1 | // © 2016 and later: Unicode, Inc. and others. |
2 | // License & terms of use: http://www.unicode.org/copyright.html |
3 | /* |
4 | ****************************************************************************** |
5 | * Copyright (C) 1997-2015, International Business Machines |
6 | * Corporation and others. All Rights Reserved. |
7 | ****************************************************************************** |
8 | * file name: nfrule.cpp |
9 | * encoding: UTF-8 |
10 | * tab size: 8 (not used) |
11 | * indentation:4 |
12 | * |
13 | * Modification history |
14 | * Date Name Comments |
15 | * 10/11/2001 Doug Ported from ICU4J |
16 | */ |
17 | |
18 | #include "nfrule.h" |
19 | |
20 | #if U_HAVE_RBNF |
21 | |
22 | #include "unicode/localpointer.h" |
23 | #include "unicode/rbnf.h" |
24 | #include "unicode/tblcoll.h" |
25 | #include "unicode/plurfmt.h" |
26 | #include "unicode/upluralrules.h" |
27 | #include "unicode/coleitr.h" |
28 | #include "unicode/uchar.h" |
29 | #include "nfrs.h" |
30 | #include "nfrlist.h" |
31 | #include "nfsubs.h" |
32 | #include "patternprops.h" |
33 | #include "putilimp.h" |
34 | |
35 | U_NAMESPACE_BEGIN |
36 | |
37 | NFRule::NFRule(const RuleBasedNumberFormat* _rbnf, const UnicodeString &_ruleText, UErrorCode &status) |
38 | : baseValue((int32_t)0) |
39 | , radix(10) |
40 | , exponent(0) |
41 | , decimalPoint(0) |
42 | , fRuleText(_ruleText) |
43 | , sub1(NULL) |
44 | , sub2(NULL) |
45 | , formatter(_rbnf) |
46 | , rulePatternFormat(NULL) |
47 | { |
48 | if (!fRuleText.isEmpty()) { |
49 | parseRuleDescriptor(fRuleText, status); |
50 | } |
51 | } |
52 | |
53 | NFRule::~NFRule() |
54 | { |
55 | if (sub1 != sub2) { |
56 | delete sub2; |
57 | sub2 = NULL; |
58 | } |
59 | delete sub1; |
60 | sub1 = NULL; |
61 | delete rulePatternFormat; |
62 | rulePatternFormat = NULL; |
63 | } |
64 | |
65 | static const UChar gLeftBracket = 0x005b; |
66 | static const UChar gRightBracket = 0x005d; |
67 | static const UChar gColon = 0x003a; |
68 | static const UChar gZero = 0x0030; |
69 | static const UChar gNine = 0x0039; |
70 | static const UChar gSpace = 0x0020; |
71 | static const UChar gSlash = 0x002f; |
72 | static const UChar gGreaterThan = 0x003e; |
73 | static const UChar gLessThan = 0x003c; |
74 | static const UChar gComma = 0x002c; |
75 | static const UChar gDot = 0x002e; |
76 | static const UChar gTick = 0x0027; |
77 | //static const UChar gMinus = 0x002d; |
78 | static const UChar gSemicolon = 0x003b; |
79 | static const UChar gX = 0x0078; |
80 | |
81 | static const UChar gMinusX[] = {0x2D, 0x78, 0}; /* "-x" */ |
82 | static const UChar gInf[] = {0x49, 0x6E, 0x66, 0}; /* "Inf" */ |
83 | static const UChar gNaN[] = {0x4E, 0x61, 0x4E, 0}; /* "NaN" */ |
84 | |
85 | static const UChar gDollarOpenParenthesis[] = {0x24, 0x28, 0}; /* "$(" */ |
86 | static const UChar gClosedParenthesisDollar[] = {0x29, 0x24, 0}; /* ")$" */ |
87 | |
88 | static const UChar gLessLess[] = {0x3C, 0x3C, 0}; /* "<<" */ |
89 | static const UChar gLessPercent[] = {0x3C, 0x25, 0}; /* "<%" */ |
90 | static const UChar gLessHash[] = {0x3C, 0x23, 0}; /* "<#" */ |
91 | static const UChar gLessZero[] = {0x3C, 0x30, 0}; /* "<0" */ |
92 | static const UChar gGreaterGreater[] = {0x3E, 0x3E, 0}; /* ">>" */ |
93 | static const UChar gGreaterPercent[] = {0x3E, 0x25, 0}; /* ">%" */ |
94 | static const UChar gGreaterHash[] = {0x3E, 0x23, 0}; /* ">#" */ |
95 | static const UChar gGreaterZero[] = {0x3E, 0x30, 0}; /* ">0" */ |
96 | static const UChar gEqualPercent[] = {0x3D, 0x25, 0}; /* "=%" */ |
97 | static const UChar gEqualHash[] = {0x3D, 0x23, 0}; /* "=#" */ |
98 | static const UChar gEqualZero[] = {0x3D, 0x30, 0}; /* "=0" */ |
99 | static const UChar gGreaterGreaterGreater[] = {0x3E, 0x3E, 0x3E, 0}; /* ">>>" */ |
100 | |
101 | static const UChar * const RULE_PREFIXES[] = { |
102 | gLessLess, gLessPercent, gLessHash, gLessZero, |
103 | gGreaterGreater, gGreaterPercent,gGreaterHash, gGreaterZero, |
104 | gEqualPercent, gEqualHash, gEqualZero, NULL |
105 | }; |
106 | |
107 | void |
108 | NFRule::makeRules(UnicodeString& description, |
109 | NFRuleSet *owner, |
110 | const NFRule *predecessor, |
111 | const RuleBasedNumberFormat *rbnf, |
112 | NFRuleList& rules, |
113 | UErrorCode& status) |
114 | { |
115 | // we know we're making at least one rule, so go ahead and |
116 | // new it up and initialize its basevalue and divisor |
117 | // (this also strips the rule descriptor, if any, off the |
118 | // descripton string) |
119 | NFRule* rule1 = new NFRule(rbnf, description, status); |
120 | /* test for NULL */ |
121 | if (rule1 == 0) { |
122 | status = U_MEMORY_ALLOCATION_ERROR; |
123 | return; |
124 | } |
125 | description = rule1->fRuleText; |
126 | |
127 | // check the description to see whether there's text enclosed |
128 | // in brackets |
129 | int32_t brack1 = description.indexOf(gLeftBracket); |
130 | int32_t brack2 = brack1 < 0 ? -1 : description.indexOf(gRightBracket); |
131 | |
132 | // if the description doesn't contain a matched pair of brackets, |
133 | // or if it's of a type that doesn't recognize bracketed text, |
134 | // then leave the description alone, initialize the rule's |
135 | // rule text and substitutions, and return that rule |
136 | if (brack2 < 0 || brack1 > brack2 |
137 | || rule1->getType() == kProperFractionRule |
138 | || rule1->getType() == kNegativeNumberRule |
139 | || rule1->getType() == kInfinityRule |
140 | || rule1->getType() == kNaNRule) |
141 | { |
142 | rule1->extractSubstitutions(owner, description, predecessor, status); |
143 | } |
144 | else { |
145 | // if the description does contain a matched pair of brackets, |
146 | // then it's really shorthand for two rules (with one exception) |
147 | NFRule* rule2 = NULL; |
148 | UnicodeString sbuf; |
149 | |
150 | // we'll actually only split the rule into two rules if its |
151 | // base value is an even multiple of its divisor (or it's one |
152 | // of the special rules) |
153 | if ((rule1->baseValue > 0 |
154 | && (rule1->baseValue % util64_pow(rule1->radix, rule1->exponent)) == 0) |
155 | || rule1->getType() == kImproperFractionRule |
156 | || rule1->getType() == kMasterRule) { |
157 | |
158 | // if it passes that test, new up the second rule. If the |
159 | // rule set both rules will belong to is a fraction rule |
160 | // set, they both have the same base value; otherwise, |
161 | // increment the original rule's base value ("rule1" actually |
162 | // goes SECOND in the rule set's rule list) |
163 | rule2 = new NFRule(rbnf, UnicodeString(), status); |
164 | /* test for NULL */ |
165 | if (rule2 == 0) { |
166 | status = U_MEMORY_ALLOCATION_ERROR; |
167 | return; |
168 | } |
169 | if (rule1->baseValue >= 0) { |
170 | rule2->baseValue = rule1->baseValue; |
171 | if (!owner->isFractionRuleSet()) { |
172 | ++rule1->baseValue; |
173 | } |
174 | } |
175 | |
176 | // if the description began with "x.x" and contains bracketed |
177 | // text, it describes both the improper fraction rule and |
178 | // the proper fraction rule |
179 | else if (rule1->getType() == kImproperFractionRule) { |
180 | rule2->setType(kProperFractionRule); |
181 | } |
182 | |
183 | // if the description began with "x.0" and contains bracketed |
184 | // text, it describes both the master rule and the |
185 | // improper fraction rule |
186 | else if (rule1->getType() == kMasterRule) { |
187 | rule2->baseValue = rule1->baseValue; |
188 | rule1->setType(kImproperFractionRule); |
189 | } |
190 | |
191 | // both rules have the same radix and exponent (i.e., the |
192 | // same divisor) |
193 | rule2->radix = rule1->radix; |
194 | rule2->exponent = rule1->exponent; |
195 | |
196 | // rule2's rule text omits the stuff in brackets: initalize |
197 | // its rule text and substitutions accordingly |
198 | sbuf.append(description, 0, brack1); |
199 | if (brack2 + 1 < description.length()) { |
200 | sbuf.append(description, brack2 + 1, description.length() - brack2 - 1); |
201 | } |
202 | rule2->extractSubstitutions(owner, sbuf, predecessor, status); |
203 | } |
204 | |
205 | // rule1's text includes the text in the brackets but omits |
206 | // the brackets themselves: initialize _its_ rule text and |
207 | // substitutions accordingly |
208 | sbuf.setTo(description, 0, brack1); |
209 | sbuf.append(description, brack1 + 1, brack2 - brack1 - 1); |
210 | if (brack2 + 1 < description.length()) { |
211 | sbuf.append(description, brack2 + 1, description.length() - brack2 - 1); |
212 | } |
