Line data Source code
1 : /*
2 : * Copyright © 2011,2012 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-normalize-private.hh"
28 : #include "hb-ot-shape-complex-private.hh"
29 : #include "hb-ot-shape-private.hh"
30 :
31 :
32 : /*
33 : * HIGHLEVEL DESIGN:
34 : *
35 : * This file exports one main function: _hb_ot_shape_normalize().
36 : *
37 : * This function closely reflects the Unicode Normalization Algorithm,
38 : * yet it's different.
39 : *
40 : * Each shaper specifies whether it prefers decomposed (NFD) or composed (NFC).
41 : * The logic however tries to use whatever the font can support.
42 : *
43 : * In general what happens is that: each grapheme is decomposed in a chain
44 : * of 1:2 decompositions, marks reordered, and then recomposed if desired,
45 : * so far it's like Unicode Normalization. However, the decomposition and
46 : * recomposition only happens if the font supports the resulting characters.
47 : *
48 : * The goals are:
49 : *
50 : * - Try to render all canonically equivalent strings similarly. To really
51 : * achieve this we have to always do the full decomposition and then
52 : * selectively recompose from there. It's kinda too expensive though, so
53 : * we skip some cases. For example, if composed is desired, we simply
54 : * don't touch 1-character clusters that are supported by the font, even
55 : * though their NFC may be different.
56 : *
57 : * - When a font has a precomposed character for a sequence but the 'ccmp'
58 : * feature in the font is not adequate, use the precomposed character
59 : * which typically has better mark positioning.
60 : *
61 : * - When a font does not support a combining mark, but supports it precomposed
62 : * with previous base, use that. This needs the itemizer to have this
63 : * knowledge too. We need to provide assistance to the itemizer.
64 : *
65 : * - When a font does not support a character but supports its canonical
66 : * decomposition, well, use the decomposition.
67 : *
68 : * - The complex shapers can customize the compose and decompose functions to
69 : * offload some of their requirements to the normalizer. For example, the
70 : * Indic shaper may want to disallow recomposing of two matras.
71 : */
72 :
73 : static bool
74 0 : decompose_unicode (const hb_ot_shape_normalize_context_t *c,
75 : hb_codepoint_t ab,
76 : hb_codepoint_t *a,
77 : hb_codepoint_t *b)
78 : {
79 0 : return (bool) c->unicode->decompose (ab, a, b);
80 : }
81 :
82 : static bool
83 0 : compose_unicode (const hb_ot_shape_normalize_context_t *c,
84 : hb_codepoint_t a,
85 : hb_codepoint_t b,
86 : hb_codepoint_t *ab)
87 : {
88 0 : return (bool) c->unicode->compose (a, b, ab);
89 : }
90 :
91 : static inline void
92 0 : set_glyph (hb_glyph_info_t &info, hb_font_t *font)
93 : {
94 0 : (void) font->get_nominal_glyph (info.codepoint, &info.glyph_index());
95 0 : }
96 :
97 : static inline void
98 0 : output_char (hb_buffer_t *buffer, hb_codepoint_t unichar, hb_codepoint_t glyph)
99 : {
100 0 : buffer->cur().glyph_index() = glyph;
101 0 : buffer->output_glyph (unichar); /* This is very confusing indeed. */
102 0 : _hb_glyph_info_set_unicode_props (&buffer->prev(), buffer);
103 0 : }
104 :
105 : static inline void
106 265 : next_char (hb_buffer_t *buffer, hb_codepoint_t glyph)
107 : {
108 265 : buffer->cur().glyph_index() = glyph;
109 265 : buffer->next_glyph ();
110 265 : }
111 :
112 : static inline void
113 0 : skip_char (hb_buffer_t *buffer)
114 : {
115 0 : buffer->skip_glyph ();
116 0 : }
117 :
