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 #include <cutils/properties.h>
32 
33 namespace android {
34 
CharacterEncodingDetector()35 CharacterEncodingDetector::CharacterEncodingDetector() {
36 
37     UErrorCode status = U_ZERO_ERROR;
38     mUtf8Conv = ucnv_open("UTF-8", &status);
39     if (U_FAILURE(status)) {
40         ALOGE("could not create UConverter for UTF-8");
41         mUtf8Conv = NULL;
42     }
43 
44     // Read system locale setting from system property and map to ICU encoding names.
45     mLocaleEnc = NULL;
46     char locale_value[PROPERTY_VALUE_MAX] = "";
47     if (property_get("persist.sys.locale", locale_value, NULL) > 0) {
48         const size_t len = strnlen(locale_value, sizeof(locale_value));
49 
50         if (len == 3 && !strncmp(locale_value, "und", 3)) {
51             // Undetermined
52         } else if (!strncmp(locale_value, "th", 2)) { // Thai
53             mLocaleEnc = "windows-874-2000";
54         }
55         if (mLocaleEnc != NULL) {
56             ALOGV("System locale encoding = %s", mLocaleEnc);
57         } else {
58             ALOGV("Didn't recognize system locale setting, defaulting to en_US");
59         }
60     } else {
61         ALOGV("Couldn't read system locale setting, assuming en_US");
62     }
63 }
64 
~CharacterEncodingDetector()65 CharacterEncodingDetector::~CharacterEncodingDetector() {
66     ucnv_close(mUtf8Conv);
67 }
68 
addTag(const char * name,const char * value)69 void CharacterEncodingDetector::addTag(const char *name, const char *value) {
70     mNames.push_back(name);
71     mValues.push_back(value);
72 }
73 
size()74 size_t CharacterEncodingDetector::size() {
75     return mNames.size();
76 }
77 
getTag(int index,const char ** name,const char ** value)78 status_t CharacterEncodingDetector::getTag(int index, const char **name, const char**value) {
79     if (index >= mNames.size()) {
80         return BAD_VALUE;
81     }
82 
83     *name = mNames.getEntry(index);
84     *value = mValues.getEntry(index);
85     return OK;
86 }
87 
isPrintableAscii(const char * value,size_t len)88 static bool isPrintableAscii(const char *value, size_t len) {
89     for (size_t i = 0; i < len; i++) {
90         if ((value[i] & 0x80) || value[i] < 0x20 || value[i] == 0x7f) {
91             return false;
92         }
93     }
94     return true;
95 }
96 
detectAndConvert()97 void CharacterEncodingDetector::detectAndConvert() {
98 
99     int size = mNames.size();
100     ALOGV("%d tags before conversion", size);
101     for (int i = 0; i < size; i++) {
102         ALOGV("%s: %s", mNames.getEntry(i), mValues.getEntry(i));
103     }
104 
105     if (size && mUtf8Conv) {
106 
107         UErrorCode status = U_ZERO_ERROR;
108         UCharsetDetector *csd = ucsdet_open(&status);
109         const UCharsetMatch *ucm;
110         bool goodmatch = true;
111         int highest = 0;
112 
113         // try combined detection of artist/album/title etc.
