
1334 | /* See https://github.com/harfbuzz/harfbuzz/issues/2298#issuecomment-615318654 */ |
1335 | |
1336 | /* 6. Otherwise, reorder reph to the end of the syllable. |
1337 | */ |
1338 | { |
1339 | new_reph_pos = end - 1; |
1340 | while (new_reph_pos > start && info[new_reph_pos].indic_position() == POS_SMVD) |
1341 | new_reph_pos--; |
1342 | |
1343 | /* |
1344 | * If the Reph is to be ending up after a Matra,Halant sequence, |
1345 | * position it before that Halant so it can interact with the Matra. |
1346 | * However, if it's a plain Consonant,Halant we shouldn't do that. |
1347 | * Uniscribe doesn't do this. |
1348 | * TEST: U+0930,U+094D,U+0915,U+094B,U+094D |
1349 | */ |
1350 | if (!indic_plan->uniscribe_bug_compatible && |
1351 | unlikely (is_halant (info[new_reph_pos]))) |
1352 | { |
1353 | for (unsigned int i = base + 1; i < new_reph_pos; i++) |
1354 | if (FLAG_UNSAFE (info[i].indic_category()) & (FLAG (I_Cat(M)) | FLAG (I_Cat(MPst)))) |
1355 | { |
1356 | /* Ok, got it. */ |
1357 | new_reph_pos--; |
1358 | } |
1359 | } |
1360 | |
1361 | goto reph_move; |
1362 | } |
1363 | |
1364 | reph_move: |
1365 | { |
1366 | /* Move */ |
1367 | buffer->merge_clusters (start, new_reph_pos + 1); |
1368 | hb_glyph_info_t reph = info[start]; |
1369 | memmove (&info[start], &info[start + 1], (new_reph_pos - start) * sizeof (info[0])); |
1370 | info[new_reph_pos] = reph; |
1371 | |
1372 | if (start < base && base <= new_reph_pos) |
1373 | base--; |
1374 | } |
1375 | } |
1376 | |
1377 | |
1378 | /* o Reorder pre-base-reordering consonants: |
1379 | * |
1380 | * If a pre-base-reordering consonant is found, reorder it according to |
1381 | * the following rules: |
1382 | */ |
1383 | |
1384 | if (try_pref && base + 1 < end) /* Otherwise there can't be any pre-base-reordering Ra. */ |
1385 | { |
1386 | for (unsigned int i = base + 1; i < end; i++) |
1387 | if ((info[i].mask & indic_plan->mask_array[INDIC_PREF]) != 0) |
1388 | { |
1389 | /* 1. Only reorder a glyph produced by substitution during application |
1390 | * of the <pref> feature. (Note that a font may shape a Ra consonant with |
1391 | * the feature generally but block it in certain contexts.) |
1392 | */ |
1393 | /* Note: We just check that something got substituted. We don't check that |
1394 | * the <pref> feature actually did it... |
1395 | * |
1396 | * Reorder pref only if it ligated. */ |
1397 | if (_hb_glyph_info_ligated_and_didnt_multiply (&info[i])) |
1398 | { |
1399 | /* |
1400 | * 2. Try to find a target position the same way as for pre-base matra. |
1401 | * If it is found, reorder pre-base consonant glyph. |
1402 | * |
1403 | * 3. If position is not found, reorder immediately before main |
1404 | * consonant. |
1405 | */ |
1406 | |
1407 | unsigned int new_pos = base; |
1408 | /* Malayalam / Tamil do not have "half" forms or explicit virama forms. |
1409 | * The glyphs formed by 'half' are Chillus or ligated explicit viramas. |
1410 | * We want to position matra after them. |
1411 | */ |
1412 | if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL) |
1413 | { |
1414 | while (new_pos > start && |
1415 | !(is_one_of (info[new_pos - 1], FLAG (I_Cat(M)) | FLAG (I_Cat(MPst)) | FLAG (I_Cat(H))))) |
1416 | new_pos--; |
1417 | } |
1418 | |
1419 | if (new_pos > start && is_halant (info[new_pos - 1])) |
1420 | { |
1421 | /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */ |
1422 | if (new_pos < end && is_joiner (info[new_pos])) |
1423 | new_pos++; |
1424 | } |
1425 | |
1426 | { |
1427 | unsigned int old_pos = i; |
1428 | |
1429 | buffer->merge_clusters (new_pos, old_pos + 1); |
1430 | hb_glyph_info_t tmp = info[old_pos]; |
1431 | memmove (&info[new_pos + 1], &info[new_pos], (old_pos - new_pos) * sizeof (info[0])); |
1432 | info[new_pos] = tmp; |
1433 | |
1434 | if (new_pos <= base && base < old_pos) |
1435 | base++; |
1436 | } |
1437 | } |
1438 | |
1439 | break; |
1440 | } |
1441 | } |
1442 | |
1443 | |
1444 | /* Apply 'init' to the Left Matra if it's a word start. */ |
1445 | if (info[start].indic_position () == POS_PRE_M) |
1446 | { |
1447 | if (!start || |
1448 | !(FLAG_UNSAFE (_hb_glyph_info_get_general_category (&info[start - 1])) & |
1449 | FLAG_RANGE (HB_UNICODE_GENERAL_CATEGORY_FORMAT, HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK))) |
