1// © 2016 and later: Unicode, Inc. and others.
2// License & terms of use: http://www.unicode.org/copyright.html
3/*
4*******************************************************************************
5* Copyright (C) 2010-2012, International Business Machines
6* Corporation and others. All Rights Reserved.
7*******************************************************************************
8* file name: bytestrie.h
9* encoding: UTF-8
10* tab size: 8 (not used)
11* indentation:4
12*
13* created on: 2010sep25
14* created by: Markus W. Scherer
15*/
16
17#ifndef __BYTESTRIE_H__
18#define __BYTESTRIE_H__
19
20/**
21 * \file
22 * \brief C++ API: Trie for mapping byte sequences to integer values.
23 */
24
25#include "unicode/utypes.h"
26
27#if U_SHOW_CPLUSPLUS_API
28
29#include "unicode/stringpiece.h"
30#include "unicode/uobject.h"
31#include "unicode/ustringtrie.h"
32
33U_NAMESPACE_BEGIN
34
35class ByteSink;
36class BytesTrieBuilder;
37class CharString;
38class UVector32;
39
40/**
41 * Light-weight, non-const reader class for a BytesTrie.
42 * Traverses a byte-serialized data structure with minimal state,
43 * for mapping byte sequences to non-negative integer values.
44 *
45 * This class owns the serialized trie data only if it was constructed by
46 * the builder's build() method.
47 * The public constructor and the copy constructor only alias the data (only copy the pointer).
48 * There is no assignment operator.
49 *
50 * This class is not intended for public subclassing.
51 * @stable ICU 4.8
52 */
53class U_COMMON_API BytesTrie : public UMemory {
54public:
55 /**
56 * Constructs a BytesTrie reader instance.
57 *
58 * The trieBytes must contain a copy of a byte sequence from the BytesTrieBuilder,
59 * starting with the first byte of that sequence.
60 * The BytesTrie object will not read more bytes than
61 * the BytesTrieBuilder generated in the corresponding build() call.
62 *
63 * The array is not copied/cloned and must not be modified while
64 * the BytesTrie object is in use.
65 *
66 * @param trieBytes The byte array that contains the serialized trie.
67 * @stable ICU 4.8
68 */
69 BytesTrie(const void *trieBytes)
70 : ownedArray_(NULL), bytes_(static_cast<const uint8_t *>(trieBytes)),
71 pos_(bytes_), remainingMatchLength_(-1) {}
72
73 /**
74 * Destructor.
75 * @stable ICU 4.8
76 */
77 ~BytesTrie();
78
79 /**
80 * Copy constructor, copies the other trie reader object and its state,
81 * but not the byte array which will be shared. (Shallow copy.)
82 * @param other Another BytesTrie object.
83 * @stable ICU 4.8
84 */
85 BytesTrie(const BytesTrie &other)
86 : ownedArray_(NULL), bytes_(other.bytes_),
87 pos_(other.pos_), remainingMatchLength_(other.remainingMatchLength_) {}
88
89 /**
90 * Resets this trie to its initial state.
91 * @return *this
92 * @stable ICU 4.8
93 */
94 BytesTrie &reset() {
95 pos_=bytes_;
96 remainingMatchLength_=-1;
97 return *this;
98 }
99
100#ifndef U_HIDE_DRAFT_API
101 /**
102 * Returns the state of this trie as a 64-bit integer.
103 * The state value is never 0.
104 *
105 * @return opaque state value
106 * @see resetToState64
107 * @draft ICU 65
108 */
109 uint64_t getState64() const {
110 return (static_cast<uint64_t>(remainingMatchLength_ + 2) << kState64RemainingShift) |
111 (uint64_t)(pos_ - bytes_);
112 }
113
114 /**
115 * Resets this trie to the saved state.
116 * Unlike resetToState(State), the 64-bit state value
117 * must be from getState64() from the same trie object or
118 * from one initialized the exact same way.
119 * Because of no validation, this method is faster.
120 *
121 * @param state The opaque trie state value from getState64().
122 * @return *this
123 * @see getState64
124 * @see resetToState
125 * @see reset
126 * @draft ICU 65
127 */
128 BytesTrie &resetToState64(uint64_t state) {
129 remainingMatchLength_ = static_cast<int32_t>(state >> kState64RemainingShift) - 2;
130 pos_ = bytes_ + (state & kState64PosMask);
131 return *this;
132 }
133#endif /* U_HIDE_DRAFT_API */
134
135 /**
136 * BytesTrie state object, for saving a trie's current state
137 * and resetting the trie back to this state later.
