const NFRule*
NFRuleSet::findDoubleRule(double number) const
{
// if this is a fraction rule set, use findFractionRuleSetRule()
if (isFractionRuleSet()) {
return findFractionRuleSetRule(number);
}
if (uprv_isNaN(number)) {
const NFRule *rule = nonNumericalRules[NAN_RULE_INDEX];
if (!rule) {
rule = owner->getDefaultNaNRule();
}
return rule;
}
// if the number is negative, return the negative number rule
// (if there isn't a negative-number rule, we pretend it's a
// positive number)
if (number < 0) {
if (nonNumericalRules[NEGATIVE_RULE_INDEX]) {
return nonNumericalRules[NEGATIVE_RULE_INDEX];
} else {
number = -number;
}
}
if (uprv_isInfinite(number)) {
const NFRule *rule = nonNumericalRules[INFINITY_RULE_INDEX];
if (!rule) {
rule = owner->getDefaultInfinityRule();
}
return rule;
}
// if the number isn't an integer, we use one of the fraction rules...
if (number != uprv_floor(number)) {
// if the number is between 0 and 1, return the proper
// fraction rule
if (number < 1 && nonNumericalRules[PROPER_FRACTION_RULE_INDEX]) {
return nonNumericalRules[PROPER_FRACTION_RULE_INDEX];
}
// otherwise, return the improper fraction rule
else if (nonNumericalRules[IMPROPER_FRACTION_RULE_INDEX]) {
return nonNumericalRules[IMPROPER_FRACTION_RULE_INDEX];
}
}
// if there's a master rule, use it to format the number
if (nonNumericalRules[MASTER_RULE_INDEX]) {
return nonNumericalRules[MASTER_RULE_INDEX];
}
// and if we haven't yet returned a rule, use findNormalRule()
// to find the applicable rule
int64_t r = util64_fromDouble(number + 0.5);
return findNormalRule(r);
}
NumeratorSubstitution(int32_t _pos,
double _denominator,
const NFRuleSet* _ruleSet,
const RuleBasedNumberFormat* formatter,
const UnicodeString& description,
UErrorCode& status)
: NFSubstitution(_pos, _ruleSet, formatter, fixdesc(description), status), denominator(_denominator)
{
ldenominator = util64_fromDouble(denominator);
withZeros = description.endsWith(LTLT, 2);
}
MultiplierSubstitution(int32_t _pos,
double _divisor,
const NFRuleSet* _ruleSet,
const RuleBasedNumberFormat* formatter,
const UnicodeString& description,
UErrorCode& status)
: NFSubstitution(_pos, _ruleSet, formatter, description, status), divisor(_divisor)
{
ldivisor = util64_fromDouble(divisor);
if (divisor == 0) {
status = U_PARSE_ERROR;
}
}
NFRule *
NFRuleSet::findDoubleRule(double number) const
{
// if this is a fraction rule set, use findFractionRuleSetRule()
if (isFractionRuleSet())
{
return findFractionRuleSetRule(number);
}
// if the number is negative, return the negative number rule
// (if there isn't a negative-number rule, we pretend it's a
// positive number)
if (number < 0)
{
if (negativeNumberRule)
{
return negativeNumberRule;
}
else
{
number = -number;
}
}
// if the number isn't an integer, we use one of the fraction rules...
if (number != uprv_floor(number))
{
// if the number is between 0 and 1, return the proper
// fraction rule
if (number < 1 && fractionRules[1])
{
return fractionRules[1];
}
// otherwise, return the improper fraction rule
else if (fractionRules[0])
{
return fractionRules[0];
}
}
// if there's a master rule, use it to format the number
if (fractionRules[2])
{
return fractionRules[2];
}
// and if we haven't yet returned a rule, use findNormalRule()
// to find the applicable rule
int64_t r = util64_fromDouble(number + 0.5);
return findNormalRule(r);
}
void
NumeratorSubstitution::doSubstitution(double number, UnicodeString& toInsertInto, int32_t apos) 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());
}
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());
// 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());
} else {
UErrorCode status = U_ZERO_ERROR;
UnicodeString temp;
getNumberFormat()->format(numberToFormat, temp, status);
toInsertInto.insert(apos + getPos(), 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) const {
// perform a transformation on the number being formatted that
// is dependent on the type of substitution this is
double numberToFormat = transformNumber(number);
// 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);
// 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);
} else if (numberFormat != NULL) {
UnicodeString temp;
numberFormat->format(numberToFormat, temp);
toInsertInto.insert(_pos + this->pos, temp);
}
}
}
UBool
NFRuleSet::parse(const UnicodeString& text, ParsePosition& pos, double upperBound, Formattable& result) const
{
// try matching each rule in the rule set against the text being
// parsed. Whichever one matches the most characters is the one
// that determines the value we return.
