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#include "hb-ot-shape-complex-arabic.hh"
29#include "hb-ot-shape.hh"
30
31
32/* buffer var allocations */
33#define arabic_shaping_action() complex_var_u8_0() /* arabic shaping action */
34
35#define HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH HB_BUFFER_SCRATCH_FLAG_COMPLEX0
36
37/* See:
38 * https://github.com/harfbuzz/harfbuzz/commit/6e6f82b6f3dde0fc6c3c7d991d9ec6cfff57823d#commitcomment-14248516 */
39#define HB_ARABIC_GENERAL_CATEGORY_IS_WORD(gen_cat) \
40 (FLAG_UNSAFE (gen_cat) & \
41 (FLAG (HB_UNICODE_GENERAL_CATEGORY_UNASSIGNED) | \
42 FLAG (HB_UNICODE_GENERAL_CATEGORY_PRIVATE_USE) | \
43 /*FLAG (HB_UNICODE_GENERAL_CATEGORY_LOWERCASE_LETTER) |*/ \
44 FLAG (HB_UNICODE_GENERAL_CATEGORY_MODIFIER_LETTER) | \
45 FLAG (HB_UNICODE_GENERAL_CATEGORY_OTHER_LETTER) | \
46 /*FLAG (HB_UNICODE_GENERAL_CATEGORY_TITLECASE_LETTER) |*/ \
47 /*FLAG (HB_UNICODE_GENERAL_CATEGORY_UPPERCASE_LETTER) |*/ \
48 FLAG (HB_UNICODE_GENERAL_CATEGORY_SPACING_MARK) | \
49 FLAG (HB_UNICODE_GENERAL_CATEGORY_ENCLOSING_MARK) | \
50 FLAG (HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK) | \
51 FLAG (HB_UNICODE_GENERAL_CATEGORY_DECIMAL_NUMBER) | \
52 FLAG (HB_UNICODE_GENERAL_CATEGORY_LETTER_NUMBER) | \
53 FLAG (HB_UNICODE_GENERAL_CATEGORY_OTHER_NUMBER) | \
54 FLAG (HB_UNICODE_GENERAL_CATEGORY_CURRENCY_SYMBOL) | \
55 FLAG (HB_UNICODE_GENERAL_CATEGORY_MODIFIER_SYMBOL) | \
56 FLAG (HB_UNICODE_GENERAL_CATEGORY_MATH_SYMBOL) | \
57 FLAG (HB_UNICODE_GENERAL_CATEGORY_OTHER_SYMBOL)))
58
59
60/*
61 * Joining types:
62 */
63
64/*
65 * Bits used in the joining tables
66 */
67enum hb_arabic_joining_type_t {
68 JOINING_TYPE_U = 0,
69 JOINING_TYPE_L = 1,
70 JOINING_TYPE_R = 2,
71 JOINING_TYPE_D = 3,
72 JOINING_TYPE_C = JOINING_TYPE_D,
73 JOINING_GROUP_ALAPH = 4,
74 JOINING_GROUP_DALATH_RISH = 5,
75 NUM_STATE_MACHINE_COLS = 6,
76
77 JOINING_TYPE_T = 7,
78 JOINING_TYPE_X = 8 /* means: use general-category to choose between U or T. */
79};
80
81#include "hb-ot-shape-complex-arabic-table.hh"
82
83static unsigned int get_joining_type (hb_codepoint_t u, hb_unicode_general_category_t gen_cat)
84{
85 unsigned int j_type = joining_type(u);
86 if (likely (j_type != JOINING_TYPE_X))
87 return j_type;
88
89 return (FLAG_UNSAFE(gen_cat) &
90 (FLAG(HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK) |
91 FLAG(HB_UNICODE_GENERAL_CATEGORY_ENCLOSING_MARK) |
92 FLAG(HB_UNICODE_GENERAL_CATEGORY_FORMAT))
93 ) ? JOINING_TYPE_T : JOINING_TYPE_U;
94}
95
96#define FEATURE_IS_SYRIAC(tag) hb_in_range<unsigned char> ((unsigned char) (tag), '2', '3')
97
98static const hb_tag_t arabic_features[] =
99{
100 HB_TAG('i','s','o','l'),
101 HB_TAG('f','i','n','a'),
102 HB_TAG('f','i','n','2'),
103 HB_TAG('f','i','n','3'),
104 HB_TAG('m','e','d','i'),
105 HB_TAG('m','e','d','2'),
106 HB_TAG('i','n','i','t'),
107 HB_TAG_NONE
108};
109
110
111/* Same order as the feature array */
