2 ******************************************************************************
3 * Copyright (C) 1997-2011, International Business Machines
4 * Corporation and others. All Rights Reserved.
5 ******************************************************************************
6 * file name: nfrule.cpp
8 * tab size: 8 (not used)
11 * Modification history
13 * 10/11/2001 Doug Ported from ICU4J
20 #include "unicode/rbnf.h"
21 #include "unicode/tblcoll.h"
22 #include "unicode/coleitr.h"
23 #include "unicode/uchar.h"
27 #include "patternprops.h"
31 NFRule::NFRule(const RuleBasedNumberFormat* _rbnf)
32 : baseValue((int32_t)0)
48 static const UChar gLeftBracket = 0x005b;
49 static const UChar gRightBracket = 0x005d;
50 static const UChar gColon = 0x003a;
51 static const UChar gZero = 0x0030;
52 static const UChar gNine = 0x0039;
53 static const UChar gSpace = 0x0020;
54 static const UChar gSlash = 0x002f;
55 static const UChar gGreaterThan = 0x003e;
56 static const UChar gLessThan = 0x003c;
57 static const UChar gComma = 0x002c;
58 static const UChar gDot = 0x002e;
59 static const UChar gTick = 0x0027;
60 //static const UChar gMinus = 0x002d;
61 static const UChar gSemicolon = 0x003b;
63 static const UChar gMinusX[] = {0x2D, 0x78, 0}; /* "-x" */
64 static const UChar gXDotX[] = {0x78, 0x2E, 0x78, 0}; /* "x.x" */
65 static const UChar gXDotZero[] = {0x78, 0x2E, 0x30, 0}; /* "x.0" */
66 static const UChar gZeroDotX[] = {0x30, 0x2E, 0x78, 0}; /* "0.x" */
68 static const UChar gLessLess[] = {0x3C, 0x3C, 0}; /* "<<" */
69 static const UChar gLessPercent[] = {0x3C, 0x25, 0}; /* "<%" */
70 static const UChar gLessHash[] = {0x3C, 0x23, 0}; /* "<#" */
71 static const UChar gLessZero[] = {0x3C, 0x30, 0}; /* "<0" */
72 static const UChar gGreaterGreater[] = {0x3E, 0x3E, 0}; /* ">>" */
73 static const UChar gGreaterPercent[] = {0x3E, 0x25, 0}; /* ">%" */
74 static const UChar gGreaterHash[] = {0x3E, 0x23, 0}; /* ">#" */
75 static const UChar gGreaterZero[] = {0x3E, 0x30, 0}; /* ">0" */
76 static const UChar gEqualPercent[] = {0x3D, 0x25, 0}; /* "=%" */
77 static const UChar gEqualHash[] = {0x3D, 0x23, 0}; /* "=#" */
78 static const UChar gEqualZero[] = {0x3D, 0x30, 0}; /* "=0" */
79 static const UChar gGreaterGreaterGreater[] = {0x3E, 0x3E, 0x3E, 0}; /* ">>>" */
81 static const UChar * const tokenStrings[] = {
82 gLessLess, gLessPercent, gLessHash, gLessZero,
83 gGreaterGreater, gGreaterPercent,gGreaterHash, gGreaterZero,
84 gEqualPercent, gEqualHash, gEqualZero, NULL
88 NFRule::makeRules(UnicodeString& description,
89 const NFRuleSet *ruleSet,
90 const NFRule *predecessor,
91 const RuleBasedNumberFormat *rbnf,
95 // we know we're making at least one rule, so go ahead and
96 // new it up and initialize its basevalue and divisor
97 // (this also strips the rule descriptor, if any, off the
99 NFRule* rule1 = new NFRule(rbnf);
102 status = U_MEMORY_ALLOCATION_ERROR;
105 rule1->parseRuleDescriptor(description, status);
107 // check the description to see whether there's text enclosed
109 int32_t brack1 = description.indexOf(gLeftBracket);
110 int32_t brack2 = description.indexOf(gRightBracket);
112 // if the description doesn't contain a matched pair of brackets,
113 // or if it's of a type that doesn't recognize bracketed text,
114 // then leave the description alone, initialize the rule's
115 // rule text and substitutions, and return that rule
116 if (brack1 == -1 || brack2 == -1 || brack1 > brack2
117 || rule1->getType() == kProperFractionRule
118 || rule1->getType() == kNegativeNumberRule) {
119 rule1->ruleText = description;
120 rule1->extractSubstitutions(ruleSet, predecessor, rbnf, status);
123 // if the description does contain a matched pair of brackets,
124 // then it's really shorthand for two rules (with one exception)
125 NFRule* rule2 = NULL;
128 // we'll actually only split the rule into two rules if its
129 // base value is an even multiple of its divisor (or it's one
130 // of the special rules)
131 if ((rule1->baseValue > 0
132 && (rule1->baseValue % util64_pow(rule1->radix, rule1->exponent)) == 0)
133 || rule1->getType() == kImproperFractionRule
134 || rule1->getType() == kMasterRule) {
136 // if it passes that test, new up the second rule. If the
137 // rule set both rules will belong to is a fraction rule
138 // set, they both have the same base value; otherwise,
139 // increment the original rule's base value ("rule1" actually
140 // goes SECOND in the rule set's rule list)
141 rule2 = new NFRule(rbnf);
144 status = U_MEMORY_ALLOCATION_ERROR;
147 if (rule1->baseValue >= 0) {
148 rule2->baseValue = rule1->baseValue;
149 if (!ruleSet->isFractionRuleSet()) {
154 // if the description began with "x.x" and contains bracketed
155 // text, it describes both the improper fraction rule and
156 // the proper fraction rule
157 else if (rule1->getType() == kImproperFractionRule) {
158 rule2->setType(kProperFractionRule);
161 // if the description began with "x.0" and contains bracketed
162 // text, it describes both the master rule and the
163 // improper fraction rule
164 else if (rule1->getType() == kMasterRule) {
165 rule2->baseValue = rule1->baseValue;
166 rule1->setType(kImproperFractionRule);
169 // both rules have the same radix and exponent (i.e., the
171 rule2->radix = rule1->radix;
172 rule2->exponent = rule1->exponent;
174 // rule2's rule text omits the stuff in brackets: initalize
175 // its rule text and substitutions accordingly
176 sbuf.append(description, 0, brack1);
177 if (brack2 + 1 < description.length()) {
178 sbuf.append(description, brack2 + 1, description.length() - brack2 - 1);
180 rule2->ruleText.setTo(sbuf);
181 rule2->extractSubstitutions(ruleSet, predecessor, rbnf, status);
184 // rule1's text includes the text in the brackets but omits
185 // the brackets themselves: initialize _its_ rule text and
186 // substitutions accordingly
187 sbuf.setTo(description, 0, brack1);
188 sbuf.append(description, brack1 + 1, brack2 - brack1 - 1);
189 if (brack2 + 1 < description.length()) {
190 sbuf.append(description, brack2 + 1, description.length() - brack2 - 1);
192 rule1->ruleText.setTo(sbuf);
193 rule1->extractSubstitutions(ruleSet, predecessor, rbnf, status);
195 // if we only have one rule, return it; if we have two, return
196 // a two-element array containing them (notice that rule2 goes
197 // BEFORE rule1 in the list: in all cases, rule2 OMITS the
198 // material in the brackets and rule1 INCLUDES the material
208 * This function parses the rule's rule descriptor (i.e., the base
209 * value and/or other tokens that precede the rule's rule text
210 * in the description) and sets the rule's base value, radix, and
211 * exponent according to the descriptor. (If the description doesn't
212 * include a rule descriptor, then this function sets everything to
213 * default values and the rule set sets the rule's real base value).
