double
DigitList::getDouble() /*const*/
{
double value;
if (fCount == 0) {
value = 0.0;
}
else {
char* end = NULL;
if (!gDecimal) {
char rep[MAX_DIGITS];
// For machines that decide to change the decimal on you,
// and try to be too smart with localization.
// This normally should be just a '.'.
sprintf(rep, "%+1.1f", 1.0);
gDecimal = rep[2];
}
*fDecimalDigits = gDecimal;
*(fDigits+fCount) = 'e'; // add an e after the digits.
formatBase10(fDecimalAt,
fDigits + fCount + 1, // skip the 'e'
MAX_DEC_DIGITS - fCount - 3); // skip the 'e' and '.'
value = uprv_strtod(fDecimalDigits, &end);
}
return fIsPositive ? value : -value;
}
/**
* Currently, getDouble() depends on atof() to do its conversion.
*
* WARNING!!
* This is an extremely costly function. ~1/2 of the conversion time
* can be linked to this function.
*/
double
DigitList::getDouble() const
{
// TODO: fix thread safety. Can probably be finessed some by analyzing
// what public const functions can see which DigitLists.
// Like precompute fDouble for DigitLists coming in from a parse
// or from a Formattable::set(), but not for any others.
if (fHaveDouble) {
return fDouble;
}
DigitList *nonConstThis = const_cast<DigitList *>(this);
if (isZero()) {
nonConstThis->fDouble = 0.0;
if (decNumberIsNegative(fDecNumber)) {
nonConstThis->fDouble /= -1;
}
} else if (isInfinite()) {
if (std::numeric_limits<double>::has_infinity) {
nonConstThis->fDouble = std::numeric_limits<double>::infinity();
} else {
nonConstThis->fDouble = std::numeric_limits<double>::max();
}
if (!isPositive()) {
nonConstThis->fDouble = -fDouble;
}
} else {
MaybeStackArray<char, MAX_DBL_DIGITS+18> s;
// Note: 14 is a magic constant from the decNumber library documentation,
// the max number of extra characters beyond the number of digits
// needed to represent the number in string form. Add a few more
// for the additional digits we retain.
// Round down to appx. double precision, if the number is longer than that.
// Copy the number first, so that we don't modify the original.
if (getCount() > MAX_DBL_DIGITS + 3) {
DigitList numToConvert(*this);
numToConvert.reduce(); // Removes any trailing zeros, so that digit count is good.
numToConvert.round(MAX_DBL_DIGITS+3);
uprv_decNumberToString(numToConvert.fDecNumber, s);
// TODO: how many extra digits should be included for an accurate conversion?
} else {
uprv_decNumberToString(this->fDecNumber, s);
}
U_ASSERT(uprv_strlen(&s[0]) < MAX_DBL_DIGITS+18);
loadDecimalChar();
if (gDecimal != '.') {
char *decimalPt = strchr(s, '.');
if (decimalPt != NULL) {
*decimalPt = gDecimal;
}
}
char *end = NULL;
nonConstThis->fDouble = uprv_strtod(s, &end);
}
nonConstThis->fHaveDouble = TRUE;
return fDouble;
}
/**
* Convert a string to a double value
*/
double
ChoiceFormat::stod(const UnicodeString& string)
{
char source[256];
char* end;
string.extract(0, string.length(), source, sizeof(source), ""); /* invariant codepage */
return uprv_strtod(source,&end);
}
U_CDECL_END
double
DigitList::decimalStrToDouble(char *decstr, char **end) {
umtx_initOnce(gCLocaleInitOnce, &initCLocale);
#if U_USE_STRTOD_L
return strtod_l(decstr, end, gCLocale);
#else
char *decimalPt = strchr(decstr, '.');
if (decimalPt) {
// We need to know the decimal separator character that will be used with strtod().
// Depends on the C runtime global locale.
// Most commonly is '.'
char rep[MAX_DIGITS];
sprintf(rep, "%+1.1f", 1.0);
*decimalPt = rep[2];
}
return uprv_strtod(decstr, end);
#endif
}
UnicodeSet&
UnicodeSet::applyPropertyAlias(const UnicodeString& prop,
const UnicodeString& value,
UErrorCode& ec) {
if (U_FAILURE(ec) || isFrozen()) return *this;
// prop and value used to be converted to char * using the default
// converter instead of the invariant conversion.
