1 | /* |
2 | * Copyright © 2010,2012 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-arabic.hh" |
32 | #include "hb-ot-shape.hh" |
33 | |
34 | |
35 | /* buffer var allocations */ |
36 | #define arabic_shaping_action() ot_shaper_var_u8_auxiliary() /* arabic shaping action */ |
37 | |
38 | #define HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH HB_BUFFER_SCRATCH_FLAG_SHAPER0 |
39 | |
40 | /* See: |
41 | * https://github.com/harfbuzz/harfbuzz/commit/6e6f82b6f3dde0fc6c3c7d991d9ec6cfff57823d#commitcomment-14248516 */ |
42 | #define HB_ARABIC_GENERAL_CATEGORY_IS_WORD(gen_cat) \ |
43 | (FLAG_UNSAFE (gen_cat) & \ |
44 | (FLAG (HB_UNICODE_GENERAL_CATEGORY_UNASSIGNED) | \ |
45 | FLAG (HB_UNICODE_GENERAL_CATEGORY_PRIVATE_USE) | \ |
46 | /*FLAG (HB_UNICODE_GENERAL_CATEGORY_LOWERCASE_LETTER) |*/ \ |
47 | FLAG (HB_UNICODE_GENERAL_CATEGORY_MODIFIER_LETTER) | \ |
48 | FLAG (HB_UNICODE_GENERAL_CATEGORY_OTHER_LETTER) | \ |
49 | /*FLAG (HB_UNICODE_GENERAL_CATEGORY_TITLECASE_LETTER) |*/ \ |
50 | /*FLAG (HB_UNICODE_GENERAL_CATEGORY_UPPERCASE_LETTER) |*/ \ |
51 | FLAG (HB_UNICODE_GENERAL_CATEGORY_SPACING_MARK) | \ |
52 | FLAG (HB_UNICODE_GENERAL_CATEGORY_ENCLOSING_MARK) | \ |
53 | FLAG (HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK) | \ |
54 | FLAG (HB_UNICODE_GENERAL_CATEGORY_DECIMAL_NUMBER) | \ |
55 | FLAG (HB_UNICODE_GENERAL_CATEGORY_LETTER_NUMBER) | \ |
56 | FLAG (HB_UNICODE_GENERAL_CATEGORY_OTHER_NUMBER) | \ |
57 | FLAG (HB_UNICODE_GENERAL_CATEGORY_CURRENCY_SYMBOL) | \ |
58 | FLAG (HB_UNICODE_GENERAL_CATEGORY_MODIFIER_SYMBOL) | \ |
59 | FLAG (HB_UNICODE_GENERAL_CATEGORY_MATH_SYMBOL) | \ |
60 | FLAG (HB_UNICODE_GENERAL_CATEGORY_OTHER_SYMBOL))) |
61 | |
62 | |
63 | /* |
64 | * Joining types: |
65 | */ |
66 | |
67 | /* |
68 | * Bits used in the joining tables |
69 | */ |
70 | enum hb_arabic_joining_type_t { |
71 | JOINING_TYPE_U = 0, |
72 | JOINING_TYPE_L = 1, |
73 | JOINING_TYPE_R = 2, |
74 | JOINING_TYPE_D = 3, |
75 | JOINING_TYPE_C = JOINING_TYPE_D, |
76 | JOINING_GROUP_ALAPH = 4, |
77 | JOINING_GROUP_DALATH_RISH = 5, |
78 | NUM_STATE_MACHINE_COLS = 6, |
79 | |
80 | JOINING_TYPE_T = 7, |
81 | JOINING_TYPE_X = 8 /* means: use general-category to choose between U or T. */ |
82 | }; |
83 | |
84 | #include "hb-ot-shaper-arabic-table.hh" |
85 | |
86 | static unsigned int get_joining_type (hb_codepoint_t u, hb_unicode_general_category_t gen_cat) |
87 | { |
88 | unsigned int j_type = joining_type(u); |
89 | if (likely (j_type != JOINING_TYPE_X)) |
90 | return j_type; |
91 | |
92 | return (FLAG_UNSAFE(gen_cat) & |
93 | (FLAG(HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK) | |
94 | FLAG(HB_UNICODE_GENERAL_CATEGORY_ENCLOSING_MARK) | |
95 | FLAG(HB_UNICODE_GENERAL_CATEGORY_FORMAT)) |
96 | ) ? JOINING_TYPE_T : JOINING_TYPE_U; |
97 | } |
98 | |
99 | #define FEATURE_IS_SYRIAC(tag) hb_in_range<unsigned char> ((unsigned char) (tag), '2', '3') |
100 | |
101 | static const hb_tag_t arabic_features[] = |
102 | { |
103 | HB_TAG('i','s','o','l'), |
104 | HB_TAG('f','i','n','a'), |
105 | HB_TAG('f','i','n','2'), |
106 | HB_TAG('f','i','n','3'), |
107 | HB_TAG('m','e','d','i'), |
108 | HB_TAG('m','e','d','2'), |
109 | HB_TAG('i','n','i','t'), |
110 | HB_TAG_NONE |
111 | }; |
112 | |
113 | |
114 | /* Same order as the feature array */ |
115 | enum arabic_action_t { |
116 | ISOL, |
117 | FINA, |
118 | FIN2, |
119 | FIN3, |
120 | MEDI, |
