1 /*
2 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
3 *
4 * This code is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License version 2 only, as
6 * published by the Free Software Foundation. Oracle designates this
7 * particular file as subject to the "Classpath" exception as provided
8 * by Oracle in the LICENSE file that accompanied this code.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 */
24
25 // This file is available under and governed by the GNU General Public
26 // License version 2 only, as published by the Free Software Foundation.
27 // However, the following notice accompanied the original version of this
28 // file:
29 //
30 /*
31 * Copyright © 2011,2012 Google, Inc.
32 *
33 * This is part of HarfBuzz, a text shaping library.
34 *
35 * Permission is hereby granted, without written agreement and without
36 * license or royalty fees, to use, copy, modify, and distribute this
37 * software and its documentation for any purpose, provided that the
38 * above copyright notice and the following two paragraphs appear in
39 * all copies of this software.
40 *
41 * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
42 * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
43 * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
44 * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
45 * DAMAGE.
46 *
47 * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
48 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
49 * FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS
50 * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
51 * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
52 *
53 * Google Author(s): Behdad Esfahbod
54 */
55
56 #include "hb-ot-shape-complex-indic-private.hh"
57 #include "hb-ot-layout-private.hh"
58
59 /* buffer var allocations */
60 #define indic_category() complex_var_u8_0() /* indic_category_t */
61 #define indic_position() complex_var_u8_1() /* indic_position_t */
62
63
64 /*
65 * Indic shaper.
66 */
67
68
69 #define IN_HALF_BLOCK(u, Base) (((u) & ~0x7Fu) == (Base))
70
71 #define IS_DEVA(u) (IN_HALF_BLOCK (u, 0x0900u))
72 #define IS_BENG(u) (IN_HALF_BLOCK (u, 0x0980u))
73 #define IS_GURU(u) (IN_HALF_BLOCK (u, 0x0A00u))
74 #define IS_GUJR(u) (IN_HALF_BLOCK (u, 0x0A80u))
75 #define IS_ORYA(u) (IN_HALF_BLOCK (u, 0x0B00u))
76 #define IS_TAML(u) (IN_HALF_BLOCK (u, 0x0B80u))
77 #define IS_TELU(u) (IN_HALF_BLOCK (u, 0x0C00u))
78 #define IS_KNDA(u) (IN_HALF_BLOCK (u, 0x0C80u))
79 #define IS_MLYM(u) (IN_HALF_BLOCK (u, 0x0D00u))
80 #define IS_SINH(u) (IN_HALF_BLOCK (u, 0x0D80u))
81 #define IS_KHMR(u) (IN_HALF_BLOCK (u, 0x1780u))
82
83
84 #define MATRA_POS_LEFT(u) POS_PRE_M
85 #define MATRA_POS_RIGHT(u) ( \
86 IS_DEVA(u) ? POS_AFTER_SUB : \
87 IS_BENG(u) ? POS_AFTER_POST : \
88 IS_GURU(u) ? POS_AFTER_POST : \
89 IS_GUJR(u) ? POS_AFTER_POST : \
90 IS_ORYA(u) ? POS_AFTER_POST : \
91 IS_TAML(u) ? POS_AFTER_POST : \
92 IS_TELU(u) ? (u <= 0x0C42u ? POS_BEFORE_SUB : POS_AFTER_SUB) : \
93 IS_KNDA(u) ? (u < 0x0CC3u || u > 0xCD6u ? POS_BEFORE_SUB : POS_AFTER_SUB) : \
94 IS_MLYM(u) ? POS_AFTER_POST : \
95 IS_SINH(u) ? POS_AFTER_SUB : \
96 IS_KHMR(u) ? POS_AFTER_POST : \
97 /*default*/ POS_AFTER_SUB \
98 )
99 #define MATRA_POS_TOP(u) ( /* BENG and MLYM don't have top matras. */ \
100 IS_DEVA(u) ? POS_AFTER_SUB : \
101 IS_GURU(u) ? POS_AFTER_POST : /* Deviate from spec */ \
102 IS_GUJR(u) ? POS_AFTER_SUB : \
103 IS_ORYA(u) ? POS_AFTER_MAIN : \
104 IS_TAML(u) ? POS_AFTER_SUB : \
105 IS_TELU(u) ? POS_BEFORE_SUB : \
106 IS_KNDA(u) ? POS_BEFORE_SUB : \
107 IS_SINH(u) ? POS_AFTER_SUB : \
108 IS_KHMR(u) ? POS_AFTER_POST : \
109 /*default*/ POS_AFTER_SUB \
110 )
111 #define MATRA_POS_BOTTOM(u) ( \
112 IS_DEVA(u) ? POS_AFTER_SUB : \
113 IS_BENG(u) ? POS_AFTER_SUB : \
114 IS_GURU(u) ? POS_AFTER_POST : \
115 IS_GUJR(u) ? POS_AFTER_POST : \
116 IS_ORYA(u) ? POS_AFTER_SUB : \
117 IS_TAML(u) ? POS_AFTER_POST : \
118 IS_TELU(u) ? POS_BEFORE_SUB : \
119 IS_KNDA(u) ? POS_BEFORE_SUB : \
120 IS_MLYM(u) ? POS_AFTER_POST : \
121 IS_SINH(u) ? POS_AFTER_SUB : \
122 IS_KHMR(u) ? POS_AFTER_POST : \
123 /*default*/ POS_AFTER_SUB \
124 )
125
126 static inline indic_position_t
127 matra_position (hb_codepoint_t u, indic_position_t side)
128 {
129 switch ((int) side)
130 {
131 case POS_PRE_C: return MATRA_POS_LEFT (u);
132 case POS_POST_C: return MATRA_POS_RIGHT (u);
133 case POS_ABOVE_C: return MATRA_POS_TOP (u);
134 case POS_BELOW_C: return MATRA_POS_BOTTOM (u);
135 };
136 return side;
137 }
138
139 /* XXX
140 * This is a hack for now. We should move this data into the main Indic table.
141 * Or completely remove it and just check in the tables.
142 */
143 static const hb_codepoint_t ra_chars[] = {
144 0x0930u, /* Devanagari */
145 0x09B0u, /* Bengali */
146 0x09F0u, /* Bengali */
147 0x0A30u, /* Gurmukhi */ /* No Reph */
148 0x0AB0u, /* Gujarati */
149 0x0B30u, /* Oriya */
150 0x0BB0u, /* Tamil */ /* No Reph */
151 0x0C30u, /* Telugu */ /* Reph formed only with ZWJ */
152 0x0CB0u, /* Kannada */
153 0x0D30u, /* Malayalam */ /* No Reph, Logical Repha */
154
155 0x0DBBu, /* Sinhala */ /* Reph formed only with ZWJ */
156
157 0x179Au, /* Khmer */ /* No Reph, Visual Repha */
158 };
159
160 static inline bool
161 is_ra (hb_codepoint_t u)
162 {
163 for (unsigned int i = 0; i < ARRAY_LENGTH (ra_chars); i++)
164 if (u == ra_chars[i])
165 return true;
166 return false;
167 }
168
169 static inline bool
170 is_one_of (const hb_glyph_info_t &info, unsigned int flags)
171 {
172 /* If it ligated, all bets are off. */
173 if (_hb_glyph_info_ligated (&info)) return false;
174 return !!(FLAG_SAFE (info.indic_category()) & flags);
175 }
176
177 static inline bool
178 is_joiner (const hb_glyph_info_t &info)
179 {
180 return is_one_of (info, JOINER_FLAGS);
181 }
182
183 static inline bool
184 is_consonant (const hb_glyph_info_t &info)
185 {
186 return is_one_of (info, CONSONANT_FLAGS);
187 }
188
189 static inline bool
190 is_halant_or_coeng (const hb_glyph_info_t &info)
191 {
192 return is_one_of (info, HALANT_OR_COENG_FLAGS);
193 }
194
195 static inline void
196 set_indic_properties (hb_glyph_info_t &info)
197 {
198 hb_codepoint_t u = info.codepoint;
199 unsigned int type = hb_indic_get_categories (u);
200 indic_category_t cat = (indic_category_t) (type & 0x7Fu);
201 indic_position_t pos = (indic_position_t) (type >> 8);
202
203
204 /*
205 * Re-assign category
206 */
207
208
209 /* The spec says U+0952 is OT_A. However, testing shows that Uniscribe
210 * treats a whole bunch of characters similarly.
211 * TESTS: For example, for U+0951:
212 * U+092E,U+0947,U+0952
213 * U+092E,U+0952,U+0947
214 * U+092E,U+0947,U+0951
215 * U+092E,U+0951,U+0947
216 * U+092E,U+0951,U+0952
217 * U+092E,U+0952,U+0951
218 */
219 if (unlikely (hb_in_ranges (u, 0x0951u, 0x0952u,
220 0x1CD0u, 0x1CD2u,
221 0x1CD4u, 0x1CE1u) ||
222 u == 0x1CF4u))
223 cat = OT_A;
224 /* The following act more like the Bindus. */
225 else if (unlikely (hb_in_range (u, 0x0953u, 0x0954u)))
226 cat = OT_SM;
227 /* The following act like consonants. */
228 else if (unlikely (hb_in_ranges (u, 0x0A72u, 0x0A73u,
229 0x1CF5u, 0x1CF6u)))
230 cat = OT_C;
231 /* TODO: The following should only be allowed after a Visarga.
