scuffed-code/icu4c/source/i18n/choicfmt.cpp
1999-10-12 00:19:25 +00:00

588 lines
17 KiB
C++

/*
********************************************************************************
* *
* COPYRIGHT: *
* (C) Copyright Taligent, Inc., 1997 *
* (C) Copyright International Business Machines Corporation, 1997-1998 *
* Licensed Material - Program-Property of IBM - All Rights Reserved. *
* US Government Users Restricted Rights - Use, duplication, or disclosure *
* restricted by GSA ADP Schedule Contract with IBM Corp. *
* *
********************************************************************************
*
* File CHOICFMT.CPP
*
* Modification History:
*
* Date Name Description
* 02/19/97 aliu Converted from java.
* 03/20/97 helena Finished first cut of implementation and got rid
* of nextDouble/previousDouble and replaced with
* boolean array.
* 4/10/97 aliu Clean up. Modified to work on AIX.
* 06/04/97 helena Fixed applyPattern(), toPattern() and not to include
* wchar.h.
* 07/09/97 helena Made ParsePosition into a class.
* 08/06/97 nos removed overloaded constructor, fixed 'format(array)'
* 07/22/98 stephen JDK 1.2 Sync - removed bool_t array (doubleFlags)
* 02/22/99 stephen Removed character literals for EBCDIC safety
********************************************************************************
*/
#include "cpputils.h"
#include "choicfmt.h"
#include "numfmt.h"
#include "locid.h"
#include "mutex.h"
// *****************************************************************************
// class ChoiceFormat
// *****************************************************************************
char ChoiceFormat::fgClassID = 0; // Value is irrelevant
NumberFormat* ChoiceFormat::fgNumberFormat = 0;
// -------------------------------------
// Creates a ChoiceFormat instance based on the pattern.
ChoiceFormat::ChoiceFormat(const UnicodeString& newPattern,
UErrorCode& status)
: fChoiceLimits(0),
fChoiceFormats(0),
fCount(0)
{
applyPattern(newPattern, status);
}
// -------------------------------------
// Creates a ChoiceFormat instance with the limit array and
// format strings for each limit.
ChoiceFormat::ChoiceFormat(const double* limits,
const UnicodeString* formats,
int32_t cnt )
: fChoiceLimits(0),
fChoiceFormats(0),
fCount(0)
{
setChoices(limits, formats, cnt );
}
// -------------------------------------
// copy constructor
ChoiceFormat::ChoiceFormat(const ChoiceFormat& that)
: fChoiceLimits(0),
fChoiceFormats(0)
{
*this = that;
}
// -------------------------------------
bool_t
ChoiceFormat::operator==(const Format& that) const
{
if (this == &that) return TRUE;
if (this->getDynamicClassID() != that.getDynamicClassID()) return FALSE; // not the same class
if (!NumberFormat::operator==(that)) return FALSE;
ChoiceFormat& thatAlias = (ChoiceFormat&)that;
if (fCount != thatAlias.fCount) return FALSE;
// Checks the limits, the corresponding format string and LE or LT flags.
// LE means less than and equal to, LT means less than.
for (int32_t i = 0; i < fCount; i++) {
if ((fChoiceLimits[i] != thatAlias.fChoiceLimits[i]) ||
(fChoiceFormats[i] != thatAlias.fChoiceFormats[i]))
return FALSE;
}
return TRUE;
}
// -------------------------------------
// copy constructor
const ChoiceFormat&
ChoiceFormat::operator=(const ChoiceFormat& that)
{
if (this != &that) {
NumberFormat::operator=(that);
fCount = that.fCount;
delete [] fChoiceLimits; fChoiceLimits = 0;
delete [] fChoiceFormats; fChoiceFormats = 0;
fChoiceLimits = new double[fCount];
fChoiceFormats = new UnicodeString[fCount];
icu_arrayCopy(that.fChoiceLimits, fChoiceLimits, fCount);
icu_arrayCopy(that.fChoiceFormats, fChoiceFormats, fCount);
}
return *this;
}
// -------------------------------------
ChoiceFormat::~ChoiceFormat()
{
delete [] fChoiceLimits;
fChoiceLimits = 0;
delete [] fChoiceFormats;
fChoiceFormats = 0;
fCount = 0;
}
// -------------------------------------
// NumberFormat cache management
NumberFormat*
ChoiceFormat::getNumberFormat(UErrorCode &status)
{
NumberFormat *theFormat = 0;
if (fgNumberFormat != 0) // if there's something in the cache
{
Mutex lock;
if (fgNumberFormat != 0) // Someone might have grabbed it.
