1 /**********************************************************************
2 * File: pango_font_info.cpp
3 * Description: Font-related objects and helper functions
4 * Author: Ranjith Unnikrishnan
5 *
6 * (C) Copyright 2013, Google Inc.
7 * Licensed under the Apache License, Version 2.0 (the "License");
8 * you may not use this file except in compliance with the License.
9 * You may obtain a copy of the License at
10 * http://www.apache.org/licenses/LICENSE-2.0
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
16 *
17 **********************************************************************/
18
19 // Include automatically generated configuration file if running autoconf.
20 #ifdef HAVE_CONFIG_H
21 # include "config_auto.h"
22 #endif
23
24 #if (defined __MINGW32__) || (defined __CYGWIN__)
25 // workaround for stdlib.h and putenv
26 # undef __STRICT_ANSI__
27 #endif
28
29 #include "commandlineflags.h"
30 #include "fileio.h"
31 #include "normstrngs.h"
32 #include "pango_font_info.h"
33 #include "tlog.h"
34
35 #include <tesseract/unichar.h>
36
37 #include "pango/pango.h"
38 #include "pango/pangocairo.h"
39 #include "pango/pangofc-font.h"
40
41 #include <algorithm>
42 #include <cstdio>
43 #include <cstdlib>
44 #include <cstring>
45
46 #ifndef _MSC_VER
47 # include <sys/param.h>
48 #endif
49
50 #define DISABLE_HEAP_LEAK_CHECK
51
52 using namespace tesseract;
53
54 namespace tesseract {
55
56 // Default assumed output resolution. Required only for providing font metrics
57 // in pixels.
58 const int kDefaultResolution = 300;
59
60 std::string PangoFontInfo::fonts_dir_;
61 std::string PangoFontInfo::cache_dir_;
62
get_glyph(PangoFont * font,gunichar wc)63 static PangoGlyph get_glyph(PangoFont *font, gunichar wc) {
64 #if PANGO_VERSION_CHECK(1, 44, 0)
65 // pango_font_get_hb_font requires Pango 1.44 or newer.
66 hb_font_t *hb_font = pango_font_get_hb_font(font);
67 hb_codepoint_t glyph;
68 hb_font_get_nominal_glyph(hb_font, wc, &glyph);
69 #else
70 // Use deprecated pango_fc_font_get_glyph for older Pango versions.
71 PangoGlyph glyph = pango_fc_font_get_glyph(PANGO_FC_FONT(font), wc);
72 #endif
73 return glyph;
74 }
75
PangoFontInfo()76 PangoFontInfo::PangoFontInfo() : desc_(nullptr), resolution_(kDefaultResolution) {
77 Clear();
78 }
79
PangoFontInfo(const std::string & desc)80 PangoFontInfo::PangoFontInfo(const std::string &desc)
81 : desc_(nullptr), resolution_(kDefaultResolution) {
82 if (!ParseFontDescriptionName(desc)) {
83 tprintf("ERROR: Could not parse %s\n", desc.c_str());
84 Clear();
85 }
86 }
87
Clear()88 void PangoFontInfo::Clear() {
89 font_size_ = 0;
90 family_name_.clear();
91 font_type_ = UNKNOWN;
92 if (desc_) {
93 pango_font_description_free(desc_);
94 desc_ = nullptr;
95 }
96 }
97
~PangoFontInfo()98 PangoFontInfo::~PangoFontInfo() {
99 pango_font_description_free(desc_);
100 }
101
DescriptionName() const102 std::string PangoFontInfo::DescriptionName() const {
103 if (!desc_) {
104 return "";
105 }
106 char *desc_str = pango_font_description_to_string(desc_);
107 std::string desc_name(desc_str);
108 g_free(desc_str);
109 return desc_name;
110 }
111
112 // If not already initialized, initializes FontConfig by setting its
113 // environment variable and creating a fonts.conf file that points to the
114 // FLAGS_fonts_dir and the cache to FLAGS_fontconfig_tmpdir.