112enum arabic_action_t {
113 ISOL,
114 FINA,
115 FIN2,
116 FIN3,
117 MEDI,
118 MED2,
119 INIT,
120
121 NONE,
122
123 ARABIC_NUM_FEATURES = NONE,
124
125 /* We abuse the same byte for other things... */
126 STCH_FIXED,
127 STCH_REPEATING,
128};
129
130static const struct arabic_state_table_entry {
131 uint8_t prev_action;
132 uint8_t curr_action;
133 uint16_t next_state;
134} arabic_state_table[][NUM_STATE_MACHINE_COLS] =
135{
136 /* jt_U, jt_L, jt_R, jt_D, jg_ALAPH, jg_DALATH_RISH */
137
138 /* State 0: prev was U, not willing to join. */
139 { {NONE,NONE,0}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,6}, },
140
141 /* State 1: prev was R or ISOL/ALAPH, not willing to join. */
142 { {NONE,NONE,0}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,FIN2,5}, {NONE,ISOL,6}, },
143
144 /* State 2: prev was D/L in ISOL form, willing to join. */
145 { {NONE,NONE,0}, {NONE,ISOL,2}, {INIT,FINA,1}, {INIT,FINA,3}, {INIT,FINA,4}, {INIT,FINA,6}, },
146
147 /* State 3: prev was D in FINA form, willing to join. */
148 { {NONE,NONE,0}, {NONE,ISOL,2}, {MEDI,FINA,1}, {MEDI,FINA,3}, {MEDI,FINA,4}, {MEDI,FINA,6}, },
149
150 /* State 4: prev was FINA ALAPH, not willing to join. */
151 { {NONE,NONE,0}, {NONE,ISOL,2}, {MED2,ISOL,1}, {MED2,ISOL,2}, {MED2,FIN2,5}, {MED2,ISOL,6}, },
152
153 /* State 5: prev was FIN2/FIN3 ALAPH, not willing to join. */
154 { {NONE,NONE,0}, {NONE,ISOL,2}, {ISOL,ISOL,1}, {ISOL,ISOL,2}, {ISOL,FIN2,5}, {ISOL,ISOL,6}, },
155
156 /* State 6: prev was DALATH/RISH, not willing to join. */
157 { {NONE,NONE,0}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,FIN3,5}, {NONE,ISOL,6}, }
158};
159
160
161static void
162arabic_fallback_shape (const hb_ot_shape_plan_t *plan,
163 hb_font_t *font,
164 hb_buffer_t *buffer);
165
166static void
167record_stch (const hb_ot_shape_plan_t *plan,
168 hb_font_t *font,
169 hb_buffer_t *buffer);
170
171static void
172collect_features_arabic (hb_ot_shape_planner_t *plan)
173{
174 hb_ot_map_builder_t *map = &plan->map;
175
176 /* We apply features according to the Arabic spec, with pauses
177 * in between most.
178 *
179 * The pause between init/medi/... and rlig is required. See eg:
180 * https://bugzilla.mozilla.org/show_bug.cgi?id=644184
181 *
182 * The pauses between init/medi/... themselves are not necessarily
183 * needed as only one of those features is applied to any character.
184 * The only difference it makes is when fonts have contextual
185 * substitutions. We now follow the order of the spec, which makes
186 * for better experience if that's what Uniscribe is doing.
187 *
188 * At least for Arabic, looks like Uniscribe has a pause between
189 * rlig and calt. Otherwise the IranNastaliq's ALLAH ligature won't
190 * work. However, testing shows that rlig and calt are applied
191 * together for Mongolian in Uniscribe. As such, we only add a
192 * pause for Arabic, not other scripts.
