1 /*
2 * Copyright © 2022 Behdad Esfahbod
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.hh"
28
29 #ifndef HB_NO_BUFFER_VERIFY
30
31 #include "hb-buffer.hh"
32
33
34 #define BUFFER_VERIFY_ERROR "buffer verify error: "
35 static inline void
36 buffer_verify_error (hb_buffer_t *buffer,
37 hb_font_t *font,
38 const char *fmt,
39 ...) HB_PRINTF_FUNC(3, 4);
40
41 static inline void
buffer_verify_error(hb_buffer_t * buffer,hb_font_t * font,const char * fmt,...)42 buffer_verify_error (hb_buffer_t *buffer,
43 hb_font_t *font,
44 const char *fmt,
45 ...)
46 {
47 va_list ap;
48 va_start (ap, fmt);
49 if (buffer->messaging ())
50 {
51 buffer->message_impl (font, fmt, ap);
52 }
53 else
54 {
55 fprintf (stderr, "harfbuzz ");
56 vfprintf (stderr, fmt, ap);
57 fprintf (stderr, "\n");
58 }
59 va_end (ap);
60 }
61
62 static bool
buffer_verify_monotone(hb_buffer_t * buffer,hb_font_t * font)63 buffer_verify_monotone (hb_buffer_t *buffer,
64 hb_font_t *font)
65 {
66 /* Check that clusters are monotone. */
67 if (buffer->cluster_level == HB_BUFFER_CLUSTER_LEVEL_MONOTONE_GRAPHEMES ||
68 buffer->cluster_level == HB_BUFFER_CLUSTER_LEVEL_MONOTONE_CHARACTERS)
69 {
70 bool is_forward = HB_DIRECTION_IS_FORWARD (hb_buffer_get_direction (buffer));
71
72 unsigned int num_glyphs;
73 hb_glyph_info_t *info = hb_buffer_get_glyph_infos (buffer, &num_glyphs);
74
75 for (unsigned int i = 1; i < num_glyphs; i++)
76 if (info[i-1].cluster != info[i].cluster &&
77 (info[i-1].cluster < info[i].cluster) != is_forward)
78 {
79 buffer_verify_error (buffer, font, BUFFER_VERIFY_ERROR "clusters are not monotone.");
80 return false;
81 }
82 }
83
84 return true;
85 }
86
87 static bool
buffer_verify_unsafe_to_break(hb_buffer_t * buffer,hb_buffer_t * text_buffer,hb_font_t * font,const hb_feature_t * features,unsigned int num_features,const char * const * shapers)88 buffer_verify_unsafe_to_break (hb_buffer_t *buffer,
89 hb_buffer_t *text_buffer,
90 hb_font_t *font,
91 const hb_feature_t *features,
92 unsigned int num_features,
93 const char * const *shapers)
94 {
95 if (buffer->cluster_level != HB_BUFFER_CLUSTER_LEVEL_MONOTONE_GRAPHEMES &&
96 buffer->cluster_level != HB_BUFFER_CLUSTER_LEVEL_MONOTONE_CHARACTERS)
97 {
98 /* Cannot perform this check without monotone clusters. */
99 return true;
100 }
101
102 /* Check that breaking up shaping at safe-to-break is indeed safe. */
103
104 hb_buffer_t *fragment = hb_buffer_create_similar (buffer);
105 hb_buffer_set_flags (fragment, hb_buffer_get_flags (fragment) & ~HB_BUFFER_FLAG_VERIFY);
106 hb_buffer_t *reconstruction = hb_buffer_create_similar (buffer);
107 hb_buffer_set_flags (reconstruction, hb_buffer_get_flags (reconstruction) & ~HB_BUFFER_FLAG_VERIFY);
108
109 unsigned int num_glyphs;
110 hb_glyph_info_t *info = hb_buffer_get_glyph_infos (buffer, &num_glyphs);
111
112 unsigned int num_chars;
113 hb_glyph_info_t *text = hb_buffer_get_glyph_infos (text_buffer, &num_chars);
114
115 /* Chop text and shape fragments. */
116 bool forward = HB_DIRECTION_IS_FORWARD (hb_buffer_get_direction (buffer));
117 unsigned int start = 0;
118 unsigned int text_start = forward ? 0 : num_chars;
119 unsigned int text_end = text_start;
120 for (unsigned int end = 1; end < num_glyphs + 1; end++)
