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.hh" |
28 | |
29 | #ifndef HB_NO_OT_SHAPE |
30 | |
31 | #include "hb-ot-shape-complex-khmer.hh" |
32 | #include "hb-ot-layout.hh" |
33 | |
34 | |
35 | /* |
36 | * Khmer shaper. |
37 | */ |
38 | |
39 | static const hb_ot_map_feature_t |
40 | khmer_features[] = |
41 | { |
42 | /* |
43 | * Basic features. |
44 | * These features are applied in order, one at a time, after reordering. |
45 | */ |
46 | {HB_TAG('p','r','e','f'), F_MANUAL_JOINERS}, |
47 | {HB_TAG('b','l','w','f'), F_MANUAL_JOINERS}, |
48 | {HB_TAG('a','b','v','f'), F_MANUAL_JOINERS}, |
49 | {HB_TAG('p','s','t','f'), F_MANUAL_JOINERS}, |
50 | {HB_TAG('c','f','a','r'), F_MANUAL_JOINERS}, |
51 | /* |
52 | * Other features. |
53 | * These features are applied all at once after clearing syllables. |
54 | */ |
55 | {HB_TAG('p','r','e','s'), F_GLOBAL_MANUAL_JOINERS}, |
56 | {HB_TAG('a','b','v','s'), F_GLOBAL_MANUAL_JOINERS}, |
57 | {HB_TAG('b','l','w','s'), F_GLOBAL_MANUAL_JOINERS}, |
58 | {HB_TAG('p','s','t','s'), F_GLOBAL_MANUAL_JOINERS}, |
59 | }; |
60 | |
61 | /* |
62 | * Must be in the same order as the khmer_features array. |
63 | */ |
64 | enum { |
65 | KHMER_PREF, |
66 | KHMER_BLWF, |
67 | KHMER_ABVF, |
68 | KHMER_PSTF, |
69 | KHMER_CFAR, |
70 | |
71 | _KHMER_PRES, |
72 | _KHMER_ABVS, |
73 | _KHMER_BLWS, |
74 | _KHMER_PSTS, |
75 | |
76 | KHMER_NUM_FEATURES, |
77 | KHMER_BASIC_FEATURES = _KHMER_PRES, /* Don't forget to update this! */ |
78 | }; |
79 | |
80 | static void |
81 | setup_syllables_khmer (const hb_ot_shape_plan_t *plan, |
82 | hb_font_t *font, |
83 | hb_buffer_t *buffer); |
84 | static void |
85 | reorder_khmer (const hb_ot_shape_plan_t *plan, |
86 | hb_font_t *font, |
87 | hb_buffer_t *buffer); |
88 | |
89 | static void |
90 | collect_features_khmer (hb_ot_shape_planner_t *plan) |
91 | { |
92 | hb_ot_map_builder_t *map = &plan->map; |
93 | |
94 | /* Do this before any lookups have been applied. */ |
95 | map->add_gsub_pause (setup_syllables_khmer); |
96 | map->add_gsub_pause (reorder_khmer); |
97 | |
98 | /* Testing suggests that Uniscribe does NOT pause between basic |
99 | * features. Test with KhmerUI.ttf and the following three |
100 | * sequences: |
101 | * |
102 | * U+1789,U+17BC |
103 | * U+1789,U+17D2,U+1789 |
104 | * U+1789,U+17D2,U+1789,U+17BC |
105 | * |
106 | * https://github.com/harfbuzz/harfbuzz/issues/974 |
107 | */ |
108 | map->enable_feature (HB_TAG('l','o','c','l')); |
109 | map->enable_feature (HB_TAG('c','c','m','p')); |
110 | |
111 | unsigned int i = 0; |
112 | for (; i < KHMER_BASIC_FEATURES; i++) |
113 | map->add_feature (khmer_features[i]); |
114 | |
115 | map->add_gsub_pause (_hb_clear_syllables); |
116 | |
117 | for (; i < KHMER_NUM_FEATURES; i++) |
118 | map->add_feature (khmer_features[i]); |
119 | } |
120 | |
121 | static void |
122 | override_features_khmer (hb_ot_shape_planner_t *plan) |
123 | { |
124 | hb_ot_map_builder_t *map = &plan->map; |
125 | |
126 | /* Khmer spec has 'clig' as part of required shaping features: |
127 | * "Apply feature 'clig' to form ligatures that are desired for |
128 | * typographical correctness.", hence in overrides... */ |
129 | map->enable_feature (HB_TAG('c','l','i','g')); |
130 | |
131 | /* Uniscribe does not apply 'kern' in Khmer. */ |
132 | if (hb_options ().uniscribe_bug_compatible) |
133 | { |
134 | map->disable_feature (HB_TAG('k','e','r','n')); |
135 | } |
136 | |
137 | map->disable_feature (HB_TAG('l','i','g','a')); |
138 | } |
139 | |
140 | |
141 | struct khmer_shape_plan_t |
142 | { |
