2 * Copyright © 2011 Google, Inc.
4 * This is part of HarfBuzz, a text shaping library.
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.
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
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.
24 * Google Author(s): Behdad Esfahbod
27 #include "hb-ot-shape-complex-private.hh"
32 /* buffer var allocations */
33 #define indic_category() complex_var_persistent_u8_0() /* indic_category_t */
34 #define indic_position() complex_var_persistent_u8_1() /* indic_matra_category_t */
36 #define INDIC_TABLE_ELEMENT_TYPE uint8_t
38 /* Cateories used in the OpenType spec:
39 * https://www.microsoft.com/typography/otfntdev/devanot/shaping.aspx
41 /* Note: This enum is duplicated in the -machine.rl source file.
42 * Not sure how to avoid duplication. */
43 enum indic_category_t {
46 OT_Ra, /* Not explicitly listed in the OT spec, but used in the grammar. */
59 /* Visual positions in a syllable from left to right. */
60 enum indic_position_t {
68 /* Categories used in IndicSyllabicCategory.txt from UCD */
69 /* The assignments are guesswork */
70 enum indic_syllabic_category_t {
71 INDIC_SYLLABIC_CATEGORY_OTHER = OT_X,
73 INDIC_SYLLABIC_CATEGORY_AVAGRAHA = OT_X,
74 INDIC_SYLLABIC_CATEGORY_BINDU = OT_SM,
75 INDIC_SYLLABIC_CATEGORY_CONSONANT = OT_C,
76 INDIC_SYLLABIC_CATEGORY_CONSONANT_DEAD = OT_C,
77 INDIC_SYLLABIC_CATEGORY_CONSONANT_FINAL = OT_C,
78 INDIC_SYLLABIC_CATEGORY_CONSONANT_HEAD_LETTER = OT_C,
79 INDIC_SYLLABIC_CATEGORY_CONSONANT_MEDIAL = OT_C,
80 INDIC_SYLLABIC_CATEGORY_CONSONANT_PLACEHOLDER = OT_NBSP,
81 INDIC_SYLLABIC_CATEGORY_CONSONANT_SUBJOINED = OT_C,
82 INDIC_SYLLABIC_CATEGORY_CONSONANT_REPHA = OT_C,
83 INDIC_SYLLABIC_CATEGORY_MODIFYING_LETTER = OT_X,
84 INDIC_SYLLABIC_CATEGORY_NUKTA = OT_N,
85 INDIC_SYLLABIC_CATEGORY_REGISTER_SHIFTER = OT_X,
86 INDIC_SYLLABIC_CATEGORY_TONE_LETTER = OT_X,
87 INDIC_SYLLABIC_CATEGORY_TONE_MARK = OT_X,
88 INDIC_SYLLABIC_CATEGORY_VIRAMA = OT_H,
89 INDIC_SYLLABIC_CATEGORY_VISARGA = OT_SM,
90 INDIC_SYLLABIC_CATEGORY_VOWEL = OT_V,
91 INDIC_SYLLABIC_CATEGORY_VOWEL_DEPENDENT = OT_M,
92 INDIC_SYLLABIC_CATEGORY_VOWEL_INDEPENDENT = OT_V
95 /* Categories used in IndicSMatraCategory.txt from UCD */
96 enum indic_matra_category_t {
97 INDIC_MATRA_CATEGORY_NOT_APPLICABLE = POS_BASE,
99 INDIC_MATRA_CATEGORY_LEFT = POS_PRE,
100 INDIC_MATRA_CATEGORY_TOP = POS_ABOVE,
101 INDIC_MATRA_CATEGORY_BOTTOM = POS_BELOW,
102 INDIC_MATRA_CATEGORY_RIGHT = POS_POST,
104 /* We don't really care much about these since we decompose them
105 * in the generic pre-shaping layer. They will only be used if
106 * the font does not cover the decomposition. In which case, we
107 * define these as aliases to the place we want the split-matra
108 * glyph to show up. Quite arbitrary. */
109 INDIC_MATRA_CATEGORY_BOTTOM_AND_RIGHT = INDIC_MATRA_CATEGORY_BOTTOM,
110 INDIC_MATRA_CATEGORY_LEFT_AND_RIGHT = INDIC_MATRA_CATEGORY_LEFT,