121 | MED2, |
122 | INIT, |
123 | |
124 | NONE, |
125 | |
126 | ARABIC_NUM_FEATURES = NONE, |
127 | |
128 | /* We abuse the same byte for other things... */ |
129 | STCH_FIXED, |
130 | STCH_REPEATING, |
131 | }; |
132 | |
133 | static const struct arabic_state_table_entry { |
134 | uint8_t prev_action; |
135 | uint8_t curr_action; |
136 | uint16_t next_state; |
137 | } arabic_state_table[][NUM_STATE_MACHINE_COLS] = |
138 | { |
139 | /* jt_U, jt_L, jt_R, jt_D, jg_ALAPH, jg_DALATH_RISH */ |
140 | |
141 | /* State 0: prev was U, not willing to join. */ |
142 | { {NONE,NONE,0}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,6}, }, |
143 | |
144 | /* State 1: prev was R or ISOL/ALAPH, not willing to join. */ |
145 | { {NONE,NONE,0}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,FIN2,5}, {NONE,ISOL,6}, }, |
146 | |
147 | /* State 2: prev was D/L in ISOL form, willing to join. */ |
148 | { {NONE,NONE,0}, {NONE,ISOL,2}, {INIT,FINA,1}, {INIT,FINA,3}, {INIT,FINA,4}, {INIT,FINA,6}, }, |
149 | |
150 | /* State 3: prev was D in FINA form, willing to join. */ |
151 | { {NONE,NONE,0}, {NONE,ISOL,2}, {MEDI,FINA,1}, {MEDI,FINA,3}, {MEDI,FINA,4}, {MEDI,FINA,6}, }, |
152 | |
153 | /* State 4: prev was FINA ALAPH, not willing to join. */ |
154 | { {NONE,NONE,0}, {NONE,ISOL,2}, {MED2,ISOL,1}, {MED2,ISOL,2}, {MED2,FIN2,5}, {MED2,ISOL,6}, }, |
155 | |
156 | /* State 5: prev was FIN2/FIN3 ALAPH, not willing to join. */ |
157 | { {NONE,NONE,0}, {NONE,ISOL,2}, {ISOL,ISOL,1}, {ISOL,ISOL,2}, {ISOL,FIN2,5}, {ISOL,ISOL,6}, }, |
158 | |
159 | /* State 6: prev was DALATH/RISH, not willing to join. */ |
160 | { {NONE,NONE,0}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,FIN3,5}, {NONE,ISOL,6}, } |
161 | }; |
162 | |
163 | |
164 | static bool |
165 | arabic_fallback_shape (const hb_ot_shape_plan_t *plan, |
166 | hb_font_t *font, |
167 | hb_buffer_t *buffer); |
168 | |
169 | static bool |
170 | record_stch (const hb_ot_shape_plan_t *plan, |
171 | hb_font_t *font, |
172 | hb_buffer_t *buffer); |
173 | |
174 | static bool |
175 | deallocate_buffer_var (const hb_ot_shape_plan_t *plan, |
176 | hb_font_t *font, |
177 | hb_buffer_t *buffer) |
178 | { |
179 | HB_BUFFER_DEALLOCATE_VAR (buffer, arabic_shaping_action); |
180 | return false; |
181 | } |
182 | |
183 | static void |
184 | collect_features_arabic (hb_ot_shape_planner_t *plan) |
185 | { |
186 | hb_ot_map_builder_t *map = &plan->map; |
187 | |
188 | /* We apply features according to the Arabic spec, with pauses |
189 | * in between most. |
190 | * |
191 | * The pause between init/medi/... and rlig is required. See eg: |
192 | * https://bugzilla.mozilla.org/show_bug.cgi?id=644184 |
193 | * |
194 | * The pauses between init/medi/... themselves are not necessarily |
195 | * needed as only one of those features is applied to any character. |
196 | * The only difference it makes is when fonts have contextual |
197 | * substitutions. We now follow the order of the spec, which makes |
198 | * for better experience if that's what Uniscribe is doing. |
199 | * |
200 | * At least for Arabic, looks like Uniscribe has a pause between |
201 | * rlig and calt. Otherwise the IranNastaliq's ALLAH ligature won't |
202 | * work. However, testing shows that rlig and calt are applied |
203 | * together for Mongolian in Uniscribe. As such, we only add a |
204 | * pause for Arabic, not other scripts. |