232 * For now, just treat them like regular tone marks. */
233 else if (unlikely (hb_in_range (u, 0x1CE2u, 0x1CE8u)))
234 cat = OT_A;
235 /* TODO: The following should only be allowed after some of
236 * the nasalization marks, maybe only for U+1CE9..U+1CF1.
237 * For now, just treat them like tone marks. */
238 else if (unlikely (u == 0x1CEDu))
239 cat = OT_A;
240 /* The following take marks in standalone clusters, similar to Avagraha. */
241 else if (unlikely (hb_in_ranges (u, 0xA8F2u, 0xA8F7u,
242 0x1CE9u, 0x1CECu,
243 0x1CEEu, 0x1CF1u)))
244 {
245 cat = OT_Symbol;
246 ASSERT_STATIC ((int) INDIC_SYLLABIC_CATEGORY_AVAGRAHA == OT_Symbol);
247 }
248 else if (unlikely (hb_in_range (u, 0x17CDu, 0x17D1u) ||
249 u == 0x17CBu || u == 0x17D3u || u == 0x17DDu)) /* Khmer Various signs */
250 {
251 /* These are like Top Matras. */
252 cat = OT_M;
253 pos = POS_ABOVE_C;
254 }
255 else if (unlikely (u == 0x17C6u)) cat = OT_N; /* Khmer Bindu doesn't like to be repositioned. */
256 else if (unlikely (u == 0x17D2u)) cat = OT_Coeng; /* Khmer coeng */
257 else if (unlikely (hb_in_range (u, 0x2010u, 0x2011u)))
258 cat = OT_PLACEHOLDER;
259 else if (unlikely (u == 0x25CCu)) cat = OT_DOTTEDCIRCLE;
260 else if (unlikely (u == 0xA982u)) cat = OT_SM; /* Javanese repha. */
261 else if (unlikely (u == 0xA9BEu)) cat = OT_CM2; /* Javanese medial ya. */
262 else if (unlikely (u == 0xA9BDu)) { cat = OT_M; pos = POS_POST_C; } /* Javanese vocalic r. */
263
264
265 /*
266 * Re-assign position.
267 */
268
269 if ((FLAG_SAFE (cat) & CONSONANT_FLAGS))
270 {
271 pos = POS_BASE_C;
272 if (is_ra (u))
273 cat = OT_Ra;
274 }
275 else if (cat == OT_M)
276 {
277 pos = matra_position (u, pos);
278 }
279 else if ((FLAG_SAFE (cat) & (FLAG (OT_SM) | FLAG (OT_VD) | FLAG (OT_A) | FLAG (OT_Symbol))))
280 {
281 pos = POS_SMVD;
282 }
283
284 if (unlikely (u == 0x0B01u)) pos = POS_BEFORE_SUB; /* Oriya Bindu is BeforeSub in the spec. */
285
286
287
288 info.indic_category() = cat;
289 info.indic_position() = pos;
290 }
291
292 /*
293 * Things above this line should ideally be moved to the Indic table itself.
294 */
295
296
297 /*
298 * Indic configurations. Note that we do not want to keep every single script-specific
299 * behavior in these tables necessarily. This should mainly be used for per-script
300 * properties that are cheaper keeping here, than in the code. Ie. if, say, one and
301 * only one script has an exception, that one script can be if'ed directly in the code,
302 * instead of adding a new flag in these structs.
303 */
304
305 enum base_position_t {
306 BASE_POS_FIRST,
307 BASE_POS_LAST_SINHALA,
308 BASE_POS_LAST
309 };
310 enum reph_position_t {
311 REPH_POS_AFTER_MAIN = POS_AFTER_MAIN,
312 REPH_POS_BEFORE_SUB = POS_BEFORE_SUB,
313 REPH_POS_AFTER_SUB = POS_AFTER_SUB,
314 REPH_POS_BEFORE_POST = POS_BEFORE_POST,
315 REPH_POS_AFTER_POST = POS_AFTER_POST,
316 REPH_POS_DONT_CARE = POS_RA_TO_BECOME_REPH
317 };
318 enum reph_mode_t {
319 REPH_MODE_IMPLICIT, /* Reph formed out of initial Ra,H sequence. */
320 REPH_MODE_EXPLICIT, /* Reph formed out of initial Ra,H,ZWJ sequence. */
321 REPH_MODE_VIS_REPHA, /* Encoded Repha character, no reordering needed. */
322 REPH_MODE_LOG_REPHA /* Encoded Repha character, needs reordering. */
323 };
324 enum blwf_mode_t {
325 BLWF_MODE_PRE_AND_POST, /* Below-forms feature applied to pre-base and post-base. */
326 BLWF_MODE_POST_ONLY /* Below-forms feature applied to post-base only. */
327 };
328 enum pref_len_t {
329 PREF_LEN_1 = 1,
330 PREF_LEN_2 = 2,
331 PREF_LEN_DONT_CARE = PREF_LEN_2
332 };
333 struct indic_config_t
334 {
335 hb_script_t script;
336 bool has_old_spec;
337 hb_codepoint_t virama;
338 base_position_t base_pos;
339 reph_position_t reph_pos;
340 reph_mode_t reph_mode;
341 blwf_mode_t blwf_mode;
342 pref_len_t pref_len;
343 };
344
345 static const indic_config_t indic_configs[] =
346 {
347 /* Default. Should be first. */
348 {HB_SCRIPT_INVALID, false, 0,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_1},
349 {HB_SCRIPT_DEVANAGARI,true, 0x094Du,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
350 {HB_SCRIPT_BENGALI, true, 0x09CDu,BASE_POS_LAST, REPH_POS_AFTER_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
351 {HB_SCRIPT_GURMUKHI, true, 0x0A4Du,BASE_POS_LAST, REPH_POS_BEFORE_SUB, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
352 {HB_SCRIPT_GUJARATI, true, 0x0ACDu,BASE_POS_LAST, REPH_POS_BEFORE_POST,REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
353 {HB_SCRIPT_ORIYA, true, 0x0B4Du,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
354 {HB_SCRIPT_TAMIL, true, 0x0BCDu,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_2},
355 {HB_SCRIPT_TELUGU, true, 0x0C4Du,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_EXPLICIT, BLWF_MODE_POST_ONLY, PREF_LEN_2},
356 {HB_SCRIPT_KANNADA, true, 0x0CCDu,BASE_POS_LAST, REPH_POS_AFTER_POST, REPH_MODE_IMPLICIT, BLWF_MODE_POST_ONLY, PREF_LEN_2},
357 {HB_SCRIPT_MALAYALAM, true, 0x0D4Du,BASE_POS_LAST, REPH_POS_AFTER_MAIN, REPH_MODE_LOG_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_2},
358 {HB_SCRIPT_SINHALA, false,0x0DCAu,BASE_POS_LAST_SINHALA,
359 REPH_POS_AFTER_MAIN, REPH_MODE_EXPLICIT, BLWF_MODE_PRE_AND_POST, PREF_LEN_DONT_CARE},
360 {HB_SCRIPT_KHMER, false,0x17D2u,BASE_POS_FIRST,REPH_POS_DONT_CARE, REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_2},
361 {HB_SCRIPT_JAVANESE, false,0xA9C0u,BASE_POS_FIRST,REPH_POS_DONT_CARE, REPH_MODE_VIS_REPHA,BLWF_MODE_PRE_AND_POST, PREF_LEN_1},
362 };
363
364
365
366 /*
367 * Indic shaper.
368 */
369
370 struct feature_list_t {
371 hb_tag_t tag;
372 hb_ot_map_feature_flags_t flags;
373 };
374
375 static const feature_list_t
376 indic_features[] =
377 {
378 /*
379 * Basic features.
380 * These features are applied in order, one at a time, after initial_reordering.
381 */
382 {HB_TAG('n','u','k','t'), F_GLOBAL},
383 {HB_TAG('a','k','h','n'), F_GLOBAL},
384 {HB_TAG('r','p','h','f'), F_NONE},
385 {HB_TAG('r','k','r','f'), F_GLOBAL},
386 {HB_TAG('p','r','e','f'), F_NONE},
387 {HB_TAG('b','l','w','f'), F_NONE},
388 {HB_TAG('a','b','v','f'), F_NONE},
389 {HB_TAG('h','a','l','f'), F_NONE},
390 {HB_TAG('p','s','t','f'), F_NONE},
391 {HB_TAG('v','a','t','u'), F_GLOBAL},
392 {HB_TAG('c','j','c','t'), F_GLOBAL},
393 {HB_TAG('c','f','a','r'), F_NONE},
394 /*
395 * Other features.
396 * These features are applied all at once, after final_reordering.
397 * Default Bengali font in Windows for example has intermixed
398 * lookups for init,pres,abvs,blws features.
399 */
400 {HB_TAG('i','n','i','t'), F_NONE},
401 {HB_TAG('p','r','e','s'), F_GLOBAL},
402 {HB_TAG('a','b','v','s'), F_GLOBAL},
403 {HB_TAG('b','l','w','s'), F_GLOBAL},
404 {HB_TAG('p','s','t','s'), F_GLOBAL},
405 {HB_TAG('h','a','l','n'), F_GLOBAL},
406 /* Positioning features, though we don't care about the types. */
407 {HB_TAG('d','i','s','t'), F_GLOBAL},
408 {HB_TAG('a','b','v','m'), F_GLOBAL},
409 {HB_TAG('b','l','w','m'), F_GLOBAL},
410 };
411
412 /*
413 * Must be in the same order as the indic_features array.
