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 <CharacterEncodingDetector.h>
22#include "CharacterEncodingDetectorTables.h"
23
24#include "utils/Vector.h"
25#include "StringArray.h"
26
27#include "unicode/ucnv.h"
28#include "unicode/ucsdet.h"
29#include "unicode/ustring.h"
30
31namespace android {
32
33CharacterEncodingDetector::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
43CharacterEncodingDetector::~CharacterEncodingDetector() {
44    ucnv_close(mUtf8Conv);
45}
46
47void CharacterEncodingDetector::addTag(const char *name, const char *value) {
48    mNames.push_back(name);
49    mValues.push_back(value);
50}
51
52size_t CharacterEncodingDetector::size() {
53    return mNames.size();
54}
55
56status_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
66static 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
75void 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
89        // try combined detection of artist/album/title etc.
90        char buf[1024];
91        buf[0] = 0;
92        bool allprintable = true;
93        for (int i = 0; i < size; i++) {
94            const char *name = mNames.getEntry(i);
95            const char *value = mValues.getEntry(i);
96            if (!isPrintableAscii(value, strlen(value)) && (
97                        !strcmp(name, "artist") ||
98                        !strcmp(name, "albumartist") ||
99                        !strcmp(name, "composer") ||
100                        !strcmp(name, "genre") ||
101                        !strcmp(name, "album") ||
102                        !strcmp(name, "title"))) {
103                strlcat(buf, value, sizeof(buf));
104                // separate tags by space so ICU's ngram detector can do its job
105                strlcat(buf, " ", sizeof(buf));
106                allprintable = false;
107            }
108        }
109
110        const char *combinedenc = "UTF-8";
111        if (allprintable) {
112            // since 'buf' is empty, ICU would return a UTF-8 matcher with low confidence, so
113            // no need to even call it
114            ALOGV("all tags are printable, assuming ascii (%zu)", strlen(buf));
115        } else {
116            ucsdet_setText(csd, buf, strlen(buf), &status);
117            int32_t matches;
118            const UCharsetMatch** ucma = ucsdet_detectAll(csd, &matches, &status);
119            bool goodmatch = true;
120            int highest = 0;
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                // use encoding determined from the combination of artist/album/title etc.
184                enc = combinedenc;
185            } else {
186                if (isPrintableAscii(s, inputLength)) {
187                    enc = "UTF-8";
188                    ALOGV("@@@@ %s is ascii", mNames.getEntry(i));
189                } else {
190                    ucsdet_setText(csd, s, inputLength, &status);
191                    ucm = ucsdet_detect(csd, &status);
192                    if (!ucm) {
193                        mValues.setEntry(i, "???");
194                        continue;
195                    }
196                    enc = ucsdet_getName(ucm, &status);
197                    ALOGV("@@@@ recognized charset: %s for %s confidence %d",
198                            enc, mNames.getEntry(i), ucsdet_getConfidence(ucm, &status));
199                }
200            }
201
202            if (strcmp(enc,"UTF-8") != 0) {
203                // only convert if the source encoding isn't already UTF-8
204                ALOGV("@@@ using converter %s for %s", enc, mNames.getEntry(i));
205                status = U_ZERO_ERROR;
206                UConverter *conv = ucnv_open(enc, &status);
207                if (U_FAILURE(status)) {
208                    ALOGW("could not create UConverter for %s (%d), falling back to ISO-8859-1",
209                            enc, status);
210                    status = U_ZERO_ERROR;
211                    conv = ucnv_open("ISO-8859-1", &status);
212                    if (U_FAILURE(status)) {
213                        ALOGW("could not create UConverter for ISO-8859-1 either");
214                        continue;
215                    }
216                }
217
218                // convert from native encoding to UTF-8
219                const char* source = mValues.getEntry(i);
220                int targetLength = len * 3 + 1;
221                char* buffer = new char[targetLength];
222                // don't normally check for NULL, but in this case targetLength may be large
223                if (!buffer)
224                    break;
225                char* target = buffer;
226
227                ucnv_convertEx(mUtf8Conv, conv, &target, target + targetLength,
228                        &source, source + strlen(source),
229                        NULL, NULL, NULL, NULL, TRUE, TRUE, &status);
230
231                if (U_FAILURE(status)) {
232                    ALOGE("ucnv_convertEx failed: %d", status);
233                    mValues.setEntry(i, "???");
234                } else {
235                    // zero terminate
236                    *target = 0;
237                    // strip trailing spaces
238                    while (--target > buffer && *target == ' ') {
239                        *target = 0;
240                    }
241                    // skip leading spaces
242                    char *start = buffer;
243                    while (*start == ' ') {
244                        start++;
245                    }
246                    mValues.setEntry(i, start);
247                }
248
249                delete[] buffer;
250
251                ucnv_close(conv);
252            }
253        }
254
255        for (int i = size - 1; i >= 0; --i) {
256            if (strlen(mValues.getEntry(i)) == 0) {
257                ALOGV("erasing %s because entry is empty", mNames.getEntry(i));
258                mNames.erase(i);
259                mValues.erase(i);
260            }
261        }
262
263        ucsdet_close(csd);
264    }
265}
266
267/*
268 * When ICU detects multiple encoding matches, apply additional heuristics to determine
269 * which one is the best match, since ICU can't always be trusted to make the right choice.