138 * @stable ICU 4.8
139 */
140 class State : public UMemory {
141 public:
142 /**
143 * Constructs an empty State.
144 * @stable ICU 4.8
145 */
146 State() { bytes=NULL; }
147 private:
148 friend class BytesTrie;
149
150 const uint8_t *bytes;
151 const uint8_t *pos;
152 int32_t remainingMatchLength;
153 };
154
155 /**
156 * Saves the state of this trie.
157 * @param state The State object to hold the trie's state.
158 * @return *this
159 * @see resetToState
160 * @stable ICU 4.8
161 */
162 const BytesTrie &saveState(State &state) const {
163 state.bytes=bytes_;
164 state.pos=pos_;
165 state.remainingMatchLength=remainingMatchLength_;
166 return *this;
167 }
168
169 /**
170 * Resets this trie to the saved state.
171 * If the state object contains no state, or the state of a different trie,
172 * then this trie remains unchanged.
173 * @param state The State object which holds a saved trie state.
174 * @return *this
175 * @see saveState
176 * @see reset
177 * @stable ICU 4.8
178 */
179 BytesTrie &resetToState(const State &state) {
180 if(bytes_==state.bytes && bytes_!=NULL) {
181 pos_=state.pos;
182 remainingMatchLength_=state.remainingMatchLength;
183 }
184 return *this;
185 }
186
187 /**
188 * Determines whether the byte sequence so far matches, whether it has a value,
189 * and whether another input byte can continue a matching byte sequence.
190 * @return The match/value Result.
191 * @stable ICU 4.8
192 */
193 UStringTrieResult current() const;
194
195 /**
196 * Traverses the trie from the initial state for this input byte.
197 * Equivalent to reset().next(inByte).
198 * @param inByte Input byte value. Values -0x100..-1 are treated like 0..0xff.
199 * Values below -0x100 and above 0xff will never match.
200 * @return The match/value Result.
201 * @stable ICU 4.8
202 */
203 inline UStringTrieResult first(int32_t inByte) {
204 remainingMatchLength_=-1;
205 if(inByte<0) {
206 inByte+=0x100;
207 }
208 return nextImpl(bytes_, inByte);
209 }
210
211 /**
212 * Traverses the trie from the current state for this input byte.
213 * @param inByte Input byte value. Values -0x100..-1 are treated like 0..0xff.
214 * Values below -0x100 and above 0xff will never match.
215 * @return The match/value Result.
216 * @stable ICU 4.8
217 */
218 UStringTrieResult next(int32_t inByte);
219
220 /**
221 * Traverses the trie from the current state for this byte sequence.
222 * Equivalent to
223 * \code
224 * Result result=current();
225 * for(each c in s)
226 * if(!USTRINGTRIE_HAS_NEXT(result)) return USTRINGTRIE_NO_MATCH;
227 * result=next(c);
228 * return result;
229 * \endcode
230 * @param s A string or byte sequence. Can be NULL if length is 0.
231 * @param length The length of the byte sequence. Can be -1 if NUL-terminated.
232 * @return The match/value Result.
233 * @stable ICU 4.8
234 */
235 UStringTrieResult next(const char *s, int32_t length);
236
237 /**
238 * Returns a matching byte sequence's value if called immediately after
239 * current()/first()/next() returned USTRINGTRIE_INTERMEDIATE_VALUE or USTRINGTRIE_FINAL_VALUE.
240 * getValue() can be called multiple times.
241 *
242 * Do not call getValue() after USTRINGTRIE_NO_MATCH or USTRINGTRIE_NO_VALUE!
243 * @return The value for the byte sequence so far.
244 * @stable ICU 4.8
245 */
246 inline int32_t getValue() const {
247 const uint8_t *pos=pos_;
248 int32_t leadByte=*pos++;
249 // U_ASSERT(leadByte>=kMinValueLead);
250 return readValue(pos, leadByte>>1);
251 }
252
253 /**
254 * Determines whether all byte sequences reachable from the current state
255 * map to the same value.
256 * @param uniqueValue Receives the unique value, if this function returns TRUE.
257 * (output-only)
258 * @return TRUE if all byte sequences reachable from the current state
259 * map to the same value.
260 * @stable ICU 4.8
261 */
262 inline UBool hasUniqueValue(int32_t &uniqueValue) const {
263 const uint8_t *pos=pos_;
264 // Skip the rest of a pending linear-match node.