213 | rule1->extractSubstitutions(owner, sbuf, predecessor, status); |
214 | |
215 | // if we only have one rule, return it; if we have two, return |
216 | // a two-element array containing them (notice that rule2 goes |
217 | // BEFORE rule1 in the list: in all cases, rule2 OMITS the |
218 | // material in the brackets and rule1 INCLUDES the material |
219 | // in the brackets) |
220 | if (rule2 != NULL) { |
221 | if (rule2->baseValue >= kNoBase) { |
222 | rules.add(rule2); |
223 | } |
224 | else { |
225 | owner->setNonNumericalRule(rule2); |
226 | } |
227 | } |
228 | } |
229 | if (rule1->baseValue >= kNoBase) { |
230 | rules.add(rule1); |
231 | } |
232 | else { |
233 | owner->setNonNumericalRule(rule1); |
234 | } |
235 | } |
236 | |
237 | /** |
238 | * This function parses the rule's rule descriptor (i.e., the base |
239 | * value and/or other tokens that precede the rule's rule text |
240 | * in the description) and sets the rule's base value, radix, and |
241 | * exponent according to the descriptor. (If the description doesn't |
242 | * include a rule descriptor, then this function sets everything to |
243 | * default values and the rule set sets the rule's real base value). |
244 | * @param description The rule's description |
245 | * @return If "description" included a rule descriptor, this is |
246 | * "description" with the descriptor and any trailing whitespace |
247 | * stripped off. Otherwise; it's "descriptor" unchangd. |
248 | */ |
249 | void |
250 | NFRule::parseRuleDescriptor(UnicodeString& description, UErrorCode& status) |
251 | { |
252 | // the description consists of a rule descriptor and a rule body, |
253 | // separated by a colon. The rule descriptor is optional. If |
254 | // it's omitted, just set the base value to 0. |
255 | int32_t p = description.indexOf(gColon); |
256 | if (p != -1) { |
257 | // copy the descriptor out into its own string and strip it, |
258 | // along with any trailing whitespace, out of the original |
259 | // description |
260 | UnicodeString descriptor; |
261 | descriptor.setTo(description, 0, p); |
262 | |
263 | ++p; |
264 | while (p < description.length() && PatternProps::isWhiteSpace(description.charAt(p))) { |
265 | ++p; |
266 | } |
267 | description.removeBetween(0, p); |
268 | |
269 | // check first to see if the rule descriptor matches the token |
270 | // for one of the special rules. If it does, set the base |
271 | // value to the correct identifier value |
272 | int descriptorLength = descriptor.length(); |
273 | UChar firstChar = descriptor.charAt(0); |
274 | UChar lastChar = descriptor.charAt(descriptorLength - 1); |
275 | if (firstChar >= gZero && firstChar <= gNine && lastChar != gX) { |
276 | // if the rule descriptor begins with a digit, it's a descriptor |
277 | // for a normal rule |
278 | // since we don't have Long.parseLong, and this isn't much work anyway, |
279 | // just build up the value as we encounter the digits. |
280 | int64_t val = 0; |
281 | p = 0; |
282 | UChar c = gSpace; |
283 | |
284 | // begin parsing the descriptor: copy digits |
285 | // into "tempValue", skip periods, commas, and spaces, |
286 | // stop on a slash or > sign (or at the end of the string), |
287 | // and throw an exception on any other character |
288 | int64_t ll_10 = 10; |
289 | while (p < descriptorLength) { |
290 | c = descriptor.charAt(p); |
291 | if (c >= gZero && c <= gNine) { |
292 | val = val * ll_10 + (int32_t)(c - gZero); |
293 | } |
294 | else if (c == gSlash || c == gGreaterThan) { |
295 | break; |
296 | } |
297 | else if (PatternProps::isWhiteSpace(c) || c == gComma || c == gDot) { |
298 | } |
299 | else { |
300 | // throw new IllegalArgumentException("Illegal character in rule descriptor"); |
301 | status = U_PARSE_ERROR; |
302 | return; |
303 | } |
304 | ++p; |
305 | } |
306 | |
307 | // we have the base value, so set it |
308 | setBaseValue(val, status); |
309 | |
310 | // if we stopped the previous loop on a slash, we're |
311 | // now parsing the rule's radix. Again, accumulate digits |
312 | // in tempValue, skip punctuation, stop on a > mark, and |
313 | // throw an exception on anything else |
314 | if (c == gSlash) { |
315 | val = 0; |
316 | ++p; |
317 | ll_10 = 10; |
318 | while (p < descriptorLength) { |
319 | c = descriptor.charAt(p); |
320 | if (c >= gZero && c <= gNine) { |
321 | val = val * ll_10 + (int32_t)(c - gZero); |
322 | } |
323 | else if (c == gGreaterThan) { |
324 | break; |
325 | } |
326 | else if (PatternProps::isWhiteSpace(c) || c == gComma || c == gDot) { |
327 | } |
328 | else { |
329 | // throw new IllegalArgumentException("Illegal character is rule descriptor"); |
330 | status = U_PARSE_ERROR; |
331 | return; |
332 | } |
333 | ++p; |
334 | } |
335 | |
336 | // tempValue now contain's the rule's radix. Set it |
337 | // accordingly, and recalculate the rule's exponent |
338 | radix = (int32_t)val; |
339 | if (radix == 0) { |
340 | // throw new IllegalArgumentException("Rule can't have radix of 0"); |
341 | status = U_PARSE_ERROR; |
342 | } |
343 | |
344 | exponent = expectedExponent(); |
345 | } |
346 | |
347 | // if we stopped the previous loop on a > sign, then continue |
348 | // for as long as we still see > signs. For each one, |
349 | // decrement the exponent (unless the exponent is already 0). |
350 | // If we see another character before reaching the end of |
351 | // the descriptor, that's also a syntax error. |
352 | if (c == gGreaterThan) { |
353 | while (p < descriptor.length()) { |
354 | c = descriptor.charAt(p); |
355 | if (c == gGreaterThan && exponent > 0) { |
356 | --exponent; |
357 | } else { |
358 | // throw new IllegalArgumentException("Illegal character in rule descriptor"); |
359 | status = U_PARSE_ERROR; |
360 | return; |
361 | } |
362 | ++p; |
363 | } |
364 | } |
365 | } |
366 | else if (0 == descriptor.compare(gMinusX, 2)) { |
367 | setType(kNegativeNumberRule); |
368 | } |
369 | else if (descriptorLength == 3) { |
370 | if (firstChar == gZero && lastChar == gX) { |
371 | setBaseValue(kProperFractionRule, status); |
372 | decimalPoint = descriptor.charAt(1); |
373 | } |
374 | else if (firstChar == gX && lastChar == gX) { |
375 | setBaseValue(kImproperFractionRule, status); |
376 | decimalPoint = descriptor.charAt(1); |
377 | } |
378 | else if (firstChar == gX && lastChar == gZero) { |
379 | setBaseValue(kMasterRule, status); |
380 | decimalPoint = descriptor.charAt(1); |
381 | } |
382 | else if (descriptor.compare(gNaN, 3) == 0) { |
383 | setBaseValue(kNaNRule, status); |
384 | } |
385 | else if (descriptor.compare(gInf, 3) == 0) { |
386 | setBaseValue(kInfinityRule, status); |
387 | } |
388 | } |
389 | } |
390 | // else use the default base value for now. |
391 | |
392 | // finally, if the rule body begins with an apostrophe, strip it off |
393 | // (this is generally used to put whitespace at the beginning of |
394 | // a rule's rule text) |
395 | if (description.length() > 0 && description.charAt(0) == gTick) { |
396 | description.removeBetween(0, 1); |
397 | } |
398 | |
399 | // return the description with all the stuff we've just waded through |
400 | // stripped off the front. It now contains just the rule body. |
401 | // return description; |
402 | } |
403 | |
404 | /** |
405 | * Searches the rule's rule text for the substitution tokens, |
406 | * creates the substitutions, and removes the substitution tokens |
407 | * from the rule's rule text. |
408 | * @param owner The rule set containing this rule |
409 | * @param predecessor The rule preseding this one in "owners" rule list |
410 | * @param ownersOwner The RuleBasedFormat that owns this rule |
411 | */ |
412 | void |
413 | NFRule::(const NFRuleSet* ruleSet, |
414 | const UnicodeString &ruleText, |
415 | const NFRule* predecessor, |
416 | UErrorCode& status) |
417 | { |
418 | if (U_FAILURE(status)) { |
419 | return; |
420 | } |
421 | fRuleText = ruleText; |
422 | sub1 = extractSubstitution(ruleSet, predecessor, status); |
423 | if (sub1 == NULL) { |
424 | // Small optimization. There is no need to create a redundant NullSubstitution. |
425 | sub2 = NULL; |
426 | } |
427 | else { |
428 | sub2 = extractSubstitution(ruleSet, predecessor, status); |
429 | } |
430 | int32_t pluralRuleStart = fRuleText.indexOf(gDollarOpenParenthesis, -1, 0); |
431 | int32_t pluralRuleEnd = (pluralRuleStart >= 0 ? fRuleText.indexOf(gClosedParenthesisDollar, -1, pluralRuleStart) : -1); |
432 | if (pluralRuleEnd >= 0) { |
