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