121 {
122 if (end < num_glyphs &&
123 (info[end].cluster == info[end-1].cluster ||
124 info[end-(forward?0:1)].mask & HB_GLYPH_FLAG_UNSAFE_TO_BREAK))
125 continue;
126
127 /* Shape segment corresponding to glyphs start..end. */
128 if (end == num_glyphs)
129 {
130 if (forward)
131 text_end = num_chars;
132 else
133 text_start = 0;
134 }
135 else
136 {
137 if (forward)
138 {
139 unsigned int cluster = info[end].cluster;
140 while (text_end < num_chars && text[text_end].cluster < cluster)
141 text_end++;
142 }
143 else
144 {
145 unsigned int cluster = info[end - 1].cluster;
146 while (text_start && text[text_start - 1].cluster >= cluster)
147 text_start--;
148 }
149 }
150 assert (text_start < text_end);
151
152 if (0)
153 printf("start %d end %d text start %d end %d\n", start, end, text_start, text_end);
154
155 hb_buffer_clear_contents (fragment);
156
157 hb_buffer_flags_t flags = hb_buffer_get_flags (fragment);
158 if (0 < text_start)
159 flags = (hb_buffer_flags_t) (flags & ~HB_BUFFER_FLAG_BOT);
160 if (text_end < num_chars)
161 flags = (hb_buffer_flags_t) (flags & ~HB_BUFFER_FLAG_EOT);
162 hb_buffer_set_flags (fragment, flags);
163
164 hb_buffer_append (fragment, text_buffer, text_start, text_end);
165 if (!hb_shape_full (font, fragment, features, num_features, shapers))
166 {
167 buffer_verify_error (buffer, font, BUFFER_VERIFY_ERROR "shaping failed while shaping fragment.");
168 hb_buffer_destroy (reconstruction);
169 hb_buffer_destroy (fragment);
170 return false;
171 }
172 hb_buffer_append (reconstruction, fragment, 0, -1);
173
174 start = end;
175 if (forward)
176 text_start = text_end;
177 else
178 text_end = text_start;
179 }
180
181 bool ret = true;
182 hb_buffer_diff_flags_t diff = hb_buffer_diff (reconstruction, buffer, (hb_codepoint_t) -1, 0);
183 if (diff)
184 {
185 buffer_verify_error (buffer, font, BUFFER_VERIFY_ERROR "unsafe-to-break test failed.");
186 ret = false;
187
188 /* Return the reconstructed result instead so it can be inspected. */
189 hb_buffer_set_length (buffer, 0);
190 hb_buffer_append (buffer, reconstruction, 0, -1);
191 }
192
193 hb_buffer_destroy (reconstruction);
194 hb_buffer_destroy (fragment);
195
196 return ret;
197 }
198
199 static bool
buffer_verify_unsafe_to_concat(hb_buffer_t * buffer,hb_buffer_t * text_buffer,hb_font_t * font,const hb_feature_t * features,unsigned int num_features,const char * const * shapers)200 buffer_verify_unsafe_to_concat (hb_buffer_t *buffer,
201 hb_buffer_t *text_buffer,
202 hb_font_t *font,
203 const hb_feature_t *features,
204 unsigned int num_features,
205 const char * const *shapers)
206 {
207 if (buffer->cluster_level != HB_BUFFER_CLUSTER_LEVEL_MONOTONE_GRAPHEMES &&
208 buffer->cluster_level != HB_BUFFER_CLUSTER_LEVEL_MONOTONE_CHARACTERS)
209 {
210 /* Cannot perform this check without monotone clusters. */
211 return true;
212 }
213
214 /* Check that shuffling up text before shaping at safe-to-concat points
215 * is indeed safe. */
216
217 /* This is what we do:
218 *
219 * 1. We shape text once. Then segment the text at all the safe-to-concat
220 * points;
221 *
222 * 2. Then we create two buffers, one containing all the even segments and
223 * one all the odd segments.
