2017-09-27 00:25:20 +00:00
|
|
|
|
// © 2017 and later: Unicode, Inc. and others.
|
|
|
|
|
// License & terms of use: http://www.unicode.org/copyright.html
|
|
|
|
|
|
2017-10-04 22:51:06 +00:00
|
|
|
|
#include "unicode/utypes.h"
|
|
|
|
|
|
2017-09-27 05:31:57 +00:00
|
|
|
|
#if !UCONFIG_NO_FORMATTING
|
|
|
|
|
#ifndef __NUMBERFORMATTER_H__
|
|
|
|
|
#define __NUMBERFORMATTER_H__
|
|
|
|
|
|
|
|
|
|
#include "unicode/appendable.h"
|
|
|
|
|
#include "unicode/dcfmtsym.h"
|
|
|
|
|
#include "unicode/currunit.h"
|
|
|
|
|
#include "unicode/fieldpos.h"
|
|
|
|
|
#include "unicode/fpositer.h"
|
|
|
|
|
#include "unicode/measunit.h"
|
2018-03-27 01:58:26 +00:00
|
|
|
|
#include "unicode/unumberformatter.h"
|
2017-09-27 05:31:57 +00:00
|
|
|
|
#include "unicode/nounit.h"
|
|
|
|
|
#include "unicode/plurrule.h"
|
|
|
|
|
#include "unicode/ucurr.h"
|
|
|
|
|
#include "unicode/unum.h"
|
2018-03-14 09:15:27 +00:00
|
|
|
|
#include "unicode/uobject.h"
|
2017-09-27 05:31:57 +00:00
|
|
|
|
|
2017-09-27 09:03:19 +00:00
|
|
|
|
#ifndef U_HIDE_DRAFT_API
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* \file
|
|
|
|
|
* \brief C++ API: Library for localized number formatting introduced in ICU 60.
|
|
|
|
|
*
|
|
|
|
|
* This library was introduced in ICU 60 to simplify the process of formatting localized number strings.
|
|
|
|
|
* Basic usage examples:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* // Most basic usage:
|
|
|
|
|
* NumberFormatter::withLocale(...).format(123).toString(); // 1,234 in en-US
|
|
|
|
|
*
|
|
|
|
|
* // Custom notation, unit, and rounding strategy:
|
|
|
|
|
* NumberFormatter::with()
|
|
|
|
|
* .notation(Notation::compactShort())
|
|
|
|
|
* .unit(CurrencyUnit("EUR", status))
|
|
|
|
|
* .rounding(Rounder::maxDigits(2))
|
|
|
|
|
* .locale(...)
|
|
|
|
|
* .format(1234)
|
|
|
|
|
* .toString(); // €1.2K in en-US
|
|
|
|
|
*
|
|
|
|
|
* // Create a formatter in a singleton for use later:
|
|
|
|
|
* static const LocalizedNumberFormatter formatter = NumberFormatter::withLocale(...)
|
|
|
|
|
* .unit(NoUnit::percent())
|
|
|
|
|
* .rounding(Rounder::fixedFraction(3));
|
|
|
|
|
* formatter.format(5.9831).toString(); // 5.983% in en-US
|
|
|
|
|
*
|
|
|
|
|
* // Create a "template" in a singleton but without setting a locale until the call site:
|
|
|
|
|
* static const UnlocalizedNumberFormatter template = NumberFormatter::with()
|
|
|
|
|
* .sign(UNumberSignDisplay::UNUM_SIGN_ALWAYS)
|
|
|
|
|
* .adoptUnit(MeasureUnit::createMeter(status))
|
|
|
|
|
* .unitWidth(UNumberUnitWidth::UNUM_UNIT_WIDTH_FULL_NAME);
|
|
|
|
|
* template.locale(...).format(1234).toString(); // +1,234 meters in en-US
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* This API offers more features than DecimalFormat and is geared toward new users of ICU.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* NumberFormatter instances are immutable and thread safe. This means that invoking a configuration method has no
|
|
|
|
|
* effect on the receiving instance; you must store and use the new number formatter instance it returns instead.
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* UnlocalizedNumberFormatter formatter = UnlocalizedNumberFormatter::with().notation(Notation::scientific());
|
|
|
|
|
* formatter.rounding(Rounder.maxFraction(2)); // does nothing!
|
|
|
|
|
* formatter.locale(Locale.getEnglish()).format(9.8765).toString(); // prints "9.8765E0", not "9.88E0"
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* This API is based on the <em>fluent</em> design pattern popularized by libraries such as Google's Guava. For
|
|
|
|
|
* extensive details on the design of this API, read <a href="https://goo.gl/szi5VB">the design doc</a>.
|
|
|
|
|
*
|
|
|
|
|
* @author Shane Carr
|
|
|
|
|
*/
|
|
|
|
|
|
2018-02-08 08:49:50 +00:00
|
|
|
|
U_NAMESPACE_BEGIN
|
|
|
|
|
|
2018-03-03 08:26:58 +00:00
|
|
|
|
// Forward declarations:
|
|
|
|
|
class IFixedDecimal;
|
|
|
|
|
|
2018-02-08 08:49:50 +00:00
|
|
|
|
namespace numparse {
|
|
|
|
|
namespace impl {
|
|
|
|
|
|
|
|
|
|
// Forward declarations:
|
|
|
|
|
class NumberParserImpl;
|
2018-04-03 04:38:16 +00:00
|
|
|
|
class MultiplierParseHandler;
|
2018-02-08 08:49:50 +00:00
|
|
|
|
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
namespace number { // icu::number
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
// Forward declarations:
|
|
|
|
|
class UnlocalizedNumberFormatter;
|
|
|
|
|
class LocalizedNumberFormatter;
|
|
|
|
|
class FormattedNumber;
|
|
|
|
|
class Notation;
|
|
|
|
|
class ScientificNotation;
|
|
|
|
|
class Rounder;
|
|
|
|
|
class FractionRounder;
|
|
|
|
|
class CurrencyRounder;
|
|
|
|
|
class IncrementRounder;
|
|
|
|
|
class IntegerWidth;
|
|
|
|
|
|
|
|
|
|
namespace impl {
|
|
|
|
|
|
2018-03-06 23:41:20 +00:00
|
|
|
|
#ifndef U_HIDE_INTERNAL_API
|
2018-02-08 06:06:08 +00:00
|
|
|
|
/**
|
|
|
|
|
* Datatype for minimum/maximum fraction digits. Must be able to hold kMaxIntFracSig.
|
|
|
|
|
*
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
|
|
|
|
typedef int16_t digits_t;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Use a default threshold of 3. This means that the third time .format() is called, the data structures get built
|
|
|
|
|
* using the "safe" code path. The first two calls to .format() will trigger the unsafe code path.
|
|
|
|
|
*
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
|
|
|
|
static constexpr int32_t DEFAULT_THRESHOLD = 3;
|
2018-03-06 23:41:20 +00:00
|
|
|
|
#endif // U_HIDE_INTERNAL_API
|
2018-02-08 06:06:08 +00:00
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
// Forward declarations:
|
|
|
|
|
class Padder;
|
|
|
|
|
struct MacroProps;
|
|
|
|
|
struct MicroProps;
|
|
|
|
|
class DecimalQuantity;
|
2018-03-27 01:58:26 +00:00
|
|
|
|
struct UFormattedNumberData;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
class NumberFormatterImpl;
|
|
|
|
|
struct ParsedPatternInfo;
|
|
|
|
|
class ScientificModifier;
|
|
|
|
|
class MultiplierProducer;
|
|
|
|
|
class MutablePatternModifier;
|
|
|
|
|
class LongNameHandler;
|
|
|
|
|
class ScientificHandler;
|
|
|
|
|
class CompactHandler;
|
|
|
|
|
class Modifier;
|
|
|
|
|
class NumberStringBuilder;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
class AffixPatternProvider;
|
|
|
|
|
class NumberPropertyMapper;
|
|
|
|
|
struct DecimalFormatProperties;
|
2018-04-03 04:38:16 +00:00
|
|
|
|
class MultiplierFormatHandler;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
class CurrencySymbols;
|
2018-03-23 07:51:52 +00:00
|
|
|
|
class GeneratorHelpers;
|
2018-04-05 21:54:04 +00:00
|
|
|
|
class DecNum;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
} // namespace impl
|
|
|
|
|
|
|
|
|
|
// Reserve extra names in case they are added as classes in the future:
|
|
|
|
|
typedef Notation CompactNotation;
|
|
|
|
|
typedef Notation SimpleNotation;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* A class that defines the notation style to be used when formatting numbers in NumberFormatter.
|
|
|
|
|
*
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API Notation : public UMemory {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
public:
|
|
|
|
|
/**
|
|
|
|
|
* Print the number using scientific notation (also known as scientific form, standard index form, or standard form
|
|
|
|
|
* in the UK). The format for scientific notation varies by locale; for example, many Western locales display the
|
|
|
|
|
* number in the form "#E0", where the number is displayed with one digit before the decimal separator, zero or more
|
|
|
|
|
* digits after the decimal separator, and the corresponding power of 10 displayed after the "E".
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Example outputs in <em>en-US</em> when printing 8.765E4 through 8.765E-3:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* 8.765E4
|
|
|
|
|
* 8.765E3
|
|
|
|
|
* 8.765E2
|
|
|
|
|
* 8.765E1
|
|
|
|
|
* 8.765E0
|
|
|
|
|
* 8.765E-1
|
|
|
|
|
* 8.765E-2
|
|
|
|
|
* 8.765E-3
|
|
|
|
|
* 0E0
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* @return A ScientificNotation for chaining or passing to the NumberFormatter notation() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static ScientificNotation scientific();
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Print the number using engineering notation, a variant of scientific notation in which the exponent must be
|
|
|
|
|
* divisible by 3.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Example outputs in <em>en-US</em> when printing 8.765E4 through 8.765E-3:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* 87.65E3
|
|
|
|
|
* 8.765E3
|
|
|
|
|
* 876.5E0
|
|
|
|
|
* 87.65E0
|
|
|
|
|
* 8.765E0
|
|
|
|
|
* 876.5E-3
|
|
|
|
|
* 87.65E-3
|
|
|
|
|
* 8.765E-3
|
|
|
|
|
* 0E0
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* @return A ScientificNotation for chaining or passing to the NumberFormatter notation() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static ScientificNotation engineering();
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Print the number using short-form compact notation.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* <em>Compact notation</em>, defined in Unicode Technical Standard #35 Part 3 Section 2.4.1, prints numbers with
|
|
|
|
|
* localized prefixes or suffixes corresponding to different powers of ten. Compact notation is similar to
|
|
|
|
|
* engineering notation in how it scales numbers.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Compact notation is ideal for displaying large numbers (over ~1000) to humans while at the same time minimizing
|
|
|
|
|
* screen real estate.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* In short form, the powers of ten are abbreviated. In <em>en-US</em>, the abbreviations are "K" for thousands, "M"
|
|
|
|
|
* for millions, "B" for billions, and "T" for trillions. Example outputs in <em>en-US</em> when printing 8.765E7
|
|
|
|
|
* through 8.765E0:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* 88M
|
|
|
|
|
* 8.8M
|
|
|
|
|
* 876K
|
|
|
|
|
* 88K
|
|
|
|
|
* 8.8K
|
|
|
|
|
* 876
|
|
|
|
|
* 88
|
|
|
|
|
* 8.8
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* When compact notation is specified without an explicit rounding strategy, numbers are rounded off to the closest
|
|
|
|
|
* integer after scaling the number by the corresponding power of 10, but with a digit shown after the decimal
|
|
|
|
|
* separator if there is only one digit before the decimal separator. The default compact notation rounding strategy
|
|
|
|
|
* is equivalent to:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* Rounder.integer().withMinDigits(2)
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* @return A CompactNotation for passing to the NumberFormatter notation() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static CompactNotation compactShort();
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Print the number using long-form compact notation. For more information on compact notation, see
|
|
|
|
|
* {@link #compactShort}.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* In long form, the powers of ten are spelled out fully. Example outputs in <em>en-US</em> when printing 8.765E7
|
|
|
|
|
* through 8.765E0:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* 88 million
|
|
|
|
|
* 8.8 million
|
|
|
|
|
* 876 thousand
|
|
|
|
|
* 88 thousand
|
|
|
|
|
* 8.8 thousand
|
|
|
|
|
* 876
|
|
|
|
|
* 88
|
|
|
|
|
* 8.8
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* @return A CompactNotation for passing to the NumberFormatter notation() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static CompactNotation compactLong();
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Print the number using simple notation without any scaling by powers of ten. This is the default behavior.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Since this is the default behavior, this method needs to be called only when it is necessary to override a
|
|
|
|
|
* previous setting.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Example outputs in <em>en-US</em> when printing 8.765E7 through 8.765E0:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* 87,650,000
|
|
|
|
|
* 8,765,000
|
|
|
|
|
* 876,500
|
|
|
|
|
* 87,650
|
|
|
|
|
* 8,765
|
|
|
|
|
* 876.5
|
|
|
|
|
* 87.65
|
|
|
|
|
* 8.765
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* @return A SimpleNotation for passing to the NumberFormatter notation() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static SimpleNotation simple();
|
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
enum NotationType {
|
|
|
|
|
NTN_SCIENTIFIC, NTN_COMPACT, NTN_SIMPLE, NTN_ERROR
|
|
|
|
|
} fType;
|
|
|
|
|
|
|
|
|
|
union NotationUnion {
|
|
|
|
|
// For NTN_SCIENTIFIC
|
|
|
|
|
struct ScientificSettings {
|
|
|
|
|
int8_t fEngineeringInterval;
|
|
|
|
|
bool fRequireMinInt;
|
2018-02-08 06:06:08 +00:00
|
|
|
|
impl::digits_t fMinExponentDigits;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
UNumberSignDisplay fExponentSignDisplay;
|
|
|
|
|
} scientific;
|
|
|
|
|
|
|
|
|
|
// For NTN_COMPACT
|
|
|
|
|
UNumberCompactStyle compactStyle;
|
|
|
|
|
|
|
|
|
|
// For NTN_ERROR
|
|
|
|
|
UErrorCode errorCode;
|
|
|
|
|
} fUnion;
|
|
|
|
|
|
|
|
|
|
typedef NotationUnion::ScientificSettings ScientificSettings;
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Notation(const NotationType &type, const NotationUnion &union_) : fType(type), fUnion(union_) {}
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Notation(UErrorCode errorCode) : fType(NTN_ERROR) {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
fUnion.errorCode = errorCode;
|
|
|
|
|
}
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Notation() : fType(NTN_SIMPLE), fUnion() {}
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
UBool copyErrorTo(UErrorCode &status) const {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
if (fType == NTN_ERROR) {
|
|
|
|
|
status = fUnion.errorCode;
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// To allow MacroProps to initialize empty instances:
|
|
|
|
|
friend struct impl::MacroProps;
|
|
|
|
|
friend class ScientificNotation;
|
|
|
|
|
|
|
|
|
|
// To allow implementation to access internal types:
|
|
|
|
|
friend class impl::NumberFormatterImpl;
|
|
|
|
|
friend class impl::ScientificModifier;
|
|
|
|
|
friend class impl::ScientificHandler;
|
2018-03-23 07:51:52 +00:00
|
|
|
|
|
|
|
|
|
// To allow access to the skeleton generation code:
|
|
|
|
|
friend class impl::GeneratorHelpers;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* A class that defines the scientific notation style to be used when formatting numbers in NumberFormatter.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* To create a ScientificNotation, use one of the factory methods in {@link Notation}.