205 | */ |
206 | |
207 | |
208 | map->enable_feature (HB_TAG('s','t','c','h')); |
209 | map->add_gsub_pause (record_stch); |
210 | |
211 | map->enable_feature (HB_TAG('c','c','m','p'), F_MANUAL_ZWJ); |
212 | map->enable_feature (HB_TAG('l','o','c','l'), F_MANUAL_ZWJ); |
213 | |
214 | map->add_gsub_pause (nullptr); |
215 | |
216 | for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++) |
217 | { |
218 | bool has_fallback = plan->props.script == HB_SCRIPT_ARABIC && !FEATURE_IS_SYRIAC (arabic_features[i]); |
219 | map->add_feature (arabic_features[i], F_MANUAL_ZWJ | (has_fallback ? F_HAS_FALLBACK : F_NONE)); |
220 | map->add_gsub_pause (nullptr); |
221 | } |
222 | map->add_gsub_pause (deallocate_buffer_var); |
223 | |
224 | /* Normally, Unicode says a ZWNJ means "don't ligate". In Arabic script |
225 | * however, it says a ZWJ should also mean "don't ligate". So we run |
226 | * the main ligating features as MANUAL_ZWJ. */ |
227 | |
228 | map->enable_feature (HB_TAG('r','l','i','g'), F_MANUAL_ZWJ | F_HAS_FALLBACK); |
229 | |
230 | if (plan->props.script == HB_SCRIPT_ARABIC) |
231 | map->add_gsub_pause (arabic_fallback_shape); |
232 | |
233 | map->enable_feature (HB_TAG('c','a','l','t'), F_MANUAL_ZWJ); |
234 | /* https://github.com/harfbuzz/harfbuzz/issues/1573 */ |
235 | if (!map->has_feature (HB_TAG('r','c','l','t'))) |
236 | { |
237 | map->add_gsub_pause (nullptr); |
238 | map->enable_feature (HB_TAG('r','c','l','t'), F_MANUAL_ZWJ); |
239 | } |
240 | |
241 | map->enable_feature (HB_TAG('l','i','g','a'), F_MANUAL_ZWJ); |
242 | map->enable_feature (HB_TAG('c','l','i','g'), F_MANUAL_ZWJ); |
243 | |
244 | /* The spec includes 'cswh'. Earlier versions of Windows |
245 | * used to enable this by default, but testing suggests |
246 | * that Windows 8 and later do not enable it by default, |
247 | * and spec now says 'Off by default'. |
248 | * We disabled this in ae23c24c32. |
249 | * Note that IranNastaliq uses this feature extensively |
250 | * to fixup broken glyph sequences. Oh well... |
251 | * Test case: U+0643,U+0640,U+0631. */ |
252 | //map->enable_feature (HB_TAG('c','s','w','h'), F_MANUAL_ZWJ); |
253 | map->enable_feature (HB_TAG('m','s','e','t'), F_MANUAL_ZWJ); |
254 | } |
255 | |
256 | #include "hb-ot-shaper-arabic-fallback.hh" |
257 | |
258 | struct arabic_shape_plan_t |
259 | { |
260 | /* The "+ 1" in the next array is to accommodate for the "NONE" command, |
261 | * which is not an OpenType feature, but this simplifies the code by not |
262 | * having to do a "if (... < NONE) ..." and just rely on the fact that |
263 | * mask_array[NONE] == 0. */ |
264 | hb_mask_t mask_array[ARABIC_NUM_FEATURES + 1]; |
265 | |
266 | hb_atomic_ptr_t<arabic_fallback_plan_t> fallback_plan; |
267 | |
268 | unsigned int do_fallback : 1; |
269 | unsigned int has_stch : 1; |
270 | }; |
271 | |
272 | void * |
273 | data_create_arabic (const hb_ot_shape_plan_t *plan) |
274 | { |
275 | arabic_shape_plan_t *arabic_plan = (arabic_shape_plan_t *) hb_calloc (1, sizeof (arabic_shape_plan_t)); |
276 | if (unlikely (!arabic_plan)) |
277 | return nullptr; |
278 | |
279 | arabic_plan->do_fallback = plan->props.script == HB_SCRIPT_ARABIC; |
280 | arabic_plan->has_stch = !!plan->map.get_1_mask (HB_TAG ('s','t','c','h')); |
281 | for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++) { |
282 | arabic_plan->mask_array[i] = plan->map.get_1_mask (arabic_features[i]); |
