|
|
// © 2018 and later: Unicode, Inc. and others.
|
|
|
// License & terms of use: http://www.unicode.org/copyright.html
|
|
|
|
|
|
#include "unicode/utypes.h"
|
|
|
|
|
|
#if !UCONFIG_NO_FORMATTING
|
|
|
|
|
|
// Allow implicit conversion from char16_t* to UnicodeString for this file:
|
|
|
// Helpful in toString methods and elsewhere.
|
|
|
#define UNISTR_FROM_STRING_EXPLICIT
|
|
|
|
|
|
#include "unicode/numberrangeformatter.h"
|
|
|
#include "numrange_impl.h"
|
|
|
#include "patternprops.h"
|
|
|
#include "pluralranges.h"
|
|
|
#include "uresimp.h"
|
|
|
#include "util.h"
|
|
|
|
|
|
using namespace icu;
|
|
|
using namespace icu::number;
|
|
|
using namespace icu::number::impl;
|
|
|
|
|
|
namespace {
|
|
|
|
|
|
// Helper function for 2-dimensional switch statement
|
|
|
constexpr int8_t identity2d(UNumberRangeIdentityFallback a, UNumberRangeIdentityResult b) {
|
|
|
return static_cast<int8_t>(a) | (static_cast<int8_t>(b) << 4);
|
|
|
}
|
|
|
|
|
|
|
|
|
struct NumberRangeData {
|
|
|
SimpleFormatter rangePattern;
|
|
|
SimpleFormatter approximatelyPattern;
|
|
|
};
|
|
|
|
|
|
class NumberRangeDataSink : public ResourceSink {
|
|
|
public:
|
|
|
NumberRangeDataSink(NumberRangeData& data) : fData(data) {}
|
|
|
|
|
|
void put(const char* key, ResourceValue& value, UBool /*noFallback*/, UErrorCode& status) U_OVERRIDE {
|
|
|
ResourceTable miscTable = value.getTable(status);
|
|
|
if (U_FAILURE(status)) { return; }
|
|
|
for (int i = 0; miscTable.getKeyAndValue(i, key, value); i++) {
|
|
|
if (uprv_strcmp(key, "range") == 0) {
|
|
|
if (hasRangeData()) {
|
|
|
continue; // have already seen this pattern
|
|
|
}
|
|
|
fData.rangePattern = {value.getUnicodeString(status), status};
|
|
|
} else if (uprv_strcmp(key, "approximately") == 0) {
|
|
|
if (hasApproxData()) {
|
|
|
continue; // have already seen this pattern
|
|
|
}
|
|
|
fData.approximatelyPattern = {value.getUnicodeString(status), status};
|
|
|
}
|
|
|
}
|
|
|
}
|
|
|
|
|
|
bool hasRangeData() {
|
|
|
return fData.rangePattern.getArgumentLimit() != 0;
|
|
|
}
|
|
|
|
|
|
bool hasApproxData() {
|
|
|
return fData.approximatelyPattern.getArgumentLimit() != 0;
|
|
|
}
|
|
|
|
|
|
bool isComplete() {
|
|
|
return hasRangeData() && hasApproxData();
|
|
|
}
|
|
|
|
|
|
void fillInDefaults(UErrorCode& status) {
|
|
|
if (!hasRangeData()) {
|
|
|
fData.rangePattern = {u"{0}–{1}", status};
|
|
|
}
|
|
|
if (!hasApproxData()) {
|
|
|
fData.approximatelyPattern = {u"~{0}", status};
|
|
|
}
|
|
|
}
|
|
|
|
|
|
private:
|
|
|
NumberRangeData& fData;
|
|
|
};
|
|
|
|
|
|
void getNumberRangeData(const char* localeName, const char* nsName, NumberRangeData& data, UErrorCode& status) {
|
|
|
if (U_FAILURE(status)) { return; }
|
|
|
