1 /*
2 * Copyright (C) 2013 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 //#define LOG_NDEBUG 0
18 #define LOG_TAG "CharacterEncodingDector"
19 #include <utils/Log.h>
20
21 #include <media/CharacterEncodingDetector.h>
22 #include "CharacterEncodingDetectorTables.h"
23
24 #include <utils/Vector.h>
25 #include <media/StringArray.h>
26
27 #include <unicode/ucnv.h>
28 #include <unicode/ucsdet.h>
29 #include <unicode/ustring.h>
30
31 namespace android {
32
CharacterEncodingDetector()33 CharacterEncodingDetector::CharacterEncodingDetector() {
34
35 UErrorCode status = U_ZERO_ERROR;
36 mUtf8Conv = ucnv_open("UTF-8", &status);
37 if (U_FAILURE(status)) {
38 ALOGE("could not create UConverter for UTF-8");
39 mUtf8Conv = NULL;
40 }
41 }
42
~CharacterEncodingDetector()43 CharacterEncodingDetector::~CharacterEncodingDetector() {
44 ucnv_close(mUtf8Conv);
45 }
46
addTag(const char * name,const char * value)47 void CharacterEncodingDetector::addTag(const char *name, const char *value) {
48 mNames.push_back(name);
49 mValues.push_back(value);
50 }
51
size()52 size_t CharacterEncodingDetector::size() {
53 return mNames.size();
54 }
55
getTag(int index,const char ** name,const char ** value)56 status_t CharacterEncodingDetector::getTag(int index, const char **name, const char**value) {
57 if (index >= mNames.size()) {
58 return BAD_VALUE;
59 }
60
61 *name = mNames.getEntry(index);
62 *value = mValues.getEntry(index);
63 return OK;
64 }
65
isPrintableAscii(const char * value,size_t len)66 static bool isPrintableAscii(const char *value, size_t len) {
67 for (size_t i = 0; i < len; i++) {
68 if ((value[i] & 0x80) || value[i] < 0x20 || value[i] == 0x7f) {
69 return false;
70 }
71 }
72 return true;
73 }
74
detectAndConvert()75 void CharacterEncodingDetector::detectAndConvert() {
76
77 int size = mNames.size();
78 ALOGV("%d tags before conversion", size);
79 for (int i = 0; i < size; i++) {
80 ALOGV("%s: %s", mNames.getEntry(i), mValues.getEntry(i));
81 }
82
83 if (size && mUtf8Conv) {
84
85 UErrorCode status = U_ZERO_ERROR;
86 UCharsetDetector *csd = ucsdet_open(&status);
87 const UCharsetMatch *ucm;
88 bool goodmatch = true;
89 int highest = 0;
90
91 // try combined detection of artist/album/title etc.
92 char buf[1024];
93 buf[0] = 0;
94 bool allprintable = true;
95 for (int i = 0; i < size; i++) {
96 const char *name = mNames.getEntry(i);
97 const char *value = mValues.getEntry(i);
98 if (!isPrintableAscii(value, strlen(value)) && (
99 !strcmp(name, "artist") ||
100 !strcmp(name, "albumartist") ||
101 !strcmp(name, "composer") ||
102 !strcmp(name, "genre") ||
103 !strcmp(name, "album") ||
104 !strcmp(name, "title"))) {
105 strlcat(buf, value, sizeof(buf));
106 // separate tags by space so ICU's ngram detector can do its job
107 strlcat(buf, " ", sizeof(buf));
108 allprintable = false;
109 }
110 }
111
112 const char *combinedenc = "UTF-8";
113 if (allprintable) {
114 // since 'buf' is empty, ICU would return a UTF-8 matcher with low confidence, so
115 // no need to even call it
116 ALOGV("all tags are printable, assuming ascii (%zu)", strlen(buf));
117 } else {
118 ucsdet_setText(csd, buf, strlen(buf), &status);
119 int32_t matches;
120 const UCharsetMatch** ucma = ucsdet_detectAll(csd, &matches, &status);
121 const UCharsetMatch* bestCombinedMatch = getPreferred(buf, strlen(buf),
122 ucma, matches, &goodmatch, &highest);
123
124 ALOGV("goodmatch: %s, highest: %d", goodmatch ? "true" : "false", highest);
125 if (!goodmatch && (highest < 15 || strlen(buf) < 20)) {
126 ALOGV("not a good match, trying with more data");
127 // This string might be too short for ICU to do anything useful with.
128 // (real world example: "Björk" in ISO-8859-1 might be detected as GB18030, because
129 // the ISO detector reports a confidence of 0, while the GB18030 detector reports
130 // a confidence of 10 with no invalid characters)
131 // Append artist, album and title if they were previously omitted because they
132 // were printable ascii.