{
theFormat = fgNumberFormat;
fgNumberFormat = 0; // We have exclusive right to this formatter.
}
}
if(theFormat == 0) // If we weren't able to pull it out of the cache, then we have to create it.
{
theFormat = NumberFormat::createInstance(Locale::US, status);
if(FAILURE(status))
return 0;
theFormat->setMinimumFractionDigits(1);
}
return theFormat;
}
void
ChoiceFormat::releaseNumberFormat(NumberFormat *adopt)
{
if(fgNumberFormat == 0) // If the cache is empty we must add it back.
{
Mutex lock;
if(fgNumberFormat == 0)
{
fgNumberFormat = adopt;
adopt = 0;
}
}
delete adopt;
}
/**
* Convert a string to a double value using a default NumberFormat object
* which is static (shared by all ChoiceFormat instances).
*/
double
ChoiceFormat::stod(const UnicodeString& string,
UErrorCode& status)
{
// Use a shared global number format to convert a double value to
// or string or vice versa.
NumberFormat *myFormat = getNumberFormat(status);
if(FAILURE(status))
return -1; // OK?
Formattable result;
myFormat->parse(string, result, status);
releaseNumberFormat(myFormat);
double value = 0.0;
if (SUCCESS(status))
{
switch(result.getType())
{
case Formattable::kLong: value = result.getLong(); break;
case Formattable::kDouble: value = result.getDouble(); break;
}
}
return value;
}
// -------------------------------------
/**
* Convert a double value to a string using a default NumberFormat object
* which is static (shared by all ChoiceFormat instances).
*/
UnicodeString&
ChoiceFormat::dtos(double value,
UnicodeString& string,
UErrorCode& status)
{
NumberFormat *myFormat = getNumberFormat(status);
if (SUCCESS(status)) {
FieldPosition fieldPos(0);
myFormat->format(value, string, fieldPos);
}
releaseNumberFormat(myFormat);
return string;
}
// -------------------------------------
// Applies the pattern to this ChoiceFormat instance.
void
ChoiceFormat::applyPattern(const UnicodeString& newPattern,
UErrorCode& status)
{
if (FAILURE(status))
return;
UnicodeString segments[2];
double newChoiceLimits[30]; // current limit
UnicodeString newChoiceFormats[30]; // later, use Vectors
int32_t count = 0;
int32_t part = 0;
double startValue = 0;
double oldStartValue = icu_getNaN();
bool_t inQuote = FALSE;
for(int i = 0; i < newPattern.size(); ++i) {
UChar ch = newPattern[i];
if(ch == 0x0027 /*'\''*/) {
// Check for "''" indicating a literal quote
if((i+1) < newPattern.size() && newPattern[i+1] == ch) {
segments[part] += ch;
++i;
}
else
inQuote = !inQuote;
}
else if (inQuote) {
segments[part] += ch;
}
else if (ch == 0x003C /*'<'*/ || ch == 0x0023 /*'#'*/ || ch == 0x2264) {
if (segments[0] == "") {
status = U_ILLEGAL_ARGUMENT_ERROR;
return;
}
UnicodeString tempBuffer = segments[0];
tempBuffer.trim();
UChar posInf = 0x221E;
UChar negInf [] = {0x002D /*'-'*/, posInf };
if (tempBuffer == UnicodeString(&posInf, 1, 1)) {
startValue = icu_getInfinity();
}
else if (tempBuffer == UnicodeString(negInf, 2, 2)) {
startValue = - icu_getInfinity();
}
else {
//segments[0].trim();
startValue = stod(tempBuffer, status);
if(FAILURE(status))
return;
}
if (ch == 0x003C /*'<'*/ && ! icu_isInfinite(startValue)) {
startValue = nextDouble(startValue);
}
// {sfb} There is a bug in MSVC 5.0 sp3 -- 0.0 <= NaN ==> TRUE
//if (startValue <= oldStartValue) {
