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 
35 /* buffer var allocations */
36 #define arabic_shaping_action() complex_var_u8_0() /* arabic shaping action */
37 
38 #define HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH HB_BUFFER_SCRATCH_FLAG_COMPLEX0
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-shape-complex-arabic-table.hh"
85 
get_joining_type(hb_codepoint_t u,hb_unicode_general_category_t gen_cat)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 void
165 arabic_fallback_shape (const hb_ot_shape_plan_t *plan,
166 		       hb_font_t *font,
167 		       hb_buffer_t *buffer);
168 
169 static void
170 record_stch (const hb_ot_shape_plan_t *plan,
171 	     hb_font_t *font,
172 	     hb_buffer_t *buffer);
173 
174 static void
collect_features_arabic(hb_ot_shape_planner_t * plan)175 collect_features_arabic (hb_ot_shape_planner_t *plan)
176 {
177   hb_ot_map_builder_t *map = &plan->map;
178 
179   /* We apply features according to the Arabic spec, with pauses
180    * in between most.
181    *
182    * The pause between init/medi/... and rlig is required.  See eg:
183    * https://bugzilla.mozilla.org/show_bug.cgi?id=644184
184    *
185    * The pauses between init/medi/... themselves are not necessarily
186    * needed as only one of those features is applied to any character.
187    * The only difference it makes is when fonts have contextual
188    * substitutions.  We now follow the order of the spec, which makes
189    * for better experience if that's what Uniscribe is doing.
190    *
191    * At least for Arabic, looks like Uniscribe has a pause between
192    * rlig and calt.  Otherwise the IranNastaliq's ALLAH ligature won't
193    * work.  However, testing shows that rlig and calt are applied
194    * together for Mongolian in Uniscribe.  As such, we only add a
195    * pause for Arabic, not other scripts.
196    *
197    * A pause after calt is required to make KFGQPC Uthmanic Script HAFS
198    * work correctly.  See https://github.com/harfbuzz/harfbuzz/issues/505
199    */
200 
201 
202   map->enable_feature (HB_TAG('s','t','c','h'));
203   map->add_gsub_pause (record_stch);
204 
205   map->enable_feature (HB_TAG('c','c','m','p'));
206   map->enable_feature (HB_TAG('l','o','c','l'));
207 
208   map->add_gsub_pause (nullptr);
209 
210   for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++)
211   {
212     bool has_fallback = plan->props.script == HB_SCRIPT_ARABIC && !FEATURE_IS_SYRIAC (arabic_features[i]);
213     map->add_feature (arabic_features[i], has_fallback ? F_HAS_FALLBACK : F_NONE);
214     map->add_gsub_pause (nullptr);
215   }
216 
217   /* Normally, Unicode says a ZWNJ means "don't ligate".  In Arabic script
218    * however, it says a ZWJ should also mean "don't ligate".  So we run
219    * the main ligating features as MANUAL_ZWJ. */
220 
221   map->enable_feature (HB_TAG('r','l','i','g'), F_MANUAL_ZWJ | F_HAS_FALLBACK);
222 
223   if (plan->props.script == HB_SCRIPT_ARABIC)
224     map->add_gsub_pause (arabic_fallback_shape);
225 
226   /* No pause after rclt.  See 98460779bae19e4d64d29461ff154b3527bf8420. */
227   map->enable_feature (HB_TAG('r','c','l','t'), F_MANUAL_ZWJ);
228   map->enable_feature (HB_TAG('c','a','l','t'), F_MANUAL_ZWJ);
229   map->add_gsub_pause (nullptr);
230 
231   /* And undo here. */
232 
233   /* The spec includes 'cswh'.  Earlier versions of Windows
234    * used to enable this by default, but testing suggests
235    * that Windows 8 and later do not enable it by default,
236    * and spec now says 'Off by default'.
237    * We disabled this in ae23c24c32.
238    * Note that IranNastaliq uses this feature extensively
239    * to fixup broken glyph sequences.  Oh well...