111 INDIC_MATRA_CATEGORY_TOP_AND_BOTTOM = INDIC_MATRA_CATEGORY_BOTTOM,
112 INDIC_MATRA_CATEGORY_TOP_AND_BOTTOM_AND_RIGHT = INDIC_MATRA_CATEGORY_BOTTOM,
113 INDIC_MATRA_CATEGORY_TOP_AND_LEFT = INDIC_MATRA_CATEGORY_LEFT,
114 INDIC_MATRA_CATEGORY_TOP_AND_LEFT_AND_RIGHT = INDIC_MATRA_CATEGORY_LEFT,
115 INDIC_MATRA_CATEGORY_TOP_AND_RIGHT = INDIC_MATRA_CATEGORY_RIGHT,
117 INDIC_MATRA_CATEGORY_INVISIBLE = INDIC_MATRA_CATEGORY_NOT_APPLICABLE,
118 INDIC_MATRA_CATEGORY_OVERSTRUCK = INDIC_MATRA_CATEGORY_NOT_APPLICABLE,
119 INDIC_MATRA_CATEGORY_VISUAL_ORDER_LEFT = INDIC_MATRA_CATEGORY_NOT_APPLICABLE
122 /* Note: We use ASSERT_STATIC_EXPR_ZERO() instead of ASSERT_STATIC_EXPR() and the comma operation
123 * because gcc fails to optimize the latter and fills the table in at runtime. */
124 #define INDIC_COMBINE_CATEGORIES(S,M) \
125 (ASSERT_STATIC_EXPR_ZERO (M == INDIC_MATRA_CATEGORY_NOT_APPLICABLE || (S == INDIC_SYLLABIC_CATEGORY_VIRAMA || S == INDIC_SYLLABIC_CATEGORY_VOWEL_DEPENDENT)) + \
126 ASSERT_STATIC_EXPR_ZERO (S < 16 && M < 16) + \
129 #include "hb-ot-shape-complex-indic-table.hh"
132 * This is a hack for now. We should:
133 * 1. Move this data into the main Indic table,
135 * 2. Probe font lookups to determine consonant positions.
137 static const struct consonant_position_t {
139 indic_position_t position;
140 } consonant_positions[] = {
238 * This is a hack for now. We should move this data into the main Indic table.
240 static const hb_codepoint_t ra_chars[] = {
241 0x0930, /* Devanagari */
242 0x09B0, /* Bengali */
243 0x09F0, /* Bengali */
244 0x09F1, /* Bengali */
245 //0x0A30, /* Gurmukhi */
246 0x0AB0, /* Gujarati */
248 //0x0BB0, /* Tamil */
249 //0x0C30, /* Telugu */
250 0x0CB0, /* Kannada */
251 //0x0D30, /* Malayalam */
255 compare_codepoint (const void *pa, const void *pb)
257 hb_codepoint_t a = * (hb_codepoint_t *) pa;
258 hb_codepoint_t b = * (hb_codepoint_t *) pb;
260 return a < b ? -1 : a == b ? 0 : +1;
263 static indic_position_t
264 consonant_position (hb_codepoint_t u)
266 consonant_position_t *record;
268 record = (consonant_position_t *) bsearch (&u, consonant_positions,
269 ARRAY_LENGTH (consonant_positions),
270 sizeof (consonant_positions[0]),
273 return record ? record->position : POS_BASE;
277 is_ra (hb_codepoint_t u)
279 return !!bsearch (&u, ra_chars,
280 ARRAY_LENGTH (ra_chars),
281 sizeof (ra_chars[0]),
286 static const struct {
289 } indic_basic_features[] =
291 {HB_TAG('n','u','k','t'), true},
292 {HB_TAG('a','k','h','n'), false},
293 {HB_TAG('r','p','h','f'), false},
294 {HB_TAG('r','k','r','f'), false},
295 {HB_TAG('p','r','e','f'), false},
296 {HB_TAG('b','l','w','f'), false},
297 {HB_TAG('h','a','l','f'), false},
298 {HB_TAG('v','a','t','u'), true},
299 {HB_TAG('p','s','t','f'), false},
300 {HB_TAG('c','j','c','t'), true},
303 /* Same order as the indic_basic_features array */
317 static const hb_tag_t indic_other_features[] =
319 HB_TAG('p','r','e','s'),
