1 /*
2 * Copyright © 2013 Google, Inc.
3 *
4 * This is part of HarfBuzz, a text shaping library.
5 *
6 * Permission is hereby granted, without written agreement and without
7 * license or royalty fees, to use, copy, modify, and distribute this
8 * software and its documentation for any purpose, provided that the
9 * above copyright notice and the following two paragraphs appear in
10 * all copies of this software.
11 *
12 * IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
13 * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
14 * ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
15 * IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
16 * DAMAGE.
17 *
18 * THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
19 * BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
20 * FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS
21 * ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
22 * PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
23 *
24 * Google Author(s): Behdad Esfahbod
25 */
26
27 #include "hb-ot-shape-complex.hh"
28
29
30 /* Hangul shaper */
31
32
33 /* Same order as the feature array below */
34 enum {
35 _JMO,
36
37 LJMO,
38 VJMO,
39 TJMO,
40
41 FIRST_HANGUL_FEATURE = LJMO,
42 HANGUL_FEATURE_COUNT = TJMO + 1
43 };
44
45 static const hb_tag_t hangul_features[HANGUL_FEATURE_COUNT] =
46 {
47 HB_TAG_NONE,
48 HB_TAG('l','j','m','o'),
49 HB_TAG('v','j','m','o'),
50 HB_TAG('t','j','m','o')
51 };
52
53 static void
collect_features_hangul(hb_ot_shape_planner_t * plan)54 collect_features_hangul (hb_ot_shape_planner_t *plan)
55 {
56 hb_ot_map_builder_t *map = &plan->map;
57
58 for (unsigned int i = FIRST_HANGUL_FEATURE; i < HANGUL_FEATURE_COUNT; i++)
59 map->add_feature (hangul_features[i]);
60 }
61
62 static void
override_features_hangul(hb_ot_shape_planner_t * plan)63 override_features_hangul (hb_ot_shape_planner_t *plan)
64 {
65 /* Uniscribe does not apply 'calt' for Hangul, and certain fonts
66 * (Noto Sans CJK, Source Sans Han, etc) apply all of jamo lookups
67 * in calt, which is not desirable. */
68 plan->map.disable_feature (HB_TAG('c','a','l','t'));
69 }
70
71 struct hangul_shape_plan_t
72 {
73 hb_mask_t mask_array[HANGUL_FEATURE_COUNT];
74 };
75
76 static void *
data_create_hangul(const hb_ot_shape_plan_t * plan)77 data_create_hangul (const hb_ot_shape_plan_t *plan)
78 {
79 hangul_shape_plan_t *hangul_plan = (hangul_shape_plan_t *) calloc (1, sizeof (hangul_shape_plan_t));
80 if (unlikely (!hangul_plan))
81 return nullptr;
82
83 for (unsigned int i = 0; i < HANGUL_FEATURE_COUNT; i++)
84 hangul_plan->mask_array[i] = plan->map.get_1_mask (hangul_features[i]);
85
86 return hangul_plan;
87 }
88
89 static void
data_destroy_hangul(void * data)90 data_destroy_hangul (void *data)
91 {
92 free (data);
93 }
94
95 /* Constants for algorithmic hangul syllable [de]composition. */
96 #define LBase 0x1100u
97 #define VBase 0x1161u
98 #define TBase 0x11A7u
99 #define LCount 19u
100 #define VCount 21u
101 #define TCount 28u
102 #define SBase 0xAC00u
103 #define NCount (VCount * TCount)
104 #define SCount (LCount * NCount)
105
106 #define isCombiningL(u) (hb_in_range<hb_codepoint_t> ((u), LBase, LBase+LCount-1))
107 #define isCombiningV(u) (hb_in_range<hb_codepoint_t> ((u), VBase, VBase+VCount-1))
108 #define isCombiningT(u) (hb_in_range<hb_codepoint_t> ((u), TBase+1, TBase+TCount-1))