114         char buf[1024];
115         buf[0] = 0;
116         bool allprintable = true;
117         for (int i = 0; i < size; i++) {
118             const char *name = mNames.getEntry(i);
119             const char *value = mValues.getEntry(i);
120             if (!isPrintableAscii(value, strlen(value)) && (
121                         !strcmp(name, "artist") ||
122                         !strcmp(name, "albumartist") ||
123                         !strcmp(name, "composer") ||
124                         !strcmp(name, "genre") ||
125                         !strcmp(name, "album") ||
126                         !strcmp(name, "title"))) {
127                 strlcat(buf, value, sizeof(buf));
128                 // separate tags by space so ICU's ngram detector can do its job
129                 strlcat(buf, " ", sizeof(buf));
130                 allprintable = false;
131             }
132         }
133 
134         const char *combinedenc = "UTF-8";
135         if (allprintable) {
136             // since 'buf' is empty, ICU would return a UTF-8 matcher with low confidence, so
137             // no need to even call it
138             ALOGV("all tags are printable, assuming ascii (%zu)", strlen(buf));
139         } else {
140             ucsdet_setText(csd, buf, strlen(buf), &status);
141             int32_t matches;
142             const UCharsetMatch** ucma = ucsdet_detectAll(csd, &matches, &status);
143             const UCharsetMatch* bestCombinedMatch = getPreferred(buf, strlen(buf),
144                     ucma, matches, &goodmatch, &highest);
145 
146             ALOGV("goodmatch: %s, highest: %d", goodmatch ? "true" : "false", highest);
147             if (!goodmatch && (highest < 15 || strlen(buf) < 20)) {
148                 ALOGV("not a good match, trying with more data");
149                 // This string might be too short for ICU to do anything useful with.
150                 // (real world example: "Björk" in ISO-8859-1 might be detected as GB18030, because
151                 //  the ISO detector reports a confidence of 0, while the GB18030 detector reports
152                 //  a confidence of 10 with no invalid characters)
153                 // Append artist, album and title if they were previously omitted because they
154                 // were printable ascii.
155                 bool added = false;
156                 for (int i = 0; i < size; i++) {
157                     const char *name = mNames.getEntry(i);
158                     const char *value = mValues.getEntry(i);
159                     if (isPrintableAscii(value, strlen(value)) && (
160                                 !strcmp(name, "artist") ||
161                                 !strcmp(name, "album") ||
162                                 !strcmp(name, "title"))) {
163                         strlcat(buf, value, sizeof(buf));
164                         strlcat(buf, " ", sizeof(buf));
165                         added = true;
166                     }
167                 }
168                 if (added) {
169                     ucsdet_setText(csd, buf, strlen(buf), &status);
170                     ucma = ucsdet_detectAll(csd, &matches, &status);
171                     bestCombinedMatch = getPreferred(buf, strlen(buf),
172                             ucma, matches, &goodmatch, &highest);
173                     if (!goodmatch && highest <= 15) {
174                         ALOGV("still not a good match after adding printable tags");
175                         bestCombinedMatch = NULL;
176                     }
177                 } else {
178                     ALOGV("no printable tags to add");
179                 }
180             }
181 
182             if (mLocaleEnc != NULL && !goodmatch && highest < 50) {
183                 combinedenc = mLocaleEnc;
184                 ALOGV("confidence is low but we have recognized predefined encoding, "
185                         "so try this (%s) instead", mLocaleEnc);
186             } else if (bestCombinedMatch != NULL) {
187                 combinedenc = ucsdet_getName(bestCombinedMatch, &status);
188             } else {
189                 combinedenc = "ISO-8859-1";
190             }
191         }
192 
193         for (int i = 0; i < size; i++) {
194             const char *name = mNames.getEntry(i);
195             uint8_t* src = (uint8_t *)mValues.getEntry(i);
196             int len = strlen((char *)src);
197 
198             ALOGV("@@@ checking %s", name);
199             const char *s = mValues.getEntry(i);
200             int32_t inputLength = strlen(s);
201             const char *enc;
202 
203             if (!allprintable && (!strcmp(name, "artist") ||
204                     !strcmp(name, "albumartist") ||
205                     !strcmp(name, "composer") ||
206                     !strcmp(name, "genre") ||
207                     !strcmp(name, "album") ||
208                     !strcmp(name, "title"))) {
209                 if (!goodmatch && highest < 0) {
210                     // Give it one more chance if there is no good match.
