blob: 2fc1b8db2232a37804c0db09a49cf0bd2ad9d444 [file] [log] [blame]
// © 2016 and later: Unicode, Inc. and others.
// License & terms of use: http://www.unicode.org/copyright.html
/*
*******************************************************************************
* Copyright (C) 2011-2012, International Business Machines
* Corporation and others. All Rights Reserved.
*******************************************************************************
* file name: messagepattern.cpp
* encoding: UTF-8
* tab size: 8 (not used)
* indentation:4
*
* created on: 2011mar14
* created by: Markus W. Scherer
*/
#include "unicode/utypes.h"
#if !UCONFIG_NO_FORMATTING
#if defined(STARBOARD)
#include "starboard/client_porting/poem/assert_poem.h"
#include "starboard/client_porting/poem/string_poem.h"
#endif // defined(STARBOARD)
#include "unicode/messagepattern.h"
#include "unicode/unistr.h"
#include "unicode/utf16.h"
#include "cmemory.h"
#include "cstring.h"
#include "messageimpl.h"
#include "patternprops.h"
#include "putilimp.h"
#include "uassert.h"
U_NAMESPACE_BEGIN
// Unicode character/code point constants ---------------------------------- ***
static const UChar u_pound=0x23;
static const UChar u_apos=0x27;
static const UChar u_plus=0x2B;
static const UChar u_comma=0x2C;
static const UChar u_minus=0x2D;
static const UChar u_dot=0x2E;
static const UChar u_colon=0x3A;
static const UChar u_lessThan=0x3C;
static const UChar u_equal=0x3D;
static const UChar u_A=0x41;
static const UChar u_C=0x43;
static const UChar u_D=0x44;
static const UChar u_E=0x45;
static const UChar u_H=0x48;
static const UChar u_I=0x49;
static const UChar u_L=0x4C;
static const UChar u_N=0x4E;
static const UChar u_O=0x4F;
static const UChar u_P=0x50;
static const UChar u_R=0x52;
static const UChar u_S=0x53;
static const UChar u_T=0x54;
static const UChar u_U=0x55;
static const UChar u_Z=0x5A;
static const UChar u_a=0x61;
static const UChar u_c=0x63;
static const UChar u_d=0x64;
static const UChar u_e=0x65;
static const UChar u_f=0x66;
static const UChar u_h=0x68;
static const UChar u_i=0x69;
static const UChar u_l=0x6C;
static const UChar u_n=0x6E;
static const UChar u_o=0x6F;
static const UChar u_p=0x70;
static const UChar u_r=0x72;
static const UChar u_s=0x73;
static const UChar u_t=0x74;
static const UChar u_u=0x75;
static const UChar u_z=0x7A;
static const UChar u_leftCurlyBrace=0x7B;
static const UChar u_pipe=0x7C;
static const UChar u_rightCurlyBrace=0x7D;
static const UChar u_lessOrEqual=0x2264; // U+2264 is <=
static const UChar kOffsetColon[]={ // "offset:"
u_o, u_f, u_f, u_s, u_e, u_t, u_colon
};
static const UChar kOther[]={ // "other"
u_o, u_t, u_h, u_e, u_r
};
// MessagePatternList ------------------------------------------------------ ***
template<typename T, int32_t stackCapacity>
class MessagePatternList : public UMemory {
public:
MessagePatternList() {}
void copyFrom(const MessagePatternList<T, stackCapacity> &other,
int32_t length,
UErrorCode &errorCode);
UBool ensureCapacityForOneMore(int32_t oldLength, UErrorCode &errorCode);
UBool equals(const MessagePatternList<T, stackCapacity> &other, int32_t length) const {
for(int32_t i=0; i<length; ++i) {
if(a[i]!=other.a[i]) { return FALSE; }
}
return TRUE;
}
MaybeStackArray<T, stackCapacity> a;
};
template<typename T, int32_t stackCapacity>
void
MessagePatternList<T, stackCapacity>::copyFrom(
const MessagePatternList<T, stackCapacity> &other,
int32_t length,
UErrorCode &errorCode) {
if(U_SUCCESS(errorCode) && length>0) {
if(length>a.getCapacity() && NULL==a.resize(length)) {
errorCode=U_MEMORY_ALLOCATION_ERROR;
return;
}
uprv_memcpy(a.getAlias(), other.a.getAlias(), (size_t)length*sizeof(T));
}
}
template<typename T, int32_t stackCapacity>
UBool
MessagePatternList<T, stackCapacity>::ensureCapacityForOneMore(int32_t oldLength, UErrorCode &errorCode) {
if(U_FAILURE(errorCode)) {
return FALSE;
}
if(a.getCapacity()>oldLength || a.resize(2*oldLength, oldLength)!=NULL) {
return TRUE;
}
errorCode=U_MEMORY_ALLOCATION_ERROR;
return FALSE;
}
// MessagePatternList specializations -------------------------------------- ***
class MessagePatternDoubleList : public MessagePatternList<double, 8> {
};
class MessagePatternPartsList : public MessagePatternList<MessagePattern::Part, 32> {
};
// MessagePattern constructors etc. ---------------------------------------- ***
MessagePattern::MessagePattern(UErrorCode &errorCode)