LocalUResourceBundlePointer rb(ures_open(NULL, localeName, &status));
|
|
|
if (U_FAILURE(status)) { return; }
|
|
|
NumberRangeDataSink sink(data);
|
|
|
|
|
|
CharString dataPath;
|
|
|
dataPath.append("NumberElements/", -1, status);
|
|
|
dataPath.append(nsName, -1, status);
|
|
|
dataPath.append("/miscPatterns", -1, status);
|
|
|
if (U_FAILURE(status)) { return; }
|
|
|
|
|
|
UErrorCode localStatus = U_ZERO_ERROR;
|
|
|
ures_getAllItemsWithFallback(rb.getAlias(), dataPath.data(), sink, localStatus);
|
|
|
if (U_FAILURE(localStatus) && localStatus != U_MISSING_RESOURCE_ERROR) {
|
|
|
status = localStatus;
|
|
|
return;
|
|
|
}
|
|
|
|
|
|
// Fall back to latn if necessary
|
|
|
if (!sink.isComplete()) {
|
|
|
ures_getAllItemsWithFallback(rb.getAlias(), "NumberElements/latn/miscPatterns", sink, status);
|
|
|
}
|
|
|
|
|
|
sink.fillInDefaults(status);
|
|
|
}
|
|
|
|
|
|
} // namespace
|
|
|
|
|
|
|
|
|
|
|
|
NumberRangeFormatterImpl::NumberRangeFormatterImpl(const RangeMacroProps& macros, UErrorCode& status)
|
|
|
: formatterImpl1(macros.formatter1.fMacros, status),
|
|
|
formatterImpl2(macros.formatter2.fMacros, status),
|
|
|
fSameFormatters(macros.singleFormatter),
|
|
|
fCollapse(macros.collapse),
|
|
|
fIdentityFallback(macros.identityFallback) {
|
|
|
|
|
|
const char* nsName = formatterImpl1.getRawMicroProps().nsName;
|
|
|
if (uprv_strcmp(nsName, formatterImpl2.getRawMicroProps().nsName) != 0) {
|
|
|
status = U_ILLEGAL_ARGUMENT_ERROR;
|
|
|
return;
|
|
|
}
|
|
|
|
|
|
NumberRangeData data;
|
|
|
getNumberRangeData(macros.locale.getName(), nsName, data, status);
|
|
|
if (U_FAILURE(status)) { return; }
|
|
|
fRangeFormatter = data.rangePattern;
|
|
|
fApproximatelyModifier = {data.approximatelyPattern, kUndefinedField, false};
|
|
|
|
|
|
// TODO: Get locale from PluralRules instead?
|
|
|
fPluralRanges = StandardPluralRanges::forLocale(macros.locale, status);
|
|
|
if (U_FAILURE(status)) { return; }
|
|
|
}
|
|
|
|
|
|
void NumberRangeFormatterImpl::format(UFormattedNumberRangeData& data, bool equalBeforeRounding, UErrorCode& status) const {
|
|
|
if (U_FAILURE(status)) {
|
|
|
return;
|
|
|
}
|
|
|
|
|
|
MicroProps micros1;
|
|
|
MicroProps micros2;
|
|
|
formatterImpl1.preProcess(data.quantity1, micros1, status);
|
|
|
if (fSameFormatters) {
|
|
|
formatterImpl1.preProcess(data.quantity2, micros2, status);
|
|
|
} else {
|
|
|
formatterImpl2.preProcess(data.quantity2, micros2, status);
|
|
|
}
|
|
|
if (U_FAILURE(status)) {
|
|
|
return;
|
|
|
}
|
|
|
|
|
|
// If any of the affixes are different, an identity is not possible
|
|
|
// and we must use formatRange().
|
|
|
// TODO: Write this as MicroProps operator==() ?
|
|
|
// TODO: Avoid the redundancy of these equality operations with the
|
|
|
// ones in formatRange?
