diff options
author | monster <monster@ydb.tech> | 2022-07-07 14:41:37 +0300 |
---|---|---|
committer | monster <monster@ydb.tech> | 2022-07-07 14:41:37 +0300 |
commit | 06e5c21a835c0e923506c4ff27929f34e00761c2 (patch) | |
tree | 75efcbc6854ef9bd476eb8bf00cc5c900da436a2 /contrib/libs/icu/i18n/nfsubs.cpp | |
parent | 03f024c4412e3aa613bb543cf1660176320ba8f4 (diff) | |
download | ydb-06e5c21a835c0e923506c4ff27929f34e00761c2.tar.gz |
fix ya.make
Diffstat (limited to 'contrib/libs/icu/i18n/nfsubs.cpp')
-rw-r--r-- | contrib/libs/icu/i18n/nfsubs.cpp | 1345 |
1 files changed, 1345 insertions, 0 deletions
diff --git a/contrib/libs/icu/i18n/nfsubs.cpp b/contrib/libs/icu/i18n/nfsubs.cpp new file mode 100644 index 0000000000..208543d1ac --- /dev/null +++ b/contrib/libs/icu/i18n/nfsubs.cpp @@ -0,0 +1,1345 @@ +// © 2016 and later: Unicode, Inc. and others. +// License & terms of use: http://www.unicode.org/copyright.html +/* +****************************************************************************** +* Copyright (C) 1997-2015, International Business Machines +* Corporation and others. All Rights Reserved. +****************************************************************************** +* file name: nfsubs.cpp +* encoding: UTF-8 +* tab size: 8 (not used) +* indentation:4 +* +* Modification history +* Date Name Comments +* 10/11/2001 Doug Ported from ICU4J +*/ + +#include <stdio.h> +#include "utypeinfo.h" // for 'typeid' to work + +#include "nfsubs.h" +#include "fmtableimp.h" +#include "putilimp.h" +#include "number_decimalquantity.h" + +#if U_HAVE_RBNF + +static const UChar gLessThan = 0x003c; +static const UChar gEquals = 0x003d; +static const UChar gGreaterThan = 0x003e; +static const UChar gPercent = 0x0025; +static const UChar gPound = 0x0023; +static const UChar gZero = 0x0030; +static const UChar gSpace = 0x0020; + +static const UChar gEqualsEquals[] = +{ + 0x3D, 0x3D, 0 +}; /* "==" */ +static const UChar gGreaterGreaterGreaterThan[] = +{ + 0x3E, 0x3E, 0x3E, 0 +}; /* ">>>" */ +static const UChar gGreaterGreaterThan[] = +{ + 0x3E, 0x3E, 0 +}; /* ">>" */ + +U_NAMESPACE_BEGIN + +using number::impl::DecimalQuantity; + +class SameValueSubstitution : public NFSubstitution { +public: + SameValueSubstitution(int32_t pos, + const NFRuleSet* ruleset, + const UnicodeString& description, + UErrorCode& status); + virtual ~SameValueSubstitution(); + + virtual int64_t transformNumber(int64_t number) const { return number; } + virtual double transformNumber(double number) const { return number; } + virtual double composeRuleValue(double newRuleValue, double /*oldRuleValue*/) const { return newRuleValue; } + virtual double calcUpperBound(double oldUpperBound) const { return oldUpperBound; } + virtual UChar tokenChar() const { return (UChar)0x003d; } // '=' + +public: + static UClassID getStaticClassID(void); + virtual UClassID getDynamicClassID(void) const; +}; + +SameValueSubstitution::~SameValueSubstitution() {} + +class MultiplierSubstitution : public NFSubstitution { + int64_t divisor; + +public: + MultiplierSubstitution(int32_t _pos, + const NFRule *rule, + const NFRuleSet* _ruleSet, + const UnicodeString& description, + UErrorCode& status) + : NFSubstitution(_pos, _ruleSet, description, status), divisor(rule->getDivisor()) + { + if (divisor == 0) { + status = U_PARSE_ERROR; + } + } + virtual ~MultiplierSubstitution(); + + virtual void setDivisor(int32_t radix, int16_t exponent, UErrorCode& status) { + divisor = util64_pow(radix, exponent); + + if(divisor == 0) { + status = U_PARSE_ERROR; + } + } + + virtual UBool operator==(const NFSubstitution& rhs) const; + + virtual int64_t transformNumber(int64_t number) const { + return number / divisor; + } + + virtual double transformNumber(double number) const { + if (getRuleSet()) { + return uprv_floor(number / divisor); + } else { + return number / divisor; + } + } + + virtual double composeRuleValue(double newRuleValue, double /*oldRuleValue*/) const { + return newRuleValue * divisor; + } + + virtual double calcUpperBound(double /*oldUpperBound*/) const { return static_cast<double>(divisor); } + + virtual UChar tokenChar() const { return (UChar)0x003c; } // '<' + +public: + static UClassID getStaticClassID(void); + virtual UClassID getDynamicClassID(void) const; +}; + +MultiplierSubstitution::~MultiplierSubstitution() {} + +class ModulusSubstitution : public NFSubstitution { + int64_t divisor; + const NFRule* ruleToUse; +public: + ModulusSubstitution(int32_t pos, + const NFRule* rule, + const NFRule* rulePredecessor, + const NFRuleSet* ruleSet, + const UnicodeString& description, + UErrorCode& status); + virtual ~ModulusSubstitution(); + + virtual void setDivisor(int32_t radix, int16_t exponent, UErrorCode& status) { + divisor = util64_pow(radix, exponent); + + if (divisor == 0) { + status = U_PARSE_ERROR; + } + } + + virtual UBool operator==(const NFSubstitution& rhs) const; + + virtual void doSubstitution(int64_t number, UnicodeString& toInsertInto, int32_t pos, int32_t recursionCount, UErrorCode& status) const; + virtual void doSubstitution(double number, UnicodeString& toInsertInto, int32_t pos, int32_t recursionCount, UErrorCode& status) const; + + virtual int64_t transformNumber(int64_t number) const { return number % divisor; } + virtual double transformNumber(double number) const { return uprv_fmod(number, static_cast<double>(divisor)); } + + virtual UBool doParse(const UnicodeString& text, + ParsePosition& parsePosition, + double baseValue, + double upperBound, + UBool lenientParse, + uint32_t nonNumericalExecutedRuleMask, + Formattable& result) const; + + virtual double composeRuleValue(double newRuleValue, double oldRuleValue) const { + return oldRuleValue - uprv_fmod(oldRuleValue, static_cast<double>(divisor)) + newRuleValue; + } + + virtual double calcUpperBound(double /*oldUpperBound*/) const { return static_cast<double>(divisor); } + + virtual UBool isModulusSubstitution() const { return TRUE; } + + virtual UChar tokenChar() const { return (UChar)0x003e; } // '>' + + virtual void toString(UnicodeString& result) const; + +public: + static UClassID getStaticClassID(void); + virtual UClassID getDynamicClassID(void) const; +}; + +ModulusSubstitution::~ModulusSubstitution() {} + +class IntegralPartSubstitution : public NFSubstitution { +public: + IntegralPartSubstitution(int32_t _pos, + const NFRuleSet* _ruleSet, + const UnicodeString& description, + UErrorCode& status) + : NFSubstitution(_pos, _ruleSet, description, status) {} + virtual ~IntegralPartSubstitution(); + + virtual int64_t transformNumber(int64_t number) const { return number; } + virtual double transformNumber(double number) const { return uprv_floor(number); } + virtual double composeRuleValue(double newRuleValue, double oldRuleValue) const { return newRuleValue + oldRuleValue; } + virtual double calcUpperBound(double /*oldUpperBound*/) const { return DBL_MAX; } + virtual UChar tokenChar() const { return (UChar)0x003c; } // '<' + +public: + static UClassID getStaticClassID(void); + virtual UClassID getDynamicClassID(void) const; +}; + +IntegralPartSubstitution::~IntegralPartSubstitution() {} + +class FractionalPartSubstitution : public NFSubstitution { + UBool byDigits; + UBool useSpaces; + enum { kMaxDecimalDigits = 8 }; +public: + FractionalPartSubstitution(int32_t pos, + const NFRuleSet* ruleSet, + const UnicodeString& description, + UErrorCode& status); + virtual ~FractionalPartSubstitution(); + + virtual UBool operator==(const NFSubstitution& rhs) const; + + virtual void doSubstitution(double number, UnicodeString& toInsertInto, int32_t pos, int32_t recursionCount, UErrorCode& status) const; + virtual void doSubstitution(int64_t /*number*/, UnicodeString& /*toInsertInto*/, int32_t /*_pos*/, int32_t /*recursionCount*/, UErrorCode& /*status*/) const {} + virtual int64_t transformNumber(int64_t /*number*/) const { return 0; } + virtual double transformNumber(double number) const { return number - uprv_floor(number); } + + virtual UBool doParse(const UnicodeString& text, + ParsePosition& parsePosition, + double baseValue, + double upperBound, + UBool lenientParse, + uint32_t nonNumericalExecutedRuleMask, + Formattable& result) const; + + virtual double composeRuleValue(double newRuleValue, double oldRuleValue) const { return newRuleValue + oldRuleValue; } + virtual double calcUpperBound(double /*oldUpperBound*/) const { return 0.0; } + virtual UChar tokenChar() const { return (UChar)0x003e; } // '>' + +public: + static UClassID getStaticClassID(void); + virtual UClassID getDynamicClassID(void) const; +}; + +FractionalPartSubstitution::~FractionalPartSubstitution() {} + +class AbsoluteValueSubstitution : public NFSubstitution { +public: + AbsoluteValueSubstitution(int32_t _pos, + const NFRuleSet* _ruleSet, + const UnicodeString& description, + UErrorCode& status) + : NFSubstitution(_pos, _ruleSet, description, status) {} + virtual ~AbsoluteValueSubstitution(); + + virtual int64_t transformNumber(int64_t number) const { return number >= 0 ? number : -number; } + virtual double transformNumber(double number) const { return uprv_fabs(number); } + virtual double composeRuleValue(double newRuleValue, double /*oldRuleValue*/) const { return -newRuleValue; } + virtual double calcUpperBound(double /*oldUpperBound*/) const { return DBL_MAX; } + virtual UChar tokenChar() const { return (UChar)0x003e; } // '>' + +public: + static UClassID getStaticClassID(void); + virtual UClassID getDynamicClassID(void) const; +}; + +AbsoluteValueSubstitution::~AbsoluteValueSubstitution() {} + +class NumeratorSubstitution : public NFSubstitution { + double denominator; + int64_t ldenominator; + UBool withZeros; +public: + static inline UnicodeString fixdesc(const UnicodeString& desc) { + if (desc.endsWith(LTLT, 2)) { + UnicodeString result(desc, 0, desc.length()-1); + return result; + } + return desc; + } + NumeratorSubstitution(int32_t _pos, + double _denominator, + NFRuleSet* _ruleSet, + const UnicodeString& description, + UErrorCode& status) + : NFSubstitution(_pos, _ruleSet, fixdesc(description), status), denominator(_denominator) + { + ldenominator = util64_fromDouble(denominator); + withZeros = description.endsWith(LTLT, 2); + } + virtual ~NumeratorSubstitution(); + + virtual UBool operator==(const NFSubstitution& rhs) const; + + virtual int64_t transformNumber(int64_t number) const { return number * ldenominator; } + virtual double transformNumber(double number) const { return uprv_round(number * denominator); } + + virtual void doSubstitution(int64_t /*number*/, UnicodeString& /*toInsertInto*/, int32_t /*_pos*/, int32_t /*recursionCount*/, UErrorCode& /*status*/) const {} + virtual void doSubstitution(double number, UnicodeString& toInsertInto, int32_t pos, int32_t recursionCount, UErrorCode& status) const; + virtual UBool doParse(const UnicodeString& text, + ParsePosition& parsePosition, + double baseValue, + double upperBound, + UBool /*lenientParse*/, + uint32_t nonNumericalExecutedRuleMask, + Formattable& result) const; + + virtual double composeRuleValue(double newRuleValue, double oldRuleValue) const { return newRuleValue / oldRuleValue; } + virtual double calcUpperBound(double /*oldUpperBound*/) const { return denominator; } + virtual UChar tokenChar() const { return (UChar)0x003c; } // '<' +private: + static const UChar LTLT[2]; + +public: + static UClassID getStaticClassID(void); + virtual UClassID getDynamicClassID(void) const; +}; + +NumeratorSubstitution::~NumeratorSubstitution() {} + +NFSubstitution* +NFSubstitution::makeSubstitution(int32_t pos, + const NFRule* rule, + const NFRule* predecessor, + const NFRuleSet* ruleSet, + const RuleBasedNumberFormat* formatter, + const UnicodeString& description, + UErrorCode& status) +{ + // if the description is empty, return a NullSubstitution + if (description.length() == 0) { + return NULL; + } + + switch (description.charAt(0)) { + // if the description begins with '<'... + case gLessThan: + // throw an exception if the rule is a negative number + // rule + if (rule->getBaseValue() == NFRule::kNegativeNumberRule) { + // throw new IllegalArgumentException("<< not