115 /* static */
SoftInitFontConfig()116 void PangoFontInfo::SoftInitFontConfig() {
117 if (fonts_dir_.empty()) {
118 HardInitFontConfig(FLAGS_fonts_dir.c_str(), FLAGS_fontconfig_tmpdir.c_str());
119 }
120 }
121
122 // Re-initializes font config, whether or not already initialized.
123 // If already initialized, any existing cache is deleted, just to be sure.
124 /* static */
HardInitFontConfig(const char * fonts_dir,const char * cache_dir)125 void PangoFontInfo::HardInitFontConfig(const char *fonts_dir, const char *cache_dir) {
126 if (!cache_dir_.empty()) {
127 File::DeleteMatchingFiles(File::JoinPath(cache_dir_.c_str(), "*cache-?").c_str());
128 }
129 const int MAX_FONTCONF_FILESIZE = 1024;
130 char fonts_conf_template[MAX_FONTCONF_FILESIZE];
131 cache_dir_ = cache_dir;
132 fonts_dir_ = fonts_dir;
133 snprintf(fonts_conf_template, MAX_FONTCONF_FILESIZE,
134 "<?xml version=\"1.0\"?>\n"
135 "<!DOCTYPE fontconfig SYSTEM \"fonts.dtd\">\n"
136 "<fontconfig>\n"
137 "<dir>%s</dir>\n"
138 "<cachedir>%s</cachedir>\n"
139 "<config></config>\n"
140 "</fontconfig>\n",
141 fonts_dir, cache_dir);
142 std::string fonts_conf_file = File::JoinPath(cache_dir, "fonts.conf");
143 File::WriteStringToFileOrDie(fonts_conf_template, fonts_conf_file);
144 #ifdef _WIN32
145 std::string env("FONTCONFIG_PATH=");
146 env.append(cache_dir);
147 _putenv(env.c_str());
148 _putenv("LANG=en_US.utf8");
149 #else
150 setenv("FONTCONFIG_PATH", cache_dir, true);
151 // Fix the locale so that the reported font names are consistent.
152 setenv("LANG", "en_US.utf8", true);
153 #endif // _WIN32
154
155 if (FcInitReinitialize() != FcTrue) {
156 tprintf("FcInitiReinitialize failed!!\n");
157 }
158 FontUtils::ReInit();
159 // Clear Pango's font cache too.
160 pango_cairo_font_map_set_default(nullptr);
161 }
162
ListFontFamilies(PangoFontFamily *** families,int * n_families)163 static void ListFontFamilies(PangoFontFamily ***families, int *n_families) {
164 PangoFontInfo::SoftInitFontConfig();
165 PangoFontMap *font_map = pango_cairo_font_map_get_default();
166 DISABLE_HEAP_LEAK_CHECK;
167 pango_font_map_list_families(font_map, families, n_families);
168 }
169
ParseFontDescription(const PangoFontDescription * desc)170 bool PangoFontInfo::ParseFontDescription(const PangoFontDescription *desc) {
171 Clear();
172 const char *family = pango_font_description_get_family(desc);
173 if (!family) {
174 char *desc_str = pango_font_description_to_string(desc);
175 tprintf("WARNING: Could not parse family name from description: '%s'\n", desc_str);
176 g_free(desc_str);
177 return false;
178 }
179 family_name_ = std::string(family);
180 desc_ = pango_font_description_copy(desc);
181
182 // Set font size in points
183 font_size_ = pango_font_description_get_size(desc);
184 if (!pango_font_description_get_size_is_absolute(desc)) {
185 font_size_ /= PANGO_SCALE;
186 }
187
188 return true;
189 }
190
ParseFontDescriptionName(const std::string & name)191 bool PangoFontInfo::ParseFontDescriptionName(const std::string &name) {
192 PangoFontDescription *desc = pango_font_description_from_string(name.c_str());
193 bool success = ParseFontDescription(desc);
194 pango_font_description_free(desc);
195 return success;
196 }
197
198 // Returns the PangoFont structure corresponding to the closest available font
199 // in the font map. Note that if the font is wholly missing, this could
200 // correspond to a completely different font family and face.