265 return pos!=NULL && findUniqueValue(pos+remainingMatchLength_+1, FALSE, uniqueValue);
266 }
267
268 /**
269 * Finds each byte which continues the byte sequence from the current state.
270 * That is, each byte b for which it would be next(b)!=USTRINGTRIE_NO_MATCH now.
271 * @param out Each next byte is appended to this object.
272 * (Only uses the out.Append(s, length) method.)
273 * @return the number of bytes which continue the byte sequence from here
274 * @stable ICU 4.8
275 */
276 int32_t getNextBytes(ByteSink &out) const;
277
278 /**
279 * Iterator for all of the (byte sequence, value) pairs in a BytesTrie.
280 * @stable ICU 4.8
281 */
282 class U_COMMON_API Iterator : public UMemory {
283 public:
284 /**
285 * Iterates from the root of a byte-serialized BytesTrie.
286 * @param trieBytes The trie bytes.
287 * @param maxStringLength If 0, the iterator returns full strings/byte sequences.
288 * Otherwise, the iterator returns strings with this maximum length.
289 * @param errorCode Standard ICU error code. Its input value must
290 * pass the U_SUCCESS() test, or else the function returns
291 * immediately. Check for U_FAILURE() on output or use with
292 * function chaining. (See User Guide for details.)
293 * @stable ICU 4.8
294 */
295 Iterator(const void *trieBytes, int32_t maxStringLength, UErrorCode &errorCode);
296
297 /**
298 * Iterates from the current state of the specified BytesTrie.
299 * @param trie The trie whose state will be copied for iteration.
300 * @param maxStringLength If 0, the iterator returns full strings/byte sequences.
301 * Otherwise, the iterator returns strings with this maximum length.
302 * @param errorCode Standard ICU error code. Its input value must
303 * pass the U_SUCCESS() test, or else the function returns
304 * immediately. Check for U_FAILURE() on output or use with
305 * function chaining. (See User Guide for details.)
306 * @stable ICU 4.8
307 */
308 Iterator(const BytesTrie &trie, int32_t maxStringLength, UErrorCode &errorCode);
309
310 /**
311 * Destructor.
312 * @stable ICU 4.8
313 */
314 ~Iterator();
315
316 /**
317 * Resets this iterator to its initial state.
318 * @return *this
319 * @stable ICU 4.8
320 */
321 Iterator &reset();
322
323 /**
324 * @return TRUE if there are more elements.
325 * @stable ICU 4.8
326 */
327 UBool hasNext() const;
328
329 /**
330 * Finds the next (byte sequence, value) pair if there is one.
331 *
332 * If the byte sequence is truncated to the maximum length and does not
333 * have a real value, then the value is set to -1.
334 * In this case, this "not a real value" is indistinguishable from
335 * a real value of -1.
336 * @param errorCode Standard ICU error code. Its input value must
337 * pass the U_SUCCESS() test, or else the function returns
338 * immediately. Check for U_FAILURE() on output or use with
339 * function chaining. (See User Guide for details.)
340 * @return TRUE if there is another element.
341 * @stable ICU 4.8
342 */
343 UBool next(UErrorCode &errorCode);
344
345 /**
346 * @return The NUL-terminated byte sequence for the last successful next().
347 * @stable ICU 4.8
348 */
349 StringPiece getString() const;
350 /**
351 * @return The value for the last successful next().
352 * @stable ICU 4.8
353 */
354 int32_t getValue() const { return value_; }
355
356 private:
357 UBool truncateAndStop();
358
359 const uint8_t *branchNext(const uint8_t *pos, int32_t length, UErrorCode &errorCode);
360
361 const uint8_t *bytes_;
362 const uint8_t *pos_;
363 const uint8_t *initialPos_;
364 int32_t remainingMatchLength_;
365 int32_t initialRemainingMatchLength_;
366
367 CharString *str_;
368 int32_t maxLength_;
369 int32_t value_;
370
371 // The stack stores pairs of integers for backtracking to another
372 // outbound edge of a branch node.
373 // The first integer is an offset from bytes_.
374 // The second integer has the str_->length() from before the node in bits 15..0,
375 // and the remaining branch length in bits 24..16. (Bits 31..25 are unused.)
376 // (We could store the remaining branch length minus 1 in bits 23..16 and not use bits 31..24,
377 // but the code looks more confusing that way.)