allowed in negative-number rule"); + status = U_PARSE_ERROR; + return NULL; + } + + // if the rule is a fraction rule, return an + // IntegralPartSubstitution + else if (rule->getBaseValue() == NFRule::kImproperFractionRule + || rule->getBaseValue() == NFRule::kProperFractionRule + || rule->getBaseValue() == NFRule::kMasterRule) { + return new IntegralPartSubstitution(pos, ruleSet, description, status); + } + + // if the rule set containing the rule is a fraction + // rule set, return a NumeratorSubstitution + else if (ruleSet->isFractionRuleSet()) { + return new NumeratorSubstitution(pos, (double)rule->getBaseValue(), + formatter->getDefaultRuleSet(), description, status); + } + + // otherwise, return a MultiplierSubstitution + else { + return new MultiplierSubstitution(pos, rule, ruleSet, + description, status); + } + + // if the description begins with '>'... + case gGreaterThan: + // if the rule is a negative-number rule, return + // an AbsoluteValueSubstitution + if (rule->getBaseValue() == NFRule::kNegativeNumberRule) { + return new AbsoluteValueSubstitution(pos, ruleSet, description, status); + } + + // if the rule is a fraction rule, return a + // FractionalPartSubstitution + else if (rule->getBaseValue() == NFRule::kImproperFractionRule + || rule->getBaseValue() == NFRule::kProperFractionRule + || rule->getBaseValue() == NFRule::kMasterRule) { + return new FractionalPartSubstitution(pos, ruleSet, description, status); + } + + // if the rule set owning the rule is a fraction rule set, + // throw an exception + else if (ruleSet->isFractionRuleSet()) { + // throw new IllegalArgumentException(">> not allowed in fraction rule set"); + status = U_PARSE_ERROR; + return NULL; + } + + // otherwise, return a ModulusSubstitution + else { + return new ModulusSubstitution(pos, rule, predecessor, + ruleSet, description, status); + } + + // if the description begins with '=', always return a + // SameValueSubstitution + case gEquals: + return new SameValueSubstitution(pos, ruleSet, description, status); + + // and if it's anything else, throw an exception + default: + // throw new IllegalArgumentException("Illegal substitution character"); + status = U_PARSE_ERROR; + } + return NULL; +} + +NFSubstitution::NFSubstitution(int32_t _pos, + const NFRuleSet* _ruleSet, + const UnicodeString& description, + UErrorCode& status) + : pos(_pos), ruleSet(NULL), numberFormat(NULL) +{ + // the description should begin and end with the same character. + // If it doesn't that's a syntax error. Otherwise, + // makeSubstitution() was the only thing that needed to know + // about these characters, so strip them off + UnicodeString workingDescription(description); + if (description.length() >= 2 + && description.charAt(0) == description.charAt(description.length() - 1)) + { + workingDescription.remove(description.length() - 1, 1); + workingDescription.remove(0, 1); + } + else if (description.length() != 0) { + // throw new IllegalArgumentException("Illegal substitution syntax"); + status = U_PARSE_ERROR; + return; + } + + if (workingDescription.length() == 0) { + // if the description was just two paired token characters + // (i.e., "<<" or ">>"), it uses the rule set it belongs to to + // format its result + this->ruleSet = _ruleSet; + } + else if (workingDescription.charAt(0) == gPercent) { + // if the description contains a rule set name, that's the rule + // set we use to format the result: get a reference to the + // names rule set + this->ruleSet = _ruleSet->getOwner()->findRuleSet(workingDescription, status); + } + else if (workingDescription.charAt(0) == gPound || workingDescription.charAt(0) ==gZero) { + // if the description begins with 0 or #, treat it as a + // DecimalFormat pattern, and initialize a DecimalFormat with + // that pattern (then set it to use the DecimalFormatSymbols + // belonging to our formatter) + const DecimalFormatSymbols* sym = _ruleSet->getOwner()->getDecimalFormatSymbols(); + if (!sym) { + status = U_MISSING_RESOURCE_ERROR; + return; + } + DecimalFormat *tempNumberFormat = new DecimalFormat(workingDescription, *sym, status); + /* test for NULL */ + if (!tempNumberFormat) { + status = U_MEMORY_ALLOCATION_ERROR; + return; + } + if (U_FAILURE(status)) { + delete tempNumberFormat; + return; + } + this->numberFormat = tempNumberFormat; + } + else if (workingDescription.charAt(0) == gGreaterThan) { + // if the description is ">>>", this substitution bypasses the + // usual rule-search process and always uses the rule that precedes + // it in its own rule set's rule list (this is used for place-value + // notations: formats where you want to see a particular part of + // a number even when it's 0) + + // this causes problems when >>> is used in a frationalPartSubstitution + // this->ruleSet = NULL; + this->ruleSet = _ruleSet; + this->numberFormat = NULL; + } + else { + // and of the description is none of these things, it's a syntax error + + // throw new IllegalArgumentException("Illegal substitution syntax"); + status = U_PARSE_ERROR; + } +} + +NFSubstitution::~NFSubstitution() +{ + delete numberFormat; + numberFormat = NULL; +} + +/** + * Set's the substitution's divisor. Used by NFRule.setBaseValue(). + * A no-op for all substitutions except multiplier and modulus + * substitutions. + * @param radix The radix of the divisor + * @param exponent The exponent of the divisor + */ +void +NFSubstitution::setDivisor(int32_t /*radix*/, int16_t /*exponent*/, UErrorCode& /*status*/) { + // a no-op for all substitutions except multiplier and modulus substitutions +} + +void +NFSubstitution::setDecimalFormatSymbols(const DecimalFormatSymbols &newSymbols, UErrorCode& /*status*/) { + if (numberFormat != NULL) { + numberFormat->setDecimalFormatSymbols(newSymbols); + } +} + +//----------------------------------------------------------------------- +// boilerplate +//----------------------------------------------------------------------- + +UOBJECT_DEFINE_RTTI_IMPLEMENTATION(NFSubstitution) + +/** + * Compares two substitutions for equality + * @param The substitution