|
|
|
|
|
*
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API ScientificNotation : public Notation {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
public:
|
|
|
|
|
/**
|
|
|
|
|
* Sets the minimum number of digits to show in the exponent of scientific notation, padding with zeros if
|
|
|
|
|
* necessary. Useful for fixed-width display.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* For example, with minExponentDigits=2, the number 123 will be printed as "1.23E02" in <em>en-US</em> instead of
|
|
|
|
|
* the default "1.23E2".
|
|
|
|
|
*
|
|
|
|
|
* @param minExponentDigits
|
|
|
|
|
* The minimum number of digits to show in the exponent.
|
|
|
|
|
* @return A ScientificNotation, for chaining.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
ScientificNotation withMinExponentDigits(int32_t minExponentDigits) const;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Sets whether to show the sign on positive and negative exponents in scientific notation. The default is AUTO,
|
|
|
|
|
* showing the minus sign but not the plus sign.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* For example, with exponentSignDisplay=ALWAYS, the number 123 will be printed as "1.23E+2" in <em>en-US</em>
|
|
|
|
|
* instead of the default "1.23E2".
|
|
|
|
|
*
|
|
|
|
|
* @param exponentSignDisplay
|
|
|
|
|
* The strategy for displaying the sign in the exponent.
|
|
|
|
|
* @return A ScientificNotation, for chaining.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
ScientificNotation withExponentSignDisplay(UNumberSignDisplay exponentSignDisplay) const;
|
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
// Inherit constructor
|
|
|
|
|
using Notation::Notation;
|
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
// Raw constructor for NumberPropertyMapper
|
|
|
|
|
ScientificNotation(int8_t fEngineeringInterval, bool fRequireMinInt, impl::digits_t fMinExponentDigits,
|
|
|
|
|
UNumberSignDisplay fExponentSignDisplay);
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
friend class Notation;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
// So that NumberPropertyMapper can create instances
|
|
|
|
|
friend class impl::NumberPropertyMapper;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
// Reserve extra names in case they are added as classes in the future:
|
|
|
|
|
typedef Rounder DigitRounder;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* A class that defines the rounding strategy to be used when formatting numbers in NumberFormatter.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* To create a Rounder, use one of the factory methods.
|
|
|
|
|
*
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API Rounder : public UMemory {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
public:
|
|
|
|
|
/**
|
|
|
|
|
* Show all available digits to full precision.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* <strong>NOTE:</strong> When formatting a <em>double</em>, this method, along with {@link #minFraction} and
|
|
|
|
|
* {@link #minDigits}, will trigger complex algorithm similar to <em>Dragon4</em> to determine the low-order digits
|
|
|
|
|
* and the number of digits to display based on the value of the double. If the number of fraction places or
|
|
|
|
|
* significant digits can be bounded, consider using {@link #maxFraction} or {@link #maxDigits} instead to maximize
|
|
|
|
|
* performance. For more information, read the following blog post.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* http://www.serpentine.com/blog/2011/06/29/here-be-dragons-advances-in-problems-you-didnt-even-know-you-had/
|
|
|
|
|
*
|
|
|
|
|
* @return A Rounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static Rounder unlimited();
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Show numbers rounded if necessary to the nearest integer.
|
|
|
|
|
*
|
|
|
|
|
* @return A FractionRounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static FractionRounder integer();
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Show numbers rounded if necessary to a certain number of fraction places (numerals after the decimal separator).
|
|
|
|
|
* Additionally, pad with zeros to ensure that this number of places are always shown.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Example output with minMaxFractionPlaces = 3:
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* 87,650.000<br>
|
|
|
|
|
* 8,765.000<br>
|
|
|
|
|
* 876.500<br>
|
|
|
|
|
* 87.650<br>
|
|
|
|
|
* 8.765<br>
|
|
|
|
|
* 0.876<br>
|
|
|
|
|
* 0.088<br>
|
|
|
|
|
* 0.009<br>
|
|
|
|
|
* 0.000 (zero)
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* This method is equivalent to {@link #minMaxFraction} with both arguments equal.
|
|
|
|
|
*
|
|
|
|
|
* @param minMaxFractionPlaces
|
|
|
|
|
* The minimum and maximum number of numerals to display after the decimal separator (rounding if too
|
|
|
|
|
* long or padding with zeros if too short).
|
|
|
|
|
* @return A FractionRounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static FractionRounder fixedFraction(int32_t minMaxFractionPlaces);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Always show at least a certain number of fraction places after the decimal separator, padding with zeros if
|
|
|
|
|
* necessary. Do not perform rounding (display numbers to their full precision).
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* <strong>NOTE:</strong> If you are formatting <em>doubles</em>, see the performance note in {@link #unlimited}.
|
|
|
|
|
*
|
|
|
|
|
* @param minFractionPlaces
|
|
|
|
|
* The minimum number of numerals to display after the decimal separator (padding with zeros if
|
|
|
|
|
* necessary).
|
|
|
|
|
* @return A FractionRounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static FractionRounder minFraction(int32_t minFractionPlaces);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Show numbers rounded if necessary to a certain number of fraction places (numerals after the decimal separator).
|
|
|
|
|
* Unlike the other fraction rounding strategies, this strategy does <em>not</em> pad zeros to the end of the
|
|
|
|
|
* number.
|
|
|
|
|
*
|
|
|
|
|
* @param maxFractionPlaces
|
|
|
|
|
* The maximum number of numerals to display after the decimal mark (rounding if necessary).
|
|
|
|
|
* @return A FractionRounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static FractionRounder maxFraction(int32_t maxFractionPlaces);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Show numbers rounded if necessary to a certain number of fraction places (numerals after the decimal separator);
|
|
|
|
|
* in addition, always show at least a certain number of places after the decimal separator, padding with zeros if
|
|
|
|
|
* necessary.
|
|
|
|
|
*
|
|
|
|
|
* @param minFractionPlaces
|
|
|
|
|
* The minimum number of numerals to display after the decimal separator (padding with zeros if
|
|
|
|
|
* necessary).
|
|
|
|
|
* @param maxFractionPlaces
|
|
|
|
|
* The maximum number of numerals to display after the decimal separator (rounding if necessary).
|
|
|
|
|
* @return A FractionRounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static FractionRounder minMaxFraction(int32_t minFractionPlaces, int32_t maxFractionPlaces);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Show numbers rounded if necessary to a certain number of significant digits or significant figures. Additionally,
|
|
|
|
|
* pad with zeros to ensure that this number of significant digits/figures are always shown.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* This method is equivalent to {@link #minMaxDigits} with both arguments equal.
|
|
|
|
|
*
|
|
|
|
|
* @param minMaxSignificantDigits
|
|
|
|
|
* The minimum and maximum number of significant digits to display (rounding if too long or padding with
|
|
|
|
|
* zeros if too short).
|
|
|
|
|
* @return A Rounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static DigitRounder fixedDigits(int32_t minMaxSignificantDigits);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Always show at least a certain number of significant digits/figures, padding with zeros if necessary. Do not
|
|
|
|
|
* perform rounding (display numbers to their full precision).
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* <strong>NOTE:</strong> If you are formatting <em>doubles</em>, see the performance note in {@link #unlimited}.
|
|
|
|
|
*
|
|
|
|
|
* @param minSignificantDigits
|
|
|
|
|
* The minimum number of significant digits to display (padding with zeros if too short).
|
|
|
|
|
* @return A Rounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static DigitRounder minDigits(int32_t minSignificantDigits);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Show numbers rounded if necessary to a certain number of significant digits/figures.
|
|
|
|
|
*
|
|
|
|
|
* @param maxSignificantDigits
|
|
|
|
|
* The maximum number of significant digits to display (rounding if too long).
|
|
|
|
|
* @return A Rounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static DigitRounder maxDigits(int32_t maxSignificantDigits);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Show numbers rounded if necessary to a certain number of significant digits/figures; in addition, always show at
|
|
|
|
|
* least a certain number of significant digits, padding with zeros if necessary.
|
|
|
|
|
*
|
|
|
|
|
* @param minSignificantDigits
|
|
|
|
|
* The minimum number of significant digits to display (padding with zeros if necessary).
|
|
|
|
|
* @param maxSignificantDigits
|
|
|
|
|
* The maximum number of significant digits to display (rounding if necessary).
|
|
|
|
|
* @return A Rounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static DigitRounder minMaxDigits(int32_t minSignificantDigits, int32_t maxSignificantDigits);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Show numbers rounded if necessary to the closest multiple of a certain rounding increment. For example, if the
|
|
|
|
|
* rounding increment is 0.5, then round 1.2 to 1 and round 1.3 to 1.5.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* In order to ensure that numbers are padded to the appropriate number of fraction places, call
|
|
|
|
|
* withMinFraction() on the return value of this method.
|
|
|
|
|
* For example, to round to the nearest 0.5 and always display 2 numerals after the
|
|
|
|
|
* decimal separator (to display 1.2 as "1.00" and 1.3 as "1.50"), you can run:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* Rounder::increment(0.5).withMinFraction(2)
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* @param roundingIncrement
|
|
|
|
|
* The increment to which to round numbers.
|
|
|
|
|
* @return A Rounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static IncrementRounder increment(double roundingIncrement);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Show numbers rounded and padded according to the rules for the currency unit. The most common rounding settings
|
|
|
|
|
* for currencies include <code>Rounder.fixedFraction(2)</code>, <code>Rounder.integer()</code>, and
|
|
|
|
|
* <code>Rounder.increment(0.05)</code> for cash transactions ("nickel rounding").
