1/*
2 * Copyright © 2011,2012 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-ot-shape-fallback.hh"
28#include "hb-kern.hh"
29
30static unsigned int
31recategorize_combining_class (hb_codepoint_t u,
32 unsigned int klass)
33{
34 if (klass >= 200)
35 return klass;
36
37 /* Thai / Lao need some per-character work. */
38 if ((u & ~0xFF) == 0x0E00u)
39 {
40 if (unlikely (klass == 0))
41 {
42 switch (u)
43 {
44 case 0x0E31u:
45 case 0x0E34u:
46 case 0x0E35u:
47 case 0x0E36u:
48 case 0x0E37u:
49 case 0x0E47u:
50 case 0x0E4Cu:
51 case 0x0E4Du:
52 case 0x0E4Eu:
53 klass = HB_UNICODE_COMBINING_CLASS_ABOVE_RIGHT;
54 break;
55
56 case 0x0EB1u:
57 case 0x0EB4u:
58 case 0x0EB5u:
59 case 0x0EB6u:
60 case 0x0EB7u:
61 case 0x0EBBu:
62 case 0x0ECCu:
63 case 0x0ECDu:
64 klass = HB_UNICODE_COMBINING_CLASS_ABOVE;
65 break;
66
67 case 0x0EBCu:
68 klass = HB_UNICODE_COMBINING_CLASS_BELOW;
69 break;
70 }
71 } else {
72 /* Thai virama is below-right */
73 if (u == 0x0E3Au)
74 klass = HB_UNICODE_COMBINING_CLASS_BELOW_RIGHT;
75 }
76 }
77
78 switch (klass)
79 {
80
81 /* Hebrew */
82
83 case HB_MODIFIED_COMBINING_CLASS_CCC10: /* sheva */
84 case HB_MODIFIED_COMBINING_CLASS_CCC11: /* hataf segol */
85 case HB_MODIFIED_COMBINING_CLASS_CCC12: /* hataf patah */
86 case HB_MODIFIED_COMBINING_CLASS_CCC13: /* hataf qamats */
87 case HB_MODIFIED_COMBINING_CLASS_CCC14: /* hiriq */
88 case HB_MODIFIED_COMBINING_CLASS_CCC15: /* tsere */
89 case HB_MODIFIED_COMBINING_CLASS_CCC16: /* segol */
90 case HB_MODIFIED_COMBINING_CLASS_CCC17: /* patah */
91 case HB_MODIFIED_COMBINING_CLASS_CCC18: /* qamats */
92 case HB_MODIFIED_COMBINING_CLASS_CCC20: /* qubuts */
93 case HB_MODIFIED_COMBINING_CLASS_CCC22: /* meteg */
94 return HB_UNICODE_COMBINING_CLASS_BELOW;
95
96 case HB_MODIFIED_COMBINING_CLASS_CCC23: /* rafe */
97 return HB_UNICODE_COMBINING_CLASS_ATTACHED_ABOVE;
98
99 case HB_MODIFIED_COMBINING_CLASS_CCC24: /* shin dot */
100 return HB_UNICODE_COMBINING_CLASS_ABOVE_RIGHT;
101
102 case HB_MODIFIED_COMBINING_CLASS_CCC25: /* sin dot */
103 case HB_MODIFIED_COMBINING_CLASS_CCC19: /* holam */
104 return HB_UNICODE_COMBINING_CLASS_ABOVE_LEFT;
105
106 case HB_MODIFIED_COMBINING_CLASS_CCC26: /* point varika */
107 return HB_UNICODE_COMBINING_CLASS_ABOVE;
108
109 case HB_MODIFIED_COMBINING_CLASS_CCC21: /* dagesh */
110 break;
111
112
113 /* Arabic and Syriac */
114
115 case HB_MODIFIED_COMBINING_CLASS_CCC27: /* fathatan */
116 case HB_MODIFIED_COMBINING_CLASS_CCC28: /* dammatan */
117 case HB_MODIFIED_COMBINING_CLASS_CCC30: /* fatha */
118 case HB_MODIFIED_COMBINING_CLASS_CCC31: /* damma */
119 case HB_MODIFIED_COMBINING_CLASS_CCC33: /* shadda */
120 case HB_MODIFIED_COMBINING_CLASS_CCC34: /* sukun */
121 case HB_MODIFIED_COMBINING_CLASS_CCC35: /* superscript alef */
122 case HB_MODIFIED_COMBINING_CLASS_CCC36: /* superscript alaph */
123 return HB_UNICODE_COMBINING_CLASS_ABOVE;
124
125 case HB_MODIFIED_COMBINING_CLASS_CCC29: /* kasratan */
