• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2015 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 #include <vector>
18 #include <memory>
19 #include <algorithm>
20 #include <string>
21 #include <unicode/uchar.h>
22 #include <unicode/uscript.h>
23 
24 // HACK: for reading pattern file
25 #include <fcntl.h>
26 
27 #define LOG_TAG "Minikin"
28 #include "utils/Log.h"
29 
30 #include "minikin/Hyphenator.h"
31 
32 using std::vector;
33 
34 namespace minikin {
35 
36 static const uint16_t CHAR_HYPHEN_MINUS = 0x002D;
37 static const uint16_t CHAR_SOFT_HYPHEN = 0x00AD;
38 static const uint16_t CHAR_MIDDLE_DOT = 0x00B7;
39 static const uint16_t CHAR_HYPHEN = 0x2010;
40 
41 // The following are structs that correspond to tables inside the hyb file format
42 
43 struct AlphabetTable0 {
44     uint32_t version;
45     uint32_t min_codepoint;
46     uint32_t max_codepoint;
47     uint8_t data[1];  // actually flexible array, size is known at runtime
48 };
49 
50 struct AlphabetTable1 {
51     uint32_t version;
52     uint32_t n_entries;
53     uint32_t data[1]; // actually flexible array, size is known at runtime
54 
codepointminikin::AlphabetTable155     static uint32_t codepoint(uint32_t entry) { return entry >> 11; }
valueminikin::AlphabetTable156     static uint32_t value(uint32_t entry) { return entry & 0x7ff; }
57 };
58 
59 struct Trie {
60     uint32_t version;
61     uint32_t char_mask;
62     uint32_t link_shift;
63     uint32_t link_mask;
64     uint32_t pattern_shift;
65     uint32_t n_entries;
66     uint32_t data[1];  // actually flexible array, size is known at runtime
67 };
68 
69 struct Pattern {
70     uint32_t version;
71     uint32_t n_entries;
72     uint32_t pattern_offset;
73     uint32_t pattern_size;
74     uint32_t data[1];  // actually flexible array, size is known at runtime
75 
76     // accessors
lenminikin::Pattern77     static uint32_t len(uint32_t entry) { return entry >> 26; }
shiftminikin::Pattern78     static uint32_t shift(uint32_t entry) { return (entry >> 20) & 0x3f; }
bufminikin::Pattern79     const uint8_t* buf(uint32_t entry) const {
80         return reinterpret_cast<const uint8_t*>(this) + pattern_offset + (entry & 0xfffff);
81     }
82 };
83 
84 struct Header {
85     uint32_t magic;
86     uint32_t version;
87     uint32_t alphabet_offset;
88     uint32_t trie_offset;
89     uint32_t pattern_offset;
90     uint32_t file_size;
91 
92     // accessors
bytesminikin::Header93     const uint8_t* bytes() const { return reinterpret_cast<const uint8_t*>(this); }
alphabetVersionminikin::Header94     uint32_t alphabetVersion() const {
95         return *reinterpret_cast<const uint32_t*>(bytes() + alphabet_offset);
96     }
alphabetTable0minikin::Header97     const AlphabetTable0* alphabetTable0() const {
98         return reinterpret_cast<const AlphabetTable0*>(bytes() + alphabet_offset);
99     }
alphabetTable1minikin::Header100     const AlphabetTable1* alphabetTable1() const {
101         return reinterpret_cast<const AlphabetTable1*>(bytes() + alphabet_offset);
102     }
trieTableminikin::Header103     const Trie* trieTable() const {
104         return reinterpret_cast<const Trie*>(bytes() + trie_offset);
105     }
patternTableminikin::Header106     const Pattern* patternTable() const {
107         return reinterpret_cast<const Pattern*>(bytes() + pattern_offset);
108     }
109 };
110 
loadBinary(const uint8_t * patternData,size_t minPrefix,size_t minSuffix)111 Hyphenator* Hyphenator::loadBinary(const uint8_t* patternData, size_t minPrefix, size_t minSuffix) {
112     Hyphenator* result = new Hyphenator;
113     result->patternData = patternData;
114     result->minPrefix = minPrefix;
115     result->minSuffix = minSuffix;
116     return result;
117 }
118 
hyphenate(vector<HyphenationType> * result,const uint16_t * word,size_t len,const icu::Locale & locale)119 void Hyphenator::hyphenate(vector<HyphenationType>* result, const uint16_t* word, size_t len,
120         const icu::Locale& locale) {
121     result->clear();
122     result->resize(len);
123     const size_t paddedLen = len + 2;  // start and stop code each count for 1
124     if (patternData != nullptr &&
125             len >= minPrefix + minSuffix && paddedLen <= MAX_HYPHENATED_SIZE) {
126         uint16_t alpha_codes[MAX_HYPHENATED_SIZE];
127         const HyphenationType hyphenValue = alphabetLookup(alpha_codes, word, len);
128         if (hyphenValue != HyphenationType::DONT_BREAK) {
129             hyphenateFromCodes(result->data(), alpha_codes, paddedLen, hyphenValue);
130             return;
131         }
132         // TODO: try NFC normalization
133         // TODO: handle non-BMP Unicode (requires remapping of offsets)
134     }
135     // Note that we will always get here if the word contains a hyphen or a soft hyphen, because the
136     // alphabet is not expected to contain a hyphen or a soft hyphen character, so alphabetLookup
137     // would return DONT_BREAK.