ToPangoFont() const201 PangoFont *PangoFontInfo::ToPangoFont() const {
202 SoftInitFontConfig();
203 PangoFontMap *font_map = pango_cairo_font_map_get_default();
204 PangoContext *context = pango_context_new();
205 pango_cairo_context_set_resolution(context, resolution_);
206 pango_context_set_font_map(context, font_map);
207 PangoFont *font = nullptr;
208 {
209 DISABLE_HEAP_LEAK_CHECK;
210 font = pango_font_map_load_font(font_map, context, desc_);
211 }
212 g_object_unref(context);
213 return font;
214 }
215
CoversUTF8Text(const char * utf8_text,int byte_length) const216 bool PangoFontInfo::CoversUTF8Text(const char *utf8_text, int byte_length) const {
217 PangoFont *font = ToPangoFont();
218 if (font == nullptr) {
219 // Font not found.
220 return false;
221 }
222 PangoCoverage *coverage = pango_font_get_coverage(font, nullptr);
223 for (UNICHAR::const_iterator it = UNICHAR::begin(utf8_text, byte_length);
224 it != UNICHAR::end(utf8_text, byte_length); ++it) {
225 if (IsWhitespace(*it) || pango_is_zero_width(*it)) {
226 continue;
227 }
228 if (pango_coverage_get(coverage, *it) != PANGO_COVERAGE_EXACT) {
229 char tmp[5];
230 int len = it.get_utf8(tmp);
231 tmp[len] = '\0';
232 tlog(2, "'%s' (U+%x) not covered by font\n", tmp, *it);
233 pango_coverage_unref(coverage);
234 g_object_unref(font);
235 return false;
236 }
237 }
238 pango_coverage_unref(coverage);
239 g_object_unref(font);
240 return true;
241 }
242
243 // This variant of strncpy permits src and dest to overlap. It will copy the
244 // first byte first.
my_strnmove(char * dest,const char * src,size_t n)245 static char *my_strnmove(char *dest, const char *src, size_t n) {
246 char *ret = dest;
247
248 // Copy characters until n reaches zero or the src byte is a nul.
249 do {
250 *dest = *src;
251 --n;
252 ++dest;
253 ++src;
254 } while (n && src[0]);
255
256 // If we reached a nul byte and there are more 'n' left, zero them out.
257 while (n) {
258 *dest = '\0';
259 --n;
260 ++dest;
261 }
262 return ret;
263 }
264
DropUncoveredChars(std::string * utf8_text) const265 int PangoFontInfo::DropUncoveredChars(std::string *utf8_text) const {
266 int num_dropped_chars = 0;
267 PangoFont *font = ToPangoFont();
268 if (font == nullptr) {
269 // Font not found, drop all characters.
270 num_dropped_chars = utf8_text->length();
271 utf8_text->clear();
272 return num_dropped_chars;
273 }
274 PangoCoverage *coverage = pango_font_get_coverage(font, nullptr);
275 // Maintain two iterators that point into the string. For space efficiency, we
276 // will repeatedly copy one covered UTF8 character from one to the other, and
277 // at the end resize the string to the right length.
278 char *out = const_cast<char *>(utf8_text->c_str());
279 const UNICHAR::const_iterator it_begin = UNICHAR::begin(utf8_text->c_str(), utf8_text->length());
280 const UNICHAR::const_iterator it_end = UNICHAR::end(utf8_text->c_str(), utf8_text->length());
281 for (UNICHAR::const_iterator it = it_begin; it != it_end;) {
282 // Skip bad utf-8.
283 if (!it.is_legal()) {
284 ++it; // One suitable error message will still be issued.
285 continue;
286 }
287 int unicode = *it;
288 int utf8_len = it.utf8_len();
289 const char *utf8_char = it.utf8_data();
290 // Move it forward before the data gets modified.