|
|
|
if (!micros1.modInner->semanticallyEquivalent(*micros2.modInner)
|
|
|
|| !micros1.modMiddle->semanticallyEquivalent(*micros2.modMiddle)
|
|
|
|| !micros1.modOuter->semanticallyEquivalent(*micros2.modOuter)) {
|
|
|
formatRange(data, micros1, micros2, status);
|
|
|
data.identityResult = UNUM_IDENTITY_RESULT_NOT_EQUAL;
|
|
|
return;
|
|
|
}
|
|
|
|
|
|
// Check for identity
|
|
|
if (equalBeforeRounding) {
|
|
|
data.identityResult = UNUM_IDENTITY_RESULT_EQUAL_BEFORE_ROUNDING;
|
|
|
} else if (data.quantity1 == data.quantity2) {
|
|
|
data.identityResult = UNUM_IDENTITY_RESULT_EQUAL_AFTER_ROUNDING;
|
|
|
} else {
|
|
|
data.identityResult = UNUM_IDENTITY_RESULT_NOT_EQUAL;
|
|
|
}
|
|
|
|
|
|
switch (identity2d(fIdentityFallback, data.identityResult)) {
|
|
|
case identity2d(UNUM_IDENTITY_FALLBACK_RANGE,
|
|
|
UNUM_IDENTITY_RESULT_NOT_EQUAL):
|
|
|
case identity2d(UNUM_IDENTITY_FALLBACK_RANGE,
|
|
|
UNUM_IDENTITY_RESULT_EQUAL_AFTER_ROUNDING):
|
|
|
case identity2d(UNUM_IDENTITY_FALLBACK_RANGE,
|
|
|
UNUM_IDENTITY_RESULT_EQUAL_BEFORE_ROUNDING):
|
|
|
case identity2d(UNUM_IDENTITY_FALLBACK_APPROXIMATELY,
|
|
|
UNUM_IDENTITY_RESULT_NOT_EQUAL):
|
|
|
case identity2d(UNUM_IDENTITY_FALLBACK_APPROXIMATELY_OR_SINGLE_VALUE,
|
|
|
UNUM_IDENTITY_RESULT_NOT_EQUAL):
|
|
|
case identity2d(UNUM_IDENTITY_FALLBACK_SINGLE_VALUE,
|
|
|
UNUM_IDENTITY_RESULT_NOT_EQUAL):
|
|
|
formatRange(data, micros1, micros2, status);
|
|
|
break;
|
|
|
|
|
|
case identity2d(UNUM_IDENTITY_FALLBACK_APPROXIMATELY,
|
|
|
UNUM_IDENTITY_RESULT_EQUAL_AFTER_ROUNDING):
|
|
|
case identity2d(UNUM_IDENTITY_FALLBACK_APPROXIMATELY,
|
|
|
UNUM_IDENTITY_RESULT_EQUAL_BEFORE_ROUNDING):
|
|
|
case identity2d(UNUM_IDENTITY_FALLBACK_APPROXIMATELY_OR_SINGLE_VALUE,
|
|
|
UNUM_IDENTITY_RESULT_EQUAL_AFTER_ROUNDING):
|
|
|
formatApproximately(data, micros1, micros2, status);
|
|
|
break;
|
|
|
|
|
|
case identity2d(UNUM_IDENTITY_FALLBACK_APPROXIMATELY_OR_SINGLE_VALUE,
|
|
|
UNUM_IDENTITY_RESULT_EQUAL_BEFORE_ROUNDING):
|
|
|
case identity2d(UNUM_IDENTITY_FALLBACK_SINGLE_VALUE,
|
|
|
UNUM_IDENTITY_RESULT_EQUAL_AFTER_ROUNDING):
|
|
|
case identity2d(UNUM_IDENTITY_FALLBACK_SINGLE_VALUE,
|
|
|
UNUM_IDENTITY_RESULT_EQUAL_BEFORE_ROUNDING):
|
|
|
formatSingleValue(data, micros1, micros2, status);
|
|
|
break;
|
|
|
|
|
|
default:
|
|
|
UPRV_UNREACHABLE;
|
|
|
}
|
|
|
}
|
|
|
|
|
|
|
|
|
void NumberRangeFormatterImpl::formatSingleValue(UFormattedNumberRangeData& data,
|
|
|
MicroProps& micros1, MicroProps& micros2,
|
|
|
UErrorCode& status) const {
|
|
|
if (U_FAILURE(status)) { return; }
|
|
|
if (fSameFormatters) {
|
|
|
int32_t length = NumberFormatterImpl::writeNumber(micros1, data.quantity1, data.getStringRef(), 0, status);