240    * Test case: U+0643,U+0640,U+0631. */
241   //map->enable_feature (HB_TAG('c','s','w','h'));
242   map->enable_feature (HB_TAG('m','s','e','t'));
243 }
244 
245 #include "hb-ot-shape-complex-arabic-fallback.hh"
246 
247 struct arabic_shape_plan_t
248 {
249   /* The "+ 1" in the next array is to accommodate for the "NONE" command,
250    * which is not an OpenType feature, but this simplifies the code by not
251    * having to do a "if (... < NONE) ..." and just rely on the fact that
252    * mask_array[NONE] == 0. */
253   hb_mask_t mask_array[ARABIC_NUM_FEATURES + 1];
254 
255   hb_atomic_ptr_t<arabic_fallback_plan_t> fallback_plan;
256 
257   unsigned int do_fallback : 1;
258   unsigned int has_stch : 1;
259 };
260 
261 void *
data_create_arabic(const hb_ot_shape_plan_t * plan)262 data_create_arabic (const hb_ot_shape_plan_t *plan)
263 {
264   arabic_shape_plan_t *arabic_plan = (arabic_shape_plan_t *) calloc (1, sizeof (arabic_shape_plan_t));
265   if (unlikely (!arabic_plan))
266     return nullptr;
267 
268   arabic_plan->do_fallback = plan->props.script == HB_SCRIPT_ARABIC;
269   arabic_plan->has_stch = !!plan->map.get_1_mask (HB_TAG ('s','t','c','h'));
270   for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++) {
271     arabic_plan->mask_array[i] = plan->map.get_1_mask (arabic_features[i]);
272     arabic_plan->do_fallback = arabic_plan->do_fallback &&
273 			       (FEATURE_IS_SYRIAC (arabic_features[i]) ||
274 			        plan->map.needs_fallback (arabic_features[i]));
275   }
276 
277   return arabic_plan;
278 }
279 
280 void
data_destroy_arabic(void * data)281 data_destroy_arabic (void *data)
282 {
283   arabic_shape_plan_t *arabic_plan = (arabic_shape_plan_t *) data;
284 
285   arabic_fallback_plan_destroy (arabic_plan->fallback_plan);
286 
287   free (data);
288 }
289 
290 static void
arabic_joining(hb_buffer_t * buffer)291 arabic_joining (hb_buffer_t *buffer)
292 {
293   unsigned int count = buffer->len;
294   hb_glyph_info_t *info = buffer->info;
295   unsigned int prev = (unsigned int) -1, state = 0;
296 
297   /* Check pre-context */
298   for (unsigned int i = 0; i < buffer->context_len[0]; i++)
299   {
300     unsigned int this_type = get_joining_type (buffer->context[0][i], buffer->unicode->general_category (buffer->context[0][i]));
301 
302     if (unlikely (this_type == JOINING_TYPE_T))
303       continue;
304 
305     const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
306     state = entry->next_state;
307     break;
308   }
309 
310   for (unsigned int i = 0; i < count; i++)
311   {
312     unsigned int this_type = get_joining_type (info[i].codepoint, _hb_glyph_info_get_general_category (&info[i]));
313 
314     if (unlikely (this_type == JOINING_TYPE_T)) {
315       info[i].arabic_shaping_action() = NONE;
316       continue;
317     }
318 
319     const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
320 
321     if (entry->prev_action != NONE && prev != (unsigned int) -1)
322     {
323       info[prev].arabic_shaping_action() = entry->prev_action;
324       buffer->unsafe_to_break (prev, i + 1);
325     }
326 
327     info[i].arabic_shaping_action() = entry->curr_action;
328 
329     prev = i;
330     state = entry->next_state;
331   }
332 
333   for (unsigned int i = 0; i < buffer->context_len[1]; i++)
334   {