291 ++it;
292 if (!IsWhitespace(unicode) && !pango_is_zero_width(unicode) &&
293 pango_coverage_get(coverage, unicode) != PANGO_COVERAGE_EXACT) {
294 if (TLOG_IS_ON(2)) {
295 UNICHAR unichar(unicode);
296 char *str = unichar.utf8_str();
297 tlog(2, "'%s' (U+%x) not covered by font\n", str, unicode);
298 delete[] str;
299 }
300 ++num_dropped_chars;
301 continue;
302 }
303 my_strnmove(out, utf8_char, utf8_len);
304 out += utf8_len;
305 }
306 pango_coverage_unref(coverage);
307 g_object_unref(font);
308 utf8_text->resize(out - utf8_text->c_str());
309 return num_dropped_chars;
310 }
311
GetSpacingProperties(const std::string & utf8_char,int * x_bearing,int * x_advance) const312 bool PangoFontInfo::GetSpacingProperties(const std::string &utf8_char, int *x_bearing,
313 int *x_advance) const {
314 // Convert to equivalent PangoFont structure
315 PangoFont *font = ToPangoFont();
316 if (!font) {
317 return false;
318 }
319 // Find the glyph index in the font for the supplied utf8 character.
320 int total_advance = 0;
321 int min_bearing = 0;
322 // Handle multi-unicode strings by reporting the left-most position of the
323 // x-bearing, and right-most position of the x-advance if the string were to
324 // be rendered.
325 const UNICHAR::const_iterator it_begin = UNICHAR::begin(utf8_char.c_str(), utf8_char.length());
326 const UNICHAR::const_iterator it_end = UNICHAR::end(utf8_char.c_str(), utf8_char.length());
327 for (UNICHAR::const_iterator it = it_begin; it != it_end; ++it) {
328 PangoGlyph glyph_index = get_glyph(font, *it);
329 if (!glyph_index) {
330 // Glyph for given unicode character doesn't exist in font.
331 g_object_unref(font);
332 return false;
333 }
334 // Find the ink glyph extents for the glyph
335 PangoRectangle ink_rect, logical_rect;
336 pango_font_get_glyph_extents(font, glyph_index, &ink_rect, &logical_rect);
337 pango_extents_to_pixels(&ink_rect, nullptr);
338 pango_extents_to_pixels(&logical_rect, nullptr);
339
340 int bearing = total_advance + PANGO_LBEARING(ink_rect);
341 if (it == it_begin || bearing < min_bearing) {
342 min_bearing = bearing;
343 }
344 total_advance += PANGO_RBEARING(logical_rect);
345 }
346 *x_bearing = min_bearing;
347 *x_advance = total_advance;
348 g_object_unref(font);
349 return true;
350 }
351
CanRenderString(const char * utf8_word,int len) const352 bool PangoFontInfo::CanRenderString(const char *utf8_word, int len) const {
353 std::vector<std::string> graphemes;
354 return CanRenderString(utf8_word, len, &graphemes);
355 }
356
CanRenderString(const char * utf8_word,int len,std::vector<std::string> * graphemes) const357 bool PangoFontInfo::CanRenderString(const char *utf8_word, int len,
358 std::vector<std::string> *graphemes) const {
359 if (graphemes) {
360 graphemes->clear();
361 }
362 // We check for font coverage of the text first, as otherwise Pango could
363 // (undesirably) fall back to another font that does have the required
364 // coverage.
365 if (!CoversUTF8Text(utf8_word, len)) {
366 return false;
367 }
368 // U+25CC dotted circle character that often (but not always) gets rendered
369 // when there is an illegal grapheme sequence.
370 const char32 kDottedCircleGlyph = 9676;
371 bool bad_glyph = false;
372 PangoFontMap *font_map = pango_cairo_font_map_get_default();
373 PangoContext *context = pango_context_new();
374 pango_context_set_font_map(context, font_map);
375 PangoLayout *layout;
376 {
377 // Pango is not releasing the cached layout.
378 DISABLE_HEAP_LEAK_CHECK;
379 layout = pango_layout_new(context);
380 }
381 if (desc_) {
382 pango_layout_set_font_description(layout, desc_);
383 } else {
384 PangoFontDescription *desc = pango_font_description_from_string(DescriptionName().c_str());
385 pango_layout_set_font_description(layout, desc);
386 pango_font_description_free(desc);
387 }
388 pango_layout_set_text(layout, utf8_word, len);
389 PangoLayoutIter *run_iter = nullptr;
390 { // Fontconfig caches some information here that is not freed before exit.
