]>
Commit | Line | Data |
---|---|---|
73c04bcf A |
1 | /** |
2 | ******************************************************************************* | |
3 | * Copyright (C) 2006, International Business Machines Corporation and others. * | |
4 | * All Rights Reserved. * | |
5 | ******************************************************************************* | |
6 | */ | |
7 | ||
8 | #include "unicode/utypes.h" | |
9 | ||
10 | #if !UCONFIG_NO_BREAK_ITERATION | |
11 | ||
12 | #include "triedict.h" | |
13 | #include "unicode/chariter.h" | |
14 | #include "unicode/uchriter.h" | |
15 | #include "unicode/strenum.h" | |
16 | #include "unicode/uenum.h" | |
17 | #include "unicode/udata.h" | |
18 | #include "cmemory.h" | |
19 | #include "udataswp.h" | |
20 | #include "uvector.h" | |
21 | #include "uvectr32.h" | |
22 | #include "uarrsort.h" | |
23 | ||
24 | //#define DEBUG_TRIE_DICT 1 | |
25 | ||
26 | #ifdef DEBUG_TRIE_DICT | |
27 | #include <sys/times.h> | |
28 | #include <limits.h> | |
29 | #include <stdio.h> | |
30 | #endif | |
31 | ||
32 | U_NAMESPACE_BEGIN | |
33 | ||
34 | /******************************************************************* | |
35 | * TrieWordDictionary | |
36 | */ | |
37 | ||
38 | TrieWordDictionary::TrieWordDictionary() { | |
39 | } | |
40 | ||
41 | TrieWordDictionary::~TrieWordDictionary() { | |
42 | } | |
43 | ||
44 | /******************************************************************* | |
45 | * MutableTrieDictionary | |
46 | */ | |
47 | ||
48 | // Node structure for the ternary, uncompressed trie | |
49 | struct TernaryNode : public UMemory { | |
50 | UChar ch; // UTF-16 code unit | |
51 | uint16_t flags; // Flag word | |
52 | TernaryNode *low; // Less-than link | |
53 | TernaryNode *equal; // Equal link | |
54 | TernaryNode *high; // Greater-than link | |
55 | ||
56 | TernaryNode(UChar uc); | |
57 | ~TernaryNode(); | |
58 | }; | |
59 | ||
60 | enum MutableTrieNodeFlags { | |
61 | kEndsWord = 0x0001 // This node marks the end of a valid word | |
62 | }; | |
63 | ||
64 | inline | |
65 | TernaryNode::TernaryNode(UChar uc) { | |
66 | ch = uc; | |
67 | flags = 0; | |
68 | low = NULL; | |
69 | equal = NULL; | |
70 | high = NULL; | |
71 | } | |
72 | ||
73 | // Not inline since it's recursive | |
74 | TernaryNode::~TernaryNode() { | |
75 | delete low; | |
76 | delete equal; | |
77 | delete high; | |
78 | } | |
79 | ||
80 | MutableTrieDictionary::MutableTrieDictionary( UChar median, UErrorCode &status ) { | |
81 | // Start the trie off with something. Having the root node already present | |
82 | // cuts a special case out of the search/insertion functions. | |
83 | // Making it a median character cuts the worse case for searches from | |
84 | // 4x a balanced trie to 2x a balanced trie. It's best to choose something | |
85 | // that starts a word that is midway in the list. | |
86 | fTrie = new TernaryNode(median); | |
87 | if (fTrie == NULL) { | |
88 | status = U_MEMORY_ALLOCATION_ERROR; | |
89 | } | |
90 | fIter = utext_openUChars(NULL, NULL, 0, &status); | |
91 | if (U_SUCCESS(status) && fIter == NULL) { | |
92 | status = U_MEMORY_ALLOCATION_ERROR; | |
93 | } | |
94 | } | |
95 | ||
96 | MutableTrieDictionary::MutableTrieDictionary( UErrorCode &status ) { | |
97 | fTrie = NULL; | |
98 | fIter = utext_openUChars(NULL, NULL, 0, &status); | |
99 | if (U_SUCCESS(status) && fIter == NULL) { | |
100 | status = U_MEMORY_ALLOCATION_ERROR; | |
101 | } | |
102 | } | |
103 | ||
104 | MutableTrieDictionary::~MutableTrieDictionary() { | |
105 | delete fTrie; | |
106 | utext_close(fIter); | |
107 | } | |
108 | ||
109 | int32_t | |
110 | MutableTrieDictionary::search( UText *text, | |
111 | int32_t maxLength, | |
112 | int32_t *lengths, | |
113 | int &count, | |
114 | int limit, | |
115 | TernaryNode *&parent, | |
116 | UBool &pMatched ) const { | |
117 | // TODO: current implementation works in UTF-16 space | |
118 | const TernaryNode *up = NULL; | |
119 | const TernaryNode *p = fTrie; | |
120 | int mycount = 0; | |
121 | pMatched = TRUE; | |
122 | int i; | |
123 | ||
124 | UChar uc = utext_current32(text); | |
125 | for (i = 0; i < maxLength && p != NULL; ++i) { | |
126 | while (p != NULL) { | |
127 | if (uc < p->ch) { | |
128 | up = p; | |
129 | p = p->low; | |
130 | } | |
131 | else if (uc == p->ch) { | |
132 | break; | |
133 | } | |
134 | else { | |
135 | up = p; | |
136 | p = p->high; | |
137 | } | |
138 | } | |
139 | if (p == NULL) { | |
140 | pMatched = FALSE; | |
141 | break; | |
142 | } | |
143 | // Must be equal to get here | |
144 | if (limit > 0 && (p->flags & kEndsWord)) { | |
145 | lengths[mycount++] = i+1; | |
146 | --limit; | |
147 | } | |
148 | up = p; | |
149 | p = p->equal; | |
150 | uc = utext_next32(text); | |
151 | uc = utext_current32(text); | |
152 | } | |
153 | ||
154 | // Note that there is no way to reach here with up == 0 unless | |
155 | // maxLength is 0 coming in. | |
156 | parent = (TernaryNode *)up; | |
157 | count = mycount; | |
158 | return i; | |
159 | } | |
160 | ||
161 | void | |
162 | MutableTrieDictionary::addWord( const UChar *word, | |
163 | int32_t length, | |
164 | UErrorCode &status ) { | |
165 | #if 0 | |
166 | if (length <= 0) { | |
167 | status = U_ILLEGAL_ARGUMENT_ERROR; | |
168 | return; | |
169 | } | |
170 | #endif | |
171 | TernaryNode *parent; | |
172 | UBool pMatched; | |
173 | int count; | |
174 | fIter = utext_openUChars(fIter, word, length, &status); | |
175 | ||
176 | int matched; | |
177 | matched = search(fIter, length, NULL, count, 0, parent, pMatched); | |
178 | ||
179 | while (matched++ < length) { | |
180 | UChar32 uc = utext_next32(fIter); // TODO: supplemetary support? | |
181 | U_ASSERT(uc != U_SENTINEL); | |
182 | TernaryNode *newNode = new TernaryNode(uc); | |
183 | if (newNode == NULL) { | |
184 | status = U_MEMORY_ALLOCATION_ERROR; | |
185 | return; | |
186 | } | |
187 | if (pMatched) { | |
188 | parent->equal = newNode; | |
189 | } | |
190 | else { | |
191 | pMatched = TRUE; | |
192 | if (uc < parent->ch) { | |
193 | parent->low = newNode; | |
194 | } | |
195 | else { | |
196 | parent->high = newNode; | |
197 | } | |
198 | } | |
199 | parent = newNode; | |
200 | } | |
201 | ||
202 | parent->flags |= kEndsWord; | |
203 | } | |
204 | ||
205 | #if 0 | |
206 | void | |
207 | MutableTrieDictionary::addWords( UEnumeration *words, | |
208 | UErrorCode &status ) { | |
209 | int32_t length; | |
210 | const UChar *word; | |
211 | while ((word = uenum_unext(words, &length, &status)) && U_SUCCESS(status)) { | |
212 | addWord(word, length, status); | |
213 | } | |
214 | } | |
215 | #endif | |
216 | ||
217 | int32_t | |
218 | MutableTrieDictionary::matches( UText *text, | |
219 | int32_t maxLength, | |