138     hyphenateWithNoPatterns(result->data(), word, len, locale);
139 }
140 
141 // This function determines whether a character is like U+2010 HYPHEN in
142 // line breaking and usage: a character immediately after which line breaks
143 // are allowed, but words containing it should not be automatically
144 // hyphenated using patterns. This is a curated set, created by manually
145 // inspecting all the characters that have the Unicode line breaking
146 // property of BA or HY and seeing which ones are hyphens.
isLineBreakingHyphen(uint32_t c)147 bool Hyphenator::isLineBreakingHyphen(uint32_t c) {
148     return (c == 0x002D || // HYPHEN-MINUS
149             c == 0x058A || // ARMENIAN HYPHEN
150             c == 0x05BE || // HEBREW PUNCTUATION MAQAF
151             c == 0x1400 || // CANADIAN SYLLABICS HYPHEN
152             c == 0x2010 || // HYPHEN
153             c == 0x2013 || // EN DASH
154             c == 0x2027 || // HYPHENATION POINT
155             c == 0x2E17 || // DOUBLE OBLIQUE HYPHEN
156             c == 0x2E40);  // DOUBLE HYPHEN
157 }
158 
159 const static uint32_t HYPHEN_STR[] = {0x2010, 0};
160 const static uint32_t ARMENIAN_HYPHEN_STR[] = {0x058A, 0};
161 const static uint32_t MAQAF_STR[] = {0x05BE, 0};
162 const static uint32_t UCAS_HYPHEN_STR[] = {0x1400, 0};
163 const static uint32_t ZWJ_STR[] = {0x200D, 0};
164 const static uint32_t ZWJ_AND_HYPHEN_STR[] = {0x200D, 0x2010, 0};
165 
getHyphenString(uint32_t hyph)166 const uint32_t* HyphenEdit::getHyphenString(uint32_t hyph) {
167     switch (hyph) {
168         case INSERT_HYPHEN_AT_END:
169         case REPLACE_WITH_HYPHEN_AT_END:
170         case INSERT_HYPHEN_AT_START:
171             return HYPHEN_STR;
172         case INSERT_ARMENIAN_HYPHEN_AT_END:
173             return ARMENIAN_HYPHEN_STR;
174         case INSERT_MAQAF_AT_END:
175             return MAQAF_STR;
176         case INSERT_UCAS_HYPHEN_AT_END:
177             return UCAS_HYPHEN_STR;
178         case INSERT_ZWJ_AND_HYPHEN_AT_END:
179             return ZWJ_AND_HYPHEN_STR;
180         case INSERT_ZWJ_AT_START:
181             return ZWJ_STR;
182         default:
183             return nullptr;
184     }
185 }
186 
editForThisLine(HyphenationType type)187 uint32_t HyphenEdit::editForThisLine(HyphenationType type) {
188     switch (type) {
189         case HyphenationType::DONT_BREAK:
190             return NO_EDIT;
191         case HyphenationType::BREAK_AND_INSERT_HYPHEN:
192             return INSERT_HYPHEN_AT_END;
193         case HyphenationType::BREAK_AND_INSERT_ARMENIAN_HYPHEN:
194             return INSERT_ARMENIAN_HYPHEN_AT_END;
195         case HyphenationType::BREAK_AND_INSERT_MAQAF:
196             return INSERT_MAQAF_AT_END;
197         case HyphenationType::BREAK_AND_INSERT_UCAS_HYPHEN:
198             return INSERT_UCAS_HYPHEN_AT_END;
199         case HyphenationType::BREAK_AND_REPLACE_WITH_HYPHEN:
200             return REPLACE_WITH_HYPHEN_AT_END;
201         case HyphenationType::BREAK_AND_INSERT_HYPHEN_AND_ZWJ:
202             return INSERT_ZWJ_AND_HYPHEN_AT_END;
203         default:
204             return BREAK_AT_END;
205     }