if (startValue <= oldStartValue && ! icu_isNaN(oldStartValue)) {
status = U_ILLEGAL_ARGUMENT_ERROR;
return;
}
segments[0].remove();
part = 1;
} else if (ch == 0x007C /*'|'*/) {
newChoiceLimits[count] = startValue;
newChoiceFormats[count] = segments[1];
++count;
oldStartValue = startValue;
segments[1].remove();
part = 0;
} else {
segments[part] += ch;
}
}
// clean up last one
if (part == 1) {
newChoiceLimits[count] = startValue;
newChoiceFormats[count] = segments[1];
++count;
}
delete [] fChoiceLimits; fChoiceLimits = 0;
delete [] fChoiceFormats; fChoiceFormats = 0;
fCount = count;
fChoiceLimits = new double[fCount];
fChoiceFormats = new UnicodeString[fCount];
icu_arrayCopy(newChoiceLimits, fChoiceLimits, fCount);
icu_arrayCopy(newChoiceFormats, fChoiceFormats, fCount);
}
// -------------------------------------
// Reconstruct the original input pattern.
UnicodeString&
ChoiceFormat::toPattern(UnicodeString& result) const
{
result.remove();
for (int32_t i = 0; i < fCount; ++i) {
if (i != 0) {
result += 0x007C /*'|'*/;
}
// choose based upon which has less precision
// approximate that by choosing the closest one to an integer.
// could do better, but it's not worth it.
double less = previousDouble(fChoiceLimits[i]);
double tryLessOrEqual = icu_fabs(icu_IEEEremainder(fChoiceLimits[i], 1.0));
double tryLess = icu_fabs(icu_IEEEremainder(less, 1.0));
UErrorCode status = U_ZERO_ERROR;
UnicodeString buf;
// {sfb} hack to get this to work on MSVC - NaN doesn't behave as it should
if (tryLessOrEqual < tryLess &&
! (icu_isNaN(tryLessOrEqual) || icu_isNaN(tryLess))) {
result += dtos(fChoiceLimits[i], buf, status);
result += 0x0023 /*'#'*/;
}
else {
if (icu_isPositiveInfinity(fChoiceLimits[i])) {
result += 0x221E;
} else if (icu_isNegativeInfinity(fChoiceLimits[i])) {
result += 0x002D /*'-'*/;
result += 0x221E;
} else {
result += dtos(less, buf, status);
}
result += 0x003C /*'<'*/;
}
// Append fChoiceFormats[i], using quotes if there are special characters.
// Single quotes themselves must be escaped in either case.
UnicodeString text = fChoiceFormats[i];
bool_t needQuote = text.indexOf(0x003C /*'<'*/) >= 0
|| text.indexOf(0x0023 /*'#'*/) >= 0
|| text.indexOf(0x2264) >= 0
|| text.indexOf(0x007C /*'|'*/) >= 0;
if (needQuote)
result += 0x0027 /*'\''*/;
if (text.indexOf(0x0027 /*'\''*/) < 0)
result += text;
else {
for (int j = 0; j < text.size(); ++j) {
UChar c = text[j];
result += c;
if (c == 0x0027 /*'\''*/)
result += c;
}
}
if (needQuote)
result += 0x0027 /*'\''*/;
}
return result;
}
// -------------------------------------
// Adopts the limit and format arrays.
void
ChoiceFormat::adoptChoices(double *limits,
UnicodeString *formats,
int32_t cnt )
{
if(limits == 0 || formats == 0)
return;
delete [] fChoiceLimits;
fChoiceLimits = 0;
delete [] fChoiceFormats;
fChoiceFormats = 0;
fChoiceLimits = limits;
fChoiceFormats = formats;
fCount = cnt;
}
// -------------------------------------
// Sets the limit and format arrays.
void
ChoiceFormat::setChoices( const double* limits,
const UnicodeString* formats,
int32_t cnt )
{
if(limits == 0 || formats == 0)
return;
delete [] fChoiceLimits; fChoiceLimits = 0;
delete [] fChoiceFormats; fChoiceFormats = 0;
// Note that the old arrays are deleted and this owns
// the created array.