220 | int32_t *lengths, | |
221 | int &count, | |
222 | int limit ) const { | |
223 | TernaryNode *parent; | |
224 | UBool pMatched; | |
225 | return search(text, maxLength, lengths, count, limit, parent, pMatched); | |
226 | } | |
227 | ||
228 | // Implementation of iteration for MutableTrieDictionary | |
229 | class MutableTrieEnumeration : public StringEnumeration { | |
230 | private: | |
231 | UStack fNodeStack; // Stack of nodes to process | |
232 | UVector32 fBranchStack; // Stack of which branch we are working on | |
233 | TernaryNode *fRoot; // Root node | |
234 | static const char fgClassID; | |
235 | enum StackBranch { | |
236 | kLessThan, | |
237 | kEqual, | |
238 | kGreaterThan, | |
239 | kDone | |
240 | }; | |
241 | ||
242 | public: | |
243 | static UClassID U_EXPORT2 getStaticClassID(void) { return (UClassID)&fgClassID; } | |
244 | virtual UClassID getDynamicClassID(void) const { return getStaticClassID(); } | |
245 | public: | |
246 | MutableTrieEnumeration(TernaryNode *root, UErrorCode &status) | |
247 | : fNodeStack(status), fBranchStack(status) { | |
248 | fRoot = root; | |
249 | fNodeStack.push(root, status); | |
250 | fBranchStack.push(kLessThan, status); | |
251 | unistr.remove(); | |
252 | } | |
253 | ||
254 | virtual ~MutableTrieEnumeration() { | |
255 | } | |
256 | ||
257 | virtual StringEnumeration *clone() const { | |
258 | UErrorCode status = U_ZERO_ERROR; | |
259 | return new MutableTrieEnumeration(fRoot, status); | |
260 | } | |
261 | ||
262 | virtual const UnicodeString *snext(UErrorCode &status) { | |
263 | if (fNodeStack.empty() || U_FAILURE(status)) { | |
264 | return NULL; | |
265 | } | |
266 | TernaryNode *node = (TernaryNode *) fNodeStack.peek(); | |
267 | StackBranch where = (StackBranch) fBranchStack.peeki(); | |
268 | while (!fNodeStack.empty() && U_SUCCESS(status)) { | |
269 | UBool emit; | |
270 | UBool equal; | |
271 | ||
272 | switch (where) { | |
273 | case kLessThan: | |
274 | if (node->low != NULL) { | |
275 | fBranchStack.setElementAt(kEqual, fBranchStack.size()-1); | |
276 | node = (TernaryNode *) fNodeStack.push(node->low, status); | |
277 | where = (StackBranch) fBranchStack.push(kLessThan, status); | |
278 | break; | |
279 | } | |
280 | case kEqual: | |
281 | emit = (node->flags & kEndsWord) != 0; | |
282 | equal = (node->equal != NULL); | |
283 | // If this node should be part of the next emitted string, append | |
284 | // the UChar to the string, and make sure we pop it when we come | |
285 | // back to this node. The character should only be in the string | |
286 | // for as long as we're traversing the equal subtree of this node | |
287 | if (equal || emit) { | |
288 | unistr.append(node->ch); | |
289 | fBranchStack.setElementAt(kGreaterThan, fBranchStack.size()-1); | |
290 | } | |
291 | if (equal) { | |
292 | node = (TernaryNode *) fNodeStack.push(node->equal, status); | |
293 | where = (StackBranch) fBranchStack.push(kLessThan, status); | |
294 | } | |
295 | if (emit) { | |
296 | return &unistr; | |
297 | } | |
298 | if (equal) { | |
299 | break; | |
300 | } | |
301 | case kGreaterThan: | |
302 | // If this node's character is in the string, remove it. | |
303 | if (node->equal != NULL || (node->flags & kEndsWord)) { | |
304 | unistr.truncate(unistr.length()-1); | |
305 | } | |
306 | if (node->high != NULL) { | |
307 | fBranchStack.setElementAt(kDone, fBranchStack.size()-1); | |
308 | node = (TernaryNode *) fNodeStack.push(node->high, status); | |
309 | where = (StackBranch) fBranchStack.push(kLessThan, status); | |
310 | break; | |
311 | } | |
312 | case kDone: | |
313 | fNodeStack.pop(); | |
314 | fBranchStack.popi(); | |
315 | node = (TernaryNode *) fNodeStack.peek(); | |
316 | where = (StackBranch) fBranchStack.peeki(); | |
317 | break; | |
318 | default: | |
319 | return NULL; | |
320 | } | |
321 | } | |
322 | return NULL; | |
323 | } | |
324 | ||
325 | // Very expensive, but this should never be used. | |
326 | virtual int32_t count(UErrorCode &status) const { | |
327 | MutableTrieEnumeration counter(fRoot, status); | |
328 | int32_t result = 0; | |
329 | while (counter.snext(status) != NULL && U_SUCCESS(status)) { | |
330 | ++result; | |
331 | } | |
332 | return result; | |
333 | } | |
334 | ||
335 | virtual void reset(UErrorCode &status) { | |
336 | fNodeStack.removeAllElements(); | |
337 | fBranchStack.removeAllElements(); | |
338 | fNodeStack.push(fRoot, status); | |
339 | fBranchStack.push(kLessThan, status); | |
340 | unistr.remove(); | |
341 | } | |
342 | }; | |
343 | ||
344 | const char MutableTrieEnumeration::fgClassID = '\0'; | |
345 | ||
346 | StringEnumeration * | |
347 | MutableTrieDictionary::openWords( UErrorCode &status ) const { | |
348 | if (U_FAILURE(status)) { | |
349 | return NULL; | |
350 | } | |
351 | return new MutableTrieEnumeration(fTrie, status); | |
352 | } | |
353 | ||
354 | /******************************************************************* | |
355 | * CompactTrieDictionary | |
356 | */ | |
357 | ||
358 | struct CompactTrieHeader { | |
359 | uint32_t size; // Size of the data in bytes | |
360 | uint32_t magic; // Magic number (including version) | |
361 | uint16_t nodeCount; // Number of entries in offsets[] | |
362 | uint16_t root; // Node number of the root node | |
363 | uint32_t offsets[1]; // Offsets to nodes from start of data | |
364 | }; | |
365 | ||
366 | // Note that to avoid platform-specific alignment issues, all members of the node | |
367 | // structures should be the same size, or should contain explicit padding to | |
368 | // natural alignment boundaries. | |
369 | ||
370 | // We can't use a bitfield for the flags+count field, because the layout of those | |
371 | // is not portable. 12 bits of count allows for up to 4096 entries in a node. | |
372 | struct CompactTrieNode { | |
373 | uint16_t flagscount; // Count of sub-entries, plus flags | |
374 | }; | |
375 | ||
376 | enum CompactTrieNodeFlags { | |
377 | kVerticalNode = 0x1000, // This is a vertical node | |
378 | kParentEndsWord = 0x2000, // The node whose equal link points to this ends a word | |
379 | kReservedFlag1 = 0x4000, | |
380 | kReservedFlag2 = 0x8000, | |
381 | kCountMask = 0x0FFF, // The count portion of flagscount | |
382 | kFlagMask = 0xF000 // The flags portion of flagscount | |
383 | }; | |
384 | ||
385 | // The two node types are distinguished by the kVerticalNode flag. | |
386 | ||
387 | struct CompactTrieHorizontalEntry { | |
388 | uint16_t ch; // UChar | |
389 | uint16_t equal; // Equal link node index | |
390 | }; | |
391 | ||
392 | // We don't use inheritance here because C++ does not guarantee that the | |
393 | // base class comes first in memory!! | |
394 | ||
395 | struct CompactTrieHorizontalNode { | |