391 DISABLE_HEAP_LEAK_CHECK;
392 run_iter = pango_layout_get_iter(layout);
393 }
394 do {
395 PangoLayoutRun *run = pango_layout_iter_get_run_readonly(run_iter);
396 if (!run) {
397 tlog(2, "Found end of line nullptr run marker\n");
398 continue;
399 }
400 PangoGlyph dotted_circle_glyph;
401 PangoFont *font = run->item->analysis.font;
402
403 dotted_circle_glyph = get_glyph(font, kDottedCircleGlyph);
404
405 if (TLOG_IS_ON(2)) {
406 PangoFontDescription *desc = pango_font_describe(font);
407 char *desc_str = pango_font_description_to_string(desc);
408 tlog(2, "Desc of font in run: %s\n", desc_str);
409 g_free(desc_str);
410 pango_font_description_free(desc);
411 }
412
413 PangoGlyphItemIter cluster_iter;
414 gboolean have_cluster;
415 for (have_cluster = pango_glyph_item_iter_init_start(&cluster_iter, run, utf8_word);
416 have_cluster && !bad_glyph;
417 have_cluster = pango_glyph_item_iter_next_cluster(&cluster_iter)) {
418 const int start_byte_index = cluster_iter.start_index;
419 const int end_byte_index = cluster_iter.end_index;
420 int start_glyph_index = cluster_iter.start_glyph;
421 int end_glyph_index = cluster_iter.end_glyph;
422 std::string cluster_text =
423 std::string(utf8_word + start_byte_index, end_byte_index - start_byte_index);
424 if (graphemes) {
425 graphemes->push_back(cluster_text);
426 }
427 if (IsUTF8Whitespace(cluster_text.c_str())) {
428 tlog(2, "Skipping whitespace\n");
429 continue;
430 }
431 if (TLOG_IS_ON(2)) {
432 printf("start_byte=%d end_byte=%d start_glyph=%d end_glyph=%d ", start_byte_index,
433 end_byte_index, start_glyph_index, end_glyph_index);
434 }
435 for (int i = start_glyph_index, step = (end_glyph_index > start_glyph_index) ? 1 : -1;
436 !bad_glyph && i != end_glyph_index; i += step) {
437 const bool unknown_glyph =
438 (cluster_iter.glyph_item->glyphs->glyphs[i].glyph & PANGO_GLYPH_UNKNOWN_FLAG);
439 const bool illegal_glyph =
440 (cluster_iter.glyph_item->glyphs->glyphs[i].glyph == dotted_circle_glyph);
441 bad_glyph = unknown_glyph || illegal_glyph;
442 if (TLOG_IS_ON(2)) {
443 printf("(%d=%d)", cluster_iter.glyph_item->glyphs->glyphs[i].glyph, bad_glyph ? 1 : 0);
444 }
445 }
446 if (TLOG_IS_ON(2)) {
447 printf(" '%s'\n", cluster_text.c_str());
448 }
449 if (bad_glyph)
450 tlog(1, "Found illegal glyph!\n");
451 }
452 } while (!bad_glyph && pango_layout_iter_next_run(run_iter));
453
454 pango_layout_iter_free(run_iter);
455 g_object_unref(context);
456 g_object_unref(layout);
457 if (bad_glyph && graphemes) {
458 graphemes->clear();
459 }
460 return !bad_glyph;
461 }
462
463 // ------------------------ FontUtils ------------------------------------
464 std::vector<std::string> FontUtils::available_fonts_; // cache list
465
466 // Returns whether the specified font description is available in the fonts
467 // directory.
468 //
469 // The generated list of font families and faces includes "synthesized" font
470 // faces that are not truly loadable. Pango versions >=1.18 have a
471 // pango_font_face_is_synthesized method that can be used to prune the list.
472 // Until then, we are restricted to using a hack where we try to load the font
473 // from the font_map, and then check what we loaded to see if it has the
474 // description we expected. If it is not, then the font is deemed unavailable.
475 //
476 // TODO: This function reports also some not synthesized fonts as not available
477 // e.g. 'Bitstream Charter Medium Italic', 'LMRoman17', so we need this hack
478 // until other solution is found.