143 | bool get_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const |
144 | { |
145 | hb_codepoint_t glyph = virama_glyph; |
146 | if (unlikely (virama_glyph == (hb_codepoint_t) -1)) |
147 | { |
148 | if (!font->get_nominal_glyph (0x17D2u, &glyph)) |
149 | glyph = 0; |
150 | /* Technically speaking, the spec says we should apply 'locl' to virama too. |
151 | * Maybe one day... */ |
152 | |
153 | /* Our get_nominal_glyph() function needs a font, so we can't get the virama glyph |
154 | * during shape planning... Instead, overwrite it here. It's safe. Don't worry! */ |
155 | virama_glyph = glyph; |
156 | } |
157 | |
158 | *pglyph = glyph; |
159 | return glyph != 0; |
160 | } |
161 | |
162 | mutable hb_codepoint_t virama_glyph; |
163 | |
164 | hb_mask_t mask_array[KHMER_NUM_FEATURES]; |
165 | }; |
166 | |
167 | static void * |
168 | data_create_khmer (const hb_ot_shape_plan_t *plan) |
169 | { |
170 | khmer_shape_plan_t *khmer_plan = (khmer_shape_plan_t *) calloc (1, sizeof (khmer_shape_plan_t)); |
171 | if (unlikely (!khmer_plan)) |
172 | return nullptr; |
173 | |
174 | khmer_plan->virama_glyph = (hb_codepoint_t) -1; |
175 | |
176 | for (unsigned int i = 0; i < ARRAY_LENGTH (khmer_plan->mask_array); i++) |
177 | khmer_plan->mask_array[i] = (khmer_features[i].flags & F_GLOBAL) ? |
178 | 0 : plan->map.get_1_mask (khmer_features[i].tag); |
179 | |
180 | return khmer_plan; |
181 | } |
182 | |
183 | static void |
184 | data_destroy_khmer (void *data) |
185 | { |
186 | free (data); |
187 | } |
188 | |
189 | |
190 | enum khmer_syllable_type_t { |
191 | khmer_consonant_syllable, |
192 | khmer_broken_cluster, |
193 | khmer_non_khmer_cluster, |
194 | }; |
195 | |
196 | #include "hb-ot-shape-complex-khmer-machine.hh" |
197 | |
198 | static void |
199 | setup_masks_khmer (const hb_ot_shape_plan_t *plan HB_UNUSED, |
200 | hb_buffer_t *buffer, |
201 | hb_font_t *font HB_UNUSED) |
202 | { |
203 | HB_BUFFER_ALLOCATE_VAR (buffer, khmer_category); |
204 | |
205 | /* We cannot setup masks here. We save information about characters |
206 | * and setup masks later on in a pause-callback. */ |
207 | |
208 | unsigned int count = buffer->len; |
209 | hb_glyph_info_t *info = buffer->info; |
210 | for (unsigned int i = 0; i < count; i++) |
211 | set_khmer_properties (info[i]); |
212 | } |
213 | |
214 | static void |
215 | setup_syllables_khmer (const hb_ot_shape_plan_t *plan HB_UNUSED, |
216 | hb_font_t *font HB_UNUSED, |
217 | hb_buffer_t *buffer) |
218 | { |
219 | find_syllables_khmer (buffer); |
220 | foreach_syllable (buffer, start, end) |
221 | buffer->unsafe_to_break (start, end); |
222 | } |
223 | |
224 | |
225 | /* Rules from: |
226 | * https://docs.microsoft.com/en-us/typography/script-development/devanagari */ |
227 | |
228 | static void |
229 | reorder_consonant_syllable (const hb_ot_shape_plan_t *plan, |
230 | hb_face_t *face HB_UNUSED, |
231 | hb_buffer_t *buffer, |
232 | unsigned int start, unsigned int end) |
233 | { |
234 | const khmer_shape_plan_t *khmer_plan = (const khmer_shape_plan_t *) plan->data; |
235 | hb_glyph_info_t *info = buffer->info; |
236 | |
237 | /* Setup masks. */ |
238 | { |
239 | /* Post-base */ |
240 | hb_mask_t mask = khmer_plan->mask_array[KHMER_BLWF] | |
241 | khmer_plan->mask_array[KHMER_ABVF] | |
242 | khmer_plan->mask_array[KHMER_PSTF]; |
243 | for (unsigned int i = start + 1; i < end; i++) |
244 | info[i].mask |= mask; |
245 | } |
246 | |
247 | unsigned int num_coengs = 0; |
248 | for (unsigned int i = start + 1; i < end; i++) |
249 | { |
250 | /* """ |