396 | uint16_t flagscount; // Count of sub-entries, plus flags | |
397 | CompactTrieHorizontalEntry entries[1]; | |
398 | }; | |
399 | ||
400 | struct CompactTrieVerticalNode { | |
401 | uint16_t flagscount; // Count of sub-entries, plus flags | |
402 | uint16_t equal; // Equal link node index | |
403 | uint16_t chars[1]; // Code units | |
404 | }; | |
405 | ||
406 | // {'Dic', 1}, version 1 | |
407 | #define COMPACT_TRIE_MAGIC_1 0x44696301 | |
408 | ||
409 | CompactTrieDictionary::CompactTrieDictionary(UDataMemory *dataObj, | |
410 | UErrorCode &status ) | |
411 | : fUData(dataObj) | |
412 | { | |
413 | fData = (const CompactTrieHeader *) udata_getMemory(dataObj); | |
414 | fOwnData = FALSE; | |
415 | if (fData->magic != COMPACT_TRIE_MAGIC_1) { | |
416 | status = U_ILLEGAL_ARGUMENT_ERROR; | |
417 | fData = NULL; | |
418 | } | |
419 | } | |
420 | CompactTrieDictionary::CompactTrieDictionary( const void *data, | |
421 | UErrorCode &status ) | |
422 | : fUData(NULL) | |
423 | { | |
424 | fData = (const CompactTrieHeader *) data; | |
425 | fOwnData = FALSE; | |
426 | if (fData->magic != COMPACT_TRIE_MAGIC_1) { | |
427 | status = U_ILLEGAL_ARGUMENT_ERROR; | |
428 | fData = NULL; | |
429 | } | |
430 | } | |
431 | ||
432 | CompactTrieDictionary::CompactTrieDictionary( const MutableTrieDictionary &dict, | |
433 | UErrorCode &status ) | |
434 | : fUData(NULL) | |
435 | { | |
436 | fData = compactMutableTrieDictionary(dict, status); | |
437 | fOwnData = !U_FAILURE(status); | |
438 | } | |
439 | ||
440 | CompactTrieDictionary::~CompactTrieDictionary() { | |
441 | if (fOwnData) { | |
442 | uprv_free((void *)fData); | |
443 | } | |
444 | if (fUData) { | |
445 | udata_close(fUData); | |
446 | } | |
447 | } | |
448 | ||
449 | uint32_t | |
450 | CompactTrieDictionary::dataSize() const { | |
451 | return fData->size; | |
452 | } | |
453 | ||
454 | const void * | |
455 | CompactTrieDictionary::data() const { | |
456 | return fData; | |
457 | } | |
458 | ||
459 | // This function finds the address of a node for us, given its node ID | |
460 | static inline const CompactTrieNode * | |
461 | getCompactNode(const CompactTrieHeader *header, uint16_t node) { | |
462 | return (const CompactTrieNode *)((const uint8_t *)header + header->offsets[node]); | |
463 | } | |
464 | ||
465 | int32_t | |
466 | CompactTrieDictionary::matches( UText *text, | |
467 | int32_t maxLength, | |
468 | int32_t *lengths, | |
469 | int &count, | |
470 | int limit ) const { | |
471 | // TODO: current implementation works in UTF-16 space | |
472 | const CompactTrieNode *node = getCompactNode(fData, fData->root); | |
473 | int mycount = 0; | |
474 | ||
475 | UChar uc = utext_current32(text); | |
476 | int i = 0; | |
477 | ||
478 | while (node != NULL) { | |
479 | // Check if the node we just exited ends a word | |
480 | if (limit > 0 && (node->flagscount & kParentEndsWord)) { | |
481 | lengths[mycount++] = i; | |
482 | --limit; | |
483 | } | |
484 | // Check that we haven't exceeded the maximum number of input characters. | |
485 | // We have to do that here rather than in the while condition so that | |
486 | // we can check for ending a word, above. | |
487 | if (i >= maxLength) { | |
488 | break; | |
489 | } | |
490 | ||
491 | int nodeCount = (node->flagscount & kCountMask); | |
492 | if (nodeCount == 0) { | |
493 | // Special terminal node; return now | |
494 | break; | |
495 | } | |
496 | if (node->flagscount & kVerticalNode) { | |
497 | // Vertical node; check all the characters in it | |
498 | const CompactTrieVerticalNode *vnode = (const CompactTrieVerticalNode *)node; | |
499 | for (int j = 0; j < nodeCount && i < maxLength; ++j) { | |
500 | if (uc != vnode->chars[j]) { | |
501 | // We hit a non-equal character; return | |
502 | goto exit; | |
503 | } | |
504 | utext_next32(text); | |
505 | uc = utext_current32(text); | |
506 | ++i; | |
507 | } | |
508 | // To get here we must have come through the whole list successfully; | |
509 | // go on to the next node. Note that a word cannot end in the middle | |
510 | // of a vertical node. | |
511 | node = getCompactNode(fData, vnode->equal); | |
512 | } | |
513 | else { | |
514 | // Horizontal node; do binary search | |
515 | const CompactTrieHorizontalNode *hnode = (const CompactTrieHorizontalNode *)node; | |
516 | int low = 0; | |
517 | int high = nodeCount-1; | |
518 | int middle; | |
519 | node = NULL; // If we don't find a match, we'll fall out of the loop | |
520 | while (high >= low) { | |
521 | middle = (high+low)/2; | |
522 | if (uc == hnode->entries[middle].ch) { | |
523 | // We hit a match; get the next node and next character | |
524 | node = getCompactNode(fData, hnode->entries[middle].equal); | |
525 | utext_next32(text); | |
526 | uc = utext_current32(text); | |
527 | ++i; | |
528 | break; | |
529 | } | |
530 | else if (uc < hnode->entries[middle].ch) { | |
531 | high = middle-1; | |
532 | } | |
533 | else { | |
534 | low = middle+1; | |
535 | } | |
536 | } | |
537 | } | |
538 | } | |
539 | exit: | |
540 | count = mycount; | |
541 | return i; | |
542 | } | |
543 | ||
544 | // Implementation of iteration for CompactTrieDictionary | |
545 | class CompactTrieEnumeration : public StringEnumeration { | |
546 | private: | |
547 | UVector32 fNodeStack; // Stack of nodes to process | |
548 | UVector32 fIndexStack; // Stack of where in node we are | |
549 | const CompactTrieHeader *fHeader; // Trie data | |
550 | static const char fgClassID; | |
551 | ||
552 | public: | |
553 | static UClassID U_EXPORT2 getStaticClassID(void) { return (UClassID)&fgClassID; } | |
554 | virtual UClassID getDynamicClassID(void) const { return getStaticClassID(); } | |
555 | public: | |
556 | CompactTrieEnumeration(const CompactTrieHeader *header, UErrorCode &status) | |
557 | : fNodeStack(status), fIndexStack(status) { | |
558 | fHeader = header; | |
559 | fNodeStack.push(header->root, status); | |
560 | fIndexStack.push(0, status); | |
561 | unistr.remove(); | |
562 | } | |
563 | ||
564 | virtual ~CompactTrieEnumeration() { | |
565 | } | |
566 | ||
567 | virtual StringEnumeration *clone() const { | |
568 | UErrorCode status = U_ZERO_ERROR; | |
569 | return new CompactTrieEnumeration(fHeader, status); | |
570 | } | |
571 | ||
572 | virtual const UnicodeString * snext(UErrorCode &status); | |
573 | ||
574 | // Very expensive, but this should never be used. | |
575 | virtual int32_t count(UErrorCode &status) const { | |
576 | CompactTrieEnumeration counter(fHeader, status); | |
577 | int32_t result = 0; | |
578 | while (counter.snext(status) != NULL && U_SUCCESS(status)) { | |
579 | ++result; | |
580 | } | |
581 | return result; | |
582 | } | |
583 | ||