193 *
194 * A pause after calt is required to make KFGQPC Uthmanic Script HAFS
195 * work correctly. See https://github.com/harfbuzz/harfbuzz/issues/505
196 */
197
198
199 map->enable_feature (HB_TAG('s','t','c','h'));
200 map->add_gsub_pause (record_stch);
201
202 map->enable_feature (HB_TAG('c','c','m','p'));
203 map->enable_feature (HB_TAG('l','o','c','l'));
204
205 map->add_gsub_pause (nullptr);
206
207 for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++)
208 {
209 bool has_fallback = plan->props.script == HB_SCRIPT_ARABIC && !FEATURE_IS_SYRIAC (arabic_features[i]);
210 map->add_feature (arabic_features[i], has_fallback ? F_HAS_FALLBACK : F_NONE);
211 map->add_gsub_pause (nullptr);
212 }
213
214 /* Normally, Unicode says a ZWNJ means "don't ligate". In Arabic script
215 * however, it says a ZWJ should also mean "don't ligate". So we run
216 * the main ligating features as MANUAL_ZWJ. */
217
218 map->enable_feature (HB_TAG('r','l','i','g'), F_MANUAL_ZWJ | F_HAS_FALLBACK);
219
220 if (plan->props.script == HB_SCRIPT_ARABIC)
221 map->add_gsub_pause (arabic_fallback_shape);
222
223 /* No pause after rclt. See 98460779bae19e4d64d29461ff154b3527bf8420. */
224 map->enable_feature (HB_TAG('r','c','l','t'), F_MANUAL_ZWJ);
225 map->enable_feature (HB_TAG('c','a','l','t'), F_MANUAL_ZWJ);
226 map->add_gsub_pause (nullptr);
227
228 /* And undo here. */
229
230 /* The spec includes 'cswh'. Earlier versions of Windows
231 * used to enable this by default, but testing suggests
232 * that Windows 8 and later do not enable it by default,
233 * and spec now says 'Off by default'.
234 * We disabled this in ae23c24c32.
235 * Note that IranNastaliq uses this feature extensively
236 * to fixup broken glyph sequences. Oh well...
237 * Test case: U+0643,U+0640,U+0631. */
238 //map->enable_feature (HB_TAG('c','s','w','h'));
239 map->enable_feature (HB_TAG('m','s','e','t'));
240}
241
242#include "hb-ot-shape-complex-arabic-fallback.hh"
243
244struct arabic_shape_plan_t
245{
246 /* The "+ 1" in the next array is to accommodate for the "NONE" command,
247 * which is not an OpenType feature, but this simplifies the code by not
248 * having to do a "if (... < NONE) ..." and just rely on the fact that
249 * mask_array[NONE] == 0. */
250 hb_mask_t mask_array[ARABIC_NUM_FEATURES + 1];
251
252 hb_atomic_ptr_t<arabic_fallback_plan_t> fallback_plan;
253
254 unsigned int do_fallback : 1;
255 unsigned int has_stch : 1;
256};
257
258void *
259data_create_arabic (const hb_ot_shape_plan_t *plan)
260{
261 arabic_shape_plan_t *arabic_plan = (arabic_shape_plan_t *) calloc (1, sizeof (arabic_shape_plan_t));
262 if (unlikely (!arabic_plan))
263 return nullptr;
264
265 arabic_plan->do_fallback = plan->props.script == HB_SCRIPT_ARABIC;
266 arabic_plan->has_stch = !!plan->map.get_1_mask (HB_TAG ('s','t','c','h'));
267 for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++) {
268 arabic_plan->mask_array[i] = plan->map.get_1_mask (arabic_features[i]);
269 arabic_plan->do_fallback = arabic_plan->do_fallback &&
270 (FEATURE_IS_SYRIAC (arabic_features[i]) ||
271 plan->map.needs_fallback (arabic_features[i]));
272 }
273
274 return arabic_plan;
275}
276
277void
278data_destroy_arabic (void *data)
279{
280 arabic_shape_plan_t *arabic_plan = (arabic_shape_plan_t *) data;
281
282 arabic_fallback_plan_destroy (arabic_plan->fallback_plan);
283
284 free (data);
285}
286
287static void
288arabic_joining (hb_buffer_t *buffer)
289{
290 unsigned int count = buffer->len;
291 hb_glyph_info_t *info = buffer->info;
292 unsigned int prev = (unsigned int) -1, state = 0;