320 HB_TAG('a','b','v','s'),
321 HB_TAG('b','l','w','s'),
322 HB_TAG('p','s','t','s'),
323 HB_TAG('h','a','l','n'),
325 HB_TAG('d','i','s','t'),
326 HB_TAG('a','b','v','m'),
327 HB_TAG('b','l','w','m'),
332 initial_reordering (const hb_ot_map_t *map,
335 void *user_data HB_UNUSED);
337 final_reordering (const hb_ot_map_t *map,
340 void *user_data HB_UNUSED);
343 _hb_ot_shape_complex_collect_features_indic (hb_ot_map_builder_t *map, const hb_segment_properties_t *props)
345 map->add_bool_feature (HB_TAG('l','o','c','l'));
346 /* The Indic specs do not require ccmp, but we apply it here since if
347 * there is a use of it, it's typically at the beginning. */
348 map->add_bool_feature (HB_TAG('c','c','m','p'));
350 map->add_gsub_pause (initial_reordering, NULL);
352 for (unsigned int i = 0; i < ARRAY_LENGTH (indic_basic_features); i++)
353 map->add_bool_feature (indic_basic_features[i].tag, indic_basic_features[i].is_global);
355 map->add_gsub_pause (final_reordering, NULL);
357 for (unsigned int i = 0; i < ARRAY_LENGTH (indic_other_features); i++)
358 map->add_bool_feature (indic_other_features[i], true);
363 _hb_ot_shape_complex_prefer_decomposed_indic (void)
365 /* We want split matras decomposed by the common shaping logic. */
371 _hb_ot_shape_complex_setup_masks_indic (hb_ot_map_t *map, hb_buffer_t *buffer)
373 HB_BUFFER_ALLOCATE_VAR (buffer, indic_category);
374 HB_BUFFER_ALLOCATE_VAR (buffer, indic_position);
376 /* We cannot setup masks here. We save information about characters
377 * and setup masks later on in a pause-callback. */
379 unsigned int count = buffer->len;
380 for (unsigned int i = 0; i < count; i++)
382 unsigned int type = get_indic_categories (buffer->info[i].codepoint);
384 buffer->info[i].indic_category() = type & 0x0F;
385 buffer->info[i].indic_position() = type >> 4;
387 if (buffer->info[i].indic_category() == OT_C) {
388 buffer->info[i].indic_position() = consonant_position (buffer->info[i].codepoint);
389 if (is_ra (buffer->info[i].codepoint))
390 buffer->info[i].indic_category() = OT_Ra;
396 compare_indic_order (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb)
398 int a = pa->indic_position();
399 int b = pb->indic_position();
401 return a < b ? -1 : a == b ? 0 : +1;
405 found_consonant_syllable (const hb_ot_map_t *map, hb_buffer_t *buffer, hb_mask_t *mask_array,
406 unsigned int start, unsigned int end)
411 * https://www.microsoft.com/typography/otfntdev/devanot/shaping.aspx */
413 /* 1. Find base consonant:
415 * The shaping engine finds the base consonant of the syllable, using the
416 * following algorithm: starting from the end of the syllable, move backwards
417 * until a consonant is found that does not have a below-base or post-base
418 * form (post-base forms have to follow below-base forms), or that is not a
419 * pre-base reordering Ra, or arrive at the first consonant. The consonant
420 * stopped at will be the base.