206 }
207 
editForNextLine(HyphenationType type)208 uint32_t HyphenEdit::editForNextLine(HyphenationType type) {
209     switch (type) {
210         case HyphenationType::DONT_BREAK:
211             return NO_EDIT;
212         case HyphenationType::BREAK_AND_INSERT_HYPHEN_AT_NEXT_LINE:
213             return INSERT_HYPHEN_AT_START;
214         case HyphenationType::BREAK_AND_INSERT_HYPHEN_AND_ZWJ:
215             return INSERT_ZWJ_AT_START;
216         default:
217             return BREAK_AT_START;
218     }
219 }
220 
getScript(uint32_t codePoint)221 static UScriptCode getScript(uint32_t codePoint) {
222     UErrorCode errorCode = U_ZERO_ERROR;
223     const UScriptCode script = uscript_getScript(static_cast<UChar32>(codePoint), &errorCode);
224     if (U_SUCCESS(errorCode)) {
225         return script;
226     } else {
227         return USCRIPT_INVALID_CODE;
228     }
229 }
230 
hyphenationTypeBasedOnScript(uint32_t codePoint)231 static HyphenationType hyphenationTypeBasedOnScript(uint32_t codePoint) {
232     // Note: It's not clear what the best hyphen for Hebrew is. While maqaf is the "correct" hyphen
233     // for Hebrew, modern practice may have shifted towards Western hyphens. We use normal hyphens
234     // for now to be safe.  BREAK_AND_INSERT_MAQAF is already implemented, so if we want to switch
235     // to maqaf for Hebrew, we can simply add a condition here.
236     const UScriptCode script = getScript(codePoint);
237     if (script == USCRIPT_KANNADA
238             || script == USCRIPT_MALAYALAM
239             || script == USCRIPT_TAMIL
240             || script == USCRIPT_TELUGU) {
241         // Grantha is not included, since we don't support non-BMP hyphenation yet.
242         return HyphenationType::BREAK_AND_DONT_INSERT_HYPHEN;
243     } else if (script == USCRIPT_ARMENIAN) {
244         return HyphenationType::BREAK_AND_INSERT_ARMENIAN_HYPHEN;
245     } else if (script == USCRIPT_CANADIAN_ABORIGINAL) {
246         return HyphenationType::BREAK_AND_INSERT_UCAS_HYPHEN;
247     } else {
248         return HyphenationType::BREAK_AND_INSERT_HYPHEN;
249     }
250 }
251 
getJoiningType(UChar32 codepoint)252 static inline int32_t getJoiningType(UChar32 codepoint) {
253     return u_getIntPropertyValue(codepoint, UCHAR_JOINING_TYPE);
254 }
255 
256 // Assumption for caller: location must be >= 2 and word[location] == CHAR_SOFT_HYPHEN.
257 // This function decides if the letters before and after the hyphen should appear as joining.
getHyphTypeForArabic(const uint16_t * word,size_t len,size_t location)258 static inline HyphenationType getHyphTypeForArabic(const uint16_t* word, size_t len,
259         size_t location) {
260     ssize_t i = location;
261     int32_t type = U_JT_NON_JOINING;
262     while (static_cast<size_t>(i) < len && (type = getJoiningType(word[i])) == U_JT_TRANSPARENT) {
263         i++;
264     }
265     if (type == U_JT_DUAL_JOINING || type == U_JT_RIGHT_JOINING || type == U_JT_JOIN_CAUSING) {
266         // The next character is of the type that may join the last character. See if the last
267         // character is also of the right type.