: aposMode(UCONFIG_MSGPAT_DEFAULT_APOSTROPHE_MODE),
partsList(NULL), parts(NULL), partsLength(0),
numericValuesList(NULL), numericValues(NULL), numericValuesLength(0),
hasArgNames(FALSE), hasArgNumbers(FALSE), needsAutoQuoting(FALSE) {
init(errorCode);
}
MessagePattern::MessagePattern(UMessagePatternApostropheMode mode, UErrorCode &errorCode)
: aposMode(mode),
partsList(NULL), parts(NULL), partsLength(0),
numericValuesList(NULL), numericValues(NULL), numericValuesLength(0),
hasArgNames(FALSE), hasArgNumbers(FALSE), needsAutoQuoting(FALSE) {
init(errorCode);
}
MessagePattern::MessagePattern(const UnicodeString &pattern, UParseError *parseError, UErrorCode &errorCode)
: aposMode(UCONFIG_MSGPAT_DEFAULT_APOSTROPHE_MODE),
partsList(NULL), parts(NULL), partsLength(0),
numericValuesList(NULL), numericValues(NULL), numericValuesLength(0),
hasArgNames(FALSE), hasArgNumbers(FALSE), needsAutoQuoting(FALSE) {
if(init(errorCode)) {
parse(pattern, parseError, errorCode);
}
}
UBool
MessagePattern::init(UErrorCode &errorCode) {
if(U_FAILURE(errorCode)) {
return FALSE;
}
partsList=new MessagePatternPartsList();
if(partsList==NULL) {
errorCode=U_MEMORY_ALLOCATION_ERROR;
return FALSE;
}
parts=partsList->a.getAlias();
return TRUE;
}
MessagePattern::MessagePattern(const MessagePattern &other)
: UObject(other), aposMode(other.aposMode), msg(other.msg),
partsList(NULL), parts(NULL), partsLength(0),
numericValuesList(NULL), numericValues(NULL), numericValuesLength(0),
hasArgNames(other.hasArgNames), hasArgNumbers(other.hasArgNumbers),
needsAutoQuoting(other.needsAutoQuoting) {
UErrorCode errorCode=U_ZERO_ERROR;
if(!copyStorage(other, errorCode)) {
clear();
}
}
MessagePattern &
MessagePattern::operator=(const MessagePattern &other) {
if(this==&other) {
return *this;
}
aposMode=other.aposMode;
msg=other.msg;
hasArgNames=other.hasArgNames;
hasArgNumbers=other.hasArgNumbers;
needsAutoQuoting=other.needsAutoQuoting;
UErrorCode errorCode=U_ZERO_ERROR;
if(!copyStorage(other, errorCode)) {
clear();
}
return *this;
}
UBool
MessagePattern::copyStorage(const MessagePattern &other, UErrorCode &errorCode) {
if(U_FAILURE(errorCode)) {
return FALSE;
}
parts=NULL;
partsLength=0;
numericValues=NULL;
numericValuesLength=0;
if(partsList==NULL) {
partsList=new MessagePatternPartsList();
if(partsList==NULL) {
errorCode=U_MEMORY_ALLOCATION_ERROR;
return FALSE;
}
parts=partsList->a.getAlias();
}
if(other.partsLength>0) {
partsList->copyFrom(*other.partsList, other.partsLength, errorCode);
if(U_FAILURE(errorCode)) {
return FALSE;
}
parts=partsList->a.getAlias();
partsLength=other.partsLength;
}
if(other.numericValuesLength>0) {
if(numericValuesList==NULL) {
numericValuesList=new MessagePatternDoubleList();
if(numericValuesList==NULL) {
errorCode=U_MEMORY_ALLOCATION_ERROR;
return FALSE;
}
numericValues=numericValuesList->a.getAlias();
}
numericValuesList->copyFrom(
*other.numericValuesList, other.numericValuesLength, errorCode);
if(U_FAILURE(errorCode)) {
return FALSE;
}
numericValues=numericValuesList->a.getAlias();
numericValuesLength=other.numericValuesLength;
}
return TRUE;
}
MessagePattern::~MessagePattern() {
delete partsList;
delete numericValuesList;
}
// MessagePattern API ------------------------------------------------------ ***
MessagePattern &
MessagePattern::parse(const UnicodeString &pattern, UParseError *parseError, UErrorCode &errorCode) {
preParse(pattern, parseError, errorCode);
parseMessage(0, 0, 0, UMSGPAT_ARG_TYPE_NONE, parseError, errorCode);
postParse();
return *this;
}
MessagePattern &
MessagePattern::parseChoiceStyle(const UnicodeString &pattern,
UParseError *parseError, UErrorCode &errorCode) {
preParse(pattern, parseError, errorCode);
parseChoiceStyle(0, 0, parseError, errorCode);
postParse();
return *this;
}
MessagePattern &
MessagePattern::parsePluralStyle(const UnicodeString &pattern,
UParseError *parseError, UErrorCode &errorCode) {
preParse(pattern, parseError, errorCode);
parsePluralOrSelectStyle(UMSGPAT_ARG_TYPE_PLURAL, 0, 0, parseError, errorCode);
postParse();
return *this;
}
MessagePattern &
MessagePattern::parseSelectStyle(const UnicodeString &pattern,
UParseError *parseError, UErrorCode &errorCode) {
preParse(pattern, parseError, errorCode);
parsePluralOrSelectStyle(UMSGPAT_ARG_TYPE_SELECT, 0, 0, parseError, errorCode);
postParse();
return *this;
}
void
MessagePattern::clear() {
// Mostly the same as preParse().