422 * o If the syllable starts with Ra + Halant (in a script that has Reph)
423 * and has more than one consonant, Ra is excluded from candidates for
427 unsigned int base = 0;
429 /* -> starting from the end of the syllable, move backwards */
433 /* -> until a consonant is found */
434 if (buffer->info[i].indic_category() == OT_C)
436 /* -> that does not have a below-base or post-base form
437 * (post-base forms have to follow below-base forms), */
438 if (buffer->info[i].indic_position() != POS_BELOW &&
439 buffer->info[i].indic_position() != POS_POST)
445 /* TODO: or that is not a pre-base reordering Ra, */
447 /* -> or arrive at the first consonant. The consonant stopped at will be the base. */
452 base = start; /* Just in case... */
455 * If the syllable starts with Ra + Halant (in a script that has Reph)
456 * and has more than one consonant, Ra is excluded from candidates for
457 * base consonants. */
460 /* 2. Decompose and reorder Matras:
462 * Each matra and any syllable modifier sign in the cluster are moved to the
463 * appropriate position relative to the consonant(s) in the cluster. The
464 * shaping engine decomposes two- or three-part matras into their constituent
465 * parts before any repositioning. Matra characters are classified by which
466 * consonant in a conjunct they have affinity for and are reordered to the
467 * following positions:
469 * o Before first half form in the syllable
470 * o After subjoined consonants
471 * o After post-form consonant
472 * o After main consonant (for above marks)
474 * IMPLEMENTATION NOTES:
476 * The normalize() routine has already decomposed matras for us, so we don't
477 * need to worry about that.
481 /* 3. Reorder marks to canonical order:
483 * Adjacent nukta and halant or nukta and vedic sign are always repositioned
484 * if necessary, so that the nukta is first.
486 * IMPLEMENTATION NOTES:
488 * We don't need to do this: the normalize() routine already did this for us.
492 /* Reorder characters */
494 for (i = start; i < base; i++)
495 buffer->info[i].indic_position() = POS_PRE;
496 buffer->info[base].indic_position() = POS_BASE;
498 /* Attach ZWJ, ZWNJ, nukta, and halant to previous char to move with them. */
499 for (i = start + 1; i < end; i++)
500 if ((FLAG (buffer->info[i].indic_category()) &
501 (FLAG (OT_ZWNJ) | FLAG (OT_ZWJ) | FLAG (OT_N) | FLAG (OT_H))))
502 buffer->info[i].indic_position() = buffer->info[i - 1].indic_position();
504 /* We do bubble-sort, skip malicious clusters attempts */
505 if (end - start > 20)
508 /* Sit tight, rock 'n roll! */
509 hb_bubble_sort (buffer->info + start, end - start, compare_indic_order);
511 /* Setup masks now */
514 for (i = start; i < base; i++)
515 buffer->info[i].mask |= mask_array[HALF] | mask_array[AKHN];
517 buffer->info[base].mask |= mask_array[AKHN];
519 for (i = base + 1; i < end; i++)
520 buffer->info[i].mask |= mask_array[BLWF] | mask_array[PSTF];
525 found_vowel_syllable (const hb_ot_map_t *map, hb_buffer_t *buffer, hb_mask_t *mask_array,
526 unsigned int start, unsigned int end)
529 * Not clear to me how this should work. Do the matras move to before the
530 * independent vowel? No idea.
535 found_standalone_cluster (const hb_ot_map_t *map, hb_buffer_t *buffer, hb_mask_t *mask_array,
536 unsigned int start, unsigned int end)
539 * Easiest thing to do here is to convert the NBSP to consonant and
540 * call found_consonant_syllable.
545 found_non_indic (const hb_ot_map_t *map, hb_buffer_t *buffer, hb_mask_t *mask_array,
546 unsigned int start, unsigned int end)
548 /* Nothing to do right now. If we ever switch to using the output
549 * buffer in the reordering process, we'd need to next_glyph() here. */
552 #include "hb-ot-shape-complex-indic-machine.hh"
555 initial_reordering (const hb_ot_map_t *map,
558 void *user_data HB_UNUSED)
560 hb_mask_t mask_array[ARRAY_LENGTH (indic_basic_features)] = {0};
561 unsigned int num_masks = ARRAY_LENGTH (indic_basic_features);
562 for (unsigned int i = 0; i < num_masks; i++)
563 mask_array[i] = map->get_1_mask (indic_basic_features[i].tag);
565 find_syllables (map, buffer, mask_array);
569 final_reordering (const hb_ot_map_t *map,
572 void *user_data HB_UNUSED)
574 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_category);
575 HB_BUFFER_DEALLOCATE_VAR (buffer, indic_position);