214 * @param description The rule's description
215 * @return If "description" included a rule descriptor, this is
216 * "description" with the descriptor and any trailing whitespace
217 * stripped off. Otherwise; it's "descriptor" unchangd.
220 NFRule::parseRuleDescriptor(UnicodeString& description, UErrorCode& status)
222 // the description consists of a rule descriptor and a rule body,
223 // separated by a colon. The rule descriptor is optional. If
224 // it's omitted, just set the base value to 0.
225 int32_t p = description.indexOf(gColon);
227 setBaseValue((int32_t)0, status);
229 // copy the descriptor out into its own string and strip it,
230 // along with any trailing whitespace, out of the original
232 UnicodeString descriptor;
233 descriptor.setTo(description, 0, p);
236 while (p < description.length() && PatternProps::isWhiteSpace(description.charAt(p))) {
239 description.removeBetween(0, p);
241 // check first to see if the rule descriptor matches the token
242 // for one of the special rules. If it does, set the base
243 // value to the correct identfier value
244 if (0 == descriptor.compare(gMinusX, 2)) {
245 setType(kNegativeNumberRule);
247 else if (0 == descriptor.compare(gXDotX, 3)) {
248 setType(kImproperFractionRule);
250 else if (0 == descriptor.compare(gZeroDotX, 3)) {
251 setType(kProperFractionRule);
253 else if (0 == descriptor.compare(gXDotZero, 3)) {
254 setType(kMasterRule);
257 // if the rule descriptor begins with a digit, it's a descriptor
259 // since we don't have Long.parseLong, and this isn't much work anyway,
260 // just build up the value as we encounter the digits.
261 else if (descriptor.charAt(0) >= gZero && descriptor.charAt(0) <= gNine) {
266 // begin parsing the descriptor: copy digits
267 // into "tempValue", skip periods, commas, and spaces,
268 // stop on a slash or > sign (or at the end of the string),
269 // and throw an exception on any other character
271 while (p < descriptor.length()) {
272 c = descriptor.charAt(p);
273 if (c >= gZero && c <= gNine) {
274 val = val * ll_10 + (int32_t)(c - gZero);
276 else if (c == gSlash || c == gGreaterThan) {
279 else if (PatternProps::isWhiteSpace(c) || c == gComma || c == gDot) {
282 // throw new IllegalArgumentException("Illegal character in rule descriptor");
283 status = U_PARSE_ERROR;
289 // we have the base value, so set it
290 setBaseValue(val, status);
292 // if we stopped the previous loop on a slash, we're
293 // now parsing the rule's radix. Again, accumulate digits
294 // in tempValue, skip punctuation, stop on a > mark, and
295 // throw an exception on anything else
300 while (p < descriptor.length()) {
301 c = descriptor.charAt(p);
302 if (c >= gZero && c <= gNine) {
303 val = val * ll_10 + (int32_t)(c - gZero);
305 else if (c == gGreaterThan) {
308 else if (PatternProps::isWhiteSpace(c) || c == gComma || c == gDot) {
311 // throw new IllegalArgumentException("Illegal character is rule descriptor");
312 status = U_PARSE_ERROR;
318 // tempValue now contain's the rule's radix. Set it
319 // accordingly, and recalculate the rule's exponent
320 radix = (int32_t)val;
322 // throw new IllegalArgumentException("Rule can't have radix of 0");
323 status = U_PARSE_ERROR;
326 exponent = expectedExponent();
329 // if we stopped the previous loop on a > sign, then continue
330 // for as long as we still see > signs. For each one,
331 // decrement the exponent (unless the exponent is already 0).
332 // If we see another character before reaching the end of
333 // the descriptor, that's also a syntax error.
334 if (c == gGreaterThan) {
335 while (p < descriptor.length()) {
336 c = descriptor.charAt(p);
337 if (c == gGreaterThan && exponent > 0) {
340 // throw new IllegalArgumentException("Illegal character in rule descriptor");
341 status = U_PARSE_ERROR;
350 // finally, if the rule body begins with an apostrophe, strip it off
351 // (this is generally used to put whitespace at the beginning of
352 // a rule's rule text)
353 if (description.length() > 0 && description.charAt(0) == gTick) {
354 description.removeBetween(0, 1);
357 // return the description with all the stuff we've just waded through
358 // stripped off the front. It now contains just the rule body.