414 */
415 enum {
416 _NUKT,
417 _AKHN,
418 RPHF,
419 _RKRF,
420 PREF,
421 BLWF,
422 ABVF,
423 HALF,
424 PSTF,
425 _VATU,
426 _CJCT,
427 CFAR,
428
429 INIT,
430 _PRES,
431 _ABVS,
432 _BLWS,
433 _PSTS,
434 _HALN,
435 _DIST,
436 _ABVM,
437 _BLWM,
438
439 INDIC_NUM_FEATURES,
440 INDIC_BASIC_FEATURES = INIT /* Don't forget to update this! */
441 };
442
443 static void
444 setup_syllables (const hb_ot_shape_plan_t *plan,
445 hb_font_t *font,
446 hb_buffer_t *buffer);
447 static void
448 initial_reordering (const hb_ot_shape_plan_t *plan,
449 hb_font_t *font,
450 hb_buffer_t *buffer);
451 static void
452 final_reordering (const hb_ot_shape_plan_t *plan,
453 hb_font_t *font,
454 hb_buffer_t *buffer);
455 static void
456 clear_syllables (const hb_ot_shape_plan_t *plan,
457 hb_font_t *font,
458 hb_buffer_t *buffer);
459
460 static void
461 collect_features_indic (hb_ot_shape_planner_t *plan)
462 {
463 hb_ot_map_builder_t *map = &plan->map;
464
465 /* Do this before any lookups have been applied. */
466 map->add_gsub_pause (setup_syllables);
467
468 map->add_global_bool_feature (HB_TAG('l','o','c','l'));
469 /* The Indic specs do not require ccmp, but we apply it here since if
470 * there is a use of it, it's typically at the beginning. */
471 map->add_global_bool_feature (HB_TAG('c','c','m','p'));
472
473
474 unsigned int i = 0;
475 map->add_gsub_pause (initial_reordering);
476 for (; i < INDIC_BASIC_FEATURES; i++) {
477 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ);
478 map->add_gsub_pause (NULL);
479 }
480 map->add_gsub_pause (final_reordering);
481 for (; i < INDIC_NUM_FEATURES; i++) {
482 map->add_feature (indic_features[i].tag, 1, indic_features[i].flags | F_MANUAL_ZWJ);
483 }
484
485 map->add_global_bool_feature (HB_TAG('c','a','l','t'));
486 map->add_global_bool_feature (HB_TAG('c','l','i','g'));
487
488 map->add_gsub_pause (clear_syllables);
489 }
490
491 static void
492 override_features_indic (hb_ot_shape_planner_t *plan)
493 {
494 /* Uniscribe does not apply 'kern' in Khmer. */
495 if (hb_options ().uniscribe_bug_compatible)
496 {
497 switch ((hb_tag_t) plan->props.script)
498 {
499 case HB_SCRIPT_KHMER:
500 plan->map.add_feature (HB_TAG('k','e','r','n'), 0, F_GLOBAL);
501 break;
502 }
503 }
504
505 plan->map.add_feature (HB_TAG('l','i','g','a'), 0, F_GLOBAL);
506 }
507
508
509 struct would_substitute_feature_t
510 {
511 inline void init (const hb_ot_map_t *map, hb_tag_t feature_tag, bool zero_context_)
512 {
513 zero_context = zero_context_;
514 map->get_stage_lookups (0/*GSUB*/,
515 map->get_feature_stage (0/*GSUB*/, feature_tag),
516 &lookups, &count);
517 }
518
519 inline bool would_substitute (const hb_codepoint_t *glyphs,
520 unsigned int glyphs_count,
521 hb_face_t *face) const
522 {
523 for (unsigned int i = 0; i < count; i++)
524 if (hb_ot_layout_lookup_would_substitute_fast (face, lookups[i].index, glyphs, glyphs_count, zero_context))
525 return true;
526 return false;
527 }
528
529 private:
530 const hb_ot_map_t::lookup_map_t *lookups;
531 unsigned int count;
532 bool zero_context;
533 };
534
535 struct indic_shape_plan_t
536 {
537 ASSERT_POD ();
538
539 inline bool get_virama_glyph (hb_font_t *font, hb_codepoint_t *pglyph) const
540 {
541 hb_codepoint_t glyph = virama_glyph;
542 if (unlikely (virama_glyph == (hb_codepoint_t) -1))
543 {
544 if (!config->virama || !font->get_glyph (config->virama, 0, &glyph))
545 glyph = 0;
546 /* Technically speaking, the spec says we should apply 'locl' to virama too.
547 * Maybe one day... */
548
549 /* Our get_glyph() function needs a font, so we can't get the virama glyph
550 * during shape planning... Instead, overwrite it here. It's safe. Don't worry! */
551 (const_cast<indic_shape_plan_t *> (this))->virama_glyph = glyph;
552 }
553
554 *pglyph = glyph;
555 return glyph != 0;
556 }
557
558 const indic_config_t *config;
559
560 bool is_old_spec;
561 hb_codepoint_t virama_glyph;
562
563 would_substitute_feature_t rphf;
564 would_substitute_feature_t pref;
565 would_substitute_feature_t blwf;
566 would_substitute_feature_t pstf;
567
568 hb_mask_t mask_array[INDIC_NUM_FEATURES];
569 };
570
571 static void *
572 data_create_indic (const hb_ot_shape_plan_t *plan)
573 {
574 indic_shape_plan_t *indic_plan = (indic_shape_plan_t *) calloc (1, sizeof (indic_shape_plan_t));
575 if (unlikely (!indic_plan))
576 return NULL;
577
578 indic_plan->config = &indic_configs[0];
579 for (unsigned int i = 1; i < ARRAY_LENGTH (indic_configs); i++)
580 if (plan->props.script == indic_configs[i].script) {
581 indic_plan->config = &indic_configs[i];
582 break;
583 }
584
585 indic_plan->is_old_spec = indic_plan->config->has_old_spec && ((plan->map.chosen_script[0] & 0x000000FFu) != '2');
586 indic_plan->virama_glyph = (hb_codepoint_t) -1;
587
588 /* Use zero-context would_substitute() matching for new-spec of the main
589 * Indic scripts, and scripts with one spec only, but not for old-specs. */
590 bool zero_context = !indic_plan->is_old_spec;
591 indic_plan->rphf.init (&plan->map, HB_TAG('r','p','h','f'), zero_context);
592 indic_plan->pref.init (&plan->map, HB_TAG('p','r','e','f'), zero_context);
593 indic_plan->blwf.init (&plan->map, HB_TAG('b','l','w','f'), zero_context);
594 indic_plan->pstf.init (&plan->map, HB_TAG('p','s','t','f'), zero_context);
595
596 for (unsigned int i = 0; i < ARRAY_LENGTH (indic_plan->mask_array); i++)
597 indic_plan->mask_array[i] = (indic_features[i].flags & F_GLOBAL) ?
598 0 : plan->map.get_1_mask (indic_features[i].tag);
599
600 return indic_plan;
601 }
602
603 static void
604 data_destroy_indic (void *data)
605 {
606 free (data);
607 }
608
609 static indic_position_t
610 consonant_position_from_face (const indic_shape_plan_t *indic_plan,
611 const hb_codepoint_t consonant,
612 const hb_codepoint_t virama,
613 hb_face_t *face)
614 {
615 /* For old-spec, the order of glyphs is Consonant,Virama,
616 * whereas for new-spec, it's Virama,Consonant. However,
617 * some broken fonts (like Free Sans) simply copied lookups
618 * from old-spec to new-spec without modification.
619 * And oddly enough, Uniscribe seems to respect those lookups.
