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
162nuke_joiners (const hb_ot_shape_plan_t *plan,
163 hb_font_t *font,
164 hb_buffer_t *buffer);
165
166static void
167arabic_fallback_shape (const hb_ot_shape_plan_t *plan,
168 hb_font_t *font,
169 hb_buffer_t *buffer);
170
171static void
172record_stch (const hb_ot_shape_plan_t *plan,
173 hb_font_t *font,
174 hb_buffer_t *buffer);
175
176static void
177collect_features_arabic (hb_ot_shape_planner_t *plan)
178{
179 hb_ot_map_builder_t *map = &plan->map;
180
181 /* We apply features according to the Arabic spec, with pauses
182 * in between most.
183 *
184 * The pause between init/medi/... and rlig is required. See eg:
185 * https://bugzilla.mozilla.org/show_bug.cgi?id=644184
186 *
187 * The pauses between init/medi/... themselves are not necessarily
188 * needed as only one of those features is applied to any character.
189 * The only difference it makes is when fonts have contextual
190 * substitutions. We now follow the order of the spec, which makes
191 * for better experience if that's what Uniscribe is doing.
192 *
193 * At least for Arabic, looks like Uniscribe has a pause between
194 * rlig and calt. Otherwise the IranNastaliq's ALLAH ligature won't
195 * work. However, testing shows that rlig and calt are applied
196 * together for Mongolian in Uniscribe. As such, we only add a
197 * pause for Arabic, not other scripts.
198 *
199 * A pause after calt is required to make KFGQPC Uthmanic Script HAFS
200 * work correctly. See https://github.com/harfbuzz/harfbuzz/issues/505
201 */
202
203 map->add_gsub_pause (nuke_joiners);
204
205 map->add_global_bool_feature (HB_TAG('s','t','c','h'));
206 map->add_gsub_pause (record_stch);
207
208 map->add_global_bool_feature (HB_TAG('c','c','m','p'));
209 map->add_global_bool_feature (HB_TAG('l','o','c','l'));
210
211 map->add_gsub_pause (nullptr);
212
213 for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++)
214 {
215 bool has_fallback = plan->props.script == HB_SCRIPT_ARABIC && !FEATURE_IS_SYRIAC (arabic_features[i]);
216 map->add_feature (arabic_features[i], 1, has_fallback ? F_HAS_FALLBACK : F_NONE);
217 map->add_gsub_pause (nullptr);
218 }
219
220 map->add_feature (HB_TAG('r','l','i','g'), 1, F_GLOBAL|F_HAS_FALLBACK);
221 if (plan->props.script == HB_SCRIPT_ARABIC)
222 map->add_gsub_pause (arabic_fallback_shape);
223
224 /* No pause after rclt. See 98460779bae19e4d64d29461ff154b3527bf8420. */
225 map->add_global_bool_feature (HB_TAG('r','c','l','t'));
226 map->add_global_bool_feature (HB_TAG('c','a','l','t'));
227 map->add_gsub_pause (nullptr);
228
229 /* The spec includes 'cswh'. Earlier versions of Windows
230 * used to enable this by default, but testing suggests
231 * that Windows 8 and later do not enable it by default,
232 * and spec now says 'Off by default'.
233 * We disabled this in ae23c24c32.
234 * Note that IranNastaliq uses this feature extensively
235 * to fixup broken glyph sequences. Oh well...