211                     ALOGV("Trying to detect %s separately", name);
212                     int32_t matches;
213                     bool goodmatchSingle = true;
214                     int highestSingle = 0;
215                     ucsdet_setText(csd, s, inputLength, &status);
216                     const UCharsetMatch** ucma = ucsdet_detectAll(csd, &matches, &status);
217                     const UCharsetMatch* bestSingleMatch = getPreferred(s, inputLength,
218                             ucma, matches, &goodmatchSingle, &highestSingle);
219                     if (goodmatchSingle || highestSingle > highest)
220                         enc = ucsdet_getName(bestSingleMatch, &status);
221                     else
222                         enc = combinedenc;
223                 } else {
224                     // use encoding determined from the combination of artist/album/title etc.
225                     enc = combinedenc;
226                 }
227             } else {
228                 if (isPrintableAscii(s, inputLength)) {
229                     enc = "UTF-8";
230                     ALOGV("@@@@ %s is ascii", mNames.getEntry(i));
231                 } else {
232                     ucsdet_setText(csd, s, inputLength, &status);
233                     ucm = ucsdet_detect(csd, &status);
234                     if (!ucm) {
235                         mValues.setEntry(i, "???");
236                         continue;
237                     }
238                     enc = ucsdet_getName(ucm, &status);
239                     ALOGV("@@@@ recognized charset: %s for %s confidence %d",
240                             enc, mNames.getEntry(i), ucsdet_getConfidence(ucm, &status));
241                 }
242             }
243 
244             if (strcmp(enc,"UTF-8") != 0) {
245                 // only convert if the source encoding isn't already UTF-8
246                 ALOGV("@@@ using converter %s for %s", enc, mNames.getEntry(i));
247                 status = U_ZERO_ERROR;
248                 UConverter *conv = ucnv_open(enc, &status);
249                 if (U_FAILURE(status)) {
250                     ALOGW("could not create UConverter for %s (%d), falling back to ISO-8859-1",
251                             enc, status);
252                     status = U_ZERO_ERROR;
253                     conv = ucnv_open("ISO-8859-1", &status);
254                     if (U_FAILURE(status)) {
255                         ALOGW("could not create UConverter for ISO-8859-1 either");
256                         continue;
257                     }
258                 }
259 
260                 // convert from native encoding to UTF-8
261                 const char* source = mValues.getEntry(i);
262                 int targetLength = len * 3 + 1;
263                 char* buffer = new char[targetLength];
264                 // don't normally check for NULL, but in this case targetLength may be large
265                 if (!buffer)
266                     break;
267                 char* target = buffer;
268 
269                 ucnv_convertEx(mUtf8Conv, conv, &target, target + targetLength,
270                         &source, source + strlen(source),
271                         NULL, NULL, NULL, NULL, TRUE, TRUE, &status);
272 
273                 if (U_FAILURE(status)) {
274                     ALOGE("ucnv_convertEx failed: %d", status);
275                     mValues.setEntry(i, "???");
276                 } else {
277                     // zero terminate
278                     *target = 0;
279                     // strip trailing spaces
280                     while (--target > buffer && *target == ' ') {
281                         *target = 0;
282                     }
283                     // skip leading spaces
284                     char *start = buffer;
285                     while (*start == ' ') {
286                         start++;
287                     }
288                     mValues.setEntry(i, start);
289                 }
290 
291                 delete[] buffer;
292 
293                 ucnv_close(conv);
294             }
295         }
296 
297         for (int i = size - 1; i >= 0; --i) {
298             if (strlen(mValues.getEntry(i)) == 0) {
299                 ALOGV("erasing %s because entry is empty", mNames.getEntry(i));
300                 mNames.erase(i);
301                 mValues.erase(i);
302             }
303         }
304 
305         ucsdet_close(csd);
306     }
307 }
308 
309 /*
310  * When ICU detects multiple encoding matches, apply additional heuristics to determine
311  * which one is the best match, since ICU can't always be trusted to make the right choice.