251 | * When a COENG + (Cons | IndV) combination are found (and subscript count |
252 | * is less than two) the character combination is handled according to the |
253 | * subscript type of the character following the COENG. |
254 | * |
255 | * ... |
256 | * |
257 | * Subscript Type 2 - The COENG + RO characters are reordered to immediately |
258 | * before the base glyph. Then the COENG + RO characters are assigned to have |
259 | * the 'pref' OpenType feature applied to them. |
260 | * """ |
261 | */ |
262 | if (info[i].khmer_category() == OT_Coeng && num_coengs <= 2 && i + 1 < end) |
263 | { |
264 | num_coengs++; |
265 | |
266 | if (info[i + 1].khmer_category() == OT_Ra) |
267 | { |
268 | for (unsigned int j = 0; j < 2; j++) |
269 | info[i + j].mask |= khmer_plan->mask_array[KHMER_PREF]; |
270 | |
271 | /* Move the Coeng,Ro sequence to the start. */ |
272 | buffer->merge_clusters (start, i + 2); |
273 | hb_glyph_info_t t0 = info[i]; |
274 | hb_glyph_info_t t1 = info[i + 1]; |
275 | memmove (&info[start + 2], &info[start], (i - start) * sizeof (info[0])); |
276 | info[start] = t0; |
277 | info[start + 1] = t1; |
278 | |
279 | /* Mark the subsequent stuff with 'cfar'. Used in Khmer. |
280 | * Read the feature spec. |
281 | * This allows distinguishing the following cases with MS Khmer fonts: |
282 | * U+1784,U+17D2,U+179A,U+17D2,U+1782 |
283 | * U+1784,U+17D2,U+1782,U+17D2,U+179A |
284 | */ |
285 | if (khmer_plan->mask_array[KHMER_CFAR]) |
286 | for (unsigned int j = i + 2; j < end; j++) |
287 | info[j].mask |= khmer_plan->mask_array[KHMER_CFAR]; |
288 | |
289 | num_coengs = 2; /* Done. */ |
290 | } |
291 | } |
292 | |
293 | /* Reorder left matra piece. */ |
294 | else if (info[i].khmer_category() == OT_VPre) |
295 | { |
296 | /* Move to the start. */ |
297 | buffer->merge_clusters (start, i + 1); |
298 | hb_glyph_info_t t = info[i]; |
299 | memmove (&info[start + 1], &info[start], (i - start) * sizeof (info[0])); |
300 | info[start] = t; |
301 | } |
302 | } |
303 | } |
304 | |
305 | static void |
306 | reorder_syllable_khmer (const hb_ot_shape_plan_t *plan, |
307 | hb_face_t *face, |
308 | hb_buffer_t *buffer, |
309 | unsigned int start, unsigned int end) |
310 | { |
311 | khmer_syllable_type_t syllable_type = (khmer_syllable_type_t) (buffer->info[start].syllable() & 0x0F); |
312 | switch (syllable_type) |
313 | { |
314 | case khmer_broken_cluster: /* We already inserted dotted-circles, so just call the consonant_syllable. */ |
315 | case khmer_consonant_syllable: |
316 | reorder_consonant_syllable (plan, face, buffer, start, end); |
317 | break; |
318 | |
319 | case khmer_non_khmer_cluster: |
320 | break; |
321 | } |
322 | } |
323 | |
324 | static inline void |
325 | insert_dotted_circles_khmer (const hb_ot_shape_plan_t *plan HB_UNUSED, |
326 | hb_font_t *font, |
327 | hb_buffer_t *buffer) |
328 | { |
329 | if (unlikely (buffer->flags & HB_BUFFER_FLAG_DO_NOT_INSERT_DOTTED_CIRCLE)) |
330 | return; |
331 | |
332 | /* Note: This loop is extra overhead, but should not be measurable. |
333 | * TODO Use a buffer scratch flag to remove the loop. */ |
334 | bool has_broken_syllables = false; |
335 | unsigned int count = buffer->len; |
336 | hb_glyph_info_t *info = buffer->info; |
337 | for (unsigned int i = 0; i < count; i++) |
338 | if ((info[i].syllable() & 0x0F) == khmer_broken_cluster) |
339 | { |
340 | has_broken_syllables = true; |
341 | break; |
342 | } |
343 | if (likely (!has_broken_syllables)) |
344 | return; |
345 | |
346 | |
347 | hb_codepoint_t dottedcircle_glyph; |
