1 | // © 2016 and later: Unicode, Inc. and others. |
2 | // License & terms of use: http://www.unicode.org/copyright.html |
3 | /* |
4 | ******************************************************************************* |
5 | * Copyright (C) 2010-2014, International Business Machines |
6 | * Corporation and others. All Rights Reserved. |
7 | ******************************************************************************* |
8 | * collationiterator.h |
9 | * |
10 | * created on: 2010oct27 |
11 | * created by: Markus W. Scherer |
12 | */ |
13 | |
14 | #ifndef __COLLATIONITERATOR_H__ |
15 | #define __COLLATIONITERATOR_H__ |
16 | |
17 | #include "unicode/utypes.h" |
18 | |
19 | #if !UCONFIG_NO_COLLATION |
20 | |
21 | #include "cmemory.h" |
22 | #include "collation.h" |
23 | #include "collationdata.h" |
24 | |
25 | U_NAMESPACE_BEGIN |
26 | |
27 | class SkippedState; |
28 | class UCharsTrie; |
29 | class UVector32; |
30 | |
31 | /* Large enough for CEs of most short strings. */ |
32 | #define CEBUFFER_INITIAL_CAPACITY 40 |
33 | |
34 | // Export an explicit template instantiation of the MaybeStackArray that |
35 | // is used as a data member of CEBuffer. |
36 | // |
37 | // When building DLLs for Windows this is required even though |
38 | // no direct access to the MaybeStackArray leaks out of the i18n library. |
39 | // |
40 | // See digitlst.h, pluralaffix.h, datefmt.h, and others for similar examples. |
41 | // |
42 | #if U_PF_WINDOWS <= U_PLATFORM && U_PLATFORM <= U_PF_CYGWIN |
43 | template class U_I18N_API MaybeStackArray<int64_t, CEBUFFER_INITIAL_CAPACITY>; |
44 | #endif |
45 | |
46 | /** |
47 | * Collation element iterator and abstract character iterator. |
48 | * |
49 | * When a method returns a code point value, it must be in 0..10FFFF, |
50 | * except it can be negative as a sentinel value. |
51 | */ |
52 | class U_I18N_API CollationIterator : public UObject { |
53 | private: |
54 | class U_I18N_API CEBuffer { |
55 | private: |
56 | /** Large enough for CEs of most short strings. */ |
57 | static const int32_t INITIAL_CAPACITY = CEBUFFER_INITIAL_CAPACITY; |
58 | public: |
59 | CEBuffer() : length(0) {} |
60 | ~CEBuffer(); |
61 | |
62 | inline void append(int64_t ce, UErrorCode &errorCode) { |
63 | if(length < INITIAL_CAPACITY || ensureAppendCapacity(1, errorCode)) { |
64 | buffer[length++] = ce; |
65 | } |
66 | } |
67 | |
68 | inline void appendUnsafe(int64_t ce) { |
69 | buffer[length++] = ce; |
70 | } |
71 | |
72 | UBool ensureAppendCapacity(int32_t appCap, UErrorCode &errorCode); |
73 | |
74 | inline UBool incLength(UErrorCode &errorCode) { |
75 | // Use INITIAL_CAPACITY for a very simple fastpath. |
76 | // (Rather than buffer.getCapacity().) |
77 | if(length < INITIAL_CAPACITY || ensureAppendCapacity(1, errorCode)) { |
78 | ++length; |
79 | return TRUE; |
80 | } else { |
81 | return FALSE; |
82 | } |
83 | } |
84 | |
85 | inline int64_t set(int32_t i, int64_t ce) { |
86 | return buffer[i] = ce; |
87 | } |
88 | inline int64_t get(int32_t i) const { return buffer[i]; } |
89 | |
90 | const int64_t *getCEs() const { return buffer.getAlias(); } |
91 | |
92 | int32_t length; |
93 | |
94 | private: |
95 | CEBuffer(const CEBuffer &); |
96 | void operator=(const CEBuffer &); |
97 | |
98 | MaybeStackArray<int64_t, INITIAL_CAPACITY> buffer; |
99 | }; |
100 | |
101 | public: |
102 | CollationIterator(const CollationData *d, UBool numeric) |
103 | : trie(d->trie), |
104 | data(d), |
105 | cesIndex(0), |
106 | skipped(NULL), |
107 | numCpFwd(-1), |
108 | isNumeric(numeric) {} |
109 | |
110 | virtual ~CollationIterator(); |
111 | |
112 | virtual UBool operator==(const CollationIterator &other) const; |
113 | inline UBool operator!=(const CollationIterator &other) const { |
114 | return !operator==(other); |
115 | } |
116 | |
117 | /** |
118 | * Resets the iterator state and sets the position to the specified offset. |
119 | * Subclasses must implement, and must call the parent class method, |
