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-private.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], 1, F_NONE);
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.add_feature (HB_TAG('c','a','l','t'), 0, F_GLOBAL);
69 }
70
71 struct hangul_shape_plan_t
72 {
73 ASSERT_POD ();
74
75 hb_mask_t mask_array[HANGUL_FEATURE_COUNT];
76 };
77
78 static void *
data_create_hangul(const hb_ot_shape_plan_t * plan)79 data_create_hangul (const hb_ot_shape_plan_t *plan)
80 {
81 hangul_shape_plan_t *hangul_plan = (hangul_shape_plan_t *) calloc (1, sizeof (hangul_shape_plan_t));
82 if (unlikely (!hangul_plan))
83 return nullptr;
84
85 for (unsigned int i = 0; i < HANGUL_FEATURE_COUNT; i++)
86 hangul_plan->mask_array[i] = plan->map.get_1_mask (hangul_features[i]);
87
88 return hangul_plan;
89 }
90
91 static void
data_destroy_hangul(void * data)92 data_destroy_hangul (void *data)
93 {
94 free (data);
95 }
96
97 /* Constants for algorithmic hangul syllable [de]composition. */
98 #define LBase 0x1100u
99 #define VBase 0x1161u
100 #define TBase 0x11A7u
101 #define LCount 19u
102 #define VCount 21u
103 #define TCount 28u
104 #define SBase 0xAC00u
105 #define NCount (VCount * TCount)
106 #define SCount (LCount * NCount)
107
108 #define isCombiningL(u) (hb_in_range<hb_codepoint_t> ((u), LBase, LBase+LCount-1))
109 #define isCombiningV(u) (hb_in_range<hb_codepoint_t> ((u), VBase, VBase+VCount-1))
110 #define isCombiningT(u) (hb_in_range<hb_codepoint_t> ((u), TBase+1, TBase+TCount-1))
111 #define isCombinedS(u) (hb_in_range<hb_codepoint_t> ((u), SBase, SBase+SCount-1))
112
113 #define isL(u) (hb_in_ranges<hb_codepoint_t> ((u), 0x1100u, 0x115Fu, 0xA960u, 0xA97Cu))
114 #define isV(u) (hb_in_ranges<hb_codepoint_t> ((u), 0x1160u, 0x11A7u, 0xD7B0u, 0xD7C6u))
115 #define isT(u) (hb_in_ranges<hb_codepoint_t> ((u), 0x11A8u, 0x11FFu, 0xD7CBu, 0xD7FBu))
116
117 #define isHangulTone(u) (hb_in_range<hb_codepoint_t> ((u), 0x302Eu, 0x302Fu))
118
119 /* buffer var allocations */
120 #define hangul_shaping_feature() complex_var_u8_0() /* hangul jamo shaping feature */
121
122 static bool
is_zero_width_char(hb_font_t * font,hb_codepoint_t unicode)123 is_zero_width_char (hb_font_t *font,
124 hb_codepoint_t unicode)
125 {
126 hb_codepoint_t glyph;
127 return hb_font_get_glyph (font, unicode, 0, &glyph) && hb_font_get_glyph_h_advance (font, glyph) == 0;
128 }
129
130 static void
preprocess_text_hangul(const hb_ot_shape_plan_t * plan,hb_buffer_t * buffer,hb_font_t * font)131 preprocess_text_hangul (const hb_ot_shape_plan_t *plan,
132 hb_buffer_t *buffer,
133 hb_font_t *font)
134 {
135 HB_BUFFER_ALLOCATE_VAR (buffer, hangul_shaping_feature);
136
137 /* Hangul syllables come in two shapes: LV, and LVT. Of those:
138 *
139 * - LV can be precomposed, or decomposed. Lets call those
140 * <LV> and <L,V>,
141 * - LVT can be fully precomposed, partically precomposed, or
142 * fully decomposed. Ie. <LVT>, <LV,T>, or <L,V,T>.
143 *
144 * The composition / decomposition is mechanical. However, not
145 * all <L,V> sequences compose, and not all <LV,T> sequences
146 * compose.