224 *
225 * 3. Because all these segments were safe-to-concat at both ends, we
226 * expect that concatenating them and shaping should NOT change the
227 * shaping results of each segment. As such, we expect that after
228 * shaping the two buffers, we still get cluster boundaries at the
229 * segment boundaries, and that those all are safe-to-concat points.
230 * Moreover, that there are NOT any safe-to-concat points within the
231 * segments.
232 *
233 * 4. Finally, we reconstruct the shaping results of the original text by
234 * simply interleaving the shaping results of the segments from the two
235 * buffers, and assert that the total shaping results is the same as
236 * the one from original buffer in step 1.
237 */
238
239 hb_buffer_t *fragments[2] {hb_buffer_create_similar (buffer),
240 hb_buffer_create_similar (buffer)};
241 hb_buffer_set_flags (fragments[0], hb_buffer_get_flags (fragments[0]) & ~HB_BUFFER_FLAG_VERIFY);
242 hb_buffer_set_flags (fragments[1], hb_buffer_get_flags (fragments[1]) & ~HB_BUFFER_FLAG_VERIFY);
243 hb_buffer_t *reconstruction = hb_buffer_create_similar (buffer);
244 hb_buffer_set_flags (reconstruction, hb_buffer_get_flags (reconstruction) & ~HB_BUFFER_FLAG_VERIFY);
245 hb_segment_properties_t props;
246 hb_buffer_get_segment_properties (buffer, &props);
247 hb_buffer_set_segment_properties (fragments[0], &props);
248 hb_buffer_set_segment_properties (fragments[1], &props);
249 hb_buffer_set_segment_properties (reconstruction, &props);
250
251 unsigned num_glyphs;
252 hb_glyph_info_t *info = hb_buffer_get_glyph_infos (buffer, &num_glyphs);
253
254 unsigned num_chars;
255 hb_glyph_info_t *text = hb_buffer_get_glyph_infos (text_buffer, &num_chars);
256
257 bool forward = HB_DIRECTION_IS_FORWARD (hb_buffer_get_direction (buffer));
258
259 if (!forward)
260 hb_buffer_reverse (buffer);
261
262 /*
263 * Split text into segments and collect into to fragment streams.
264 */
265 {
266 unsigned fragment_idx = 0;
267 unsigned start = 0;
268 unsigned text_start = 0;
269 unsigned text_end = 0;
270 for (unsigned end = 1; end < num_glyphs + 1; end++)
271 {
272 if (end < num_glyphs &&
273 (info[end].cluster == info[end-1].cluster ||
274 info[end].mask & HB_GLYPH_FLAG_UNSAFE_TO_CONCAT))
275 continue;
276
277 /* Accumulate segment corresponding to glyphs start..end. */
278 if (end == num_glyphs)
279 text_end = num_chars;
280 else
281 {
282 unsigned cluster = info[end].cluster;
283 while (text_end < num_chars && text[text_end].cluster < cluster)
284 text_end++;
285 }
286 assert (text_start < text_end);
287
288 if (0)
289 printf("start %d end %d text start %d end %d\n", start, end, text_start, text_end);
290
291 #if 0
292 hb_buffer_flags_t flags = hb_buffer_get_flags (fragment);
293 if (0 < text_start)
294 flags = (hb_buffer_flags_t) (flags & ~HB_BUFFER_FLAG_BOT);
295 if (text_end < num_chars)
296 flags = (hb_buffer_flags_t) (flags & ~HB_BUFFER_FLAG_EOT);
297 hb_buffer_set_flags (fragment, flags);
298 #endif
299
300 hb_buffer_append (fragments[fragment_idx], text_buffer, text_start, text_end);
301
302 start = end;
303 text_start = text_end;
304 fragment_idx = 1 - fragment_idx;
305 }
306 }
307
308 bool ret = true;
309 hb_buffer_diff_flags_t diff;
310
311 /*
312 * Shape the two fragment streams.