msg.remove();
hasArgNames=hasArgNumbers=FALSE;
needsAutoQuoting=FALSE;
partsLength=0;
numericValuesLength=0;
}
UBool
MessagePattern::operator==(const MessagePattern &other) const {
if(this==&other) {
return TRUE;
}
return
aposMode==other.aposMode &&
msg==other.msg &&
// parts.equals(o.parts)
partsLength==other.partsLength &&
(partsLength==0 || partsList->equals(*other.partsList, partsLength));
// No need to compare numericValues if msg and parts are the same.
}
int32_t
MessagePattern::hashCode() const {
int32_t hash=(aposMode*37+msg.hashCode())*37+partsLength;
for(int32_t i=0; i<partsLength; ++i) {
hash=hash*37+parts[i].hashCode();
}
return hash;
}
int32_t
MessagePattern::validateArgumentName(const UnicodeString &name) {
if(!PatternProps::isIdentifier(name.getBuffer(), name.length())) {
return UMSGPAT_ARG_NAME_NOT_VALID;
}
return parseArgNumber(name, 0, name.length());
}
UnicodeString
MessagePattern::autoQuoteApostropheDeep() const {
if(!needsAutoQuoting) {
return msg;
}
UnicodeString modified(msg);
// Iterate backward so that the insertion indexes do not change.
int32_t count=countParts();
for(int32_t i=count; i>0;) {
const Part &part=getPart(--i);
if(part.getType()==UMSGPAT_PART_TYPE_INSERT_CHAR) {
modified.insert(part.index, (UChar)part.value);
}
}
return modified;
}
double
MessagePattern::getNumericValue(const Part &part) const {
UMessagePatternPartType type=part.type;
if(type==UMSGPAT_PART_TYPE_ARG_INT) {
return part.value;
} else if(type==UMSGPAT_PART_TYPE_ARG_DOUBLE) {
return numericValues[part.value];
} else {
return UMSGPAT_NO_NUMERIC_VALUE;
}
}
/**
* Returns the "offset:" value of a PluralFormat argument, or 0 if none is specified.
* @param pluralStart the index of the first PluralFormat argument style part. (0..countParts()-1)
* @return the "offset:" value.
* @draft ICU 4.8
*/
double
MessagePattern::getPluralOffset(int32_t pluralStart) const {
const Part &part=getPart(pluralStart);
if(Part::hasNumericValue(part.type)) {
return getNumericValue(part);
} else {
return 0;
}
}
// MessagePattern::Part ---------------------------------------------------- ***
UBool
MessagePattern::Part::operator==(const Part &other) const {
if(this==&other) {
return TRUE;
}
return
type==other.type &&
index==other.index &&
length==other.length &&
value==other.value &&
limitPartIndex==other.limitPartIndex;
}
// MessagePattern parser --------------------------------------------------- ***
void
MessagePattern::preParse(const UnicodeString &pattern, UParseError *parseError, UErrorCode &errorCode) {
if(U_FAILURE(errorCode)) {
return;
}
if(parseError!=NULL) {
parseError->line=0;
parseError->offset=0;
parseError->preContext[0]=0;
parseError->postContext[0]=0;
}
msg=pattern;
hasArgNames=hasArgNumbers=FALSE;
needsAutoQuoting=FALSE;
partsLength=0;
numericValuesLength=0;
}
void
MessagePattern::postParse() {
if(partsList!=NULL) {
parts=partsList->a.getAlias();
}
if(numericValuesList!=NULL) {
numericValues=numericValuesList->a.getAlias();
}
}
int32_t
MessagePattern::parseMessage(int32_t index, int32_t msgStartLength,
int32_t nestingLevel, UMessagePatternArgType parentType,
UParseError *parseError, UErrorCode &errorCode) {
if(U_FAILURE(errorCode)) {
return 0;
}
if(nestingLevel>Part::MAX_VALUE) {
errorCode=U_INDEX_OUTOFBOUNDS_ERROR;
return 0;
}
int32_t msgStart=partsLength;
addPart(UMSGPAT_PART_TYPE_MSG_START, index, msgStartLength, nestingLevel, errorCode);
index+=msgStartLength;
for(;;) { // while(index<msg.length()) with U_FAILURE(errorCode) check
if(U_FAILURE(errorCode)) {
return 0;
}
if(index>=msg.length()) {
break;
}
UChar c=msg.charAt(index++);
if(c==u_apos) {
if(index==msg.length()) {
// The apostrophe is the last character in the pattern.