359 // return description;
363 * Searches the rule's rule text for the substitution tokens,
364 * creates the substitutions, and removes the substitution tokens
365 * from the rule's rule text.
366 * @param owner The rule set containing this rule
367 * @param predecessor The rule preseding this one in "owners" rule list
368 * @param ownersOwner The RuleBasedFormat that owns this rule
371 NFRule::extractSubstitutions(const NFRuleSet* ruleSet,
372 const NFRule* predecessor,
373 const RuleBasedNumberFormat* rbnf,
376 if (U_SUCCESS(status)) {
377 sub1 = extractSubstitution(ruleSet, predecessor, rbnf, status);
378 sub2 = extractSubstitution(ruleSet, predecessor, rbnf, status);
383 * Searches the rule's rule text for the first substitution token,
384 * creates a substitution based on it, and removes the token from
385 * the rule's rule text.
386 * @param owner The rule set containing this rule
387 * @param predecessor The rule preceding this one in the rule set's
389 * @param ownersOwner The RuleBasedNumberFormat that owns this rule
390 * @return The newly-created substitution. This is never null; if
391 * the rule text doesn't contain any substitution tokens, this will
392 * be a NullSubstitution.
395 NFRule::extractSubstitution(const NFRuleSet* ruleSet,
396 const NFRule* predecessor,
397 const RuleBasedNumberFormat* rbnf,
400 NFSubstitution* result = NULL;
402 // search the rule's rule text for the first two characters of
403 // a substitution token
404 int32_t subStart = indexOfAny(tokenStrings);
405 int32_t subEnd = subStart;
407 // if we didn't find one, create a null substitution positioned
408 // at the end of the rule text
409 if (subStart == -1) {
410 return NFSubstitution::makeSubstitution(ruleText.length(), this, predecessor,
411 ruleSet, rbnf, UnicodeString(), status);
414 // special-case the ">>>" token, since searching for the > at the
415 // end will actually find the > in the middle
416 if (ruleText.indexOf(gGreaterGreaterGreater, 3, 0) == subStart) {
417 subEnd = subStart + 2;
419 // otherwise the substitution token ends with the same character
422 UChar c = ruleText.charAt(subStart);
423 subEnd = ruleText.indexOf(c, subStart + 1);
424 // special case for '<%foo<<'
425 if (c == gLessThan && subEnd != -1 && subEnd < ruleText.length() - 1 && ruleText.charAt(subEnd+1) == c) {
426 // ordinals use "=#,##0==%abbrev=" as their rule. Notice that the '==' in the middle
427 // occurs because of the juxtaposition of two different rules. The check for '<' is a hack
428 // to get around this. Having the duplicate at the front would cause problems with
429 // rules like "<<%" to format, say, percents...
434 // if we don't find the end of the token (i.e., if we're on a single,
435 // unmatched token character), create a null substitution positioned
436 // at the end of the rule
438 return NFSubstitution::makeSubstitution(ruleText.length(), this, predecessor,
439 ruleSet, rbnf, UnicodeString(), status);
442 // if we get here, we have a real substitution token (or at least
443 // some text bounded by substitution token characters). Use
444 // makeSubstitution() to create the right kind of substitution
445 UnicodeString subToken;
446 subToken.setTo(ruleText, subStart, subEnd + 1 - subStart);
447 result = NFSubstitution::makeSubstitution(subStart, this, predecessor, ruleSet,
448 rbnf, subToken, status);
450 // remove the substitution from the rule text
451 ruleText.removeBetween(subStart, subEnd+1);
457 * Sets the rule's base value, and causes the radix and exponent
458 * to be recalculated. This is used during construction when we
459 * don't know the rule's base value until after it's been
460 * constructed. It should be used at any other time.
461 * @param The new base value for the rule.
464 NFRule::setBaseValue(int64_t newBaseValue, UErrorCode& status)
466 // set the base value
467 baseValue = newBaseValue;
469 // if this isn't a special rule, recalculate the radix and exponent
470 // (the radix always defaults to 10; if it's supposed to be something
471 // else, it's cleaned up by the caller and the exponent is
472 // recalculated again-- the only function that does this is
473 // NFRule.parseRuleDescriptor() )
474 if (baseValue >= 1) {
476 exponent = expectedExponent();
478 // this function gets called on a fully-constructed rule whose
479 // description didn't specify a base value. This means it
480 // has substitutions, and some substitutions hold on to copies
481 // of the rule's divisor. Fix their copies of the divisor.
483 sub1->setDivisor(radix, exponent, status);
486 sub2->setDivisor(radix, exponent, status);
489 // if this is a special rule, its radix and exponent are basically
490 // ignored. Set them to "safe" default values
498 * This calculates the rule's exponent based on its radix and base
499 * value. This will be the highest power the radix can be raised to
500 * and still produce a result less than or equal to the base value.
503 NFRule::expectedExponent() const
505 // since the log of 0, or the log base 0 of something, causes an
506 // error, declare the exponent in these cases to be 0 (we also
507 // deal with the special-rule identifiers here)
508 if (radix == 0 || baseValue < 1) {
512 // we get rounding error in some cases-- for example, log 1000 / log 10
513 // gives us 1.9999999996 instead of 2. The extra logic here is to take
515 int16_t tempResult = (int16_t)(uprv_log((double)baseValue) / uprv_log((double)radix));
516 int64_t temp = util64_pow(radix, tempResult + 1);
517 if (temp <= baseValue) {
524 * Searches the rule's rule text for any of the specified strings.
525 * @param strings An array of strings to search the rule's rule
527 * @return The index of the first match in the rule's rule text
528 * (i.e., the first substring in the rule's rule text that matches
529 * _any_ of the strings in "strings"). If none of the strings in
530 * "strings" is found in the rule's rule text, returns -1.
533 NFRule::indexOfAny(const UChar* const strings[]) const
536 for (int i = 0; strings[i]; i++) {
537 int32_t pos = ruleText.indexOf(*strings[i]);
538 if (pos != -1 && (result == -1 || pos < result)) {
545 //-----------------------------------------------------------------------
547 //-----------------------------------------------------------------------
550 * Tests two rules for equality.