433 | int32_t endType = fRuleText.indexOf(gComma, pluralRuleStart); |
434 | if (endType < 0) { |
435 | status = U_PARSE_ERROR; |
436 | return; |
437 | } |
438 | UnicodeString type(fRuleText.tempSubString(pluralRuleStart + 2, endType - pluralRuleStart - 2)); |
439 | UPluralType pluralType; |
440 | if (type.startsWith(UNICODE_STRING_SIMPLE("cardinal" ))) { |
441 | pluralType = UPLURAL_TYPE_CARDINAL; |
442 | } |
443 | else if (type.startsWith(UNICODE_STRING_SIMPLE("ordinal" ))) { |
444 | pluralType = UPLURAL_TYPE_ORDINAL; |
445 | } |
446 | else { |
447 | status = U_ILLEGAL_ARGUMENT_ERROR; |
448 | return; |
449 | } |
450 | rulePatternFormat = formatter->createPluralFormat(pluralType, |
451 | fRuleText.tempSubString(endType + 1, pluralRuleEnd - endType - 1), status); |
452 | } |
453 | } |
454 | |
455 | /** |
456 | * Searches the rule's rule text for the first substitution token, |
457 | * creates a substitution based on it, and removes the token from |
458 | * the rule's rule text. |
459 | * @param owner The rule set containing this rule |
460 | * @param predecessor The rule preceding this one in the rule set's |
461 | * rule list |
462 | * @param ownersOwner The RuleBasedNumberFormat that owns this rule |
463 | * @return The newly-created substitution. This is never null; if |
464 | * the rule text doesn't contain any substitution tokens, this will |
465 | * be a NullSubstitution. |
466 | */ |
467 | NFSubstitution * |
468 | NFRule::(const NFRuleSet* ruleSet, |
469 | const NFRule* predecessor, |
470 | UErrorCode& status) |
471 | { |
472 | NFSubstitution* result = NULL; |
473 | |
474 | // search the rule's rule text for the first two characters of |
475 | // a substitution token |
476 | int32_t subStart = indexOfAnyRulePrefix(); |
477 | int32_t subEnd = subStart; |
478 | |
479 | // if we didn't find one, create a null substitution positioned |
480 | // at the end of the rule text |
481 | if (subStart == -1) { |
482 | return NULL; |
483 | } |
484 | |
485 | // special-case the ">>>" token, since searching for the > at the |
486 | // end will actually find the > in the middle |
487 | if (fRuleText.indexOf(gGreaterGreaterGreater, 3, 0) == subStart) { |
488 | subEnd = subStart + 2; |
489 | |
490 | // otherwise the substitution token ends with the same character |
491 | // it began with |
492 | } else { |
493 | UChar c = fRuleText.charAt(subStart); |
494 | subEnd = fRuleText.indexOf(c, subStart + 1); |
495 | // special case for '<%foo<<' |
496 | if (c == gLessThan && subEnd != -1 && subEnd < fRuleText.length() - 1 && fRuleText.charAt(subEnd+1) == c) { |
497 | // ordinals use "=#,##0==%abbrev=" as their rule. Notice that the '==' in the middle |
498 | // occurs because of the juxtaposition of two different rules. The check for '<' is a hack |
499 | // to get around this. Having the duplicate at the front would cause problems with |
500 | // rules like "<<%" to format, say, percents... |
501 | ++subEnd; |
502 | } |
503 | } |
504 | |
505 | // if we don't find the end of the token (i.e., if we're on a single, |
506 | // unmatched token character), create a null substitution positioned |
507 | // at the end of the rule |
508 | if (subEnd == -1) { |
509 | return NULL; |
510 | } |
511 | |
512 | // if we get here, we have a real substitution token (or at least |
513 | // some text bounded by substitution token characters). Use |
514 | // makeSubstitution() to create the right kind of substitution |
515 | UnicodeString subToken; |
516 | subToken.setTo(fRuleText, subStart, subEnd + 1 - subStart); |
517 | result = NFSubstitution::makeSubstitution(subStart, this, predecessor, ruleSet, |
518 | this->formatter, subToken, status); |
519 | |
520 | // remove the substitution from the rule text |
521 | fRuleText.removeBetween(subStart, subEnd+1); |
522 | |
523 | return result; |
524 | } |
525 | |
526 | /** |
527 | * Sets the rule's base value, and causes the radix and exponent |
528 | * to be recalculated. This is used during construction when we |
529 | * don't know the rule's base value until after it's been |
530 | * constructed. It should be used at any other time. |
531 | * @param The new base value for the rule. |
532 | */ |
533 | void |
534 | NFRule::setBaseValue(int64_t newBaseValue, UErrorCode& status) |
535 | { |
536 | // set the base value |
537 | baseValue = newBaseValue; |
538 | radix = 10; |
539 | |
540 | // if this isn't a special rule, recalculate the radix and exponent |
541 | // (the radix always defaults to 10; if it's supposed to be something |
542 | // else, it's cleaned up by the caller and the exponent is |
543 | // recalculated again-- the only function that does this is |
544 | // NFRule.parseRuleDescriptor() ) |
545 | if (baseValue >= 1) { |
546 | exponent = expectedExponent(); |
547 | |
548 | // this function gets called on a fully-constructed rule whose |
549 | // description didn't specify a base value. This means it |
550 | // has substitutions, and some substitutions hold on to copies |
551 | // of the rule's divisor. Fix their copies of the divisor. |
552 | if (sub1 != NULL) { |
553 | sub1->setDivisor(radix, exponent, status); |
554 | } |
555 | if (sub2 != NULL) { |
556 | sub2->setDivisor(radix, exponent, status); |
557 | } |
558 | |
559 | // if this is a special rule, its radix and exponent are basically |
560 | // ignored. Set them to "safe" default values |
561 | } else { |
562 | exponent = 0; |
563 | } |
564 | } |
565 | |
566 | /** |
567 | * This calculates the rule's exponent based on its radix and base |
568 | * value. This will be the highest power the radix can be raised to |
569 | * and still produce a result less than or equal to the base value. |
570 | */ |
571 | int16_t |
572 | NFRule::expectedExponent() const |
573 | { |
574 | // since the log of 0, or the log base 0 of something, causes an |
575 | // error, declare the exponent in these cases to be 0 (we also |
576 | // deal with the special-rule identifiers here) |
577 | if (radix == 0 || baseValue < 1) { |
578 | return 0; |
579 | } |
580 | |
581 | // we get rounding error in some cases-- for example, log 1000 / log 10 |
582 | // gives us 1.9999999996 instead of 2. The extra logic here is to take |
583 | // that into account |
584 | int16_t tempResult = (int16_t)(uprv_log((double)baseValue) / uprv_log((double)radix)); |
585 | int64_t temp = util64_pow(radix, tempResult + 1); |
586 | if (temp <= baseValue) { |
587 | tempResult += 1; |
588 | } |
589 | return tempResult; |
590 | } |
591 | |
592 | /** |
593 | * Searches the rule's rule text for any of the specified strings. |
594 | * @return The index of the first match in the rule's rule text |
595 | * (i.e., the first substring in the rule's rule text that matches |
596 | * _any_ of the strings in "strings"). If none of the strings in |
597 | * "strings" is found in the rule's rule text, returns -1. |
598 | */ |
599 | int32_t |
600 | NFRule::indexOfAnyRulePrefix() const |
601 | { |
602 | int result = -1; |
603 | for (int i = 0; RULE_PREFIXES[i]; i++) { |
604 | int32_t pos = fRuleText.indexOf(*RULE_PREFIXES[i]); |
605 | if (pos != -1 && (result == -1 || pos < result)) { |
606 | result = pos; |
607 | } |
608 | } |
609 | return result; |
610 | } |
611 | |
612 | //----------------------------------------------------------------------- |
613 | // boilerplate |
614 | //----------------------------------------------------------------------- |
615 | |
616 | static UBool |
617 | util_equalSubstitutions(const NFSubstitution* sub1, const NFSubstitution* sub2) |
618 | { |
619 | if (sub1) { |
620 | if (sub2) { |
621 | return *sub1 == *sub2; |
622 | } |
623 | } else if (!sub2) { |
624 | return TRUE; |
625 | } |
626 | return FALSE; |
627 | } |
628 | |
629 | /** |
630 | * Tests two rules for equality. |
631 | * @param that The rule to compare this one against |
632 | * @return True is the two rules are functionally equivalent |
633 | */ |
634 | UBool |
635 | NFRule::operator==(const NFRule& rhs) const |
636 | { |
637 | return baseValue == rhs.baseValue |
638 | && radix == rhs.radix |
639 | && exponent == rhs.exponent |
640 | && fRuleText == rhs.fRuleText |
641 | && util_equalSubstitutions(sub1, rhs.sub1) |
642 | && util_equalSubstitutions(sub2, rhs.sub2); |
643 | } |
644 | |
645 | /** |
646 | * Returns a textual representation of the rule. This won't |
647 | * necessarily be the same as the description that this rule |
648 | * was created with, but it will produce the same result. |
649 | * @return A textual description of the rule |