335     unsigned int this_type = get_joining_type (buffer->context[1][i], buffer->unicode->general_category (buffer->context[1][i]));
336 
337     if (unlikely (this_type == JOINING_TYPE_T))
338       continue;
339 
340     const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
341     if (entry->prev_action != NONE && prev != (unsigned int) -1)
342       info[prev].arabic_shaping_action() = entry->prev_action;
343     break;
344   }
345 }
346 
347 static void
mongolian_variation_selectors(hb_buffer_t * buffer)348 mongolian_variation_selectors (hb_buffer_t *buffer)
349 {
350   /* Copy arabic_shaping_action() from base to Mongolian variation selectors. */
351   unsigned int count = buffer->len;
352   hb_glyph_info_t *info = buffer->info;
353   for (unsigned int i = 1; i < count; i++)
354     if (unlikely (hb_in_range<hb_codepoint_t> (info[i].codepoint, 0x180Bu, 0x180Du)))
355       info[i].arabic_shaping_action() = info[i - 1].arabic_shaping_action();
356 }
357 
358 void
setup_masks_arabic_plan(const arabic_shape_plan_t * arabic_plan,hb_buffer_t * buffer,hb_script_t script)359 setup_masks_arabic_plan (const arabic_shape_plan_t *arabic_plan,
360 			 hb_buffer_t               *buffer,
361 			 hb_script_t                script)
362 {
363   HB_BUFFER_ALLOCATE_VAR (buffer, arabic_shaping_action);
364 
365   arabic_joining (buffer);
366   if (script == HB_SCRIPT_MONGOLIAN)
367     mongolian_variation_selectors (buffer);
368 
369   unsigned int count = buffer->len;
370   hb_glyph_info_t *info = buffer->info;
371   for (unsigned int i = 0; i < count; i++)
372     info[i].mask |= arabic_plan->mask_array[info[i].arabic_shaping_action()];
373 }
374 
375 static void
setup_masks_arabic(const hb_ot_shape_plan_t * plan,hb_buffer_t * buffer,hb_font_t * font HB_UNUSED)376 setup_masks_arabic (const hb_ot_shape_plan_t *plan,
377 		    hb_buffer_t              *buffer,
378 		    hb_font_t                *font HB_UNUSED)
379 {
380   const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
381   setup_masks_arabic_plan (arabic_plan, buffer, plan->props.script);
382 }
383 
384 static void
arabic_fallback_shape(const hb_ot_shape_plan_t * plan,hb_font_t * font,hb_buffer_t * buffer)385 arabic_fallback_shape (const hb_ot_shape_plan_t *plan,
386 		       hb_font_t *font,
387 		       hb_buffer_t *buffer)
388 {
389 #ifdef HB_NO_OT_SHAPE_COMPLEX_ARABIC_FALLBACK
390   return;
391 #endif
392 
393   const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
394 
395   if (!arabic_plan->do_fallback)
396     return;
397 
398 retry:
399   arabic_fallback_plan_t *fallback_plan = arabic_plan->fallback_plan;
400   if (unlikely (!fallback_plan))
401   {
402     /* This sucks.  We need a font to build the fallback plan... */
403     fallback_plan = arabic_fallback_plan_create (plan, font);
404     if (unlikely (!arabic_plan->fallback_plan.cmpexch (nullptr, fallback_plan)))
405     {
406       arabic_fallback_plan_destroy (fallback_plan);
407       goto retry;
408     }
409   }
410 
411   arabic_fallback_plan_shape (fallback_plan, font, buffer);
412 }
413 
414 /*
415  * Stretch feature: "stch".
416  * See example here:
417  * https://docs.microsoft.com/en-us/typography/script-development/syriac
418  * We implement this in a generic way, such that the Arabic subtending
419  * marks can use it as well.