270 *
271 * What this method does is:
272 * - decode the input using each of the matches found
273 * - recalculate the starting confidence level for multibyte encodings using a different
274 *   algorithm and larger frequent character lists than ICU
275 * - devalue encoding where the conversion contains unlikely characters (symbols, reserved, etc)
276 * - pick the highest match
277 * - signal to the caller whether this match is considered good: confidence > 15, and confidence
278 *   delta with the next runner up > 15
279 */
280const UCharsetMatch *CharacterEncodingDetector::getPreferred(
281        const char *input, size_t len,
282        const UCharsetMatch** ucma, size_t nummatches,
283        bool *goodmatch, int *highestmatch) {
284
285    *goodmatch = false;
286    Vector<const UCharsetMatch*> matches;
287    UErrorCode status = U_ZERO_ERROR;
288
289    ALOGV("%zu matches", nummatches);
290    for (size_t i = 0; i < nummatches; i++) {
291        const char *encname = ucsdet_getName(ucma[i], &status);
292        int confidence = ucsdet_getConfidence(ucma[i], &status);
293        ALOGV("%zu: %s %d", i, encname, confidence);
294        matches.push_back(ucma[i]);
295    }
296
297    size_t num = matches.size();
298    if (num == 0) {
299        return NULL;
300    }
301    if (num == 1) {
302        int confidence = ucsdet_getConfidence(matches[0], &status);
303        if (confidence > 15) {
304            *goodmatch = true;
305        }
306        return matches[0];
307    }
308
309    ALOGV("considering %zu matches", num);
310
311    // keep track of how many "special" characters result when converting the input using each
312    // encoding
313    Vector<int> newconfidence;
314    for (size_t i = 0; i < num; i++) {
315        const uint16_t *freqdata = NULL;
316        float freqcoverage = 0;
317        status = U_ZERO_ERROR;
318        const char *encname = ucsdet_getName(matches[i], &status);
319        int confidence = ucsdet_getConfidence(matches[i], &status);
320        if (!strcmp("GB18030", encname)) {
321            freqdata = frequent_zhCN;
322            freqcoverage = frequent_zhCN_coverage;
323        } else if (!strcmp("Big5", encname)) {
324            freqdata = frequent_zhTW;
325            freqcoverage = frequent_zhTW_coverage;
326        } else if (!strcmp("EUC-KR", encname)) {
327            freqdata = frequent_ko;
328            freqcoverage = frequent_ko_coverage;
329        } else if (!strcmp("EUC-JP", encname)) {
330            freqdata = frequent_ja;
331            freqcoverage = frequent_ja_coverage;
332        } else if (!strcmp("Shift_JIS", encname)) {
333            freqdata = frequent_ja;
334            freqcoverage = frequent_ja_coverage;
335        }
336
337        ALOGV("%zu: %s %d", i, encname, confidence);
338        status = U_ZERO_ERROR;
339        UConverter *conv = ucnv_open(encname, &status);
340        int demerit = 0;
341        if (U_FAILURE(status)) {
342            ALOGV("failed to open %s: %d", encname, status);
343            confidence = 0;
344            demerit += 1000;
345        }
346        const char *source = input;
347        const char *sourceLimit = input + len;
348        status = U_ZERO_ERROR;
349        int frequentchars = 0;
350        int totalchars = 0;
351        while (true) {
352            // demerit the current encoding for each "special" character found after conversion.