236 * Test case: U+0643,U+0640,U+0631. */
237 //map->add_global_bool_feature (HB_TAG('c','s','w','h'));
238 map->add_global_bool_feature (HB_TAG('m','s','e','t'));
239}
240
241#include "hb-ot-shape-complex-arabic-fallback.hh"
242
243struct arabic_shape_plan_t
244{
245 ASSERT_POD ();
246
247 /* The "+ 1" in the next array is to accommodate for the "NONE" command,
248 * which is not an OpenType feature, but this simplifies the code by not
249 * having to do a "if (... < NONE) ..." and just rely on the fact that
250 * mask_array[NONE] == 0. */
251 hb_mask_t mask_array[ARABIC_NUM_FEATURES + 1];
252
253 hb_atomic_ptr_t<arabic_fallback_plan_t> fallback_plan;
254
255 unsigned int do_fallback : 1;
256 unsigned int has_stch : 1;
257};
258
259void *
260data_create_arabic (const hb_ot_shape_plan_t *plan)
261{
262 arabic_shape_plan_t *arabic_plan = (arabic_shape_plan_t *) calloc (1, sizeof (arabic_shape_plan_t));
263 if (unlikely (!arabic_plan))
264 return nullptr;
265
266 arabic_plan->do_fallback = plan->props.script == HB_SCRIPT_ARABIC;
267 arabic_plan->has_stch = !!plan->map.get_1_mask (HB_TAG ('s','t','c','h'));
268 for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++) {
269 arabic_plan->mask_array[i] = plan->map.get_1_mask (arabic_features[i]);
270 arabic_plan->do_fallback = arabic_plan->do_fallback &&
271 (FEATURE_IS_SYRIAC (arabic_features[i]) ||
272 plan->map.needs_fallback (arabic_features[i]));
273 }
274
275 return arabic_plan;
276}
277
278void
279data_destroy_arabic (void *data)
280{
281 arabic_shape_plan_t *arabic_plan = (arabic_shape_plan_t *) data;
282
283 arabic_fallback_plan_destroy (arabic_plan->fallback_plan.get ());
284
285 free (data);
286}
287
288static void
289arabic_joining (hb_buffer_t *buffer)
290{
291 unsigned int count = buffer->len;
292 hb_glyph_info_t *info = buffer->info;
293 unsigned int prev = (unsigned int) -1, state = 0;
294
295 /* Check pre-context */
296 for (unsigned int i = 0; i < buffer->context_len[0]; i++)
297 {
298 unsigned int this_type = get_joining_type (buffer->context[0][i], buffer->unicode->general_category (buffer->context[0][i]));
299
300 if (unlikely (this_type == JOINING_TYPE_T))
301 continue;
302
303 const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
304 state = entry->next_state;
305 break;
306 }
307
308 for (unsigned int i = 0; i < count; i++)
309 {
310 unsigned int this_type = get_joining_type (info[i].codepoint, _hb_glyph_info_get_general_category (&info[i]));
311
312 if (unlikely (this_type == JOINING_TYPE_T)) {
313 info[i].arabic_shaping_action() = NONE;
314 continue;
315 }
316
317 const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
318
319 if (entry->prev_action != NONE && prev != (unsigned int) -1)
320 {
321 info[prev].arabic_shaping_action() = entry->prev_action;
322 buffer->unsafe_to_break (prev, i + 1);
323 }
324
325 info[i].arabic_shaping_action() = entry->curr_action;
326
327 prev = i;
328 state = entry->next_state;
329 }
330
331 for (unsigned int i = 0; i < buffer->context_len[1]; i++)
332 {
333 unsigned int this_type = get_joining_type (buffer->context[1][i], buffer->unicode->general_category (buffer->context[1][i]));
334
335 if (unlikely (this_type == JOINING_TYPE_T))
336 continue;
337
338 const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
339 if (entry->prev_action != NONE && prev != (unsigned int) -1)
340 info[prev].arabic_shaping_action() = entry->prev_action;
341 break;
342 }
343}
344
345static void
346mongolian_variation_selectors (hb_buffer_t *buffer)
347{
348 /* Copy arabic_shaping_action() from base to Mongolian variation selectors. */
349 unsigned int count = buffer->len;
350 hb_glyph_info_t *info = buffer->info;
351 for (unsigned int i = 1; i < count; i++)
352 if (unlikely (hb_in_range<hb_codepoint_t> (info[i].codepoint, 0x180Bu, 0x180Du)))
353 info[i].arabic_shaping_action() = info[i - 1].arabic_shaping_action();
354}
355
356void
357setup_masks_arabic_plan (const arabic_shape_plan_t *arabic_plan,
358 hb_buffer_t *buffer,
359 hb_script_t script)
360{
361 HB_BUFFER_ALLOCATE_VAR (buffer, arabic_shaping_action);
362
363 arabic_joining (buffer);
364 if (script == HB_SCRIPT_MONGOLIAN)
365 mongolian_variation_selectors (buffer);
366
367 unsigned int count = buffer->len;
368 hb_glyph_info_t *info = buffer->info;
369 for (unsigned int i = 0; i < count; i++)
370 info[i].mask |= arabic_plan->mask_array[info[i].arabic_shaping_action()];
371}
372
373static void
374setup_masks_arabic (const hb_ot_shape_plan_t *plan,
375 hb_buffer_t *buffer,
376 hb_font_t *font HB_UNUSED)
377{
378 const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
379 setup_masks_arabic_plan (arabic_plan, buffer, plan->props.script);
380}
381
382
383static void
384nuke_joiners (const hb_ot_shape_plan_t *plan HB_UNUSED,