to compare this one to + * @return true if the two substitutions are functionally equivalent + */ +UBool +NFSubstitution::operator==(const NFSubstitution& rhs) const +{ + // compare class and all of the fields all substitutions have + // in common + // this should be called by subclasses before their own equality tests + return typeid(*this) == typeid(rhs) + && pos == rhs.pos + && (ruleSet == NULL) == (rhs.ruleSet == NULL) + // && ruleSet == rhs.ruleSet causes circularity, other checks to make instead? + && (numberFormat == NULL + ? (rhs.numberFormat == NULL) + : (*numberFormat == *rhs.numberFormat)); +} + +/** + * Returns a textual description of the substitution + * @return A textual description of the substitution. This might + * not be identical to the description it was created from, but + * it'll produce the same result. + */ +void +NFSubstitution::toString(UnicodeString& text) const +{ + // use tokenChar() to get the character at the beginning and + // end of the substitutin token. In between them will go + // either the name of the rule set it uses, or the pattern of + // the DecimalFormat it uses + text.remove(); + text.append(tokenChar()); + + UnicodeString temp; + if (ruleSet != NULL) { + ruleSet->getName(temp); + } else if (numberFormat != NULL) { + numberFormat->toPattern(temp); + } + text.append(temp); + text.append(tokenChar()); +} + +//----------------------------------------------------------------------- +// formatting +//----------------------------------------------------------------------- + +/** + * Performs a mathematical operation on the number, formats it using + * either ruleSet or decimalFormat, and inserts the result into + * toInsertInto. + * @param number The number being formatted. + * @param toInsertInto The string we insert the result into + * @param pos The position in toInsertInto where the owning rule's + * rule text begins (this value is added to this substitution's + * position to determine exactly where to insert the new text) + */ +void +NFSubstitution::doSubstitution(int64_t number, UnicodeString& toInsertInto, int32_t _pos, int32_t recursionCount, UErrorCode& status) const +{ + if (ruleSet != NULL) { + // Perform a transformation on the number that is dependent + // on the type of substitution this is, then just call its + // rule set's format() method to format the result + ruleSet->format(transformNumber(number), toInsertInto, _pos + this->pos, recursionCount, status); + } else if (numberFormat != NULL) { + if (number <= MAX_INT64_IN_DOUBLE) { + // or perform the transformation on the number (preserving + // the result's fractional part if the formatter it set + // to show it), then use that formatter's format() method + // to format the result + double numberToFormat = transformNumber((double)number); + if (numberFormat->getMaximumFractionDigits() == 0) { + numberToFormat = uprv_floor(numberToFormat); + } + + UnicodeString temp; + numberFormat->format(numberToFormat, temp, status); + toInsertInto.insert(_pos + this->pos, temp); + } + else { + // We have gone beyond double precision. Something has to give. + // We're favoring accuracy of the large number over potential rules + // that round like a CompactDecimalFormat, which is not a common use case. + // + // Perform a transformation on the number that is dependent + // on the type of substitution this is, then just call its + // rule set's format() method to format the result + int64_t numberToFormat = transformNumber(number); + UnicodeString temp; + numberFormat->format(numberToFormat, temp, status); + toInsertInto.insert(_pos + this->pos, temp); + } + } +} + +/** + * Performs a mathematical operation on the number, formats it using + * either ruleSet or decimalFormat, and inserts the result into + * toInsertInto. + * @param number The number being formatted. + * @param toInsertInto The string we insert the result into + * @param pos The position in toInsertInto where the owning rule's + * rule text begins (this value is added to this substitution's + * position to determine exactly where to insert the new text) + */ +void +NFSubstitution::doSubstitution(double number, UnicodeString& toInsertInto, int32_t _pos, int32_t recursionCount, UErrorCode& status) const { + // perform a transformation on the number being formatted that + // is dependent on the type of substitution this is + double numberToFormat = transformNumber(number); + + if (uprv_isInfinite(numberToFormat)) { + // This is probably a minus rule. Combine it with an infinite rule. + const NFRule *infiniteRule = ruleSet->findDoubleRule(uprv_getInfinity()); + infiniteRule->doFormat(numberToFormat, toInsertInto, _pos + this->pos, recursionCount, status); + return; + } + + // if the result is an integer, from here on out we work in integer + // space (saving time and memory and preserving accuracy) + if (numberToFormat == uprv_floor(numberToFormat) && ruleSet != NULL) { + ruleSet->format(util64_fromDouble(numberToFormat), toInsertInto, _pos + this->pos, recursionCount, status); + + // if the result isn't an integer, then call either our rule set's + // format() method or our DecimalFormat's format() method to + // format the result + } else { + if (ruleSet != NULL) { + ruleSet->format(numberToFormat, toInsertInto, _pos + this->pos, recursionCount, status); + } else if (numberFormat != NULL) { + UnicodeString temp; + numberFormat->format(numberToFormat, temp); + toInsertInto.insert(_pos + this->pos, temp); + } + } +} + + + //----------------------------------------------------------------------- + // parsing + //----------------------------------------------------------------------- + +#ifdef RBNF_DEBUG +#include <stdio.h> +#endif + +/** + * Parses a string using the rule set or DecimalFormat belonging + * to this substitution. If there's a match, a mathematical + * operation (the inverse of the one used in formatting) is + * performed on the result of the parse and the value passed in + * and returned as the result. The parse position is updated to + * point to the first unmatched character in the