353            // The amount of demerit is somewhat arbitrarily chosen.
354            int inchar;
355            if (source != sourceLimit) {
356                inchar = (source[0] << 8) + source[1];
357            }
358            UChar32 c = ucnv_getNextUChar(conv, &source, sourceLimit, &status);
359            if (!U_SUCCESS(status)) {
360                break;
361            }
362            if (c < 0x20 || (c >= 0x7f && c <= 0x009f)) {
363                ALOGV("control character %x", c);
364                demerit += 100;
365            } else if ((c == 0xa0)                      // no-break space
366                    || (c >= 0xa2 && c <= 0xbe)         // symbols, superscripts
367                    || (c == 0xd7) || (c == 0xf7)       // multiplication and division signs
368                    || (c >= 0x2000 && c <= 0x209f)) {  // punctuation, superscripts
369                ALOGV("unlikely character %x", c);
370                demerit += 10;
371            } else if (c >= 0xe000 && c <= 0xf8ff) {
372                ALOGV("private use character %x", c);
373                demerit += 30;
374            } else if (c >= 0x2190 && c <= 0x2bff) {
375                // this range comprises various symbol ranges that are unlikely to appear in
376                // music file metadata.
377                ALOGV("symbol %x", c);
378                demerit += 10;
379            } else if (c == 0xfffd) {
380                ALOGV("replacement character");
381                demerit += 50;
382            } else if (c >= 0xfff0 && c <= 0xfffc) {
383                ALOGV("unicode special %x", c);
384                demerit += 50;
385            } else if (freqdata != NULL) {
386                totalchars++;
387                if (isFrequent(freqdata, c)) {
388                    frequentchars++;
389                }
390            }
391        }
392        if (freqdata != NULL && totalchars != 0) {
393            int myconfidence = 10 + float((100 * frequentchars) / totalchars) / freqcoverage;
394            ALOGV("ICU confidence: %d, my confidence: %d (%d %d)", confidence, myconfidence,
395                    totalchars, frequentchars);
396            if (myconfidence > 100) myconfidence = 100;
397            if (myconfidence < 0) myconfidence = 0;
398            confidence = myconfidence;
399        }
400        ALOGV("%d-%d=%d", confidence, demerit, confidence - demerit);
401        newconfidence.push_back(confidence - demerit);
402        ucnv_close(conv);
403        if (i == 0 && (confidence - demerit) == 100) {
404            // no need to check any further, we'll end up using this match anyway
405            break;
406        }
407    }
408
409    // find match with highest confidence after adjusting for unlikely characters
410    int highest = newconfidence[0];
411    size_t highestidx = 0;
412    int runnerup = -10000;
413    int runnerupidx = -10000;
414    num = newconfidence.size();
415    for (size_t i = 1; i < num; i++) {
416        if (newconfidence[i] > highest) {
417            runnerup = highest;
418            runnerupidx = highestidx;
419            highest = newconfidence[i];
420            highestidx = i;
421        } else if (newconfidence[i] > runnerup){
422            runnerup = newconfidence[i];
423            runnerupidx = i;
424        }
425    }
426    status = U_ZERO_ERROR;
427    ALOGV("selecting: '%s' w/ %d confidence",
428            ucsdet_getName(matches[highestidx], &status), highest);
429    if (runnerupidx < 0) {
430        ALOGV("no runner up");
431        if (highest > 15) {
432            *goodmatch = true;
433        }
434    } else {
435        ALOGV("runner up: '%s' w/ %d confidence",
436                ucsdet_getName(matches[runnerupidx], &status), runnerup);
437        if (runnerup < 0) {
438            runnerup = 0;
439        }
440        if ((highest - runnerup) > 15) {
441            *goodmatch = true;
442        }
443    }
444    *highestmatch = highest;
445    return matches[highestidx];
446}
447
448
449bool CharacterEncodingDetector::isFrequent(const uint16_t *values, uint32_t c) {
450
451    int start = 0;
452    int end = 511; // All the tables have 512 entries
453    int mid = (start+end)/2;
454
455    while(start <= end) {
456        if(c == values[mid]) {
457            return true;
458        } else if (c > values[mid]) {
459            start = mid + 1;
460        } else {
461            end = mid - 1;
462        }
463
464        mid = (start + end) / 2;
465    }
466
467    return false;
468}
469
470
471}  // namespace android
472