268         i = location - 2; // Skip the soft hyphen
269         type = U_JT_NON_JOINING;
270         while (i >= 0 && (type = getJoiningType(word[i])) == U_JT_TRANSPARENT) {
271             i--;
272         }
273         if (type == U_JT_DUAL_JOINING || type == U_JT_LEFT_JOINING || type == U_JT_JOIN_CAUSING) {
274             return HyphenationType::BREAK_AND_INSERT_HYPHEN_AND_ZWJ;
275         }
276     }
277     return HyphenationType::BREAK_AND_INSERT_HYPHEN;
278 }
279 
280 // Use various recommendations of UAX #14 Unicode Line Breaking Algorithm for hyphenating words
281 // that didn't match patterns, especially words that contain hyphens or soft hyphens (See sections
282 // 5.3, Use of Hyphen, and 5.4, Use of Soft Hyphen).
hyphenateWithNoPatterns(HyphenationType * result,const uint16_t * word,size_t len,const icu::Locale & locale)283 void Hyphenator::hyphenateWithNoPatterns(HyphenationType* result, const uint16_t* word, size_t len,
284         const icu::Locale& locale) {
285     result[0] = HyphenationType::DONT_BREAK;
286     for (size_t i = 1; i < len; i++) {
287         const uint16_t prevChar = word[i - 1];
288         if (i > 1 && isLineBreakingHyphen(prevChar)) {
289             // Break after hyphens, but only if they don't start the word.
290 
291             if ((prevChar == CHAR_HYPHEN_MINUS || prevChar == CHAR_HYPHEN)
292                     && strcmp(locale.getLanguage(), "pl") == 0
293                     && getScript(word[i]) == USCRIPT_LATIN ) {
294                 // In Polish, hyphens get repeated at the next line. To be safe,
295                 // we will do this only if the next character is Latin.
296                 result[i] = HyphenationType::BREAK_AND_INSERT_HYPHEN_AT_NEXT_LINE;
297             } else {
298                 result[i] = HyphenationType::BREAK_AND_DONT_INSERT_HYPHEN;
299             }
300         } else if (i > 1 && prevChar == CHAR_SOFT_HYPHEN) {
301             // Break after soft hyphens, but only if they don't start the word (a soft hyphen
302             // starting the word doesn't give any useful break opportunities). The type of the break
303             // is based on the script of the character we break on.
304             if (getScript(word[i]) == USCRIPT_ARABIC) {
305                 // For Arabic, we need to look and see if the characters around the soft hyphen
306                 // actually join. If they don't, we'll just insert a normal hyphen.
307                 result[i] = getHyphTypeForArabic(word, len, i);
308             } else {
309                 result[i] = hyphenationTypeBasedOnScript(word[i]);
310             }
311         } else if (prevChar == CHAR_MIDDLE_DOT
312                 && minPrefix < i && i <= len - minSuffix
313                 && ((word[i - 2] == 'l' && word[i] == 'l')
314                         || (word[i - 2] == 'L' && word[i] == 'L'))
315                 && strcmp(locale.getLanguage(), "ca") == 0) {
316             // In Catalan, "l·l" should break as "l-" on the first line
317             // and "l" on the next line.
318             result[i] = HyphenationType::BREAK_AND_REPLACE_WITH_HYPHEN;
319         } else {
320             result[i] = HyphenationType::DONT_BREAK;
321         }
322      }
323 }
324 
alphabetLookup(uint16_t * alpha_codes,const uint16_t * word,size_t len)325 HyphenationType Hyphenator::alphabetLookup(uint16_t* alpha_codes, const uint16_t* word,
326         size_t len) {
327     const Header* header = getHeader();
328     HyphenationType result = HyphenationType::BREAK_AND_INSERT_HYPHEN;
329     // TODO: check header magic
330     uint32_t alphabetVersion = header->alphabetVersion();
331     if (alphabetVersion == 0) {
332         const AlphabetTable0* alphabet = header->alphabetTable0();
333         uint32_t min_codepoint = alphabet->min_codepoint;
334         uint32_t max_codepoint = alphabet->max_codepoint;
335         alpha_codes[0] = 0;  // word start
336         for (size_t i = 0; i < len; i++) {
337             uint16_t c = word[i];
338             if (c < min_codepoint || c >= max_codepoint) {
339                 return HyphenationType::DONT_BREAK;
340             }
341             uint8_t code = alphabet->data[c - min_codepoint];
342             if (code == 0) {
343                 return HyphenationType::DONT_BREAK;
344             }
345             if (result == HyphenationType::BREAK_AND_INSERT_HYPHEN) {
346                 result = hyphenationTypeBasedOnScript(c);