133 bool added = false;
134 for (int i = 0; i < size; i++) {
135 const char *name = mNames.getEntry(i);
136 const char *value = mValues.getEntry(i);
137 if (isPrintableAscii(value, strlen(value)) && (
138 !strcmp(name, "artist") ||
139 !strcmp(name, "album") ||
140 !strcmp(name, "title"))) {
141 strlcat(buf, value, sizeof(buf));
142 strlcat(buf, " ", sizeof(buf));
143 added = true;
144 }
145 }
146 if (added) {
147 ucsdet_setText(csd, buf, strlen(buf), &status);
148 ucma = ucsdet_detectAll(csd, &matches, &status);
149 bestCombinedMatch = getPreferred(buf, strlen(buf),
150 ucma, matches, &goodmatch, &highest);
151 if (!goodmatch && highest <= 15) {
152 ALOGV("still not a good match after adding printable tags");
153 bestCombinedMatch = NULL;
154 }
155 } else {
156 ALOGV("no printable tags to add");
157 }
158 }
159
160 if (bestCombinedMatch != NULL) {
161 combinedenc = ucsdet_getName(bestCombinedMatch, &status);
162 } else {
163 combinedenc = "ISO-8859-1";
164 }
165 }
166
167 for (int i = 0; i < size; i++) {
168 const char *name = mNames.getEntry(i);
169 uint8_t* src = (uint8_t *)mValues.getEntry(i);
170 int len = strlen((char *)src);
171
172 ALOGV("@@@ checking %s", name);
173 const char *s = mValues.getEntry(i);
174 int32_t inputLength = strlen(s);
175 const char *enc;
176
177 if (!allprintable && (!strcmp(name, "artist") ||
178 !strcmp(name, "albumartist") ||
179 !strcmp(name, "composer") ||
180 !strcmp(name, "genre") ||
181 !strcmp(name, "album") ||
182 !strcmp(name, "title"))) {
183 if (!goodmatch && highest < 0) {
184 // Give it one more chance if there is no good match.
185 ALOGV("Trying to detect %s separately", name);
186 int32_t matches;
187 bool goodmatchSingle = true;
188 int highestSingle = 0;
189 ucsdet_setText(csd, s, inputLength, &status);
190 const UCharsetMatch** ucma = ucsdet_detectAll(csd, &matches, &status);
191 const UCharsetMatch* bestSingleMatch = getPreferred(s, inputLength,
192 ucma, matches, &goodmatchSingle, &highestSingle);
193 if (goodmatchSingle || highestSingle > highest)
194 enc = ucsdet_getName(bestSingleMatch, &status);
195 else
196 enc = combinedenc;
197 } else {
198 // use encoding determined from the combination of artist/album/title etc.
199 enc = combinedenc;
200 }
201 } else {
202 if (isPrintableAscii(s, inputLength)) {
203 enc = "UTF-8";
204 ALOGV("@@@@ %s is ascii", mNames.getEntry(i));
205 } else {
206 ucsdet_setText(csd, s, inputLength, &status);
207 ucm = ucsdet_detect(csd, &status);
208 if (!ucm) {
209 mValues.setEntry(i, "???");
210 continue;
211 }
212 enc = ucsdet_getName(ucm, &status);
213 ALOGV("@@@@ recognized charset: %s for %s confidence %d",
214 enc, mNames.getEntry(i), ucsdet_getConfidence(ucm, &status));
215 }
216 }
217
218 if (strcmp(enc,"UTF-8") != 0) {
219 // only convert if the source encoding isn't already UTF-8
220 ALOGV("@@@ using converter %s for %s", enc, mNames.getEntry(i));
221 status = U_ZERO_ERROR;
222 UConverter *conv = ucnv_open(enc, &status);
223 if (U_FAILURE(status)) {
224 ALOGW("could not create UConverter for %s (%d), falling back to ISO-8859-1",
225 enc, status);
226 status = U_ZERO_ERROR;
227 conv = ucnv_open("ISO-8859-1", &status);
228 if (U_FAILURE(status)) {
229 ALOGW("could not create UConverter for ISO-8859-1 either");
230 continue;
231 }
232 }
233
234 // convert from native encoding to UTF-8
235 const char* source = mValues.getEntry(i);
236 int targetLength = len * 3 + 1;
237 char* buffer = new char[targetLength];
238 // don't normally check for NULL, but in this case targetLength may be large
239 if (!buffer)
240 break;
241 char* target = buffer;
242
243 ucnv_convertEx(mUtf8Conv, conv, &target, target + targetLength,
244 &source, source + strlen(source),
245 NULL, NULL, NULL, NULL, TRUE, TRUE, &status);
246
247 if (U_FAILURE(status)) {
248 ALOGE("ucnv_convertEx failed: %d", status);
249 mValues.setEntry(i, "???");
250 } else {
251 // zero terminate
252 *target = 0;
253 // strip trailing spaces
254 while (--target > buffer && *target == ' ') {
255 *target = 0;
256 }
257 // skip leading spaces
258 char *start = buffer;
259 while (*start == ' ') {
260 start++;
261 }
262 mValues.setEntry(i, start);
263 }
264
265 delete[] buffer;
266
267 ucnv_close(conv);
268 }
269 }
270
271 for (int i = size - 1; i >= 0; --i) {
272 if (strlen(mValues.getEntry(i)) == 0) {
273 ALOGV("erasing %s because entry is empty", mNames.getEntry(i));
274 mNames.erase(i);
275 mValues.erase(i);
276 }
277 }
278
279 ucsdet_close(csd);
280 }
281 }
282
283 /*
284 * When ICU detects multiple encoding matches, apply additional heuristics to determine
285 * which one is the best match, since ICU can't always be trusted to make the right choice.