118 : /* Returns 0 if didn't decompose, number of resulting characters otherwise. */
119 : static inline unsigned int
120 0 : decompose (const hb_ot_shape_normalize_context_t *c, bool shortest, hb_codepoint_t ab)
121 : {
122 : hb_codepoint_t a, b, a_glyph, b_glyph;
123 0 : hb_buffer_t * const buffer = c->buffer;
124 0 : hb_font_t * const font = c->font;
125 :
126 0 : if (!c->decompose (c, ab, &a, &b) ||
127 0 : (b && !font->get_nominal_glyph (b, &b_glyph)))
128 0 : return 0;
129 :
130 0 : bool has_a = (bool) font->get_nominal_glyph (a, &a_glyph);
131 0 : if (shortest && has_a) {
132 : /* Output a and b */
133 0 : output_char (buffer, a, a_glyph);
134 0 : if (likely (b)) {
135 0 : output_char (buffer, b, b_glyph);
136 0 : return 2;
137 : }
138 0 : return 1;
139 : }
140 :
141 : unsigned int ret;
142 0 : if ((ret = decompose (c, shortest, a))) {
143 0 : if (b) {
144 0 : output_char (buffer, b, b_glyph);
145 0 : return ret + 1;
146 : }
147 0 : return ret;
148 : }
149 :
150 0 : if (has_a) {
151 0 : output_char (buffer, a, a_glyph);
152 0 : if (likely (b)) {
153 0 : output_char (buffer, b, b_glyph);
154 0 : return 2;
155 : }
156 0 : return 1;
157 : }
158 :
159 0 : return 0;
160 : }
161 :
162 : static inline void
163 265 : decompose_current_character (const hb_ot_shape_normalize_context_t *c, bool shortest)
164 : {
165 265 : hb_buffer_t * const buffer = c->buffer;
166 265 : hb_codepoint_t u = buffer->cur().codepoint;
167 : hb_codepoint_t glyph;
168 :
169 265 : if (shortest && c->font->get_nominal_glyph (u, &glyph))
170 : {
171 265 : next_char (buffer, glyph);
172 265 : return;
173 : }
174 :
175 0 : if (decompose (c, shortest, u))
176 : {
177 0 : skip_char (buffer);
178 0 : return;
179 : }
180 :
181 0 : if (!shortest && c->font->get_nominal_glyph (u, &glyph))
182 : {
183 0 : next_char (buffer, glyph);
184 0 : return;
185 : }
186 :
187 0 : if (_hb_glyph_info_is_unicode_space (&buffer->cur()))
188 : {
189 : hb_codepoint_t space_glyph;
190 0 : hb_unicode_funcs_t::space_t space_type = buffer->unicode->space_fallback_type (u);
191 0 : if (space_type != hb_unicode_funcs_t::NOT_SPACE && c->font->get_nominal_glyph (0x0020u, &space_glyph))
192 : {
193 0 : _hb_glyph_info_set_unicode_space_fallback_type (&buffer->cur(), space_type);
194 0 : next_char (buffer, space_glyph);
195 0 : buffer->scratch_flags |= HB_BUFFER_SCRATCH_FLAG_HAS_SPACE_FALLBACK;
196 0 : return;
197 : }
198 : }
199 :
200 0 : if (u == 0x2011u)
201 : {
202 : /* U+2011 is the only sensible character that is a no-break version of another character
203 : * and not a space. The space ones are handled already. Handle this lone one. */
204 : hb_codepoint_t other_glyph;
205 0 : if (c->font->get_nominal_glyph (0x2010u, &other_glyph))
206 : {
207 0 : next_char (buffer, other_glyph);
208 0 : return;
209 : }
210 : }
211 :
212 0 : next_char (buffer, glyph); /* glyph is initialized in earlier branches. */
213 : }
214 :
215 : static inline void
216 0 : handle_variation_selector_cluster (const hb_ot_shape_normalize_context_t *c, unsigned int end, bool short_circuit)
217 : {
218 : /* TODO Currently if there's a variation-selector we give-up, it's just too hard. */
219 0 : hb_buffer_t * const buffer = c->buffer;
220 0 : hb_font_t * const font = c->font;
221 0 : for (; buffer->idx < end - 1 && !buffer->in_error;) {
222 0 : if (unlikely (buffer->unicode->is_variation_selector (buffer->cur(+1).codepoint))) {
223 : /* The next two lines are some ugly lines... But work. */
224 0 : if (font->get_variation_glyph (buffer->cur().codepoint, buffer->cur(+1).codepoint, &buffer->cur().glyph_index()))