584 | virtual void reset(UErrorCode &status) { | |
585 | fNodeStack.removeAllElements(); | |
586 | fIndexStack.removeAllElements(); | |
587 | fNodeStack.push(fHeader->root, status); | |
588 | fIndexStack.push(0, status); | |
589 | unistr.remove(); | |
590 | } | |
591 | }; | |
592 | ||
593 | const char CompactTrieEnumeration::fgClassID = '\0'; | |
594 | ||
595 | const UnicodeString * | |
596 | CompactTrieEnumeration::snext(UErrorCode &status) { | |
597 | if (fNodeStack.empty() || U_FAILURE(status)) { | |
598 | return NULL; | |
599 | } | |
600 | const CompactTrieNode *node = getCompactNode(fHeader, fNodeStack.peeki()); | |
601 | int where = fIndexStack.peeki(); | |
602 | while (!fNodeStack.empty() && U_SUCCESS(status)) { | |
603 | int nodeCount = (node->flagscount & kCountMask); | |
604 | UBool goingDown = FALSE; | |
605 | if (nodeCount == 0) { | |
606 | // Terminal node; go up immediately | |
607 | fNodeStack.popi(); | |
608 | fIndexStack.popi(); | |
609 | node = getCompactNode(fHeader, fNodeStack.peeki()); | |
610 | where = fIndexStack.peeki(); | |
611 | } | |
612 | else if (node->flagscount & kVerticalNode) { | |
613 | // Vertical node | |
614 | const CompactTrieVerticalNode *vnode = (const CompactTrieVerticalNode *)node; | |
615 | if (where == 0) { | |
616 | // Going down | |
617 | unistr.append((const UChar *)vnode->chars, (int32_t) nodeCount); | |
618 | fIndexStack.setElementAt(1, fIndexStack.size()-1); | |
619 | node = getCompactNode(fHeader, fNodeStack.push(vnode->equal, status)); | |
620 | where = fIndexStack.push(0, status); | |
621 | goingDown = TRUE; | |
622 | } | |
623 | else { | |
624 | // Going up | |
625 | unistr.truncate(unistr.length()-nodeCount); | |
626 | fNodeStack.popi(); | |
627 | fIndexStack.popi(); | |
628 | node = getCompactNode(fHeader, fNodeStack.peeki()); | |
629 | where = fIndexStack.peeki(); | |
630 | } | |
631 | } | |
632 | else { | |
633 | // Horizontal node | |
634 | const CompactTrieHorizontalNode *hnode = (const CompactTrieHorizontalNode *)node; | |
635 | if (where > 0) { | |
636 | // Pop previous char | |
637 | unistr.truncate(unistr.length()-1); | |
638 | } | |
639 | if (where < nodeCount) { | |
640 | // Push on next node | |
641 | unistr.append((UChar)hnode->entries[where].ch); | |
642 | fIndexStack.setElementAt(where+1, fIndexStack.size()-1); | |
643 | node = getCompactNode(fHeader, fNodeStack.push(hnode->entries[where].equal, status)); | |
644 | where = fIndexStack.push(0, status); | |
645 | goingDown = TRUE; | |
646 | } | |
647 | else { | |
648 | // Going up | |
649 | fNodeStack.popi(); | |
650 | fIndexStack.popi(); | |
651 | node = getCompactNode(fHeader, fNodeStack.peeki()); | |
652 | where = fIndexStack.peeki(); | |
653 | } | |
654 | } | |
655 | // Check if the parent of the node we've just gone down to ends a | |
656 | // word. If so, return it. | |
657 | if (goingDown && (node->flagscount & kParentEndsWord)) { | |
658 | return &unistr; | |
659 | } | |
660 | } | |
661 | return NULL; | |
662 | } | |
663 | ||
664 | StringEnumeration * | |
665 | CompactTrieDictionary::openWords( UErrorCode &status ) const { | |
666 | if (U_FAILURE(status)) { | |
667 | return NULL; | |
668 | } | |
669 | return new CompactTrieEnumeration(fData, status); | |
670 | } | |
671 | ||
672 | // | |
673 | // Below here is all code related to converting a ternary trie to a compact trie | |
674 | // and back again | |
675 | // | |
676 | ||
677 | // Helper classes to construct the compact trie | |
678 | class BuildCompactTrieNode: public UMemory { | |
679 | public: | |
680 | UBool fParentEndsWord; | |
681 | UBool fVertical; | |
682 | UBool fHasDuplicate; | |
683 | int32_t fNodeID; | |
684 | UnicodeString fChars; | |
685 | ||
686 | public: | |
687 | BuildCompactTrieNode(UBool parentEndsWord, UBool vertical, UStack &nodes, UErrorCode &status) { | |
688 | fParentEndsWord = parentEndsWord; | |
689 | fHasDuplicate = FALSE; | |
690 | fVertical = vertical; | |
691 | fNodeID = nodes.size(); | |
692 | nodes.push(this, status); | |
693 | } | |
694 | ||
695 | virtual ~BuildCompactTrieNode() { | |
696 | } | |
697 | ||
698 | virtual uint32_t size() { | |
699 | return sizeof(uint16_t); | |
700 | } | |
701 | ||
702 | virtual void write(uint8_t *bytes, uint32_t &offset, const UVector32 &/*translate*/) { | |
703 | // Write flag/count | |
704 | *((uint16_t *)(bytes+offset)) = (fChars.length() & kCountMask) | |
705 | | (fVertical ? kVerticalNode : 0) | (fParentEndsWord ? kParentEndsWord : 0 ); | |
706 | offset += sizeof(uint16_t); | |
707 | } | |
708 | }; | |
709 | ||
710 | class BuildCompactTrieHorizontalNode: public BuildCompactTrieNode { | |
711 | public: | |
712 | UStack fLinks; | |
713 | ||
714 | public: | |
715 | BuildCompactTrieHorizontalNode(UBool parentEndsWord, UStack &nodes, UErrorCode &status) | |
716 | : BuildCompactTrieNode(parentEndsWord, FALSE, nodes, status), fLinks(status) { | |
717 | } | |
718 | ||
719 | virtual ~BuildCompactTrieHorizontalNode() { | |
720 | } | |
721 | ||
722 | virtual uint32_t size() { | |
723 | return offsetof(CompactTrieHorizontalNode,entries) + | |
724 | (fChars.length()*sizeof(CompactTrieHorizontalEntry)); | |
725 | } | |
726 | ||
727 | virtual void write(uint8_t *bytes, uint32_t &offset, const UVector32 &translate) { | |
728 | BuildCompactTrieNode::write(bytes, offset, translate); | |
729 | int32_t count = fChars.length(); | |
730 | for (int32_t i = 0; i < count; ++i) { | |
731 | CompactTrieHorizontalEntry *entry = (CompactTrieHorizontalEntry *)(bytes+offset); | |
732 | entry->ch = fChars[i]; | |
733 | entry->equal = translate.elementAti(((BuildCompactTrieNode *)fLinks[i])->fNodeID); | |
734 | #ifdef DEBUG_TRIE_DICT | |
735 | if (entry->equal == 0) { | |
736 | fprintf(stderr, "ERROR: horizontal link %d, logical node %d maps to physical node zero\n", | |
737 | i, ((BuildCompactTrieNode *)fLinks[i])->fNodeID); | |
738 | } | |
739 | #endif | |
740 | offset += sizeof(CompactTrieHorizontalEntry); | |
741 | } | |
742 | } | |
743 | ||
744 | void addNode(UChar ch, BuildCompactTrieNode *link, UErrorCode &status) { | |
745 | fChars.append(ch); | |
746 | fLinks.push(link, status); | |
747 | } | |
748 | }; | |
749 | ||
750 | class BuildCompactTrieVerticalNode: public BuildCompactTrieNode { | |
751 | public: | |
752 | BuildCompactTrieNode *fEqual; | |
753 | ||
754 | public: | |
755 | BuildCompactTrieVerticalNode(UBool parentEndsWord, UStack &nodes, UErrorCode &status) | |
756 | : BuildCompactTrieNode(parentEndsWord, TRUE, nodes, status) { | |
757 | fEqual = NULL; | |
758 | } | |
759 | ||
760 | virtual ~BuildCompactTrieVerticalNode() { | |
761 | } | |
762 | ||
763 | virtual uint32_t size() { | |
764 | return offsetof(CompactTrieVerticalNode,chars) + (fChars.length()*sizeof(uint16_t)); | |