293
294 /* Check pre-context */
295 for (unsigned int i = 0; i < buffer->context_len[0]; i++)
296 {
297 unsigned int this_type = get_joining_type (buffer->context[0][i], buffer->unicode->general_category (buffer->context[0][i]));
298
299 if (unlikely (this_type == JOINING_TYPE_T))
300 continue;
301
302 const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
303 state = entry->next_state;
304 break;
305 }
306
307 for (unsigned int i = 0; i < count; i++)
308 {
309 unsigned int this_type = get_joining_type (info[i].codepoint, _hb_glyph_info_get_general_category (&info[i]));
310
311 if (unlikely (this_type == JOINING_TYPE_T)) {
312 info[i].arabic_shaping_action() = NONE;
313 continue;
314 }
315
316 const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
317
318 if (entry->prev_action != NONE && prev != (unsigned int) -1)
319 {
320 info[prev].arabic_shaping_action() = entry->prev_action;
321 buffer->unsafe_to_break (prev, i + 1);
322 }
323
324 info[i].arabic_shaping_action() = entry->curr_action;
325
326 prev = i;
327 state = entry->next_state;
328 }
329
330 for (unsigned int i = 0; i < buffer->context_len[1]; i++)
331 {
332 unsigned int this_type = get_joining_type (buffer->context[1][i], buffer->unicode->general_category (buffer->context[1][i]));
333
334 if (unlikely (this_type == JOINING_TYPE_T))
335 continue;
336
337 const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
338 if (entry->prev_action != NONE && prev != (unsigned int) -1)
339 info[prev].arabic_shaping_action() = entry->prev_action;
340 break;
341 }
342}
343
344static void
345mongolian_variation_selectors (hb_buffer_t *buffer)
346{
347 /* Copy arabic_shaping_action() from base to Mongolian variation selectors. */
348 unsigned int count = buffer->len;
349 hb_glyph_info_t *info = buffer->info;
350 for (unsigned int i = 1; i < count; i++)
351 if (unlikely (hb_in_range<hb_codepoint_t> (info[i].codepoint, 0x180Bu, 0x180Du)))
352 info[i].arabic_shaping_action() = info[i - 1].arabic_shaping_action();
353}
354
355void
356setup_masks_arabic_plan (const arabic_shape_plan_t *arabic_plan,
357 hb_buffer_t *buffer,
358 hb_script_t script)
359{
360 HB_BUFFER_ALLOCATE_VAR (buffer, arabic_shaping_action);
361
362 arabic_joining (buffer);
363 if (script == HB_SCRIPT_MONGOLIAN)
364 mongolian_variation_selectors (buffer);
365
366 unsigned int count = buffer->len;
367 hb_glyph_info_t *info = buffer->info;
368 for (unsigned int i = 0; i < count; i++)
369 info[i].mask |= arabic_plan->mask_array[info[i].arabic_shaping_action()];
370}
371
372static void
373setup_masks_arabic (const hb_ot_shape_plan_t *plan,
374 hb_buffer_t *buffer,
375 hb_font_t *font HB_UNUSED)
376{
377 const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
378 setup_masks_arabic_plan (arabic_plan, buffer, plan->props.script);
379}
380
381static void
382arabic_fallback_shape (const hb_ot_shape_plan_t *plan,
383 hb_font_t *font,
384 hb_buffer_t *buffer)
385{
386 const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
387
388 if (!arabic_plan->do_fallback)
389 return;
390
391retry:
392 arabic_fallback_plan_t *fallback_plan = arabic_plan->fallback_plan;
393 if (unlikely (!fallback_plan))
394 {
395 /* This sucks. We need a font to build the fallback plan... */
396 fallback_plan = arabic_fallback_plan_create (plan, font);
397 if (unlikely (!arabic_plan->fallback_plan.cmpexch (nullptr, fallback_plan)))
398 {
399 arabic_fallback_plan_destroy (fallback_plan);
400 goto retry;
401 }
402 }
403
404 arabic_fallback_plan_shape (fallback_plan, font, buffer);
405}
406
407/*
408 * Stretch feature: "stch".