385 hb_font_t *font HB_UNUSED,
386 hb_buffer_t *buffer)
387{
388 unsigned int count = buffer->len;
389 hb_glyph_info_t *info = buffer->info;
390 for (unsigned int i = 0; i < count; i++)
391 if (_hb_glyph_info_is_zwj (&info[i]))
392 _hb_glyph_info_flip_joiners (&info[i]);
393}
394
395static void
396arabic_fallback_shape (const hb_ot_shape_plan_t *plan,
397 hb_font_t *font,
398 hb_buffer_t *buffer)
399{
400 const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
401
402 if (!arabic_plan->do_fallback)
403 return;
404
405retry:
406 arabic_fallback_plan_t *fallback_plan = arabic_plan->fallback_plan.get ();
407 if (unlikely (!fallback_plan))
408 {
409 /* This sucks. We need a font to build the fallback plan... */
410 fallback_plan = arabic_fallback_plan_create (plan, font);
411 if (unlikely (!arabic_plan->fallback_plan.cmpexch (nullptr, fallback_plan)))
412 {
413 arabic_fallback_plan_destroy (fallback_plan);
414 goto retry;
415 }
416 }
417
418 arabic_fallback_plan_shape (fallback_plan, font, buffer);
419}
420
421/*
422 * Stretch feature: "stch".
423 * See example here:
424 * https://docs.microsoft.com/en-us/typography/script-development/syriac
425 * We implement this in a generic way, such that the Arabic subtending
426 * marks can use it as well.
427 */
428
429static void
430record_stch (const hb_ot_shape_plan_t *plan,
431 hb_font_t *font,
432 hb_buffer_t *buffer)
433{
434 const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
435 if (!arabic_plan->has_stch)
436 return;
437
438 /* 'stch' feature was just applied. Look for anything that multiplied,
439 * and record it for stch treatment later. Note that rtlm, frac, etc
440 * are applied before stch, but we assume that they didn't result in
441 * anything multiplying into 5 pieces, so it's safe-ish... */
442
443 unsigned int count = buffer->len;
444 hb_glyph_info_t *info = buffer->info;
445 for (unsigned int i = 0; i < count; i++)
446 if (unlikely (_hb_glyph_info_multiplied (&info[i])))
447 {
448 unsigned int comp = _hb_glyph_info_get_lig_comp (&info[i]);
449 info[i].arabic_shaping_action() = comp % 2 ? STCH_REPEATING : STCH_FIXED;
450 buffer->scratch_flags |= HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH;
451 }
452}
453
454static void
455apply_stch (const hb_ot_shape_plan_t *plan,
456 hb_buffer_t *buffer,
457 hb_font_t *font)
458{
459 if (likely (!(buffer->scratch_flags & HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH)))
460 return;
461
462 /* The Arabic shaper currently always processes in RTL mode, so we should
463 * stretch / position the stretched pieces to the left / preceding glyphs. */
464
465 /* We do a two pass implementation:
466 * First pass calculates the exact number of extra glyphs we need,
467 * We then enlarge buffer to have that much room,
468 * Second pass applies the stretch, copying things to the end of buffer.
469 */
470
471 int sign = font->x_scale < 0 ? -1 : +1;
472 unsigned int extra_glyphs_needed = 0; // Set during MEASURE, used during CUT
473 typedef enum { MEASURE, CUT } step_t;
474
475 for (step_t step = MEASURE; step <= CUT; step = (step_t) (step + 1))
476 {
477 unsigned int count = buffer->len;
478 hb_glyph_info_t *info = buffer->info;
479 hb_glyph_position_t *pos = buffer->pos;
480 unsigned int new_len = count + extra_glyphs_needed; // write head during CUT
481 unsigned int j = new_len;
482 for (unsigned int i = count; i; i--)
483 {
484 if (!hb_in_range<unsigned> (info[i - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING))
485 {
486 if (step == CUT)
487 {
488 --j;
489 info[j] = info[i - 1];
490 pos[j] = pos[i - 1];
491 }
492 continue;
493 }
494
495 /* Yay, justification! */
496
497 hb_position_t w_total = 0; // Total to be filled
498 hb_position_t w_fixed = 0; // Sum of fixed tiles
499 hb_position_t w_repeating = 0; // Sum of repeating tiles
500 int n_fixed = 0;
501 int n_repeating = 0;
502
503 unsigned int end = i;
504 while (i &&
505 hb_in_range<unsigned> (info[i - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING))
506 {
507 i--;
508 hb_position_t width = font->get_glyph_h_advance (info[i].codepoint);
509 if (info[i].arabic_shaping_action() == STCH_FIXED)
510 {
511 w_fixed += width;
512 n_fixed++;
513 }
514 else
515 {
516 w_repeating += width;
517 n_repeating++;
518 }
519 }
520 unsigned int start = i;
521 unsigned int context = i;
522 while (context &&
523 !hb_in_range<unsigned> (info[context - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING) &&
524 (_hb_glyph_info_is_default_ignorable (&info[context - 1]) ||
525 HB_ARABIC_GENERAL_CATEGORY_IS_WORD (_hb_glyph_info_get_general_category (&info[context - 1]))))
526 {
527 context--;
528 w_total += pos[context].x_advance;
529 }
530 i++; // Don't touch i again.