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* The exact rounding details will be resolved at runtime based on the currency unit specified in the
|
|
|
|
|
* NumberFormatter chain. To round according to the rules for one currency while displaying the symbol for another
|
|
|
|
|
* currency, the withCurrency() method can be called on the return value of this method.
|
|
|
|
|
*
|
|
|
|
|
* @param currencyUsage
|
|
|
|
|
* Either STANDARD (for digital transactions) or CASH (for transactions where the rounding increment may
|
|
|
|
|
* be limited by the available denominations of cash or coins).
|
|
|
|
|
* @return A CurrencyRounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static CurrencyRounder currency(UCurrencyUsage currencyUsage);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Sets the rounding mode to use when picking the direction to round (up or down). Common values
|
|
|
|
|
* include HALF_EVEN, HALF_UP, and FLOOR. The default is HALF_EVEN.
|
|
|
|
|
*
|
|
|
|
|
* @param roundingMode
|
|
|
|
|
* The RoundingMode to use.
|
|
|
|
|
* @return A Rounder for passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
Rounder withMode(UNumberFormatRoundingMode roundingMode) const;
|
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
enum RounderType {
|
|
|
|
|
RND_BOGUS,
|
|
|
|
|
RND_NONE,
|
|
|
|
|
RND_FRACTION,
|
|
|
|
|
RND_SIGNIFICANT,
|
|
|
|
|
RND_FRACTION_SIGNIFICANT,
|
|
|
|
|
RND_INCREMENT,
|
|
|
|
|
RND_CURRENCY,
|
|
|
|
|
RND_PASS_THROUGH,
|
|
|
|
|
RND_ERROR
|
|
|
|
|
} fType;
|
|
|
|
|
|
|
|
|
|
union RounderUnion {
|
|
|
|
|
struct FractionSignificantSettings {
|
|
|
|
|
// For RND_FRACTION, RND_SIGNIFICANT, and RND_FRACTION_SIGNIFICANT
|
2018-02-08 06:06:08 +00:00
|
|
|
|
impl::digits_t fMinFrac;
|
|
|
|
|
impl::digits_t fMaxFrac;
|
|
|
|
|
impl::digits_t fMinSig;
|
|
|
|
|
impl::digits_t fMaxSig;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
} fracSig;
|
|
|
|
|
struct IncrementSettings {
|
|
|
|
|
double fIncrement;
|
2018-02-08 06:06:08 +00:00
|
|
|
|
impl::digits_t fMinFrac;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
impl::digits_t fMaxFrac;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
} increment; // For RND_INCREMENT
|
|
|
|
|
UCurrencyUsage currencyUsage; // For RND_CURRENCY
|
|
|
|
|
UErrorCode errorCode; // For RND_ERROR
|
|
|
|
|
} fUnion;
|
|
|
|
|
|
|
|
|
|
typedef RounderUnion::FractionSignificantSettings FractionSignificantSettings;
|
|
|
|
|
typedef RounderUnion::IncrementSettings IncrementSettings;
|
|
|
|
|
|
|
|
|
|
UNumberFormatRoundingMode fRoundingMode;
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Rounder(const RounderType &type, const RounderUnion &union_, UNumberFormatRoundingMode roundingMode)
|
2017-09-27 00:25:20 +00:00
|
|
|
|
: fType(type), fUnion(union_), fRoundingMode(roundingMode) {}
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Rounder(UErrorCode errorCode) : fType(RND_ERROR) {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
fUnion.errorCode = errorCode;
|
|
|
|
|
}
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Rounder() : fType(RND_BOGUS) {}
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
bool isBogus() const {
|
|
|
|
|
return fType == RND_BOGUS;
|
|
|
|
|
}
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
UBool copyErrorTo(UErrorCode &status) const {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
if (fType == RND_ERROR) {
|
|
|
|
|
status = fUnion.errorCode;
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// On the parent type so that this method can be called internally on Rounder instances.
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Rounder withCurrency(const CurrencyUnit ¤cy, UErrorCode &status) const;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/** NON-CONST: mutates the current instance. */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
void setLocaleData(const CurrencyUnit ¤cy, UErrorCode &status);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
void apply(impl::DecimalQuantity &value, UErrorCode &status) const;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/** Version of {@link #apply} that obeys minInt constraints. Used for scientific notation compatibility mode. */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
void apply(impl::DecimalQuantity &value, int32_t minInt, UErrorCode status);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2017-12-14 00:47:43 +00:00
|
|
|
|
/**
|
|
|
|
|
* Rounding endpoint used by Engineering and Compact notation. Chooses the most appropriate multiplier (magnitude
|
|
|
|
|
* adjustment), applies the adjustment, rounds, and returns the chosen multiplier.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* In most cases, this is simple. However, when rounding the number causes it to cross a multiplier boundary, we
|
|
|
|
|
* need to re-do the rounding. For example, to display 999,999 in Engineering notation with 2 sigfigs, first you
|
|
|
|
|
* guess the multiplier to be -3. However, then you end up getting 1000E3, which is not the correct output. You then
|
|
|
|
|
* change your multiplier to be -6, and you get 1.0E6, which is correct.
|
|
|
|
|
*
|
|
|
|
|
* @param input The quantity to process.
|
|
|
|
|
* @param producer Function to call to return a multiplier based on a magnitude.
|
|
|
|
|
* @return The number of orders of magnitude the input was adjusted by this method.
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
int32_t
|
|
|
|
|
chooseMultiplierAndApply(impl::DecimalQuantity &input, const impl::MultiplierProducer &producer,
|
|
|
|
|
UErrorCode &status);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
static FractionRounder constructFraction(int32_t minFrac, int32_t maxFrac);
|
|
|
|
|
|
|
|
|
|
static Rounder constructSignificant(int32_t minSig, int32_t maxSig);
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
static Rounder
|
|
|
|
|
constructFractionSignificant(const FractionRounder &base, int32_t minSig, int32_t maxSig);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
static IncrementRounder constructIncrement(double increment, int32_t minFrac);
|
|
|
|
|
|
|
|
|
|
static CurrencyRounder constructCurrency(UCurrencyUsage usage);
|
|
|
|
|
|
|
|
|
|
static Rounder constructPassThrough();
|
|
|
|
|
|
|
|
|
|
// To allow MacroProps/MicroProps to initialize bogus instances:
|
|
|
|
|
friend struct impl::MacroProps;
|
|
|
|
|
friend struct impl::MicroProps;
|
|
|
|
|
|
|
|
|
|
// To allow NumberFormatterImpl to access isBogus() and other internal methods:
|
|
|
|
|
friend class impl::NumberFormatterImpl;
|
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
// To allow NumberPropertyMapper to create instances from DecimalFormatProperties:
|
|
|
|
|
friend class impl::NumberPropertyMapper;
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
// To give access to apply() and chooseMultiplierAndApply():
|
|
|
|
|
friend class impl::MutablePatternModifier;
|
|
|
|
|
friend class impl::LongNameHandler;
|
|
|
|
|
friend class impl::ScientificHandler;
|
|
|
|
|
friend class impl::CompactHandler;
|
|
|
|
|
|
|
|
|
|
// To allow child classes to call private methods:
|
|
|
|
|
friend class FractionRounder;
|
|
|
|
|
friend class CurrencyRounder;
|
|
|
|
|
friend class IncrementRounder;
|
2018-03-23 07:51:52 +00:00
|
|
|
|
|
|
|
|
|
// To allow access to the skeleton generation code:
|
|
|
|
|
friend class impl::GeneratorHelpers;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* A class that defines a rounding strategy based on a number of fraction places and optionally significant digits to be
|
|
|
|
|
* used when formatting numbers in NumberFormatter.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* To create a FractionRounder, use one of the factory methods on Rounder.
|
|
|
|
|
*
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API FractionRounder : public Rounder {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
public:
|
|
|
|
|
/**
|
|
|
|
|
* Ensure that no less than this number of significant digits are retained when rounding according to fraction
|
|
|
|
|
* rules.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* For example, with integer rounding, the number 3.141 becomes "3". However, with minimum figures set to 2, 3.141
|
|
|
|
|
* becomes "3.1" instead.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* This setting does not affect the number of trailing zeros. For example, 3.01 would print as "3", not "3.0".
|
|
|
|
|
*
|
|
|
|
|
* @param minSignificantDigits
|
|
|
|
|
* The number of significant figures to guarantee.
|
|
|
|
|
* @return A Rounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
Rounder withMinDigits(int32_t minSignificantDigits) const;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Ensure that no more than this number of significant digits are retained when rounding according to fraction
|
|
|
|
|
* rules.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* For example, with integer rounding, the number 123.4 becomes "123". However, with maximum figures set to 2, 123.4
|
|
|
|
|
* becomes "120" instead.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* This setting does not affect the number of trailing zeros. For example, with fixed fraction of 2, 123.4 would
|
|
|
|
|
* become "120.00".
|
|
|
|
|
*
|
|
|
|
|
* @param maxSignificantDigits
|
|
|
|
|
* Round the number to no more than this number of significant figures.
|
|
|
|
|
* @return A Rounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
Rounder withMaxDigits(int32_t maxSignificantDigits) const;
|
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
// Inherit constructor
|
|
|
|
|
using Rounder::Rounder;
|
|
|
|
|
|
|
|
|
|
// To allow parent class to call this class's constructor:
|
|
|
|
|
friend class Rounder;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* A class that defines a rounding strategy parameterized by a currency to be used when formatting numbers in
|
|
|
|
|
* NumberFormatter.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* To create a CurrencyRounder, use one of the factory methods on Rounder.
|
|
|
|
|
*
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API CurrencyRounder : public Rounder {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
public:
|
|
|
|
|
/**
|
|
|
|
|
* Associates a currency with this rounding strategy.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* <strong>Calling this method is <em>not required</em></strong>, because the currency specified in unit()
|
|
|
|
|
* is automatically applied to currency rounding strategies. However,
|
|
|
|
|
* this method enables you to override that automatic association.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* This method also enables numbers to be formatted using currency rounding rules without explicitly using a
|
|
|
|
|
* currency format.
|
|
|
|
|
*
|
|
|
|
|
* @param currency
|
|
|
|
|
* The currency to associate with this rounding strategy.
|
|
|
|
|
* @return A Rounder for chaining or passing to the NumberFormatter rounding() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Rounder withCurrency(const CurrencyUnit ¤cy) const;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
// Inherit constructor
|
|
|
|
|
using Rounder::Rounder;
|
|
|
|
|
|
|
|
|
|
// To allow parent class to call this class's constructor:
|
|
|
|
|
friend class Rounder;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* A class that defines a rounding strategy parameterized by a rounding increment to be used when formatting numbers in
|
|
|
|
|
* NumberFormatter.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* To create an IncrementRounder, use one of the factory methods on Rounder.
|
|
|
|
|
*
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API IncrementRounder : public Rounder {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
public:
|
|
|
|
|
/**
|
|
|
|
|
* Specifies the minimum number of fraction digits to render after the decimal separator, padding with zeros if
|
|
|
|
|
* necessary. By default, no trailing zeros are added.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* For example, if the rounding increment is 0.5 and minFrac is 2, then the resulting strings include "0.00",
|
|
|
|
|
* "0.50", "1.00", and "1.50".
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Note: In ICU4J, this functionality is accomplished via the scale of the BigDecimal rounding increment.
|
|
|
|
|
*
|
|
|
|
|
* @param minFrac The minimum number of digits after the decimal separator.
|
|
|
|
|
* @return A Rounder for chaining or passing to the NumberFormatter rounding() setter.
|
2017-10-05 23:03:19 +00:00
|
|
|
|
* @draft ICU 60
|
2017-09-27 00:25:20 +00:00
|
|
|
|
*/
|
|
|
|
|
Rounder withMinFraction(int32_t minFrac) const;
|
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
// Inherit constructor
|
|
|
|
|
using Rounder::Rounder;
|
|
|
|
|
|
|
|
|
|
// To allow parent class to call this class's constructor:
|
|
|
|
|
friend class Rounder;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* A class that defines the strategy for padding and truncating integers before the decimal separator.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* To create an IntegerWidth, use one of the factory methods.
|
|
|
|
|
*
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
* @see NumberFormatter
|
|
|
|
|
*/
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API IntegerWidth : public UMemory {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
public:
|
|
|
|
|
/**
|
|
|
|
|
* Pad numbers at the beginning with zeros to guarantee a certain number of numerals before the decimal separator.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* For example, with minInt=3, the number 55 will get printed as "055".
|
|
|
|
|
*
|
|
|
|
|
* @param minInt
|
|
|
|
|
* The minimum number of places before the decimal separator.
|
|
|
|
|
* @return An IntegerWidth for chaining or passing to the NumberFormatter integerWidth() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static IntegerWidth zeroFillTo(int32_t minInt);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Truncate numbers exceeding a certain number of numerals before the decimal separator.
|
|
|
|
|
*
|
|
|
|
|
* For example, with maxInt=3, the number 1234 will get printed as "234".
|
|
|
|
|
*
|
|
|
|
|
* @param maxInt
|
2018-02-08 06:06:08 +00:00
|
|
|
|
* The maximum number of places before the decimal separator. maxInt == -1 means no
|
|
|
|
|
* truncation.