765 | } | |
766 | ||
767 | virtual void write(uint8_t *bytes, uint32_t &offset, const UVector32 &translate) { | |
768 | CompactTrieVerticalNode *node = (CompactTrieVerticalNode *)(bytes+offset); | |
769 | BuildCompactTrieNode::write(bytes, offset, translate); | |
770 | node->equal = translate.elementAti(fEqual->fNodeID); | |
771 | offset += sizeof(node->equal); | |
772 | #ifdef DEBUG_TRIE_DICT | |
773 | if (node->equal == 0) { | |
774 | fprintf(stderr, "ERROR: vertical link, logical node %d maps to physical node zero\n", | |
775 | fEqual->fNodeID); | |
776 | } | |
777 | #endif | |
778 | fChars.extract(0, fChars.length(), (UChar *)node->chars); | |
779 | offset += sizeof(uint16_t)*fChars.length(); | |
780 | } | |
781 | ||
782 | void addChar(UChar ch) { | |
783 | fChars.append(ch); | |
784 | } | |
785 | ||
786 | void setLink(BuildCompactTrieNode *node) { | |
787 | fEqual = node; | |
788 | } | |
789 | }; | |
790 | ||
791 | // Forward declaration | |
792 | static void walkHorizontal(const TernaryNode *node, | |
793 | BuildCompactTrieHorizontalNode *building, | |
794 | UStack &nodes, | |
795 | UErrorCode &status); | |
796 | ||
797 | // Convert one node. Uses recursion. | |
798 | ||
799 | static BuildCompactTrieNode * | |
800 | compactOneNode(const TernaryNode *node, UBool parentEndsWord, UStack &nodes, UErrorCode &status) { | |
801 | if (U_FAILURE(status)) { | |
802 | return NULL; | |
803 | } | |
804 | BuildCompactTrieNode *result = NULL; | |
805 | UBool horizontal = (node->low != NULL || node->high != NULL); | |
806 | if (horizontal) { | |
807 | BuildCompactTrieHorizontalNode *hResult = | |
808 | new BuildCompactTrieHorizontalNode(parentEndsWord, nodes, status); | |
809 | if (hResult == NULL) { | |
810 | status = U_MEMORY_ALLOCATION_ERROR; | |
811 | } | |
812 | if (U_SUCCESS(status)) { | |
813 | walkHorizontal(node, hResult, nodes, status); | |
814 | result = hResult; | |
815 | } | |
816 | } | |
817 | else { | |
818 | BuildCompactTrieVerticalNode *vResult = | |
819 | new BuildCompactTrieVerticalNode(parentEndsWord, nodes, status); | |
820 | if (vResult == NULL) { | |
821 | status = U_MEMORY_ALLOCATION_ERROR; | |
822 | } | |
823 | if (U_SUCCESS(status)) { | |
824 | UBool endsWord = FALSE; | |
825 | // Take up nodes until we end a word, or hit a node with < or > links | |
826 | do { | |
827 | vResult->addChar(node->ch); | |
828 | endsWord = (node->flags & kEndsWord) != 0; | |
829 | node = node->equal; | |
830 | } | |
831 | while(node != NULL && !endsWord && node->low == NULL && node->high == NULL); | |
832 | if (node == NULL) { | |
833 | if (!endsWord) { | |
834 | status = U_ILLEGAL_ARGUMENT_ERROR; // Corrupt input trie | |
835 | } | |
836 | else { | |
837 | vResult->setLink((BuildCompactTrieNode *)nodes[1]); | |
838 | } | |
839 | } | |
840 | else { | |
841 | vResult->setLink(compactOneNode(node, endsWord, nodes, status)); | |
842 | } | |
843 | result = vResult; | |
844 | } | |
845 | } | |
846 | return result; | |
847 | } | |
848 | ||
849 | // Walk the set of peers at the same level, to build a horizontal node. | |
850 | // Uses recursion. | |
851 | ||
852 | static void walkHorizontal(const TernaryNode *node, | |
853 | BuildCompactTrieHorizontalNode *building, | |
854 | UStack &nodes, | |
855 | UErrorCode &status) { | |
856 | while (U_SUCCESS(status) && node != NULL) { | |
857 | if (node->low != NULL) { | |
858 | walkHorizontal(node->low, building, nodes, status); | |
859 | } | |
860 | BuildCompactTrieNode *link = NULL; | |
861 | if (node->equal != NULL) { | |
862 | link = compactOneNode(node->equal, (node->flags & kEndsWord) != 0, nodes, status); | |
863 | } | |
864 | else if (node->flags & kEndsWord) { | |
865 | link = (BuildCompactTrieNode *)nodes[1]; | |
866 | } | |
867 | if (U_SUCCESS(status) && link != NULL) { | |
868 | building->addNode(node->ch, link, status); | |
869 | } | |
870 | // Tail recurse manually instead of leaving it to the compiler. | |
871 | //if (node->high != NULL) { | |
872 | // walkHorizontal(node->high, building, nodes, status); | |
873 | //} | |
874 | node = node->high; | |
875 | } | |
876 | } | |
877 | ||
878 | U_NAMESPACE_END | |
879 | U_CDECL_BEGIN | |
880 | static int32_t U_CALLCONV | |
881 | _sortBuildNodes(const void * /*context*/, const void *voidl, const void *voidr) { | |
882 | BuildCompactTrieNode *left = *(BuildCompactTrieNode **)voidl; | |
883 | BuildCompactTrieNode *right = *(BuildCompactTrieNode **)voidr; | |
884 | // Check for comparing a node to itself, to avoid spurious duplicates | |
885 | if (left == right) { | |
886 | return 0; | |
887 | } | |
888 | // Most significant is type of node. Can never coalesce. | |
889 | if (left->fVertical != right->fVertical) { | |
890 | return left->fVertical - right->fVertical; | |
891 | } | |
892 | // Next, the "parent ends word" flag. If that differs, we cannot coalesce. | |
893 | if (left->fParentEndsWord != right->fParentEndsWord) { | |
894 | return left->fParentEndsWord - right->fParentEndsWord; | |
895 | } | |
896 | // Next, the string. If that differs, we can never coalesce. | |
897 | int32_t result = left->fChars.compare(right->fChars); | |
898 | if (result != 0) { | |
899 | return result; | |
900 | } | |
901 | // We know they're both the same node type, so branch for the two cases. | |
902 | if (left->fVertical) { | |
903 | result = ((BuildCompactTrieVerticalNode *)left)->fEqual->fNodeID | |
904 | - ((BuildCompactTrieVerticalNode *)right)->fEqual->fNodeID; | |
905 | } | |
906 | else { | |
907 | // We need to compare the links vectors. They should be the | |
908 | // same size because the strings were equal. | |
909 | // We compare the node IDs instead of the pointers, to handle | |
910 | // coalesced nodes. | |
911 | BuildCompactTrieHorizontalNode *hleft, *hright; | |
912 | hleft = (BuildCompactTrieHorizontalNode *)left; | |
913 | hright = (BuildCompactTrieHorizontalNode *)right; | |
914 | int32_t count = hleft->fLinks.size(); | |
915 | for (int32_t i = 0; i < count && result == 0; ++i) { | |
916 | result = ((BuildCompactTrieNode *)(hleft->fLinks[i]))->fNodeID - | |
917 | ((BuildCompactTrieNode *)(hright->fLinks[i]))->fNodeID; | |
918 | } | |
919 | } | |
920 | // If they are equal to each other, mark them (speeds coalescing) | |
921 | if (result == 0) { | |
922 | left->fHasDuplicate = TRUE; | |
923 | right->fHasDuplicate = TRUE; | |
924 | } | |
925 | return result; | |
926 | } | |
927 | U_CDECL_END | |
928 | U_NAMESPACE_BEGIN | |
929 | ||
930 | static void coalesceDuplicates(UStack &nodes, UErrorCode &status) { | |
931 | // We sort the array of nodes to place duplicates next to each other | |
932 | if (U_FAILURE(status)) { | |
933 | return; | |