286 *
287 * What this method does is:
288 * - decode the input using each of the matches found
289 * - recalculate the starting confidence level for multibyte encodings using a different
290 * algorithm and larger frequent character lists than ICU
291 * - devalue encoding where the conversion contains unlikely characters (symbols, reserved, etc)
292 * - pick the highest match
293 * - signal to the caller whether this match is considered good: confidence > 15, and confidence
294 * delta with the next runner up > 15
295 */
getPreferred(const char * input,size_t len,const UCharsetMatch ** ucma,size_t nummatches,bool * goodmatch,int * highestmatch)296 const UCharsetMatch *CharacterEncodingDetector::getPreferred(
297 const char *input, size_t len,
298 const UCharsetMatch** ucma, size_t nummatches,
299 bool *goodmatch, int *highestmatch) {
300
301 *goodmatch = false;
302 Vector<const UCharsetMatch*> matches;
303 UErrorCode status = U_ZERO_ERROR;
304
305 ALOGV("%zu matches", nummatches);
306 for (size_t i = 0; i < nummatches; i++) {
307 const char *encname = ucsdet_getName(ucma[i], &status);
308 int confidence = ucsdet_getConfidence(ucma[i], &status);
309 ALOGV("%zu: %s %d", i, encname, confidence);
310 matches.push_back(ucma[i]);
311 }
312
313 size_t num = matches.size();
314 if (num == 0) {
315 return NULL;
316 }
317 if (num == 1) {
318 int confidence = ucsdet_getConfidence(matches[0], &status);
319 if (confidence > 15) {
320 *goodmatch = true;
321 }
322 return matches[0];
323 }
324
325 ALOGV("considering %zu matches", num);
326
327 // keep track of how many "special" characters result when converting the input using each
328 // encoding
329 Vector<int> newconfidence;
330 for (size_t i = 0; i < num; i++) {
331 const uint16_t *freqdata = NULL;
332 float freqcoverage = 0;
333 status = U_ZERO_ERROR;
334 const char *encname = ucsdet_getName(matches[i], &status);
335 int confidence = ucsdet_getConfidence(matches[i], &status);
336 if (!strcmp("GB18030", encname)) {
337 freqdata = frequent_zhCN;
338 freqcoverage = frequent_zhCN_coverage;
339 } else if (!strcmp("Big5", encname)) {
340 freqdata = frequent_zhTW;
341 freqcoverage = frequent_zhTW_coverage;
342 } else if (!strcmp("EUC-KR", encname)) {
343 freqdata = frequent_ko;
344 freqcoverage = frequent_ko_coverage;
345 } else if (!strcmp("EUC-JP", encname)) {
346 freqdata = frequent_ja;
347 freqcoverage = frequent_ja_coverage;
348 } else if (!strcmp("Shift_JIS", encname)) {
349 freqdata = frequent_ja;
350 freqcoverage = frequent_ja_coverage;
351 }
352
353 ALOGV("%zu: %s %d", i, encname, confidence);
354 status = U_ZERO_ERROR;
355 UConverter *conv = ucnv_open(encname, &status);
356 int demerit = 0;
357 if (U_FAILURE(status)) {
358 ALOGV("failed to open %s: %d", encname, status);
359 confidence = 0;
360 demerit += 1000;
361 }
362 const char *source = input;
363 const char *sourceLimit = input + len;
364 status = U_ZERO_ERROR;
365 int frequentchars = 0;
366 int totalchars = 0;
367 while (true) {
368 // demerit the current encoding for each "special" character found after conversion.