283 | arabic_plan->do_fallback = arabic_plan->do_fallback && |
284 | (FEATURE_IS_SYRIAC (arabic_features[i]) || |
285 | plan->map.needs_fallback (arabic_features[i])); |
286 | } |
287 | |
288 | return arabic_plan; |
289 | } |
290 | |
291 | void |
292 | data_destroy_arabic (void *data) |
293 | { |
294 | arabic_shape_plan_t *arabic_plan = (arabic_shape_plan_t *) data; |
295 | |
296 | arabic_fallback_plan_destroy (arabic_plan->fallback_plan); |
297 | |
298 | hb_free (data); |
299 | } |
300 | |
301 | static void |
302 | arabic_joining (hb_buffer_t *buffer) |
303 | { |
304 | unsigned int count = buffer->len; |
305 | hb_glyph_info_t *info = buffer->info; |
306 | unsigned int prev = UINT_MAX, state = 0; |
307 | |
308 | /* Check pre-context */ |
309 | for (unsigned int i = 0; i < buffer->context_len[0]; i++) |
310 | { |
311 | unsigned int this_type = get_joining_type (buffer->context[0][i], buffer->unicode->general_category (buffer->context[0][i])); |
312 | |
313 | if (unlikely (this_type == JOINING_TYPE_T)) |
314 | continue; |
315 | |
316 | const arabic_state_table_entry *entry = &arabic_state_table[state][this_type]; |
317 | state = entry->next_state; |
318 | break; |
319 | } |
320 | |
321 | for (unsigned int i = 0; i < count; i++) |
322 | { |
323 | unsigned int this_type = get_joining_type (info[i].codepoint, _hb_glyph_info_get_general_category (&info[i])); |
324 | |
325 | if (unlikely (this_type == JOINING_TYPE_T)) { |
326 | info[i].arabic_shaping_action() = NONE; |
327 | continue; |
328 | } |
329 | |
330 | const arabic_state_table_entry *entry = &arabic_state_table[state][this_type]; |
331 | |
332 | if (entry->prev_action != NONE && prev != UINT_MAX) |
333 | { |
334 | info[prev].arabic_shaping_action() = entry->prev_action; |
335 | buffer->safe_to_insert_tatweel (prev, i + 1); |
336 | } |
337 | else |
338 | { |
339 | if (prev == UINT_MAX) |
340 | { |
341 | if (this_type >= JOINING_TYPE_R) |
342 | buffer->unsafe_to_concat_from_outbuffer (0, i + 1); |
343 | } |
344 | else |
345 | { |
346 | if (this_type >= JOINING_TYPE_R || |
347 | (2 <= state && state <= 5) /* States that have a possible prev_action. */) |
348 | buffer->unsafe_to_concat (prev, i + 1); |
349 | } |
350 | } |
351 | |
352 | info[i].arabic_shaping_action() = entry->curr_action; |
353 | |
354 | prev = i; |
355 | state = entry->next_state; |
356 | } |
357 | |
358 | for (unsigned int i = 0; i < buffer->context_len[1]; i++) |
359 | { |
360 | unsigned int this_type = get_joining_type (buffer->context[1][i], buffer->unicode->general_category (buffer->context[1][i])); |
361 | |
362 | if (unlikely (this_type == JOINING_TYPE_T)) |
363 | continue; |
364 | |
365 | const arabic_state_table_entry *entry = &arabic_state_table[state][this_type]; |
366 | if (entry->prev_action != NONE && prev != UINT_MAX) |
367 | { |
368 | info[prev].arabic_shaping_action() = entry->prev_action; |
369 | buffer->safe_to_insert_tatweel (prev, buffer->len); |
370 | } |
371 | else if (2 <= state && state <= 5) /* States that have a possible prev_action. */ |
372 | { |
373 | buffer->unsafe_to_concat (prev, buffer->len); |
374 | } |
375 | break; |
376 | } |
377 | } |
378 | |
379 | static void |
380 | mongolian_variation_selectors (hb_buffer_t *buffer) |
381 | { |
382 | /* Copy arabic_shaping_action() from base to Mongolian variation selectors. */ |
383 | unsigned int count = buffer->len; |