string. + * @param text The string to parse + * @param parsePosition On entry, ignored, but assumed to be 0. + * On exit, this is updated to point to the first unmatched + * character (or 0 if the substitution didn't match) + * @param baseValue A partial parse result that should be + * combined with the result of this parse + * @param upperBound When searching the rule set for a rule + * matching the string passed in, only rules with base values + * lower than this are considered + * @param lenientParse If true and matching against rules fails, + * the substitution will also try matching the text against + * numerals using a default-costructed NumberFormat. If false, + * no extra work is done. (This value is false whenever the + * formatter isn't in lenient-parse mode, but is also false + * under some conditions even when the formatter _is_ in + * lenient-parse mode.) + * @return If there's a match, this is the result of composing + * baseValue with whatever was returned from matching the + * characters. This will be either a Long or a Double. If there's + * no match this is new Long(0) (not null), and parsePosition + * is left unchanged. + */ +UBool +NFSubstitution::doParse(const UnicodeString& text, + ParsePosition& parsePosition, + double baseValue, + double upperBound, + UBool lenientParse, + uint32_t nonNumericalExecutedRuleMask, + Formattable& result) const +{ +#ifdef RBNF_DEBUG + fprintf(stderr, "<nfsubs> %x bv: %g ub: %g\n", this, baseValue, upperBound); +#endif + // figure out the highest base value a rule can have and match + // the text being parsed (this varies according to the type of + // substitutions: multiplier, modulus, and numerator substitutions + // restrict the search to rules with base values lower than their + // own; same-value substitutions leave the upper bound wherever + // it was, and the others allow any rule to match + upperBound = calcUpperBound(upperBound); + + // use our rule set to parse the text. If that fails and + // lenient parsing is enabled (this is always false if the + // formatter's lenient-parsing mode is off, but it may also + // be false even when the formatter's lenient-parse mode is + // on), then also try parsing the text using a default- + // constructed NumberFormat + if (ruleSet != NULL) { + ruleSet->parse(text, parsePosition, upperBound, nonNumericalExecutedRuleMask, result); + if (lenientParse && !ruleSet->isFractionRuleSet() && parsePosition.getIndex() == 0) { + UErrorCode status = U_ZERO_ERROR; + NumberFormat* fmt = NumberFormat::createInstance(status); + if (U_SUCCESS(status)) { + fmt->parse(text, result, parsePosition); + } + delete fmt; + } + + // ...or use our DecimalFormat to parse the text + } else if (numberFormat != NULL) { + numberFormat->parse(text, result, parsePosition); + } + + // if the parse was successful, we've already advanced the caller's + // parse position (this is the one function that doesn't have one + // of its own). Derive a parse result and return it as a Long, + // if possible, or a Double + if (parsePosition.getIndex() != 0) { + UErrorCode status = U_ZERO_ERROR; + double tempResult = result.getDouble(status); + + // composeRuleValue() produces a full parse result from + // the partial parse result passed to this function from + // the caller (this is either the owning rule's base value + // or the partial result obtained from composing the + // owning rule's base value with its other substitution's + // parse result) and the partial parse result obtained by + // matching the substitution (which will be the same value + // the caller would get by parsing just this part of the + // text with RuleBasedNumberFormat.parse() ). How the two + // values are used to derive the full parse result depends + // on the types of substitutions: For a regular rule, the + // ultimate result is its multiplier substitution's result + // times the rule's divisor (or the rule's base value) plus + // the modulus substitution's result (which will actually + // supersede part of the rule's base value). For a negative- + // number rule, the result is the negative of its substitution's + // result. For a fraction rule, it's the sum of its two + // substitution results. For a rule in a fraction rule set, + // it's the numerator substitution's result divided by + // the rule's base value. Results from same-value substitutions + // propagate back upard, and null substitutions don't affect + // the result. + tempResult = composeRuleValue(tempResult, baseValue); + result.setDouble(tempResult); + return TRUE; + // if the parse was UNsuccessful, return 0 + } else { + result.setLong(0); + return FALSE; + } +} + + /** + * Returns true if this is a modulus substitution. (We didn't do this + * with instanceof partially because it causes source files to + * proliferate and partially because we have to port this to C++.) + * @return true if this object is an instance of ModulusSubstitution + */ +UBool +NFSubstitution::isModulusSubstitution() const { + return FALSE; +} + +//=================================================================== +// SameValueSubstitution +//=================================================================== + +/** + * A substitution that passes the value passed to it through unchanged. + * Represented by == in rule descriptions. + */ +SameValueSubstitution::SameValueSubstitution(int32_t _pos, + const NFRuleSet* _ruleSet, + const UnicodeString& description, + UErrorCode& status) +: NFSubstitution(_pos, _ruleSet, description, status) +{ + if (0 == description.compare(gEqualsEquals, 2)) { + // throw new IllegalArgumentException("== is not a legal token"); + status = U_PARSE_ERROR; + } +} + +UOBJECT_DEFINE_RTTI_IMPLEMENTATION(SameValueSubstitution) + +//=================================================================== +// MultiplierSubstitution +//=================================================================== + +UOBJECT_DEFINE_RTTI_IMPLEMENTATION(MultiplierSubstitution) + +UBool MultiplierSubstitution::operator==(const NFSubstitution& rhs) const +{ + return NFSubstitution::operator==(rhs) && + divisor == ((const MultiplierSubstitution*)&rhs)->divisor; +} + + +//=================================================================== +// ModulusSubstitution +//=================================================================== + +/** + * A substitution that divides the number being formatted by the its rule's + * divisor and formats the remainder. Represented by ">>" in a + * regular rule. + */ +ModulusSubstitution::ModulusSubstitution(int32_t _pos, + const NFRule* rule, + const NFRule* predecessor, + const NFRuleSet* _ruleSet, + const UnicodeString& description, + UErrorCode& status) + : NFSubstitution(_pos, _ruleSet, description, status) + , divisor(rule->getDivisor()) + , ruleToUse(NULL) +{ + // the owning rule's divisor controls the behavior of this + // substitution: rather than keeping a backpointer to the rule, + // we keep a copy of the divisor + + if (divisor == 0) { + status = U_PARSE_ERROR; + } + + if (0 == description.compare(gGreaterGreaterGreaterThan, 3)) { + // the >>> token doesn't alter how this substituion calculates the + // values it uses for formatting and parsing, but it changes + // what's done with that value after it's obtained: >>> short- + // circuits the rule-search process and goes straight to the + // specified rule to format the substitution value + ruleToUse = predecessor; + } +} + +UOBJECT_DEFINE_RTTI_IMPLEMENTATION(ModulusSubstitution) + +UBool ModulusSubstitution::operator==(const NFSubstitution& rhs) const +{ + return NFSubstitution::operator==(rhs) && + divisor == ((const ModulusSubstitution*)&rhs)->divisor && + ruleToUse == ((const ModulusSubstitution*)&rhs)->ruleToUse; +} + +//----------------------------------------------------------------------- +// formatting +//----------------------------------------------------------------------- + + +/** + * If this is a >>> substitution, use ruleToUse to fill in + * the substitution. Otherwise, just use the superclass function. + * @param number The number being formatted + * @toInsertInto The string to insert the result of this substitution + * into + * @param pos The position of the rule text in toInsertInto + */ +void +ModulusSubstitution::doSubstitution(int64_t number, UnicodeString& toInsertInto, int32_t _pos, int32_t recursionCount, UErrorCode& status) const +{ + // if this isn't a >>> substitution, just use the inherited version + // of this function (which uses either a rule set or a DecimalFormat + // to format its substitution value) + if (ruleToUse == NULL) { + NFSubstitution::doSubstitution(number, toInsertInto, _pos, recursionCount, status); + + // a >>> substitution goes straight to a particular rule to + // format the substitution value + } else { + int64_t numberToFormat = transformNumber(number); + ruleToUse->doFormat(numberToFormat, toInsertInto, _pos + getPos(), recursionCount, status); + } +} + +/** +* If this is a >>> substitution, use ruleToUse to fill in +* the substitution. Otherwise, just use the superclass function. +* @param number The number being formatted +* @toInsertInto The string to insert the result of this substitution +* into +* @param pos The position of the rule text in toInsertInto +*/ +void +ModulusSubstitution::doSubstitution(double number, UnicodeString& toInsertInto, int32_t _pos, int32_t recursionCount, UErrorCode& status) const +{ + // if this isn't a >>> substitution, just use the inherited version + // of this function (which uses either a rule set or a DecimalFormat + // to format its substitution value) + if (ruleToUse == NULL) { + NFSubstitution::doSubstitution(number, toInsertInto, _pos, recursionCount, status); + + // a >>> substitution goes straight to a particular rule to + // format the substitution value + } else { + double numberToFormat = transformNumber(number); + + ruleToUse->doFormat(numberToFormat, toInsertInto, _pos + getPos(), recursionCount, status); + } +} + +//----------------------------------------------------------------------- +// parsing +//----------------------------------------------------------------------- + +/** + * If this is a >>> substitution, match only against ruleToUse. + * Otherwise, use the superclass function. + * @param text The string to parse + * @param parsePosition Ignored on entry, updated on exit to point to + * the first unmatched character. + * @param baseValue The partial parse result prior to calling this + * routine. + */ +UBool +ModulusSubstitution::doParse(const UnicodeString& text, + ParsePosition& parsePosition, + double baseValue, + double upperBound, + UBool lenientParse, + uint32_t nonNumericalExecutedRuleMask, + Formattable& result) const +{ + // if this isn't a >>> substitution, we can just use the + // inherited parse() routine to do the parsing + if (ruleToUse == NULL) { + return NFSubstitution::doParse(text, parsePosition, baseValue, upperBound, lenientParse, nonNumericalExecutedRuleMask, result); + + // but if it IS a >>> substitution, we have to do it here: we + // use the specific rule's doParse() method, and then we have to + // do some of the other work of NFRuleSet.parse() + } else { + ruleToUse->doParse(text, parsePosition, FALSE, upperBound, nonNumericalExecutedRuleMask, result); + + if (parsePosition.getIndex() != 0) { + UErrorCode status = U_ZERO_ERROR; + double tempResult = result.getDouble(status); + tempResult = composeRuleValue(tempResult, baseValue); + result.setDouble(tempResult); + } + + return TRUE; + } +} +/** + * Returns a textual description of the substitution + * @return A textual description of the substitution. This might + * not be identical to the description it was created from, but + * it'll produce the same result. + */ +void +ModulusSubstitution::toString(UnicodeString& text) const +{ + // use tokenChar() to get the character at the beginning and + // end of the substitutin token. In between them will go + // either the name of the rule set it uses, or the pattern of + // the DecimalFormat it uses + + if ( ruleToUse != NULL ) { // Must have been a >>> substitution. + text.remove(); + text.append(tokenChar()); + text.append(tokenChar()); + text.append(tokenChar()); + } else { // Otherwise just use the super-class function. + NFSubstitution::toString(text); + } +} +//=================================================================== +// IntegralPartSubstitution +//=================================================================== + +UOBJECT_DEFINE_RTTI_IMPLEMENTATION(IntegralPartSubstitution) + + +//=================================================================== +// FractionalPartSubstitution +//=================================================================== + + + /** + * Constructs a FractionalPartSubstitution. This object keeps a flag + * telling whether it should format by digits or not. In addition, + * it marks the rule set it calls (if any) as a fraction rule set. + */ +FractionalPartSubstitution::FractionalPartSubstitution(int32_t _pos, + const NFRuleSet* _ruleSet, + const UnicodeString& description, + UErrorCode& status) + : NFSubstitution(_pos, _ruleSet, description, status) + , byDigits(FALSE) + , useSpaces(TRUE) + +{ + // akk, ruleSet can change in superclass constructor + if (0 == description.compare(gGreaterGreaterThan, 2) || + 0 == description.compare(gGreaterGreaterGreaterThan, 3) || + _ruleSet == getRuleSet()) { + byDigits = TRUE; + if (0 == description.compare(gGreaterGreaterGreaterThan, 3)) { + useSpaces = FALSE; + } + } else { + // cast away const + ((NFRuleSet*)getRuleSet())->makeIntoFractionRuleSet(); + } +} + +//----------------------------------------------------------------------- +// formatting +//----------------------------------------------------------------------- + +/** + * If in "by digits" mode, fills in the substitution one decimal digit + * at a time using the rule set containing this substitution. + * Otherwise, uses the superclass function. + * @param number The number being formatted + * @param toInsertInto The string to insert the result of formatting + * the substitution into + * @param pos The position of the owning rule's rule text in + * toInsertInto + */ +void +FractionalPartSubstitution::doSubstitution(double number, UnicodeString& toInsertInto, + int32_t _pos, int32_t recursionCount, UErrorCode& status) const +{ + // if we're not in "byDigits" mode, just use the inherited + // doSubstitution() routine + if (!byDigits) { + NFSubstitution::doSubstitution(number, toInsertInto, _pos, recursionCount, status); + + // if we're in "byDigits" mode, transform the value into an integer + // by moving the decimal point eight places to the right and + // pulling digits off the right one at a time, formatting each digit + // as an integer using this substitution's owning rule set + // (this is slower, but more accurate, than doing it from the + // other end) + } else { + // int32_t numberToFormat = (int32_t)uprv_round(transformNumber(number) * uprv_pow(10, kMaxDecimalDigits)); + // // this flag keeps us from formatting trailing zeros. It starts + // // out false because we're pulling from the right, and switches + // // to true the first time we encounter a non-zero digit + // UBool doZeros = FALSE; + // for (int32_t i = 0; i < kMaxDecimalDigits; i++) { + // int64_t digit = numberToFormat % 10; + // if (digit != 0 || doZeros) { + // if (doZeros && useSpaces) { + // toInsertInto.insert(_pos + getPos(), gSpace); + // } + // doZeros = TRUE; + // getRuleSet()->format(digit, toInsertInto, _pos + getPos()); + // } + // numberToFormat /= 10; + // } + + DecimalQuantity dl; + dl.setToDouble(number); + dl.roundToMagnitude(-20, UNUM_ROUND_HALFEVEN, status); // round to 20 fraction digits. + + UBool pad = FALSE; + for (int32_t didx = dl.getLowerDisplayMagnitude(); didx<0; didx++) { + // Loop iterates over fraction digits, starting with the LSD. + // include both real digits from the number, and zeros + // to the left of the MSD but to the right of the decimal point. + if (pad && useSpaces) { + toInsertInto.insert(_pos + getPos(), gSpace); + } else { + pad = TRUE; + } + int64_t digit = dl.getDigit(didx); + getRuleSet()->format(digit, toInsertInto, _pos + getPos(), recursionCount, status); + } + + if (!pad) { + // hack around lack of precision in digitlist. if we would end up with + // "foo point" make sure we add a " zero" to the end. + getRuleSet()->format((int64_t)0, toInsertInto, _pos + getPos(), recursionCount, status); + } + } +} + +//----------------------------------------------------------------------- +// parsing +//----------------------------------------------------------------------- + +/** + * If in "by digits" mode, parses the string as if it were a string + * of individual digits; otherwise, uses the superclass function. + * @param text The string to parse + * @param parsePosition Ignored on entry, but updated on exit to point + * to the first unmatched character + * @param baseValue The partial parse result prior to entering this + * function + * @param upperBound Only consider rules with base values lower than + * this when filling in the substitution + * @param lenientParse If true, try matching the text as numerals if + * matching as words doesn't work + * @return If the match was successful, the current partial parse + * result; otherwise new Long(0). The result is either a Long or + * a Double. + */ + +UBool +FractionalPartSubstitution::doParse(const UnicodeString& text, + ParsePosition& parsePosition, + double baseValue, + double /*upperBound*/, + UBool lenientParse, + uint32_t nonNumericalExecutedRuleMask, + Formattable& resVal) const +{ + // if we're not in byDigits mode, we can just use the inherited + // doParse() + if (!byDigits) { + return NFSubstitution::doParse(text, parsePosition, baseValue, 0, lenientParse, nonNumericalExecutedRuleMask, resVal); + + // if we ARE in byDigits mode, parse the text one digit at a time + // using this substitution's owning rule set (we do this by setting + // upperBound to 10 when calling doParse() ) until we reach + // nonmatching text + } else { + UnicodeString workText(text); + ParsePosition workPos(1); + double result = 0; + int32_t digit; +// double p10 = 0.1; + + DecimalQuantity