348 | if (!font->get_nominal_glyph (0x25CCu, &dottedcircle_glyph)) |
349 | return; |
350 | |
351 | hb_glyph_info_t dottedcircle = {0}; |
352 | dottedcircle.codepoint = 0x25CCu; |
353 | set_khmer_properties (dottedcircle); |
354 | dottedcircle.codepoint = dottedcircle_glyph; |
355 | |
356 | buffer->clear_output (); |
357 | |
358 | buffer->idx = 0; |
359 | unsigned int last_syllable = 0; |
360 | while (buffer->idx < buffer->len && buffer->successful) |
361 | { |
362 | unsigned int syllable = buffer->cur().syllable(); |
363 | khmer_syllable_type_t syllable_type = (khmer_syllable_type_t) (syllable & 0x0F); |
364 | if (unlikely (last_syllable != syllable && syllable_type == khmer_broken_cluster)) |
365 | { |
366 | last_syllable = syllable; |
367 | |
368 | hb_glyph_info_t ginfo = dottedcircle; |
369 | ginfo.cluster = buffer->cur().cluster; |
370 | ginfo.mask = buffer->cur().mask; |
371 | ginfo.syllable() = buffer->cur().syllable(); |
372 | |
373 | /* Insert dottedcircle after possible Repha. */ |
374 | while (buffer->idx < buffer->len && buffer->successful && |
375 | last_syllable == buffer->cur().syllable() && |
376 | buffer->cur().khmer_category() == OT_Repha) |
377 | buffer->next_glyph (); |
378 | |
379 | buffer->output_info (ginfo); |
380 | } |
381 | else |
382 | buffer->next_glyph (); |
383 | } |
384 | buffer->swap_buffers (); |
385 | } |
386 | |
387 | static void |
388 | reorder_khmer (const hb_ot_shape_plan_t *plan, |
389 | hb_font_t *font, |
390 | hb_buffer_t *buffer) |
391 | { |
392 | insert_dotted_circles_khmer (plan, font, buffer); |
393 | |
394 | foreach_syllable (buffer, start, end) |
395 | reorder_syllable_khmer (plan, font->face, buffer, start, end); |
396 | |
397 | HB_BUFFER_DEALLOCATE_VAR (buffer, khmer_category); |
398 | } |
399 | |
400 | |
401 | static bool |
402 | decompose_khmer (const hb_ot_shape_normalize_context_t *c, |
403 | hb_codepoint_t ab, |
404 | hb_codepoint_t *a, |
405 | hb_codepoint_t *b) |
406 | { |
407 | switch (ab) |
408 | { |
409 | /* |
410 | * Decompose split matras that don't have Unicode decompositions. |
411 | */ |
412 | |
413 | /* Khmer */ |
414 | case 0x17BEu : *a = 0x17C1u; *b= 0x17BEu; return true; |
415 | case 0x17BFu : *a = 0x17C1u; *b= 0x17BFu; return true; |
416 | case 0x17C0u : *a = 0x17C1u; *b= 0x17C0u; return true; |
417 | case 0x17C4u : *a = 0x17C1u; *b= 0x17C4u; return true; |
418 | case 0x17C5u : *a = 0x17C1u; *b= 0x17C5u; return true; |
419 | } |
420 | |
421 | return (bool) c->unicode->decompose (ab, a, b); |
422 | } |
423 | |
424 | static bool |
425 | compose_khmer (const hb_ot_shape_normalize_context_t *c, |
426 | hb_codepoint_t a, |
427 | hb_codepoint_t b, |
428 | hb_codepoint_t *ab) |
429 | { |
430 | /* Avoid recomposing split matras. */ |
431 | if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a))) |
432 | return false; |
433 | |
434 | return (bool) c->unicode->compose (a, b, ab); |
435 | } |
436 | |
437 | |
438 | const hb_ot_complex_shaper_t _hb_ot_complex_shaper_khmer = |
439 | { |
440 | collect_features_khmer, |
441 | override_features_khmer, |
442 | data_create_khmer, |
443 | data_destroy_khmer, |
444 | nullptr, /* preprocess_text */ |
445 | nullptr, /* postprocess_glyphs */ |
446 | HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT, |
447 | decompose_khmer, |
448 | compose_khmer, |
449 | setup_masks_khmer, |
450 | HB_TAG_NONE, /* gpos_tag */ |
451 | nullptr, /* reorder_marks */ |
452 | HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE, |
453 | false, /* fallback_position */ |
454 | }; |
455 | |
456 | |
457 | #endif |
458 | |