313 */
314 if (!hb_shape_full (font, fragments[0], features, num_features, shapers))
315 {
316 buffer_verify_error (buffer, font, BUFFER_VERIFY_ERROR "shaping failed while shaping fragment.");
317 ret = false;
318 goto out;
319 }
320 if (!hb_shape_full (font, fragments[1], features, num_features, shapers))
321 {
322 buffer_verify_error (buffer, font, BUFFER_VERIFY_ERROR "shaping failed while shaping fragment.");
323 ret = false;
324 goto out;
325 }
326
327 if (!forward)
328 {
329 hb_buffer_reverse (fragments[0]);
330 hb_buffer_reverse (fragments[1]);
331 }
332
333 /*
334 * Reconstruct results.
335 */
336 {
337 unsigned fragment_idx = 0;
338 unsigned fragment_start[2] {0, 0};
339 unsigned fragment_num_glyphs[2];
340 hb_glyph_info_t *fragment_info[2];
341 for (unsigned i = 0; i < 2; i++)
342 fragment_info[i] = hb_buffer_get_glyph_infos (fragments[i], &fragment_num_glyphs[i]);
343 while (fragment_start[0] < fragment_num_glyphs[0] ||
344 fragment_start[1] < fragment_num_glyphs[1])
345 {
346 unsigned fragment_end = fragment_start[fragment_idx] + 1;
347 while (fragment_end < fragment_num_glyphs[fragment_idx] &&
348 (fragment_info[fragment_idx][fragment_end].cluster == fragment_info[fragment_idx][fragment_end - 1].cluster ||
349 fragment_info[fragment_idx][fragment_end].mask & HB_GLYPH_FLAG_UNSAFE_TO_CONCAT))
350 fragment_end++;
351
352 hb_buffer_append (reconstruction, fragments[fragment_idx], fragment_start[fragment_idx], fragment_end);
353
354 fragment_start[fragment_idx] = fragment_end;
355 fragment_idx = 1 - fragment_idx;
356 }
357 }
358
359 if (!forward)
360 {
361 hb_buffer_reverse (buffer);
362 hb_buffer_reverse (reconstruction);
363 }
364
365 /*
366 * Diff results.
367 */
368 diff = hb_buffer_diff (reconstruction, buffer, (hb_codepoint_t) -1, 0);
369 if (diff)
370 {
371 buffer_verify_error (buffer, font, BUFFER_VERIFY_ERROR "unsafe-to-concat test failed.");
372 ret = false;
373
374 /* Return the reconstructed result instead so it can be inspected. */
375 hb_buffer_set_length (buffer, 0);
376 hb_buffer_append (buffer, reconstruction, 0, -1);
377 }
378
379
380 out:
381 hb_buffer_destroy (reconstruction);
382 hb_buffer_destroy (fragments[0]);
383 hb_buffer_destroy (fragments[1]);
384
385 return ret;
386 }
387
388 bool
verify(hb_buffer_t * text_buffer,hb_font_t * font,const hb_feature_t * features,unsigned int num_features,const char * const * shapers)389 hb_buffer_t::verify (hb_buffer_t *text_buffer,
390 hb_font_t *font,
391 const hb_feature_t *features,
392 unsigned int num_features,
393 const char * const *shapers)
394 {
395 bool ret = true;
396 if (!buffer_verify_monotone (this, font))
397 ret = false;
398 if (!buffer_verify_unsafe_to_break (this, text_buffer, font, features, num_features, shapers))
399 ret = false;
400 if (!buffer_verify_unsafe_to_concat (this, text_buffer, font, features, num_features, shapers))
401 ret = false;
402 if (!ret)
403 {
404 unsigned len = text_buffer->len;
405 hb_vector_t<char> bytes;
406 if (likely (bytes.resize (len * 10 + 16)))
407 {
408 hb_buffer_serialize_unicode (text_buffer,
409 0, len,
410 bytes.arrayZ, bytes.length,
411 &len,
412 HB_BUFFER_SERIALIZE_FORMAT_TEXT,
413 HB_BUFFER_SERIALIZE_FLAG_NO_CLUSTERS);
414 buffer_verify_error (this, font, BUFFER_VERIFY_ERROR "text was: %s.", bytes.arrayZ);
415 }
416 }
417 return ret;
418 }
419
420
421 #endif
422