934 | } | |
935 | int32_t size = nodes.size(); | |
936 | void **array = (void **)uprv_malloc(sizeof(void *)*size); | |
937 | if (array == NULL) { | |
938 | status = U_MEMORY_ALLOCATION_ERROR; | |
939 | return; | |
940 | } | |
941 | (void) nodes.toArray(array); | |
942 | ||
943 | // Now repeatedly identify duplicates until there are no more | |
944 | int32_t dupes = 0; | |
945 | long passCount = 0; | |
946 | #ifdef DEBUG_TRIE_DICT | |
947 | long totalDupes = 0; | |
948 | #endif | |
949 | do { | |
950 | BuildCompactTrieNode *node; | |
951 | BuildCompactTrieNode *first = NULL; | |
952 | BuildCompactTrieNode **p; | |
953 | BuildCompactTrieNode **pFirst = NULL; | |
954 | int32_t counter = size - 2; | |
955 | // Sort the array, skipping nodes 0 and 1. Use quicksort for the first | |
956 | // pass for speed. For the second and subsequent passes, we use stable | |
957 | // (insertion) sort for two reasons: | |
958 | // 1. The array is already mostly ordered, so we get better performance. | |
959 | // 2. The way we find one and only one instance of a set of duplicates is to | |
960 | // check that the node ID equals the array index. If we used an unstable | |
961 | // sort for the second or later passes, it's possible that none of the | |
962 | // duplicates would wind up with a node ID equal to its array index. | |
963 | // The sort stability guarantees that, because as we coalesce more and | |
964 | // more groups, the first element of the resultant group will be one of | |
965 | // the first elements of the groups being coalesced. | |
966 | // To use quicksort for the second and subsequent passes, we would have to | |
967 | // find the minimum of the node numbers in a group, and set all the nodes | |
968 | // in the group to that node number. | |
969 | uprv_sortArray(array+2, counter, sizeof(void *), _sortBuildNodes, NULL, (passCount > 0), &status); | |
970 | dupes = 0; | |
971 | for (p = (BuildCompactTrieNode **)array + 2; counter > 0; --counter, ++p) { | |
972 | node = *p; | |
973 | if (node->fHasDuplicate) { | |
974 | if (first == NULL) { | |
975 | first = node; | |
976 | pFirst = p; | |
977 | } | |
978 | else if (_sortBuildNodes(NULL, pFirst, p) != 0) { | |
979 | // Starting a new run of dupes | |
980 | first = node; | |
981 | pFirst = p; | |
982 | } | |
983 | else if (node->fNodeID != first->fNodeID) { | |
984 | // Slave one to the other, note duplicate | |
985 | node->fNodeID = first->fNodeID; | |
986 | dupes += 1; | |
987 | } | |
988 | } | |
989 | else { | |
990 | // This node has no dupes | |
991 | first = NULL; | |
992 | pFirst = NULL; | |
993 | } | |
994 | } | |
995 | passCount += 1; | |
996 | #ifdef DEBUG_TRIE_DICT | |
997 | totalDupes += dupes; | |
998 | fprintf(stderr, "Trie node dupe removal, pass %d: %d nodes tagged\n", passCount, dupes); | |
999 | #endif | |
1000 | } | |
1001 | while (dupes > 0); | |
1002 | #ifdef DEBUG_TRIE_DICT | |
1003 | fprintf(stderr, "Trie node dupe removal complete: %d tagged in %d passes\n", totalDupes, passCount); | |
1004 | #endif | |
1005 | ||
1006 | // We no longer need the temporary array, as the nodes have all been marked appropriately. | |
1007 | uprv_free(array); | |
1008 | } | |
1009 | ||
1010 | U_NAMESPACE_END | |
1011 | U_CDECL_BEGIN | |
1012 | static void U_CALLCONV _deleteBuildNode(void *obj) { | |
1013 | delete (BuildCompactTrieNode *) obj; | |
1014 | } | |
1015 | U_CDECL_END | |
1016 | U_NAMESPACE_BEGIN | |
1017 | ||
1018 | CompactTrieHeader * | |
1019 | CompactTrieDictionary::compactMutableTrieDictionary( const MutableTrieDictionary &dict, | |
1020 | UErrorCode &status ) { | |
1021 | if (U_FAILURE(status)) { | |
1022 | return NULL; | |
1023 | } | |
1024 | #ifdef DEBUG_TRIE_DICT | |
1025 | struct tms timing; | |
1026 | struct tms previous; | |
1027 | (void) ::times(&previous); | |
1028 | #endif | |
1029 | UStack nodes(_deleteBuildNode, NULL, status); // Index of nodes | |
1030 | ||
1031 | // Add node 0, used as the NULL pointer/sentinel. | |
1032 | nodes.addElement((int32_t)0, status); | |
1033 | ||
1034 | // Start by creating the special empty node we use to indicate that the parent | |
1035 | // terminates a word. This must be node 1, because the builder assumes | |
1036 | // that. | |
1037 | if (U_FAILURE(status)) { | |
1038 | return NULL; | |
1039 | } | |
1040 | BuildCompactTrieNode *terminal = new BuildCompactTrieNode(TRUE, FALSE, nodes, status); | |
1041 | if (terminal == NULL) { | |
1042 | status = U_MEMORY_ALLOCATION_ERROR; | |
1043 | } | |
1044 | ||
1045 | // This call does all the work of building the new trie structure. The root | |
1046 | // will be node 2. | |
1047 | BuildCompactTrieNode *root = compactOneNode(dict.fTrie, FALSE, nodes, status); | |
1048 | #ifdef DEBUG_TRIE_DICT | |
1049 | (void) ::times(&timing); | |
1050 | fprintf(stderr, "Compact trie built, %d nodes, time user %f system %f\n", | |
1051 | nodes.size(), (double)(timing.tms_utime-previous.tms_utime)/CLK_TCK, | |
1052 | (double)(timing.tms_stime-previous.tms_stime)/CLK_TCK); | |
1053 | previous = timing; | |
1054 | #endif | |
1055 | ||
1056 | // Now coalesce all duplicate nodes. | |
1057 | coalesceDuplicates(nodes, status); | |
1058 | #ifdef DEBUG_TRIE_DICT | |
1059 | (void) ::times(&timing); | |
1060 | fprintf(stderr, "Duplicates coalesced, time user %f system %f\n", | |
1061 | (double)(timing.tms_utime-previous.tms_utime)/CLK_TCK, | |
1062 | (double)(timing.tms_stime-previous.tms_stime)/CLK_TCK); | |
1063 | previous = timing; | |
1064 | #endif | |
1065 | ||
1066 | // Next, build the output trie. | |
1067 | // First we compute all the sizes and build the node ID translation table. | |
1068 | uint32_t totalSize = offsetof(CompactTrieHeader,offsets); | |
1069 | int32_t count = nodes.size(); | |
1070 | int32_t nodeCount = 1; // The sentinel node we already have | |
1071 | BuildCompactTrieNode *node; | |
1072 | UVector32 translate(count, status); // Should be no growth needed after this | |
1073 | translate.push(0, status); // The sentinel node | |
1074 | ||
1075 | if (U_FAILURE(status)) { | |
1076 | return NULL; | |
1077 | } | |
1078 | ||
1079 | for (int32_t i = 1; i < count; ++i) { | |
1080 | node = (BuildCompactTrieNode *)nodes[i]; | |
1081 | if (node->fNodeID == i) { | |
1082 | // Only one node out of each duplicate set is used | |
1083 | if (i >= translate.size()) { | |
1084 | // Logically extend the mapping table | |
1085 | translate.setSize(i+1); | |
1086 | } | |
1087 | translate.setElementAt(nodeCount++, i); | |
1088 | totalSize += node->size(); | |
1089 | } | |
1090 | } | |
1091 | ||
1092 | // Check for overflowing 16 bits worth of nodes. | |
1093 | if (nodeCount > 0x10000) { | |
1094 | status = U_ILLEGAL_ARGUMENT_ERROR; | |