126 case HB_MODIFIED_COMBINING_CLASS_CCC32: /* kasra */
127 return HB_UNICODE_COMBINING_CLASS_BELOW;
128
129
130 /* Thai */
131
132 case HB_MODIFIED_COMBINING_CLASS_CCC103: /* sara u / sara uu */
133 return HB_UNICODE_COMBINING_CLASS_BELOW_RIGHT;
134
135 case HB_MODIFIED_COMBINING_CLASS_CCC107: /* mai */
136 return HB_UNICODE_COMBINING_CLASS_ABOVE_RIGHT;
137
138
139 /* Lao */
140
141 case HB_MODIFIED_COMBINING_CLASS_CCC118: /* sign u / sign uu */
142 return HB_UNICODE_COMBINING_CLASS_BELOW;
143
144 case HB_MODIFIED_COMBINING_CLASS_CCC122: /* mai */
145 return HB_UNICODE_COMBINING_CLASS_ABOVE;
146
147
148 /* Tibetan */
149
150 case HB_MODIFIED_COMBINING_CLASS_CCC129: /* sign aa */
151 return HB_UNICODE_COMBINING_CLASS_BELOW;
152
153 case HB_MODIFIED_COMBINING_CLASS_CCC130: /* sign i*/
154 return HB_UNICODE_COMBINING_CLASS_ABOVE;
155
156 case HB_MODIFIED_COMBINING_CLASS_CCC132: /* sign u */
157 return HB_UNICODE_COMBINING_CLASS_BELOW;
158
159 }
160
161 return klass;
162}
163
164void
165_hb_ot_shape_fallback_mark_position_recategorize_marks (const hb_ot_shape_plan_t *plan HB_UNUSED,
166 hb_font_t *font HB_UNUSED,
167 hb_buffer_t *buffer)
168{
169 unsigned int count = buffer->len;
170 hb_glyph_info_t *info = buffer->info;
171 for (unsigned int i = 0; i < count; i++)
172 if (_hb_glyph_info_get_general_category (&info[i]) == HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK) {
173 unsigned int combining_class = _hb_glyph_info_get_modified_combining_class (&info[i]);
174 combining_class = recategorize_combining_class (info[i].codepoint, combining_class);
175 _hb_glyph_info_set_modified_combining_class (&info[i], combining_class);
176 }
177}
178
179
180static void
181zero_mark_advances (hb_buffer_t *buffer,
182 unsigned int start,
183 unsigned int end,
184 bool adjust_offsets_when_zeroing)
185{
186 hb_glyph_info_t *info = buffer->info;
187 for (unsigned int i = start; i < end; i++)
188 if (_hb_glyph_info_get_general_category (&info[i]) == HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK)
189 {
190 if (adjust_offsets_when_zeroing)
191 {
192 buffer->pos[i].x_offset -= buffer->pos[i].x_advance;
193 buffer->pos[i].y_offset -= buffer->pos[i].y_advance;
194 }
195 buffer->pos[i].x_advance = 0;
196 buffer->pos[i].y_advance = 0;
197 }
198}
199
200static inline void
201position_mark (const hb_ot_shape_plan_t *plan HB_UNUSED,
202 hb_font_t *font,
203 hb_buffer_t *buffer,
204 hb_glyph_extents_t &base_extents,
205 unsigned int i,
206 unsigned int combining_class)
207{
208 hb_glyph_extents_t mark_extents;
209 if (!font->get_glyph_extents (buffer->info[i].codepoint, &mark_extents))
210 return;
211
212 hb_position_t y_gap = font->y_scale / 16;
213
214 hb_glyph_position_t &pos = buffer->pos[i];
215 pos.x_offset = pos.y_offset = 0;
216
217
218 /* We don't position LEFT and RIGHT marks. */
219
220 /* X positioning */
221 switch (combining_class)
222 {
223 case HB_UNICODE_COMBINING_CLASS_DOUBLE_BELOW:
224 case HB_UNICODE_COMBINING_CLASS_DOUBLE_ABOVE:
225 if (buffer->props.direction == HB_DIRECTION_LTR) {
226 pos.x_offset += base_extents.x_bearing + base_extents.width - mark_extents.width / 2 - mark_extents.x_bearing;
227 break;
228 } else if (buffer->props.direction == HB_DIRECTION_RTL) {
229 pos.x_offset += base_extents.x_bearing - mark_extents.width / 2 - mark_extents.x_bearing;
230 break;
231 }
232 HB_FALLTHROUGH;
233
234 default:
235 case HB_UNICODE_COMBINING_CLASS_ATTACHED_BELOW:
236 case HB_UNICODE_COMBINING_CLASS_ATTACHED_ABOVE:
237 case HB_UNICODE_COMBINING_CLASS_BELOW:
238 case HB_UNICODE_COMBINING_CLASS_ABOVE:
239 /* Center align. */
240 pos.x_offset += base_extents.x_bearing + (base_extents.width - mark_extents.width) / 2 - mark_extents.x_bearing;
241 break;
242
243 case HB_UNICODE_COMBINING_CLASS_ATTACHED_BELOW_LEFT:
244 case HB_UNICODE_COMBINING_CLASS_BELOW_LEFT:
245 case HB_UNICODE_COMBINING_CLASS_ABOVE_LEFT:
246 /* Left align. */
247 pos.x_offset += base_extents.x_bearing - mark_extents.x_bearing;
248 break;
249
250 case HB_UNICODE_COMBINING_CLASS_ATTACHED_ABOVE_RIGHT:
251 case HB_UNICODE_COMBINING_CLASS_BELOW_RIGHT:
252 case HB_UNICODE_COMBINING_CLASS_ABOVE_RIGHT:
253 /* Right align. */
254 pos.x_offset += base_extents.x_bearing + base_extents.width - mark_extents.width - mark_extents.x_bearing;
255 break;
256 }
257
258 /* Y positioning */
259 switch (combining_class)
260 {
261 case HB_UNICODE_COMBINING_CLASS_DOUBLE_BELOW:
262 case HB_UNICODE_COMBINING_CLASS_BELOW_LEFT:
263 case HB_UNICODE_COMBINING_CLASS_BELOW:
264 case HB_UNICODE_COMBINING_CLASS_BELOW_RIGHT:
265 /* Add gap, fall-through. */
266 base_extents.height -= y_gap;
267 HB_FALLTHROUGH;
268
269 case HB_UNICODE_COMBINING_CLASS_ATTACHED_BELOW_LEFT:
270 case HB_UNICODE_COMBINING_CLASS_ATTACHED_BELOW:
271 pos.y_offset = base_extents.y_bearing + base_extents.height - mark_extents.y_bearing;
272 /* Never shift up "below" marks. */
273 if ((y_gap > 0) == (pos.y_offset > 0))
274 {
275 base_extents.height -= pos.y_offset;
276 pos.y_offset = 0;
277 }
278 base_extents.height += mark_extents.height;
279 break;
280
281 case HB_UNICODE_COMBINING_CLASS_DOUBLE_ABOVE:
282 case HB_UNICODE_COMBINING_CLASS_ABOVE_LEFT:
283 case HB_UNICODE_COMBINING_CLASS_ABOVE:
284 case HB_UNICODE_COMBINING_CLASS_ABOVE_RIGHT:
285 /* Add gap, fall-through. */
286 base_extents.y_bearing += y_gap;
287 base_extents.height -= y_gap;
288 HB_FALLTHROUGH;
289
290 case HB_UNICODE_COMBINING_CLASS_ATTACHED_ABOVE:
291 case HB_UNICODE_COMBINING_CLASS_ATTACHED_ABOVE_RIGHT:
292 pos.y_offset = base_extents.y_bearing - (mark_extents.y_bearing + mark_extents.height);
293 /* Don't shift down "above" marks too much. */
294 if ((y_gap > 0) != (pos.y_offset > 0))
295 {
296 unsigned int correction = -pos.y_offset / 2;
297 base_extents.y_bearing += correction;
298 base_extents.height -= correction;
299 pos.y_offset += correction;
300 }
301 base_extents.y_bearing -= mark_extents.height;
302 base_extents.height += mark_extents.height;
303 break;
304 }
