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-normalize-private.hh"
57 #include "hb-ot-shape-complex-private.hh"
58 #include "hb-ot-shape-private.hh"
59
60
61 /*
62 * HIGHLEVEL DESIGN:
63 *
64 * This file exports one main function: _hb_ot_shape_normalize().
65 *
66 * This function closely reflects the Unicode Normalization Algorithm,
67 * yet it's different.
68 *
69 * Each shaper specifies whether it prefers decomposed (NFD) or composed (NFC).
70 * The logic however tries to use whatever the font can support.
71 *
72 * In general what happens is that: each grapheme is decomposed in a chain
73 * of 1:2 decompositions, marks reordered, and then recomposed if desired,
74 * so far it's like Unicode Normalization. However, the decomposition and
75 * recomposition only happens if the font supports the resulting characters.
76 *
77 * The goals are:
78 *
79 * - Try to render all canonically equivalent strings similarly. To really
80 * achieve this we have to always do the full decomposition and then
81 * selectively recompose from there. It's kinda too expensive though, so
82 * we skip some cases. For example, if composed is desired, we simply
83 * don't touch 1-character clusters that are supported by the font, even
84 * though their NFC may be different.
85 *
86 * - When a font has a precomposed character for a sequence but the 'ccmp'
87 * feature in the font is not adequate, use the precomposed character
88 * which typically has better mark positioning.
89 *
90 * - When a font does not support a combining mark, but supports it precomposed
91 * with previous base, use that. This needs the itemizer to have this
92 * knowledge too. We need to provide assistance to the itemizer.
93 *
94 * - When a font does not support a character but supports its decomposition,
95 * well, use the decomposition (preferring the canonical decomposition, but
96 * falling back to the compatibility decomposition if necessary). The
97 * compatibility decomposition is really nice to have, for characters like
98 * ellipsis, or various-sized space characters.
99 *
100 * - The complex shapers can customize the compose and decompose functions to
101 * offload some of their requirements to the normalizer. For example, the
102 * Indic shaper may want to disallow recomposing of two matras.
103 *
104 * - We try compatibility decomposition if decomposing through canonical
105 * decomposition alone failed to find a sequence that the font supports.
106 * We don't try compatibility decomposition recursively during the canonical
107 * decomposition phase. This has minimal impact. There are only a handful
108 * of Greek letter that have canonical decompositions that include characters
109 * with compatibility decomposition. Those can be found using this command:
110 *
111 * egrep "`echo -n ';('; grep ';<' UnicodeData.txt | cut -d';' -f1 | tr '\n' '|'; echo ') '`" UnicodeData.txt
112 */
113
114 static bool
115 decompose_unicode (const hb_ot_shape_normalize_context_t *c,
116 hb_codepoint_t ab,
117 hb_codepoint_t *a,
118 hb_codepoint_t *b)
119 {
120 return c->unicode->decompose (ab, a, b);
121 }
122
123 static bool
124 compose_unicode (const hb_ot_shape_normalize_context_t *c,
125 hb_codepoint_t a,
126 hb_codepoint_t b,
127 hb_codepoint_t *ab)
128 {
129 return c->unicode->compose (a, b, ab);
130 }
131
132 static inline void
133 set_glyph (hb_glyph_info_t &info, hb_font_t *font)
134 {
135 font->get_glyph (info.codepoint, 0, &info.glyph_index());
136 }
137
138 static inline void
139 output_char (hb_buffer_t *buffer, hb_codepoint_t unichar, hb_codepoint_t glyph)
140 {
141 buffer->cur().glyph_index() = glyph;
142 buffer->output_glyph (unichar);
143 _hb_glyph_info_set_unicode_props (&buffer->prev(), buffer->unicode);
144 }
145
146 static inline void
147 next_char (hb_buffer_t *buffer, hb_codepoint_t glyph)
148 {
149 buffer->cur().glyph_index() = glyph;
150 buffer->next_glyph ();
151 }
152
153 static inline void
154 skip_char (hb_buffer_t *buffer)
155 {
156 buffer->skip_glyph ();
157 }
158
159 /* Returns 0 if didn't decompose, number of resulting characters otherwise. */
160 static inline unsigned int
161 decompose (const hb_ot_shape_normalize_context_t *c, bool shortest, hb_codepoint_t ab)
162 {
163 hb_codepoint_t a, b, a_glyph, b_glyph;
164 hb_buffer_t * const buffer = c->buffer;
165 hb_font_t * const font = c->font;
166
167 if (!c->decompose (c, ab, &a, &b) ||
168 (b && !font->get_glyph (b, 0, &b_glyph)))
169 return 0;