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* @return An IntegerWidth for passing to the NumberFormatter integerWidth() setter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
IntegerWidth truncateAt(int32_t maxInt);
|
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
union {
|
|
|
|
|
struct {
|
2018-02-08 06:06:08 +00:00
|
|
|
|
impl::digits_t fMinInt;
|
|
|
|
|
impl::digits_t fMaxInt;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
} minMaxInt;
|
|
|
|
|
UErrorCode errorCode;
|
|
|
|
|
} fUnion;
|
|
|
|
|
bool fHasError = false;
|
|
|
|
|
|
2018-02-08 06:06:08 +00:00
|
|
|
|
IntegerWidth(impl::digits_t minInt, impl::digits_t maxInt);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
IntegerWidth(UErrorCode errorCode) { // NOLINT
|
|
|
|
|
fUnion.errorCode = errorCode;
|
|
|
|
|
fHasError = true;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
IntegerWidth() { // NOLINT
|
|
|
|
|
fUnion.minMaxInt.fMinInt = -1;
|
|
|
|
|
}
|
|
|
|
|
|
2018-03-23 07:51:52 +00:00
|
|
|
|
/** Returns the default instance. */
|
|
|
|
|
static IntegerWidth standard() {
|
|
|
|
|
return IntegerWidth::zeroFillTo(1);
|
|
|
|
|
}
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
bool isBogus() const {
|
|
|
|
|
return !fHasError && fUnion.minMaxInt.fMinInt == -1;
|
|
|
|
|
}
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
UBool copyErrorTo(UErrorCode &status) const {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
if (fHasError) {
|
|
|
|
|
status = fUnion.errorCode;
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
void apply(impl::DecimalQuantity &quantity, UErrorCode &status) const;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-23 07:51:52 +00:00
|
|
|
|
bool operator==(const IntegerWidth& other) const;
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
// To allow MacroProps/MicroProps to initialize empty instances:
|
|
|
|
|
friend struct impl::MacroProps;
|
|
|
|
|
friend struct impl::MicroProps;
|
|
|
|
|
|
|
|
|
|
// To allow NumberFormatterImpl to access isBogus() and perform other operations:
|
|
|
|
|
friend class impl::NumberFormatterImpl;
|
2018-03-23 07:51:52 +00:00
|
|
|
|
|
|
|
|
|
// To allow access to the skeleton generation code:
|
|
|
|
|
friend class impl::GeneratorHelpers;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
};
|
|
|
|
|
|
2018-04-03 04:38:16 +00:00
|
|
|
|
/**
|
|
|
|
|
* A class that defines a quantity by which a number should be multiplied when formatting.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* To create a Multiplier, use one of the factory methods.
|
|
|
|
|
*
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
class U_I18N_API Multiplier : public UMemory {
|
|
|
|
|
public:
|
|
|
|
|
/**
|
|
|
|
|
* Do not change the value of numbers when formatting or parsing.
|
|
|
|
|
*
|
|
|
|
|
* @return A Multiplier to prevent any multiplication.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
static Multiplier none();
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Multiply numbers by 100 before formatting. Useful for combining with a percent unit:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with().unit(NoUnit::percent()).multiplier(Multiplier::powerOfTen(2))
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* @return A Multiplier for passing to the setter in NumberFormatter.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
static Multiplier powerOfTen(int32_t power);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Multiply numbers by an arbitrary value before formatting. Useful for unit conversions.
|
|
|
|
|
*
|
|
|
|
|
* This method takes a string in a decimal number format with syntax
|
|
|
|
|
* as defined in the Decimal Arithmetic Specification, available at
|
|
|
|
|
* http://speleotrove.com/decimal
|
|
|
|
|
*
|
|
|
|
|
* Also see the version of this method that takes a double.
|
|
|
|
|
*
|
|
|
|
|
* @return A Multiplier for passing to the setter in NumberFormatter.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
static Multiplier arbitraryDecimal(StringPiece multiplicand);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Multiply numbers by an arbitrary value before formatting. Useful for unit conversions.
|
|
|
|
|
*
|
|
|
|
|
* This method takes a double; also see the version of this method that takes an exact decimal.
|
|
|
|
|
*
|
|
|
|
|
* @return A Multiplier for passing to the setter in NumberFormatter.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
static Multiplier arbitraryDouble(double multiplicand);
|
|
|
|
|
|
2018-04-05 21:54:04 +00:00
|
|
|
|
// We need a custom destructor for the DecNum, which means we need to declare
|
|
|
|
|
// the copy/move constructor/assignment quartet.
|
|
|
|
|
|
|
|
|
|
/** @draft ICU 62 */
|
|
|
|
|
Multiplier(const Multiplier& other);
|
|
|
|
|
|
|
|
|
|
/** @draft ICU 62 */
|
|
|
|
|
Multiplier& operator=(const Multiplier& other);
|
|
|
|
|
|
|
|
|
|
/** @draft ICU 62 */
|
|
|
|
|
Multiplier(Multiplier&& src) U_NOEXCEPT;
|
|
|
|
|
|
|
|
|
|
/** @draft ICU 62 */
|
|
|
|
|
Multiplier& operator=(Multiplier&& src) U_NOEXCEPT;
|
|
|
|
|
|
|
|
|
|
/** @draft ICU 62 */
|
|
|
|
|
~Multiplier();
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
Multiplier(int32_t magnitude, impl::DecNum* arbitraryToAdopt);
|
|
|
|
|
|
2018-04-03 04:38:16 +00:00
|
|
|
|
private:
|
|
|
|
|
int32_t fMagnitude;
|
2018-04-05 21:54:04 +00:00
|
|
|
|
impl::DecNum* fArbitrary;
|
|
|
|
|
UErrorCode fError;
|
2018-04-03 04:38:16 +00:00
|
|
|
|
|
2018-04-05 21:54:04 +00:00
|
|
|
|
Multiplier(UErrorCode error) : fMagnitude(0), fArbitrary(nullptr), fError(error) {}
|
2018-04-03 04:38:16 +00:00
|
|
|
|
|
2018-04-05 21:54:04 +00:00
|
|
|
|
Multiplier() : fMagnitude(0), fArbitrary(nullptr), fError(U_ZERO_ERROR) {}
|
2018-04-03 04:38:16 +00:00
|
|
|
|
|
|
|
|
|
bool isValid() const {
|
2018-04-05 21:54:04 +00:00
|
|
|
|
return fMagnitude != 0 || fArbitrary != nullptr;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
UBool copyErrorTo(UErrorCode &status) const {
|
|
|
|
|
if (fError != U_ZERO_ERROR) {
|
|
|
|
|
status = fError;
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
return FALSE;
|
2018-04-03 04:38:16 +00:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void applyTo(impl::DecimalQuantity& quantity) const;
|
|
|
|
|
|
|
|
|
|
void applyReciprocalTo(impl::DecimalQuantity& quantity) const;
|
|
|
|
|
|
|
|
|
|
// To allow MacroProps/MicroProps to initialize empty instances:
|
|
|
|
|
friend struct impl::MacroProps;
|
|
|
|
|
friend struct impl::MicroProps;
|
|
|
|
|
|
|
|
|
|
// To allow NumberFormatterImpl to access isBogus() and perform other operations:
|
|
|
|
|
friend class impl::NumberFormatterImpl;
|
|
|
|
|
|
|
|
|
|
// To allow the helper class MultiplierFormatHandler access to private fields:
|
|
|
|
|
friend class impl::MultiplierFormatHandler;
|
|
|
|
|
|
|
|
|
|
// To allow access to the skeleton generation code:
|
|
|
|
|
friend class impl::GeneratorHelpers;
|
|
|
|
|
|
|
|
|
|
// To allow access to parsing code:
|
|
|
|
|
friend class ::icu::numparse::impl::NumberParserImpl;
|
|
|
|
|
friend class ::icu::numparse::impl::MultiplierParseHandler;
|
|
|
|
|
};
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
namespace impl {
|
|
|
|
|
|
2018-03-06 23:41:20 +00:00
|
|
|
|
// Do not enclose entire SymbolsWrapper with #ifndef U_HIDE_INTERNAL_API, needed for a protected field
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/** @internal */
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API SymbolsWrapper : public UMemory {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
public:
|
|
|
|
|
/** @internal */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
SymbolsWrapper() : fType(SYMPTR_NONE), fPtr{nullptr} {}
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/** @internal */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
SymbolsWrapper(const SymbolsWrapper &other);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
/** @internal */
|
2018-04-05 21:54:04 +00:00
|
|
|
|
SymbolsWrapper &operator=(const SymbolsWrapper &other);
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/** @internal */
|
2018-04-05 21:54:04 +00:00
|
|
|
|
SymbolsWrapper(SymbolsWrapper&& src) U_NOEXCEPT;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/** @internal */
|
2018-04-05 21:54:04 +00:00
|
|
|
|
SymbolsWrapper &operator=(SymbolsWrapper&& src) U_NOEXCEPT;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
/** @internal */
|
2018-04-05 21:54:04 +00:00
|
|
|
|
~SymbolsWrapper();
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
2018-03-06 23:41:20 +00:00
|
|
|
|
#ifndef U_HIDE_INTERNAL_API
|
2018-03-28 01:08:47 +00:00
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/**
|
|
|
|
|
* The provided object is copied, but we do not adopt it.
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
void setTo(const DecimalFormatSymbols &dfs);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Adopt the provided object.
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
void setTo(const NumberingSystem *ns);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Whether the object is currently holding a DecimalFormatSymbols.
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
|
|
|
|
bool isDecimalFormatSymbols() const;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Whether the object is currently holding a NumberingSystem.
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
|
|
|
|
bool isNumberingSystem() const;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Get the DecimalFormatSymbols pointer. No ownership change.
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
const DecimalFormatSymbols *getDecimalFormatSymbols() const;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Get the NumberingSystem pointer. No ownership change.
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
const NumberingSystem *getNumberingSystem() const;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/** @internal */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
UBool copyErrorTo(UErrorCode &status) const {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
if (fType == SYMPTR_DFS && fPtr.dfs == nullptr) {
|
|
|
|
|
status = U_MEMORY_ALLOCATION_ERROR;
|
|
|
|
|
return TRUE;
|
|
|
|
|
} else if (fType == SYMPTR_NS && fPtr.ns == nullptr) {
|
|
|
|
|
status = U_MEMORY_ALLOCATION_ERROR;
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
2018-03-06 23:41:20 +00:00
|
|
|
|
#endif // U_HIDE_INTERNAL_API
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
enum SymbolsPointerType {
|
|
|
|
|
SYMPTR_NONE, SYMPTR_DFS, SYMPTR_NS
|
|
|
|
|
} fType;
|
|
|
|
|
|
|
|
|
|
union {
|
2018-03-28 00:39:28 +00:00
|
|
|
|
const DecimalFormatSymbols *dfs;
|
|
|
|
|
const NumberingSystem *ns;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
} fPtr;
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
void doCopyFrom(const SymbolsWrapper &other);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
void doMoveFrom(SymbolsWrapper&& src);
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
void doCleanup();
|
|
|
|
|
};
|
|
|
|
|
|
2018-03-06 23:41:20 +00:00
|
|
|
|
// Do not enclose entire Grouper with #ifndef U_HIDE_INTERNAL_API, needed for a protected field
|
2018-02-06 03:08:17 +00:00
|
|
|
|
/** @internal */
|
|
|
|
|
class U_I18N_API Grouper : public UMemory {
|
|
|
|
|
public:
|
2018-03-06 23:41:20 +00:00
|
|
|
|
#ifndef U_HIDE_INTERNAL_API
|
2018-02-06 03:08:17 +00:00
|
|
|
|
/** @internal */
|
|
|
|
|
static Grouper forStrategy(UGroupingStrategy grouping);
|
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
/**
|
|
|
|
|
* Resolve the values in Properties to a Grouper object.