1095 | return NULL; | |
1096 | } | |
1097 | ||
1098 | // Add enough room for the offsets. | |
1099 | totalSize += nodeCount*sizeof(uint32_t); | |
1100 | #ifdef DEBUG_TRIE_DICT | |
1101 | (void) ::times(&timing); | |
1102 | fprintf(stderr, "Sizes/mapping done, time user %f system %f\n", | |
1103 | (double)(timing.tms_utime-previous.tms_utime)/CLK_TCK, | |
1104 | (double)(timing.tms_stime-previous.tms_stime)/CLK_TCK); | |
1105 | previous = timing; | |
1106 | fprintf(stderr, "%d nodes, %d unique, %d bytes\n", nodes.size(), nodeCount, totalSize); | |
1107 | #endif | |
1108 | uint8_t *bytes = (uint8_t *)uprv_malloc(totalSize); | |
1109 | if (bytes == NULL) { | |
1110 | status = U_MEMORY_ALLOCATION_ERROR; | |
1111 | return NULL; | |
1112 | } | |
1113 | ||
1114 | CompactTrieHeader *header = (CompactTrieHeader *)bytes; | |
1115 | header->size = totalSize; | |
1116 | header->nodeCount = nodeCount; | |
1117 | header->offsets[0] = 0; // Sentinel | |
1118 | header->root = translate.elementAti(root->fNodeID); | |
1119 | #ifdef DEBUG_TRIE_DICT | |
1120 | if (header->root == 0) { | |
1121 | fprintf(stderr, "ERROR: root node %d translate to physical zero\n", root->fNodeID); | |
1122 | } | |
1123 | #endif | |
1124 | uint32_t offset = offsetof(CompactTrieHeader,offsets)+(nodeCount*sizeof(uint32_t)); | |
1125 | nodeCount = 1; | |
1126 | // Now write the data | |
1127 | for (int32_t i = 1; i < count; ++i) { | |
1128 | node = (BuildCompactTrieNode *)nodes[i]; | |
1129 | if (node->fNodeID == i) { | |
1130 | header->offsets[nodeCount++] = offset; | |
1131 | node->write(bytes, offset, translate); | |
1132 | } | |
1133 | } | |
1134 | #ifdef DEBUG_TRIE_DICT | |
1135 | (void) ::times(&timing); | |
1136 | fprintf(stderr, "Trie built, time user %f system %f\n", | |
1137 | (double)(timing.tms_utime-previous.tms_utime)/CLK_TCK, | |
1138 | (double)(timing.tms_stime-previous.tms_stime)/CLK_TCK); | |
1139 | previous = timing; | |
1140 | fprintf(stderr, "Final offset is %d\n", offset); | |
1141 | ||
1142 | // Collect statistics on node types and sizes | |
1143 | int hCount = 0; | |
1144 | int vCount = 0; | |
1145 | size_t hSize = 0; | |
1146 | size_t vSize = 0; | |
1147 | size_t hItemCount = 0; | |
1148 | size_t vItemCount = 0; | |
1149 | uint32_t previousOff = offset; | |
1150 | for (uint16_t i = nodeCount-1; i >= 2; --i) { | |
1151 | const CompactTrieNode *node = getCompactNode(header, i); | |
1152 | if (node->flagscount & kVerticalNode) { | |
1153 | vCount += 1; | |
1154 | vItemCount += (node->flagscount & kCountMask); | |
1155 | vSize += previousOff-header->offsets[i]; | |
1156 | } | |
1157 | else { | |
1158 | hCount += 1; | |
1159 | hItemCount += (node->flagscount & kCountMask); | |
1160 | hSize += previousOff-header->offsets[i]; | |
1161 | } | |
1162 | previousOff = header->offsets[i]; | |
1163 | } | |
1164 | fprintf(stderr, "Horizontal nodes: %d total, average %f bytes with %f items\n", hCount, | |
1165 | (double)hSize/hCount, (double)hItemCount/hCount); | |
1166 | fprintf(stderr, "Vertical nodes: %d total, average %f bytes with %f items\n", vCount, | |
1167 | (double)vSize/vCount, (double)vItemCount/vCount); | |
1168 | #endif | |
1169 | ||
1170 | if (U_FAILURE(status)) { | |
1171 | uprv_free(bytes); | |
1172 | header = NULL; | |
1173 | } | |
1174 | else { | |
1175 | header->magic = COMPACT_TRIE_MAGIC_1; | |
1176 | } | |
1177 | return header; | |
1178 | } | |
1179 | ||
1180 | // Forward declaration | |
1181 | static TernaryNode * | |
1182 | unpackOneNode( const CompactTrieHeader *header, const CompactTrieNode *node, UErrorCode &status ); | |
1183 | ||
1184 | ||
1185 | // Convert a horizontal node (or subarray thereof) into a ternary subtrie | |
1186 | static TernaryNode * | |
1187 | unpackHorizontalArray( const CompactTrieHeader *header, const CompactTrieHorizontalEntry *array, | |
1188 | int low, int high, UErrorCode &status ) { | |
1189 | if (U_FAILURE(status) || low > high) { | |
1190 | return NULL; | |
1191 | } | |
1192 | int middle = (low+high)/2; | |
1193 | TernaryNode *result = new TernaryNode(array[middle].ch); | |
1194 | if (result == NULL) { | |
1195 | status = U_MEMORY_ALLOCATION_ERROR; | |
1196 | return NULL; | |
1197 | } | |
1198 | const CompactTrieNode *equal = getCompactNode(header, array[middle].equal); | |
1199 | if (equal->flagscount & kParentEndsWord) { | |
1200 | result->flags |= kEndsWord; | |
1201 | } | |
1202 | result->low = unpackHorizontalArray(header, array, low, middle-1, status); | |
1203 | result->high = unpackHorizontalArray(header, array, middle+1, high, status); | |
1204 | result->equal = unpackOneNode(header, equal, status); | |
1205 | return result; | |
1206 | } | |
1207 | ||
1208 | // Convert one compact trie node into a ternary subtrie | |
1209 | static TernaryNode * | |
1210 | unpackOneNode( const CompactTrieHeader *header, const CompactTrieNode *node, UErrorCode &status ) { | |
1211 | int nodeCount = (node->flagscount & kCountMask); | |
1212 | if (nodeCount == 0 || U_FAILURE(status)) { | |
1213 | // Failure, or terminal node | |
1214 | return NULL; | |
1215 | } | |
1216 | if (node->flagscount & kVerticalNode) { | |
1217 | const CompactTrieVerticalNode *vnode = (const CompactTrieVerticalNode *)node; | |
1218 | TernaryNode *head = NULL; | |
1219 | TernaryNode *previous = NULL; | |
1220 | TernaryNode *latest = NULL; | |
1221 | for (int i = 0; i < nodeCount; ++i) { | |
1222 | latest = new TernaryNode(vnode->chars[i]); | |
1223 | if (latest == NULL) { | |
1224 | status = U_MEMORY_ALLOCATION_ERROR; | |
1225 | break; | |
1226 | } | |
1227 | if (head == NULL) { | |
1228 | head = latest; | |
1229 | } | |
1230 | if (previous != NULL) { | |
1231 | previous->equal = latest; | |
1232 | } | |
1233 | previous = latest; | |
1234 | } | |
1235 | if (latest != NULL) { | |
1236 | const CompactTrieNode *equal = getCompactNode(header, vnode->equal); | |
1237 | if (equal->flagscount & kParentEndsWord) { | |
1238 | latest->flags |= kEndsWord; | |
1239 | } | |
1240 | latest->equal = unpackOneNode(header, equal, status); | |
1241 | } | |
1242 | return head; | |
1243 | } | |
1244 | else { | |
1245 | // Horizontal node | |
1246 | const CompactTrieHorizontalNode *hnode = (const CompactTrieHorizontalNode *)node; | |
1247 | return unpackHorizontalArray(header, &hnode->entries[0], 0, nodeCount-1, status); | |
1248 | } | |
1249 | } | |
1250 | ||
1251 | MutableTrieDictionary * | |
1252 | CompactTrieDictionary::cloneMutable( UErrorCode &status ) const { | |
1253 | MutableTrieDictionary *result = new MutableTrieDictionary( status ); | |
1254 | if (result == NULL) { | |
1255 | status = U_MEMORY_ALLOCATION_ERROR; | |
1256 | return NULL; | |
1257 | } | |