170
171 bool has_a = font->get_glyph (a, 0, &a_glyph);
172 if (shortest && has_a) {
173 /* Output a and b */
174 output_char (buffer, a, a_glyph);
175 if (likely (b)) {
176 output_char (buffer, b, b_glyph);
177 return 2;
178 }
179 return 1;
180 }
181
182 unsigned int ret;
183 if ((ret = decompose (c, shortest, a))) {
184 if (b) {
185 output_char (buffer, b, b_glyph);
186 return ret + 1;
187 }
188 return ret;
189 }
190
191 if (has_a) {
192 output_char (buffer, a, a_glyph);
193 if (likely (b)) {
194 output_char (buffer, b, b_glyph);
195 return 2;
196 }
197 return 1;
198 }
199
200 return 0;
201 }
202
203 /* Returns 0 if didn't decompose, number of resulting characters otherwise. */
204 static inline unsigned int
205 decompose_compatibility (const hb_ot_shape_normalize_context_t *c, hb_codepoint_t u)
206 {
207 unsigned int len, i;
208 hb_codepoint_t decomposed[HB_UNICODE_MAX_DECOMPOSITION_LEN];
209 hb_codepoint_t glyphs[HB_UNICODE_MAX_DECOMPOSITION_LEN];
210
211 len = c->buffer->unicode->decompose_compatibility (u, decomposed);
212 if (!len)
213 return 0;
214
215 for (i = 0; i < len; i++)
216 if (!c->font->get_glyph (decomposed[i], 0, &glyphs[i]))
217 return 0;
218
219 for (i = 0; i < len; i++)
220 output_char (c->buffer, decomposed[i], glyphs[i]);
221
222 return len;
223 }
224
225 static inline void
226 decompose_current_character (const hb_ot_shape_normalize_context_t *c, bool shortest)
227 {
228 hb_buffer_t * const buffer = c->buffer;
229 hb_codepoint_t u = buffer->cur().codepoint;
230 hb_codepoint_t glyph;
231
232 /* Kind of a cute waterfall here... */
233 if (shortest && c->font->get_glyph (u, 0, &glyph))
234 next_char (buffer, glyph);
235 else if (decompose (c, shortest, u))
236 skip_char (buffer);
237 else if (!shortest && c->font->get_glyph (u, 0, &glyph))
238 next_char (buffer, glyph);
239 else if (decompose_compatibility (c, u))
240 skip_char (buffer);
241 else
242 next_char (buffer, glyph); /* glyph is initialized in earlier branches. */
243 }
244
245 static inline void
246 handle_variation_selector_cluster (const hb_ot_shape_normalize_context_t *c, unsigned int end, bool short_circuit)
247 {
248 /* TODO Currently if there's a variation-selector we give-up, it's just too hard. */
249 hb_buffer_t * const buffer = c->buffer;
250 hb_font_t * const font = c->font;
251 for (; buffer->idx < end - 1;) {
252 if (unlikely (buffer->unicode->is_variation_selector (buffer->cur(+1).codepoint))) {
253 /* The next two lines are some ugly lines... But work. */
254 if (font->get_glyph (buffer->cur().codepoint, buffer->cur(+1).codepoint, &buffer->cur().glyph_index()))
255 {
256 buffer->replace_glyphs (2, 1, &buffer->cur().codepoint);
257 }
258 else
259 {
260 /* Just pass on the two characters separately, let GSUB do its magic. */
261 set_glyph (buffer->cur(), font);
262 buffer->next_glyph ();
263 set_glyph (buffer->cur(), font);
264 buffer->next_glyph ();
265 }
266 /* Skip any further variation selectors. */
267 while (buffer->idx < end && unlikely (buffer->unicode->is_variation_selector (buffer->cur().codepoint)))
268 {
269 set_glyph (buffer->cur(), font);
270 buffer->next_glyph ();
271 }
272 } else {
273 set_glyph (buffer->cur(), font);
274 buffer->next_glyph ();
275 }
276 }
277 if (likely (buffer->idx < end)) {
278 set_glyph (buffer->cur(), font);
279 buffer->next_glyph ();
280 }
281 }
282
283 static inline void
284 decompose_multi_char_cluster (const hb_ot_shape_normalize_context_t *c, unsigned int end, bool short_circuit)
285 {
286 hb_buffer_t * const buffer = c->buffer;
287 for (unsigned int i = buffer->idx; i < end; i++)
288 if (unlikely (buffer->unicode->is_variation_selector (buffer->info[i].codepoint))) {
289 handle_variation_selector_cluster (c, end, short_circuit);
290 return;
291 }
292
293 while (buffer->idx < end)
294 decompose_current_character (c, short_circuit);
295 }
296
297 static inline void
298 decompose_cluster (const hb_ot_shape_normalize_context_t *c, unsigned int end, bool might_short_circuit, bool always_short_circuit)
299 {
300 if (likely (c->buffer->idx + 1 == end))
301 decompose_current_character (c, might_short_circuit);
302 else
303 decompose_multi_char_cluster (c, end, always_short_circuit);
304 }
305
306
307 static int
308 compare_combining_class (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb)
309 {
310 unsigned int a = _hb_glyph_info_get_modified_combining_class (pa);
311 unsigned int b = _hb_glyph_info_get_modified_combining_class (pb);