420  */
421 
422 static void
record_stch(const hb_ot_shape_plan_t * plan,hb_font_t * font HB_UNUSED,hb_buffer_t * buffer)423 record_stch (const hb_ot_shape_plan_t *plan,
424 	     hb_font_t *font HB_UNUSED,
425 	     hb_buffer_t *buffer)
426 {
427   const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
428   if (!arabic_plan->has_stch)
429     return;
430 
431   /* 'stch' feature was just applied.  Look for anything that multiplied,
432    * and record it for stch treatment later.  Note that rtlm, frac, etc
433    * are applied before stch, but we assume that they didn't result in
434    * anything multiplying into 5 pieces, so it's safe-ish... */
435 
436   unsigned int count = buffer->len;
437   hb_glyph_info_t *info = buffer->info;
438   for (unsigned int i = 0; i < count; i++)
439     if (unlikely (_hb_glyph_info_multiplied (&info[i])))
440     {
441       unsigned int comp = _hb_glyph_info_get_lig_comp (&info[i]);
442       info[i].arabic_shaping_action() = comp % 2 ? STCH_REPEATING : STCH_FIXED;
443       buffer->scratch_flags |= HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH;
444     }
445 }
446 
447 static void
apply_stch(const hb_ot_shape_plan_t * plan HB_UNUSED,hb_buffer_t * buffer,hb_font_t * font)448 apply_stch (const hb_ot_shape_plan_t *plan HB_UNUSED,
449 	    hb_buffer_t              *buffer,
450 	    hb_font_t                *font)
451 {
452   if (likely (!(buffer->scratch_flags & HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH)))
453     return;
454 
455   /* The Arabic shaper currently always processes in RTL mode, so we should
456    * stretch / position the stretched pieces to the left / preceding glyphs. */
457 
458   /* We do a two pass implementation:
459    * First pass calculates the exact number of extra glyphs we need,
460    * We then enlarge buffer to have that much room,
461    * Second pass applies the stretch, copying things to the end of buffer.
462    */
463 
464   int sign = font->x_scale < 0 ? -1 : +1;
465   unsigned int extra_glyphs_needed = 0; // Set during MEASURE, used during CUT
466   enum { MEASURE, CUT } /* step_t */;
467 
468   for (unsigned int step = MEASURE; step <= CUT; step = step + 1)
469   {
470     unsigned int count = buffer->len;
471     hb_glyph_info_t *info = buffer->info;
472     hb_glyph_position_t *pos = buffer->pos;
473     unsigned int new_len = count + extra_glyphs_needed; // write head during CUT
474     unsigned int j = new_len;
475     for (unsigned int i = count; i; i--)
476     {
477       if (!hb_in_range<uint8_t> (info[i - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING))
478       {
479 	if (step == CUT)
480 	{
481 	  --j;
482 	  info[j] = info[i - 1];
483 	  pos[j] = pos[i - 1];
484 	}
485 	continue;
486       }
487 
488       /* Yay, justification! */
489 
490       hb_position_t w_total = 0; // Total to be filled
491       hb_position_t w_fixed = 0; // Sum of fixed tiles
492       hb_position_t w_repeating = 0; // Sum of repeating tiles
493       int n_fixed = 0;
494       int n_repeating = 0;
495 
496       unsigned int end = i;
497       while (i &&
498 	     hb_in_range<uint8_t> (info[i - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING))
499       {
500 	i--;
501 	hb_position_t width = font->get_glyph_h_advance (info[i].codepoint);
502 	if (info[i].arabic_shaping_action() == STCH_FIXED)
503 	{
504 	  w_fixed += width;
505 	  n_fixed++;
506 	}
507 	else
508 	{
509 	  w_repeating += width;
510 	  n_repeating++;
511 	}
512       }
513       unsigned int start = i;
514       unsigned int context = i;
515       while (context &&
516 	     !hb_in_range<uint8_t> (info[context - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING) &&
517 	     (_hb_glyph_info_is_default_ignorable (&info[context - 1]) ||
518 	      HB_ARABIC_GENERAL_CATEGORY_IS_WORD (_hb_glyph_info_get_general_category (&info[context - 1]))))
519       {
520 	context--;
521 	w_total += pos[context].x_advance;
522       }
523       i++; // Don't touch i again.