|
|
|
NumberFormatterImpl::writeAffixes(micros1, data.getStringRef(), 0, length, status);
|
|
|
} else {
|
|
|
formatRange(data, micros1, micros2, status);
|
|
|
}
|
|
|
}
|
|
|
|
|
|
|
|
|
void NumberRangeFormatterImpl::formatApproximately (UFormattedNumberRangeData& data,
|
|
|
MicroProps& micros1, MicroProps& micros2,
|
|
|
UErrorCode& status) const {
|
|
|
if (U_FAILURE(status)) { return; }
|
|
|
if (fSameFormatters) {
|
|
|
int32_t length = NumberFormatterImpl::writeNumber(micros1, data.quantity1, data.getStringRef(), 0, status);
|
|
|
// HEURISTIC: Desired modifier order: inner, middle, approximately, outer.
|
|
|
length += micros1.modInner->apply(data.getStringRef(), 0, length, status);
|
|
|
length += micros1.modMiddle->apply(data.getStringRef(), 0, length, status);
|
|
|
length += fApproximatelyModifier.apply(data.getStringRef(), 0, length, status);
|
|
|
micros1.modOuter->apply(data.getStringRef(), 0, length, status);
|
|
|
} else {
|
|
|
formatRange(data, micros1, micros2, status);
|
|
|
}
|
|
|
}
|
|
|
|
|
|
|
|
|
void NumberRangeFormatterImpl::formatRange(UFormattedNumberRangeData& data,
|
|
|
MicroProps& micros1, MicroProps& micros2,
|
|
|
UErrorCode& status) const {
|
|
|
if (U_FAILURE(status)) { return; }
|
|
|
|
|
|
// modInner is always notation (scientific); collapsable in ALL.
|
|
|
// modOuter is always units; collapsable in ALL, AUTO, and UNIT.
|
|
|
// modMiddle could be either; collapsable in ALL and sometimes AUTO and UNIT.
|
|
|
// Never collapse an outer mod but not an inner mod.
|
|
|
bool collapseOuter, collapseMiddle, collapseInner;
|
|
|
switch (fCollapse) {
|
|
|
case UNUM_RANGE_COLLAPSE_ALL:
|
|
|
case UNUM_RANGE_COLLAPSE_AUTO:
|
|
|
case UNUM_RANGE_COLLAPSE_UNIT:
|
|
|
{
|
|
|
// OUTER MODIFIER
|
|
|
collapseOuter = micros1.modOuter->semanticallyEquivalent(*micros2.modOuter);
|
|
|
|
|
|
if (!collapseOuter) {
|
|
|
// Never collapse inner mods if outer mods are not collapsable
|
|
|
collapseMiddle = false;
|
|
|
collapseInner = false;
|
|
|
break;
|
|
|
}
|
|
|
|
|
|
// MIDDLE MODIFIER
|
|
|
collapseMiddle = micros1.modMiddle->semanticallyEquivalent(*micros2.modMiddle);
|
|
|
|
|
|
if (!collapseMiddle) {
|
|
|
// Never collapse inner mods if outer mods are not collapsable
|
|
|
collapseInner = false;
|
|
|
break;
|
|
|
}
|
|
|
|
|
|
// MIDDLE MODIFIER HEURISTICS
|
|
|
// (could disable collapsing of the middle modifier)
|
|
|
// The modifiers are equal by this point, so we can look at just one of them.
|
|
|
const Modifier* mm = micros1.modMiddle;
|
|
|
if (fCollapse == UNUM_RANGE_COLLAPSE_UNIT) {
|
|
|
// Only collapse if the modifier is a unit.