650 | */ |
651 | static void util_append64(UnicodeString& result, int64_t n) |
652 | { |
653 | UChar buffer[256]; |
654 | int32_t len = util64_tou(n, buffer, sizeof(buffer)); |
655 | UnicodeString temp(buffer, len); |
656 | result.append(temp); |
657 | } |
658 | |
659 | void |
660 | NFRule::_appendRuleText(UnicodeString& result) const |
661 | { |
662 | switch (getType()) { |
663 | case kNegativeNumberRule: result.append(gMinusX, 2); break; |
664 | case kImproperFractionRule: result.append(gX).append(decimalPoint == 0 ? gDot : decimalPoint).append(gX); break; |
665 | case kProperFractionRule: result.append(gZero).append(decimalPoint == 0 ? gDot : decimalPoint).append(gX); break; |
666 | case kMasterRule: result.append(gX).append(decimalPoint == 0 ? gDot : decimalPoint).append(gZero); break; |
667 | case kInfinityRule: result.append(gInf, 3); break; |
668 | case kNaNRule: result.append(gNaN, 3); break; |
669 | default: |
670 | // for a normal rule, write out its base value, and if the radix is |
671 | // something other than 10, write out the radix (with the preceding |
672 | // slash, of course). Then calculate the expected exponent and if |
673 | // if isn't the same as the actual exponent, write an appropriate |
674 | // number of > signs. Finally, terminate the whole thing with |
675 | // a colon. |
676 | util_append64(result, baseValue); |
677 | if (radix != 10) { |
678 | result.append(gSlash); |
679 | util_append64(result, radix); |
680 | } |
681 | int numCarets = expectedExponent() - exponent; |
682 | for (int i = 0; i < numCarets; i++) { |
683 | result.append(gGreaterThan); |
684 | } |
685 | break; |
686 | } |
687 | result.append(gColon); |
688 | result.append(gSpace); |
689 | |
690 | // if the rule text begins with a space, write an apostrophe |
691 | // (whitespace after the rule descriptor is ignored; the |
692 | // apostrophe is used to make the whitespace significant) |
693 | if (fRuleText.charAt(0) == gSpace && (sub1 == NULL || sub1->getPos() != 0)) { |
694 | result.append(gTick); |
695 | } |
696 | |
697 | // now, write the rule's rule text, inserting appropriate |
698 | // substitution tokens in the appropriate places |
699 | UnicodeString ruleTextCopy; |
700 | ruleTextCopy.setTo(fRuleText); |
701 | |
702 | UnicodeString temp; |
703 | if (sub2 != NULL) { |
704 | sub2->toString(temp); |
705 | ruleTextCopy.insert(sub2->getPos(), temp); |
706 | } |
707 | if (sub1 != NULL) { |
708 | sub1->toString(temp); |
709 | ruleTextCopy.insert(sub1->getPos(), temp); |
710 | } |
711 | |
712 | result.append(ruleTextCopy); |
713 | |
714 | // and finally, top the whole thing off with a semicolon and |
715 | // return the result |
716 | result.append(gSemicolon); |
717 | } |
718 | |
719 | int64_t NFRule::getDivisor() const |
720 | { |
721 | return util64_pow(radix, exponent); |
722 | } |
723 | |
724 | |
725 | //----------------------------------------------------------------------- |
726 | // formatting |
727 | //----------------------------------------------------------------------- |
728 | |
729 | /** |
730 | * Formats the number, and inserts the resulting text into |
731 | * toInsertInto. |
732 | * @param number The number being formatted |
733 | * @param toInsertInto The string where the resultant text should |
734 | * be inserted |
735 | * @param pos The position in toInsertInto where the resultant text |
736 | * should be inserted |
737 | */ |
738 | void |
739 | NFRule::doFormat(int64_t number, UnicodeString& toInsertInto, int32_t pos, int32_t recursionCount, UErrorCode& status) const |
740 | { |
741 | // first, insert the rule's rule text into toInsertInto at the |
742 | // specified position, then insert the results of the substitutions |
743 | // into the right places in toInsertInto (notice we do the |
744 | // substitutions in reverse order so that the offsets don't get |
745 | // messed up) |
746 | int32_t pluralRuleStart = fRuleText.length(); |
747 | int32_t lengthOffset = 0; |
748 | if (!rulePatternFormat) { |
749 | toInsertInto.insert(pos, fRuleText); |
750 | } |
751 | else { |
752 | pluralRuleStart = fRuleText.indexOf(gDollarOpenParenthesis, -1, 0); |
753 | int pluralRuleEnd = fRuleText.indexOf(gClosedParenthesisDollar, -1, pluralRuleStart); |
754 | int initialLength = toInsertInto.length(); |
755 | if (pluralRuleEnd < fRuleText.length() - 1) { |
756 | toInsertInto.insert(pos, fRuleText.tempSubString(pluralRuleEnd + 2)); |
757 | } |
758 | toInsertInto.insert(pos, |
759 | rulePatternFormat->format((int32_t)(number/util64_pow(radix, exponent)), status)); |
760 | if (pluralRuleStart > 0) { |
761 | toInsertInto.insert(pos, fRuleText.tempSubString(0, pluralRuleStart)); |
762 | } |
763 | lengthOffset = fRuleText.length() - (toInsertInto.length() - initialLength); |
764 | } |
765 | |
766 | if (sub2 != NULL) { |
767 | sub2->doSubstitution(number, toInsertInto, pos - (sub2->getPos() > pluralRuleStart ? lengthOffset : 0), recursionCount, status); |
768 | } |
769 | if (sub1 != NULL) { |
770 | sub1->doSubstitution(number, toInsertInto, pos - (sub1->getPos() > pluralRuleStart ? lengthOffset : 0), recursionCount, status); |
771 | } |
772 | } |
773 | |
774 | /** |
775 | * Formats the number, and inserts the resulting text into |
776 | * toInsertInto. |
777 | * @param number The number being formatted |
778 | * @param toInsertInto The string where the resultant text should |
779 | * be inserted |
780 | * @param pos The position in toInsertInto where the resultant text |
781 | * should be inserted |
782 | */ |
783 | void |
784 | NFRule::doFormat(double number, UnicodeString& toInsertInto, int32_t pos, int32_t recursionCount, UErrorCode& status) const |
785 | { |
786 | // first, insert the rule's rule text into toInsertInto at the |
787 | // specified position, then insert the results of the substitutions |
788 | // into the right places in toInsertInto |
789 | // [again, we have two copies of this routine that do the same thing |
790 | // so that we don't sacrifice precision in a long by casting it |
791 | // to a double] |
792 | int32_t pluralRuleStart = fRuleText.length(); |
793 | int32_t lengthOffset = 0; |
794 | if (!rulePatternFormat) { |
795 | toInsertInto.insert(pos, fRuleText); |
796 | } |
797 | else { |
798 | pluralRuleStart = fRuleText.indexOf(gDollarOpenParenthesis, -1, 0); |
799 | int pluralRuleEnd = fRuleText.indexOf(gClosedParenthesisDollar, -1, pluralRuleStart); |
800 | int initialLength = toInsertInto.length(); |
801 | if (pluralRuleEnd < fRuleText.length() - 1) { |
802 | toInsertInto.insert(pos, fRuleText.tempSubString(pluralRuleEnd + 2)); |
803 | } |
804 | double pluralVal = number; |
805 | if (0 <= pluralVal && pluralVal < 1) { |
806 | // We're in a fractional rule, and we have to match the NumeratorSubstitution behavior. |
807 | // 2.3 can become 0.2999999999999998 for the fraction due to rounding errors. |
808 | pluralVal = uprv_round(pluralVal * util64_pow(radix, exponent)); |
809 | } |
810 | else { |
811 | pluralVal = pluralVal / util64_pow(radix, exponent); |
812 | } |
813 | toInsertInto.insert(pos, rulePatternFormat->format((int32_t)(pluralVal), status)); |
814 | if (pluralRuleStart > 0) { |
815 | toInsertInto.insert(pos, fRuleText.tempSubString(0, pluralRuleStart)); |
816 | } |
817 | lengthOffset = fRuleText.length() - (toInsertInto.length() - initialLength); |
818 | } |
819 | |
820 | if (sub2 != NULL) { |
821 | sub2->doSubstitution(number, toInsertInto, pos - (sub2->getPos() > pluralRuleStart ? lengthOffset : 0), recursionCount, status); |
822 | } |
823 | if (sub1 != NULL) { |
824 | sub1->doSubstitution(number, toInsertInto, pos - (sub1->getPos() > pluralRuleStart ? lengthOffset : 0), recursionCount, status); |
825 | } |
826 | } |
827 | |
828 | /** |
829 | * Used by the owning rule set to determine whether to invoke the |
830 | * rollback rule (i.e., whether this rule or the one that precedes |
831 | * it in the rule set's list should be used to format the number) |
832 | * @param The number being formatted |
833 | * @return True if the rule set should use the rule that precedes |
834 | * this one in its list; false if it should use this rule |
835 | */ |
836 | UBool |
837 | NFRule::shouldRollBack(int64_t number) const |
838 | { |
839 | // we roll back if the rule contains a modulus substitution, |
840 | // the number being formatted is an even multiple of the rule's |
841 | // divisor, and the rule's base value is NOT an even multiple |
842 | // of its divisor |
843 | // In other words, if the original description had |