531
532 DEBUG_MSG (ARABIC, nullptr, "%s stretch at (%d,%d,%d)",
533 step == MEASURE ? "measuring" : "cutting", context, start, end);
534 DEBUG_MSG (ARABIC, nullptr, "rest of word: count=%d width %d", start - context, w_total);
535 DEBUG_MSG (ARABIC, nullptr, "fixed tiles: count=%d width=%d", n_fixed, w_fixed);
536 DEBUG_MSG (ARABIC, nullptr, "repeating tiles: count=%d width=%d", n_repeating, w_repeating);
537
538 /* Number of additional times to repeat each repeating tile. */
539 int n_copies = 0;
540
541 hb_position_t w_remaining = w_total - w_fixed;
542 if (sign * w_remaining > sign * w_repeating && sign * w_repeating > 0)
543 n_copies = (sign * w_remaining) / (sign * w_repeating) - 1;
544
545 /* See if we can improve the fit by adding an extra repeat and squeezing them together a bit. */
546 hb_position_t extra_repeat_overlap = 0;
547 hb_position_t shortfall = sign * w_remaining - sign * w_repeating * (n_copies + 1);
548 if (shortfall > 0 && n_repeating > 0)
549 {
550 ++n_copies;
551 hb_position_t excess = (n_copies + 1) * sign * w_repeating - sign * w_remaining;
552 if (excess > 0)
553 extra_repeat_overlap = excess / (n_copies * n_repeating);
554 }
555
556 if (step == MEASURE)
557 {
558 extra_glyphs_needed += n_copies * n_repeating;
559 DEBUG_MSG (ARABIC, nullptr, "will add extra %d copies of repeating tiles", n_copies);
560 }
561 else
562 {
563 buffer->unsafe_to_break (context, end);
564 hb_position_t x_offset = 0;
565 for (unsigned int k = end; k > start; k--)
566 {
567 hb_position_t width = font->get_glyph_h_advance (info[k - 1].codepoint);
568
569 unsigned int repeat = 1;
570 if (info[k - 1].arabic_shaping_action() == STCH_REPEATING)
571 repeat += n_copies;
572
573 DEBUG_MSG (ARABIC, nullptr, "appending %d copies of glyph %d; j=%d",
574 repeat, info[k - 1].codepoint, j);
575 for (unsigned int n = 0; n < repeat; n++)
576 {
577 x_offset -= width;
578 if (n > 0)
579 x_offset += extra_repeat_overlap;
580 pos[k - 1].x_offset = x_offset;
581 /* Append copy. */
582 --j;
583 info[j] = info[k - 1];
584 pos[j] = pos[k - 1];
585 }
586 }
587 }
588 }
589
590 if (step == MEASURE)
591 {
592 if (unlikely (!buffer->ensure (count + extra_glyphs_needed)))
593 break;
594 }
595 else
596 {
597 assert (j == 0);
598 buffer->len = new_len;
599 }
600 }
601}
602
603
604static void
605postprocess_glyphs_arabic (const hb_ot_shape_plan_t *plan,
606 hb_buffer_t *buffer,
607 hb_font_t *font)
608{
609 apply_stch (plan, buffer, font);
610
611 HB_BUFFER_DEALLOCATE_VAR (buffer, arabic_shaping_action);
612}
613
614/* https://unicode.org/reports/tr53/tr53-1.pdf */
615
616static hb_codepoint_t
617modifier_combining_marks[] =
618{
619 0x0654u, /* ARABIC HAMZA ABOVE */
620 0x0655u, /* ARABIC HAMZA BELOW */
621 0x0658u, /* ARABIC MARK NOON GHUNNA */
622 0x06DCu, /* ARABIC SMALL HIGH SEEN */
623 0x06E3u, /* ARABIC SMALL LOW SEEN */
624 0x06E7u, /* ARABIC SMALL HIGH YEH */
625 0x06E8u, /* ARABIC SMALL HIGH NOON */
626 0x08F3u, /* ARABIC SMALL HIGH WAW */
627};
628
629static inline bool
630info_is_mcm (const hb_glyph_info_t &info)
631{