109 #define isCombinedS(u) (hb_in_range<hb_codepoint_t> ((u), SBase, SBase+SCount-1))
110
111 #define isL(u) (hb_in_ranges<hb_codepoint_t> ((u), 0x1100u, 0x115Fu, 0xA960u, 0xA97Cu))
112 #define isV(u) (hb_in_ranges<hb_codepoint_t> ((u), 0x1160u, 0x11A7u, 0xD7B0u, 0xD7C6u))
113 #define isT(u) (hb_in_ranges<hb_codepoint_t> ((u), 0x11A8u, 0x11FFu, 0xD7CBu, 0xD7FBu))
114
115 #define isHangulTone(u) (hb_in_range<hb_codepoint_t> ((u), 0x302Eu, 0x302Fu))
116
117 /* buffer var allocations */
118 #define hangul_shaping_feature() complex_var_u8_0() /* hangul jamo shaping feature */
119
120 static bool
is_zero_width_char(hb_font_t * font,hb_codepoint_t unicode)121 is_zero_width_char (hb_font_t *font,
122 hb_codepoint_t unicode)
123 {
124 hb_codepoint_t glyph;
125 return hb_font_get_glyph (font, unicode, 0, &glyph) && hb_font_get_glyph_h_advance (font, glyph) == 0;
126 }
127
128 static void
preprocess_text_hangul(const hb_ot_shape_plan_t * plan HB_UNUSED,hb_buffer_t * buffer,hb_font_t * font)129 preprocess_text_hangul (const hb_ot_shape_plan_t *plan HB_UNUSED,
130 hb_buffer_t *buffer,
131 hb_font_t *font)
132 {
133 HB_BUFFER_ALLOCATE_VAR (buffer, hangul_shaping_feature);
134
135 /* Hangul syllables come in two shapes: LV, and LVT. Of those:
136 *
137 * - LV can be precomposed, or decomposed. Lets call those
138 * <LV> and <L,V>,
139 * - LVT can be fully precomposed, partically precomposed, or
140 * fully decomposed. Ie. <LVT>, <LV,T>, or <L,V,T>.
141 *
142 * The composition / decomposition is mechanical. However, not
143 * all <L,V> sequences compose, and not all <LV,T> sequences
144 * compose.
145 *
146 * Here are the specifics:
147 *
148 * - <L>: U+1100..115F, U+A960..A97F
149 * - <V>: U+1160..11A7, U+D7B0..D7C7
150 * - <T>: U+11A8..11FF, U+D7CB..D7FB
151 *
152 * - Only the <L,V> sequences for some of the U+11xx ranges combine.
153 * - Only <LV,T> sequences for some of the Ts in U+11xx range combine.
154 *
155 * Here is what we want to accomplish in this shaper:
156 *
157 * - If the whole syllable can be precomposed, do that,
158 * - Otherwise, fully decompose and apply ljmo/vjmo/tjmo features.
159 * - If a valid syllable is followed by a Hangul tone mark, reorder the tone
160 * mark to precede the whole syllable - unless it is a zero-width glyph, in
161 * which case we leave it untouched, assuming it's designed to overstrike.
162 *
163 * That is, of the different possible syllables:
164 *
165 * <L>
166 * <L,V>
167 * <L,V,T>
168 * <LV>
169 * <LVT>
170 * <LV, T>
171 *
172 * - <L> needs no work.
173 *
174 * - <LV> and <LVT> can stay the way they are if the font supports them, otherwise we
175 * should fully decompose them if font supports.
176 *
177 * - <L,V> and <L,V,T> we should compose if the whole thing can be composed.
178 *
179 * - <LV,T> we should compose if the whole thing can be composed, otherwise we should
180 * decompose.
181 */
182
183 buffer->clear_output ();
184 unsigned int start = 0, end = 0; /* Extent of most recently seen syllable;
185 * valid only if start < end
186 */
187 unsigned int count = buffer->len;
188
189 for (buffer->idx = 0; buffer->idx < count && buffer->successful;)
190 {
191 hb_codepoint_t u = buffer->cur().codepoint;
192
193 if (isHangulTone (u))
194 {
195 /*
196 * We could cache the width of the tone marks and the existence of dotted-circle,
197 * but the use of the Hangul tone mark characters seems to be rare enough that
198 * I didn't bother for now.