225 : {
226 0 : buffer->replace_glyphs (2, 1, &buffer->cur().codepoint);
227 : }
228 : else
229 : {
230 : /* Just pass on the two characters separately, let GSUB do its magic. */
231 0 : set_glyph (buffer->cur(), font);
232 0 : buffer->next_glyph ();
233 0 : set_glyph (buffer->cur(), font);
234 0 : buffer->next_glyph ();
235 : }
236 : /* Skip any further variation selectors. */
237 0 : while (buffer->idx < end && unlikely (buffer->unicode->is_variation_selector (buffer->cur().codepoint)))
238 : {
239 0 : set_glyph (buffer->cur(), font);
240 0 : buffer->next_glyph ();
241 : }
242 : } else {
243 0 : set_glyph (buffer->cur(), font);
244 0 : buffer->next_glyph ();
245 : }
246 : }
247 0 : if (likely (buffer->idx < end)) {
248 0 : set_glyph (buffer->cur(), font);
249 0 : buffer->next_glyph ();
250 : }
251 0 : }
252 :
253 : static inline void
254 0 : decompose_multi_char_cluster (const hb_ot_shape_normalize_context_t *c, unsigned int end, bool short_circuit)
255 : {
256 0 : hb_buffer_t * const buffer = c->buffer;
257 0 : for (unsigned int i = buffer->idx; i < end && !buffer->in_error; i++)
258 0 : if (unlikely (buffer->unicode->is_variation_selector (buffer->info[i].codepoint))) {
259 0 : handle_variation_selector_cluster (c, end, short_circuit);
260 0 : return;
261 : }
262 :
263 0 : while (buffer->idx < end && !buffer->in_error)
264 0 : decompose_current_character (c, short_circuit);
265 : }
266 :
267 : static inline void
268 265 : decompose_cluster (const hb_ot_shape_normalize_context_t *c, unsigned int end, bool might_short_circuit, bool always_short_circuit)
269 : {
270 265 : if (likely (c->buffer->idx + 1 == end))
271 265 : decompose_current_character (c, might_short_circuit);
272 : else
273 0 : decompose_multi_char_cluster (c, end, always_short_circuit);
274 265 : }
275 :
276 :
277 : static int
278 0 : compare_combining_class (const hb_glyph_info_t *pa, const hb_glyph_info_t *pb)
279 : {
280 0 : unsigned int a = _hb_glyph_info_get_modified_combining_class (pa);
281 0 : unsigned int b = _hb_glyph_info_get_modified_combining_class (pb);
282 :
283 0 : return a < b ? -1 : a == b ? 0 : +1;
284 : }
285 :
286 :
287 : void
288 34 : _hb_ot_shape_normalize (const hb_ot_shape_plan_t *plan,
289 : hb_buffer_t *buffer,
290 : hb_font_t *font)
291 : {
292 34 : if (unlikely (!buffer->len)) return;
293 :
294 34 : _hb_buffer_assert_unicode_vars (buffer);
295 :
296 34 : hb_ot_shape_normalization_mode_t mode = plan->shaper->normalization_preference;
297 : const hb_ot_shape_normalize_context_t c = {
298 : plan,
299 : buffer,
300 : font,
301 34 : buffer->unicode,
302 34 : plan->shaper->decompose ? plan->shaper->decompose : decompose_unicode,
303 34 : plan->shaper->compose ? plan->shaper->compose : compose_unicode
304 136 : };
305 :
306 34 : bool always_short_circuit = mode == HB_OT_SHAPE_NORMALIZATION_MODE_NONE;
307 68 : bool might_short_circuit = always_short_circuit ||
308 34 : (mode != HB_OT_SHAPE_NORMALIZATION_MODE_DECOMPOSED &&
309 34 : mode != HB_OT_SHAPE_NORMALIZATION_MODE_COMPOSED_DIACRITICS_NO_SHORT_CIRCUIT);
310 : unsigned int count;
311 :
312 : /* We do a fairly straightforward yet custom normalization process in three
313 : * separate rounds: decompose, reorder, recompose (if desired). Currently
314 : * this makes two buffer swaps. We can make it faster by moving the last
315 : * two rounds into the inner loop for the first round, but it's more readable
316 : * this way. */
317 :
318 :
319 : /* First round, decompose */
320 :