312
313 return a < b ? -1 : a == b ? 0 : +1;
314 }
315
316
317 void
318 _hb_ot_shape_normalize (const hb_ot_shape_plan_t *plan,
319 hb_buffer_t *buffer,
320 hb_font_t *font)
321 {
322 if (unlikely (!buffer->len)) return;
323
324 _hb_buffer_assert_unicode_vars (buffer);
325
326 hb_ot_shape_normalization_mode_t mode = plan->shaper->normalization_preference;
327 const hb_ot_shape_normalize_context_t c = {
328 plan,
329 buffer,
330 font,
331 buffer->unicode,
332 plan->shaper->decompose ? plan->shaper->decompose : decompose_unicode,
333 plan->shaper->compose ? plan->shaper->compose : compose_unicode
334 };
335
336 bool always_short_circuit = mode == HB_OT_SHAPE_NORMALIZATION_MODE_NONE;
337 bool might_short_circuit = always_short_circuit ||
338 (mode != HB_OT_SHAPE_NORMALIZATION_MODE_DECOMPOSED &&
339 mode != HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT);
340 unsigned int count;
341
342 /* We do a fairly straightforward yet custom normalization process in three
343 * separate rounds: decompose, reorder, recompose (if desired). Currently
344 * this makes two buffer swaps. We can make it faster by moving the last
345 * two rounds into the inner loop for the first round, but it's more readable
346 * this way. */
347
348
349 /* First round, decompose */
350
351 buffer->clear_output ();
352 count = buffer->len;
353 for (buffer->idx = 0; buffer->idx < count;)
354 {
355 unsigned int end;
356 for (end = buffer->idx + 1; end < count; end++)
357 if (likely (!HB_UNICODE_GENERAL_CATEGORY_IS_MARK (_hb_glyph_info_get_general_category (&buffer->info[end]))))
358 break;
359
360 decompose_cluster (&c, end, might_short_circuit, always_short_circuit);
361 }
362 buffer->swap_buffers ();
363
364
365 /* Second round, reorder (inplace) */
366
367 count = buffer->len;
368 for (unsigned int i = 0; i < count; i++)
369 {
370 if (_hb_glyph_info_get_modified_combining_class (&buffer->info[i]) == 0)
371 continue;
372
373 unsigned int end;
374 for (end = i + 1; end < count; end++)
375 if (_hb_glyph_info_get_modified_combining_class (&buffer->info[end]) == 0)
376 break;
377
378 /* We are going to do a O(n^2). Only do this if the sequence is short. */
379 if (end - i > 10) {
380 i = end;
381 continue;
382 }
383
384 buffer->sort (i, end, compare_combining_class);
385
386 i = end;
387 }
388
389
390 if (mode == HB_OT_SHAPE_NORMALIZATION_MODE_NONE ||
391 mode == HB_OT_SHAPE_NORMALIZATION_MODE_DECOMPOSED)
392 return;
393
394 /* Third round, recompose */
395
396 /* As noted in the comment earlier, we don't try to combine
397 * ccc=0 chars with their previous Starter. */
398
399 buffer->clear_output ();
400 count = buffer->len;
401 unsigned int starter = 0;
402 buffer->next_glyph ();
403 while (buffer->idx < count)
404 {
405 hb_codepoint_t composed, glyph;
406 if (/* We don't try to compose a non-mark character with it's preceding starter.
407 * This is both an optimization to avoid trying to compose every two neighboring
408 * glyphs in most scripts AND a desired feature for Hangul. Apparently Hangul
409 * fonts are not designed to mix-and-match pre-composed syllables and Jamo. */
410 HB_UNICODE_GENERAL_CATEGORY_IS_MARK (_hb_glyph_info_get_general_category (&buffer->cur())) &&
411 /* If there's anything between the starter and this char, they should have CCC
412 * smaller than this character's. */
413 (starter == buffer->out_len - 1 ||
414 _hb_glyph_info_get_modified_combining_class (&buffer->prev()) < _hb_glyph_info_get_modified_combining_class (&buffer->cur())) &&
415 /* And compose. */
416 c.compose (&c,
417 buffer->out_info[starter].codepoint,
418 buffer->cur().codepoint,
419 &composed) &&
420 /* And the font has glyph for the composite. */
421 font->get_glyph (composed, 0, &glyph))
422 {
423 /* Composes. */
424 buffer->next_glyph (); /* Copy to out-buffer. */
425 if (unlikely (buffer->in_error))
426 return;
427 buffer->merge_out_clusters (starter, buffer->out_len);
428 buffer->out_len--; /* Remove the second composable. */
429 /* Modify starter and carry on. */
430 buffer->out_info[starter].codepoint = composed;
431 buffer->out_info[starter].glyph_index() = glyph;
432 _hb_glyph_info_set_unicode_props (&buffer->out_info[starter], buffer->unicode);
433
434 continue;
435 }
436
437 /* Blocked, or doesn't compose. */
438 buffer->next_glyph ();
439
440 if (_hb_glyph_info_get_modified_combining_class (&buffer->prev()) == 0)
441 starter = buffer->out_len - 1;
442 }
443 buffer->swap_buffers ();
444
445 }