dl; + int32_t totalDigits = 0; + NumberFormat* fmt = NULL; + while (workText.length() > 0 && workPos.getIndex() != 0) { + workPos.setIndex(0); + Formattable temp; + getRuleSet()->parse(workText, workPos, 10, nonNumericalExecutedRuleMask, temp); + UErrorCode status = U_ZERO_ERROR; + digit = temp.getLong(status); +// digit = temp.getType() == Formattable::kLong ? +// temp.getLong() : +// (int32_t)temp.getDouble(); + + if (lenientParse && workPos.getIndex() == 0) { + if (!fmt) { + status = U_ZERO_ERROR; + fmt = NumberFormat::createInstance(status); + if (U_FAILURE(status)) { + delete fmt; + fmt = NULL; + } + } + if (fmt) { + fmt->parse(workText, temp, workPos); + digit = temp.getLong(status); + } + } + + if (workPos.getIndex() != 0) { + dl.appendDigit(static_cast<int8_t>(digit), 0, true); + totalDigits++; +// result += digit * p10; +// p10 /= 10; + parsePosition.setIndex(parsePosition.getIndex() + workPos.getIndex()); + workText.removeBetween(0, workPos.getIndex()); + while (workText.length() > 0 && workText.charAt(0) == gSpace) { + workText.removeBetween(0, 1); + parsePosition.setIndex(parsePosition.getIndex() + 1); + } + } + } + delete fmt; + + dl.adjustMagnitude(-totalDigits); + result = dl.toDouble(); + result = composeRuleValue(result, baseValue); + resVal.setDouble(result); + return TRUE; + } +} + +UBool +FractionalPartSubstitution::operator==(const NFSubstitution& rhs) const +{ + return NFSubstitution::operator==(rhs) && + ((const FractionalPartSubstitution*)&rhs)->byDigits == byDigits; +} + +UOBJECT_DEFINE_RTTI_IMPLEMENTATION(FractionalPartSubstitution) + + +//=================================================================== +// AbsoluteValueSubstitution +//=================================================================== + +UOBJECT_DEFINE_RTTI_IMPLEMENTATION(AbsoluteValueSubstitution) + +//=================================================================== +// NumeratorSubstitution +//=================================================================== + +void +NumeratorSubstitution::doSubstitution(double number, UnicodeString& toInsertInto, int32_t apos, int32_t recursionCount, UErrorCode& status) const { + // perform a transformation on the number being formatted that + // is dependent on the type of substitution this is + + double numberToFormat = transformNumber(number); + int64_t longNF = util64_fromDouble(numberToFormat); + + const NFRuleSet* aruleSet = getRuleSet(); + if (withZeros && aruleSet != NULL) { + // if there are leading zeros in the decimal expansion then emit them + int64_t nf =longNF; + int32_t len = toInsertInto.length(); + while ((nf *= 10) < denominator) { + toInsertInto.insert(apos + getPos(), gSpace); + aruleSet->format((int64_t)0, toInsertInto, apos + getPos(), recursionCount, status); + } + apos += toInsertInto.length() - len; + } + + // if the result is an integer, from here on out we work in integer + // space (saving time and memory and preserving accuracy) + if (numberToFormat == longNF && aruleSet != NULL) { + aruleSet->format(longNF, toInsertInto, apos + getPos(), recursionCount, status); + + // if the result isn't an integer, then call either our rule set's + // format() method or our DecimalFormat's format() method to + // format the result + } else { + if (aruleSet != NULL) { + aruleSet->format(numberToFormat, toInsertInto, apos + getPos(), recursionCount, status); + } else { + UnicodeString temp; + getNumberFormat()->format(numberToFormat, temp, status); + toInsertInto.insert(apos + getPos(), temp); + } + } +} + +UBool +NumeratorSubstitution::doParse(const UnicodeString& text, + ParsePosition& parsePosition, + double baseValue, + double upperBound, + UBool /*lenientParse*/, + uint32_t nonNumericalExecutedRuleMask, + Formattable& result) const +{ + // we don't have to do anything special to do the parsing here, + // but we have to turn lenient parsing off-- if we leave it on, + // it SERIOUSLY messes up the algorithm + + // if withZeros is true, we need to count the zeros + // and use that to adjust the parse result + UErrorCode status = U_ZERO_ERROR; + int32_t zeroCount = 0; + UnicodeString workText(text); + + if (withZeros) { + ParsePosition workPos(1); + Formattable temp; + + while (workText.length() > 0 && workPos.getIndex() != 0) { + workPos.setIndex(0); + getRuleSet()->parse(workText, workPos, 1, nonNumericalExecutedRuleMask, temp); // parse zero or nothing at all + if (workPos.getIndex() == 0) { + // we failed, either there were no more zeros, or the number was formatted with digits + // either way, we're done + break; + } + + ++zeroCount; + parsePosition.setIndex(parsePosition.getIndex() + workPos.getIndex()); + workText.remove(0, workPos.getIndex()); + while (workText.length() > 0 && workText.charAt(0) == gSpace) { + workText.remove(0, 1); + parsePosition.setIndex(parsePosition.getIndex() + 1); + } + } + + workText = text; + workText.remove(0, (int32_t)parsePosition.getIndex()); + parsePosition.setIndex(0); + } + + // we've parsed off the zeros, now let's parse the rest from our current position + NFSubstitution::doParse(workText, parsePosition, withZeros ? 1 : baseValue, upperBound, FALSE, nonNumericalExecutedRuleMask, result); + + if (withZeros) { + // any base value will do in this case. is there a way to + // force this to not bother trying all the base values? + + // compute the 'effective' base and prescale the value down + int64_t n = result.getLong(status); // force conversion! + int64_t d = 1; + int32_t pow = 0; + while (d <= n) { + d *= 10; + ++pow; + } + // now add the zeros + while (zeroCount > 0) { + d *= 10; + --zeroCount; + } + // d is now our true denominator + result.setDouble((double)n/(double)d); + } + + return TRUE; +} + +UBool +NumeratorSubstitution::operator==(const NFSubstitution& rhs) const +{ + return NFSubstitution::operator==(rhs) && + denominator == ((const NumeratorSubstitution*)&rhs)->denominator; +} + +UOBJECT_DEFINE_RTTI_IMPLEMENTATION(NumeratorSubstitution) + +const UChar NumeratorSubstitution::LTLT[] = { 0x003c, 0x003c }; + +U_NAMESPACE_END + +/* U_HAVE_RBNF */ +#endif + |