120 | * or CollationIterator::reset(). |
121 | */ |
122 | virtual void resetToOffset(int32_t newOffset) = 0; |
123 | |
124 | virtual int32_t getOffset() const = 0; |
125 | |
126 | /** |
127 | * Returns the next collation element. |
128 | */ |
129 | inline int64_t nextCE(UErrorCode &errorCode) { |
130 | if(cesIndex < ceBuffer.length) { |
131 | // Return the next buffered CE. |
132 | return ceBuffer.get(cesIndex++); |
133 | } |
134 | // assert cesIndex == ceBuffer.length; |
135 | if(!ceBuffer.incLength(errorCode)) { |
136 | return Collation::NO_CE; |
137 | } |
138 | UChar32 c; |
139 | uint32_t ce32 = handleNextCE32(c, errorCode); |
140 | uint32_t t = ce32 & 0xff; |
141 | if(t < Collation::SPECIAL_CE32_LOW_BYTE) { // Forced-inline of isSpecialCE32(ce32). |
142 | // Normal CE from the main data. |
143 | // Forced-inline of ceFromSimpleCE32(ce32). |
144 | return ceBuffer.set(cesIndex++, |
145 | ((int64_t)(ce32 & 0xffff0000) << 32) | ((ce32 & 0xff00) << 16) | (t << 8)); |
146 | } |
147 | const CollationData *d; |
148 | // The compiler should be able to optimize the previous and the following |
149 | // comparisons of t with the same constant. |
150 | if(t == Collation::SPECIAL_CE32_LOW_BYTE) { |
151 | if(c < 0) { |
152 | return ceBuffer.set(cesIndex++, Collation::NO_CE); |
153 | } |
154 | d = data->base; |
155 | ce32 = d->getCE32(c); |
156 | t = ce32 & 0xff; |
157 | if(t < Collation::SPECIAL_CE32_LOW_BYTE) { |
158 | // Normal CE from the base data. |
159 | return ceBuffer.set(cesIndex++, |
160 | ((int64_t)(ce32 & 0xffff0000) << 32) | ((ce32 & 0xff00) << 16) | (t << 8)); |
161 | } |
162 | } else { |
163 | d = data; |
164 | } |
165 | if(t == Collation::LONG_PRIMARY_CE32_LOW_BYTE) { |
166 | // Forced-inline of ceFromLongPrimaryCE32(ce32). |
167 | return ceBuffer.set(cesIndex++, |
168 | ((int64_t)(ce32 - t) << 32) | Collation::COMMON_SEC_AND_TER_CE); |
169 | } |
170 | return nextCEFromCE32(d, c, ce32, errorCode); |
171 | } |
172 | |
173 | /** |
174 | * Fetches all CEs. |
175 | * @return getCEsLength() |
176 | */ |
177 | int32_t fetchCEs(UErrorCode &errorCode); |
178 | |
179 | /** |
180 | * Overwrites the current CE (the last one returned by nextCE()). |
181 | */ |
182 | void setCurrentCE(int64_t ce) { |
183 | // assert cesIndex > 0; |
184 | ceBuffer.set(cesIndex - 1, ce); |
185 | } |
186 | |
187 | /** |
188 | * Returns the previous collation element. |
189 | */ |
190 | int64_t previousCE(UVector32 &offsets, UErrorCode &errorCode); |
191 | |
192 | inline int32_t getCEsLength() const { |
193 | return ceBuffer.length; |
194 | } |
195 | |
196 | inline int64_t getCE(int32_t i) const { |
197 | return ceBuffer.get(i); |
198 | } |
199 | |
200 | const int64_t *getCEs() const { |
201 | return ceBuffer.getCEs(); |
202 | } |
203 | |
204 | void clearCEs() { |
205 | cesIndex = ceBuffer.length = 0; |
206 | } |
207 | |
208 | void clearCEsIfNoneRemaining() { |
209 | if(cesIndex == ceBuffer.length) { clearCEs(); } |
210 | } |
211 | |
212 | /** |
213 | * Returns the next code point (with post-increment). |
214 | * Public for identical-level comparison and for testing. |
215 | */ |
216 | virtual UChar32 nextCodePoint(UErrorCode &errorCode) = 0; |
217 | |
218 | /** |
219 | * Returns the previous code point (with pre-decrement). |
220 | * Public for identical-level comparison and for testing. |
221 | */ |
222 | virtual UChar32 previousCodePoint(UErrorCode &errorCode) = 0; |
223 | |
224 | protected: |
225 | CollationIterator(const CollationIterator &other); |
226 | |
227 | void reset(); |
228 | |
229 | /** |
230 | * Returns the next code point and its local CE32 value. |
231 | * Returns Collation::FALLBACK_CE32 at the end of the text (c<0) |
232 | * or when c's CE32 value is to be looked up in the base data (fallback). |
233 | * |
234 | * The code point is used for fallbacks, context and implicit weights. |
235 | * It is ignored when the returned CE32 is not special (e.g., FFFD_CE32). |