620 * Eg. in the sequence U+0924,U+094D,U+0930, Uniscribe finds
621 * base at 0. The font however, only has lookups matching
622 * 930,94D in 'blwf', not the expected 94D,930 (with new-spec
623 * table). As such, we simply match both sequences. Seems
624 * to work. */
625 hb_codepoint_t glyphs[3] = {virama, consonant, virama};
626 if (indic_plan->blwf.would_substitute (glyphs , 2, face) ||
627 indic_plan->blwf.would_substitute (glyphs+1, 2, face))
628 return POS_BELOW_C;
629 if (indic_plan->pstf.would_substitute (glyphs , 2, face) ||
630 indic_plan->pstf.would_substitute (glyphs+1, 2, face))
631 return POS_POST_C;
632 unsigned int pref_len = indic_plan->config->pref_len;
633 if ((pref_len == PREF_LEN_2 &&
634 (indic_plan->pref.would_substitute (glyphs , 2, face) ||
635 indic_plan->pref.would_substitute (glyphs+1, 2, face)))
636 || (pref_len == PREF_LEN_1 &&
637 indic_plan->pref.would_substitute (glyphs+1, 1, face)))
638 return POS_POST_C;
639 return POS_BASE_C;
640 }
641
642
643 enum syllable_type_t {
644 consonant_syllable,
645 vowel_syllable,
646 standalone_cluster,
647 symbol_cluster,
648 broken_cluster,
649 non_indic_cluster,
650 };
651
652 #include "hb-ot-shape-complex-indic-machine.hh"
653
654
655 static void
656 setup_masks_indic (const hb_ot_shape_plan_t *plan HB_UNUSED,
657 hb_buffer_t *buffer,
658 hb_font_t *font HB_UNUSED)
659 {
660 HB_BUFFER_ALLOCATE_VAR (buffer, indic_category);
661 HB_BUFFER_ALLOCATE_VAR (buffer, indic_position);
662
663 /* We cannot setup masks here. We save information about characters
664 * and setup masks later on in a pause-callback. */
665
666 unsigned int count = buffer->len;
667 hb_glyph_info_t *info = buffer->info;
668 for (unsigned int i = 0; i < count; i++)
669 set_indic_properties (info[i]);
670 }
671
672 static void
673 setup_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
674 hb_font_t *font HB_UNUSED,
675 hb_buffer_t *buffer)
676 {
677 find_syllables (buffer);
678 }
679
680 static int
681 compare_indic_order (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb)
682 {
683 int a = pa->indic_position();
684 int b = pb->indic_position();
685
686 return a < b ? -1 : a == b ? 0 : +1;
687 }
688
689
690
691 static void
692 update_consonant_positions (const hb_ot_shape_plan_t *plan,
693 hb_font_t *font,
694 hb_buffer_t *buffer)
695 {
696 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
697
698 if (indic_plan->config->base_pos != BASE_POS_LAST)
699 return;
700
701 hb_codepoint_t virama;
702 if (indic_plan->get_virama_glyph (font, &virama))
703 {
704 hb_face_t *face = font->face;
705 unsigned int count = buffer->len;
706 hb_glyph_info_t *info = buffer->info;
707 for (unsigned int i = 0; i < count; i++)
708 if (info[i].indic_position() == POS_BASE_C)
709 {
710 hb_codepoint_t consonant = info[i].codepoint;
711 info[i].indic_position() = consonant_position_from_face (indic_plan, consonant, virama, face);
712 }
713 }
714 }
715
716
717 /* Rules from:
718 * https://www.microsoft.com/typography/otfntdev/devanot/shaping.aspx */
719
720 static void
721 initial_reordering_consonant_syllable (const hb_ot_shape_plan_t *plan,
722 hb_face_t *face,
723 hb_buffer_t *buffer,
724 unsigned int start, unsigned int end)
725 {
726 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
727 hb_glyph_info_t *info = buffer->info;
728
729
730 /* 1. Find base consonant:
731 *
732 * The shaping engine finds the base consonant of the syllable, using the
733 * following algorithm: starting from the end of the syllable, move backwards
734 * until a consonant is found that does not have a below-base or post-base
735 * form (post-base forms have to follow below-base forms), or that is not a
736 * pre-base reordering Ra, or arrive at the first consonant. The consonant
737 * stopped at will be the base.
738 *
739 * o If the syllable starts with Ra + Halant (in a script that has Reph)
740 * and has more than one consonant, Ra is excluded from candidates for
741 * base consonants.
742 */
743
744 unsigned int base = end;
745 bool has_reph = false;
746
747 {
748 /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
749 * and has more than one consonant, Ra is excluded from candidates for
750 * base consonants. */
751 unsigned int limit = start;
752 if (indic_plan->config->reph_pos != REPH_POS_DONT_CARE &&
753 indic_plan->mask_array[RPHF] &&
754 start + 3 <= end &&
755 (
756 (indic_plan->config->reph_mode == REPH_MODE_IMPLICIT && !is_joiner (info[start + 2])) ||
757 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT && info[start + 2].indic_category() == OT_ZWJ)
758 ))
759 {
760 /* See if it matches the 'rphf' feature. */
761 hb_codepoint_t glyphs[3] = {info[start].codepoint,
762 info[start + 1].codepoint,
763 indic_plan->config->reph_mode == REPH_MODE_EXPLICIT ?
764 info[start + 2].codepoint : 0};
765 if (indic_plan->rphf.would_substitute (glyphs, 2, face) ||
766 (indic_plan->config->reph_mode == REPH_MODE_EXPLICIT &&
767 indic_plan->rphf.would_substitute (glyphs, 3, face)))
768 {
769 limit += 2;
770 while (limit < end && is_joiner (info[limit]))
771 limit++;
772 base = start;
773 has_reph = true;
774 }
775 } else if (indic_plan->config->reph_mode == REPH_MODE_LOG_REPHA && info[start].indic_category() == OT_Repha)
776 {
777 limit += 1;
778 while (limit < end && is_joiner (info[limit]))
779 limit++;
780 base = start;
781 has_reph = true;
782 }
783
784 switch (indic_plan->config->base_pos)
785 {
786 default:
787 assert (false);
788 /* fallthrough */
789
790 case BASE_POS_LAST:
791 {
792 /* -> starting from the end of the syllable, move backwards */
793 unsigned int i = end;
794 bool seen_below = false;
795 do {
796 i--;
797 /* -> until a consonant is found */
798 if (is_consonant (info[i]))
799 {
800 /* -> that does not have a below-base or post-base form
801 * (post-base forms have to follow below-base forms), */
802 if (info[i].indic_position() != POS_BELOW_C &&
803 (info[i].indic_position() != POS_POST_C || seen_below))
804 {
805 base = i;
806 break;
807 }
808 if (info[i].indic_position() == POS_BELOW_C)
809 seen_below = true;
810
811 /* -> or that is not a pre-base reordering Ra,
812 *
813 * IMPLEMENTATION NOTES:
814 *
815 * Our pre-base reordering Ra's are marked POS_POST_C, so will be skipped
816 * by the logic above already.
817 */
818
819 /* -> or arrive at the first consonant. The consonant stopped at will
820 * be the base. */
821 base = i;
822 }
823 else
824 {
825 /* A ZWJ after a Halant stops the base search, and requests an explicit
826 * half form.
827 * A ZWJ before a Halant, requests a subjoined form instead, and hence
828 * search continues. This is particularly important for Bengali
829 * sequence Ra,H,Ya that should form Ya-Phalaa by subjoining Ya. */
830 if (start < i &&
831 info[i].indic_category() == OT_ZWJ &&
832 info[i - 1].indic_category() == OT_H)
833 break;
834 }
835 } while (i > limit);
836 }
837 break;
838
839 case BASE_POS_LAST_SINHALA:
840 {
841 /* Sinhala base positioning is slightly different from main Indic, in that:
842 * 1. Its ZWJ behavior is different,
843 * 2. We don't need to look into the font for consonant positions.
844 */
845
846 if (!has_reph)
847 base = limit;
848
849 /* Find the last base consonant that is not blocked by ZWJ. If there is
850 * a ZWJ right before a base consonant, that would request a subjoined form. */
851 for (unsigned int i = limit; i < end; i++)
852 if (is_consonant (info[i]))
853 {
854 if (limit < i && info[i - 1].indic_category() == OT_ZWJ)
855 break;
856 else
857 base = i;
858 }
859
860 /* Mark all subsequent consonants as below. */
861 for (unsigned int i = base + 1; i < end; i++)
862 if (is_consonant (info[i]))
863 info[i].indic_position() = POS_BELOW_C;
864 }
865 break;
866
867 case BASE_POS_FIRST:
868 {
869 /* The first consonant is always the base. */
870
871 assert (indic_plan->config->reph_mode == REPH_MODE_VIS_REPHA);
872 assert (!has_reph);
873
874 base = start;
875
876 /* Mark all subsequent consonants as below. */
877 for (unsigned int i = base + 1; i < end; i++)
878 if (is_consonant (info[i]))
879 info[i].indic_position() = POS_BELOW_C;
880 }
881 break;
882 }
883
884 /* -> If the syllable starts with Ra + Halant (in a script that has Reph)
885 * and has more than one consonant, Ra is excluded from candidates for
886 * base consonants.
887 *
888 * Only do this for unforced Reph. (ie. not for Ra,H,ZWJ. */
889 if (has_reph && base == start && limit - base <= 2) {
890 /* Have no other consonant, so Reph is not formed and Ra becomes base. */
891 has_reph = false;
892 }
893 }
894
895
896 /* 2. Decompose and reorder Matras:
897 *
898 * Each matra and any syllable modifier sign in the cluster are moved to the
899 * appropriate position relative to the consonant(s) in the cluster. The
900 * shaping engine decomposes two- or three-part matras into their constituent
901 * parts before any repositioning. Matra characters are classified by which
902 * consonant in a conjunct they have affinity for and are reordered to the
903 * following positions:
904 *
905 * o Before first half form in the syllable
906 * o After subjoined consonants
907 * o After post-form consonant
908 * o After main consonant (for above marks)
909 *
910 * IMPLEMENTATION NOTES:
911 *
912 * The normalize() routine has already decomposed matras for us, so we don't
913 * need to worry about that.
914 */
915
916
917 /* 3. Reorder marks to canonical order:
918 *
919 * Adjacent nukta and halant or nukta and vedic sign are always repositioned
920 * if necessary, so that the nukta is first.
921 *
922 * IMPLEMENTATION NOTES:
923 *
924 * We don't need to do this: the normalize() routine already did this for us.