// This should not be necessary because all Unicode property and value
// names use only invariant characters.
// If there are any variant characters, then we won't find them anyway.
// Checking first avoids assertion failures in the conversion.
if( !uprv_isInvariantUString(prop.getBuffer(), prop.length()) ||
!uprv_isInvariantUString(value.getBuffer(), value.length())
) {
FAIL(ec);
}
CharString pname, vname;
pname.appendInvariantChars(prop, ec);
vname.appendInvariantChars(value, ec);
if (U_FAILURE(ec)) return *this;
UProperty p;
int32_t v;
UBool mustNotBeEmpty = FALSE, invert = FALSE;
if (value.length() > 0) {
p = u_getPropertyEnum(pname.data());
if (p == UCHAR_INVALID_CODE) FAIL(ec);
// Treat gc as gcm
if (p == UCHAR_GENERAL_CATEGORY) {
p = UCHAR_GENERAL_CATEGORY_MASK;
}
if ((p >= UCHAR_BINARY_START && p < UCHAR_BINARY_LIMIT) ||
(p >= UCHAR_INT_START && p < UCHAR_INT_LIMIT) ||
(p >= UCHAR_MASK_START && p < UCHAR_MASK_LIMIT)) {
v = u_getPropertyValueEnum(p, vname.data());
if (v == UCHAR_INVALID_CODE) {
// Handle numeric CCC
if (p == UCHAR_CANONICAL_COMBINING_CLASS ||
p == UCHAR_TRAIL_CANONICAL_COMBINING_CLASS ||
p == UCHAR_LEAD_CANONICAL_COMBINING_CLASS) {
char* end;
double value = uprv_strtod(vname.data(), &end);
v = (int32_t) value;
if (v != value || v < 0 || *end != 0) {
// non-integral or negative value, or trailing junk
FAIL(ec);
}
// If the resultant set is empty then the numeric value
// was invalid.
mustNotBeEmpty = TRUE;
} else {
FAIL(ec);
}
}
}
else {
switch (p) {
case UCHAR_NUMERIC_VALUE:
{
char* end;
double value = uprv_strtod(vname.data(), &end);
if (*end != 0) {
FAIL(ec);
}
applyFilter(numericValueFilter, &value, UPROPS_SRC_CHAR, ec);
return *this;
}
case UCHAR_NAME:
{
// Must munge name, since u_charFromName() does not do
// 'loose' matching.
char buf[128]; // it suffices that this be > uprv_getMaxCharNameLength
if (!mungeCharName(buf, vname.data(), sizeof(buf))) FAIL(ec);
UChar32 ch = u_charFromName(U_EXTENDED_CHAR_NAME, buf, &ec);
if (U_SUCCESS(ec)) {
clear();
add(ch);
return *this;
} else {
FAIL(ec);
}
}
case UCHAR_UNICODE_1_NAME:
// ICU 49 deprecates the Unicode_1_Name property APIs.
FAIL(ec);
case UCHAR_AGE:
{
// Must munge name, since u_versionFromString() does not do
// 'loose' matching.
char buf[128];
if (!mungeCharName(buf, vname.data(), sizeof(buf))) FAIL(ec);
UVersionInfo version;
u_versionFromString(version, buf);
applyFilter(versionFilter, &version, UPROPS_SRC_PROPSVEC, ec);
return *this;
}
case UCHAR_SCRIPT_EXTENSIONS:
v = u_getPropertyValueEnum(UCHAR_SCRIPT, vname.data());
if (v == UCHAR_INVALID_CODE) {
FAIL(ec);
}
// fall through to calling applyIntPropertyValue()
break;
default:
// p is a non-binary, non-enumerated property that we
// don't support (yet).