384 | hb_glyph_info_t *info = buffer->info; |
385 | for (unsigned int i = 1; i < count; i++) |
386 | if (unlikely (hb_in_ranges<hb_codepoint_t> (info[i].codepoint, 0x180Bu, 0x180Du, 0x180Fu, 0x180Fu))) |
387 | info[i].arabic_shaping_action() = info[i - 1].arabic_shaping_action(); |
388 | } |
389 | |
390 | void |
391 | setup_masks_arabic_plan (const arabic_shape_plan_t *arabic_plan, |
392 | hb_buffer_t *buffer, |
393 | hb_script_t script) |
394 | { |
395 | HB_BUFFER_ALLOCATE_VAR (buffer, arabic_shaping_action); |
396 | |
397 | arabic_joining (buffer); |
398 | if (script == HB_SCRIPT_MONGOLIAN) |
399 | mongolian_variation_selectors (buffer); |
400 | |
401 | unsigned int count = buffer->len; |
402 | hb_glyph_info_t *info = buffer->info; |
403 | for (unsigned int i = 0; i < count; i++) |
404 | info[i].mask |= arabic_plan->mask_array[info[i].arabic_shaping_action()]; |
405 | } |
406 | |
407 | static void |
408 | setup_masks_arabic (const hb_ot_shape_plan_t *plan, |
409 | hb_buffer_t *buffer, |
410 | hb_font_t *font HB_UNUSED) |
411 | { |
412 | const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data; |
413 | setup_masks_arabic_plan (arabic_plan, buffer, plan->props.script); |
414 | } |
415 | |
416 | static bool |
417 | arabic_fallback_shape (const hb_ot_shape_plan_t *plan, |
418 | hb_font_t *font, |
419 | hb_buffer_t *buffer) |
420 | { |
421 | #ifdef HB_NO_OT_SHAPER_ARABIC_FALLBACK |
422 | return false; |
423 | #endif |
424 | |
425 | const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data; |
426 | |
427 | if (!arabic_plan->do_fallback) |
428 | return false; |
429 | |
430 | retry: |
431 | arabic_fallback_plan_t *fallback_plan = arabic_plan->fallback_plan; |
432 | if (unlikely (!fallback_plan)) |
433 | { |
434 | /* This sucks. We need a font to build the fallback plan... */ |
435 | fallback_plan = arabic_fallback_plan_create (plan, font); |
436 | if (unlikely (!arabic_plan->fallback_plan.cmpexch (nullptr, fallback_plan))) |
437 | { |
438 | arabic_fallback_plan_destroy (fallback_plan); |
439 | goto retry; |
440 | } |
441 | } |
442 | |
443 | arabic_fallback_plan_shape (fallback_plan, font, buffer); |
444 | return true; |
445 | } |
446 | |
447 | /* |
448 | * Stretch feature: "stch". |
449 | * See example here: |
450 | * https://docs.microsoft.com/en-us/typography/script-development/syriac |
451 | * We implement this in a generic way, such that the Arabic subtending |
452 | * marks can use it as well. |
453 | */ |
454 | |
455 | static bool |
456 | record_stch (const hb_ot_shape_plan_t *plan, |
457 | hb_font_t *font HB_UNUSED, |
458 | hb_buffer_t *buffer) |
459 | { |
460 | const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data; |
461 | if (!arabic_plan->has_stch) |
462 | return false; |
463 | |
464 | /* 'stch' feature was just applied. Look for anything that multiplied, |
465 | * and record it for stch treatment later. Note that rtlm, frac, etc |
466 | * are applied before stch, but we assume that they didn't result in |
467 | * anything multiplying into 5 pieces, so it's safe-ish... */ |
468 | |
469 | unsigned int count = buffer->len; |
470 | hb_glyph_info_t *info = buffer->info; |
471 | for (unsigned int i = 0; i < count; i++) |
472 | if (unlikely (_hb_glyph_info_multiplied (&info[i]))) |
473 | { |
474 | unsigned int comp = _hb_glyph_info_get_lig_comp (&info[i]); |