409 * See example here:
410 * https://docs.microsoft.com/en-us/typography/script-development/syriac
411 * We implement this in a generic way, such that the Arabic subtending
412 * marks can use it as well.
413 */
414
415static void
416record_stch (const hb_ot_shape_plan_t *plan,
417 hb_font_t *font HB_UNUSED,
418 hb_buffer_t *buffer)
419{
420 const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
421 if (!arabic_plan->has_stch)
422 return;
423
424 /* 'stch' feature was just applied. Look for anything that multiplied,
425 * and record it for stch treatment later. Note that rtlm, frac, etc
426 * are applied before stch, but we assume that they didn't result in
427 * anything multiplying into 5 pieces, so it's safe-ish... */
428
429 unsigned int count = buffer->len;
430 hb_glyph_info_t *info = buffer->info;
431 for (unsigned int i = 0; i < count; i++)
432 if (unlikely (_hb_glyph_info_multiplied (&info[i])))
433 {
434 unsigned int comp = _hb_glyph_info_get_lig_comp (&info[i]);
435 info[i].arabic_shaping_action() = comp % 2 ? STCH_REPEATING : STCH_FIXED;
436 buffer->scratch_flags |= HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH;
437 }
438}
439
440static void
441apply_stch (const hb_ot_shape_plan_t *plan HB_UNUSED,
442 hb_buffer_t *buffer,
443 hb_font_t *font)
444{
445 if (likely (!(buffer->scratch_flags & HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH)))
446 return;
447
448 /* The Arabic shaper currently always processes in RTL mode, so we should
449 * stretch / position the stretched pieces to the left / preceding glyphs. */
450
451 /* We do a two pass implementation:
452 * First pass calculates the exact number of extra glyphs we need,
453 * We then enlarge buffer to have that much room,
454 * Second pass applies the stretch, copying things to the end of buffer.
455 */
456
457 int sign = font->x_scale < 0 ? -1 : +1;
458 unsigned int extra_glyphs_needed = 0; // Set during MEASURE, used during CUT
459 enum { MEASURE, CUT } /* step_t */;
460
461 for (unsigned int step = MEASURE; step <= CUT; step = step + 1)
462 {
463 unsigned int count = buffer->len;
464 hb_glyph_info_t *info = buffer->info;
465 hb_glyph_position_t *pos = buffer->pos;
466 unsigned int new_len = count + extra_glyphs_needed; // write head during CUT
467 unsigned int j = new_len;
468 for (unsigned int i = count; i; i--)
469 {
470 if (!hb_in_range<unsigned> (info[i - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING))
471 {
472 if (step == CUT)
473 {
474 --j;
475 info[j] = info[i - 1];
476 pos[j] = pos[i - 1];
477 }
478 continue;
479 }
480
481 /* Yay, justification! */
482
483 hb_position_t w_total = 0; // Total to be filled
484 hb_position_t w_fixed = 0; // Sum of fixed tiles
485 hb_position_t w_repeating = 0; // Sum of repeating tiles
486 int n_fixed = 0;
487 int n_repeating = 0;
488
489 unsigned int end = i;
490 while (i &&
491 hb_in_range<unsigned> (info[i - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING))
492 {
493 i--;
494 hb_position_t width = font->get_glyph_h_advance (info[i].codepoint);
495 if (info[i].arabic_shaping_action() == STCH_FIXED)
496 {
497 w_fixed += width;
498 n_fixed++;
499 }
500 else
501 {
502 w_repeating += width;
503 n_repeating++;
504 }
505 }
506 unsigned int start = i;
507 unsigned int context = i;
508 while (context &&
509 !hb_in_range<unsigned> (info[context - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING) &&
510 (_hb_glyph_info_is_default_ignorable (&info[context - 1]) ||
511 HB_ARABIC_GENERAL_CATEGORY_IS_WORD (_hb_glyph_info_get_general_category (&info[context - 1]))))
512 {
513 context--;
514 w_total += pos[context].x_advance;
515 }
516 i++; // Don't touch i again.