305}
306
307static inline void
308position_around_base (const hb_ot_shape_plan_t *plan,
309 hb_font_t *font,
310 hb_buffer_t *buffer,
311 unsigned int base,
312 unsigned int end,
313 bool adjust_offsets_when_zeroing)
314{
315 hb_direction_t horiz_dir = HB_DIRECTION_INVALID;
316
317 buffer->unsafe_to_break (base, end);
318
319 hb_glyph_extents_t base_extents;
320 if (!font->get_glyph_extents (buffer->info[base].codepoint,
321 &base_extents))
322 {
323 /* If extents don't work, zero marks and go home. */
324 zero_mark_advances (buffer, base + 1, end, adjust_offsets_when_zeroing);
325 return;
326 }
327 base_extents.y_bearing += buffer->pos[base].y_offset;
328 /* Use horizontal advance for horizontal positioning.
329 * Generally a better idea. Also works for zero-ink glyphs. See:
330 * https://github.com/harfbuzz/harfbuzz/issues/1532 */
331 base_extents.x_bearing = 0;
332 base_extents.width = font->get_glyph_h_advance (buffer->info[base].codepoint);
333
334 unsigned int lig_id = _hb_glyph_info_get_lig_id (&buffer->info[base]);
335 /* Use integer for num_lig_components such that it doesn't convert to unsigned
336 * when we divide or multiply by it. */
337 int num_lig_components = _hb_glyph_info_get_lig_num_comps (&buffer->info[base]);
338
339 hb_position_t x_offset = 0, y_offset = 0;
340 if (HB_DIRECTION_IS_FORWARD (buffer->props.direction)) {
341 x_offset -= buffer->pos[base].x_advance;
342 y_offset -= buffer->pos[base].y_advance;
343 }
344
345 hb_glyph_extents_t component_extents = base_extents;
346 int last_lig_component = -1;
347 unsigned int last_combining_class = 255;
348 hb_glyph_extents_t cluster_extents = base_extents; /* Initialization is just to shut gcc up. */
349 hb_glyph_info_t *info = buffer->info;
350 for (unsigned int i = base + 1; i < end; i++)
351 if (_hb_glyph_info_get_modified_combining_class (&info[i]))
352 {
353 if (num_lig_components > 1) {
354 unsigned int this_lig_id = _hb_glyph_info_get_lig_id (&info[i]);
355 int this_lig_component = _hb_glyph_info_get_lig_comp (&info[i]) - 1;
356 /* Conditions for attaching to the last component. */
357 if (!lig_id || lig_id != this_lig_id || this_lig_component >= num_lig_components)
358 this_lig_component = num_lig_components - 1;
359 if (last_lig_component != this_lig_component)
360 {
361 last_lig_component = this_lig_component;
362 last_combining_class = 255;
363 component_extents = base_extents;
364 if (unlikely (horiz_dir == HB_DIRECTION_INVALID)) {
365 if (HB_DIRECTION_IS_HORIZONTAL (plan->props.direction))
366 horiz_dir = plan->props.direction;
367 else
368 horiz_dir = hb_script_get_horizontal_direction (plan->props.script);
369 }
370 if (horiz_dir == HB_DIRECTION_LTR)
371 component_extents.x_bearing += (this_lig_component * component_extents.width) / num_lig_components;
372 else
373 component_extents.x_bearing += ((num_lig_components - 1 - this_lig_component) * component_extents.width) / num_lig_components;
374 component_extents.width /= num_lig_components;