632 hb_codepoint_t u = info.codepoint;
633 for (unsigned int i = 0; i < ARRAY_LENGTH (modifier_combining_marks); i++)
634 if (u == modifier_combining_marks[i])
635 return true;
636 return false;
637}
638
639static void
640reorder_marks_arabic (const hb_ot_shape_plan_t *plan,
641 hb_buffer_t *buffer,
642 unsigned int start,
643 unsigned int end)
644{
645 hb_glyph_info_t *info = buffer->info;
646
647 DEBUG_MSG (ARABIC, buffer, "Reordering marks from %d to %d", start, end);
648
649 unsigned int i = start;
650 for (unsigned int cc = 220; cc <= 230; cc += 10)
651 {
652 DEBUG_MSG (ARABIC, buffer, "Looking for %d's starting at %d", cc, i);
653 while (i < end && info_cc(info[i]) < cc)
654 i++;
655 DEBUG_MSG (ARABIC, buffer, "Looking for %d's stopped at %d", cc, i);
656
657 if (i == end)
658 break;
659
660 if (info_cc(info[i]) > cc)
661 continue;
662
663 unsigned int j = i;
664 while (j < end && info_cc(info[j]) == cc && info_is_mcm (info[j]))
665 j++;
666
667 if (i == j)
668 continue;
669
670 DEBUG_MSG (ARABIC, buffer, "Found %d's from %d to %d", cc, i, j);
671
672 /* Shift it! */
673 DEBUG_MSG (ARABIC, buffer, "Shifting %d's: %d %d", cc, i, j);
674 hb_glyph_info_t temp[HB_OT_SHAPE_COMPLEX_MAX_COMBINING_MARKS];
675 assert (j - i <= ARRAY_LENGTH (temp));
676 buffer->merge_clusters (start, j);
677 memmove (temp, &info[i], (j - i) * sizeof (hb_glyph_info_t));
678 memmove (&info[start + j - i], &info[start], (i - start) * sizeof (hb_glyph_info_t));
679 memmove (&info[start], temp, (j - i) * sizeof (hb_glyph_info_t));
680
681 /* Renumber CC such that the reordered sequence is still sorted.
682 * 22 and 26 are chosen because they are smaller than all Arabic categories,
683 * and are folded back to 220/230 respectively during fallback mark positioning.
684 *
685 * We do this because the CGJ-handling logic in the normalizer relies on
686 * mark sequences having an increasing order even after this reordering.
687 * https://github.com/harfbuzz/harfbuzz/issues/554
688 * This, however, does break some obscure sequences, where the normalizer
689 * might compose a sequence that it should not. For example, in the seequence
690 * ALEF, HAMZAH, MADDAH, we should NOT try to compose ALEF+MADDAH, but with this
691 * renumbering, we will.
692 */
693 unsigned int new_start = start + j - i;
694 unsigned int new_cc = cc == 220 ? HB_MODIFIED_COMBINING_CLASS_CCC22 : HB_MODIFIED_COMBINING_CLASS_CCC26;
695 while (start < new_start)
696 {
697 _hb_glyph_info_set_modified_combining_class (&info[start], new_cc);
698 start++;
699 }
700
701 i = j;
702 }
703}
704
705const hb_ot_complex_shaper_t _hb_ot_complex_shaper_arabic =
706{
707 collect_features_arabic,
708 nullptr, /* override_features */
709 data_create_arabic,
710 data_destroy_arabic,
711 nullptr, /* preprocess_text */
712 postprocess_glyphs_arabic,
713 HB_OT_SHAPE_NORMALIZATION_MODE_DEFAULT,
714 nullptr, /* decompose */
715 nullptr, /* compose */
716 setup_masks_arabic,
717 nullptr, /* disable_otl */
718 reorder_marks_arabic,
719 HB_OT_SHAPE_ZERO_WIDTH_MARKS_BY_GDEF_LATE,
720 true, /* fallback_position */
721};
722