517
518 DEBUG_MSG (ARABIC, nullptr, "%s stretch at (%d,%d,%d)",
519 step == MEASURE ? "measuring" : "cutting", context, start, end);
520 DEBUG_MSG (ARABIC, nullptr, "rest of word: count=%d width %d", start - context, w_total);
521 DEBUG_MSG (ARABIC, nullptr, "fixed tiles: count=%d width=%d", n_fixed, w_fixed);
522 DEBUG_MSG (ARABIC, nullptr, "repeating tiles: count=%d width=%d", n_repeating, w_repeating);
523
524 /* Number of additional times to repeat each repeating tile. */
525 int n_copies = 0;
526
527 hb_position_t w_remaining = w_total - w_fixed;
528 if (sign * w_remaining > sign * w_repeating && sign * w_repeating > 0)
529 n_copies = (sign * w_remaining) / (sign * w_repeating) - 1;
530
531 /* See if we can improve the fit by adding an extra repeat and squeezing them together a bit. */
532 hb_position_t extra_repeat_overlap = 0;
533 hb_position_t shortfall = sign * w_remaining - sign * w_repeating * (n_copies + 1);
534 if (shortfall > 0 && n_repeating > 0)
535 {
536 ++n_copies;
537 hb_position_t excess = (n_copies + 1) * sign * w_repeating - sign * w_remaining;
538 if (excess > 0)
539 extra_repeat_overlap = excess / (n_copies * n_repeating);
540 }
541
542 if (step == MEASURE)
543 {
544 extra_glyphs_needed += n_copies * n_repeating;
545 DEBUG_MSG (ARABIC, nullptr, "will add extra %d copies of repeating tiles", n_copies);
546 }
547 else
548 {
549 buffer->unsafe_to_break (context, end);
550 hb_position_t x_offset = 0;
551 for (unsigned int k = end; k > start; k--)
552 {
553 hb_position_t width = font->get_glyph_h_advance (info[k - 1].codepoint);
554
555 unsigned int repeat = 1;
556 if (info[k - 1].arabic_shaping_action() == STCH_REPEATING)
557 repeat += n_copies;
558
559 DEBUG_MSG (ARABIC, nullptr, "appending %d copies of glyph %d; j=%d",
560 repeat, info[k - 1].codepoint, j);
561 for (unsigned int n = 0; n < repeat; n++)
562 {
563 x_offset -= width;
564 if (n > 0)
565 x_offset += extra_repeat_overlap;
566 pos[k - 1].x_offset = x_offset;
567 /* Append copy. */
568 --j;
569 info[j] = info[k - 1];
570 pos[j] = pos[k - 1];
571 }
572 }
573 }
574 }
575
576 if (step == MEASURE)
577 {
578 if (unlikely (!buffer->ensure (count + extra_glyphs_needed)))
579 break;
580 }
581 else
582 {
583 assert (j == 0);
584 buffer->len = new_len;
585 }
586 }
587}
588
589
590static void
591postprocess_glyphs_arabic (const hb_ot_shape_plan_t *plan,
592 hb_buffer_t *buffer,
593 hb_font_t *font)
594{
595 apply_stch (plan, buffer, font);
596
597 HB_BUFFER_DEALLOCATE_VAR (buffer, arabic_shaping_action);
598}
599
600/* http://www.unicode.org/reports/tr53/ */
601
602static hb_codepoint_t
603modifier_combining_marks[] =
604{
605 0x0654u, /* ARABIC HAMZA ABOVE */
606 0x0655u, /* ARABIC HAMZA BELOW */
607 0x0658u, /* ARABIC MARK NOON GHUNNA */
608 0x06DCu, /* ARABIC SMALL HIGH SEEN */
609 0x06E3u, /* ARABIC SMALL LOW SEEN */
610 0x06E7u, /* ARABIC SMALL HIGH YEH */
611 0x06E8u, /* ARABIC SMALL HIGH NOON */
612 0x08D3u, /* ARABIC SMALL LOW WAW */
613 0x08F3u, /* ARABIC SMALL HIGH WAW */
614};
615
616static inline bool
617info_is_mcm (const hb_glyph_info_t &info)
618{