551 * @param that The rule to compare this one against
552 * @return True is the two rules are functionally equivalent
555 NFRule::operator==(const NFRule& rhs) const
557 return baseValue == rhs.baseValue
558 && radix == rhs.radix
559 && exponent == rhs.exponent
560 && ruleText == rhs.ruleText
561 && *sub1 == *rhs.sub1
562 && *sub2 == *rhs.sub2;
566 * Returns a textual representation of the rule. This won't
567 * necessarily be the same as the description that this rule
568 * was created with, but it will produce the same result.
569 * @return A textual description of the rule
571 static void util_append64(UnicodeString& result, int64_t n)
574 int32_t len = util64_tou(n, buffer, sizeof(buffer));
575 UnicodeString temp(buffer, len);
580 NFRule::_appendRuleText(UnicodeString& result) const
583 case kNegativeNumberRule: result.append(gMinusX, 2); break;
584 case kImproperFractionRule: result.append(gXDotX, 3); break;
585 case kProperFractionRule: result.append(gZeroDotX, 3); break;
586 case kMasterRule: result.append(gXDotZero, 3); break;
588 // for a normal rule, write out its base value, and if the radix is
589 // something other than 10, write out the radix (with the preceding
590 // slash, of course). Then calculate the expected exponent and if
591 // if isn't the same as the actual exponent, write an appropriate
592 // number of > signs. Finally, terminate the whole thing with
594 util_append64(result, baseValue);
596 result.append(gSlash);
597 util_append64(result, radix);
599 int numCarets = expectedExponent() - exponent;
600 for (int i = 0; i < numCarets; i++) {
601 result.append(gGreaterThan);
605 result.append(gColon);
606 result.append(gSpace);
608 // if the rule text begins with a space, write an apostrophe
609 // (whitespace after the rule descriptor is ignored; the
610 // apostrophe is used to make the whitespace significant)
611 if (ruleText.charAt(0) == gSpace && sub1->getPos() != 0) {
612 result.append(gTick);
615 // now, write the rule's rule text, inserting appropriate
616 // substitution tokens in the appropriate places
617 UnicodeString ruleTextCopy;
618 ruleTextCopy.setTo(ruleText);
621 sub2->toString(temp);
622 ruleTextCopy.insert(sub2->getPos(), temp);
623 sub1->toString(temp);
624 ruleTextCopy.insert(sub1->getPos(), temp);
626 result.append(ruleTextCopy);
628 // and finally, top the whole thing off with a semicolon and
630 result.append(gSemicolon);
633 //-----------------------------------------------------------------------
635 //-----------------------------------------------------------------------
638 * Formats the number, and inserts the resulting text into
640 * @param number The number being formatted
641 * @param toInsertInto The string where the resultant text should
643 * @param pos The position in toInsertInto where the resultant text
647 NFRule::doFormat(int64_t number, UnicodeString& toInsertInto, int32_t pos) const
649 // first, insert the rule's rule text into toInsertInto at the
650 // specified position, then insert the results of the substitutions
651 // into the right places in toInsertInto (notice we do the
652 // substitutions in reverse order so that the offsets don't get
654 toInsertInto.insert(pos, ruleText);
655 sub2->doSubstitution(number, toInsertInto, pos);
656 sub1->doSubstitution(number, toInsertInto, pos);
660 * Formats the number, and inserts the resulting text into
662 * @param number The number being formatted
663 * @param toInsertInto The string where the resultant text should
665 * @param pos The position in toInsertInto where the resultant text
669 NFRule::doFormat(double number, UnicodeString& toInsertInto, int32_t pos) const
671 // first, insert the rule's rule text into toInsertInto at the
672 // specified position, then insert the results of the substitutions
673 // into the right places in toInsertInto
674 // [again, we have two copies of this routine that do the same thing
675 // so that we don't sacrifice precision in a long by casting it
677 toInsertInto.insert(pos, ruleText);
678 sub2->doSubstitution(number, toInsertInto, pos);
679 sub1->doSubstitution(number, toInsertInto, pos);
683 * Used by the owning rule set to determine whether to invoke the
684 * rollback rule (i.e., whether this rule or the one that precedes
685 * it in the rule set's list should be used to format the number)
686 * @param The number being formatted
687 * @return True if the rule set should use the rule that precedes
688 * this one in its list; false if it should use this rule
691 NFRule::shouldRollBack(double number) const
693 // we roll back if the rule contains a modulus substitution,
694 // the number being formatted is an even multiple of the rule's
695 // divisor, and the rule's base value is NOT an even multiple
697 // In other words, if the original description had
698 // 100: << hundred[ >>];
701 // 101: << hundred >>;
702 // internally. But when we're formatting 200, if we use the rule
703 // at 101, which would normally apply, we get "two hundred zero".
704 // To prevent this, we roll back and use the rule at 100 instead.
705 // This is the logic that makes this happen: the rule at 101 has
706 // a modulus substitution, its base value isn't an even multiple
707 // of 100, and the value we're trying to format _is_ an even
708 // multiple of 100. This is called the "rollback rule."
709 if ((sub1->isModulusSubstitution()) || (sub2->isModulusSubstitution())) {
710 int64_t re = util64_pow(radix, exponent);
711 return uprv_fmod(number, (double)re) == 0 && (baseValue % re) != 0;
716 //-----------------------------------------------------------------------
718 //-----------------------------------------------------------------------
721 * Attempts to parse the string with this rule.
722 * @param text The string being parsed
723 * @param parsePosition On entry, the value is ignored and assumed to
724 * be 0. On exit, this has been updated with the position of the first
725 * character not consumed by matching the text against this rule
726 * (if this rule doesn't match the text at all, the parse position
727 * if left unchanged (presumably at 0) and the function returns
729 * @param isFractionRule True if this rule is contained within a
730 * fraction rule set. This is only used if the rule has no
732 * @return If this rule matched the text, this is the rule's base value
733 * combined appropriately with the results of parsing the substitutions.
734 * If nothing matched, this is new Long(0) and the parse position is
735 * left unchanged. The result will be an instance of Long if the
736 * result is an integer and Double otherwise. The result is never null.