925 */
926
927
928 /* Reorder characters */
929
930 for (unsigned int i = start; i < base; i++)
931 info[i].indic_position() = MIN (POS_PRE_C, (indic_position_t) info[i].indic_position());
932
933 if (base < end)
934 info[base].indic_position() = POS_BASE_C;
935
936 /* Mark final consonants. A final consonant is one appearing after a matra,
937 * like in Khmer. */
938 for (unsigned int i = base + 1; i < end; i++)
939 if (info[i].indic_category() == OT_M) {
940 for (unsigned int j = i + 1; j < end; j++)
941 if (is_consonant (info[j])) {
942 info[j].indic_position() = POS_FINAL_C;
943 break;
944 }
945 break;
946 }
947
948 /* Handle beginning Ra */
949 if (has_reph)
950 info[start].indic_position() = POS_RA_TO_BECOME_REPH;
951
952 /* For old-style Indic script tags, move the first post-base Halant after
953 * last consonant.
954 *
955 * Reports suggest that in some scripts Uniscribe does this only if there
956 * is *not* a Halant after last consonant already (eg. Kannada), while it
957 * does it unconditionally in other scripts (eg. Malayalam). We don't
958 * currently know about other scripts, so we single out Malayalam for now.
959 *
960 * Kannada test case:
961 * U+0C9A,U+0CCD,U+0C9A,U+0CCD
962 * With some versions of Lohit Kannada.
963 * https://bugs.freedesktop.org/show_bug.cgi?id=59118
964 *
965 * Malayalam test case:
966 * U+0D38,U+0D4D,U+0D31,U+0D4D,U+0D31,U+0D4D
967 * With lohit-ttf-20121122/Lohit-Malayalam.ttf
968 */
969 if (indic_plan->is_old_spec)
970 {
971 bool disallow_double_halants = buffer->props.script != HB_SCRIPT_MALAYALAM;
972 for (unsigned int i = base + 1; i < end; i++)
973 if (info[i].indic_category() == OT_H)
974 {
975 unsigned int j;
976 for (j = end - 1; j > i; j--)
977 if (is_consonant (info[j]) ||
978 (disallow_double_halants && info[j].indic_category() == OT_H))
979 break;
980 if (info[j].indic_category() != OT_H && j > i) {
981 /* Move Halant to after last consonant. */
982 hb_glyph_info_t t = info[i];
983 memmove (&info[i], &info[i + 1], (j - i) * sizeof (info[0]));
984 info[j] = t;
985 }
986 break;
987 }
988 }
989
990 /* Attach misc marks to previous char to move with them. */
991 {
992 indic_position_t last_pos = POS_START;
993 for (unsigned int i = start; i < end; i++)
994 {
995 if ((FLAG_SAFE (info[i].indic_category()) & (JOINER_FLAGS | FLAG (OT_N) | FLAG (OT_RS) | MEDIAL_FLAGS | HALANT_OR_COENG_FLAGS)))
996 {
997 info[i].indic_position() = last_pos;
998 if (unlikely (info[i].indic_category() == OT_H &&
999 info[i].indic_position() == POS_PRE_M))
1000 {
1001 /*
1002 * Uniscribe doesn't move the Halant with Left Matra.
1003 * TEST: U+092B,U+093F,U+094DE
1004 * We follow. This is important for the Sinhala
1005 * U+0DDA split matra since it decomposes to U+0DD9,U+0DCA
1006 * where U+0DD9 is a left matra and U+0DCA is the virama.
1007 * We don't want to move the virama with the left matra.
1008 * TEST: U+0D9A,U+0DDA
1009 */
1010 for (unsigned int j = i; j > start; j--)
1011 if (info[j - 1].indic_position() != POS_PRE_M) {
1012 info[i].indic_position() = info[j - 1].indic_position();
1013 break;
1014 }
1015 }
1016 } else if (info[i].indic_position() != POS_SMVD) {
1017 last_pos = (indic_position_t) info[i].indic_position();
1018 }
1019 }
1020 }
1021 /* For post-base consonants let them own anything before them
1022 * since the last consonant or matra. */
1023 {
1024 unsigned int last = base;
1025 for (unsigned int i = base + 1; i < end; i++)
1026 if (is_consonant (info[i]))
1027 {
1028 for (unsigned int j = last + 1; j < i; j++)
1029 if (info[j].indic_position() < POS_SMVD)
1030 info[j].indic_position() = info[i].indic_position();
1031 last = i;
1032 } else if (info[i].indic_category() == OT_M)
1033 last = i;
1034 }
1035
1036
1037 {
1038 /* Use syllable() for sort accounting temporarily. */
1039 unsigned int syllable = info[start].syllable();
1040 for (unsigned int i = start; i < end; i++)
1041 info[i].syllable() = i - start;
1042
1043 /* Sit tight, rock 'n roll! */
1044 hb_stable_sort (info + start, end - start, compare_indic_order);
1045 /* Find base again */
1046 base = end;
1047 for (unsigned int i = start; i < end; i++)
1048 if (info[i].indic_position() == POS_BASE_C)
1049 {
1050 base = i;
1051 break;
1052 }
1053 /* Things are out-of-control for post base positions, they may shuffle
1054 * around like crazy. In old-spec mode, we move halants around, so in
1055 * that case merge all clusters after base. Otherwise, check the sort
1056 * order and merge as needed.
1057 * For pre-base stuff, we handle cluster issues in final reordering.
1058 *
1059 * We could use buffer->sort() for this, if there was no special
1060 * reordering of pre-base stuff happening later...
1061 */
1062 if (indic_plan->is_old_spec || end - base > 127)
1063 buffer->merge_clusters (base, end);
1064 else
1065 {
1066 /* Note! syllable() is a one-byte field. */
1067 for (unsigned int i = base; i < end; i++)
1068 if (info[i].syllable() != 255)
1069 {
1070 unsigned int max = i;
1071 unsigned int j = start + info[i].syllable();
1072 while (j != i)
1073 {
1074 max = MAX (max, j);
1075 unsigned int next = start + info[j].syllable();
1076 info[j].syllable() = 255; /* So we don't process j later again. */
1077 j = next;
1078 }
1079 if (i != max)
1080 buffer->merge_clusters (i, max + 1);
1081 }
1082 }
1083
1084 /* Put syllable back in. */
1085 for (unsigned int i = start; i < end; i++)
1086 info[i].syllable() = syllable;
1087 }
1088
1089 /* Setup masks now */
1090
1091 {
1092 hb_mask_t mask;
1093
1094 /* Reph */
1095 for (unsigned int i = start; i < end && info[i].indic_position() == POS_RA_TO_BECOME_REPH; i++)
1096 info[i].mask |= indic_plan->mask_array[RPHF];
1097
1098 /* Pre-base */
1099 mask = indic_plan->mask_array[HALF];
1100 if (!indic_plan->is_old_spec &&
1101 indic_plan->config->blwf_mode == BLWF_MODE_PRE_AND_POST)
1102 mask |= indic_plan->mask_array[BLWF];
1103 for (unsigned int i = start; i < base; i++)
1104 info[i].mask |= mask;
1105 /* Base */
1106 mask = 0;
1107 if (base < end)
1108 info[base].mask |= mask;
1109 /* Post-base */
1110 mask = indic_plan->mask_array[BLWF] | indic_plan->mask_array[ABVF] | indic_plan->mask_array[PSTF];
1111 for (unsigned int i = base + 1; i < end; i++)
1112 info[i].mask |= mask;
1113 }
1114
1115 if (indic_plan->is_old_spec &&
1116 buffer->props.script == HB_SCRIPT_DEVANAGARI)
1117 {
1118 /* Old-spec eye-lash Ra needs special handling. From the
1119 * spec:
1120 *
1121 * "The feature 'below-base form' is applied to consonants
1122 * having below-base forms and following the base consonant.
1123 * The exception is vattu, which may appear below half forms
1124 * as well as below the base glyph. The feature 'below-base
1125 * form' will be applied to all such occurrences of Ra as well."
1126 *
1127 * Test case: U+0924,U+094D,U+0930,U+094d,U+0915
1128 * with Sanskrit 2003 font.
1129 *
1130 * However, note that Ra,Halant,ZWJ is the correct way to
1131 * request eyelash form of Ra, so we wouldbn't inhibit it
1132 * in that sequence.
1133 *
1134 * Test case: U+0924,U+094D,U+0930,U+094d,U+200D,U+0915
1135 */
1136 for (unsigned int i = start; i + 1 < base; i++)
1137 if (info[i ].indic_category() == OT_Ra &&
1138 info[i+1].indic_category() == OT_H &&
1139 (i + 2 == base ||
1140 info[i+2].indic_category() != OT_ZWJ))
1141 {
1142 info[i ].mask |= indic_plan->mask_array[BLWF];
1143 info[i+1].mask |= indic_plan->mask_array[BLWF];
1144 }
1145 }
1146
1147 unsigned int pref_len = indic_plan->config->pref_len;
1148 if (indic_plan->mask_array[PREF] && base + pref_len < end)
1149 {
1150 assert (1 <= pref_len && pref_len <= 2);
1151 /* Find a Halant,Ra sequence and mark it for pre-base reordering processing. */
1152 for (unsigned int i = base + 1; i + pref_len - 1 < end; i++) {
1153 hb_codepoint_t glyphs[2];
1154 for (unsigned int j = 0; j < pref_len; j++)
1155 glyphs[j] = info[i + j].codepoint;
1156 if (indic_plan->pref.would_substitute (glyphs, pref_len, face))
1157 {
1158 for (unsigned int j = 0; j < pref_len; j++)
1159 info[i++].mask |= indic_plan->mask_array[PREF];
1160
1161 /* Mark the subsequent stuff with 'cfar'. Used in Khmer.
1162 * Read the feature spec.
