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
39static const hb_ot_map_feature_t
40khmer_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 */
64enum {
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
80static void
81setup_syllables_khmer (const hb_ot_shape_plan_t *plan,
82 hb_font_t *font,
83 hb_buffer_t *buffer);
84static void
85reorder_khmer (const hb_ot_shape_plan_t *plan,
86 hb_font_t *font,
87 hb_buffer_t *buffer);
88
89static void
90collect_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
121static void
122override_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
141struct 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_indic_would_substitute_feature_t pref;
165
166 hb_mask_t mask_array[KHMER_NUM_FEATURES];
167};
168
169static void *
170data_create_khmer (const hb_ot_shape_plan_t *plan)
171{
172 khmer_shape_plan_t *khmer_plan = (khmer_shape_plan_t *) calloc (1, sizeof (khmer_shape_plan_t));
173 if (unlikely (!khmer_plan))
174 return nullptr;
175
176 khmer_plan->virama_glyph = (hb_codepoint_t) -1;
177
178 khmer_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), true);
179
180 for (unsigned int i = 0; i < ARRAY_LENGTH (khmer_plan->mask_array); i++)
181 khmer_plan->mask_array[i] = (khmer_features[i].flags & F_GLOBAL) ?
182 0 : plan->map.get_1_mask (khmer_features[i].tag);
183
184 return khmer_plan;
185}
186
187static void
188data_destroy_khmer (void *data)
189{
190 free (data);
191}
192
193
194enum khmer_syllable_type_t {
195 khmer_consonant_syllable,
196 khmer_broken_cluster,
197 khmer_non_khmer_cluster,
198};
199
200#include "hb-ot-shape-complex-khmer-machine.hh"
201
202static void
203setup_masks_khmer (const hb_ot_shape_plan_t *plan HB_UNUSED,
204 hb_buffer_t *buffer,
205 hb_font_t *font HB_UNUSED)
206{
207 HB_BUFFER_ALLOCATE_VAR (buffer, khmer_category);
208
209 /* We cannot setup masks here. We save information about characters
210 * and setup masks later on in a pause-callback. */
211
212 unsigned int count = buffer->len;
213 hb_glyph_info_t *info = buffer->info;
214 for (unsigned int i = 0; i < count; i++)
215 set_khmer_properties (info[i]);
216}
217
218static void
219setup_syllables_khmer (const hb_ot_shape_plan_t *plan HB_UNUSED,
220 hb_font_t *font HB_UNUSED,
221 hb_buffer_t *buffer)
222{
223 find_syllables_khmer (buffer);
224 foreach_syllable (buffer, start, end)
225 buffer->unsafe_to_break (start, end);
226}
227
228
229/* Rules from:
230 * https://docs.microsoft.com/en-us/typography/script-development/devanagari */
231
232static void
233reorder_consonant_syllable (const hb_ot_shape_plan_t *plan,
234 hb_face_t *face HB_UNUSED,
235 hb_buffer_t *buffer,
236 unsigned int start, unsigned int end)
237{
238 const khmer_shape_plan_t *khmer_plan = (const khmer_shape_plan_t *) plan->data;
239 hb_glyph_info_t *info = buffer->info;
240
241 /* Setup masks. */
242 {
243 /* Post-base */
244 hb_mask_t mask = khmer_plan->mask_array[KHMER_BLWF] |
245 khmer_plan->mask_array[KHMER_ABVF] |
246 khmer_plan->mask_array[KHMER_PSTF];
247 for (unsigned int i = start + 1; i < end; i++)
248 info[i].mask |= mask;
249 }
250
251 unsigned int num_coengs = 0;
252 for (unsigned int i = start + 1; i < end; i++)
253 {
254 /* """
255 * When a COENG + (Cons | IndV) combination are found (and subscript count
256 * is less than two) the character combination is handled according to the
257 * subscript type of the character following the COENG.
258 *
259 * ...
260 *
261 * Subscript Type 2 - The COENG + RO characters are reordered to immediately
262 * before the base glyph. Then the COENG + RO characters are assigned to have
263 * the 'pref' OpenType feature applied to them.
264 * """
265 */
266 if (info[i].khmer_category() == OT_Coeng && num_coengs <= 2 && i + 1 < end)
267 {
268 num_coengs++;
269
270 if (info[i + 1].khmer_category() == OT_Ra)
271 {
272 for (unsigned int j = 0; j < 2; j++)
273 info[i + j].mask |= khmer_plan->mask_array[KHMER_PREF];
274
275 /* Move the Coeng,Ro sequence to the start. */
276 buffer->merge_clusters (start, i + 2);
277 hb_glyph_info_t t0 = info[i];
278 hb_glyph_info_t t1 = info[i + 1];
279 memmove (&info[start + 2], &info[start], (i - start) * sizeof (info[0]));
280 info[start] = t0;
281 info[start + 1] = t1;
282
283 /* Mark the subsequent stuff with 'cfar'. Used in Khmer.
284 * Read the feature spec.