369 // The amount of demerit is somewhat arbitrarily chosen.
370 int inchar;
371 if (source != sourceLimit) {
372 inchar = (source[0] << 8) + source[1];
373 }
374 UChar32 c = ucnv_getNextUChar(conv, &source, sourceLimit, &status);
375 if (!U_SUCCESS(status)) {
376 break;
377 }
378 if (c < 0x20 || (c >= 0x7f && c <= 0x009f)) {
379 ALOGV("control character %x", c);
380 demerit += 100;
381 } else if ((c == 0xa0) // no-break space
382 || (c >= 0xa2 && c <= 0xbe) // symbols, superscripts
383 || (c == 0xd7) || (c == 0xf7) // multiplication and division signs
384 || (c >= 0x2000 && c <= 0x209f)) { // punctuation, superscripts
385 ALOGV("unlikely character %x", c);
386 demerit += 10;
387 } else if (c >= 0xe000 && c <= 0xf8ff) {
388 ALOGV("private use character %x", c);
389 demerit += 30;
390 } else if (c >= 0x2190 && c <= 0x2bff) {
391 // this range comprises various symbol ranges that are unlikely to appear in
392 // music file metadata.
393 ALOGV("symbol %x", c);
394 demerit += 10;
395 } else if (c == 0xfffd) {
396 ALOGV("replacement character");
397 demerit += 50;
398 } else if (c >= 0xfff0 && c <= 0xfffc) {
399 ALOGV("unicode special %x", c);
400 demerit += 50;
401 } else if (freqdata != NULL) {
402 totalchars++;
403 if (isFrequent(freqdata, c)) {
404 frequentchars++;
405 }
406 }
407 }
408 if (freqdata != NULL && totalchars != 0) {
409 int myconfidence = 10 + float((100 * frequentchars) / totalchars) / freqcoverage;
410 ALOGV("ICU confidence: %d, my confidence: %d (%d %d)", confidence, myconfidence,
411 totalchars, frequentchars);
412 if (myconfidence > 100) myconfidence = 100;
413 if (myconfidence < 0) myconfidence = 0;
414 confidence = myconfidence;
415 }
416 ALOGV("%d-%d=%d", confidence, demerit, confidence - demerit);
417 newconfidence.push_back(confidence - demerit);
418 ucnv_close(conv);
419 if (i == 0 && (confidence - demerit) == 100) {
420 // no need to check any further, we'll end up using this match anyway
421 break;
422 }
423 }
424
425 // find match with highest confidence after adjusting for unlikely characters
426 int highest = newconfidence[0];
427 size_t highestidx = 0;
428 int runnerup = -10000;
429 int runnerupidx = -10000;
430 num = newconfidence.size();
431 for (size_t i = 1; i < num; i++) {
432 if (newconfidence[i] > highest) {
433 runnerup = highest;
434 runnerupidx = highestidx;
435 highest = newconfidence[i];
436 highestidx = i;
437 } else if (newconfidence[i] > runnerup){
438 runnerup = newconfidence[i];
439 runnerupidx = i;
440 }
441 }
442 status = U_ZERO_ERROR;
443 ALOGV("selecting: '%s' w/ %d confidence",
444 ucsdet_getName(matches[highestidx], &status), highest);
445 if (runnerupidx < 0) {
446 ALOGV("no runner up");
447 if (highest > 15) {
448 *goodmatch = true;
449 }
450 } else {
451 ALOGV("runner up: '%s' w/ %d confidence",
452 ucsdet_getName(matches[runnerupidx], &status), runnerup);
453 if (runnerup < 0) {
454 runnerup = 0;
455 }
456 if ((highest - runnerup) > 15) {
457 *goodmatch = true;
458 }
459 }
460 *highestmatch = highest;
461 return matches[highestidx];
462 }
463
464
isFrequent(const uint16_t * values,uint32_t c)465 bool CharacterEncodingDetector::isFrequent(const uint16_t *values, uint32_t c) {
466
467 int start = 0;
468 int end = 511; // All the tables have 512 entries
469 int mid = (start+end)/2;
470
471 while(start <= end) {
472 if(c == values[mid]) {
473 return true;
474 } else if (c > values[mid]) {
475 start = mid + 1;
476 } else {
477 end = mid - 1;
478 }
479
480 mid = (start + end) / 2;
481 }
482
483 return false;
484 }
485
486
487 } // namespace android
488