741 static void dumpUS(FILE* f, const UnicodeString& us) {
742 int len = us.length();
743 char* buf = (char *)uprv_malloc((len+1)*sizeof(char)); //new char[len+1];
745 us.extract(0, len, buf);
747 fprintf(f, "%s", buf);
748 uprv_free(buf); //delete[] buf;
754 NFRule::doParse(const UnicodeString& text,
755 ParsePosition& parsePosition,
756 UBool isFractionRule,
758 Formattable& resVal) const
760 // internally we operate on a copy of the string being parsed
761 // (because we're going to change it) and use our own ParsePosition
763 UnicodeString workText(text);
765 // check to see whether the text before the first substitution
766 // matches the text at the beginning of the string being
767 // parsed. If it does, strip that off the front of workText;
768 // otherwise, dump out with a mismatch
769 UnicodeString prefix;
770 prefix.setTo(ruleText, 0, sub1->getPos());
773 fprintf(stderr, "doParse %x ", this);
780 fprintf(stderr, " text: '", this);
781 dumpUS(stderr, text);
782 fprintf(stderr, "' prefix: '");
783 dumpUS(stderr, prefix);
785 stripPrefix(workText, prefix, pp);
786 int32_t prefixLength = text.length() - workText.length();
789 fprintf(stderr, "' pl: %d ppi: %d s1p: %d\n", prefixLength, pp.getIndex(), sub1->getPos());
792 if (pp.getIndex() == 0 && sub1->getPos() != 0) {
793 // commented out because ParsePosition doesn't have error index in 1.1.x
794 // restored for ICU4C port
795 parsePosition.setErrorIndex(pp.getErrorIndex());
800 // this is the fun part. The basic guts of the rule-matching
801 // logic is matchToDelimiter(), which is called twice. The first
802 // time it searches the input string for the rule text BETWEEN
803 // the substitutions and tries to match the intervening text
804 // in the input string with the first substitution. If that
805 // succeeds, it then calls it again, this time to look for the
806 // rule text after the second substitution and to match the
807 // intervening input text against the second substitution.
809 // For example, say we have a rule that looks like this:
810 // first << middle >> last;
811 // and input text that looks like this:
812 // first one middle two last
813 // First we use stripPrefix() to match "first " in both places and
814 // strip it off the front, leaving
815 // one middle two last
816 // Then we use matchToDelimiter() to match " middle " and try to
817 // match "one" against a substitution. If it's successful, we now
820 // We use matchToDelimiter() a second time to match " last" and
821 // try to match "two" against a substitution. If "two" matches
822 // the substitution, we have a successful parse.
824 // Since it's possible in many cases to find multiple instances
825 // of each of these pieces of rule text in the input string,
826 // we need to try all the possible combinations of these
827 // locations. This prevents us from prematurely declaring a mismatch,
828 // and makes sure we match as much input text as we can.
829 int highWaterMark = 0;
832 double tempBaseValue = (double)(baseValue <= 0 ? 0 : baseValue);
836 // our partial parse result starts out as this rule's base
837 // value. If it finds a successful match, matchToDelimiter()
838 // will compose this in some way with what it gets back from
839 // the substitution, giving us a new partial parse result
842 temp.setTo(ruleText, sub1->getPos(), sub2->getPos() - sub1->getPos());
843 double partialResult = matchToDelimiter(workText, start, tempBaseValue,
847 // if we got a successful match (or were trying to match a
848 // null substitution), pp is now pointing at the first unmatched
849 // character. Take note of that, and try matchToDelimiter()
850 // on the input text again
851 if (pp.getIndex() != 0 || sub1->isNullSubstitution()) {
852 start = pp.getIndex();
854 UnicodeString workText2;
855 workText2.setTo(workText, pp.getIndex(), workText.length() - pp.getIndex());
858 // the second matchToDelimiter() will compose our previous
859 // partial result with whatever it gets back from its
860 // substitution if there's a successful match, giving us
862 temp.setTo(ruleText, sub2->getPos(), ruleText.length() - sub2->getPos());
863 partialResult = matchToDelimiter(workText2, 0, partialResult,
867 // if we got a successful match on this second
868 // matchToDelimiter() call, update the high-water mark
869 // and result (if necessary)
870 if (pp2.getIndex() != 0 || sub2->isNullSubstitution()) {
871 if (prefixLength + pp.getIndex() + pp2.getIndex() > highWaterMark) {
872 highWaterMark = prefixLength + pp.getIndex() + pp2.getIndex();
873 result = partialResult;
876 // commented out because ParsePosition doesn't have error index in 1.1.x
877 // restored for ICU4C port
879 int32_t temp = pp2.getErrorIndex() + sub1->getPos() + pp.getIndex();
880 if (temp> parsePosition.getErrorIndex()) {
881 parsePosition.setErrorIndex(temp);
885 // commented out because ParsePosition doesn't have error index in 1.1.x
886 // restored for ICU4C port
888 int32_t temp = sub1->getPos() + pp.getErrorIndex();
889 if (temp > parsePosition.getErrorIndex()) {
890 parsePosition.setErrorIndex(temp);
893 // keep trying to match things until the outer matchToDelimiter()
894 // call fails to make a match (each time, it picks up where it
895 // left off the previous time)
896 } while (sub1->getPos() != sub2->getPos()
898 && pp.getIndex() < workText.length()
899 && pp.getIndex() != start);
901 // update the caller's ParsePosition with our high-water mark
902 // (i.e., it now points at the first character this function
903 // didn't match-- the ParsePosition is therefore unchanged if
904 // we didn't match anything)
905 parsePosition.setIndex(highWaterMark);
906 // commented out because ParsePosition doesn't have error index in 1.1.x
907 // restored for ICU4C port
908 if (highWaterMark > 0) {
909 parsePosition.setErrorIndex(0);
912 // this is a hack for one unusual condition: Normally, whether this
913 // rule belong to a fraction rule set or not is handled by its
914 // substitutions. But if that rule HAS NO substitutions, then
915 // we have to account for it here. By definition, if the matching
916 // rule in a fraction rule set has no substitutions, its numerator
917 // is 1, and so the result is the reciprocal of its base value.