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
|
|
|
|
static Grouper forProperties(const DecimalFormatProperties& properties);
|
|
|
|
|
|
2018-02-06 03:08:17 +00:00
|
|
|
|
// Future: static Grouper forProperties(DecimalFormatProperties& properties);
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
2018-03-23 07:51:52 +00:00
|
|
|
|
Grouper(int16_t grouping1, int16_t grouping2, int16_t minGrouping, UGroupingStrategy strategy)
|
|
|
|
|
: fGrouping1(grouping1),
|
|
|
|
|
fGrouping2(grouping2),
|
|
|
|
|
fMinGrouping(minGrouping),
|
|
|
|
|
fStrategy(strategy) {}
|
2018-03-06 23:41:20 +00:00
|
|
|
|
#endif // U_HIDE_INTERNAL_API
|
2018-02-06 03:08:17 +00:00
|
|
|
|
|
2018-02-08 08:49:50 +00:00
|
|
|
|
/** @internal */
|
|
|
|
|
int16_t getPrimary() const;
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
int16_t getSecondary() const;
|
|
|
|
|
|
2018-02-06 03:08:17 +00:00
|
|
|
|
private:
|
|
|
|
|
/**
|
|
|
|
|
* The grouping sizes, with the following special values:
|
|
|
|
|
* <ul>
|
|
|
|
|
* <li>-1 = no grouping
|
|
|
|
|
* <li>-2 = needs locale data
|
|
|
|
|
* <li>-4 = fall back to Western grouping if not in locale
|
|
|
|
|
* </ul>
|
|
|
|
|
*/
|
|
|
|
|
int16_t fGrouping1;
|
|
|
|
|
int16_t fGrouping2;
|
|
|
|
|
|
|
|
|
|
/**
|
2018-03-23 07:51:52 +00:00
|
|
|
|
* The minimum grouping size, with the following special values:
|
2018-02-06 03:08:17 +00:00
|
|
|
|
* <ul>
|
|
|
|
|
* <li>-2 = needs locale data
|
|
|
|
|
* <li>-3 = no less than 2
|
|
|
|
|
* </ul>
|
|
|
|
|
*/
|
|
|
|
|
int16_t fMinGrouping;
|
|
|
|
|
|
2018-03-23 07:51:52 +00:00
|
|
|
|
/**
|
|
|
|
|
* The UGroupingStrategy that was used to create this Grouper, or UNUM_GROUPING_COUNT if this
|
|
|
|
|
* was not created from a UGroupingStrategy.
|
|
|
|
|
*/
|
|
|
|
|
UGroupingStrategy fStrategy;
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Grouper() : fGrouping1(-3) {};
|
2018-02-06 03:08:17 +00:00
|
|
|
|
|
|
|
|
|
bool isBogus() const {
|
|
|
|
|
return fGrouping1 == -3;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/** NON-CONST: mutates the current instance. */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
void setLocaleData(const impl::ParsedPatternInfo &patternInfo, const Locale& locale);
|
2018-02-06 03:08:17 +00:00
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
bool groupAtPosition(int32_t position, const impl::DecimalQuantity &value) const;
|
2018-02-06 03:08:17 +00:00
|
|
|
|
|
|
|
|
|
// To allow MacroProps/MicroProps to initialize empty instances:
|
|
|
|
|
friend struct MacroProps;
|
|
|
|
|
friend struct MicroProps;
|
|
|
|
|
|
|
|
|
|
// To allow NumberFormatterImpl to access isBogus() and perform other operations:
|
|
|
|
|
friend class NumberFormatterImpl;
|
2018-02-08 08:49:50 +00:00
|
|
|
|
|
|
|
|
|
// To allow NumberParserImpl to perform setLocaleData():
|
|
|
|
|
friend class ::icu::numparse::impl::NumberParserImpl;
|
2018-03-23 07:51:52 +00:00
|
|
|
|
|
|
|
|
|
// To allow access to the skeleton generation code:
|
|
|
|
|
friend class impl::GeneratorHelpers;
|
2018-02-06 03:08:17 +00:00
|
|
|
|
};
|
|
|
|
|
|
2018-03-06 23:41:20 +00:00
|
|
|
|
// Do not enclose entire Padder with #ifndef U_HIDE_INTERNAL_API, needed for a protected field
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/** @internal */
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API Padder : public UMemory {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
public:
|
2018-03-06 23:41:20 +00:00
|
|
|
|
#ifndef U_HIDE_INTERNAL_API
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/** @internal */
|
|
|
|
|
static Padder none();
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
static Padder codePoints(UChar32 cp, int32_t targetWidth, UNumberFormatPadPosition position);
|
2018-03-06 23:41:20 +00:00
|
|
|
|
#endif // U_HIDE_INTERNAL_API
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
/** @internal */
|
|
|
|
|
static Padder forProperties(const DecimalFormatProperties& properties);
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
private:
|
|
|
|
|
UChar32 fWidth; // -3 = error; -2 = bogus; -1 = no padding
|
|
|
|
|
union {
|
|
|
|
|
struct {
|
|
|
|
|
int32_t fCp;
|
|
|
|
|
UNumberFormatPadPosition fPosition;
|
|
|
|
|
} padding;
|
|
|
|
|
UErrorCode errorCode;
|
|
|
|
|
} fUnion;
|
|
|
|
|
|
|
|
|
|
Padder(UChar32 cp, int32_t width, UNumberFormatPadPosition position);
|
|
|
|
|
|
|
|
|
|
Padder(int32_t width);
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Padder(UErrorCode errorCode) : fWidth(-3) { // NOLINT
|
2017-09-27 00:25:20 +00:00
|
|
|
|
fUnion.errorCode = errorCode;
|
|
|
|
|
}
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Padder() : fWidth(-2) {} // NOLINT
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
bool isBogus() const {
|
|
|
|
|
return fWidth == -2;
|
|
|
|
|
}
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
UBool copyErrorTo(UErrorCode &status) const {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
if (fWidth == -3) {
|
|
|
|
|
status = fUnion.errorCode;
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
bool isValid() const {
|
|
|
|
|
return fWidth > 0;
|
|
|
|
|
}
|
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
int32_t padAndApply(const impl::Modifier &mod1, const impl::Modifier &mod2,
|
|
|
|
|
impl::NumberStringBuilder &string, int32_t leftIndex, int32_t rightIndex,
|
|
|
|
|
UErrorCode &status) const;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
// To allow MacroProps/MicroProps to initialize empty instances:
|
|
|
|
|
friend struct MacroProps;
|
|
|
|
|
friend struct MicroProps;
|
|
|
|
|
|
|
|
|
|
// To allow NumberFormatterImpl to access isBogus() and perform other operations:
|
|
|
|
|
friend class impl::NumberFormatterImpl;
|
2018-03-23 07:51:52 +00:00
|
|
|
|
|
|
|
|
|
// To allow access to the skeleton generation code:
|
|
|
|
|
friend class impl::GeneratorHelpers;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
};
|
|
|
|
|
|
2018-03-06 23:41:20 +00:00
|
|
|
|
// Do not enclose entire MacroProps with #ifndef U_HIDE_INTERNAL_API, needed for a protected field
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/** @internal */
|
2017-09-27 18:50:44 +00:00
|
|
|
|
struct U_I18N_API MacroProps : public UMemory {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/** @internal */
|
|
|
|
|
Notation notation;
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
MeasureUnit unit; // = NoUnit::base();
|
|
|
|
|
|
2017-12-22 00:02:01 +00:00
|
|
|
|
/** @internal */
|
|
|
|
|
MeasureUnit perUnit; // = NoUnit::base();
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/** @internal */
|
|
|
|
|
Rounder rounder; // = Rounder(); (bogus)
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
Grouper grouper; // = Grouper(); (bogus)
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
Padder padder; // = Padder(); (bogus)
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
IntegerWidth integerWidth; // = IntegerWidth(); (bogus)
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
SymbolsWrapper symbols;
|
|
|
|
|
|
|
|
|
|
// UNUM_XYZ_COUNT denotes null (bogus) values.
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
UNumberUnitWidth unitWidth = UNUM_UNIT_WIDTH_COUNT;
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
UNumberSignDisplay sign = UNUM_SIGN_COUNT;
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
UNumberDecimalSeparatorDisplay decimal = UNUM_DECIMAL_SEPARATOR_COUNT;
|
|
|
|
|
|
2018-04-03 04:38:16 +00:00
|
|
|
|
/** @internal */
|
|
|
|
|
Multiplier multiplier; // = Multiplier(); (benign value)
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
2018-04-03 04:38:16 +00:00
|
|
|
|
/** @internal */
|
2018-03-17 07:24:02 +00:00
|
|
|
|
AffixPatternProvider* affixProvider = nullptr; // no ownership
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/** @internal */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
PluralRules *rules = nullptr; // no ownership
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
/** @internal */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
CurrencySymbols *currencySymbols = nullptr; // no ownership
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/** @internal */
|
2017-10-04 01:15:54 +00:00
|
|
|
|
int32_t threshold = DEFAULT_THRESHOLD;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/** @internal */
|
2017-09-27 00:25:20 +00:00
|
|
|
|
Locale locale;
|
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
// NOTE: Uses default copy and move constructors.
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/**
|
|
|
|
|
* Check all members for errors.
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
bool copyErrorTo(UErrorCode &status) const {
|
|
|
|
|
return notation.copyErrorTo(status) || rounder.copyErrorTo(status) ||
|
|
|
|
|
padder.copyErrorTo(status) || integerWidth.copyErrorTo(status) ||
|
2018-04-05 21:54:04 +00:00
|
|
|
|
symbols.copyErrorTo(status) || multiplier.copyErrorTo(status);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
}
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
} // namespace impl
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* An abstract base class for specifying settings related to number formatting. This class is implemented by
|
|
|
|
|
* {@link UnlocalizedNumberFormatter} and {@link LocalizedNumberFormatter}.
|
|
|
|
|
*/
|
|
|
|
|
template<typename Derived>
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API NumberFormatterSettings {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
public:
|
|
|
|
|
/**
|
|
|
|
|
* Specifies the notation style (simple, scientific, or compact) for rendering numbers.
|
|
|
|
|
*
|
|
|
|
|
* <ul>
|
|
|
|
|
* <li>Simple notation: "12,300"
|
|
|
|
|
* <li>Scientific notation: "1.23E4"
|
|
|
|
|
* <li>Compact notation: "12K"
|
|
|
|
|
* </ul>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* All notation styles will be properly localized with locale data, and all notation styles are compatible with
|
|
|
|
|
* units, rounding strategies, and other number formatter settings.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Pass this method the return value of a {@link Notation} factory method. For example:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with().notation(Notation::compactShort())
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* The default is to use simple notation.
|
|
|
|
|
*
|
|
|
|
|
* @param notation
|
|
|
|
|
* The notation strategy to use.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see Notation
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived notation(const Notation ¬ation) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of notation() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param notation
|
|
|
|
|
* The notation strategy to use.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #notation
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived notation(const Notation ¬ation) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Specifies the unit (unit of measure, currency, or percent) to associate with rendered numbers.
|
|
|
|
|
*
|
|
|
|
|
* <ul>
|
|
|
|
|
* <li>Unit of measure: "12.3 meters"
|
|
|
|
|
* <li>Currency: "$12.30"
|
|
|
|
|
* <li>Percent: "12.3%"
|
|
|
|
|
* </ul>
|
|
|
|
|
*
|
|
|
|
|
* All units will be properly localized with locale data, and all units are compatible with notation styles,
|
|
|
|
|
* rounding strategies, and other number formatter settings.
|
|
|
|
|
*
|
2018-02-09 22:52:42 +00:00
|
|
|
|
* Pass this method any instance of {@link MeasureUnit}. For units of measure (which often involve the
|
|
|
|
|
* factory methods that return a pointer):
|
2017-09-27 00:25:20 +00:00
|
|
|
|
*
|
|
|
|
|
* <pre>
|
2017-12-22 00:02:01 +00:00
|
|
|
|
* NumberFormatter::with().adoptUnit(MeasureUnit::createMeter(status))
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* Currency:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
2017-12-22 00:02:01 +00:00
|
|
|
|
* NumberFormatter::with().unit(CurrencyUnit(u"USD", status))
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* Percent:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
2017-12-22 00:02:01 +00:00
|
|
|
|
* NumberFormatter::with().unit(NoUnit.percent())
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* </pre>
|
|
|
|
|
*
|
2017-12-22 00:02:01 +00:00
|
|
|
|
* See {@link #perUnit} for information on how to format strings like "5 meters per second".
|
|
|
|
|
*
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* The default is to render without units (equivalent to NoUnit.base()).
|
|
|
|
|
*
|
|
|
|
|
* @param unit
|
|
|
|
|
* The unit to render.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see MeasureUnit
|
|
|
|
|
* @see Currency
|
|
|
|
|
* @see NoUnit
|
2017-12-22 00:02:01 +00:00
|
|
|
|
* @see #perUnit
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived unit(const icu::MeasureUnit &unit) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of unit() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param unit
|
|
|
|
|
* The unit to render.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #unit
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived unit(const icu::MeasureUnit &unit) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Like unit(), but takes ownership of a pointer. Convenient for use with the MeasureFormat factory
|
2018-02-09 22:52:42 +00:00
|
|
|
|
* methods, which return pointers that need ownership. Example:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with().adoptUnit(MeasureUnit::createMeter(status))
|
|
|
|
|
* </pre>
|
2017-09-27 00:25:20 +00:00
|
|
|
|
*
|
|
|
|
|
* @param unit
|
2017-12-22 00:02:01 +00:00
|
|
|
|
* The unit to render.