199 */
200 if (start < end && end == buffer->out_len)
201 {
202 /* Tone mark follows a valid syllable; move it in front, unless it's zero width. */
203 buffer->unsafe_to_break_from_outbuffer (start, buffer->idx);
204 buffer->next_glyph ();
205 if (!is_zero_width_char (font, u))
206 {
207 buffer->merge_out_clusters (start, end + 1);
208 hb_glyph_info_t *info = buffer->out_info;
209 hb_glyph_info_t tone = info[end];
210 memmove (&info[start + 1], &info[start], (end - start) * sizeof (hb_glyph_info_t));
211 info[start] = tone;
212 }
213 }
214 else
215 {
216 /* No valid syllable as base for tone mark; try to insert dotted circle. */
217 if (font->has_glyph (0x25CCu))
218 {
219 hb_codepoint_t chars[2];
220 if (!is_zero_width_char (font, u)) {
221 chars[0] = u;
222 chars[1] = 0x25CCu;
223 } else {
224 chars[0] = 0x25CCu;
225 chars[1] = u;
226 }
227 buffer->replace_glyphs (1, 2, chars);
228 }
229 else
230 {
231 /* No dotted circle available in the font; just leave tone mark untouched. */
232 buffer->next_glyph ();
233 }
234 }
235 start = end = buffer->out_len;
236 continue;
237 }
238
239 start = buffer->out_len; /* Remember current position as a potential syllable start;
240 * will only be used if we set end to a later position.
241 */
242
243 if (isL (u) && buffer->idx + 1 < count)
244 {
245 hb_codepoint_t l = u;
246 hb_codepoint_t v = buffer->cur(+1).codepoint;
247 if (isV (v))
248 {
249 /* Have <L,V> or <L,V,T>. */
250 hb_codepoint_t t = 0;
251 unsigned int tindex = 0;
252 if (buffer->idx + 2 < count)
253 {
254 t = buffer->cur(+2).codepoint;
255 if (isT (t))
256 tindex = t - TBase; /* Only used if isCombiningT (t); otherwise invalid. */
257 else
258 t = 0; /* The next character was not a trailing jamo. */
259 }
260 buffer->unsafe_to_break (buffer->idx, buffer->idx + (t ? 3 : 2));
261
262 /* We've got a syllable <L,V,T?>; see if it can potentially be composed. */
263 if (isCombiningL (l) && isCombiningV (v) && (t == 0 || isCombiningT (t)))
264 {
265 /* Try to compose; if this succeeds, end is set to start+1. */
266 hb_codepoint_t s = SBase + (l - LBase) * NCount + (v - VBase) * TCount + tindex;
267 if (font->has_glyph (s))
268 {
269 buffer->replace_glyphs (t ? 3 : 2, 1, &s);
270 if (unlikely (!buffer->successful))
271 return;
272 end = start + 1;
273 continue;
274 }
275 }
276
277 /* We didn't compose, either because it's an Old Hangul syllable without a
278 * precomposed character in Unicode, or because the font didn't support the
279 * necessary precomposed glyph.
280 * Set jamo features on the individual glyphs, and advance past them.
281 */
282 buffer->cur().hangul_shaping_feature() = LJMO;
283 buffer->next_glyph ();
284 buffer->cur().hangul_shaping_feature() = VJMO;
285 buffer->next_glyph ();
286 if (t)
287 {
288 buffer->cur().hangul_shaping_feature() = TJMO;
289 buffer->next_glyph ();
290 end = start + 3;
291 }
292 else
293 end = start + 2;
294 if (buffer->cluster_level == HB_BUFFER_CLUSTER_LEVEL_MONOTONE_GRAPHEMES)
295 buffer->merge_out_clusters (start, end);
296 continue;
297 }
298 }
299
300 else if (isCombinedS (u))
301 {
302 /* Have <LV>, <LVT>, or <LV,T> */
303 hb_codepoint_t s = u;
304 bool has_glyph = font->has_glyph (s);
305 unsigned int lindex = (s - SBase) / NCount;
306 unsigned int nindex = (s - SBase) % NCount;
307 unsigned int vindex = nindex / TCount;
308 unsigned int tindex = nindex % TCount;
309
310 if (!tindex &&
311 buffer->idx + 1 < count &&
312 isCombiningT (buffer->cur(+1).codepoint))
313 {
314 /* <LV,T>, try to combine. */
315 unsigned int new_tindex = buffer->cur(+1).codepoint - TBase;
316 hb_codepoint_t new_s = s + new_tindex;
317 if (font->has_glyph (new_s))
318 {
319 buffer->replace_glyphs (2, 1, &new_s);
320 if (unlikely (!buffer->successful))
321 return;
322 end = start + 1;
323 continue;
324 }
325 else
326 buffer->unsafe_to_break (buffer->idx, buffer->idx + 2); /* Mark unsafe between LV and T. */
327 }
328
329 /* Otherwise, decompose if font doesn't support <LV> or <LVT>,
330 * or if having non-combining <LV,T>. Note that we already handled
331 * combining <LV,T> above. */
332 if (!has_glyph ||
333 (!tindex &&
334 buffer->idx + 1 < count &&
335 isT (buffer->cur(+1).codepoint)))
336 {
337 hb_codepoint_t decomposed[3] = {LBase + lindex,
338 VBase + vindex,
339 TBase + tindex};
340 if (font->has_glyph (decomposed[0]) &&
341 font->has_glyph (decomposed[1]) &&
342 (!tindex || font->has_glyph (decomposed[2])))
343 {
344 unsigned int s_len = tindex ? 3 : 2;
345 buffer->replace_glyphs (1, s_len, decomposed);
346
347 /* If we decomposed an LV because of a non-combining T following,
348 * we want to include this T in the syllable.