1163 * This allows distinguishing the following cases with MS Khmer fonts:
1164 * U+1784,U+17D2,U+179A,U+17D2,U+1782
1165 * U+1784,U+17D2,U+1782,U+17D2,U+179A
1166 */
1167 if (indic_plan->mask_array[CFAR])
1168 for (; i < end; i++)
1169 info[i].mask |= indic_plan->mask_array[CFAR];
1170
1171 break;
1172 }
1173 }
1174 }
1175
1176 /* Apply ZWJ/ZWNJ effects */
1177 for (unsigned int i = start + 1; i < end; i++)
1178 if (is_joiner (info[i])) {
1179 bool non_joiner = info[i].indic_category() == OT_ZWNJ;
1180 unsigned int j = i;
1181
1182 do {
1183 j--;
1184
1185 /* ZWJ/ZWNJ should disable CJCT. They do that by simply
1186 * being there, since we don't skip them for the CJCT
1187 * feature (ie. F_MANUAL_ZWJ) */
1188
1189 /* A ZWNJ disables HALF. */
1190 if (non_joiner)
1191 info[j].mask &= ~indic_plan->mask_array[HALF];
1192
1193 } while (j > start && !is_consonant (info[j]));
1194 }
1195 }
1196
1197 static void
1198 initial_reordering_standalone_cluster (const hb_ot_shape_plan_t *plan,
1199 hb_face_t *face,
1200 hb_buffer_t *buffer,
1201 unsigned int start, unsigned int end)
1202 {
1203 /* We treat placeholder/dotted-circle as if they are consonants, so we
1204 * should just chain. Only if not in compatibility mode that is... */
1205
1206 if (hb_options ().uniscribe_bug_compatible)
1207 {
1208 /* For dotted-circle, this is what Uniscribe does:
1209 * If dotted-circle is the last glyph, it just does nothing.
1210 * Ie. It doesn't form Reph. */
1211 if (buffer->info[end - 1].indic_category() == OT_DOTTEDCIRCLE)
1212 return;
1213 }
1214
1215 initial_reordering_consonant_syllable (plan, face, buffer, start, end);
1216 }
1217
1218 static void
1219 initial_reordering_syllable (const hb_ot_shape_plan_t *plan,
1220 hb_face_t *face,
1221 hb_buffer_t *buffer,
1222 unsigned int start, unsigned int end)
1223 {
1224 syllable_type_t syllable_type = (syllable_type_t) (buffer->info[start].syllable() & 0x0F);
1225 switch (syllable_type)
1226 {
1227 case vowel_syllable: /* We made the vowels look like consonants. So let's call the consonant logic! */
1228 case consonant_syllable:
1229 initial_reordering_consonant_syllable (plan, face, buffer, start, end);
1230 break;
1231
1232 case broken_cluster: /* We already inserted dotted-circles, so just call the standalone_cluster. */
1233 case standalone_cluster:
1234 initial_reordering_standalone_cluster (plan, face, buffer, start, end);
1235 break;
1236
1237 case symbol_cluster:
1238 case non_indic_cluster:
1239 break;
1240 }
1241 }
1242
1243 static inline void
1244 insert_dotted_circles (const hb_ot_shape_plan_t *plan HB_UNUSED,
1245 hb_font_t *font,
1246 hb_buffer_t *buffer)
1247 {
1248 /* Note: This loop is extra overhead, but should not be measurable. */
1249 bool has_broken_syllables = false;
1250 unsigned int count = buffer->len;
1251 hb_glyph_info_t *info = buffer->info;
1252 for (unsigned int i = 0; i < count; i++)
1253 if ((info[i].syllable() & 0x0F) == broken_cluster)
1254 {
1255 has_broken_syllables = true;
1256 break;
1257 }
1258 if (likely (!has_broken_syllables))
1259 return;
1260
1261
1262 hb_codepoint_t dottedcircle_glyph;
1263 if (!font->get_glyph (0x25CCu, 0, &dottedcircle_glyph))
1264 return;
1265
1266 hb_glyph_info_t dottedcircle = {0};
1267 dottedcircle.codepoint = 0x25CCu;
1268 set_indic_properties (dottedcircle);
1269 dottedcircle.codepoint = dottedcircle_glyph;
1270
1271 buffer->clear_output ();
1272
1273 buffer->idx = 0;
1274 unsigned int last_syllable = 0;
1275 while (buffer->idx < buffer->len)
1276 {
1277 unsigned int syllable = buffer->cur().syllable();
1278 syllable_type_t syllable_type = (syllable_type_t) (syllable & 0x0F);
1279 if (unlikely (last_syllable != syllable && syllable_type == broken_cluster))
1280 {
1281 last_syllable = syllable;
1282
1283 hb_glyph_info_t info = dottedcircle;
1284 info.cluster = buffer->cur().cluster;
1285 info.mask = buffer->cur().mask;
1286 info.syllable() = buffer->cur().syllable();
1287 /* TODO Set glyph_props? */
1288
1289 /* Insert dottedcircle after possible Repha. */
1290 while (buffer->idx < buffer->len &&
1291 last_syllable == buffer->cur().syllable() &&
1292 buffer->cur().indic_category() == OT_Repha)
1293 buffer->next_glyph ();
1294
1295 buffer->output_info (info);
1296 }
1297 else
1298 buffer->next_glyph ();
1299 }
1300
1301 buffer->swap_buffers ();
1302 }
1303
1304 static void
1305 initial_reordering (const hb_ot_shape_plan_t *plan,
1306 hb_font_t *font,
1307 hb_buffer_t *buffer)
1308 {
1309 update_consonant_positions (plan, font, buffer);
1310 insert_dotted_circles (plan, font, buffer);
1311
1312 foreach_syllable (buffer, start, end)
1313 initial_reordering_syllable (plan, font->face, buffer, start, end);
1314 }
1315
1316 static void
1317 final_reordering_syllable (const hb_ot_shape_plan_t *plan,
1318 hb_buffer_t *buffer,
1319 unsigned int start, unsigned int end)
1320 {
1321 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) plan->data;
1322 hb_glyph_info_t *info = buffer->info;
1323
1324
1325 /* This function relies heavily on halant glyphs. Lots of ligation
1326 * and possibly multiplication substitutions happened prior to this
1327 * phase, and that might have messed up our properties. Recover
1328 * from a particular case of that where we're fairly sure that a
1329 * class of OT_H is desired but has been lost. */
1330 if (indic_plan->virama_glyph)
1331 {
1332 unsigned int virama_glyph = indic_plan->virama_glyph;
1333 for (unsigned int i = start; i < end; i++)
1334 if (info[i].codepoint == virama_glyph &&
1335 _hb_glyph_info_ligated (&info[i]) &&
1336 _hb_glyph_info_multiplied (&info[i]))
1337 {
1338 /* This will make sure that this glyph passes is_halant_or_coeng() test. */
1339 info[i].indic_category() = OT_H;
1340 _hb_glyph_info_clear_ligated_and_multiplied (&info[i]);
1341 }
1342 }
1343
1344
1345 /* 4. Final reordering:
1346 *
1347 * After the localized forms and basic shaping forms GSUB features have been
1348 * applied (see below), the shaping engine performs some final glyph
1349 * reordering before applying all the remaining font features to the entire
1350 * cluster.
1351 */
1352
1353 bool try_pref = !!indic_plan->mask_array[PREF];
1354
1355 /* Find base again */
1356 unsigned int base;
1357 for (base = start; base < end; base++)
1358 if (info[base].indic_position() >= POS_BASE_C)
1359 {
1360 if (try_pref && base + 1 < end && indic_plan->config->pref_len == 2)
1361 {
1362 for (unsigned int i = base + 1; i < end; i++)
1363 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0)
1364 {
1365 if (!(_hb_glyph_info_substituted (&info[i]) &&
1366 _hb_glyph_info_ligated_and_didnt_multiply (&info[i])))
1367 {
1368 /* Ok, this was a 'pref' candidate but didn't form any.
1369 * Base is around here... */
1370 base = i;
1371 while (base < end && is_halant_or_coeng (info[base]))
1372 base++;
1373 info[base].indic_position() = POS_BASE_C;
1374
1375 try_pref = false;
1376 }
1377 break;
1378 }
1379 }
1380
1381 if (start < base && info[base].indic_position() > POS_BASE_C)
1382 base--;
1383 break;
1384 }
1385 if (base == end && start < base &&
1386 is_one_of (info[base - 1], FLAG (OT_ZWJ)))
1387 base--;
1388 if (base < end)
1389 while (start < base &&
1390 is_one_of (info[base], (FLAG (OT_N) | HALANT_OR_COENG_FLAGS)))
1391 base--;
1392
1393
1394 /* o Reorder matras:
1395 *
1396 * If a pre-base matra character had been reordered before applying basic
1397 * features, the glyph can be moved closer to the main consonant based on
1398 * whether half-forms had been formed. Actual position for the matra is
1399 * defined as “after last standalone halant glyph, after initial matra
1400 * position and before the main consonant”. If ZWJ or ZWNJ follow this
1401 * halant, position is moved after it.
1402 */
1403
1404 if (start + 1 < end && start < base) /* Otherwise there can't be any pre-base matra characters. */
1405 {
1406 /* If we lost track of base, alas, position before last thingy. */
1407 unsigned int new_pos = base == end ? base - 2 : base - 1;
1408
1409 /* Malayalam / Tamil do not have "half" forms or explicit virama forms.