FAIL(ec);
}
}
}
else {
// value is empty. Interpret as General Category, Script, or
// Binary property.
p = UCHAR_GENERAL_CATEGORY_MASK;
v = u_getPropertyValueEnum(p, pname.data());
if (v == UCHAR_INVALID_CODE) {
p = UCHAR_SCRIPT;
v = u_getPropertyValueEnum(p, pname.data());
if (v == UCHAR_INVALID_CODE) {
p = u_getPropertyEnum(pname.data());
if (p >= UCHAR_BINARY_START && p < UCHAR_BINARY_LIMIT) {
v = 1;
} else if (0 == uprv_comparePropertyNames(ANY, pname.data())) {
set(MIN_VALUE, MAX_VALUE);
return *this;
} else if (0 == uprv_comparePropertyNames(ASCII, pname.data())) {
set(0, 0x7F);
return *this;
} else if (0 == uprv_comparePropertyNames(ASSIGNED, pname.data())) {
// [:Assigned:]=[:^Cn:]
p = UCHAR_GENERAL_CATEGORY_MASK;
v = U_GC_CN_MASK;
invert = TRUE;
} else {
FAIL(ec);
}
}
}
}
applyIntPropertyValue(p, v, ec);
if(invert) {
complement();
}
if (U_SUCCESS(ec) && (mustNotBeEmpty && isEmpty())) {
// mustNotBeEmpty is set to true if an empty set indicates
// invalid input.
ec = U_ILLEGAL_ARGUMENT_ERROR;
}
if (isBogus() && U_SUCCESS(ec)) {
// We likely ran out of memory. AHHH!
ec = U_MEMORY_ALLOCATION_ERROR;
}
return *this;
}
/**
* Currently, getDouble() depends on strtod() to do its conversion.
*
* WARNING!!
* This is an extremely costly function. ~1/2 of the conversion time
* can be linked to this function.
*/
double
DigitList::getDouble() const
{
static char gDecimal = 0;
char decimalSeparator;
{
Mutex mutex;
if (fHave == kDouble) {
return fUnion.fDouble;
} else if(fHave == kInt64) {
return (double)fUnion.fInt64;
}
decimalSeparator = gDecimal;
}
if (decimalSeparator == 0) {
// We need to know the decimal separator character that will be used with strtod().
// Depends on the C runtime global locale.
// Most commonly is '.'
// TODO: caching could fail if the global locale is changed on the fly.
char rep[MAX_DIGITS];
sprintf(rep, "%+1.1f", 1.0);
decimalSeparator = rep[2];
}
double tDouble = 0.0;
if (isZero()) {
tDouble = 0.0;
if (decNumberIsNegative(fDecNumber)) {
tDouble /= -1;
}
} else if (isInfinite()) {
if (std::numeric_limits<double>::has_infinity) {
tDouble = std::numeric_limits<double>::infinity();
} else {
tDouble = std::numeric_limits<double>::max();
}
if (!isPositive()) {
tDouble = -tDouble; //this was incorrectly "-fDouble" originally.
}
} else {
MaybeStackArray<char, MAX_DBL_DIGITS+18> s;
// Note: 14 is a magic constant from the decNumber library documentation,
// the max number of extra characters beyond the number of digits
// needed to represent the number in string form. Add a few more
// for the additional digits we retain.
// Round down to appx. double precision, if the number is longer than that.
// Copy the number first, so that we don't modify the original.
if (getCount() > MAX_DBL_DIGITS + 3) {
DigitList numToConvert(*this);
numToConvert.reduce(); // Removes any trailing zeros, so that digit count is good.
numToConvert.round(MAX_DBL_DIGITS+3);
uprv_decNumberToString(numToConvert.fDecNumber, s);
// TODO: how many extra digits should be included for an accurate conversion?
} else {
uprv_decNumberToString(this->fDecNumber, s);
}
U_ASSERT(uprv_strlen(&s[0]) < MAX_DBL_DIGITS+18);
if (decimalSeparator != '.') {
char *decimalPt = strchr(s, '.');
if (decimalPt != NULL) {
*decimalPt = decimalSeparator;
}
}
char *end = NULL;
tDouble = uprv_strtod(s, &end);
}
{
Mutex mutex;
DigitList *nonConstThis = const_cast<DigitList *>(this);
nonConstThis->internalSetDouble(tDouble);
gDecimal = decimalSeparator;
}
return tDouble;
}
// populatePower10 grabs data for a particular power of 10 from CLDR.