479 /* static */
IsAvailableFont(const char * input_query_desc,std::string * best_match)480 bool FontUtils::IsAvailableFont(const char *input_query_desc, std::string *best_match) {
481 std::string query_desc(input_query_desc);
482 PangoFontDescription *desc = pango_font_description_from_string(query_desc.c_str());
483 PangoFont *selected_font = nullptr;
484 {
485 PangoFontInfo::SoftInitFontConfig();
486 PangoFontMap *font_map = pango_cairo_font_map_get_default();
487 PangoContext *context = pango_context_new();
488 pango_context_set_font_map(context, font_map);
489 {
490 DISABLE_HEAP_LEAK_CHECK;
491 selected_font = pango_font_map_load_font(font_map, context, desc);
492 }
493 g_object_unref(context);
494 }
495 if (selected_font == nullptr) {
496 pango_font_description_free(desc);
497 tlog(4, "** Font '%s' failed to load from font map!\n", input_query_desc);
498 return false;
499 }
500 PangoFontDescription *selected_desc = pango_font_describe(selected_font);
501
502 bool equal = pango_font_description_equal(desc, selected_desc);
503 tlog(3, "query weight = %d \t selected weight =%d\n", pango_font_description_get_weight(desc),
504 pango_font_description_get_weight(selected_desc));
505
506 char *selected_desc_str = pango_font_description_to_string(selected_desc);
507 tlog(2, "query_desc: '%s' Selected: '%s'\n", query_desc.c_str(), selected_desc_str);
508 if (!equal && best_match != nullptr) {
509 *best_match = selected_desc_str;
510 // Clip the ending ' 0' if there is one. It seems that, if there is no
511 // point size on the end of the fontname, then Pango always appends ' 0'.
512 int len = best_match->size();
513 if (len > 2 && best_match->at(len - 1) == '0' && best_match->at(len - 2) == ' ') {
514 *best_match = best_match->substr(0, len - 2);
515 }
516 }
517 g_free(selected_desc_str);
518 pango_font_description_free(selected_desc);
519 g_object_unref(selected_font);
520 pango_font_description_free(desc);
521 if (!equal)
522 tlog(4, "** Font '%s' failed pango_font_description_equal!\n", input_query_desc);
523 return equal;
524 }
525
ShouldIgnoreFontFamilyName(const char * query)526 static bool ShouldIgnoreFontFamilyName(const char *query) {
527 static const char *kIgnoredFamilyNames[] = {"Sans", "Serif", "Monospace", nullptr};
528 const char **list = kIgnoredFamilyNames;
529 for (; *list != nullptr; ++list) {
530 if (!strcmp(*list, query)) {
531 return true;
532 }
533 }
534 return false;
535 }
536
537 // Outputs description names of available fonts.
538 /* static */
ListAvailableFonts()539 const std::vector<std::string> &FontUtils::ListAvailableFonts() {
540 if (!available_fonts_.empty()) {
541 return available_fonts_;
542 }
543
544 PangoFontFamily **families = nullptr;
545 int n_families = 0;
546 ListFontFamilies(&families, &n_families);
547 for (int i = 0; i < n_families; ++i) {
548 const char *family_name = pango_font_family_get_name(families[i]);
549 tlog(2, "Listing family %s\n", family_name);
550 if (ShouldIgnoreFontFamilyName(family_name)) {
551 continue;
552 }
553
554 int n_faces;
555 PangoFontFace **faces = nullptr;
556 pango_font_family_list_faces(families[i], &faces, &n_faces);
557 for (int j = 0; j < n_faces; ++j) {
558 PangoFontDescription *desc = pango_font_face_describe(faces[j]);
559 char *desc_str = pango_font_description_to_string(desc);
560 // "synthesized" font faces that are not truly loadable, so we skip it