1410 * The glyphs formed by 'half' are Chillus or ligated explicit viramas.
1411 * We want to position matra after them.
1412 */
1413 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL)
1414 {
1415 while (new_pos > start &&
1416 !(is_one_of (info[new_pos], (FLAG (OT_M) | HALANT_OR_COENG_FLAGS))))
1417 new_pos--;
1418
1419 /* If we found no Halant we are done.
1420 * Otherwise only proceed if the Halant does
1421 * not belong to the Matra itself! */
1422 if (is_halant_or_coeng (info[new_pos]) &&
1423 info[new_pos].indic_position() != POS_PRE_M)
1424 {
1425 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */
1426 if (new_pos + 1 < end && is_joiner (info[new_pos + 1]))
1427 new_pos++;
1428 }
1429 else
1430 new_pos = start; /* No move. */
1431 }
1432
1433 if (start < new_pos && info[new_pos].indic_position () != POS_PRE_M)
1434 {
1435 /* Now go see if there's actually any matras... */
1436 for (unsigned int i = new_pos; i > start; i--)
1437 if (info[i - 1].indic_position () == POS_PRE_M)
1438 {
1439 unsigned int old_pos = i - 1;
1440 if (old_pos < base && base <= new_pos) /* Shouldn't actually happen. */
1441 base--;
1442
1443 hb_glyph_info_t tmp = info[old_pos];
1444 memmove (&info[old_pos], &info[old_pos + 1], (new_pos - old_pos) * sizeof (info[0]));
1445 info[new_pos] = tmp;
1446
1447 /* Note: this merge_clusters() is intentionally *after* the reordering.
1448 * Indic matra reordering is special and tricky... */
1449 buffer->merge_clusters (new_pos, MIN (end, base + 1));
1450
1451 new_pos--;
1452 }
1453 } else {
1454 for (unsigned int i = start; i < base; i++)
1455 if (info[i].indic_position () == POS_PRE_M) {
1456 buffer->merge_clusters (i, MIN (end, base + 1));
1457 break;
1458 }
1459 }
1460 }
1461
1462
1463 /* o Reorder reph:
1464 *
1465 * Reph’s original position is always at the beginning of the syllable,
1466 * (i.e. it is not reordered at the character reordering stage). However,
1467 * it will be reordered according to the basic-forms shaping results.
1468 * Possible positions for reph, depending on the script, are; after main,
1469 * before post-base consonant forms, and after post-base consonant forms.
1470 */
1471
1472 /* Two cases:
1473 *
1474 * - If repha is encoded as a sequence of characters (Ra,H or Ra,H,ZWJ), then
1475 * we should only move it if the sequence ligated to the repha form.
1476 *
1477 * - If repha is encoded separately and in the logical position, we should only
1478 * move it if it did NOT ligate. If it ligated, it's probably the font trying
1479 * to make it work without the reordering.
1480 */
1481 if (start + 1 < end &&
1482 info[start].indic_position() == POS_RA_TO_BECOME_REPH &&
1483 ((info[start].indic_category() == OT_Repha) ^
1484 _hb_glyph_info_ligated_and_didnt_multiply (&info[start])))
1485 {
1486 unsigned int new_reph_pos;
1487 reph_position_t reph_pos = indic_plan->config->reph_pos;
1488
1489 assert (reph_pos != REPH_POS_DONT_CARE);
1490
1491 /* 1. If reph should be positioned after post-base consonant forms,
1492 * proceed to step 5.
1493 */
1494 if (reph_pos == REPH_POS_AFTER_POST)
1495 {
1496 goto reph_step_5;
1497 }
1498
1499 /* 2. If the reph repositioning class is not after post-base: target
1500 * position is after the first explicit halant glyph between the
1501 * first post-reph consonant and last main consonant. If ZWJ or ZWNJ
1502 * are following this halant, position is moved after it. If such
1503 * position is found, this is the target position. Otherwise,
1504 * proceed to the next step.
1505 *
1506 * Note: in old-implementation fonts, where classifications were
1507 * fixed in shaping engine, there was no case where reph position
1508 * will be found on this step.
1509 */
1510 {
1511 new_reph_pos = start + 1;
1512 while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos]))
1513 new_reph_pos++;
1514
1515 if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos]))
1516 {
1517 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */
1518 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1]))
1519 new_reph_pos++;
1520 goto reph_move;
1521 }
1522 }
1523
1524 /* 3. If reph should be repositioned after the main consonant: find the
1525 * first consonant not ligated with main, or find the first
1526 * consonant that is not a potential pre-base reordering Ra.
1527 */
1528 if (reph_pos == REPH_POS_AFTER_MAIN)
1529 {
1530 new_reph_pos = base;
1531 while (new_reph_pos + 1 < end && info[new_reph_pos + 1].indic_position() <= POS_AFTER_MAIN)
1532 new_reph_pos++;
1533 if (new_reph_pos < end)
1534 goto reph_move;
1535 }
1536
1537 /* 4. If reph should be positioned before post-base consonant, find
1538 * first post-base classified consonant not ligated with main. If no
1539 * consonant is found, the target position should be before the
1540 * first matra, syllable modifier sign or vedic sign.
1541 */
1542 /* This is our take on what step 4 is trying to say (and failing, BADLY). */
1543 if (reph_pos == REPH_POS_AFTER_SUB)
1544 {
1545 new_reph_pos = base;
1546 while (new_reph_pos < end &&
1547 !( FLAG_SAFE (info[new_reph_pos + 1].indic_position()) & (FLAG (POS_POST_C) | FLAG (POS_AFTER_POST) | FLAG (POS_SMVD))))
1548 new_reph_pos++;
1549 if (new_reph_pos < end)
1550 goto reph_move;
1551 }
1552
1553 /* 5. If no consonant is found in steps 3 or 4, move reph to a position
1554 * immediately before the first post-base matra, syllable modifier
1555 * sign or vedic sign that has a reordering class after the intended
1556 * reph position. For example, if the reordering position for reph
1557 * is post-main, it will skip above-base matras that also have a
1558 * post-main position.
1559 */
1560 reph_step_5:
1561 {
1562 /* Copied from step 2. */
1563 new_reph_pos = start + 1;
1564 while (new_reph_pos < base && !is_halant_or_coeng (info[new_reph_pos]))
1565 new_reph_pos++;
1566
1567 if (new_reph_pos < base && is_halant_or_coeng (info[new_reph_pos]))
1568 {
1569 /* ->If ZWJ or ZWNJ are following this halant, position is moved after it. */
1570 if (new_reph_pos + 1 < base && is_joiner (info[new_reph_pos + 1]))
1571 new_reph_pos++;
1572 goto reph_move;
1573 }
1574 }
1575
1576 /* 6. Otherwise, reorder reph to the end of the syllable.
1577 */
1578 {
1579 new_reph_pos = end - 1;
1580 while (new_reph_pos > start && info[new_reph_pos].indic_position() == POS_SMVD)
1581 new_reph_pos--;
1582
1583 /*
1584 * If the Reph is to be ending up after a Matra,Halant sequence,
1585 * position it before that Halant so it can interact with the Matra.
1586 * However, if it's a plain Consonant,Halant we shouldn't do that.
1587 * Uniscribe doesn't do this.
1588 * TEST: U+0930,U+094D,U+0915,U+094B,U+094D
1589 */
1590 if (!hb_options ().uniscribe_bug_compatible &&
1591 unlikely (is_halant_or_coeng (info[new_reph_pos]))) {
1592 for (unsigned int i = base + 1; i < new_reph_pos; i++)
1593 if (info[i].indic_category() == OT_M) {
1594 /* Ok, got it. */
1595 new_reph_pos--;
1596 }
1597 }
1598 goto reph_move;
1599 }
1600
1601 reph_move:
1602 {
1603 /* Move */
1604 buffer->merge_clusters (start, new_reph_pos + 1);
1605 hb_glyph_info_t reph = info[start];
1606 memmove (&info[start], &info[start + 1], (new_reph_pos - start) * sizeof (info[0]));
1607 info[new_reph_pos] = reph;
1608
1609 if (start < base && base <= new_reph_pos)
1610 base--;
1611 }
1612 }
1613
1614
1615 /* o Reorder pre-base reordering consonants:
1616 *
1617 * If a pre-base reordering consonant is found, reorder it according to
1618 * the following rules:
1619 */
1620
1621 if (try_pref && base + 1 < end) /* Otherwise there can't be any pre-base reordering Ra. */
1622 {
1623 unsigned int pref_len = indic_plan->config->pref_len;
1624 for (unsigned int i = base + 1; i < end; i++)
1625 if ((info[i].mask & indic_plan->mask_array[PREF]) != 0)
1626 {
1627 /* 1. Only reorder a glyph produced by substitution during application
1628 * of the <pref> feature. (Note that a font may shape a Ra consonant with
1629 * the feature generally but block it in certain contexts.)
1630 */
1631 /* Note: We just check that something got substituted. We don't check that
1632 * the <pref> feature actually did it...
1633 *
1634 * If pref len is longer than one, then only reorder if it ligated. If
1635 * pref len is one, only reorder if it didn't ligate with other things. */
1636 if (_hb_glyph_info_substituted (&info[i]) &&
1637 ((pref_len == 1) ^ _hb_glyph_info_ligated_and_didnt_multiply (&info[i])))
1638 {
1639 /*
1640 * 2. Try to find a target position the same way as for pre-base matra.