378 UVector32 *stack_;
379 };
380
381private:
382 friend class BytesTrieBuilder;
383
384 /**
385 * Constructs a BytesTrie reader instance.
386 * Unlike the public constructor which just aliases an array,
387 * this constructor adopts the builder's array.
388 * This constructor is only called by the builder.
389 */
390 BytesTrie(void *adoptBytes, const void *trieBytes)
391 : ownedArray_(static_cast<uint8_t *>(adoptBytes)),
392 bytes_(static_cast<const uint8_t *>(trieBytes)),
393 pos_(bytes_), remainingMatchLength_(-1) {}
394
395 // No assignment operator.
396 BytesTrie &operator=(const BytesTrie &other);
397
398 inline void stop() {
399 pos_=NULL;
400 }
401
402 // Reads a compact 32-bit integer.
403 // pos is already after the leadByte, and the lead byte is already shifted right by 1.
404 static int32_t readValue(const uint8_t *pos, int32_t leadByte);
405 static inline const uint8_t *skipValue(const uint8_t *pos, int32_t leadByte) {
406 // U_ASSERT(leadByte>=kMinValueLead);
407 if(leadByte>=(kMinTwoByteValueLead<<1)) {
408 if(leadByte<(kMinThreeByteValueLead<<1)) {
409 ++pos;
410 } else if(leadByte<(kFourByteValueLead<<1)) {
411 pos+=2;
412 } else {
413 pos+=3+((leadByte>>1)&1);
414 }
415 }
416 return pos;
417 }
418 static inline const uint8_t *skipValue(const uint8_t *pos) {
419 int32_t leadByte=*pos++;
420 return skipValue(pos, leadByte);
421 }
422
423 // Reads a jump delta and jumps.
424 static const uint8_t *jumpByDelta(const uint8_t *pos);
425
426 static inline const uint8_t *skipDelta(const uint8_t *pos) {
427 int32_t delta=*pos++;
428 if(delta>=kMinTwoByteDeltaLead) {
429 if(delta<kMinThreeByteDeltaLead) {
430 ++pos;
431 } else if(delta<kFourByteDeltaLead) {
432 pos+=2;
433 } else {
434 pos+=3+(delta&1);
435 }
436 }
437 return pos;
438 }
439
440 static inline UStringTrieResult valueResult(int32_t node) {
441 return (UStringTrieResult)(USTRINGTRIE_INTERMEDIATE_VALUE-(node&kValueIsFinal));
442 }
443
444 // Handles a branch node for both next(byte) and next(string).
445 UStringTrieResult branchNext(const uint8_t *pos, int32_t length, int32_t inByte);
446
447 // Requires remainingLength_<0.
448 UStringTrieResult nextImpl(const uint8_t *pos, int32_t inByte);
449
450 // Helper functions for hasUniqueValue().
451 // Recursively finds a unique value (or whether there is not a unique one)
452 // from a branch.
453 static const uint8_t *findUniqueValueFromBranch(const uint8_t *pos, int32_t length,
454 UBool haveUniqueValue, int32_t &uniqueValue);
455 // Recursively finds a unique value (or whether there is not a unique one)
456 // starting from a position on a node lead byte.
457 static UBool findUniqueValue(const uint8_t *pos, UBool haveUniqueValue, int32_t &uniqueValue);
458
459 // Helper functions for getNextBytes().
460 // getNextBytes() when pos is on a branch node.
461 static void getNextBranchBytes(const uint8_t *pos, int32_t length, ByteSink &out);
462 static void append(ByteSink &out, int c);
463
464 // BytesTrie data structure
465 //
466 // The trie consists of a series of byte-serialized nodes for incremental
467 // string/byte sequence matching. The root node is at the beginning of the trie data.
468 //
469 // Types of nodes are distinguished by their node lead byte ranges.
470 // After each node, except a final-value node, another node follows to
471 // encode match values or continue matching further bytes.
472 //
473 // Node types:
474 // - Value node: Stores a 32-bit integer in a compact, variable-length format.
475 // The value is for the string/byte sequence so far.
476 // One node bit indicates whether the value is final or whether
477 // matching continues with the next node.
478 // - Linear-match node: Matches a number of bytes.
479 // - Branch node: Branches to other nodes according to the current input byte.
480 // The node byte is the length of the branch (number of bytes to select from)
481 // minus 1. It is followed by a sub-node:
482 // - If the length is at most kMaxBranchLinearSubNodeLength, then
483 // there are length-1 (key, value) pairs and then one more comparison byte.