844 | // 100: << hundred[ >>]; |
845 | // that expands into |
846 | // 100: << hundred; |
847 | // 101: << hundred >>; |
848 | // internally. But when we're formatting 200, if we use the rule |
849 | // at 101, which would normally apply, we get "two hundred zero". |
850 | // To prevent this, we roll back and use the rule at 100 instead. |
851 | // This is the logic that makes this happen: the rule at 101 has |
852 | // a modulus substitution, its base value isn't an even multiple |
853 | // of 100, and the value we're trying to format _is_ an even |
854 | // multiple of 100. This is called the "rollback rule." |
855 | if ((sub1 != NULL && sub1->isModulusSubstitution()) || (sub2 != NULL && sub2->isModulusSubstitution())) { |
856 | int64_t re = util64_pow(radix, exponent); |
857 | return (number % re) == 0 && (baseValue % re) != 0; |
858 | } |
859 | return FALSE; |
860 | } |
861 | |
862 | //----------------------------------------------------------------------- |
863 | // parsing |
864 | //----------------------------------------------------------------------- |
865 | |
866 | /** |
867 | * Attempts to parse the string with this rule. |
868 | * @param text The string being parsed |
869 | * @param parsePosition On entry, the value is ignored and assumed to |
870 | * be 0. On exit, this has been updated with the position of the first |
871 | * character not consumed by matching the text against this rule |
872 | * (if this rule doesn't match the text at all, the parse position |
873 | * if left unchanged (presumably at 0) and the function returns |
874 | * new Long(0)). |
875 | * @param isFractionRule True if this rule is contained within a |
876 | * fraction rule set. This is only used if the rule has no |
877 | * substitutions. |
878 | * @return If this rule matched the text, this is the rule's base value |
879 | * combined appropriately with the results of parsing the substitutions. |
880 | * If nothing matched, this is new Long(0) and the parse position is |
881 | * left unchanged. The result will be an instance of Long if the |
882 | * result is an integer and Double otherwise. The result is never null. |
883 | */ |
884 | #ifdef RBNF_DEBUG |
885 | #include <stdio.h> |
886 | |
887 | static void dumpUS(FILE* f, const UnicodeString& us) { |
888 | int len = us.length(); |
889 | char* buf = (char *)uprv_malloc((len+1)*sizeof(char)); //new char[len+1]; |
890 | if (buf != NULL) { |
891 | us.extract(0, len, buf); |
892 | buf[len] = 0; |
893 | fprintf(f, "%s" , buf); |
894 | uprv_free(buf); //delete[] buf; |
895 | } |
896 | } |
897 | #endif |
898 | UBool |
899 | NFRule::doParse(const UnicodeString& text, |
900 | ParsePosition& parsePosition, |
901 | UBool isFractionRule, |
902 | double upperBound, |
903 | uint32_t nonNumericalExecutedRuleMask, |
904 | Formattable& resVal) const |
905 | { |
906 | // internally we operate on a copy of the string being parsed |
907 | // (because we're going to change it) and use our own ParsePosition |
908 | ParsePosition pp; |
909 | UnicodeString workText(text); |
910 | |
911 | int32_t sub1Pos = sub1 != NULL ? sub1->getPos() : fRuleText.length(); |
912 | int32_t sub2Pos = sub2 != NULL ? sub2->getPos() : fRuleText.length(); |
913 | |
914 | // check to see whether the text before the first substitution |
915 | // matches the text at the beginning of the string being |
916 | // parsed. If it does, strip that off the front of workText; |
917 | // otherwise, dump out with a mismatch |
918 | UnicodeString prefix; |
919 | prefix.setTo(fRuleText, 0, sub1Pos); |
920 | |
921 | #ifdef RBNF_DEBUG |
922 | fprintf(stderr, "doParse %p " , this); |
923 | { |
924 | UnicodeString rt; |
925 | _appendRuleText(rt); |
926 | dumpUS(stderr, rt); |
927 | } |
928 | |
929 | fprintf(stderr, " text: '" ); |
930 | dumpUS(stderr, text); |
931 | fprintf(stderr, "' prefix: '" ); |
932 | dumpUS(stderr, prefix); |
933 | #endif |
934 | stripPrefix(workText, prefix, pp); |
935 | int32_t prefixLength = text.length() - workText.length(); |
936 | |
937 | #ifdef RBNF_DEBUG |
938 | fprintf(stderr, "' pl: %d ppi: %d s1p: %d\n" , prefixLength, pp.getIndex(), sub1Pos); |
939 | #endif |
940 | |
941 | if (pp.getIndex() == 0 && sub1Pos != 0) { |
942 | // commented out because ParsePosition doesn't have error index in 1.1.x |
943 | // restored for ICU4C port |
944 | parsePosition.setErrorIndex(pp.getErrorIndex()); |
945 | resVal.setLong(0); |
946 | return TRUE; |
947 | } |
948 | if (baseValue == kInfinityRule) { |
949 | // If you match this, don't try to perform any calculations on it. |
950 | parsePosition.setIndex(pp.getIndex()); |
951 | resVal.setDouble(uprv_getInfinity()); |
952 | return TRUE; |
953 | } |
954 | if (baseValue == kNaNRule) { |
955 | // If you match this, don't try to perform any calculations on it. |
956 | parsePosition.setIndex(pp.getIndex()); |
957 | resVal.setDouble(uprv_getNaN()); |
958 | return TRUE; |
959 | } |
960 | |
961 | // this is the fun part. The basic guts of the rule-matching |
962 | // logic is matchToDelimiter(), which is called twice. The first |
963 | // time it searches the input string for the rule text BETWEEN |
964 | // the substitutions and tries to match the intervening text |
965 | // in the input string with the first substitution. If that |
966 | // succeeds, it then calls it again, this time to look for the |
967 | // rule text after the second substitution and to match the |
968 | // intervening input text against the second substitution. |
969 | // |
970 | // For example, say we have a rule that looks like this: |
971 | // first << middle >> last; |
972 | // and input text that looks like this: |
973 | // first one middle two last |
974 | // First we use stripPrefix() to match "first " in both places and |
975 | // strip it off the front, leaving |
976 | // one middle two last |
977 | // Then we use matchToDelimiter() to match " middle " and try to |
978 | // match "one" against a substitution. If it's successful, we now |
979 | // have |
980 | // two last |
981 | // We use matchToDelimiter() a second time to match " last" and |
982 | // try to match "two" against a substitution. If "two" matches |
983 | // the substitution, we have a successful parse. |
984 | // |
985 | // Since it's possible in many cases to find multiple instances |
986 | // of each of these pieces of rule text in the input string, |
987 | // we need to try all the possible combinations of these |
988 | // locations. This prevents us from prematurely declaring a mismatch, |
989 | // and makes sure we match as much input text as we can. |
990 | int highWaterMark = 0; |
991 | double result = 0; |
992 | int start = 0; |
993 | double tempBaseValue = (double)(baseValue <= 0 ? 0 : baseValue); |
994 | |
995 | UnicodeString temp; |
996 | do { |
997 | // our partial parse result starts out as this rule's base |
998 | // value. If it finds a successful match, matchToDelimiter() |
999 | // will compose this in some way with what it gets back from |
1000 | // the substitution, giving us a new partial parse result |
1001 | pp.setIndex(0); |
1002 | |
1003 | temp.setTo(fRuleText, sub1Pos, sub2Pos - sub1Pos); |
1004 | double partialResult = matchToDelimiter(workText, start, tempBaseValue, |
1005 | temp, pp, sub1, |
1006 | nonNumericalExecutedRuleMask, |
1007 | upperBound); |
1008 | |
1009 | // if we got a successful match (or were trying to match a |
1010 | // null substitution), pp is now pointing at the first unmatched |
1011 | // character. Take note of that, and try matchToDelimiter() |
1012 | // on the input text again |
1013 | if (pp.getIndex() != 0 || sub1 == NULL) { |
1014 | start = pp.getIndex(); |
1015 | |
1016 | UnicodeString workText2; |
1017 | workText2.setTo(workText, pp.getIndex(), workText.length() - pp.getIndex()); |
1018 | ParsePosition pp2; |
1019 | |
1020 | // the second matchToDelimiter() will compose our previous |
1021 | // partial result with whatever it gets back from its |
1022 | // substitution if there's a successful match, giving us |
1023 | // a real result |
1024 | temp.setTo(fRuleText, sub2Pos, fRuleText.length() - sub2Pos); |
1025 | partialResult = matchToDelimiter(workText2, 0, partialResult, |
1026 | temp, pp2, sub2, |
1027 | nonNumericalExecutedRuleMask, |
1028 | upperBound); |
1029 | |
1030 | // if we got a successful match on this second |
1031 | // matchToDelimiter() call, update the high-water mark |
1032 | // and result (if necessary) |
1033 | if (pp2.getIndex() != 0 || sub2 == NULL) { |