1641 * If it is found, reorder pre-base consonant glyph.
1642 *
1643 * 3. If position is not found, reorder immediately before main
1644 * consonant.
1645 */
1646
1647 unsigned int new_pos = base;
1648 /* Malayalam / Tamil do not have "half" forms or explicit virama forms.
1649 * The glyphs formed by 'half' are Chillus or ligated explicit viramas.
1650 * We want to position matra after them.
1651 */
1652 if (buffer->props.script != HB_SCRIPT_MALAYALAM && buffer->props.script != HB_SCRIPT_TAMIL)
1653 {
1654 while (new_pos > start &&
1655 !(is_one_of (info[new_pos - 1], FLAG(OT_M) | HALANT_OR_COENG_FLAGS)))
1656 new_pos--;
1657
1658 /* In Khmer coeng model, a H,Ra can go *after* matras. If it goes after a
1659 * split matra, it should be reordered to *before* the left part of such matra. */
1660 if (new_pos > start && info[new_pos - 1].indic_category() == OT_M)
1661 {
1662 unsigned int old_pos = i;
1663 for (unsigned int i = base + 1; i < old_pos; i++)
1664 if (info[i].indic_category() == OT_M)
1665 {
1666 new_pos--;
1667 break;
1668 }
1669 }
1670 }
1671
1672 if (new_pos > start && is_halant_or_coeng (info[new_pos - 1]))
1673 {
1674 /* -> If ZWJ or ZWNJ follow this halant, position is moved after it. */
1675 if (new_pos < end && is_joiner (info[new_pos]))
1676 new_pos++;
1677 }
1678
1679 {
1680 unsigned int old_pos = i;
1681
1682 buffer->merge_clusters (new_pos, old_pos + 1);
1683 hb_glyph_info_t tmp = info[old_pos];
1684 memmove (&info[new_pos + 1], &info[new_pos], (old_pos - new_pos) * sizeof (info[0]));
1685 info[new_pos] = tmp;
1686
1687 if (new_pos <= base && base < old_pos)
1688 base++;
1689 }
1690 }
1691
1692 break;
1693 }
1694 }
1695
1696
1697 /* Apply 'init' to the Left Matra if it's a word start. */
1698 if (info[start].indic_position () == POS_PRE_M &&
1699 (!start ||
1700 !(FLAG_SAFE (_hb_glyph_info_get_general_category (&info[start - 1])) &
1701 FLAG_RANGE (HB_UNICODE_GENERAL_CATEGORY_FORMAT, HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK))))
1702 info[start].mask |= indic_plan->mask_array[INIT];
1703
1704
1705 /*
1706 * Finish off the clusters and go home!
1707 */
1708 if (hb_options ().uniscribe_bug_compatible)
1709 {
1710 switch ((hb_tag_t) plan->props.script)
1711 {
1712 case HB_SCRIPT_TAMIL:
1713 case HB_SCRIPT_SINHALA:
1714 break;
1715
1716 default:
1717 /* Uniscribe merges the entire cluster... Except for Tamil & Sinhala.
1718 * This means, half forms are submerged into the main consonants cluster.
1719 * This is unnecessary, and makes cursor positioning harder, but that's what
1720 * Uniscribe does. */
1721 buffer->merge_clusters (start, end);
1722 break;
1723 }
1724 }
1725 }
1726
1727
1728 static void
1729 final_reordering (const hb_ot_shape_plan_t *plan,
1730 hb_font_t *font HB_UNUSED,
1731 hb_buffer_t *buffer)
1732 {
1733 unsigned int count = buffer->len;
1734 if (unlikely (!count)) return;
1735
1736 foreach_syllable (buffer, start, end)
1737 final_reordering_syllable (plan, buffer, start, end);
1738
1739 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_category);
1740 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_position);
1741 }
1742
1743
1744 static void
1745 clear_syllables (const hb_ot_shape_plan_t *plan HB_UNUSED,
1746 hb_font_t *font HB_UNUSED,
1747 hb_buffer_t *buffer)
1748 {
1749 hb_glyph_info_t *info = buffer->info;
1750 unsigned int count = buffer->len;
1751 for (unsigned int i = 0; i < count; i++)
1752 info[i].syllable() = 0;
1753 }
1754
1755
1756 static bool
1757 decompose_indic (const hb_ot_shape_normalize_context_t *c,
1758 hb_codepoint_t ab,
1759 hb_codepoint_t *a,
1760 hb_codepoint_t *b)
1761 {
1762 switch (ab)
1763 {
1764 /* Don't decompose these. */
1765 case 0x0931u : return false;
1766 case 0x0B94u : return false;
1767
1768
1769 /*
1770 * Decompose split matras that don't have Unicode decompositions.
1771 */
1772
1773 case 0x0F77u : *a = 0x0FB2u; *b= 0x0F81u; return true;
1774 case 0x0F79u : *a = 0x0FB3u; *b= 0x0F81u; return true;
1775 case 0x17BEu : *a = 0x17C1u; *b= 0x17BEu; return true;
1776 case 0x17BFu : *a = 0x17C1u; *b= 0x17BFu; return true;
1777 case 0x17C0u : *a = 0x17C1u; *b= 0x17C0u; return true;
1778 case 0x17C4u : *a = 0x17C1u; *b= 0x17C4u; return true;
1779 case 0x17C5u : *a = 0x17C1u; *b= 0x17C5u; return true;
1780 case 0x1925u : *a = 0x1920u; *b= 0x1923u; return true;
1781 case 0x1926u : *a = 0x1920u; *b= 0x1924u; return true;
1782 case 0x1B3Cu : *a = 0x1B42u; *b= 0x1B3Cu; return true;
1783 case 0x1112Eu : *a = 0x11127u; *b= 0x11131u; return true;
1784 case 0x1112Fu : *a = 0x11127u; *b= 0x11132u; return true;
1785 #if 0
1786 /* This one has no decomposition in Unicode, but needs no decomposition either. */
1787 /* case 0x0AC9u : return false; */
1788 case 0x0B57u : *a = no decomp, -> RIGHT; return true;
1789 case 0x1C29u : *a = no decomp, -> LEFT; return true;
1790 case 0xA9C0u : *a = no decomp, -> RIGHT; return true;
1791 case 0x111BuF : *a = no decomp, -> ABOVE; return true;
1792 #endif
1793 }
1794
1795 if ((ab == 0x0DDAu || hb_in_range (ab, 0x0DDCu, 0x0DDEu)))
1796 {
1797 /*
1798 * Sinhala split matras... Let the fun begin.
1799 *
1800 * These four characters have Unicode decompositions. However, Uniscribe
1801 * decomposes them "Khmer-style", that is, it uses the character itself to
1802 * get the second half. The first half of all four decompositions is always
1803 * U+0DD9.
1804 *
1805 * Now, there are buggy fonts, namely, the widely used lklug.ttf, that are
1806 * broken with Uniscribe. But we need to support them. As such, we only
1807 * do the Uniscribe-style decomposition if the character is transformed into
1808 * its "sec.half" form by the 'pstf' feature. Otherwise, we fall back to
1809 * Unicode decomposition.
1810 *
1811 * Note that we can't unconditionally use Unicode decomposition. That would
1812 * break some other fonts, that are designed to work with Uniscribe, and
1813 * don't have positioning features for the Unicode-style decomposition.
1814 *
1815 * Argh...
1816 *
1817 * The Uniscribe behavior is now documented in the newly published Sinhala
1818 * spec in 2012:
1819 *
1820 * http://www.microsoft.com/typography/OpenTypeDev/sinhala/intro.htm#shaping
1821 */
1822
1823 const indic_shape_plan_t *indic_plan = (const indic_shape_plan_t *) c->plan->data;
1824
1825 hb_codepoint_t glyph;
1826
1827 if (hb_options ().uniscribe_bug_compatible ||
1828 (c->font->get_glyph (ab, 0, &glyph) &&
1829 indic_plan->pstf.would_substitute (&glyph, 1, c->font->face)))
1830 {
1831 /* Ok, safe to use Uniscribe-style decomposition. */
1832 *a = 0x0DD9u;
1833 *b = ab;
1834 return true;
1835 }
1836 }
1837
1838 return c->unicode->decompose (ab, a, b);
1839 }
1840
1841 static bool
1842 compose_indic (const hb_ot_shape_normalize_context_t *c,
1843 hb_codepoint_t a,
1844 hb_codepoint_t b,
1845 hb_codepoint_t *ab)
1846 {
1847 /* Avoid recomposing split matras. */
1848 if (HB_UNICODE_GENERAL_CATEGORY_IS_MARK (c->unicode->general_category (a)))
1849 return false;
1850
1851 /* Composition-exclusion exceptions that we want to recompose. */
1852 if (a == 0x09AFu && b == 0x09BCu) { *ab = 0x09DFu; return true; }
1853
1854 return c->unicode->compose (a, b, ab);
1855 }
1856
1857
1858 const hb_ot_complex_shaper_t _hb_ot_complex_shaper_indic =
1859 {
1860 "indic",
1861 collect_features_indic,
1862 override_features_indic,
1863 data_create_indic,
1864 data_destroy_indic,
1865 NULL, /* preprocess_text */
1866 HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT,
1867 decompose_indic,
1868 compose_indic,
1869 setup_masks_indic,
1870 HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE,
1871 false, /* fallback_position */
1872 };