524 
525       DEBUG_MSG (ARABIC, nullptr, "%s stretch at (%d,%d,%d)",
526 		 step == MEASURE ? "measuring" : "cutting", context, start, end);
527       DEBUG_MSG (ARABIC, nullptr, "rest of word:    count=%d width %d", start - context, w_total);
528       DEBUG_MSG (ARABIC, nullptr, "fixed tiles:     count=%d width=%d", n_fixed, w_fixed);
529       DEBUG_MSG (ARABIC, nullptr, "repeating tiles: count=%d width=%d", n_repeating, w_repeating);
530 
531       /* Number of additional times to repeat each repeating tile. */
532       int n_copies = 0;
533 
534       hb_position_t w_remaining = w_total - w_fixed;
535       if (sign * w_remaining > sign * w_repeating && sign * w_repeating > 0)
536 	n_copies = (sign * w_remaining) / (sign * w_repeating) - 1;
537 
538       /* See if we can improve the fit by adding an extra repeat and squeezing them together a bit. */
539       hb_position_t extra_repeat_overlap = 0;
540       hb_position_t shortfall = sign * w_remaining - sign * w_repeating * (n_copies + 1);
541       if (shortfall > 0 && n_repeating > 0)
542       {
543 	++n_copies;
544 	hb_position_t excess = (n_copies + 1) * sign * w_repeating - sign * w_remaining;
545 	if (excess > 0)
546 	  extra_repeat_overlap = excess / (n_copies * n_repeating);
547       }
548 
549       if (step == MEASURE)
550       {
551 	extra_glyphs_needed += n_copies * n_repeating;
552 	DEBUG_MSG (ARABIC, nullptr, "will add extra %d copies of repeating tiles", n_copies);
553       }
554       else
555       {
556 	buffer->unsafe_to_break (context, end);
557 	hb_position_t x_offset = 0;
558 	for (unsigned int k = end; k > start; k--)
559 	{
560 	  hb_position_t width = font->get_glyph_h_advance (info[k - 1].codepoint);
561 
562 	  unsigned int repeat = 1;
563 	  if (info[k - 1].arabic_shaping_action() == STCH_REPEATING)
564 	    repeat += n_copies;
565 
566 	  DEBUG_MSG (ARABIC, nullptr, "appending %d copies of glyph %d; j=%d",
567 		     repeat, info[k - 1].codepoint, j);
568 	  for (unsigned int n = 0; n < repeat; n++)
569 	  {
570 	    x_offset -= width;
571 	    if (n > 0)
572 	      x_offset += extra_repeat_overlap;
573 	    pos[k - 1].x_offset = x_offset;
574 	    /* Append copy. */
575 	    --j;
576 	    info[j] = info[k - 1];
577 	    pos[j] = pos[k - 1];
578 	  }
579 	}
580       }
581     }
582 
583     if (step == MEASURE)
584     {
585       if (unlikely (!buffer->ensure (count + extra_glyphs_needed)))
586 	break;
587     }
588     else
589     {
590       assert (j == 0);
591       buffer->len = new_len;
592     }
593   }
594 }
595 
596 
597 static void
postprocess_glyphs_arabic(const hb_ot_shape_plan_t * plan,hb_buffer_t * buffer,hb_font_t * font)598 postprocess_glyphs_arabic (const hb_ot_shape_plan_t *plan,
599 			   hb_buffer_t              *buffer,
600 			   hb_font_t                *font)
601 {
602   apply_stch (plan, buffer, font);
603 
604   HB_BUFFER_DEALLOCATE_VAR (buffer, arabic_shaping_action);
605 }
606 
607 /* http://www.unicode.org/reports/tr53/ */
608 
609 static hb_codepoint_t
610 modifier_combining_marks[] =
611 {
612   0x0654u, /* ARABIC HAMZA ABOVE */
613   0x0655u, /* ARABIC HAMZA BELOW */
614   0x0658u, /* ARABIC MARK NOON GHUNNA */
615   0x06DCu, /* ARABIC SMALL HIGH SEEN */
616   0x06E3u, /* ARABIC SMALL LOW SEEN */
617   0x06E7u, /* ARABIC SMALL HIGH YEH */
618   0x06E8u, /* ARABIC SMALL HIGH NOON */
619   0x08D3u, /* ARABIC SMALL LOW WAW */
620   0x08F3u, /* ARABIC SMALL HIGH WAW */
621 };
622 
623 static inline bool
info_is_mcm(const hb_glyph_info_t & info)624 info_is_mcm (const hb_glyph_info_t &info)