1034 | if (prefixLength + pp.getIndex() + pp2.getIndex() > highWaterMark) { |
1035 | highWaterMark = prefixLength + pp.getIndex() + pp2.getIndex(); |
1036 | result = partialResult; |
1037 | } |
1038 | } |
1039 | else { |
1040 | // commented out because ParsePosition doesn't have error index in 1.1.x |
1041 | // restored for ICU4C port |
1042 | int32_t i_temp = pp2.getErrorIndex() + sub1Pos + pp.getIndex(); |
1043 | if (i_temp> parsePosition.getErrorIndex()) { |
1044 | parsePosition.setErrorIndex(i_temp); |
1045 | } |
1046 | } |
1047 | } |
1048 | else { |
1049 | // commented out because ParsePosition doesn't have error index in 1.1.x |
1050 | // restored for ICU4C port |
1051 | int32_t i_temp = sub1Pos + pp.getErrorIndex(); |
1052 | if (i_temp > parsePosition.getErrorIndex()) { |
1053 | parsePosition.setErrorIndex(i_temp); |
1054 | } |
1055 | } |
1056 | // keep trying to match things until the outer matchToDelimiter() |
1057 | // call fails to make a match (each time, it picks up where it |
1058 | // left off the previous time) |
1059 | } while (sub1Pos != sub2Pos |
1060 | && pp.getIndex() > 0 |
1061 | && pp.getIndex() < workText.length() |
1062 | && pp.getIndex() != start); |
1063 | |
1064 | // update the caller's ParsePosition with our high-water mark |
1065 | // (i.e., it now points at the first character this function |
1066 | // didn't match-- the ParsePosition is therefore unchanged if |
1067 | // we didn't match anything) |
1068 | parsePosition.setIndex(highWaterMark); |
1069 | // commented out because ParsePosition doesn't have error index in 1.1.x |
1070 | // restored for ICU4C port |
1071 | if (highWaterMark > 0) { |
1072 | parsePosition.setErrorIndex(0); |
1073 | } |
1074 | |
1075 | // this is a hack for one unusual condition: Normally, whether this |
1076 | // rule belong to a fraction rule set or not is handled by its |
1077 | // substitutions. But if that rule HAS NO substitutions, then |
1078 | // we have to account for it here. By definition, if the matching |
1079 | // rule in a fraction rule set has no substitutions, its numerator |
1080 | // is 1, and so the result is the reciprocal of its base value. |
1081 | if (isFractionRule && highWaterMark > 0 && sub1 == NULL) { |
1082 | result = 1 / result; |
1083 | } |
1084 | |
1085 | resVal.setDouble(result); |
1086 | return TRUE; // ??? do we need to worry if it is a long or a double? |
1087 | } |
1088 | |
1089 | /** |
1090 | * This function is used by parse() to match the text being parsed |
1091 | * against a possible prefix string. This function |
1092 | * matches characters from the beginning of the string being parsed |
1093 | * to characters from the prospective prefix. If they match, pp is |
1094 | * updated to the first character not matched, and the result is |
1095 | * the unparsed part of the string. If they don't match, the whole |
1096 | * string is returned, and pp is left unchanged. |
1097 | * @param text The string being parsed |
1098 | * @param prefix The text to match against |
1099 | * @param pp On entry, ignored and assumed to be 0. On exit, points |
1100 | * to the first unmatched character (assuming the whole prefix matched), |
1101 | * or is unchanged (if the whole prefix didn't match). |
1102 | * @return If things match, this is the unparsed part of "text"; |
1103 | * if they didn't match, this is "text". |
1104 | */ |
1105 | void |
1106 | NFRule::stripPrefix(UnicodeString& text, const UnicodeString& prefix, ParsePosition& pp) const |
1107 | { |
1108 | // if the prefix text is empty, dump out without doing anything |
1109 | if (prefix.length() != 0) { |
1110 | UErrorCode status = U_ZERO_ERROR; |
1111 | // use prefixLength() to match the beginning of |
1112 | // "text" against "prefix". This function returns the |
1113 | // number of characters from "text" that matched (or 0 if |
1114 | // we didn't match the whole prefix) |
1115 | int32_t pfl = prefixLength(text, prefix, status); |
1116 | if (U_FAILURE(status)) { // Memory allocation error. |
1117 | return; |
1118 | } |
1119 | if (pfl != 0) { |
1120 | // if we got a successful match, update the parse position |
1121 | // and strip the prefix off of "text" |
1122 | pp.setIndex(pp.getIndex() + pfl); |
1123 | text.remove(0, pfl); |
1124 | } |
1125 | } |
1126 | } |
1127 | |
1128 | /** |
1129 | * Used by parse() to match a substitution and any following text. |
1130 | * "text" is searched for instances of "delimiter". For each instance |
1131 | * of delimiter, the intervening text is tested to see whether it |
1132 | * matches the substitution. The longest match wins. |
1133 | * @param text The string being parsed |
1134 | * @param startPos The position in "text" where we should start looking |
1135 | * for "delimiter". |
1136 | * @param baseValue A partial parse result (often the rule's base value), |
1137 | * which is combined with the result from matching the substitution |
1138 | * @param delimiter The string to search "text" for. |
1139 | * @param pp Ignored and presumed to be 0 on entry. If there's a match, |
1140 | * on exit this will point to the first unmatched character. |
1141 | * @param sub If we find "delimiter" in "text", this substitution is used |
1142 | * to match the text between the beginning of the string and the |
1143 | * position of "delimiter." (If "delimiter" is the empty string, then |
1144 | * this function just matches against this substitution and updates |
1145 | * everything accordingly.) |
1146 | * @param upperBound When matching the substitution, it will only |
1147 | * consider rules with base values lower than this value. |
1148 | * @return If there's a match, this is the result of composing |
1149 | * baseValue with the result of matching the substitution. Otherwise, |
1150 | * this is new Long(0). It's never null. If the result is an integer, |
1151 | * this will be an instance of Long; otherwise, it's an instance of |
1152 | * Double. |
1153 | * |
1154 | * !!! note {dlf} in point of fact, in the java code the caller always converts |
1155 | * the result to a double, so we might as well return one. |
1156 | */ |
1157 | double |
1158 | NFRule::matchToDelimiter(const UnicodeString& text, |
1159 | int32_t startPos, |
1160 | double _baseValue, |
1161 | const UnicodeString& delimiter, |
1162 | ParsePosition& pp, |
1163 | const NFSubstitution* sub, |
1164 | uint32_t nonNumericalExecutedRuleMask, |
1165 | double upperBound) const |
1166 | { |
1167 | UErrorCode status = U_ZERO_ERROR; |
1168 | // if "delimiter" contains real (i.e., non-ignorable) text, search |
1169 | // it for "delimiter" beginning at "start". If that succeeds, then |
1170 | // use "sub"'s doParse() method to match the text before the |
1171 | // instance of "delimiter" we just found. |
1172 | if (!allIgnorable(delimiter, status)) { |
1173 | if (U_FAILURE(status)) { //Memory allocation error. |
1174 | return 0; |
1175 | } |
1176 | ParsePosition tempPP; |
1177 | Formattable result; |
1178 | |
1179 | // use findText() to search for "delimiter". It returns a two- |
1180 | // element array: element 0 is the position of the match, and |
1181 | // element 1 is the number of characters that matched |
1182 | // "delimiter". |
1183 | int32_t dLen; |
1184 | int32_t dPos = findText(text, delimiter, startPos, &dLen); |
1185 | |
1186 | // if findText() succeeded, isolate the text preceding the |
1187 | // match, and use "sub" to match that text |
1188 | while (dPos >= 0) { |
1189 | UnicodeString subText; |
1190 | subText.setTo(text, 0, dPos); |
1191 | if (subText.length() > 0) { |
1192 | UBool success = sub->doParse(subText, tempPP, _baseValue, upperBound, |
1193 | #if UCONFIG_NO_COLLATION |
1194 | FALSE, |
1195 | #else |
1196 | formatter->isLenient(), |
1197 | #endif |
1198 | nonNumericalExecutedRuleMask, |
1199 | result); |
1200 | |
1201 | // if the substitution could match all the text up to |
1202 | // where we found "delimiter", then this function has |
1203 | // a successful match. Bump the caller's parse position |
1204 | // to point to the first character after the text |
1205 | // that matches "delimiter", and return the result |
1206 | // we got from parsing the substitution. |
1207 | if (success && tempPP.getIndex() == dPos) { |
1208 | pp.setIndex(dPos + dLen); |
1209 | return result.getDouble(); |
1210 | } |
1211 | else { |
1212 | // commented out because ParsePosition doesn't have error index in 1.1.x |
1213 | // restored for ICU4C port |
1214 | if (tempPP.getErrorIndex() > 0) { |