236 | */ |
237 | virtual uint32_t handleNextCE32(UChar32 &c, UErrorCode &errorCode); |
238 | |
239 | /** |
240 | * Called when handleNextCE32() returns a LEAD_SURROGATE_TAG for a lead surrogate code unit. |
241 | * Returns the trail surrogate in that case and advances past it, |
242 | * if a trail surrogate follows the lead surrogate. |
243 | * Otherwise returns any other code unit and does not advance. |
244 | */ |
245 | virtual UChar handleGetTrailSurrogate(); |
246 | |
247 | /** |
248 | * Called when handleNextCE32() returns with c==0, to see whether it is a NUL terminator. |
249 | * (Not needed in Java.) |
250 | */ |
251 | virtual UBool foundNULTerminator(); |
252 | |
253 | /** |
254 | * @return FALSE if surrogate code points U+D800..U+DFFF |
255 | * map to their own implicit primary weights (for UTF-16), |
256 | * or TRUE if they map to CE(U+FFFD) (for UTF-8) |
257 | */ |
258 | virtual UBool forbidSurrogateCodePoints() const; |
259 | |
260 | virtual void forwardNumCodePoints(int32_t num, UErrorCode &errorCode) = 0; |
261 | |
262 | virtual void backwardNumCodePoints(int32_t num, UErrorCode &errorCode) = 0; |
263 | |
264 | /** |
265 | * Returns the CE32 from the data trie. |
266 | * Normally the same as data->getCE32(), but overridden in the builder. |
267 | * Call this only when the faster data->getCE32() cannot be used. |
268 | */ |
269 | virtual uint32_t getDataCE32(UChar32 c) const; |
270 | |
271 | virtual uint32_t getCE32FromBuilderData(uint32_t ce32, UErrorCode &errorCode); |
272 | |
273 | void appendCEsFromCE32(const CollationData *d, UChar32 c, uint32_t ce32, |
274 | UBool forward, UErrorCode &errorCode); |
275 | |
276 | // Main lookup trie of the data object. |
277 | const UTrie2 *trie; |
278 | const CollationData *data; |
279 | |
280 | private: |
281 | int64_t nextCEFromCE32(const CollationData *d, UChar32 c, uint32_t ce32, |
282 | UErrorCode &errorCode); |
283 | |
284 | uint32_t getCE32FromPrefix(const CollationData *d, uint32_t ce32, |
285 | UErrorCode &errorCode); |
286 | |
287 | UChar32 nextSkippedCodePoint(UErrorCode &errorCode); |
288 | |
289 | void backwardNumSkipped(int32_t n, UErrorCode &errorCode); |
290 | |
291 | uint32_t nextCE32FromContraction( |
292 | const CollationData *d, uint32_t contractionCE32, |
293 | const UChar *p, uint32_t ce32, UChar32 c, |
294 | UErrorCode &errorCode); |
295 | |
296 | uint32_t nextCE32FromDiscontiguousContraction( |
297 | const CollationData *d, UCharsTrie &suffixes, uint32_t ce32, |
298 | int32_t lookAhead, UChar32 c, |
299 | UErrorCode &errorCode); |
300 | |
301 | /** |
302 | * Returns the previous CE when data->isUnsafeBackward(c, isNumeric). |
303 | */ |
304 | int64_t previousCEUnsafe(UChar32 c, UVector32 &offsets, UErrorCode &errorCode); |
305 | |
306 | /** |
307 | * Turns a string of digits (bytes 0..9) |
308 | * into a sequence of CEs that will sort in numeric order. |
309 | * |
310 | * Starts from this ce32's digit value and consumes the following/preceding digits. |
311 | * The digits string must not be empty and must not have leading zeros. |
312 | */ |
313 | void appendNumericCEs(uint32_t ce32, UBool forward, UErrorCode &errorCode); |
314 | |
315 | /** |
316 | * Turns 1..254 digits into a sequence of CEs. |
317 | * Called by appendNumericCEs() for each segment of at most 254 digits. |
318 | */ |
319 | void appendNumericSegmentCEs(const char *digits, int32_t length, UErrorCode &errorCode); |
320 | |
321 | CEBuffer ceBuffer; |
322 | int32_t cesIndex; |
323 | |
324 | SkippedState *skipped; |
325 | |
326 | // Number of code points to read forward, or -1. |
327 | // Used as a forward iteration limit in previousCEUnsafe(). |
328 | int32_t numCpFwd; |
329 | // Numeric collation (CollationSettings::NUMERIC). |
330 | UBool isNumeric; |
331 | }; |
332 | |
333 | U_NAMESPACE_END |
334 | |
335 | #endif // !UCONFIG_NO_COLLATION |
336 | #endif // __COLLATIONITERATOR_H__ |
337 | |