475 | info[i].arabic_shaping_action() = comp % 2 ? STCH_REPEATING : STCH_FIXED; |
476 | buffer->scratch_flags |= HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH; |
477 | } |
478 | return false; |
479 | } |
480 | |
481 | static void |
482 | apply_stch (const hb_ot_shape_plan_t *plan HB_UNUSED, |
483 | hb_buffer_t *buffer, |
484 | hb_font_t *font) |
485 | { |
486 | if (likely (!(buffer->scratch_flags & HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH))) |
487 | return; |
488 | |
489 | /* The Arabic shaper currently always processes in RTL mode, so we should |
490 | * stretch / position the stretched pieces to the left / preceding glyphs. */ |
491 | |
492 | /* We do a two pass implementation: |
493 | * First pass calculates the exact number of extra glyphs we need, |
494 | * We then enlarge buffer to have that much room, |
495 | * Second pass applies the stretch, copying things to the end of buffer. |
496 | */ |
497 | |
498 | int sign = font->x_scale < 0 ? -1 : +1; |
499 | unsigned int = 0; // Set during MEASURE, used during CUT |
500 | enum { MEASURE, CUT } /* step_t */; |
501 | |
502 | for (unsigned int step = MEASURE; step <= CUT; step = step + 1) |
503 | { |
504 | unsigned int count = buffer->len; |
505 | hb_glyph_info_t *info = buffer->info; |
506 | hb_glyph_position_t *pos = buffer->pos; |
507 | unsigned int new_len = count + extra_glyphs_needed; // write head during CUT |
508 | unsigned int j = new_len; |
509 | for (unsigned int i = count; i; i--) |
510 | { |
511 | if (!hb_in_range<uint8_t> (info[i - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING)) |
512 | { |
513 | if (step == CUT) |
514 | { |
515 | --j; |
516 | info[j] = info[i - 1]; |
517 | pos[j] = pos[i - 1]; |
518 | } |
519 | continue; |
520 | } |
521 | |
522 | /* Yay, justification! */ |
523 | |
524 | hb_position_t w_total = 0; // Total to be filled |
525 | hb_position_t w_fixed = 0; // Sum of fixed tiles |
526 | hb_position_t w_repeating = 0; // Sum of repeating tiles |
527 | int n_fixed = 0; |
528 | int n_repeating = 0; |
529 | |
530 | unsigned int end = i; |
531 | while (i && |
532 | hb_in_range<uint8_t> (info[i - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING)) |
533 | { |
534 | i--; |
535 | hb_position_t width = font->get_glyph_h_advance (info[i].codepoint); |
536 | if (info[i].arabic_shaping_action() == STCH_FIXED) |
537 | { |
538 | w_fixed += width; |
539 | n_fixed++; |
540 | } |
541 | else |
542 | { |
543 | w_repeating += width; |
544 | n_repeating++; |
545 | } |
546 | } |
547 | unsigned int start = i; |
548 | unsigned int context = i; |
549 | while (context && |
550 | !hb_in_range<uint8_t> (info[context - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING) && |
551 | (_hb_glyph_info_is_default_ignorable (&info[context - 1]) || |
552 | HB_ARABIC_GENERAL_CATEGORY_IS_WORD (_hb_glyph_info_get_general_category (&info[context - 1])))) |
553 | { |
554 | context--; |
555 | w_total += pos[context].x_advance; |
556 | } |
557 | i++; // Don't touch i again. |
558 | |
559 | DEBUG_MSG (ARABIC, nullptr, "%s stretch at (%u,%u,%u)" , |
560 | step == MEASURE ? "measuring" : "cutting" , context, start, end); |
561 | DEBUG_MSG (ARABIC, nullptr, "rest of word: count=%u width %d" , start - context, w_total); |
562 | DEBUG_MSG (ARABIC, nullptr, "fixed tiles: count=%d width=%d" , n_fixed, w_fixed); |
563 | DEBUG_MSG (ARABIC, nullptr, "repeating tiles: count=%d width=%d" , n_repeating, w_repeating); |