375 }
376 }
377
378 unsigned int this_combining_class = _hb_glyph_info_get_modified_combining_class (&info[i]);
379 if (last_combining_class != this_combining_class)
380 {
381 last_combining_class = this_combining_class;
382 cluster_extents = component_extents;
383 }
384
385 position_mark (plan, font, buffer, cluster_extents, i, this_combining_class);
386
387 buffer->pos[i].x_advance = 0;
388 buffer->pos[i].y_advance = 0;
389 buffer->pos[i].x_offset += x_offset;
390 buffer->pos[i].y_offset += y_offset;
391
392 } else {
393 if (HB_DIRECTION_IS_FORWARD (buffer->props.direction)) {
394 x_offset -= buffer->pos[i].x_advance;
395 y_offset -= buffer->pos[i].y_advance;
396 } else {
397 x_offset += buffer->pos[i].x_advance;
398 y_offset += buffer->pos[i].y_advance;
399 }
400 }
401}
402
403static inline void
404position_cluster (const hb_ot_shape_plan_t *plan,
405 hb_font_t *font,
406 hb_buffer_t *buffer,
407 unsigned int start,
408 unsigned int end,
409 bool adjust_offsets_when_zeroing)
410{
411 if (end - start < 2)
412 return;
413
414 /* Find the base glyph */
415 hb_glyph_info_t *info = buffer->info;
416 for (unsigned int i = start; i < end; i++)
417 if (!HB_UNICODE_GENERAL_CATEGORY_IS_MARK (_hb_glyph_info_get_general_category (&info[i])))
418 {
419 /* Find mark glyphs */
420 unsigned int j;
421 for (j = i + 1; j < end; j++)
422 if (!HB_UNICODE_GENERAL_CATEGORY_IS_MARK (_hb_glyph_info_get_general_category (&info[j])))
423 break;
424
425 position_around_base (plan, font, buffer, i, j, adjust_offsets_when_zeroing);
426
427 i = j - 1;
428 }
429}
430
431void
432_hb_ot_shape_fallback_mark_position (const hb_ot_shape_plan_t *plan,
433 hb_font_t *font,
434 hb_buffer_t *buffer,
435 bool adjust_offsets_when_zeroing)
436{
437 _hb_buffer_assert_gsubgpos_vars (buffer);
438
439 unsigned int start = 0;
440 unsigned int count = buffer->len;
441 hb_glyph_info_t *info = buffer->info;
442 for (unsigned int i = 1; i < count; i++)
443 if (likely (!HB_UNICODE_GENERAL_CATEGORY_IS_MARK (_hb_glyph_info_get_general_category (&info[i])))) {
444 position_cluster (plan, font, buffer, start, i, adjust_offsets_when_zeroing);
445 start = i;
446 }
447 position_cluster (plan, font, buffer, start, count, adjust_offsets_when_zeroing);
448}
449
450
451struct hb_ot_shape_fallback_kern_driver_t
452{
453 hb_ot_shape_fallback_kern_driver_t (hb_font_t *font_,
454 hb_buffer_t *buffer) :
455 font (font_), direction (buffer->props.direction) {}
456
457 hb_position_t get_kerning (hb_codepoint_t first, hb_codepoint_t second) const
458 {
459 hb_position_t kern = 0;
460 font->get_glyph_kerning_for_direction (first, second,
461 direction,
462 &kern, &kern);
463 return kern;
464 }
465
466 hb_font_t *font;
467 hb_direction_t direction;
468};
469
470/* Performs font-assisted kerning. */
471void
472_hb_ot_shape_fallback_kern (const hb_ot_shape_plan_t *plan,
473 hb_font_t *font,
474 hb_buffer_t *buffer)
475{
476 if (HB_DIRECTION_IS_HORIZONTAL (buffer->props.direction) ?