918 if (isFractionRule &&
920 sub1->isNullSubstitution()) {
924 resVal.setDouble(result);
925 return TRUE; // ??? do we need to worry if it is a long or a double?
929 * This function is used by parse() to match the text being parsed
930 * against a possible prefix string. This function
931 * matches characters from the beginning of the string being parsed
932 * to characters from the prospective prefix. If they match, pp is
933 * updated to the first character not matched, and the result is
934 * the unparsed part of the string. If they don't match, the whole
935 * string is returned, and pp is left unchanged.
936 * @param text The string being parsed
937 * @param prefix The text to match against
938 * @param pp On entry, ignored and assumed to be 0. On exit, points
939 * to the first unmatched character (assuming the whole prefix matched),
940 * or is unchanged (if the whole prefix didn't match).
941 * @return If things match, this is the unparsed part of "text";
942 * if they didn't match, this is "text".
945 NFRule::stripPrefix(UnicodeString& text, const UnicodeString& prefix, ParsePosition& pp) const
947 // if the prefix text is empty, dump out without doing anything
948 if (prefix.length() != 0) {
949 UErrorCode status = U_ZERO_ERROR;
950 // use prefixLength() to match the beginning of
951 // "text" against "prefix". This function returns the
952 // number of characters from "text" that matched (or 0 if
953 // we didn't match the whole prefix)
954 int32_t pfl = prefixLength(text, prefix, status);
955 if (U_FAILURE(status)) { // Memory allocation error.
959 // if we got a successful match, update the parse position
960 // and strip the prefix off of "text"
961 pp.setIndex(pp.getIndex() + pfl);
968 * Used by parse() to match a substitution and any following text.
969 * "text" is searched for instances of "delimiter". For each instance
970 * of delimiter, the intervening text is tested to see whether it
971 * matches the substitution. The longest match wins.
972 * @param text The string being parsed
973 * @param startPos The position in "text" where we should start looking
975 * @param baseValue A partial parse result (often the rule's base value),
976 * which is combined with the result from matching the substitution
977 * @param delimiter The string to search "text" for.
978 * @param pp Ignored and presumed to be 0 on entry. If there's a match,
979 * on exit this will point to the first unmatched character.
980 * @param sub If we find "delimiter" in "text", this substitution is used
981 * to match the text between the beginning of the string and the
982 * position of "delimiter." (If "delimiter" is the empty string, then
983 * this function just matches against this substitution and updates
984 * everything accordingly.)
985 * @param upperBound When matching the substitution, it will only
986 * consider rules with base values lower than this value.
987 * @return If there's a match, this is the result of composing
988 * baseValue with the result of matching the substitution. Otherwise,
989 * this is new Long(0). It's never null. If the result is an integer,
990 * this will be an instance of Long; otherwise, it's an instance of
993 * !!! note {dlf} in point of fact, in the java code the caller always converts
994 * the result to a double, so we might as well return one.
997 NFRule::matchToDelimiter(const UnicodeString& text,
1000 const UnicodeString& delimiter,
1002 const NFSubstitution* sub,
1003 double upperBound) const
1005 UErrorCode status = U_ZERO_ERROR;
1006 // if "delimiter" contains real (i.e., non-ignorable) text, search
1007 // it for "delimiter" beginning at "start". If that succeeds, then
1008 // use "sub"'s doParse() method to match the text before the
1009 // instance of "delimiter" we just found.
1010 if (!allIgnorable(delimiter, status)) {
1011 if (U_FAILURE(status)) { //Memory allocation error.
1014 ParsePosition tempPP;
1017 // use findText() to search for "delimiter". It returns a two-
1018 // element array: element 0 is the position of the match, and
1019 // element 1 is the number of characters that matched
1022 int32_t dPos = findText(text, delimiter, startPos, &dLen);
1024 // if findText() succeeded, isolate the text preceding the
1025 // match, and use "sub" to match that text
1027 UnicodeString subText;
1028 subText.setTo(text, 0, dPos);
1029 if (subText.length() > 0) {
1030 UBool success = sub->doParse(subText, tempPP, _baseValue, upperBound,
1031 #if UCONFIG_NO_COLLATION
1034 formatter->isLenient(),
1038 // if the substitution could match all the text up to
1039 // where we found "delimiter", then this function has
1040 // a successful match. Bump the caller's parse position
1041 // to point to the first character after the text
1042 // that matches "delimiter", and return the result
1043 // we got from parsing the substitution.
1044 if (success && tempPP.getIndex() == dPos) {
1045 pp.setIndex(dPos + dLen);
1046 return result.getDouble();
1048 // commented out because ParsePosition doesn't have error index in 1.1.x
1049 // restored for ICU4C port
1051 if (tempPP.getErrorIndex() > 0) {
1052 pp.setErrorIndex(tempPP.getErrorIndex());
1054 pp.setErrorIndex(tempPP.getIndex());
1059 // if we didn't match the substitution, search for another
1060 // copy of "delimiter" in "text" and repeat the loop if
1063 dPos = findText(text, delimiter, dPos + dLen, &dLen);
1065 // if we make it here, this was an unsuccessful match, and we
1066 // leave pp unchanged and return 0
1070 // if "delimiter" is empty, or consists only of ignorable characters
1071 // (i.e., is semantically empty), thwe we obviously can't search
1072 // for "delimiter". Instead, just use "sub" to parse as much of
1073 // "text" as possible.