564 | |
565 | /* Number of additional times to repeat each repeating tile. */ |
566 | int n_copies = 0; |
567 | |
568 | hb_position_t w_remaining = w_total - w_fixed; |
569 | if (sign * w_remaining > sign * w_repeating && sign * w_repeating > 0) |
570 | n_copies = (sign * w_remaining) / (sign * w_repeating) - 1; |
571 | |
572 | /* See if we can improve the fit by adding an extra repeat and squeezing them together a bit. */ |
573 | hb_position_t = 0; |
574 | hb_position_t shortfall = sign * w_remaining - sign * w_repeating * (n_copies + 1); |
575 | if (shortfall > 0 && n_repeating > 0) |
576 | { |
577 | ++n_copies; |
578 | hb_position_t excess = (n_copies + 1) * sign * w_repeating - sign * w_remaining; |
579 | if (excess > 0) |
580 | extra_repeat_overlap = excess / (n_copies * n_repeating); |
581 | } |
582 | |
583 | if (step == MEASURE) |
584 | { |
585 | extra_glyphs_needed += n_copies * n_repeating; |
586 | DEBUG_MSG (ARABIC, nullptr, "will add extra %d copies of repeating tiles" , n_copies); |
587 | } |
588 | else |
589 | { |
590 | buffer->unsafe_to_break (context, end); |
591 | hb_position_t x_offset = 0; |
592 | for (unsigned int k = end; k > start; k--) |
593 | { |
594 | hb_position_t width = font->get_glyph_h_advance (info[k - 1].codepoint); |
595 | |
596 | unsigned int repeat = 1; |
597 | if (info[k - 1].arabic_shaping_action() == STCH_REPEATING) |
598 | repeat += n_copies; |
599 | |
600 | DEBUG_MSG (ARABIC, nullptr, "appending %u copies of glyph %u; j=%u" , |
601 | repeat, info[k - 1].codepoint, j); |
602 | for (unsigned int n = 0; n < repeat; n++) |
603 | { |
604 | x_offset -= width; |
605 | if (n > 0) |
606 | x_offset += extra_repeat_overlap; |
607 | pos[k - 1].x_offset = x_offset; |
608 | /* Append copy. */ |
609 | --j; |
610 | info[j] = info[k - 1]; |
611 | pos[j] = pos[k - 1]; |
612 | } |
613 | } |
614 | } |
615 | } |
616 | |
617 | if (step == MEASURE) |
618 | { |
619 | if (unlikely (!buffer->ensure (count + extra_glyphs_needed))) |
620 | break; |
621 | } |
622 | else |
623 | { |
624 | assert (j == 0); |
625 | buffer->len = new_len; |
626 | } |
627 | } |
628 | } |
629 | |
630 | |
631 | static void |
632 | postprocess_glyphs_arabic (const hb_ot_shape_plan_t *plan, |
633 | hb_buffer_t *buffer, |
634 | hb_font_t *font) |
635 | { |
636 | apply_stch (plan, buffer, font); |
637 | } |
638 | |
639 | /* https://www.unicode.org/reports/tr53/ */ |
640 | |
641 | static hb_codepoint_t |
642 | modifier_combining_marks[] = |
643 | { |
644 | 0x0654u, /* ARABIC HAMZA ABOVE */ |
645 | 0x0655u, /* ARABIC HAMZA BELOW */ |
646 | 0x0658u, /* ARABIC MARK NOON GHUNNA */ |
647 | 0x06DCu, /* ARABIC SMALL HIGH SEEN */ |
648 | 0x06E3u, /* ARABIC SMALL LOW SEEN */ |
649 | 0x06E7u, /* ARABIC SMALL HIGH YEH */ |
650 | 0x06E8u, /* ARABIC SMALL HIGH NOON */ |
651 | 0x08CAu, /* ARABIC SMALL HIGH FARSI YEH */ |
652 | 0x08CBu, /* ARABIC SMALL HIGH YEH BARREE WITH TWO DOTS BELOW */ |
653 | 0x08CDu, /* ARABIC SMALL HIGH ZAH */ |
654 | 0x08CEu, /* ARABIC LARGE ROUND DOT ABOVE */ |
655 | 0x08CFu, /* ARABIC LARGE ROUND DOT BELOW */ |
656 | 0x08D3u, /* ARABIC SMALL LOW WAW */ |
657 | 0x08F3u, /* ARABIC SMALL HIGH WAW */ |
658 | }; |
659 | |
660 | static inline bool |
661 | info_is_mcm (const hb_glyph_info_t &info) |
662 | { |
663 | hb_codepoint_t u = info.codepoint; |
664 | for (unsigned int i = 0; i < ARRAY_LENGTH (modifier_combining_marks); i++) |