result.setLong(0);
// dump out if there's no text to parse
if (text.length() == 0) {
return 0;
}
ParsePosition highWaterMark;
ParsePosition workingPos = pos;
#ifdef RBNF_DEBUG
fprintf(stderr, "<nfrs> %x '", this);
dumpUS(stderr, name);
fprintf(stderr, "' text '");
dumpUS(stderr, text);
fprintf(stderr, "'\n");
fprintf(stderr, " parse negative: %d\n", this, negativeNumberRule != 0);
#endif
// start by trying the negative number rule (if there is one)
if (negativeNumberRule) {
Formattable tempResult;
#ifdef RBNF_DEBUG
fprintf(stderr, " <nfrs before negative> %x ub: %g\n", negativeNumberRule, upperBound);
#endif
UBool success = negativeNumberRule->doParse(text, workingPos, 0, upperBound, tempResult);
#ifdef RBNF_DEBUG
fprintf(stderr, " <nfrs after negative> success: %d wpi: %d\n", success, workingPos.getIndex());
#endif
if (success && workingPos.getIndex() > highWaterMark.getIndex()) {
result = tempResult;
highWaterMark = workingPos;
}
workingPos = pos;
}
#ifdef RBNF_DEBUG
fprintf(stderr, "<nfrs> continue fractional with text '");
dumpUS(stderr, text);
fprintf(stderr, "' hwm: %d\n", highWaterMark.getIndex());
#endif
// then try each of the fraction rules
{
for (int i = 0; i < 3; i++) {
if (fractionRules[i]) {
Formattable tempResult;
UBool success = fractionRules[i]->doParse(text, workingPos, 0, upperBound, tempResult);
if (success && (workingPos.getIndex() > highWaterMark.getIndex())) {
result = tempResult;
highWaterMark = workingPos;
}
workingPos = pos;
}
}
}
#ifdef RBNF_DEBUG
fprintf(stderr, "<nfrs> continue other with text '");
dumpUS(stderr, text);
fprintf(stderr, "' hwm: %d\n", highWaterMark.getIndex());
#endif
// finally, go through the regular rules one at a time. We start
// at the end of the list because we want to try matching the most
// sigificant rule first (this helps ensure that we parse
// "five thousand three hundred six" as
// "(five thousand) (three hundred) (six)" rather than
// "((five thousand three) hundred) (six)"). Skip rules whose
// base values are higher than the upper bound (again, this helps
// limit ambiguity by making sure the rules that match a rule's
// are less significant than the rule containing the substitutions)/
{
int64_t ub = util64_fromDouble(upperBound);
#ifdef RBNF_DEBUG
{
char ubstr[64];
util64_toa(ub, ubstr, 64);
char ubstrhex[64];
util64_toa(ub, ubstrhex, 64, 16);
fprintf(stderr, "ub: %g, i64: %s (%s)\n", upperBound, ubstr, ubstrhex);
}
#endif
for (int32_t i = rules.size(); --i >= 0 && highWaterMark.getIndex() < text.length();) {
if ((!fIsFractionRuleSet) && (rules[i]->getBaseValue() >= ub)) {
continue;
}
Formattable tempResult;
UBool success = rules[i]->doParse(text, workingPos, fIsFractionRuleSet, upperBound, tempResult);
if (success && workingPos.getIndex() > highWaterMark.getIndex()) {
result = tempResult;
highWaterMark = workingPos;
}
workingPos = pos;
}
}
#ifdef RBNF_DEBUG
fprintf(stderr, "<nfrs> exit\n");
#endif
// finally, update the parse postion we were passed to point to the
// first character we didn't use, and return the result that
// corresponds to that string of characters
pos = highWaterMark;
return 1;
}
/**
* If this rule is a fraction rule set, this function is used by
* findRule() to select the most appropriate rule for formatting
* the number. Basically, the base value of each rule in the rule
* set is treated as the denominator of a fraction. Whichever
* denominator can produce the fraction closest in value to the
* number passed in is the result. If there's a tie, the earlier
* one in the list wins. (If there are two rules in a row with the
* same base value, the first one is used when the numerator of the
* fraction would be 1, and the second rule is used the rest of the
* time.
* @param number The number being formatted (which will always be
* a number between 0 and 1)
* @return The rule to use to format this number
*/
NFRule*
NFRuleSet::findFractionRuleSetRule(double number) const
{
// the obvious way to do this (multiply the value being formatted
// by each rule's base value until you get an integral result)
// doesn't work because of rounding error. This method is more
// accurate
// find the least common multiple of the rules' base values
// and multiply this by the number being formatted. This is
// all the precision we need, and we can do all of the rest
// of the math using integer arithmetic
int64_t leastCommonMultiple = rules[0]->getBaseValue();
int64_t numerator;
{
for (uint32_t i = 1; i < rules.size(); ++i) {
leastCommonMultiple = util_lcm(leastCommonMultiple, rules[i]->getBaseValue());
}
numerator = util64_fromDouble(number * (double)leastCommonMultiple + 0.5);
}
// for each rule, do the following...
int64_t tempDifference;
int64_t difference = util64_fromDouble(uprv_maxMantissa());
int32_t winner = 0;
for (uint32_t i = 0; i < rules.size(); ++i) {
// "numerator" is the numerator of the fraction if the
// denominator is the LCD. The numerator if the rule's
// base value is the denominator is "numerator" times the
// base value divided bythe LCD. Here we check to see if
// that's an integer, and if not, how close it is to being
// an integer.
tempDifference = numerator * rules[i]->getBaseValue() % leastCommonMultiple;
// normalize the result of the above calculation: we want
// the numerator's distance from the CLOSEST multiple
// of the LCD
if (leastCommonMultiple - tempDifference < tempDifference) {
tempDifference = leastCommonMultiple - tempDifference;
}
// if this is as close as we've come, keep track of how close
// that is, and the line number of the rule that did it. If
// we've scored a direct hit, we don't have to look at any more
// rules
if (tempDifference < difference) {
difference = tempDifference;
winner = i;
if (difference == 0) {
break;
}
}
}
// if we have two successive rules that both have the winning base
// value, then the first one (the one we found above) is used if
// the numerator of the fraction is 1 and the second one is used if
// the numerator of the fraction is anything else (this lets us
// do things like "one third"/"two thirds" without haveing to define
// a whole bunch of extra rule sets)
if ((unsigned)(winner + 1) < rules.size() &&
rules[winner + 1]->getBaseValue() == rules[winner]->getBaseValue()) {
double n = ((double)rules[winner]->getBaseValue()) * number;
if (n < 0.5 || n >= 2) {
++winner;
}
}
// finally, return the winning rule
return rules[winner];
}