561 if (!pango_font_face_is_synthesized(faces[j]) && IsAvailableFont(desc_str)) {
562 available_fonts_.emplace_back(desc_str);
563 }
564 pango_font_description_free(desc);
565 g_free(desc_str);
566 }
567 g_free(faces);
568 }
569 g_free(families);
570 std::sort(available_fonts_.begin(), available_fonts_.end());
571 return available_fonts_;
572 }
573
574 // Utilities written to be backward compatible with StringRender
575
576 /* static */
FontScore(const std::unordered_map<char32,int64_t> & ch_map,const std::string & fontname,int * raw_score,std::vector<bool> * ch_flags)577 int FontUtils::FontScore(const std::unordered_map<char32, int64_t> &ch_map,
578 const std::string &fontname, int *raw_score, std::vector<bool> *ch_flags) {
579 PangoFontInfo font_info;
580 if (!font_info.ParseFontDescriptionName(fontname)) {
581 tprintf("ERROR: Could not parse %s\n", fontname.c_str());
582 }
583 PangoFont *font = font_info.ToPangoFont();
584 PangoCoverage *coverage = nullptr;
585 if (font != nullptr) {
586 coverage = pango_font_get_coverage(font, nullptr);
587 }
588 if (ch_flags) {
589 ch_flags->clear();
590 ch_flags->reserve(ch_map.size());
591 }
592 *raw_score = 0;
593 int ok_chars = 0;
594 for (auto it : ch_map) {
595 bool covered =
596 (coverage != nullptr) && (IsWhitespace(it.first) ||
597 (pango_coverage_get(coverage, it.first) == PANGO_COVERAGE_EXACT));
598 if (covered) {
599 ++(*raw_score);
600 ok_chars += it.second;
601 }
602 if (ch_flags) {
603 ch_flags->push_back(covered);
604 }
605 }
606 pango_coverage_unref(coverage);
607 g_object_unref(font);
608 return ok_chars;
609 }
610
611 /* static */
BestFonts(const std::unordered_map<char32,int64_t> & ch_map,std::vector<std::pair<const char *,std::vector<bool>>> * fonts)612 std::string FontUtils::BestFonts(const std::unordered_map<char32, int64_t> &ch_map,
613 std::vector<std::pair<const char *, std::vector<bool>>> *fonts) {
614 const double kMinOKFraction = 0.99;
615 // Weighted fraction of characters that must be renderable in a font to make
616 // it OK even if the raw count is not good.
617 const double kMinWeightedFraction = 0.99995;
618
619 fonts->clear();
620 std::vector<std::vector<bool>> font_flags;
621 std::vector<int> font_scores;
622 std::vector<int> raw_scores;
623 int most_ok_chars = 0;
624 int best_raw_score = 0;
625 const std::vector<std::string> &font_names = FontUtils::ListAvailableFonts();
626 for (const auto &font_name : font_names) {
627 std::vector<bool> ch_flags;
628 int raw_score = 0;
629 int ok_chars = FontScore(ch_map, font_name, &raw_score, &ch_flags);
630 most_ok_chars = std::max(ok_chars, most_ok_chars);
631 best_raw_score = std::max(raw_score, best_raw_score);
632
633 font_flags.push_back(ch_flags);
634 font_scores.push_back(ok_chars);
635 raw_scores.push_back(raw_score);
636 }
637
638 // Now select the fonts with a score above a threshold fraction
639 // of both the raw and weighted best scores. To prevent bogus fonts being
640 // selected for CJK, we require a high fraction (kMinOKFraction = 0.99) of
641 // BOTH weighted and raw scores.
642 // In low character-count scripts, the issue is more getting enough fonts,
643 // when only 1 or 2 might have all those rare dingbats etc in them, so we
644 // allow a font with a very high weighted (coverage) score
645 // (kMinWeightedFraction = 0.99995) to be used even if its raw score is poor.