1258 | TernaryNode *root = unpackOneNode(fData, getCompactNode(fData, fData->root), status); | |
1259 | if (U_FAILURE(status)) { | |
1260 | delete root; // Clean up | |
1261 | delete result; | |
1262 | return NULL; | |
1263 | } | |
1264 | result->fTrie = root; | |
1265 | return result; | |
1266 | } | |
1267 | ||
1268 | U_NAMESPACE_END | |
1269 | ||
1270 | U_CAPI int32_t U_EXPORT2 | |
1271 | triedict_swap(const UDataSwapper *ds, const void *inData, int32_t length, void *outData, | |
1272 | UErrorCode *status) { | |
1273 | ||
1274 | if (status == NULL || U_FAILURE(*status)) { | |
1275 | return 0; | |
1276 | } | |
1277 | if(ds==NULL || inData==NULL || length<-1 || (length>0 && outData==NULL)) { | |
1278 | *status=U_ILLEGAL_ARGUMENT_ERROR; | |
1279 | return 0; | |
1280 | } | |
1281 | ||
1282 | // | |
1283 | // Check that the data header is for for dictionary data. | |
1284 | // (Header contents are defined in genxxx.cpp) | |
1285 | // | |
1286 | const UDataInfo *pInfo = (const UDataInfo *)((const uint8_t *)inData+4); | |
1287 | if(!( pInfo->dataFormat[0]==0x54 && /* dataFormat="TrDc" */ | |
1288 | pInfo->dataFormat[1]==0x72 && | |
1289 | pInfo->dataFormat[2]==0x44 && | |
1290 | pInfo->dataFormat[3]==0x63 && | |
1291 | pInfo->formatVersion[0]==1 )) { | |
1292 | udata_printError(ds, "triedict_swap(): data format %02x.%02x.%02x.%02x (format version %02x) is not recognized\n", | |
1293 | pInfo->dataFormat[0], pInfo->dataFormat[1], | |
1294 | pInfo->dataFormat[2], pInfo->dataFormat[3], | |
1295 | pInfo->formatVersion[0]); | |
1296 | *status=U_UNSUPPORTED_ERROR; | |
1297 | return 0; | |
1298 | } | |
1299 | ||
1300 | // | |
1301 | // Swap the data header. (This is the generic ICU Data Header, not the | |
1302 | // CompactTrieHeader). This swap also conveniently gets us | |
1303 | // the size of the ICU d.h., which lets us locate the start | |
1304 | // of the RBBI specific data. | |
1305 | // | |
1306 | int32_t headerSize=udata_swapDataHeader(ds, inData, length, outData, status); | |
1307 | ||
1308 | // | |
1309 | // Get the CompactTrieHeader, and check that it appears to be OK. | |
1310 | // | |
1311 | const uint8_t *inBytes =(const uint8_t *)inData+headerSize; | |
1312 | const CompactTrieHeader *header = (const CompactTrieHeader *)inBytes; | |
1313 | if (ds->readUInt32(header->magic) != COMPACT_TRIE_MAGIC_1 | |
1314 | || ds->readUInt32(header->size) < sizeof(CompactTrieHeader)) | |
1315 | { | |
1316 | udata_printError(ds, "triedict_swap(): CompactTrieHeader is invalid.\n"); | |
1317 | *status=U_UNSUPPORTED_ERROR; | |
1318 | return 0; | |
1319 | } | |
1320 | ||
1321 | // | |
1322 | // Prefight operation? Just return the size | |
1323 | // | |
1324 | uint32_t totalSize = ds->readUInt32(header->size); | |
1325 | int32_t sizeWithUData = (int32_t)totalSize + headerSize; | |
1326 | if (length < 0) { | |
1327 | return sizeWithUData; | |
1328 | } | |
1329 | ||
1330 | // | |
1331 | // Check that length passed in is consistent with length from RBBI data header. | |
1332 | // | |
1333 | if (length < sizeWithUData) { | |
1334 | udata_printError(ds, "triedict_swap(): too few bytes (%d after ICU Data header) for trie data.\n", | |
1335 | totalSize); | |
1336 | *status=U_INDEX_OUTOFBOUNDS_ERROR; | |
1337 | return 0; | |
1338 | } | |
1339 | ||
1340 | // | |
1341 | // Swap the Data. Do the data itself first, then the CompactTrieHeader, because | |
1342 | // we need to reference the header to locate the data, and an | |
1343 | // inplace swap of the header leaves it unusable. | |
1344 | // | |
1345 | uint8_t *outBytes = (uint8_t *)outData + headerSize; | |
1346 | CompactTrieHeader *outputHeader = (CompactTrieHeader *)outBytes; | |
1347 | ||
1348 | #if 0 | |
1349 | // | |
1350 | // If not swapping in place, zero out the output buffer before starting. | |
1351 | // | |
1352 | if (inBytes != outBytes) { | |
1353 | uprv_memset(outBytes, 0, totalSize); | |
1354 | } | |
1355 | ||
1356 | // We need to loop through all the nodes in the offset table, and swap each one. | |
1357 | uint16_t nodeCount = ds->readUInt16(header->nodeCount); | |
1358 | // Skip node 0, which should always be 0. | |
1359 | for (int i = 1; i < nodeCount; ++i) { | |
1360 | uint32_t nodeOff = ds->readUInt32(header->offsets[i]); | |
1361 | const CompactTrieNode *inNode = (const CompactTrieNode *)(inBytes + nodeOff); | |
1362 | CompactTrieNode *outNode = (CompactTrieNode *)(outBytes + nodeOff); | |
1363 | uint16_t flagscount = ds->readUInt16(inNode->flagscount); | |
1364 | uint16_t itemCount = flagscount & kCountMask; | |
1365 | ds->writeUInt16(&outNode->flagscount, flagscount); | |
1366 | if (itemCount > 0) { | |
1367 | if (flagscount & kVerticalNode) { | |
1368 | ds->swapArray16(ds, inBytes+nodeOff+offsetof(CompactTrieVerticalNode,chars), | |
1369 | itemCount*sizeof(uint16_t), | |
1370 | outBytes+nodeOff+offsetof(CompactTrieVerticalNode,chars), status); | |
1371 | uint16_t equal = ds->readUInt16(inBytes+nodeOff+offsetof(CompactTrieVerticalNode,equal); | |
1372 | ds->writeUInt16(outBytes+nodeOff+offsetof(CompactTrieVerticalNode,equal)); | |
1373 | } | |
1374 | else { | |
1375 | const CompactTrieHorizontalNode *inHNode = (const CompactTrieHorizontalNode *)inNode; | |
1376 | CompactTrieHorizontalNode *outHNode = (CompactTrieHorizontalNode *)outNode; | |
1377 | for (int j = 0; j < itemCount; ++j) { | |
1378 | uint16_t word = ds->readUInt16(inHNode->entries[j].ch); | |
1379 | ds->writeUInt16(&outHNode->entries[j].ch, word); | |
1380 | word = ds->readUInt16(inHNode->entries[j].equal); | |
1381 | ds->writeUInt16(&outHNode->entries[j].equal, word); | |
1382 | } | |
1383 | } | |
1384 | } | |
1385 | } | |
1386 | #endif | |
1387 | ||
1388 | // All the data in all the nodes consist of 16 bit items. Swap them all at once. | |
1389 | uint16_t nodeCount = ds->readUInt16(header->nodeCount); | |
1390 | uint32_t nodesOff = offsetof(CompactTrieHeader,offsets)+((uint32_t)nodeCount*sizeof(uint32_t)); | |
1391 | ds->swapArray16(ds, inBytes+nodesOff, totalSize-nodesOff, outBytes+nodesOff, status); | |
1392 | ||
1393 | // Swap the header | |
1394 | ds->writeUInt32(&outputHeader->size, totalSize); | |
1395 | uint32_t magic = ds->readUInt32(header->magic); | |
1396 | ds->writeUInt32(&outputHeader->magic, magic); | |
1397 | ds->writeUInt16(&outputHeader->nodeCount, nodeCount); | |
1398 | uint16_t root = ds->readUInt16(header->root); | |
1399 | ds->writeUInt16(&outputHeader->root, root); | |
1400 | ds->swapArray32(ds, inBytes+offsetof(CompactTrieHeader,offsets), | |
1401 | sizeof(uint32_t)*(int32_t)nodeCount, | |
1402 | outBytes+offsetof(CompactTrieHeader,offsets), status); | |
1403 | ||
1404 | return sizeWithUData; | |
1405 | } | |
1406 | ||
1407 | #endif /* #if !UCONFIG_NO_BREAK_ITERATION */ |