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #unit
|
|
|
|
|
* @see MeasureUnit
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived adoptUnit(icu::MeasureUnit *unit) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of adoptUnit() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param unit
|
|
|
|
|
* The unit to render.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #adoptUnit
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived adoptUnit(icu::MeasureUnit *unit) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2017-12-22 00:02:01 +00:00
|
|
|
|
/**
|
|
|
|
|
* Sets a unit to be used in the denominator. For example, to format "3 m/s", pass METER to the unit and SECOND to
|
|
|
|
|
* the perUnit.
|
|
|
|
|
*
|
2018-02-09 22:52:42 +00:00
|
|
|
|
* Pass this method any instance of {@link MeasureUnit}. Since MeasureUnit factory methods return pointers, the
|
|
|
|
|
* {@link #adoptPerUnit} version of this method is often more useful.
|
2017-12-22 00:02:01 +00:00
|
|
|
|
*
|
|
|
|
|
* The default is not to display any unit in the denominator.
|
|
|
|
|
*
|
|
|
|
|
* If a per-unit is specified without a primary unit via {@link #unit}, the behavior is undefined.
|
|
|
|
|
*
|
|
|
|
|
* @param perUnit
|
|
|
|
|
* The unit to render in the denominator.
|
|
|
|
|
* @return The fluent chain
|
|
|
|
|
* @see #unit
|
|
|
|
|
* @draft ICU 61
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived perUnit(const icu::MeasureUnit &perUnit) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of perUnit() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param perUnit
|
|
|
|
|
* The unit to render in the denominator.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #perUnit
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived perUnit(const icu::MeasureUnit &perUnit) &&;
|
2017-12-22 00:02:01 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Like perUnit(), but takes ownership of a pointer. Convenient for use with the MeasureFormat factory
|
2018-02-09 22:52:42 +00:00
|
|
|
|
* methods, which return pointers that need ownership. Example:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with()
|
|
|
|
|
* .adoptUnit(MeasureUnit::createMeter(status))
|
|
|
|
|
* .adoptPerUnit(MeasureUnit::createSecond(status))
|
|
|
|
|
* </pre>
|
2017-12-22 00:02:01 +00:00
|
|
|
|
*
|
|
|
|
|
* @param perUnit
|
|
|
|
|
* The unit to render in the denominator.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #perUnit
|
|
|
|
|
* @see MeasureUnit
|
|
|
|
|
* @draft ICU 61
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived adoptPerUnit(icu::MeasureUnit *perUnit) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of adoptPerUnit() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param perUnit
|
|
|
|
|
* The unit to render in the denominator.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #adoptPerUnit
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived adoptPerUnit(icu::MeasureUnit *perUnit) &&;
|
2017-12-22 00:02:01 +00:00
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/**
|
|
|
|
|
* Specifies the rounding strategy to use when formatting numbers.
|
|
|
|
|
*
|
|
|
|
|
* <ul>
|
|
|
|
|
* <li>Round to 3 decimal places: "3.142"
|
|
|
|
|
* <li>Round to 3 significant figures: "3.14"
|
|
|
|
|
* <li>Round to the closest nickel: "3.15"
|
|
|
|
|
* <li>Do not perform rounding: "3.1415926..."
|
|
|
|
|
* </ul>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Pass this method the return value of one of the factory methods on {@link Rounder}. For example:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with().rounding(Rounder::fixedFraction(2))
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
2017-09-27 09:03:19 +00:00
|
|
|
|
* <p>
|
|
|
|
|
* In most cases, the default rounding strategy is to round to 6 fraction places; i.e.,
|
|
|
|
|
* <code>Rounder.maxFraction(6)</code>. The exceptions are if compact notation is being used, then the compact
|
|
|
|
|
* notation rounding strategy is used (see {@link Notation#compactShort} for details), or if the unit is a currency,
|
|
|
|
|
* then standard currency rounding is used, which varies from currency to currency (see {@link Rounder#currency} for
|
|
|
|
|
* details).
|
2017-09-27 00:25:20 +00:00
|
|
|
|
*
|
|
|
|
|
* @param rounder
|
|
|
|
|
* The rounding strategy to use.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see Rounder
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived rounding(const Rounder &rounder) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of rounding() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param rounder
|
|
|
|
|
* The rounding strategy to use.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #rounding
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived rounding(const Rounder& rounder) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Specifies the grouping strategy to use when formatting numbers.
|
|
|
|
|
*
|
|
|
|
|
* <ul>
|
|
|
|
|
* <li>Default grouping: "12,300" and "1,230"
|
|
|
|
|
* <li>Grouping with at least 2 digits: "12,300" and "1230"
|
|
|
|
|
* <li>No grouping: "12300" and "1230"
|
|
|
|
|
* </ul>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* The exact grouping widths will be chosen based on the locale.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
2018-02-06 03:08:17 +00:00
|
|
|
|
* Pass this method an element from the {@link UGroupingStrategy} enum. For example:
|
2017-09-27 00:25:20 +00:00
|
|
|
|
*
|
|
|
|
|
* <pre>
|
2018-02-06 03:08:17 +00:00
|
|
|
|
* NumberFormatter::with().grouping(UNUM_GROUPING_MIN2)
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* </pre>
|
|
|
|
|
*
|
2018-02-06 03:08:17 +00:00
|
|
|
|
* The default is to perform grouping according to locale data; most locales, but not all locales,
|
|
|
|
|
* enable it by default.
|
2017-09-27 00:25:20 +00:00
|
|
|
|
*
|
2018-02-06 03:08:17 +00:00
|
|
|
|
* @param strategy
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* The grouping strategy to use.
|
|
|
|
|
* @return The fluent chain.
|
2018-02-06 03:08:17 +00:00
|
|
|
|
* @draft ICU 61
|
2017-09-27 00:25:20 +00:00
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived grouping(const UGroupingStrategy &strategy) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of grouping() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param rounder
|
|
|
|
|
* The grouping strategy to use.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #grouping
|
|
|
|
|
* @provisional This API might change or be removed in a future release.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived grouping(const UGroupingStrategy& rounder) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Specifies the minimum and maximum number of digits to render before the decimal mark.
|
|
|
|
|
*
|
|
|
|
|
* <ul>
|
|
|
|
|
* <li>Zero minimum integer digits: ".08"
|
|
|
|
|
* <li>One minimum integer digit: "0.08"
|
|
|
|
|
* <li>Two minimum integer digits: "00.08"
|
|
|
|
|
* </ul>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Pass this method the return value of {@link IntegerWidth#zeroFillTo(int)}. For example:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with().integerWidth(IntegerWidth::zeroFillTo(2))
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* The default is to have one minimum integer digit.
|
|
|
|
|
*
|
|
|
|
|
* @param style
|
|
|
|
|
* The integer width to use.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see IntegerWidth
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived integerWidth(const IntegerWidth &style) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of integerWidth() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param style
|
|
|
|
|
* The integer width to use.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #integerWidth
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived integerWidth(const IntegerWidth &style) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Specifies the symbols (decimal separator, grouping separator, percent sign, numerals, etc.) to use when rendering
|
|
|
|
|
* numbers.
|
|
|
|
|
*
|
|
|
|
|
* <ul>
|
|
|
|
|
* <li><em>en_US</em> symbols: "12,345.67"
|
|
|
|
|
* <li><em>fr_FR</em> symbols: "12 345,67"
|
|
|
|
|
* <li><em>de_CH</em> symbols: "12’345.67"
|
|
|
|
|
* <li><em>my_MY</em> symbols: "၁၂,၃၄၅.၆၇"
|
|
|
|
|
* </ul>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Pass this method an instance of {@link DecimalFormatSymbols}. For example:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with().symbols(DecimalFormatSymbols(Locale("de_CH"), status))
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* <strong>Note:</strong> DecimalFormatSymbols automatically chooses the best numbering system based on the locale.
|
|
|
|
|
* In the examples above, the first three are using the Latin numbering system, and the fourth is using the Myanmar
|
|
|
|
|
* numbering system.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* <strong>Note:</strong> The instance of DecimalFormatSymbols will be copied: changes made to the symbols object
|
|
|
|
|
* after passing it into the fluent chain will not be seen.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* <strong>Note:</strong> Calling this method will override the NumberingSystem previously specified in
|
|
|
|
|
* {@link #symbols(NumberingSystem)}.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* The default is to choose the symbols based on the locale specified in the fluent chain.
|
|
|
|
|
*
|
|
|
|
|
* @param symbols
|
|
|
|
|
* The DecimalFormatSymbols to use.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see DecimalFormatSymbols
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived symbols(const DecimalFormatSymbols &symbols) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of symbols() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param symbols
|
|
|
|
|
* The DecimalFormatSymbols to use.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #symbols
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived symbols(const DecimalFormatSymbols &symbols) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Specifies that the given numbering system should be used when fetching symbols.
|
|
|
|
|
*
|
|
|
|
|
* <ul>
|
|
|
|
|
* <li>Latin numbering system: "12,345"
|
|
|
|
|
* <li>Myanmar numbering system: "၁၂,၃၄၅"
|
|
|
|
|
* <li>Math Sans Bold numbering system: "𝟭𝟮,𝟯𝟰𝟱"
|
|
|
|
|
* </ul>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Pass this method an instance of {@link NumberingSystem}. For example, to force the locale to always use the Latin
|
|
|
|
|
* alphabet numbering system (ASCII digits):
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with().adoptSymbols(NumberingSystem::createInstanceByName("latn", status))
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* <strong>Note:</strong> Calling this method will override the DecimalFormatSymbols previously specified in
|
|
|
|
|
* {@link #symbols(DecimalFormatSymbols)}.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* The default is to choose the best numbering system for the locale.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* This method takes ownership of a pointer in order to work nicely with the NumberingSystem factory methods.
|
|
|
|
|
*
|
2017-10-09 20:32:40 +00:00
|
|
|
|
* @param symbols
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* The NumberingSystem to use.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see NumberingSystem
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived adoptSymbols(NumberingSystem *symbols) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of adoptSymbols() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param symbols
|
|
|
|
|
* The NumberingSystem to use.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #adoptSymbols
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived adoptSymbols(NumberingSystem *symbols) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Sets the width of the unit (measure unit or currency). Most common values:
|
|
|
|
|
*
|
|
|
|
|
* <ul>
|
|
|
|
|
* <li>Short: "$12.00", "12 m"
|
|
|
|
|
* <li>ISO Code: "USD 12.00"
|
|
|
|
|
* <li>Full name: "12.00 US dollars", "12 meters"
|
|
|
|
|
* </ul>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Pass an element from the {@link UNumberUnitWidth} enum to this setter. For example:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with().unitWidth(UNumberUnitWidth::UNUM_UNIT_WIDTH_FULL_NAME)
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* The default is the SHORT width.
|
|
|
|
|
*
|
2017-10-09 20:32:40 +00:00
|
|
|
|
* @param width
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* The width to use when rendering numbers.
|
|
|
|
|
* @return The fluent chain
|
|
|
|
|
* @see UNumberUnitWidth
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived unitWidth(const UNumberUnitWidth &width) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of unitWidth() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param width
|
|
|
|
|
* The width to use when rendering numbers.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #unitWidth
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived unitWidth(const UNumberUnitWidth &width) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Sets the plus/minus sign display strategy. Most common values:
|
|
|
|
|
*
|
|
|
|
|
* <ul>
|
|
|
|
|
* <li>Auto: "123", "-123"
|
|
|
|
|
* <li>Always: "+123", "-123"
|
|
|
|
|
* <li>Accounting: "$123", "($123)"
|
|
|
|
|
* </ul>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Pass an element from the {@link UNumberSignDisplay} enum to this setter. For example:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with().sign(UNumberSignDisplay::UNUM_SIGN_ALWAYS)
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* The default is AUTO sign display.
|
|
|
|
|
*
|
2018-03-17 07:24:02 +00:00
|
|
|
|
* @param style
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* The sign display strategy to use when rendering numbers.
|
|
|
|
|
* @return The fluent chain
|
|
|
|
|
* @see UNumberSignDisplay
|
|
|
|
|
* @provisional This API might change or be removed in a future release.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived sign(const UNumberSignDisplay &style) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of sign() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param style
|
|
|
|
|
* The sign display strategy to use when rendering numbers.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #sign
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived sign(const UNumberSignDisplay &style) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Sets the decimal separator display strategy. This affects integer numbers with no fraction part. Most common
|
|
|
|
|
* values:
|
|
|
|
|
*
|
|
|
|
|
* <ul>
|
|
|
|
|
* <li>Auto: "1"
|
|
|
|
|
* <li>Always: "1."