1450 | info[start].mask |= indic_plan->mask_array[INDIC_INIT]; |
1451 | else |
1452 | buffer->unsafe_to_break (start - 1, start + 1); |
1453 | } |
1454 | |
1455 | |
1456 | /* |
1457 | * Finish off the clusters and go home! |
1458 | */ |
1459 | if (indic_plan->uniscribe_bug_compatible) |
1460 | { |
1461 | switch ((hb_tag_t) plan->props.script) |
1462 | { |
1463 | case HB_SCRIPT_TAMIL: |
1464 | break; |
1465 | |
1466 | default: |
1467 | /* Uniscribe merges the entire syllable into a single cluster... Except for Tamil. |
1468 | * This means, half forms are submerged into the main consonant's cluster. |
1469 | * This is unnecessary, and makes cursor positioning harder, but that's what |
1470 | * Uniscribe does. */ |
1471 | buffer->merge_clusters (start, end); |
1472 | break; |
1473 | } |
1474 | } |
1475 | } |
1476 | |
1477 | |
1478 | static bool |
1479 | final_reordering_indic (const hb_ot_shape_plan_t *plan, |
1480 | hb_font_t *font HB_UNUSED, |
1481 | hb_buffer_t *buffer) |
1482 | { |
1483 | unsigned int count = buffer->len; |
1484 | if (unlikely (!count)) return false; |
1485 | |
1486 | if (buffer->message (font, "start reordering indic final" )) { |
1487 | foreach_syllable (buffer, start, end) |
1488 | final_reordering_syllable_indic (plan, buffer, start, end); |
1489 | (void) buffer->message (font, "end reordering indic final" ); |
1490 | } |
1491 | |
1492 | HB_BUFFER_DEALLOCATE_VAR (buffer, indic_category); |
1493 | HB_BUFFER_DEALLOCATE_VAR (buffer, indic_position); |
1494 | |
1495 | return false; |
1496 | } |
1497 | |
1498 | |
1499 | static void |
1500 | preprocess_text_indic (const hb_ot_shape_plan_t *plan, |
1501 | hb_buffer_t *buffer, |
1502 | hb_font_t *font) |
1503 | { |
1504 | const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data; |
1505 | if (!indic_plan->uniscribe_bug_compatible) |
1506 | _hb_preprocess_text_vowel_constraints (plan, buffer, font); |
1507 | } |
1508 | |
1509 | static bool |
1510 | decompose_indic (const hb_ot_shape_normalize_context_t *c, |
1511 | hb_codepoint_t ab, |
1512 | hb_codepoint_t *a, |
1513 | hb_codepoint_t *b) |
1514 | { |
1515 | switch (ab) |
1516 | { |
1517 | /* Don't decompose these. */ |
1518 | case 0x0931u : return false; /* DEVANAGARI LETTER RRA */ |
1519 | // https://github.com/harfbuzz/harfbuzz/issues/779 |
1520 | case 0x09DCu : return false; /* BENGALI LETTER RRA */ |
1521 | case 0x09DDu : return false; /* BENGALI LETTER RHA */ |
1522 | case 0x0B94u : return false; /* TAMIL LETTER AU */ |
1523 | |
1524 | |
1525 | /* |
1526 | * Decompose split matras that don't have Unicode decompositions. |
1527 | */ |
1528 | |
1529 | #if 0 |
1530 | /* Gujarati */ |
1531 | /* This one has no decomposition in Unicode, but needs no decomposition either. */ |
1532 | /* case 0x0AC9u : return false; */ |
1533 | |
1534 | /* Oriya */ |
1535 | case 0x0B57u : *a = no decomp, -> RIGHT; return true; |
1536 | #endif |
1537 | } |
1538 | |
1539 | return (bool) c->unicode->decompose (ab, a, b); |
1540 | } |
1541 | |
1542 | static bool |
1543 | compose_indic (const hb_ot_shape_normalize_context_t *c, |
1544 | hb_codepoint_t a, |
1545 | hb_codepoint_t b, |
1546 | hb_codepoint_t *ab) |
1547 | { |
1548 | /* Avoid recomposing split matras. */ |
1549 | if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a))) |
1550 | return false; |
1551 | |
1552 | /* Composition-exclusion exceptions that we want to recompose. */ |
1553 | if (a == 0x09AFu && b == 0x09BCu) { *ab = 0x09DFu; return true; } |
1554 | |
1555 | return (bool) c->unicode->compose (a, b, ab); |
1556 | } |
1557 | |
1558 | |
1559 | const hb_ot_shaper_t _hb_ot_shaper_indic = |
1560 | { |
1561 | collect_features_indic, |
1562 | override_features_indic, |
1563 | data_create_indic, |
1564 | data_destroy_indic, |
1565 | preprocess_text_indic, |
1566 | nullptr, /* postprocess_glyphs */ |
1567 | decompose_indic, |
1568 | compose_indic, |
1569 | setup_masks_indic, |
1570 | nullptr, /* reorder_marks */ |
1571 | HB_TAG_NONE, /* gpos_tag */ |
1572 | HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT, |
1573 | HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE, |
1574 | false, /* fallback_position */ |
1575 | }; |
1576 | |
1577 | |
1578 | #endif |
1579 | |