477 !font->has_glyph_h_kerning_func () :
478 !font->has_glyph_v_kerning_func ())
479 return;
480
481 bool reverse = HB_DIRECTION_IS_BACKWARD (buffer->props.direction);
482
483 if (reverse)
484 buffer->reverse ();
485
486 hb_ot_shape_fallback_kern_driver_t driver (font, buffer);
487 OT::hb_kern_machine_t<hb_ot_shape_fallback_kern_driver_t> machine (driver);
488 machine.kern (font, buffer, plan->kern_mask, false);
489
490 if (reverse)
491 buffer->reverse ();
492}
493
494
495/* Adjusts width of various spaces. */
496void
497_hb_ot_shape_fallback_spaces (const hb_ot_shape_plan_t *plan HB_UNUSED,
498 hb_font_t *font,
499 hb_buffer_t *buffer)
500{
501 hb_glyph_info_t *info = buffer->info;
502 hb_glyph_position_t *pos = buffer->pos;
503 bool horizontal = HB_DIRECTION_IS_HORIZONTAL (buffer->props.direction);
504 unsigned int count = buffer->len;
505 for (unsigned int i = 0; i < count; i++)
506 if (_hb_glyph_info_is_unicode_space (&info[i]) && !_hb_glyph_info_ligated (&info[i]))
507 {
508 hb_unicode_funcs_t::space_t space_type = _hb_glyph_info_get_unicode_space_fallback_type (&info[i]);
509 hb_codepoint_t glyph;
510 typedef hb_unicode_funcs_t t;
511 switch (space_type)
512 {
513 case t::NOT_SPACE: /* Shouldn't happen. */
514 case t::SPACE:
515 break;
516
517 case t::SPACE_EM:
518 case t::SPACE_EM_2:
519 case t::SPACE_EM_3:
520 case t::SPACE_EM_4:
521 case t::SPACE_EM_5:
522 case t::SPACE_EM_6:
523 case t::SPACE_EM_16:
524 if (horizontal)
525 pos[i].x_advance = +(font->x_scale + ((int) space_type)/2) / (int) space_type;
526 else
527 pos[i].y_advance = -(font->y_scale + ((int) space_type)/2) / (int) space_type;
528 break;
529
530 case t::SPACE_4_EM_18:
531 if (horizontal)
532 pos[i].x_advance = (int64_t) +font->x_scale * 4 / 18;
533 else
534 pos[i].y_advance = (int64_t) -font->y_scale * 4 / 18;
535 break;
536
537 case t::SPACE_FIGURE:
538 for (char u = '0'; u <= '9'; u++)
539 if (font->get_nominal_glyph (u, &glyph))
540 {
541 if (horizontal)
542 pos[i].x_advance = font->get_glyph_h_advance (glyph);
543 else
544 pos[i].y_advance = font->get_glyph_v_advance (glyph);
545 break;
546 }
547 break;
548
549 case t::SPACE_PUNCTUATION:
550 if (font->get_nominal_glyph ('.', &glyph) ||
551 font->get_nominal_glyph (',', &glyph))
552 {
553 if (horizontal)
554 pos[i].x_advance = font->get_glyph_h_advance (glyph);
555 else
556 pos[i].y_advance = font->get_glyph_v_advance (glyph);
557 }
558 break;
559
560 case t::SPACE_NARROW:
561 /* Half-space?
562 * Unicode doc https://unicode.org/charts/PDF/U2000.pdf says ~1/4 or 1/5 of EM.
563 * However, in my testing, many fonts have their regular space being about that
564 * size. To me, a percentage of the space width makes more sense. Half is as
565 * good as any. */
566 if (horizontal)
567 pos[i].x_advance /= 2;
568 else
569 pos[i].y_advance /= 2;
570 break;
571 }
572 }
573}
574