665 | if (u == modifier_combining_marks[i]) |
666 | return true; |
667 | return false; |
668 | } |
669 | |
670 | static void |
671 | reorder_marks_arabic (const hb_ot_shape_plan_t *plan HB_UNUSED, |
672 | hb_buffer_t *buffer, |
673 | unsigned int start, |
674 | unsigned int end) |
675 | { |
676 | hb_glyph_info_t *info = buffer->info; |
677 | |
678 | DEBUG_MSG (ARABIC, buffer, "Reordering marks from %u to %u" , start, end); |
679 | |
680 | unsigned int i = start; |
681 | for (unsigned int cc = 220; cc <= 230; cc += 10) |
682 | { |
683 | DEBUG_MSG (ARABIC, buffer, "Looking for %u's starting at %u" , cc, i); |
684 | while (i < end && info_cc(info[i]) < cc) |
685 | i++; |
686 | DEBUG_MSG (ARABIC, buffer, "Looking for %u's stopped at %u" , cc, i); |
687 | |
688 | if (i == end) |
689 | break; |
690 | |
691 | if (info_cc(info[i]) > cc) |
692 | continue; |
693 | |
694 | unsigned int j = i; |
695 | while (j < end && info_cc(info[j]) == cc && info_is_mcm (info[j])) |
696 | j++; |
697 | |
698 | if (i == j) |
699 | continue; |
700 | |
701 | DEBUG_MSG (ARABIC, buffer, "Found %u's from %u to %u" , cc, i, j); |
702 | |
703 | /* Shift it! */ |
704 | DEBUG_MSG (ARABIC, buffer, "Shifting %u's: %u %u" , cc, i, j); |
705 | hb_glyph_info_t temp[HB_OT_SHAPE_MAX_COMBINING_MARKS]; |
706 | assert (j - i <= ARRAY_LENGTH (temp)); |
707 | buffer->merge_clusters (start, j); |
708 | memmove (temp, &info[i], (j - i) * sizeof (hb_glyph_info_t)); |
709 | memmove (&info[start + j - i], &info[start], (i - start) * sizeof (hb_glyph_info_t)); |
710 | memmove (&info[start], temp, (j - i) * sizeof (hb_glyph_info_t)); |
711 | |
712 | /* Renumber CC such that the reordered sequence is still sorted. |
713 | * 22 and 26 are chosen because they are smaller than all Arabic categories, |
714 | * and are folded back to 220/230 respectively during fallback mark positioning. |
715 | * |
716 | * We do this because the CGJ-handling logic in the normalizer relies on |
717 | * mark sequences having an increasing order even after this reordering. |
718 | * https://github.com/harfbuzz/harfbuzz/issues/554 |
719 | * This, however, does break some obscure sequences, where the normalizer |
720 | * might compose a sequence that it should not. For example, in the seequence |
721 | * ALEF, HAMZAH, MADDAH, we should NOT try to compose ALEF+MADDAH, but with this |
722 | * renumbering, we will. |
723 | */ |
724 | unsigned int new_start = start + j - i; |
725 | unsigned int new_cc = cc == 220 ? HB_MODIFIED_COMBINING_CLASS_CCC22 : HB_MODIFIED_COMBINING_CLASS_CCC26; |
726 | while (start < new_start) |
727 | { |
728 | _hb_glyph_info_set_modified_combining_class (&info[start], new_cc); |
729 | start++; |
730 | } |
731 | |
732 | i = j; |
733 | } |
734 | } |
735 | |
736 | const hb_ot_shaper_t _hb_ot_shaper_arabic = |
737 | { |
738 | collect_features_arabic, |
739 | nullptr, /* override_features */ |
740 | data_create_arabic, |
741 | data_destroy_arabic, |
742 | nullptr, /* preprocess_text */ |
743 | postprocess_glyphs_arabic, |
744 | nullptr, /* decompose */ |
745 | nullptr, /* compose */ |
746 | setup_masks_arabic, |
747 | reorder_marks_arabic, |
748 | HB_TAG_NONE, /* gpos_tag */ |
749 | HB_OT_SHAPE_NORMALIZATION_MODE_DEFAULT, |
750 | HB_OT_SHAPE_ZERO_WIDTH_MARKS_BY_GDEF_LATE, |
751 | true, /* fallback_position */ |
752 | }; |
753 | |
754 | |
755 | #endif |
756 | |