1215 | pp.setErrorIndex(tempPP.getErrorIndex()); |
1216 | } else { |
1217 | pp.setErrorIndex(tempPP.getIndex()); |
1218 | } |
1219 | } |
1220 | } |
1221 | |
1222 | // if we didn't match the substitution, search for another |
1223 | // copy of "delimiter" in "text" and repeat the loop if |
1224 | // we find it |
1225 | tempPP.setIndex(0); |
1226 | dPos = findText(text, delimiter, dPos + dLen, &dLen); |
1227 | } |
1228 | // if we make it here, this was an unsuccessful match, and we |
1229 | // leave pp unchanged and return 0 |
1230 | pp.setIndex(0); |
1231 | return 0; |
1232 | |
1233 | // if "delimiter" is empty, or consists only of ignorable characters |
1234 | // (i.e., is semantically empty), thwe we obviously can't search |
1235 | // for "delimiter". Instead, just use "sub" to parse as much of |
1236 | // "text" as possible. |
1237 | } |
1238 | else if (sub == NULL) { |
1239 | return _baseValue; |
1240 | } |
1241 | else { |
1242 | ParsePosition tempPP; |
1243 | Formattable result; |
1244 | |
1245 | // try to match the whole string against the substitution |
1246 | UBool success = sub->doParse(text, tempPP, _baseValue, upperBound, |
1247 | #if UCONFIG_NO_COLLATION |
1248 | FALSE, |
1249 | #else |
1250 | formatter->isLenient(), |
1251 | #endif |
1252 | nonNumericalExecutedRuleMask, |
1253 | result); |
1254 | if (success && (tempPP.getIndex() != 0)) { |
1255 | // if there's a successful match (or it's a null |
1256 | // substitution), update pp to point to the first |
1257 | // character we didn't match, and pass the result from |
1258 | // sub.doParse() on through to the caller |
1259 | pp.setIndex(tempPP.getIndex()); |
1260 | return result.getDouble(); |
1261 | } |
1262 | else { |
1263 | // commented out because ParsePosition doesn't have error index in 1.1.x |
1264 | // restored for ICU4C port |
1265 | pp.setErrorIndex(tempPP.getErrorIndex()); |
1266 | } |
1267 | |
1268 | // and if we get to here, then nothing matched, so we return |
1269 | // 0 and leave pp alone |
1270 | return 0; |
1271 | } |
1272 | } |
1273 | |
1274 | /** |
1275 | * Used by stripPrefix() to match characters. If lenient parse mode |
1276 | * is off, this just calls startsWith(). If lenient parse mode is on, |
1277 | * this function uses CollationElementIterators to match characters in |
1278 | * the strings (only primary-order differences are significant in |
1279 | * determining whether there's a match). |
1280 | * @param str The string being tested |
1281 | * @param prefix The text we're hoping to see at the beginning |
1282 | * of "str" |
1283 | * @return If "prefix" is found at the beginning of "str", this |
1284 | * is the number of characters in "str" that were matched (this |
1285 | * isn't necessarily the same as the length of "prefix" when matching |
1286 | * text with a collator). If there's no match, this is 0. |
1287 | */ |
1288 | int32_t |
1289 | NFRule::prefixLength(const UnicodeString& str, const UnicodeString& prefix, UErrorCode& status) const |
1290 | { |
1291 | // if we're looking for an empty prefix, it obviously matches |
1292 | // zero characters. Just go ahead and return 0. |
1293 | if (prefix.length() == 0) { |
1294 | return 0; |
1295 | } |
1296 | |
1297 | #if !UCONFIG_NO_COLLATION |
1298 | // go through all this grief if we're in lenient-parse mode |
1299 | if (formatter->isLenient()) { |
1300 | // get the formatter's collator and use it to create two |
1301 | // collation element iterators, one over the target string |
1302 | // and another over the prefix (right now, we'll throw an |
1303 | // exception if the collator we get back from the formatter |
1304 | // isn't a RuleBasedCollator, because RuleBasedCollator defines |
1305 | // the CollationElementIterator protocol. Hopefully, this |
1306 | // will change someday.) |
1307 | const RuleBasedCollator* collator = formatter->getCollator(); |
1308 | if (collator == NULL) { |
1309 | status = U_MEMORY_ALLOCATION_ERROR; |
1310 | return 0; |
1311 | } |
1312 | LocalPointer<CollationElementIterator> strIter(collator->createCollationElementIterator(str)); |
1313 | LocalPointer<CollationElementIterator> prefixIter(collator->createCollationElementIterator(prefix)); |
1314 | // Check for memory allocation error. |
1315 | if (strIter.isNull() || prefixIter.isNull()) { |
1316 | status = U_MEMORY_ALLOCATION_ERROR; |
1317 | return 0; |
1318 | } |
1319 | |
1320 | UErrorCode err = U_ZERO_ERROR; |
1321 | |
1322 | // The original code was problematic. Consider this match: |
1323 | // prefix = "fifty-" |
1324 | // string = " fifty-7" |
1325 | // The intent is to match string up to the '7', by matching 'fifty-' at position 1 |
1326 | // in the string. Unfortunately, we were getting a match, and then computing where |
1327 | // the match terminated by rematching the string. The rematch code was using as an |
1328 | // initial guess the substring of string between 0 and prefix.length. Because of |
1329 | // the leading space and trailing hyphen (both ignorable) this was succeeding, leaving |
1330 | // the position before the hyphen in the string. Recursing down, we then parsed the |
1331 | // remaining string '-7' as numeric. The resulting number turned out as 43 (50 - 7). |
1332 | // This was not pretty, especially since the string "fifty-7" parsed just fine. |
1333 | // |
1334 | // We have newer APIs now, so we can use calls on the iterator to determine what we |
1335 | // matched up to. If we terminate because we hit the last element in the string, |
1336 | // our match terminates at this length. If we terminate because we hit the last element |
1337 | // in the target, our match terminates at one before the element iterator position. |
1338 | |
1339 | // match collation elements between the strings |
1340 | int32_t oStr = strIter->next(err); |
1341 | int32_t oPrefix = prefixIter->next(err); |
1342 | |
1343 | while (oPrefix != CollationElementIterator::NULLORDER) { |
1344 | // skip over ignorable characters in the target string |
1345 | while (CollationElementIterator::primaryOrder(oStr) == 0 |
1346 | && oStr != CollationElementIterator::NULLORDER) { |
1347 | oStr = strIter->next(err); |
1348 | } |
1349 | |
1350 | // skip over ignorable characters in the prefix |
1351 | while (CollationElementIterator::primaryOrder(oPrefix) == 0 |
1352 | && oPrefix != CollationElementIterator::NULLORDER) { |
1353 | oPrefix = prefixIter->next(err); |
1354 | } |
1355 | |
1356 | // dlf: move this above following test, if we consume the |
1357 | // entire target, aren't we ok even if the source was also |
1358 | // entirely consumed? |
1359 | |
1360 | // if skipping over ignorables brought to the end of |
1361 | // the prefix, we DID match: drop out of the loop |
1362 | if (oPrefix == CollationElementIterator::NULLORDER) { |
1363 | break; |
1364 | } |
1365 | |
1366 | // if skipping over ignorables brought us to the end |
1367 | // of the target string, we didn't match and return 0 |
1368 | if (oStr == CollationElementIterator::NULLORDER) { |
1369 | return 0; |
1370 | } |
1371 | |
1372 | // match collation elements from the two strings |
1373 | // (considering only primary differences). If we |
1374 | // get a mismatch, dump out and return 0 |
1375 | if (CollationElementIterator::primaryOrder(oStr) |
1376 | != CollationElementIterator::primaryOrder(oPrefix)) { |
1377 | return 0; |
1378 | |
1379 | // otherwise, advance to the next character in each string |
1380 | // and loop (we drop out of the loop when we exhaust |
1381 | // collation elements in the prefix) |
1382 | } else { |
1383 | oStr = strIter->next(err); |
1384 | oPrefix = prefixIter->next(err); |
1385 | } |
1386 | } |
1387 | |
1388 | int32_t result = strIter->getOffset(); |
1389 | if (oStr != CollationElementIterator::NULLORDER) { |
1390 | --result; // back over character that we don't want to consume; |
1391 | } |
1392 | |
1393 | #ifdef RBNF_DEBUG |
1394 | fprintf(stderr, "prefix length: %d\n" , result); |
1395 | #endif |
1396 | return result; |
1397 | #if 0 |
1398 | //---------------------------------------------------------------- |
1399 | // JDK 1.2-specific API call |
1400 | // return strIter.getOffset(); |
1401 | //---------------------------------------------------------------- |
1402 | // JDK 1.1 HACK (take out for 1.2-specific code) |
1403 | |
1404 | // if we make it to here, we have a successful match. Now we |
1405 | // have to find out HOW MANY characters from the target string |