|
|
|
|
|
* </ul>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Pass an element from the {@link UNumberDecimalSeparatorDisplay} enum to this setter. For example:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with().decimal(UNumberDecimalSeparatorDisplay::UNUM_DECIMAL_SEPARATOR_ALWAYS)
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* The default is AUTO decimal separator display.
|
|
|
|
|
*
|
2018-03-17 07:24:02 +00:00
|
|
|
|
* @param style
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* The decimal separator display strategy to use when rendering numbers.
|
|
|
|
|
* @return The fluent chain
|
|
|
|
|
* @see UNumberDecimalSeparatorDisplay
|
|
|
|
|
* @provisional This API might change or be removed in a future release.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived decimal(const UNumberDecimalSeparatorDisplay &style) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of decimal() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param style
|
|
|
|
|
* The decimal separator display strategy to use when rendering numbers.
|
|
|
|
|
* @return The fluent chain.
|
2018-04-03 04:38:16 +00:00
|
|
|
|
* @see #decimal
|
2018-03-17 07:24:02 +00:00
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived decimal(const UNumberDecimalSeparatorDisplay &style) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-04-03 04:38:16 +00:00
|
|
|
|
/**
|
|
|
|
|
* Sets a multiplier to be used to scale the number by an arbitrary amount before formatting. Most
|
|
|
|
|
* common values:
|
|
|
|
|
*
|
|
|
|
|
* <ul>
|
|
|
|
|
* <li>Multiply by 100: useful for percentages.
|
|
|
|
|
* <li>Multiply by an arbitrary value: useful for unit conversions.
|
|
|
|
|
* </ul>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* Pass an element from a {@link Multiplier} factory method to this setter. For example:
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with().multiplier(Multiplier::powerOfTen(2))
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* The default is to not apply any multiplier.
|
|
|
|
|
*
|
|
|
|
|
* @param style
|
|
|
|
|
* The decimal separator display strategy to use when rendering numbers.
|
|
|
|
|
* @return The fluent chain
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
Derived multiplier(const Multiplier &style) const &;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of multiplier() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param style
|
|
|
|
|
* The multiplier separator display strategy to use when rendering numbers.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #multiplier
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
Derived multiplier(const Multiplier &style) &&;
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
#ifndef U_HIDE_INTERNAL_API
|
|
|
|
|
|
|
|
|
|
/**
|
2018-03-17 07:24:02 +00:00
|
|
|
|
* Set the padding strategy. May be added in the future; see #13338.
|
2017-09-27 00:25:20 +00:00
|
|
|
|
*
|
|
|
|
|
* @internal ICU 60: This API is ICU internal only.
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived padding(const impl::Padder &padder) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/** @internal */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived padding(const impl::Padder &padder) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Internal fluent setter to support a custom regulation threshold. A threshold of 1 causes the data structures to
|
|
|
|
|
* be built right away. A threshold of 0 prevents the data structures from being built.
|
|
|
|
|
*
|
|
|
|
|
* @internal ICU 60: This API is ICU internal only.
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived threshold(int32_t threshold) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/** @internal */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived threshold(int32_t threshold) &&;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-13 09:25:41 +00:00
|
|
|
|
/**
|
|
|
|
|
* Internal fluent setter to overwrite the entire macros object.
|
|
|
|
|
*
|
|
|
|
|
* @internal ICU 60: This API is ICU internal only.
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived macros(const impl::MacroProps& macros) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/** @internal */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived macros(const impl::MacroProps& macros) &&;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/** @internal */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived macros(impl::MacroProps&& macros) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/** @internal */
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Derived macros(impl::MacroProps&& macros) &&;
|
2018-03-13 09:25:41 +00:00
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
#endif /* U_HIDE_INTERNAL_API */
|
|
|
|
|
|
2018-03-24 05:46:28 +00:00
|
|
|
|
/**
|
|
|
|
|
* Creates a skeleton string representation of this number formatter. A skeleton string is a
|
|
|
|
|
* locale-agnostic serialized form of a number formatter.
|
2018-03-29 00:47:26 +00:00
|
|
|
|
*
|
2018-03-24 05:46:28 +00:00
|
|
|
|
* Not all options are capable of being represented in the skeleton string; for example, a
|
2018-03-29 00:47:26 +00:00
|
|
|
|
* DecimalFormatSymbols object. If any such option is encountered, the error code is set to
|
|
|
|
|
* U_UNSUPPORTED_ERROR.
|
|
|
|
|
*
|
2018-03-24 05:46:28 +00:00
|
|
|
|
* The returned skeleton is in normalized form, such that two number formatters with equivalent
|
|
|
|
|
* behavior should produce the same skeleton.
|
|
|
|
|
*
|
|
|
|
|
* @return A number skeleton string with behavior corresponding to this number formatter.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
UnicodeString toSkeleton(UErrorCode& status) const;
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
/**
|
|
|
|
|
* Sets the UErrorCode if an error occurred in the fluent chain.
|
|
|
|
|
* Preserves older error codes in the outErrorCode.
|
|
|
|
|
* @return TRUE if U_FAILURE(outErrorCode)
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
UBool copyErrorTo(UErrorCode &outErrorCode) const {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
if (U_FAILURE(outErrorCode)) {
|
|
|
|
|
// Do not overwrite the older error code
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
fMacros.copyErrorTo(outErrorCode);
|
|
|
|
|
return U_FAILURE(outErrorCode);
|
2018-03-24 05:46:28 +00:00
|
|
|
|
};
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
// NOTE: Uses default copy and move constructors.
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
protected:
|
|
|
|
|
impl::MacroProps fMacros;
|
|
|
|
|
|
|
|
|
|
private:
|
|
|
|
|
// Don't construct me directly! Use (Un)LocalizedNumberFormatter.
|
|
|
|
|
NumberFormatterSettings() = default;
|
|
|
|
|
|
|
|
|
|
friend class LocalizedNumberFormatter;
|
|
|
|
|
friend class UnlocalizedNumberFormatter;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* A NumberFormatter that does not yet have a locale. In order to format numbers, a locale must be specified.
|
|
|
|
|
*
|
|
|
|
|
* @see NumberFormatter
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API UnlocalizedNumberFormatter
|
2017-09-27 00:25:20 +00:00
|
|
|
|
: public NumberFormatterSettings<UnlocalizedNumberFormatter>, public UMemory {
|
|
|
|
|
|
|
|
|
|
public:
|
2017-10-05 23:03:19 +00:00
|
|
|
|
/**
|
|
|
|
|
* Associate the given locale with the number formatter. The locale is used for picking the appropriate symbols,
|
|
|
|
|
* formats, and other data for number display.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* To use the Java default locale, call Locale::getDefault():
|
|
|
|
|
*
|
|
|
|
|
* <pre>
|
|
|
|
|
* NumberFormatter::with(). ... .locale(Locale::getDefault())
|
|
|
|
|
* </pre>
|
|
|
|
|
*
|
|
|
|
|
* @param locale
|
|
|
|
|
* The locale to use when loading data for number formatting.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
LocalizedNumberFormatter locale(const icu::Locale &locale) const &;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Overload of locale() for use on an rvalue reference.
|
|
|
|
|
*
|
|
|
|
|
* @param locale
|
|
|
|
|
* The locale to use when loading data for number formatting.
|
|
|
|
|
* @return The fluent chain.
|
|
|
|
|
* @see #locale
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
LocalizedNumberFormatter locale(const icu::Locale &locale) &&;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Default constructor: puts the formatter into a valid but undefined state.
|
|
|
|
|
*
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
UnlocalizedNumberFormatter() = default;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
// Make default copy constructor call the NumberFormatterSettings copy constructor.
|
2017-10-05 23:03:19 +00:00
|
|
|
|
/**
|
|
|
|
|
* Returns a copy of this UnlocalizedNumberFormatter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
UnlocalizedNumberFormatter(const UnlocalizedNumberFormatter &other);
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Move constructor:
|
|
|
|
|
* The source UnlocalizedNumberFormatter will be left in a valid but undefined state.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
UnlocalizedNumberFormatter(UnlocalizedNumberFormatter&& src) U_NOEXCEPT;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Copy assignment operator.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
UnlocalizedNumberFormatter& operator=(const UnlocalizedNumberFormatter& other);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Move assignment operator:
|
|
|
|
|
* The source UnlocalizedNumberFormatter will be left in a valid but undefined state.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
UnlocalizedNumberFormatter& operator=(UnlocalizedNumberFormatter&& src) U_NOEXCEPT;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
private:
|
2018-03-17 07:24:02 +00:00
|
|
|
|
explicit UnlocalizedNumberFormatter(const NumberFormatterSettings<UnlocalizedNumberFormatter>& other);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
explicit UnlocalizedNumberFormatter(
|
2018-03-17 07:24:02 +00:00
|
|
|
|
NumberFormatterSettings<UnlocalizedNumberFormatter>&& src) U_NOEXCEPT;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
// To give the fluent setters access to this class's constructor:
|
2017-09-27 06:44:13 +00:00
|
|
|
|
friend class NumberFormatterSettings<UnlocalizedNumberFormatter>;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
// To give NumberFormatter::with() access to this class's constructor:
|
|
|
|
|
friend class NumberFormatter;
|
|
|
|
|
};
|
|
|
|
|
|
2017-09-27 09:03:19 +00:00
|
|
|
|
/**
|
|
|
|
|
* A NumberFormatter that has a locale associated with it; this means .format() methods are available.
|
|
|
|
|
*
|
|
|
|
|
* @see NumberFormatter
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API LocalizedNumberFormatter
|
2017-09-27 00:25:20 +00:00
|
|
|
|
: public NumberFormatterSettings<LocalizedNumberFormatter>, public UMemory {
|
|
|
|
|
public:
|
|
|
|
|
/**
|
|
|
|
|
* Format the given integer number to a string using the settings specified in the NumberFormatter fluent
|
|
|
|
|
* setting chain.
|
|
|
|
|
*
|
2017-10-09 20:32:40 +00:00
|
|
|
|
* @param value
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* The number to format.
|
|
|
|
|
* @param status
|
2017-10-06 22:44:25 +00:00
|
|
|
|
* Set to an ErrorCode if one occurred in the setter chain or during formatting.
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* @return A FormattedNumber object; call .toString() to get the string.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
FormattedNumber formatInt(int64_t value, UErrorCode &status) const;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Format the given float or double to a string using the settings specified in the NumberFormatter fluent setting
|
|
|
|
|
* chain.
|
|
|
|
|
*
|
2017-10-09 20:32:40 +00:00
|
|
|
|
* @param value
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* The number to format.
|
|
|
|
|
* @param status
|
2017-10-06 22:44:25 +00:00
|
|
|
|
* Set to an ErrorCode if one occurred in the setter chain or during formatting.
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* @return A FormattedNumber object; call .toString() to get the string.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
FormattedNumber formatDouble(double value, UErrorCode &status) const;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Format the given decimal number to a string using the settings
|
|
|
|
|
* specified in the NumberFormatter fluent setting chain.
|
|
|
|
|
* The syntax of the unformatted number is a "numeric string"
|
|
|
|
|
* as defined in the Decimal Arithmetic Specification, available at
|
|
|
|
|
* http://speleotrove.com/decimal
|
|
|
|
|
*
|
2017-10-09 20:32:40 +00:00
|
|
|
|
* @param value
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* The number to format.
|
|
|
|
|
* @param status
|
2017-10-06 22:44:25 +00:00
|
|
|
|
* Set to an ErrorCode if one occurred in the setter chain or during formatting.
|
2017-09-27 00:25:20 +00:00
|
|
|
|
* @return A FormattedNumber object; call .toString() to get the string.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-17 07:24:02 +00:00
|
|
|
|
FormattedNumber formatDecimal(StringPiece value, UErrorCode& status) const;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-13 10:11:36 +00:00
|
|
|
|
#ifndef U_HIDE_INTERNAL_API
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
2018-03-13 10:11:36 +00:00
|
|
|
|
/** Internal method.
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
|
|
|
|
FormattedNumber formatDecimalQuantity(const impl::DecimalQuantity& dq, UErrorCode& status) const;
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Internal method for testing.
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
|
|
|
|
const impl::NumberFormatterImpl* getCompiled() const;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Internal method for testing.