147 *
148 * Here are the specifics:
149 *
150 * - <L>: U+1100..115F, U+A960..A97F
151 * - <V>: U+1160..11A7, U+D7B0..D7C7
152 * - <T>: U+11A8..11FF, U+D7CB..D7FB
153 *
154 * - Only the <L,V> sequences for the 11xx ranges combine.
155 * - Only <LV,T> sequences for T in U+11A8..11C3 combine.
156 *
157 * Here is what we want to accomplish in this shaper:
158 *
159 * - If the whole syllable can be precomposed, do that,
160 * - Otherwise, fully decompose and apply ljmo/vjmo/tjmo features.
161 * - If a valid syllable is followed by a Hangul tone mark, reorder the tone
162 * mark to precede the whole syllable - unless it is a zero-width glyph, in
163 * which case we leave it untouched, assuming it's designed to overstrike.
164 *
165 * That is, of the different possible syllables:
166 *
167 * <L>
168 * <L,V>
169 * <L,V,T>
170 * <LV>
171 * <LVT>
172 * <LV, T>
173 *
174 * - <L> needs no work.
175 *
176 * - <LV> and <LVT> can stay the way they are if the font supports them, otherwise we
177 * should fully decompose them if font supports.
178 *
179 * - <L,V> and <L,V,T> we should compose if the whole thing can be composed.
180 *
181 * - <LV,T> we should compose if the whole thing can be composed, otherwise we should
182 * decompose.
183 */
184
185 buffer->clear_output ();
186 unsigned int start = 0, end = 0; /* Extent of most recently seen syllable;
187 * valid only if start < end
188 */
189 unsigned int count = buffer->len;
190
191 for (buffer->idx = 0; buffer->idx < count && !buffer->in_error;)
192 {
193 hb_codepoint_t u = buffer->cur().codepoint;
194
195 if (isHangulTone (u))
196 {
197 /*
198 * We could cache the width of the tone marks and the existence of dotted-circle,
199 * but the use of the Hangul tone mark characters seems to be rare enough that
200 * I didn't bother for now.
201 */
202 if (start < end && end == buffer->out_len)
203 {
204 /* Tone mark follows a valid syllable; move it in front, unless it's zero width. */
205 buffer->unsafe_to_break_from_outbuffer (start, buffer->idx);
206 buffer->next_glyph ();
207 if (!is_zero_width_char (font, u))
208 {
209 buffer->merge_out_clusters (start, end + 1);
210 hb_glyph_info_t *info = buffer->out_info;
211 hb_glyph_info_t tone = info[end];
212 memmove (&info[start + 1], &info[start], (end - start) * sizeof (hb_glyph_info_t));
213 info[start] = tone;
214 }
215 }
216 else
217 {
218 /* No valid syllable as base for tone mark; try to insert dotted circle. */
219 if (font->has_glyph (0x25CCu))
220 {
221 hb_codepoint_t chars[2];
222 if (!is_zero_width_char (font, u)) {
223 chars[0] = u;
224 chars[1] = 0x25CCu;
225 } else {
226 chars[0] = 0x25CCu;
227 chars[1] = u;
228 }
229 buffer->replace_glyphs (1, 2, chars);
230 }
231 else
232 {
233 /* No dotted circle available in the font; just leave tone mark untouched. */
234 buffer->next_glyph ();
235 }
236 }
237 start = end = buffer->out_len;
238 continue;
239 }
240
241 start = buffer->out_len; /* Remember current position as a potential syllable start;
242 * will only be used if we set end to a later position.
243 */
244
245 if (isL (u) && buffer->idx + 1 < count)
246 {
247 hb_codepoint_t l = u;
248 hb_codepoint_t v = buffer->cur(+1).codepoint;
249 if (isV (v))
250 {
251 /* Have <L,V> or <L,V,T>. */
252 hb_codepoint_t t = 0;
253 unsigned int tindex = 0;
254 if (buffer->idx + 2 < count)
255 {
256 t = buffer->cur(+2).codepoint;
257 if (isT (t))
258 tindex = t - TBase; /* Only used if isCombiningT (t); otherwise invalid. */
259 else
260 t = 0; /* The next character was not a trailing jamo. */
261 }
262 buffer->unsafe_to_break (buffer->idx, buffer->idx + (t ? 3 : 2));
263
264 /* We've got a syllable <L,V,T?>; see if it can potentially be composed. */
265 if (isCombiningL (l) && isCombiningV (v) && (t == 0 || isCombiningT (t)))
266 {
267 /* Try to compose; if this succeeds, end is set to start+1. */
268 hb_codepoint_t s = SBase + (l - LBase) * NCount + (v - VBase) * TCount + tindex;
269 if (font->has_glyph (s))
270 {
271 buffer->replace_glyphs (t ? 3 : 2, 1, &s);
272 if (unlikely (buffer->in_error))
273 return;
274 end = start + 1;
275 continue;
276 }
277 }
278
279 /* We didn't compose, either because it's an Old Hangul syllable without a
280 * precomposed character in Unicode, or because the font didn't support the
281 * necessary precomposed glyph.