625 {
626   hb_codepoint_t u = info.codepoint;
627   for (unsigned int i = 0; i < ARRAY_LENGTH (modifier_combining_marks); i++)
628     if (u == modifier_combining_marks[i])
629       return true;
630   return false;
631 }
632 
633 static void
reorder_marks_arabic(const hb_ot_shape_plan_t * plan HB_UNUSED,hb_buffer_t * buffer,unsigned int start,unsigned int end)634 reorder_marks_arabic (const hb_ot_shape_plan_t *plan HB_UNUSED,
635 		      hb_buffer_t              *buffer,
636 		      unsigned int              start,
637 		      unsigned int              end)
638 {
639   hb_glyph_info_t *info = buffer->info;
640 
641   DEBUG_MSG (ARABIC, buffer, "Reordering marks from %d to %d", start, end);
642 
643   unsigned int i = start;
644   for (unsigned int cc = 220; cc <= 230; cc += 10)
645   {
646     DEBUG_MSG (ARABIC, buffer, "Looking for %d's starting at %d", cc, i);
647     while (i < end && info_cc(info[i]) < cc)
648       i++;
649     DEBUG_MSG (ARABIC, buffer, "Looking for %d's stopped at %d", cc, i);
650 
651     if (i == end)
652       break;
653 
654     if (info_cc(info[i]) > cc)
655       continue;
656 
657     unsigned int j = i;
658     while (j < end && info_cc(info[j]) == cc && info_is_mcm (info[j]))
659       j++;
660 
661     if (i == j)
662       continue;
663 
664     DEBUG_MSG (ARABIC, buffer, "Found %d's from %d to %d", cc, i, j);
665 
666     /* Shift it! */
667     DEBUG_MSG (ARABIC, buffer, "Shifting %d's: %d %d", cc, i, j);
668     hb_glyph_info_t temp[HB_OT_SHAPE_COMPLEX_MAX_COMBINING_MARKS];
669     assert (j - i <= ARRAY_LENGTH (temp));
670     buffer->merge_clusters (start, j);
671     memmove (temp, &info[i], (j - i) * sizeof (hb_glyph_info_t));
672     memmove (&info[start + j - i], &info[start], (i - start) * sizeof (hb_glyph_info_t));
673     memmove (&info[start], temp, (j - i) * sizeof (hb_glyph_info_t));
674 
675     /* Renumber CC such that the reordered sequence is still sorted.
676      * 22 and 26 are chosen because they are smaller than all Arabic categories,
677      * and are folded back to 220/230 respectively during fallback mark positioning.
678      *
679      * We do this because the CGJ-handling logic in the normalizer relies on
680      * mark sequences having an increasing order even after this reordering.
681      * https://github.com/harfbuzz/harfbuzz/issues/554
682      * This, however, does break some obscure sequences, where the normalizer
683      * might compose a sequence that it should not.  For example, in the seequence
684      * ALEF, HAMZAH, MADDAH, we should NOT try to compose ALEF+MADDAH, but with this
685      * renumbering, we will.
686      */
687     unsigned int new_start = start + j - i;
688     unsigned int new_cc = cc == 220 ? HB_MODIFIED_COMBINING_CLASS_CCC22 : HB_MODIFIED_COMBINING_CLASS_CCC26;
689     while (start < new_start)
690     {
691       _hb_glyph_info_set_modified_combining_class (&info[start], new_cc);
692       start++;
693     }
694 
695     i = j;
696   }
697 }
698 
699 const hb_ot_complex_shaper_t _hb_ot_complex_shaper_arabic =
700 {
701   collect_features_arabic,
702   nullptr, /* override_features */
703   data_create_arabic,
704   data_destroy_arabic,
705   nullptr, /* preprocess_text */
706   postprocess_glyphs_arabic,
707   HB_OT_SHAPE_NORMALIZATION_MODE_DEFAULT,
708   nullptr, /* decompose */
709   nullptr, /* compose */
710   setup_masks_arabic,
711   HB_TAG_NONE, /* gpos_tag */
712   reorder_marks_arabic,
713   HB_OT_SHAPE_ZERO_WIDTH_MARKS_BY_GDEF_LATE,
714   true, /* fallback_position */
715 };
716 
717 
718 #endif
719