|
|
|
// TODO: Make a better way to check for a unit?
|
|
|
// TODO: Handle case where the modifier has both notation and unit (compact currency)?
|
|
|
if (!mm->containsField({UFIELD_CATEGORY_NUMBER, UNUM_CURRENCY_FIELD})
|
|
|
&& !mm->containsField({UFIELD_CATEGORY_NUMBER, UNUM_PERCENT_FIELD})) {
|
|
|
collapseMiddle = false;
|
|
|
}
|
|
|
} else if (fCollapse == UNUM_RANGE_COLLAPSE_AUTO) {
|
|
|
// Heuristic as of ICU 63: collapse only if the modifier is more than one code point.
|
|
|
if (mm->getCodePointCount() <= 1) {
|
|
|
collapseMiddle = false;
|
|
|
}
|
|
|
}
|
|
|
|
|
|
if (!collapseMiddle || fCollapse != UNUM_RANGE_COLLAPSE_ALL) {
|
|
|
collapseInner = false;
|
|
|
break;
|
|
|
}
|
|
|
|
|
|
// INNER MODIFIER
|
|
|
collapseInner = micros1.modInner->semanticallyEquivalent(*micros2.modInner);
|
|
|
|
|
|
// All done checking for collapsability.
|
|
|
break;
|
|
|
}
|
|
|
|
|
|
default:
|
|
|
collapseOuter = false;
|
|
|
collapseMiddle = false;
|
|
|
collapseInner = false;
|
|
|
break;
|
|
|
}
|
|
|
|
|
|
FormattedStringBuilder& string = data.getStringRef();
|
|
|
int32_t lengthPrefix = 0;
|
|
|
int32_t length1 = 0;
|
|
|
int32_t lengthInfix = 0;
|
|
|
int32_t length2 = 0;
|
|
|
int32_t lengthSuffix = 0;
|
|
|
|
|
|
// Use #define so that these are evaluated at the call site.
|
|
|
#define UPRV_INDEX_0 (lengthPrefix)
|
|
|
#define UPRV_INDEX_1 (lengthPrefix + length1)
|
|
|
#define UPRV_INDEX_2 (lengthPrefix + length1 + lengthInfix)
|
|
|
#define UPRV_INDEX_3 (lengthPrefix + length1 + lengthInfix + length2)
|
|
|
|
|
|
int32_t lengthRange = SimpleModifier::formatTwoArgPattern(
|
|
|
fRangeFormatter,
|
|
|
string,
|
|
|
0,
|
|
|
&lengthPrefix,
|
|
|
&lengthSuffix,
|
|
|
kUndefinedField,
|
|
|
status);
|
|
|
if (U_FAILURE(status)) { return; }
|
|
|
lengthInfix = lengthRange - lengthPrefix - lengthSuffix;
|
|
|
U_ASSERT(lengthInfix > 0);
|
|
|
|
|
|
// SPACING HEURISTIC
|
|
|
// Add spacing unless all modifiers are collapsed.
|
|
|
// TODO: add API to control this?
|
|
|
// TODO: Use a data-driven heuristic like currency spacing?
|
|
|
// TODO: Use Unicode [:whitespace:] instead of PatternProps whitespace? (consider speed implications)
|
|
|
{
|
|
|
bool repeatInner = !collapseInner && micros1.modInner->getCodePointCount() > 0;
|
|
|
bool repeatMiddle = !collapseMiddle && micros1.modMiddle->getCodePointCount() > 0;
|
|
|
bool repeatOuter = !collapseOuter && micros1.modOuter->getCodePointCount() > 0;
|
|
|
if (repeatInner || repeatMiddle || repeatOuter) {
|
|
|
// Add spacing if there is not already spacing
|
|
|
if (!PatternProps::isWhiteSpace(string.charAt(UPRV_INDEX_1))) {
|
|
|
lengthInfix += string.insertCodePoint(UPRV_INDEX_1, u'\u0020', kUndefinedField, status);
|
|
|
}
|
|
|
if (!PatternProps::isWhiteSpace(string.charAt(UPRV_INDEX_2 - 1))) {
|
|
|
lengthInfix += string.insertCodePoint(UPRV_INDEX_2, u'\u0020', kUndefinedField, status);
|
|
|
}
|
|
|
}
|
|
|
}
|
|
|
|
|
|
length1 += NumberFormatterImpl::writeNumber(micros1, data.quantity1, string, UPRV_INDEX_0, status);
|
|
|
length2 += NumberFormatterImpl::writeNumber(micros2, data.quantity2, string, UPRV_INDEX_2, status);
|
|
|
|
|
|
// TODO: Support padding?