347             }
348             alpha_codes[i + 1] = code;
349         }
350         alpha_codes[len + 1] = 0;  // word termination
351         return result;
352     } else if (alphabetVersion == 1) {
353         const AlphabetTable1* alphabet = header->alphabetTable1();
354         size_t n_entries = alphabet->n_entries;
355         const uint32_t* begin = alphabet->data;
356         const uint32_t* end = begin + n_entries;
357         alpha_codes[0] = 0;
358         for (size_t i = 0; i < len; i++) {
359             uint16_t c = word[i];
360             auto p = std::lower_bound(begin, end, c << 11);
361             if (p == end) {
362                 return HyphenationType::DONT_BREAK;
363             }
364             uint32_t entry = *p;
365             if (AlphabetTable1::codepoint(entry) != c) {
366                 return HyphenationType::DONT_BREAK;
367             }
368             if (result == HyphenationType::BREAK_AND_INSERT_HYPHEN) {
369                 result = hyphenationTypeBasedOnScript(c);
370             }
371             alpha_codes[i + 1] = AlphabetTable1::value(entry);
372         }
373         alpha_codes[len + 1] = 0;
374         return result;
375     }
376     return HyphenationType::DONT_BREAK;
377 }
378 
379 /**
380  * Internal implementation, after conversion to codes. All case folding and normalization
381  * has been done by now, and all characters have been found in the alphabet.
382  * Note: len here is the padded length including 0 codes at start and end.
383  **/
hyphenateFromCodes(HyphenationType * result,const uint16_t * codes,size_t len,HyphenationType hyphenValue)384 void Hyphenator::hyphenateFromCodes(HyphenationType* result, const uint16_t* codes, size_t len,
385         HyphenationType hyphenValue) {
386     static_assert(sizeof(HyphenationType) == sizeof(uint8_t), "HyphnationType must be uint8_t.");
387     // Reuse the result array as a buffer for calculating intermediate hyphenation numbers.
388     uint8_t* buffer = reinterpret_cast<uint8_t*>(result);
389 
390     const Header* header = getHeader();
391     const Trie* trie = header->trieTable();
392     const Pattern* pattern = header->patternTable();
393     uint32_t char_mask = trie->char_mask;
394     uint32_t link_shift = trie->link_shift;
395     uint32_t link_mask = trie->link_mask;
396     uint32_t pattern_shift = trie->pattern_shift;
397     size_t maxOffset = len - minSuffix - 1;
398     for (size_t i = 0; i < len - 1; i++) {
399         uint32_t node = 0;  // index into Trie table
400         for (size_t j = i; j < len; j++) {
401             uint16_t c = codes[j];
402             uint32_t entry = trie->data[node + c];
403             if ((entry & char_mask) == c) {
404                 node = (entry & link_mask) >> link_shift;
405             } else {
406                 break;
407             }
408             uint32_t pat_ix = trie->data[node] >> pattern_shift;
409             // pat_ix contains a 3-tuple of length, shift (number of trailing zeros), and an offset
410             // into the buf pool. This is the pattern for the substring (i..j) we just matched,
411             // which we combine (via point-wise max) into the buffer vector.
412             if (pat_ix != 0) {
413                 uint32_t pat_entry = pattern->data[pat_ix];
414                 int pat_len = Pattern::len(pat_entry);
415                 int pat_shift = Pattern::shift(pat_entry);
416                 const uint8_t* pat_buf = pattern->buf(pat_entry);
417                 int offset = j + 1 - (pat_len + pat_shift);
418                 // offset is the index within buffer that lines up with the start of pat_buf
419                 int start = std::max((int)minPrefix - offset, 0);
420                 int end = std::min(pat_len, (int)maxOffset - offset);
421                 for (int k = start; k < end; k++) {
422                     buffer[offset + k] = std::max(buffer[offset + k], pat_buf[k]);
423                 }
424             }
425         }
426     }
427     // Since the above calculation does not modify values outside
428     // [minPrefix, len - minSuffix], they are left as 0 = DONT_BREAK.
429     for (size_t i = minPrefix; i < maxOffset; i++) {
430         // Hyphenation opportunities happen when the hyphenation numbers are odd.
431         result[i] = (buffer[i] & 1u) ? hyphenValue : HyphenationType::DONT_BREAK;
432     }
433 }
434 
435 }  // namespace minikin
436