349 */
350 if (has_glyph && !tindex)
351 {
352 buffer->next_glyph ();
353 s_len++;
354 }
355
356 if (unlikely (!buffer->successful))
357 return;
358
359 /* We decomposed S: apply jamo features to the individual glyphs
360 * that are now in buffer->out_info.
361 */
362 hb_glyph_info_t *info = buffer->out_info;
363 end = start + s_len;
364
365 unsigned int i = start;
366 info[i++].hangul_shaping_feature() = LJMO;
367 info[i++].hangul_shaping_feature() = VJMO;
368 if (i < end)
369 info[i++].hangul_shaping_feature() = TJMO;
370
371 if (buffer->cluster_level == HB_BUFFER_CLUSTER_LEVEL_MONOTONE_GRAPHEMES)
372 buffer->merge_out_clusters (start, end);
373 continue;
374 }
375 else if ((!tindex && buffer->idx + 1 < count && isT (buffer->cur(+1).codepoint)))
376 buffer->unsafe_to_break (buffer->idx, buffer->idx + 2); /* Mark unsafe between LV and T. */
377 }
378
379 if (has_glyph)
380 {
381 /* We didn't decompose the S, so just advance past it. */
382 end = start + 1;
383 buffer->next_glyph ();
384 continue;
385 }
386 }
387
388 /* Didn't find a recognizable syllable, so we leave end <= start;
389 * this will prevent tone-mark reordering happening.
390 */
391 buffer->next_glyph ();
392 }
393 buffer->swap_buffers ();
394 }
395
396 static void
setup_masks_hangul(const hb_ot_shape_plan_t * plan,hb_buffer_t * buffer,hb_font_t * font HB_UNUSED)397 setup_masks_hangul (const hb_ot_shape_plan_t *plan,
398 hb_buffer_t *buffer,
399 hb_font_t *font HB_UNUSED)
400 {
401 const hangul_shape_plan_t *hangul_plan = (const hangul_shape_plan_t *) plan->data;
402
403 if (likely (hangul_plan))
404 {
405 unsigned int count = buffer->len;
406 hb_glyph_info_t *info = buffer->info;
407 for (unsigned int i = 0; i < count; i++, info++)
408 info->mask |= hangul_plan->mask_array[info->hangul_shaping_feature()];
409 }
410
411 HB_BUFFER_DEALLOCATE_VAR (buffer, hangul_shaping_feature);
412 }
413
414
415 const hb_ot_complex_shaper_t _hb_ot_complex_shaper_hangul =
416 {
417 collect_features_hangul,
418 override_features_hangul,
419 data_create_hangul,
420 data_destroy_hangul,
421 preprocess_text_hangul,
422 nullptr, /* postprocess_glyphs */
423 HB_OT_SHAPE_NORMALIZATION_MODE_NONE,
424 nullptr, /* decompose */
425 nullptr, /* compose */
426 setup_masks_hangul,
427 HB_TAG_NONE, /* gpos_tag */
428 nullptr, /* reorder_marks */
429 HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE,
430 false, /* fallback_position */
431 };
432