fCount = cnt;
fChoiceLimits = new double[fCount];
fChoiceFormats = new UnicodeString[fCount];
icu_arrayCopy(limits, fChoiceLimits, fCount);
icu_arrayCopy(formats, fChoiceFormats, fCount);
}
// -------------------------------------
// Gets the limit array.
const double*
ChoiceFormat::getLimits(int32_t& cnt) const
{
cnt = fCount;
return fChoiceLimits;
}
// -------------------------------------
// Gets the format array.
const UnicodeString*
ChoiceFormat::getFormats(int32_t& cnt) const
{
cnt = fCount;
return fChoiceFormats;
}
// -------------------------------------
// Formats a long number, it's actually formatted as
// a double. The returned format string may differ
// from the input number because of this.
UnicodeString&
ChoiceFormat::format(int32_t number,
UnicodeString& toAppendTo,
FieldPosition& status) const
{
return format((double) number, toAppendTo, status);
}
// -------------------------------------
// Formats a double number.
UnicodeString&
ChoiceFormat::format(double number,
UnicodeString& toAppendTo,
FieldPosition& status) const
{
// find the number
int32_t i;
for (i = 0; i < fCount; ++i) {
if (!(number >= fChoiceLimits[i])) {
// same as number < fChoiceLimits, except catches NaN
break;
}
}
--i;
if (i < 0)
i = 0;
// return either a formatted number, or a string
return (toAppendTo += fChoiceFormats[i]);
}
// -------------------------------------
// Formats an array of objects. Checks if the data type of the objects
// to get the right value for formatting.
UnicodeString&
ChoiceFormat::format(const Formattable* objs,
int32_t cnt,
UnicodeString& toAppendTo,
FieldPosition& pos,
UErrorCode& status) const
{
if(cnt < 0) {
status = U_ILLEGAL_ARGUMENT_ERROR;
return toAppendTo;
}
UnicodeString buffer;
for (int32_t i = 0; i < cnt; i++) {
buffer.remove();
toAppendTo += format((objs[i].getType() == Formattable::kLong) ? objs[i].getLong() : objs[i].getDouble(),
buffer, pos);
}
return toAppendTo;
}
// -------------------------------------
// Formats an array of objects. Checks if the data type of the objects
// to get the right value for formatting.
UnicodeString&
ChoiceFormat::format(const Formattable& obj,
UnicodeString& toAppendTo,
FieldPosition& pos,
UErrorCode& status) const
{
return NumberFormat::format(obj, toAppendTo, pos, status);
}
// -------------------------------------
void
ChoiceFormat::parse(const UnicodeString& text,
Formattable& result,
ParsePosition& status) const
{
// find the best number (defined as the one with the longest parse)
int32_t start = status.getIndex();
int32_t furthest = start;
double bestNumber = icu_getNaN();
double tempNumber = 0.0;
for (int i = 0; i < fCount; ++i) {
UnicodeString tempString = fChoiceFormats[i];
if(text.compareBetween(start, tempString.size(), tempString, 0, tempString.size()) == 0) {
status.setIndex(start + tempString.size());
tempNumber = fChoiceLimits[i];
if (status.getIndex() > furthest) {
furthest = status.getIndex();
bestNumber = tempNumber;
if (furthest == text.size())
break;
}
}
}
status.setIndex(furthest);
if (status.getIndex() == start) {
status.setErrorIndex(furthest);
}
result.setDouble(bestNumber);
}
// -------------------------------------
// Parses the text and return the Formattable object.
void
ChoiceFormat::parse(const UnicodeString& text,
Formattable& result,
UErrorCode& status) const
{
NumberFormat::parse(text, result, status);
}
// -------------------------------------
Format*
ChoiceFormat::clone() const
{
ChoiceFormat *aCopy = new ChoiceFormat(*this);
return aCopy;
}
// -------------------------------------
double
ChoiceFormat::nextDouble( double d, bool_t positive )
{
return icu_nextDouble( d, positive );
}
//eof