// The loaded data is stored in result.
static void populatePower10(const UResourceBundle* power10Bundle, CDFLocaleStyleData* result, UErrorCode& status) {
if (U_FAILURE(status)) {
return;
}
char* endPtr = NULL;
double power10 = uprv_strtod(ures_getKey(power10Bundle), &endPtr);
if (*endPtr != 0) {
status = U_INTERNAL_PROGRAM_ERROR;
return;
}
int32_t log10Value = computeLog10(power10, FALSE);
// Silently ignore divisors that are too big.
if (log10Value == MAX_DIGITS) {
return;
}
int32_t size = ures_getSize(power10Bundle);
int32_t numZeros = 0;
UBool otherVariantDefined = FALSE;
UResourceBundle* variantBundle = NULL;
// Iterate over all the plural variants for the power of 10
for (int32_t i = 0; i < size; ++i) {
variantBundle = ures_getByIndex(power10Bundle, i, variantBundle, &status);
if (U_FAILURE(status)) {
ures_close(variantBundle);
return;
}
const char* variant = ures_getKey(variantBundle);
int32_t resLen;
const UChar* formatStrP = ures_getString(variantBundle, &resLen, &status);
if (U_FAILURE(status)) {
ures_close(variantBundle);
return;
}
UnicodeString formatStr(false, formatStrP, resLen);
if (uprv_strcmp(variant, gOther) == 0) {
otherVariantDefined = TRUE;
}
int32_t nz = populatePrefixSuffix(
variant, log10Value, formatStr, result->unitsByVariant, status);
if (U_FAILURE(status)) {
ures_close(variantBundle);
return;
}
if (nz != numZeros) {
// We expect all format strings to have the same number of 0's
// left of the decimal point.
if (numZeros != 0) {
status = U_INTERNAL_PROGRAM_ERROR;
ures_close(variantBundle);
return;
}
numZeros = nz;
}
}
ures_close(variantBundle);
// We expect to find an OTHER variant for each power of 10.
if (!otherVariantDefined) {
status = U_INTERNAL_PROGRAM_ERROR;
return;
}
double divisor = power10;
for (int32_t i = 1; i < numZeros; ++i) {
divisor /= 10.0;
}
result->divisors[log10Value] = divisor;
}
void
MessagePattern::parseDouble(int32_t start, int32_t limit, UBool allowInfinity,
UParseError *parseError, UErrorCode &errorCode) {
if(U_FAILURE(errorCode)) {
return;
}
U_ASSERT(start<limit);
// fake loop for easy exit and single throw statement
for(;;) { /*loop doesn't iterate*/
// fast path for small integers and infinity
int32_t value=0;
int32_t isNegative=0; // not boolean so that we can easily add it to value
int32_t index=start;
UChar c=msg.charAt(index++);
if(c==u_minus) {
isNegative=1;
if(index==limit) {
break; // no number
}
c=msg.charAt(index++);
} else if(c==u_plus) {
if(index==limit) {
break; // no number
}
c=msg.charAt(index++);
}
if(c==0x221e) { // infinity
if(allowInfinity && index==limit) {
double infinity=uprv_getInfinity();
addArgDoublePart(
isNegative!=0 ? -infinity : infinity,
start, limit-start, errorCode);
return;
} else {
break;
}
}
// try to parse the number as a small integer but fall back to a double
while('0'<=c && c<='9') {
value=value*10+(c-'0');
if(value>(Part::MAX_VALUE+isNegative)) {
break; // not a small-enough integer
}
if(index==limit) {
addPart(UMSGPAT_PART_TYPE_ARG_INT, start, limit-start,
isNegative!=0 ? -value : value, errorCode);
return;
}
c=msg.charAt(index++);
}
// Let Double.parseDouble() throw a NumberFormatException.
char numberChars[128];
int32_t capacity=(int32_t)sizeof(numberChars);
int32_t length=limit-start;
if(length>=capacity) {
break; // number too long
}
msg.extract(start, length, numberChars, capacity, US_INV);
if((int32_t)uprv_strlen(numberChars)<length) {
break; // contains non-invariant character that was turned into NUL
}
char *end;
double numericValue=uprv_strtod(numberChars, &end);
if(end!=(numberChars+length)) {
break; // parsing error
}
addArgDoublePart(numericValue, start, length, errorCode);
return;
}
setParseError(parseError, start /*, limit*/); // Bad syntax for numeric value.
errorCode=U_PATTERN_SYNTAX_ERROR;
return;
}