282 * Set jamo features on the individual glyphs, and advance past them.
283 */
284 buffer->cur().hangul_shaping_feature() = LJMO;
285 buffer->next_glyph ();
286 buffer->cur().hangul_shaping_feature() = VJMO;
287 buffer->next_glyph ();
288 if (t)
289 {
290 buffer->cur().hangul_shaping_feature() = TJMO;
291 buffer->next_glyph ();
292 end = start + 3;
293 }
294 else
295 end = start + 2;
296 if (buffer->cluster_level == HB_BUFFER_CLUSTER_LEVEL_MONOTONE_GRAPHEMES)
297 buffer->merge_out_clusters (start, end);
298 continue;
299 }
300 }
301
302 else if (isCombinedS (u))
303 {
304 /* Have <LV>, <LVT>, or <LV,T> */
305 hb_codepoint_t s = u;
306 bool has_glyph = font->has_glyph (s);
307 unsigned int lindex = (s - SBase) / NCount;
308 unsigned int nindex = (s - SBase) % NCount;
309 unsigned int vindex = nindex / TCount;
310 unsigned int tindex = nindex % TCount;
311
312 if (!tindex &&
313 buffer->idx + 1 < count &&
314 isCombiningT (buffer->cur(+1).codepoint))
315 {
316 /* <LV,T>, try to combine. */
317 unsigned int new_tindex = buffer->cur(+1).codepoint - TBase;
318 hb_codepoint_t new_s = s + new_tindex;
319 if (font->has_glyph (new_s))
320 {
321 buffer->replace_glyphs (2, 1, &new_s);
322 if (unlikely (buffer->in_error))
323 return;
324 end = start + 1;
325 continue;
326 }
327 else
328 buffer->unsafe_to_break (buffer->idx, buffer->idx + 2); /* Mark unsafe between LV and T. */
329 }
330
331 /* Otherwise, decompose if font doesn't support <LV> or <LVT>,
332 * or if having non-combining <LV,T>. Note that we already handled
333 * combining <LV,T> above. */
334 if (!has_glyph ||
335 (!tindex &&
336 buffer->idx + 1 < count &&
337 isT (buffer->cur(+1).codepoint)))
338 {
339 hb_codepoint_t decomposed[3] = {LBase + lindex,
340 VBase + vindex,
341 TBase + tindex};
342 if (font->has_glyph (decomposed[0]) &&
343 font->has_glyph (decomposed[1]) &&
344 (!tindex || font->has_glyph (decomposed[2])))
345 {
346 unsigned int s_len = tindex ? 3 : 2;
347 buffer->replace_glyphs (1, s_len, decomposed);
348 if (unlikely (buffer->in_error))
349 return;
350
351 /* We decomposed S: apply jamo features to the individual glyphs
352 * that are now in buffer->out_info.
353 */
354 hb_glyph_info_t *info = buffer->out_info;
355
356 /* If we decomposed an LV because of a non-combining T following,
357 * we want to include this T in the syllable.
358 */
359 if (has_glyph && !tindex)
360 {
361 buffer->next_glyph ();
362 s_len++;
363 }
364 end = start + s_len;
365
366 unsigned int i = start;
367 info[i++].hangul_shaping_feature() = LJMO;
368 info[i++].hangul_shaping_feature() = VJMO;
369 if (i < end)
370 info[i++].hangul_shaping_feature() = TJMO;
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 nullptr, /* disable_otl */
428 nullptr, /* reorder_marks */
429 HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE,
430 false, /* fallback_position */
431 };
432