1406 | // matched the prefix (there isn't necessarily a one-to-one |
1407 | // mapping between collation elements and characters). |
1408 | // In JDK 1.2, there's a simple getOffset() call we can use. |
1409 | // In JDK 1.1, on the other hand, we have to go through some |
1410 | // ugly contortions. First, use the collator to compare the |
1411 | // same number of characters from the prefix and target string. |
1412 | // If they're equal, we're done. |
1413 | collator->setStrength(Collator::PRIMARY); |
1414 | if (str.length() >= prefix.length()) { |
1415 | UnicodeString temp; |
1416 | temp.setTo(str, 0, prefix.length()); |
1417 | if (collator->equals(temp, prefix)) { |
1418 | #ifdef RBNF_DEBUG |
1419 | fprintf(stderr, "returning: %d\n" , prefix.length()); |
1420 | #endif |
1421 | return prefix.length(); |
1422 | } |
1423 | } |
1424 | |
1425 | // if they're not equal, then we have to compare successively |
1426 | // larger and larger substrings of the target string until we |
1427 | // get to one that matches the prefix. At that point, we know |
1428 | // how many characters matched the prefix, and we can return. |
1429 | int32_t p = 1; |
1430 | while (p <= str.length()) { |
1431 | UnicodeString temp; |
1432 | temp.setTo(str, 0, p); |
1433 | if (collator->equals(temp, prefix)) { |
1434 | return p; |
1435 | } else { |
1436 | ++p; |
1437 | } |
1438 | } |
1439 | |
1440 | // SHOULD NEVER GET HERE!!! |
1441 | return 0; |
1442 | //---------------------------------------------------------------- |
1443 | #endif |
1444 | |
1445 | // If lenient parsing is turned off, forget all that crap above. |
1446 | // Just use String.startsWith() and be done with it. |
1447 | } else |
1448 | #endif |
1449 | { |
1450 | if (str.startsWith(prefix)) { |
1451 | return prefix.length(); |
1452 | } else { |
1453 | return 0; |
1454 | } |
1455 | } |
1456 | } |
1457 | |
1458 | /** |
1459 | * Searches a string for another string. If lenient parsing is off, |
1460 | * this just calls indexOf(). If lenient parsing is on, this function |
1461 | * uses CollationElementIterator to match characters, and only |
1462 | * primary-order differences are significant in determining whether |
1463 | * there's a match. |
1464 | * @param str The string to search |
1465 | * @param key The string to search "str" for |
1466 | * @param startingAt The index into "str" where the search is to |
1467 | * begin |
1468 | * @return A two-element array of ints. Element 0 is the position |
1469 | * of the match, or -1 if there was no match. Element 1 is the |
1470 | * number of characters in "str" that matched (which isn't necessarily |
1471 | * the same as the length of "key") |
1472 | */ |
1473 | int32_t |
1474 | NFRule::findText(const UnicodeString& str, |
1475 | const UnicodeString& key, |
1476 | int32_t startingAt, |
1477 | int32_t* length) const |
1478 | { |
1479 | if (rulePatternFormat) { |
1480 | Formattable result; |
1481 | FieldPosition position(UNUM_INTEGER_FIELD); |
1482 | position.setBeginIndex(startingAt); |
1483 | rulePatternFormat->parseType(str, this, result, position); |
1484 | int start = position.getBeginIndex(); |
1485 | if (start >= 0) { |
1486 | int32_t pluralRuleStart = fRuleText.indexOf(gDollarOpenParenthesis, -1, 0); |
1487 | int32_t pluralRuleSuffix = fRuleText.indexOf(gClosedParenthesisDollar, -1, pluralRuleStart) + 2; |
1488 | int32_t matchLen = position.getEndIndex() - start; |
1489 | UnicodeString prefix(fRuleText.tempSubString(0, pluralRuleStart)); |
1490 | UnicodeString suffix(fRuleText.tempSubString(pluralRuleSuffix)); |
1491 | if (str.compare(start - prefix.length(), prefix.length(), prefix, 0, prefix.length()) == 0 |
1492 | && str.compare(start + matchLen, suffix.length(), suffix, 0, suffix.length()) == 0) |
1493 | { |
1494 | *length = matchLen + prefix.length() + suffix.length(); |
1495 | return start - prefix.length(); |
1496 | } |
1497 | } |
1498 | *length = 0; |
1499 | return -1; |
1500 | } |
1501 | if (!formatter->isLenient()) { |
1502 | // if lenient parsing is turned off, this is easy: just call |
1503 | // String.indexOf() and we're done |
1504 | *length = key.length(); |
1505 | return str.indexOf(key, startingAt); |
1506 | } |
1507 | else { |
1508 | // but if lenient parsing is turned ON, we've got some work |
1509 | // ahead of us |
1510 | return findTextLenient(str, key, startingAt, length); |
1511 | } |
1512 | } |
1513 | |
1514 | int32_t |
1515 | NFRule::findTextLenient(const UnicodeString& str, |
1516 | const UnicodeString& key, |
1517 | int32_t startingAt, |
1518 | int32_t* length) const |
1519 | { |
1520 | //---------------------------------------------------------------- |
1521 | // JDK 1.1 HACK (take out of 1.2-specific code) |
1522 | |
1523 | // in JDK 1.2, CollationElementIterator provides us with an |
1524 | // API to map between character offsets and collation elements |
1525 | // and we can do this by marching through the string comparing |
1526 | // collation elements. We can't do that in JDK 1.1. Insted, |
1527 | // we have to go through this horrible slow mess: |
1528 | int32_t p = startingAt; |
1529 | int32_t keyLen = 0; |
1530 | |
1531 | // basically just isolate smaller and smaller substrings of |
1532 | // the target string (each running to the end of the string, |
1533 | // and with the first one running from startingAt to the end) |
1534 | // and then use prefixLength() to see if the search key is at |
1535 | // the beginning of each substring. This is excruciatingly |
1536 | // slow, but it will locate the key and tell use how long the |
1537 | // matching text was. |
1538 | UnicodeString temp; |
1539 | UErrorCode status = U_ZERO_ERROR; |
1540 | while (p < str.length() && keyLen == 0) { |
1541 | temp.setTo(str, p, str.length() - p); |
1542 | keyLen = prefixLength(temp, key, status); |
1543 | if (U_FAILURE(status)) { |
1544 | break; |
1545 | } |
1546 | if (keyLen != 0) { |
1547 | *length = keyLen; |
1548 | return p; |
1549 | } |
1550 | ++p; |
1551 | } |
1552 | // if we make it to here, we didn't find it. Return -1 for the |
1553 | // location. The length should be ignored, but set it to 0, |
1554 | // which should be "safe" |
1555 | *length = 0; |
1556 | return -1; |
1557 | } |
1558 | |
1559 | /** |
1560 | * Checks to see whether a string consists entirely of ignorable |
1561 | * characters. |
1562 | * @param str The string to test. |
1563 | * @return true if the string is empty of consists entirely of |
1564 | * characters that the number formatter's collator says are |
1565 | * ignorable at the primary-order level. false otherwise. |
1566 | */ |
1567 | UBool |
1568 | NFRule::allIgnorable(const UnicodeString& str, UErrorCode& status) const |
1569 | { |
1570 | // if the string is empty, we can just return true |
1571 | if (str.length() == 0) { |
1572 | return TRUE; |
1573 | } |
1574 | |
1575 | #if !UCONFIG_NO_COLLATION |
1576 | // if lenient parsing is turned on, walk through the string with |
1577 | // a collation element iterator and make sure each collation |
1578 | // element is 0 (ignorable) at the primary level |
1579 | if (formatter->isLenient()) { |
1580 | const RuleBasedCollator* collator = formatter->getCollator(); |
1581 | if (collator == NULL) { |
1582 | status = U_MEMORY_ALLOCATION_ERROR; |
1583 | return FALSE; |
1584 | } |
1585 | LocalPointer<CollationElementIterator> iter(collator->createCollationElementIterator(str)); |
1586 | |
1587 | // Memory allocation error check. |
1588 | if (iter.isNull()) { |
1589 | status = U_MEMORY_ALLOCATION_ERROR; |
1590 | return FALSE; |
1591 | } |
1592 | |
1593 | UErrorCode err = U_ZERO_ERROR; |
1594 | int32_t o = iter->next(err); |
1595 | while (o != CollationElementIterator::NULLORDER |
1596 | && CollationElementIterator::primaryOrder(o) == 0) { |
1597 | o = iter->next(err); |
1598 | } |
1599 | |
1600 | return o == CollationElementIterator::NULLORDER; |
1601 | } |
1602 | #endif |
1603 | |
1604 | // if lenient parsing is turned off, there is no such thing as |
1605 | // an ignorable character: return true only if the string is empty |
1606 | return FALSE; |
1607 | } |
1608 | |
1609 | void |
1610 | NFRule::setDecimalFormatSymbols(const DecimalFormatSymbols& newSymbols, UErrorCode& status) { |
1611 | if (sub1 != NULL) { |
1612 | sub1->setDecimalFormatSymbols(newSymbols, status); |
1613 | } |
1614 | if (sub2 != NULL) { |
1615 | sub2->setDecimalFormatSymbols(newSymbols, status); |
1616 | } |
1617 | } |
1618 | |
1619 | U_NAMESPACE_END |
1620 | |
1621 | /* U_HAVE_RBNF */ |
1622 | #endif |
1623 | |