1075 ParsePosition tempPP;
1078 // try to match the whole string against the substitution
1079 UBool success = sub->doParse(text, tempPP, _baseValue, upperBound,
1080 #if UCONFIG_NO_COLLATION
1083 formatter->isLenient(),
1086 if (success && (tempPP.getIndex() != 0 || sub->isNullSubstitution())) {
1087 // if there's a successful match (or it's a null
1088 // substitution), update pp to point to the first
1089 // character we didn't match, and pass the result from
1090 // sub.doParse() on through to the caller
1091 pp.setIndex(tempPP.getIndex());
1092 return result.getDouble();
1094 // commented out because ParsePosition doesn't have error index in 1.1.x
1095 // restored for ICU4C port
1097 pp.setErrorIndex(tempPP.getErrorIndex());
1100 // and if we get to here, then nothing matched, so we return
1101 // 0 and leave pp alone
1107 * Used by stripPrefix() to match characters. If lenient parse mode
1108 * is off, this just calls startsWith(). If lenient parse mode is on,
1109 * this function uses CollationElementIterators to match characters in
1110 * the strings (only primary-order differences are significant in
1111 * determining whether there's a match).
1112 * @param str The string being tested
1113 * @param prefix The text we're hoping to see at the beginning
1115 * @return If "prefix" is found at the beginning of "str", this
1116 * is the number of characters in "str" that were matched (this
1117 * isn't necessarily the same as the length of "prefix" when matching
1118 * text with a collator). If there's no match, this is 0.
1121 NFRule::prefixLength(const UnicodeString& str, const UnicodeString& prefix, UErrorCode& status) const
1123 // if we're looking for an empty prefix, it obviously matches
1124 // zero characters. Just go ahead and return 0.
1125 if (prefix.length() == 0) {
1129 #if !UCONFIG_NO_COLLATION
1130 // go through all this grief if we're in lenient-parse mode
1131 if (formatter->isLenient()) {
1132 // get the formatter's collator and use it to create two
1133 // collation element iterators, one over the target string
1134 // and another over the prefix (right now, we'll throw an
1135 // exception if the collator we get back from the formatter
1136 // isn't a RuleBasedCollator, because RuleBasedCollator defines
1137 // the CollationElementIterator protocol. Hopefully, this
1138 // will change someday.)
1139 RuleBasedCollator* collator = (RuleBasedCollator*)formatter->getCollator();
1140 CollationElementIterator* strIter = collator->createCollationElementIterator(str);
1141 CollationElementIterator* prefixIter = collator->createCollationElementIterator(prefix);
1142 // Check for memory allocation error.
1143 if (collator == NULL || strIter == NULL || prefixIter == NULL) {
1147 status = U_MEMORY_ALLOCATION_ERROR;
1151 UErrorCode err = U_ZERO_ERROR;
1153 // The original code was problematic. Consider this match:
1154 // prefix = "fifty-"
1155 // string = " fifty-7"
1156 // The intent is to match string up to the '7', by matching 'fifty-' at position 1
1157 // in the string. Unfortunately, we were getting a match, and then computing where
1158 // the match terminated by rematching the string. The rematch code was using as an
1159 // initial guess the substring of string between 0 and prefix.length. Because of
1160 // the leading space and trailing hyphen (both ignorable) this was succeeding, leaving
1161 // the position before the hyphen in the string. Recursing down, we then parsed the
1162 // remaining string '-7' as numeric. The resulting number turned out as 43 (50 - 7).
1163 // This was not pretty, especially since the string "fifty-7" parsed just fine.
1165 // We have newer APIs now, so we can use calls on the iterator to determine what we
1166 // matched up to. If we terminate because we hit the last element in the string,
1167 // our match terminates at this length. If we terminate because we hit the last element
1168 // in the target, our match terminates at one before the element iterator position.
1170 // match collation elements between the strings
1171 int32_t oStr = strIter->next(err);
1172 int32_t oPrefix = prefixIter->next(err);
1174 while (oPrefix != CollationElementIterator::NULLORDER) {
1175 // skip over ignorable characters in the target string
1176 while (CollationElementIterator::primaryOrder(oStr) == 0
1177 && oStr != CollationElementIterator::NULLORDER) {
1178 oStr = strIter->next(err);
1181 // skip over ignorable characters in the prefix
1182 while (CollationElementIterator::primaryOrder(oPrefix) == 0
1183 && oPrefix != CollationElementIterator::NULLORDER) {
1184 oPrefix = prefixIter->next(err);
1187 // dlf: move this above following test, if we consume the
1188 // entire target, aren't we ok even if the source was also
1189 // entirely consumed?
1191 // if skipping over ignorables brought to the end of
1192 // the prefix, we DID match: drop out of the loop
1193 if (oPrefix == CollationElementIterator::NULLORDER) {
1197 // if skipping over ignorables brought us to the end
1198 // of the target string, we didn't match and return 0
1199 if (oStr == CollationElementIterator::NULLORDER) {
1205 // match collation elements from the two strings
1206 // (considering only primary differences). If we
1207 // get a mismatch, dump out and return 0
1208 if (CollationElementIterator::primaryOrder(oStr)
1209 != CollationElementIterator::primaryOrder(oPrefix)) {
1214 // otherwise, advance to the next character in each string
1215 // and loop (we drop out of the loop when we exhaust
1216 // collation elements in the prefix)
1218 oStr = strIter->next(err);
1219 oPrefix = prefixIter->next(err);
1223 int32_t result = strIter->getOffset();
1224 if (oStr != CollationElementIterator::NULLORDER) {
1225 --result; // back over character that we don't want to consume;
1229 fprintf(stderr, "prefix length: %d\n", result);
1236 //----------------------------------------------------------------
1237 // JDK 1.2-specific API call
1238 // return strIter.getOffset();
1239 //----------------------------------------------------------------
1240 // JDK 1.1 HACK (take out for 1.2-specific code)
1242 // if we make it to here, we have a successful match. Now we
1243 // have to find out HOW MANY characters from the target string
1244 // matched the prefix (there isn't necessarily a one-to-one
1245 // mapping between collation elements and characters).
1246 // In JDK 1.2, there's a simple getOffset() call we can use.
1247 // In JDK 1.1, on the other hand, we have to go through some
1248 // ugly contortions. First, use the collator to compare the
1249 // same number of characters from the prefix and target string.
1250 // If they're equal, we're done.