619 hb_codepoint_t u = info.codepoint;
620 for (unsigned int i = 0; i < ARRAY_LENGTH (modifier_combining_marks); i++)
621 if (u == modifier_combining_marks[i])
622 return true;
623 return false;
624}
625
626static void
627reorder_marks_arabic (const hb_ot_shape_plan_t *plan HB_UNUSED,
628 hb_buffer_t *buffer,
629 unsigned int start,
630 unsigned int end)
631{
632 hb_glyph_info_t *info = buffer->info;
633
634 DEBUG_MSG (ARABIC, buffer, "Reordering marks from %d to %d", start, end);
635
636 unsigned int i = start;
637 for (unsigned int cc = 220; cc <= 230; cc += 10)
638 {
639 DEBUG_MSG (ARABIC, buffer, "Looking for %d's starting at %d", cc, i);
640 while (i < end && info_cc(info[i]) < cc)
641 i++;
642 DEBUG_MSG (ARABIC, buffer, "Looking for %d's stopped at %d", cc, i);
643
644 if (i == end)
645 break;
646
647 if (info_cc(info[i]) > cc)
648 continue;
649
650 unsigned int j = i;
651 while (j < end && info_cc(info[j]) == cc && info_is_mcm (info[j]))
652 j++;
653
654 if (i == j)
655 continue;
656
657 DEBUG_MSG (ARABIC, buffer, "Found %d's from %d to %d", cc, i, j);
658
659 /* Shift it! */
660 DEBUG_MSG (ARABIC, buffer, "Shifting %d's: %d %d", cc, i, j);
661 hb_glyph_info_t temp[HB_OT_SHAPE_COMPLEX_MAX_COMBINING_MARKS];
662 assert (j - i <= ARRAY_LENGTH (temp));
663 buffer->merge_clusters (start, j);
664 memmove (temp, &info[i], (j - i) * sizeof (hb_glyph_info_t));
665 memmove (&info[start + j - i], &info[start], (i - start) * sizeof (hb_glyph_info_t));
666 memmove (&info[start], temp, (j - i) * sizeof (hb_glyph_info_t));
667
668 /* Renumber CC such that the reordered sequence is still sorted.
669 * 22 and 26 are chosen because they are smaller than all Arabic categories,
670 * and are folded back to 220/230 respectively during fallback mark positioning.
671 *
672 * We do this because the CGJ-handling logic in the normalizer relies on
673 * mark sequences having an increasing order even after this reordering.
674 * https://github.com/harfbuzz/harfbuzz/issues/554
675 * This, however, does break some obscure sequences, where the normalizer
676 * might compose a sequence that it should not. For example, in the seequence
677 * ALEF, HAMZAH, MADDAH, we should NOT try to compose ALEF+MADDAH, but with this
678 * renumbering, we will.
679 */
680 unsigned int new_start = start + j - i;
681 unsigned int new_cc = cc == 220 ? HB_MODIFIED_COMBINING_CLASS_CCC22 : HB_MODIFIED_COMBINING_CLASS_CCC26;
682 while (start < new_start)
683 {
684 _hb_glyph_info_set_modified_combining_class (&info[start], new_cc);
685 start++;
686 }
687
688 i = j;
689 }
690}
691
692const hb_ot_complex_shaper_t _hb_ot_complex_shaper_arabic =
693{
694 collect_features_arabic,
695 nullptr, /* override_features */
696 data_create_arabic,
697 data_destroy_arabic,
698 nullptr, /* preprocess_text */
699 postprocess_glyphs_arabic,
700 HB_OT_SHAPE_NORMALIZATION_MODE_DEFAULT,
701 nullptr, /* decompose */
702 nullptr, /* compose */
703 setup_masks_arabic,
704 HB_TAG_NONE, /* gpos_tag */
705 reorder_marks_arabic,
706 HB_OT_SHAPE_ZERO_WIDTH_MARKS_BY_GDEF_LATE,
707 true, /* fallback_position */
708};
709