312  *
313  * What this method does is:
314  * - decode the input using each of the matches found
315  * - recalculate the starting confidence level for multibyte encodings using a different
316  *   algorithm and larger frequent character lists than ICU
317  * - devalue encoding where the conversion contains unlikely characters (symbols, reserved, etc)
318  * - pick the highest match
319  * - signal to the caller whether this match is considered good: confidence > 15, and confidence
320  *   delta with the next runner up > 15
321  */
getPreferred(const char * input,size_t len,const UCharsetMatch ** ucma,size_t nummatches,bool * goodmatch,int * highestmatch)322 const UCharsetMatch *CharacterEncodingDetector::getPreferred(
323         const char *input, size_t len,
324         const UCharsetMatch** ucma, size_t nummatches,
325         bool *goodmatch, int *highestmatch) {
326 
327     *goodmatch = false;
328     Vector<const UCharsetMatch*> matches;
329     UErrorCode status = U_ZERO_ERROR;
330 
331     ALOGV("%zu matches", nummatches);
332     for (size_t i = 0; i < nummatches; i++) {
333         const char *encname = ucsdet_getName(ucma[i], &status);
334         int confidence = ucsdet_getConfidence(ucma[i], &status);
335         ALOGV("%zu: %s %d", i, encname, confidence);
336         matches.push_back(ucma[i]);
337     }
338 
339     size_t num = matches.size();
340     if (num == 0) {
341         return NULL;
342     }
343     if (num == 1) {
344         int confidence = ucsdet_getConfidence(matches[0], &status);
345         if (confidence > 15) {
346             *goodmatch = true;
347         }
348         return matches[0];
349     }
350 
351     ALOGV("considering %zu matches", num);
352 
353     // keep track of how many "special" characters result when converting the input using each
354     // encoding
355     Vector<int> newconfidence;
356     for (size_t i = 0; i < num; i++) {
357         const uint16_t *freqdata = NULL;
358         float freqcoverage = 0;
359         status = U_ZERO_ERROR;
360         const char *encname = ucsdet_getName(matches[i], &status);
361         int confidence = ucsdet_getConfidence(matches[i], &status);
362         if (!strcmp("GB18030", encname)) {
363             freqdata = frequent_zhCN;
364             freqcoverage = frequent_zhCN_coverage;
365         } else if (!strcmp("Big5", encname)) {
366             freqdata = frequent_zhTW;
367             freqcoverage = frequent_zhTW_coverage;
368         } else if (!strcmp("EUC-KR", encname)) {
369             freqdata = frequent_ko;
370             freqcoverage = frequent_ko_coverage;
371         } else if (!strcmp("EUC-JP", encname)) {
372             freqdata = frequent_ja;
373             freqcoverage = frequent_ja_coverage;
374         } else if (!strcmp("Shift_JIS", encname)) {
375             freqdata = frequent_ja;
376             freqcoverage = frequent_ja_coverage;
377         }
378 
379         ALOGV("%zu: %s %d", i, encname, confidence);
380         status = U_ZERO_ERROR;
381         UConverter *conv = ucnv_open(encname, &status);
382         int demerit = 0;
383         if (U_FAILURE(status)) {
384             ALOGV("failed to open %s: %d", encname, status);
385             confidence = 0;
386             demerit += 1000;
387         }
388         const char *source = input;
389         const char *sourceLimit = input + len;
390         status = U_ZERO_ERROR;
391         int frequentchars = 0;
392         int totalchars = 0;
393         while (true) {
394             // demerit the current encoding for each "special" character found after conversion.
395             // The amount of demerit is somewhat arbitrarily chosen.