484 // If one of the key bytes matches, then the value is either a final value for
485 // the string/byte sequence so far, or a "jump" delta to the next node.
486 // If the last byte matches, then matching continues with the next node.
487 // (Values have the same encoding as value nodes.)
488 // - If the length is greater than kMaxBranchLinearSubNodeLength, then
489 // there is one byte and one "jump" delta.
490 // If the input byte is less than the sub-node byte, then "jump" by delta to
491 // the next sub-node which will have a length of length/2.
492 // (The delta has its own compact encoding.)
493 // Otherwise, skip the "jump" delta to the next sub-node
494 // which will have a length of length-length/2.
495
496 // Node lead byte values.
497
498 // 00..0f: Branch node. If node!=0 then the length is node+1, otherwise
499 // the length is one more than the next byte.
500
501 // For a branch sub-node with at most this many entries, we drop down
502 // to a linear search.
503 static const int32_t kMaxBranchLinearSubNodeLength=5;
504
505 // 10..1f: Linear-match node, match 1..16 bytes and continue reading the next node.
506 static const int32_t kMinLinearMatch=0x10;
507 static const int32_t kMaxLinearMatchLength=0x10;
508
509 // 20..ff: Variable-length value node.
510 // If odd, the value is final. (Otherwise, intermediate value or jump delta.)
511 // Then shift-right by 1 bit.
512 // The remaining lead byte value indicates the number of following bytes (0..4)
513 // and contains the value's top bits.
514 static const int32_t kMinValueLead=kMinLinearMatch+kMaxLinearMatchLength; // 0x20
515 // It is a final value if bit 0 is set.
516 static const int32_t kValueIsFinal=1;
517
518 // Compact value: After testing bit 0, shift right by 1 and then use the following thresholds.
519 static const int32_t kMinOneByteValueLead=kMinValueLead/2; // 0x10
520 static const int32_t kMaxOneByteValue=0x40; // At least 6 bits in the first byte.
521
522 static const int32_t kMinTwoByteValueLead=kMinOneByteValueLead+kMaxOneByteValue+1; // 0x51
523 static const int32_t kMaxTwoByteValue=0x1aff;
524
525 static const int32_t kMinThreeByteValueLead=kMinTwoByteValueLead+(kMaxTwoByteValue>>8)+1; // 0x6c
526 static const int32_t kFourByteValueLead=0x7e;
527
528 // A little more than Unicode code points. (0x11ffff)
529 static const int32_t kMaxThreeByteValue=((kFourByteValueLead-kMinThreeByteValueLead)<<16)-1;
530
531 static const int32_t kFiveByteValueLead=0x7f;
532
533 // Compact delta integers.
534 static const int32_t kMaxOneByteDelta=0xbf;
535 static const int32_t kMinTwoByteDeltaLead=kMaxOneByteDelta+1; // 0xc0
536 static const int32_t kMinThreeByteDeltaLead=0xf0;
537 static const int32_t kFourByteDeltaLead=0xfe;
538 static const int32_t kFiveByteDeltaLead=0xff;
539
540 static const int32_t kMaxTwoByteDelta=((kMinThreeByteDeltaLead-kMinTwoByteDeltaLead)<<8)-1; // 0x2fff
541 static const int32_t kMaxThreeByteDelta=((kFourByteDeltaLead-kMinThreeByteDeltaLead)<<16)-1; // 0xdffff
542
543 // For getState64():
544 // The remainingMatchLength_ is -1..14=(kMaxLinearMatchLength=0x10)-2
545 // so we need at least 5 bits for that.
546 // We add 2 to store it as a positive value 1..16=kMaxLinearMatchLength.
547 static constexpr int32_t kState64RemainingShift = 59;
548 static constexpr uint64_t kState64PosMask = (UINT64_C(1) << kState64RemainingShift) - 1;
549
550 uint8_t *ownedArray_;
551
552 // Fixed value referencing the BytesTrie bytes.
553 const uint8_t *bytes_;
554
555 // Iterator variables.
556
557 // Pointer to next trie byte to read. NULL if no more matches.
558 const uint8_t *pos_;
559 // Remaining length of a linear-match node, minus 1. Negative if not in such a node.
560 int32_t remainingMatchLength_;
561};
562
563U_NAMESPACE_END
564
565#endif /* U_SHOW_CPLUSPLUS_API */
566
567#endif // __BYTESTRIE_H__
568