285 * This allows distinguishing the following cases with MS Khmer fonts:
286 * U+1784,U+17D2,U+179A,U+17D2,U+1782
287 * U+1784,U+17D2,U+1782,U+17D2,U+179A
288 */
289 if (khmer_plan->mask_array[KHMER_CFAR])
290 for (unsigned int j = i + 2; j < end; j++)
291 info[j].mask |= khmer_plan->mask_array[KHMER_CFAR];
292
293 num_coengs = 2; /* Done. */
294 }
295 }
296
297 /* Reorder left matra piece. */
298 else if (info[i].khmer_category() == OT_VPre)
299 {
300 /* Move to the start. */
301 buffer->merge_clusters (start, i + 1);
302 hb_glyph_info_t t = info[i];
303 memmove (&info[start + 1], &info[start], (i - start) * sizeof (info[0]));
304 info[start] = t;
305 }
306 }
307}
308
309static void
310reorder_syllable_khmer (const hb_ot_shape_plan_t *plan,
311 hb_face_t *face,
312 hb_buffer_t *buffer,
313 unsigned int start, unsigned int end)
314{
315 khmer_syllable_type_t syllable_type = (khmer_syllable_type_t) (buffer->info[start].syllable() & 0x0F);
316 switch (syllable_type)
317 {
318 case khmer_broken_cluster: /* We already inserted dotted-circles, so just call the consonant_syllable. */
319 case khmer_consonant_syllable:
320 reorder_consonant_syllable (plan, face, buffer, start, end);
321 break;
322
323 case khmer_non_khmer_cluster:
324 break;
325 }
326}
327
328static inline void
329insert_dotted_circles_khmer (const hb_ot_shape_plan_t *plan HB_UNUSED,
330 hb_font_t *font,
331 hb_buffer_t *buffer)
332{
333 if (unlikely (buffer->flags & HB_BUFFER_FLAG_DO_NOT_INSERT_DOTTED_CIRCLE))
334 return;
335
336 /* Note: This loop is extra overhead, but should not be measurable.
337 * TODO Use a buffer scratch flag to remove the loop. */
338 bool has_broken_syllables = false;
339 unsigned int count = buffer->len;
340 hb_glyph_info_t *info = buffer->info;
341 for (unsigned int i = 0; i < count; i++)
342 if ((info[i].syllable() & 0x0F) == khmer_broken_cluster)
343 {
344 has_broken_syllables = true;
345 break;
346 }
347 if (likely (!has_broken_syllables))
348 return;
349
350
351 hb_codepoint_t dottedcircle_glyph;
352 if (!font->get_nominal_glyph (0x25CCu, &dottedcircle_glyph))
353 return;
354
355 hb_glyph_info_t dottedcircle = {0};
356 dottedcircle.codepoint = 0x25CCu;
357 set_khmer_properties (dottedcircle);
358 dottedcircle.codepoint = dottedcircle_glyph;
359
360 buffer->clear_output ();
361
362 buffer->idx = 0;
363 unsigned int last_syllable = 0;
364 while (buffer->idx < buffer->len && buffer->successful)
365 {
366 unsigned int syllable = buffer->cur().syllable();
367 khmer_syllable_type_t syllable_type = (khmer_syllable_type_t) (syllable & 0x0F);
368 if (unlikely (last_syllable != syllable && syllable_type == khmer_broken_cluster))
369 {
370 last_syllable = syllable;
371
372 hb_glyph_info_t ginfo = dottedcircle;
373 ginfo.cluster = buffer->cur().cluster;
374 ginfo.mask = buffer->cur().mask;
375 ginfo.syllable() = buffer->cur().syllable();
376
377 /* Insert dottedcircle after possible Repha. */
378 while (buffer->idx < buffer->len && buffer->successful &&
379 last_syllable == buffer->cur().syllable() &&
380 buffer->cur().khmer_category() == OT_Repha)
381 buffer->next_glyph ();
382
383 buffer->output_info (ginfo);
384 }
385 else
386 buffer->next_glyph ();
387 }
388 buffer->swap_buffers ();
389}
390
391static void
392reorder_khmer (const hb_ot_shape_plan_t *plan,
393 hb_font_t *font,
394 hb_buffer_t *buffer)
395{
396 insert_dotted_circles_khmer (plan, font, buffer);
397
398 foreach_syllable (buffer, start, end)
399 reorder_syllable_khmer (plan, font->face, buffer, start, end);
400
401 HB_BUFFER_DEALLOCATE_VAR (buffer, khmer_category);
402}
403
404
405static bool
406decompose_khmer (const hb_ot_shape_normalize_context_t *c,
407 hb_codepoint_t ab,
408 hb_codepoint_t *a,
409 hb_codepoint_t *b)
410{
411 switch (ab)
412 {
413 /*
414 * Decompose split matras that don't have Unicode decompositions.
415 */
416
417 /* Khmer */
418 case 0x17BEu : *a = 0x17C1u; *b= 0x17BEu; return true;
419 case 0x17BFu : *a = 0x17C1u; *b= 0x17BFu; return true;
420 case 0x17C0u : *a = 0x17C1u; *b= 0x17C0u; return true;
421 case 0x17C4u : *a = 0x17C1u; *b= 0x17C4u; return true;
422 case 0x17C5u : *a = 0x17C1u; *b= 0x17C5u; return true;
423 }
424
425 return (bool) c->unicode->decompose (ab, a, b);
426}
427
428static bool
429compose_khmer (const hb_ot_shape_normalize_context_t *c,
430 hb_codepoint_t a,
431 hb_codepoint_t b,
432 hb_codepoint_t *ab)
433{
434 /* Avoid recomposing split matras. */
435 if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a)))
436 return false;
437
438 return (bool) c->unicode->compose (a, b, ab);
439}
440
441
442const hb_ot_complex_shaper_t _hb_ot_complex_shaper_khmer =
443{
444 collect_features_khmer,
445 override_features_khmer,
446 data_create_khmer,
447 data_destroy_khmer,
448 nullptr, /* preprocess_text */
449 nullptr, /* postprocess_glyphs */
450 HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT,
451 decompose_khmer,
452 compose_khmer,
453 setup_masks_khmer,
454 HB_TAG_NONE, /* gpos_tag */
455 nullptr, /* reorder_marks */
456 HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE,
457 false, /* fallback_position */
458};
459
460
461#endif
462