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
|
|
|
|
int32_t getCallCount() const;
|
|
|
|
|
|
2018-03-13 10:11:36 +00:00
|
|
|
|
#endif
|
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
/**
|
|
|
|
|
* Default constructor: puts the formatter into a valid but undefined state.
|
|
|
|
|
*
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
LocalizedNumberFormatter() = default;
|
|
|
|
|
|
2017-09-27 00:25:20 +00:00
|
|
|
|
// Make default copy constructor call the NumberFormatterSettings copy constructor.
|
2017-10-05 23:03:19 +00:00
|
|
|
|
/**
|
|
|
|
|
* Returns a copy of this LocalizedNumberFormatter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
LocalizedNumberFormatter(const LocalizedNumberFormatter &other);
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Move constructor:
|
|
|
|
|
* The source LocalizedNumberFormatter will be left in a valid but undefined state.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
LocalizedNumberFormatter(LocalizedNumberFormatter&& src) U_NOEXCEPT;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Copy assignment operator.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
LocalizedNumberFormatter& operator=(const LocalizedNumberFormatter& other);
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Move assignment operator:
|
|
|
|
|
* The source LocalizedNumberFormatter will be left in a valid but undefined state.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
LocalizedNumberFormatter& operator=(LocalizedNumberFormatter&& src) U_NOEXCEPT;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-27 01:58:26 +00:00
|
|
|
|
#ifndef U_HIDE_INTERNAL_API
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* This is the core entrypoint to the number formatting pipeline. It performs self-regulation: a static code path
|
|
|
|
|
* for the first few calls, and compiling a more efficient data structure if called repeatedly.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* This function is very hot, being called in every call to the number formatting pipeline.
|
|
|
|
|
*
|
|
|
|
|
* @param results
|
|
|
|
|
* The results object. This method will mutate it to save the results.
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
void formatImpl(impl::UFormattedNumberData *results, UErrorCode &status) const;
|
2018-03-27 01:58:26 +00:00
|
|
|
|
|
|
|
|
|
#endif
|
|
|
|
|
|
2017-10-05 23:03:19 +00:00
|
|
|
|
/**
|
|
|
|
|
* Destruct this LocalizedNumberFormatter, cleaning up any memory it might own.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 00:25:20 +00:00
|
|
|
|
~LocalizedNumberFormatter();
|
|
|
|
|
|
|
|
|
|
private:
|
2018-03-17 07:24:02 +00:00
|
|
|
|
// Note: fCompiled can't be a LocalPointer because impl::NumberFormatterImpl is defined in an internal
|
|
|
|
|
// header, and LocalPointer needs the full class definition in order to delete the instance.
|
2018-03-28 00:39:28 +00:00
|
|
|
|
const impl::NumberFormatterImpl* fCompiled {nullptr};
|
|
|
|
|
char fUnsafeCallCount[8] {}; // internally cast to u_atomic_int32_t
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
explicit LocalizedNumberFormatter(const NumberFormatterSettings<LocalizedNumberFormatter>& other);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
explicit LocalizedNumberFormatter(NumberFormatterSettings<LocalizedNumberFormatter>&& src) U_NOEXCEPT;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
LocalizedNumberFormatter(const impl::MacroProps ¯os, const Locale &locale);
|
2018-03-17 07:24:02 +00:00
|
|
|
|
|
2018-03-28 00:39:28 +00:00
|
|
|
|
LocalizedNumberFormatter(impl::MacroProps &¯os, const Locale &locale);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
// To give the fluent setters access to this class's constructor:
|
2017-09-27 06:44:13 +00:00
|
|
|
|
friend class NumberFormatterSettings<UnlocalizedNumberFormatter>;
|
|
|
|
|
friend class NumberFormatterSettings<LocalizedNumberFormatter>;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
// To give UnlocalizedNumberFormatter::locale() access to this class's constructor:
|
|
|
|
|
friend class UnlocalizedNumberFormatter;
|
|
|
|
|
};
|
|
|
|
|
|
2017-09-27 09:03:19 +00:00
|
|
|
|
/**
|
|
|
|
|
* The result of a number formatting operation. This class allows the result to be exported in several data types,
|
|
|
|
|
* including a UnicodeString and a FieldPositionIterator.
|
|
|
|
|
*
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API FormattedNumber : public UMemory {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
public:
|
2017-09-27 09:03:19 +00:00
|
|
|
|
/**
|
2017-10-06 22:44:25 +00:00
|
|
|
|
* Returns a UnicodeString representation of the formatted number.
|
2017-09-27 09:03:19 +00:00
|
|
|
|
*
|
|
|
|
|
* @return a UnicodeString containing the localized number.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 00:25:20 +00:00
|
|
|
|
UnicodeString toString() const;
|
|
|
|
|
|
2017-09-27 09:03:19 +00:00
|
|
|
|
/**
|
|
|
|
|
* Appends the formatted number to an Appendable.
|
|
|
|
|
*
|
|
|
|
|
* @param appendable
|
|
|
|
|
* The Appendable to which to append the formatted number string.
|
|
|
|
|
* @return The same Appendable, for chaining.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
* @see Appendable
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
Appendable &appendTo(Appendable &appendable);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2017-09-27 09:03:19 +00:00
|
|
|
|
/**
|
|
|
|
|
* Determine the start and end indices of the first occurrence of the given <em>field</em> in the output string.
|
|
|
|
|
* This allows you to determine the locations of the integer part, fraction part, and sign.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* If multiple different field attributes are needed, this method can be called repeatedly, or if <em>all</em> field
|
|
|
|
|
* attributes are needed, consider using populateFieldPositionIterator().
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* If a field occurs multiple times in an output string, such as a grouping separator, this method will only ever
|
|
|
|
|
* return the first occurrence. Use populateFieldPositionIterator() to access all occurrences of an attribute.
|
|
|
|
|
*
|
|
|
|
|
* @param fieldPosition
|
|
|
|
|
* The FieldPosition to populate with the start and end indices of the desired field.
|
|
|
|
|
* @param status
|
|
|
|
|
* Set if an error occurs while populating the FieldPosition.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
* @see UNumberFormatFields
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
void populateFieldPosition(FieldPosition &fieldPosition, UErrorCode &status);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2017-09-27 09:03:19 +00:00
|
|
|
|
/**
|
|
|
|
|
* Export the formatted number to a FieldPositionIterator. This allows you to determine which characters in
|
|
|
|
|
* the output string correspond to which <em>fields</em>, such as the integer part, fraction part, and sign.
|
|
|
|
|
*
|
|
|
|
|
* <p>
|
|
|
|
|
* If information on only one field is needed, consider using populateFieldPosition() instead.
|
|
|
|
|
*
|
|
|
|
|
* @param iterator
|
|
|
|
|
* The FieldPositionIterator to populate with all of the fields present in the formatted number.
|
|
|
|
|
* @param status
|
|
|
|
|
* Set if an error occurs while populating the FieldPositionIterator.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
* @see UNumberFormatFields
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
void populateFieldPositionIterator(FieldPositionIterator &iterator, UErrorCode &status);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-03 08:26:58 +00:00
|
|
|
|
#ifndef U_HIDE_INTERNAL_API
|
2018-03-14 09:15:27 +00:00
|
|
|
|
|
2018-03-03 08:26:58 +00:00
|
|
|
|
/**
|
2018-03-14 10:41:27 +00:00
|
|
|
|
* Gets the raw DecimalQuantity for plural rule selection.
|
2018-03-03 08:26:58 +00:00
|
|
|
|
* @internal
|
|
|
|
|
*/
|
2018-03-14 10:41:27 +00:00
|
|
|
|
void getDecimalQuantity(impl::DecimalQuantity& output, UErrorCode& status) const;
|
2018-03-14 09:15:27 +00:00
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
const UnicodeString getPrefix(UErrorCode& status) const;
|
|
|
|
|
|
|
|
|
|
/** @internal */
|
|
|
|
|
const UnicodeString getSuffix(UErrorCode& status) const;
|
|
|
|
|
|
2018-03-03 08:26:58 +00:00
|
|
|
|
#endif
|
|
|
|
|
|
2018-03-17 07:24:02 +00:00
|
|
|
|
// Don't allow copying of FormattedNumber, but moving is okay.
|
|
|
|
|
FormattedNumber(const FormattedNumber&) = delete;
|
|
|
|
|
FormattedNumber& operator=(const FormattedNumber&) = delete;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Move constructor:
|
|
|
|
|
* Leaves the source FormattedNumber in an undefined state.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
FormattedNumber(FormattedNumber&& src) U_NOEXCEPT;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Move assignment:
|
|
|
|
|
* Leaves the source FormattedNumber in an undefined state.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
FormattedNumber& operator=(FormattedNumber&& src) U_NOEXCEPT;
|
|
|
|
|
|
2017-10-05 23:03:19 +00:00
|
|
|
|
/**
|
|
|
|
|
* Destruct an instance of FormattedNumber, cleaning up any memory it might own.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 00:25:20 +00:00
|
|
|
|
~FormattedNumber();
|
|
|
|
|
|
|
|
|
|
private:
|
2018-03-27 01:58:26 +00:00
|
|
|
|
// Can't use LocalPointer because UFormattedNumberData is forward-declared
|
2018-03-28 00:39:28 +00:00
|
|
|
|
const impl::UFormattedNumberData *fResults;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2017-10-25 01:26:48 +00:00
|
|
|
|
// Error code for the terminal methods
|
|
|
|
|
UErrorCode fErrorCode;
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-27 01:58:26 +00:00
|
|
|
|
/**
|
|
|
|
|
* Internal constructor from data type. Adopts the data pointer.
|
|
|
|
|
* @internal
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
explicit FormattedNumber(impl::UFormattedNumberData *results)
|
|
|
|
|
: fResults(results), fErrorCode(U_ZERO_ERROR) {};
|
2017-10-25 01:26:48 +00:00
|
|
|
|
|
|
|
|
|
explicit FormattedNumber(UErrorCode errorCode)
|
2018-03-28 00:39:28 +00:00
|
|
|
|
: fResults(nullptr), fErrorCode(errorCode) {};
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
|
|
|
|
// To give LocalizedNumberFormatter format methods access to this class's constructor:
|
|
|
|
|
friend class LocalizedNumberFormatter;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/**
|
2017-09-27 09:03:19 +00:00
|
|
|
|
* See the main description in numberformatter.h for documentation and examples.
|
2017-09-27 00:25:20 +00:00
|
|
|
|
*
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 18:50:44 +00:00
|
|
|
|
class U_I18N_API NumberFormatter final {
|
2017-09-27 00:25:20 +00:00
|
|
|
|
public:
|
|
|
|
|
/**
|
|
|
|
|
* Call this method at the beginning of a NumberFormatter fluent chain in which the locale is not currently known at
|
|
|
|
|
* the call site.
|
|
|
|
|
*
|
|
|
|
|
* @return An {@link UnlocalizedNumberFormatter}, to be used for chaining.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
|
|
|
|
static UnlocalizedNumberFormatter with();
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Call this method at the beginning of a NumberFormatter fluent chain in which the locale is known at the call
|
|
|
|
|
* site.
|
|
|
|
|
*
|
|
|
|
|
* @param locale
|
|
|
|
|
* The locale from which to load formats and symbols for number formatting.
|
|
|
|
|
* @return A {@link LocalizedNumberFormatter}, to be used for chaining.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2018-03-28 00:39:28 +00:00
|
|
|
|
static LocalizedNumberFormatter withLocale(const Locale &locale);
|
2017-09-27 00:25:20 +00:00
|
|
|
|
|
2018-03-24 05:46:28 +00:00
|
|
|
|
/**
|
|
|
|
|
* Call this method at the beginning of a NumberFormatter fluent chain to create an instance based
|
|
|
|
|
* on a given number skeleton string.
|
|
|
|
|
*
|
|
|
|
|
* @param skeleton
|
|
|
|
|
* The skeleton string off of which to base this NumberFormatter.
|
2018-03-29 00:47:26 +00:00
|
|
|
|
* @param status
|
|
|
|
|
* Set to U_NUMBER_SKELETON_SYNTAX_ERROR if the skeleton was invalid.
|
2018-03-24 05:46:28 +00:00
|
|
|
|
* @return An UnlocalizedNumberFormatter, to be used for chaining.
|
|
|
|
|
* @draft ICU 62
|
|
|
|
|
*/
|
|
|
|
|
static UnlocalizedNumberFormatter fromSkeleton(const UnicodeString& skeleton, UErrorCode& status);
|
|
|
|
|
|
2017-10-05 23:03:19 +00:00
|
|
|
|
/**
|
|
|
|
|
* Use factory methods instead of the constructor to create a NumberFormatter.
|
|
|
|
|
* @draft ICU 60
|
|
|
|
|
*/
|
2017-09-27 00:25:20 +00:00
|
|
|
|
NumberFormatter() = delete;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
} // namespace number
|
|
|
|
|
U_NAMESPACE_END
|
|
|
|
|
|
2017-09-27 09:03:19 +00:00
|
|
|
|
#endif // U_HIDE_DRAFT_API
|
|
|
|
|
|
2017-09-27 05:31:57 +00:00
|
|
|
|
#endif // __NUMBERFORMATTER_H__
|
|
|
|
|
|
|
|
|
|
#endif /* #if !UCONFIG_NO_FORMATTING */
|