396             int inchar;
397             if (source != sourceLimit) {
398                 inchar = (source[0] << 8) + source[1];
399             }
400             UChar32 c = ucnv_getNextUChar(conv, &source, sourceLimit, &status);
401             if (!U_SUCCESS(status)) {
402                 break;
403             }
404             if (c < 0x20 || (c >= 0x7f && c <= 0x009f)) {
405                 ALOGV("control character %x", c);
406                 demerit += 100;
407             } else if ((c == 0xa0)                      // no-break space
408                     || (c >= 0xa2 && c <= 0xbe)         // symbols, superscripts
409                     || (c == 0xd7) || (c == 0xf7)       // multiplication and division signs
410                     || (c >= 0x2000 && c <= 0x209f)) {  // punctuation, superscripts
411                 ALOGV("unlikely character %x", c);
412                 demerit += 10;
413             } else if (c >= 0xe000 && c <= 0xf8ff) {
414                 ALOGV("private use character %x", c);
415                 demerit += 30;
416             } else if (c >= 0x2190 && c <= 0x2bff) {
417                 // this range comprises various symbol ranges that are unlikely to appear in
418                 // music file metadata.
419                 ALOGV("symbol %x", c);
420                 demerit += 10;
421             } else if (c == 0xfffd) {
422                 ALOGV("replacement character");
423                 demerit += 50;
424             } else if (c >= 0xfff0 && c <= 0xfffc) {
425                 ALOGV("unicode special %x", c);
426                 demerit += 50;
427             } else if (freqdata != NULL) {
428                 totalchars++;
429                 if (isFrequent(freqdata, c)) {
430                     frequentchars++;
431                 }
432             }
433         }
434         if (freqdata != NULL && totalchars != 0) {
435             int myconfidence = 10 + float((100 * frequentchars) / totalchars) / freqcoverage;
436             ALOGV("ICU confidence: %d, my confidence: %d (%d %d)", confidence, myconfidence,
437                     totalchars, frequentchars);
438             if (myconfidence > 100) myconfidence = 100;
439             if (myconfidence < 0) myconfidence = 0;
440             confidence = myconfidence;
441         }
442         ALOGV("%d-%d=%d", confidence, demerit, confidence - demerit);
443         newconfidence.push_back(confidence - demerit);
444         ucnv_close(conv);
445         if (i == 0 && (confidence - demerit) == 100) {
446             // no need to check any further, we'll end up using this match anyway
447             break;
448         }
449     }
450 
451     // find match with highest confidence after adjusting for unlikely characters
452     int highest = newconfidence[0];
453     size_t highestidx = 0;
454     int runnerup = -10000;
455     int runnerupidx = -10000;
456     num = newconfidence.size();
457     for (size_t i = 1; i < num; i++) {
458         if (newconfidence[i] > highest) {
459             runnerup = highest;
460             runnerupidx = highestidx;
461             highest = newconfidence[i];
462             highestidx = i;
463         } else if (newconfidence[i] > runnerup){
464             runnerup = newconfidence[i];
465             runnerupidx = i;
466         }
467     }
468     status = U_ZERO_ERROR;
469     ALOGV("selecting: '%s' w/ %d confidence",
470             ucsdet_getName(matches[highestidx], &status), highest);
471     if (runnerupidx < 0) {
472         ALOGV("no runner up");
473         if (highest > 15) {
474             *goodmatch = true;
475         }
476     } else {
477         ALOGV("runner up: '%s' w/ %d confidence",
478                 ucsdet_getName(matches[runnerupidx], &status), runnerup);
479         if (runnerup < 0) {
480             runnerup = 0;
481         }
482         if ((highest - runnerup) > 15) {
483             *goodmatch = true;
484         }
485     }
486     *highestmatch = highest;
487     return matches[highestidx];
488 }
489 
490 
isFrequent(const uint16_t * values,uint32_t c)491 bool CharacterEncodingDetector::isFrequent(const uint16_t *values, uint32_t c) {
492 
493     int start = 0;
494     int end = 511; // All the tables have 512 entries
495     int mid = (start+end)/2;
496 
497     while(start <= end) {
498         if(c == values[mid]) {
499             return true;
500         } else if (c > values[mid]) {
501             start = mid + 1;
502         } else {
503             end = mid - 1;
504         }
505 
506         mid = (start + end) / 2;
507     }
508 
509     return false;
510 }
511 
512 
513 }  // namespace android
514