|
|
|
|
|
|
if (collapseInner) {
|
|
|
// Note: this is actually a mix of prefix and suffix, but adding to infix length works
|
|
|
const Modifier& mod = resolveModifierPlurals(*micros1.modInner, *micros2.modInner);
|
|
|
lengthInfix += mod.apply(string, UPRV_INDEX_0, UPRV_INDEX_3, status);
|
|
|
} else {
|
|
|
length1 += micros1.modInner->apply(string, UPRV_INDEX_0, UPRV_INDEX_1, status);
|
|
|
length2 += micros2.modInner->apply(string, UPRV_INDEX_2, UPRV_INDEX_3, status);
|
|
|
}
|
|
|
|
|
|
if (collapseMiddle) {
|
|
|
// Note: this is actually a mix of prefix and suffix, but adding to infix length works
|
|
|
const Modifier& mod = resolveModifierPlurals(*micros1.modMiddle, *micros2.modMiddle);
|
|
|
lengthInfix += mod.apply(string, UPRV_INDEX_0, UPRV_INDEX_3, status);
|
|
|
} else {
|
|
|
length1 += micros1.modMiddle->apply(string, UPRV_INDEX_0, UPRV_INDEX_1, status);
|
|
|
length2 += micros2.modMiddle->apply(string, UPRV_INDEX_2, UPRV_INDEX_3, status);
|
|
|
}
|
|
|
|
|
|
if (collapseOuter) {
|
|
|
// Note: this is actually a mix of prefix and suffix, but adding to infix length works
|
|
|
const Modifier& mod = resolveModifierPlurals(*micros1.modOuter, *micros2.modOuter);
|
|
|
lengthInfix += mod.apply(string, UPRV_INDEX_0, UPRV_INDEX_3, status);
|
|
|
} else {
|
|
|
length1 += micros1.modOuter->apply(string, UPRV_INDEX_0, UPRV_INDEX_1, status);
|
|
|
length2 += micros2.modOuter->apply(string, UPRV_INDEX_2, UPRV_INDEX_3, status);
|
|
|
}
|
|
|
}
|
|
|
|
|
|
|
|
|
const Modifier&
|
|
|
NumberRangeFormatterImpl::resolveModifierPlurals(const Modifier& first, const Modifier& second) const {
|
|
|
Modifier::Parameters parameters;
|
|
|
first.getParameters(parameters);
|
|
|
if (parameters.obj == nullptr) {
|
|
|
// No plural form; return a fallback (e.g., the first)
|
|
|
return first;
|
|
|
}
|
|
|
StandardPlural::Form firstPlural = parameters.plural;
|
|
|
|
|
|
second.getParameters(parameters);
|
|
|
if (parameters.obj == nullptr) {
|
|
|
// No plural form; return a fallback (e.g., the first)
|
|
|
return first;
|
|
|
}
|
|
|
StandardPlural::Form secondPlural = parameters.plural;
|
|
|
|
|
|
// Get the required plural form from data
|
|
|
StandardPlural::Form resultPlural = fPluralRanges.resolve(firstPlural, secondPlural);
|
|
|
|
|
|
// Get and return the new Modifier
|
|
|
const Modifier* mod = parameters.obj->getModifier(parameters.signum, resultPlural);
|
|
|
U_ASSERT(mod != nullptr);
|
|
|
return *mod;
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
#endif /* #if !UCONFIG_NO_FORMATTING */
|