1251 collator->setStrength(Collator::PRIMARY);
1252 if (str.length() >= prefix.length()) {
1254 temp.setTo(str, 0, prefix.length());
1255 if (collator->equals(temp, prefix)) {
1257 fprintf(stderr, "returning: %d\n", prefix.length());
1259 return prefix.length();
1263 // if they're not equal, then we have to compare successively
1264 // larger and larger substrings of the target string until we
1265 // get to one that matches the prefix. At that point, we know
1266 // how many characters matched the prefix, and we can return.
1268 while (p <= str.length()) {
1270 temp.setTo(str, 0, p);
1271 if (collator->equals(temp, prefix)) {
1278 // SHOULD NEVER GET HERE!!!
1280 //----------------------------------------------------------------
1283 // If lenient parsing is turned off, forget all that crap above.
1284 // Just use String.startsWith() and be done with it.
1288 if (str.startsWith(prefix)) {
1289 return prefix.length();
1297 * Searches a string for another string. If lenient parsing is off,
1298 * this just calls indexOf(). If lenient parsing is on, this function
1299 * uses CollationElementIterator to match characters, and only
1300 * primary-order differences are significant in determining whether
1302 * @param str The string to search
1303 * @param key The string to search "str" for
1304 * @param startingAt The index into "str" where the search is to
1306 * @return A two-element array of ints. Element 0 is the position
1307 * of the match, or -1 if there was no match. Element 1 is the
1308 * number of characters in "str" that matched (which isn't necessarily
1309 * the same as the length of "key")
1312 NFRule::findText(const UnicodeString& str,
1313 const UnicodeString& key,
1315 int32_t* length) const
1317 #if !UCONFIG_NO_COLLATION
1318 // if lenient parsing is turned off, this is easy: just call
1319 // String.indexOf() and we're done
1320 if (!formatter->isLenient()) {
1321 *length = key.length();
1322 return str.indexOf(key, startingAt);
1324 // but if lenient parsing is turned ON, we've got some work
1329 //----------------------------------------------------------------
1330 // JDK 1.1 HACK (take out of 1.2-specific code)
1332 // in JDK 1.2, CollationElementIterator provides us with an
1333 // API to map between character offsets and collation elements
1334 // and we can do this by marching through the string comparing
1335 // collation elements. We can't do that in JDK 1.1. Insted,
1336 // we have to go through this horrible slow mess:
1337 int32_t p = startingAt;
1340 // basically just isolate smaller and smaller substrings of
1341 // the target string (each running to the end of the string,
1342 // and with the first one running from startingAt to the end)
1343 // and then use prefixLength() to see if the search key is at
1344 // the beginning of each substring. This is excruciatingly
1345 // slow, but it will locate the key and tell use how long the
1346 // matching text was.
1348 UErrorCode status = U_ZERO_ERROR;
1349 while (p < str.length() && keyLen == 0) {
1350 temp.setTo(str, p, str.length() - p);
1351 keyLen = prefixLength(temp, key, status);
1352 if (U_FAILURE(status)) {
1361 // if we make it to here, we didn't find it. Return -1 for the
1362 // location. The length should be ignored, but set it to 0,
1363 // which should be "safe"
1367 //----------------------------------------------------------------
1368 // JDK 1.2 version of this routine
1369 //RuleBasedCollator collator = (RuleBasedCollator)formatter.getCollator();
1371 //CollationElementIterator strIter = collator.getCollationElementIterator(str);
1372 //CollationElementIterator keyIter = collator.getCollationElementIterator(key);
1374 //int keyStart = -1;
1376 //str.setOffset(startingAt);
1378 //int oStr = strIter.next();
1379 //int oKey = keyIter.next();
1380 //while (oKey != CollationElementIterator.NULLORDER) {
1381 // while (oStr != CollationElementIterator.NULLORDER &&
1382 // CollationElementIterator.primaryOrder(oStr) == 0)
1383 // oStr = strIter.next();
1385 // while (oKey != CollationElementIterator.NULLORDER &&
1386 // CollationElementIterator.primaryOrder(oKey) == 0)
1387 // oKey = keyIter.next();
1389 // if (oStr == CollationElementIterator.NULLORDER) {
1390 // return new int[] { -1, 0 };
1393 // if (oKey == CollationElementIterator.NULLORDER) {
1397 // if (CollationElementIterator.primaryOrder(oStr) ==
1398 // CollationElementIterator.primaryOrder(oKey)) {
1399 // keyStart = strIter.getOffset();
1400 // oStr = strIter.next();
1401 // oKey = keyIter.next();
1403 // if (keyStart != -1) {
1407 // oStr = strIter.next();
1412 //if (oKey == CollationElementIterator.NULLORDER) {
1413 // return new int[] { keyStart, strIter.getOffset() - keyStart };
1415 // return new int[] { -1, 0 };
1421 * Checks to see whether a string consists entirely of ignorable
1423 * @param str The string to test.
1424 * @return true if the string is empty of consists entirely of
1425 * characters that the number formatter's collator says are
1426 * ignorable at the primary-order level. false otherwise.
1429 NFRule::allIgnorable(const UnicodeString& str, UErrorCode& status) const
1431 // if the string is empty, we can just return true
1432 if (str.length() == 0) {
1436 #if !UCONFIG_NO_COLLATION
1437 // if lenient parsing is turned on, walk through the string with
1438 // a collation element iterator and make sure each collation
1439 // element is 0 (ignorable) at the primary level
1440 if (formatter->isLenient()) {
1441 RuleBasedCollator* collator = (RuleBasedCollator*)(formatter->getCollator());
1442 CollationElementIterator* iter = collator->createCollationElementIterator(str);
1444 // Memory allocation error check.
1445 if (collator == NULL || iter == NULL) {
1448 status = U_MEMORY_ALLOCATION_ERROR;
1452 UErrorCode err = U_ZERO_ERROR;
1453 int32_t o = iter->next(err);
1454 while (o != CollationElementIterator::NULLORDER
1455 && CollationElementIterator::primaryOrder(o) == 0) {
1456 o = iter->next(err);
1460 return o == CollationElementIterator::NULLORDER;
1464 // if lenient parsing is turned off, there is no such thing as
1465 // an ignorable character: return true only if the string is empty