646 int least_good_enough = static_cast<int>(most_ok_chars * kMinOKFraction);
647 int least_raw_enough = static_cast<int>(best_raw_score * kMinOKFraction);
648 int override_enough = static_cast<int>(most_ok_chars * kMinWeightedFraction);
649
650 std::string font_list;
651 for (unsigned i = 0; i < font_names.size(); ++i) {
652 int score = font_scores[i];
653 int raw_score = raw_scores[i];
654 if ((score >= least_good_enough && raw_score >= least_raw_enough) || score >= override_enough) {
655 fonts->push_back(std::make_pair(font_names[i].c_str(), font_flags[i]));
656 tlog(1, "OK font %s = %.4f%%, raw = %d = %.2f%%\n", font_names[i].c_str(),
657 100.0 * score / most_ok_chars, raw_score, 100.0 * raw_score / best_raw_score);
658 font_list += font_names[i];
659 font_list += "\n";
660 } else if (score >= least_good_enough || raw_score >= least_raw_enough) {
661 tlog(1, "Runner-up font %s = %.4f%%, raw = %d = %.2f%%\n", font_names[i].c_str(),
662 100.0 * score / most_ok_chars, raw_score, 100.0 * raw_score / best_raw_score);
663 }
664 }
665 return font_list;
666 }
667
668 /* static */
SelectFont(const char * utf8_word,const int utf8_len,std::string * font_name,std::vector<std::string> * graphemes)669 bool FontUtils::SelectFont(const char *utf8_word, const int utf8_len, std::string *font_name,
670 std::vector<std::string> *graphemes) {
671 return SelectFont(utf8_word, utf8_len, ListAvailableFonts(), font_name, graphemes);
672 }
673
674 /* static */
SelectFont(const char * utf8_word,const int utf8_len,const std::vector<std::string> & all_fonts,std::string * font_name,std::vector<std::string> * graphemes)675 bool FontUtils::SelectFont(const char *utf8_word, const int utf8_len,
676 const std::vector<std::string> &all_fonts, std::string *font_name,
677 std::vector<std::string> *graphemes) {
678 if (font_name) {
679 font_name->clear();
680 }
681 if (graphemes) {
682 graphemes->clear();
683 }
684 for (const auto &all_font : all_fonts) {
685 PangoFontInfo font;
686 std::vector<std::string> found_graphemes;
687 ASSERT_HOST_MSG(font.ParseFontDescriptionName(all_font), "Could not parse font desc name %s\n",
688 all_font.c_str());
689 if (font.CanRenderString(utf8_word, utf8_len, &found_graphemes)) {
690 if (graphemes) {
691 graphemes->swap(found_graphemes);
692 }
693 if (font_name) {
694 *font_name = all_font;
695 }
696 return true;
697 }
698 }
699 return false;
700 }
701
702 // PangoFontInfo is reinitialized, so clear the static list of fonts.
703 /* static */
ReInit()704 void FontUtils::ReInit() {
705 available_fonts_.clear();
706 }
707
708 // Print info about used font backend
709 /* static */
PangoFontTypeInfo()710 void FontUtils::PangoFontTypeInfo() {
711 PangoFontMap *font_map = pango_cairo_font_map_get_default();
712 if (pango_cairo_font_map_get_font_type(reinterpret_cast<PangoCairoFontMap *>(font_map)) ==
713 CAIRO_FONT_TYPE_TOY) {
714 printf("Using CAIRO_FONT_TYPE_TOY.\n");
715 } else if (pango_cairo_font_map_get_font_type(reinterpret_cast<PangoCairoFontMap *>(font_map)) ==
716 CAIRO_FONT_TYPE_FT) {
717 printf("Using CAIRO_FONT_TYPE_FT.\n");
718 } else if (pango_cairo_font_map_get_font_type(reinterpret_cast<PangoCairoFontMap *>(font_map)) ==
719 CAIRO_FONT_TYPE_WIN32) {
720 printf("Using CAIRO_FONT_TYPE_WIN32.\n");
721 } else if (pango_cairo_font_map_get_font_type(reinterpret_cast<PangoCairoFontMap *>(font_map)) ==
722 CAIRO_FONT_TYPE_QUARTZ) {
723 printf("Using CAIRO_FONT_TYPE_QUARTZ.\n");
724 } else if (pango_cairo_font_map_get_font_type(reinterpret_cast<PangoCairoFontMap *>(font_map)) ==
725 CAIRO_FONT_TYPE_USER) {
726 printf("Using CAIRO_FONT_TYPE_USER.\n");
727 } else if (!font_map) {
728 printf("Can not create pango cairo font map!\n");
729 }
730 }
731
732 } // namespace tesseract
733