321 34 : buffer->clear_output ();
322 34 : count = buffer->len;
323 299 : for (buffer->idx = 0; buffer->idx < count && !buffer->in_error;)
324 : {
325 : unsigned int end;
326 265 : for (end = buffer->idx + 1; end < count; end++)
327 231 : if (likely (!HB_UNICODE_GENERAL_CATEGORY_IS_MARK (_hb_glyph_info_get_general_category (&buffer->info[end]))))
328 231 : break;
329 :
330 265 : decompose_cluster (&c, end, might_short_circuit, always_short_circuit);
331 : }
332 34 : buffer->swap_buffers ();
333 :
334 :
335 : /* Second round, reorder (inplace) */
336 :
337 34 : count = buffer->len;
338 299 : for (unsigned int i = 0; i < count; i++)
339 : {
340 265 : if (_hb_glyph_info_get_modified_combining_class (&buffer->info[i]) == 0)
341 265 : continue;
342 :
343 : unsigned int end;
344 0 : for (end = i + 1; end < count; end++)
345 0 : if (_hb_glyph_info_get_modified_combining_class (&buffer->info[end]) == 0)
346 0 : break;
347 :
348 : /* We are going to do a O(n^2). Only do this if the sequence is short. */
349 0 : if (end - i > 10) {
350 0 : i = end;
351 0 : continue;
352 : }
353 :
354 0 : buffer->sort (i, end, compare_combining_class);
355 :
356 0 : i = end;
357 : }
358 :
359 :
360 34 : if (mode == HB_OT_SHAPE_NORMALIZATION_MODE_NONE ||
361 : mode == HB_OT_SHAPE_NORMALIZATION_MODE_DECOMPOSED)
362 0 : return;
363 :
364 : /* Third round, recompose */
365 :
366 : /* As noted in the comment earlier, we don't try to combine
367 : * ccc=0 chars with their previous Starter. */
368 :
369 34 : buffer->clear_output ();
370 34 : count = buffer->len;
371 34 : unsigned int starter = 0;
372 34 : buffer->next_glyph ();
373 496 : while (buffer->idx < count && !buffer->in_error)
374 : {
375 : hb_codepoint_t composed, glyph;
376 231 : if (/* We don't try to compose a non-mark character with it's preceding starter.
377 : * This is both an optimization to avoid trying to compose every two neighboring
378 : * glyphs in most scripts AND a desired feature for Hangul. Apparently Hangul
379 : * fonts are not designed to mix-and-match pre-composed syllables and Jamo. */
380 231 : HB_UNICODE_GENERAL_CATEGORY_IS_MARK (_hb_glyph_info_get_general_category (&buffer->cur())) &&
381 : /* If there's anything between the starter and this char, they should have CCC
382 : * smaller than this character's. */
383 0 : (starter == buffer->out_len - 1 ||
384 231 : _hb_glyph_info_get_modified_combining_class (&buffer->prev()) < _hb_glyph_info_get_modified_combining_class (&buffer->cur())) &&
385 : /* And compose. */
386 0 : c.compose (&c,
387 0 : buffer->out_info[starter].codepoint,
388 0 : buffer->cur().codepoint,
389 231 : &composed) &&
390 : /* And the font has glyph for the composite. */
391 0 : font->get_nominal_glyph (composed, &glyph))
392 : {
393 : /* Composes. */
394 0 : buffer->next_glyph (); /* Copy to out-buffer. */
395 0 : if (unlikely (buffer->in_error))
396 0 : return;
397 0 : buffer->merge_out_clusters (starter, buffer->out_len);
398 0 : buffer->out_len--; /* Remove the second composable. */
399 : /* Modify starter and carry on. */
400 0 : buffer->out_info[starter].codepoint = composed;
401 0 : buffer->out_info[starter].glyph_index() = glyph;
402 0 : _hb_glyph_info_set_unicode_props (&buffer->out_info[starter], buffer);
403 :
404 0 : continue;
405 : }
406 :
407 : /* Blocked, or doesn't compose. */
408 231 : buffer->next_glyph ();
409 :
410 231 : if (_hb_glyph_info_get_modified_combining_class (&buffer->prev()) == 0)
411 231 : starter = buffer->out_len - 1;
412 : }
413 34 : buffer->swap_buffers ();
414 :
415 : }
|