// Add a Part for auto-quoting.
addPart(UMSGPAT_PART_TYPE_INSERT_CHAR, index, 0,
u_apos, errorCode); // value=char to be inserted
needsAutoQuoting=TRUE;
} else {
c=msg.charAt(index);
if(c==u_apos) {
// double apostrophe, skip the second one
addPart(UMSGPAT_PART_TYPE_SKIP_SYNTAX, index++, 1, 0, errorCode);
} else if(
aposMode==UMSGPAT_APOS_DOUBLE_REQUIRED ||
c==u_leftCurlyBrace || c==u_rightCurlyBrace ||
(parentType==UMSGPAT_ARG_TYPE_CHOICE && c==u_pipe) ||
(UMSGPAT_ARG_TYPE_HAS_PLURAL_STYLE(parentType) && c==u_pound)
) {
// skip the quote-starting apostrophe
addPart(UMSGPAT_PART_TYPE_SKIP_SYNTAX, index-1, 1, 0, errorCode);
// find the end of the quoted literal text
for(;;) {
index=msg.indexOf(u_apos, index+1);
if(index>=0) {
if(/*(index+1)<msg.length() &&*/ msg.charAt(index+1)==u_apos) {
// double apostrophe inside quoted literal text
// still encodes a single apostrophe, skip the second one
addPart(UMSGPAT_PART_TYPE_SKIP_SYNTAX, ++index, 1, 0, errorCode);
} else {
// skip the quote-ending apostrophe
addPart(UMSGPAT_PART_TYPE_SKIP_SYNTAX, index++, 1, 0, errorCode);
break;
}
} else {
// The quoted text reaches to the end of the of the message.
index=msg.length();
// Add a Part for auto-quoting.
addPart(UMSGPAT_PART_TYPE_INSERT_CHAR, index, 0,
u_apos, errorCode); // value=char to be inserted
needsAutoQuoting=TRUE;
break;
}
}
} else {
// Interpret the apostrophe as literal text.
// Add a Part for auto-quoting.
addPart(UMSGPAT_PART_TYPE_INSERT_CHAR, index, 0,
u_apos, errorCode); // value=char to be inserted
needsAutoQuoting=TRUE;
}
}
} else if(UMSGPAT_ARG_TYPE_HAS_PLURAL_STYLE(parentType) && c==u_pound) {
// The unquoted # in a plural message fragment will be replaced
// with the (number-offset).
addPart(UMSGPAT_PART_TYPE_REPLACE_NUMBER, index-1, 1, 0, errorCode);
} else if(c==u_leftCurlyBrace) {
index=parseArg(index-1, 1, nestingLevel, parseError, errorCode);
} else if((nestingLevel>0 && c==u_rightCurlyBrace) ||
(parentType==UMSGPAT_ARG_TYPE_CHOICE && c==u_pipe)) {
// Finish the message before the terminator.
// In a choice style, report the "}" substring only for the following ARG_LIMIT,
// not for this MSG_LIMIT.
int32_t limitLength=(parentType==UMSGPAT_ARG_TYPE_CHOICE && c==u_rightCurlyBrace) ? 0 : 1;
addLimitPart(msgStart, UMSGPAT_PART_TYPE_MSG_LIMIT, index-1, limitLength,
nestingLevel, errorCode);
if(parentType==UMSGPAT_ARG_TYPE_CHOICE) {
// Let the choice style parser see the '}' or '|'.
return index-1;
} else {
// continue parsing after the '}'
return index;
}
} // else: c is part of literal text
}
if(nestingLevel>0 && !inTopLevelChoiceMessage(nestingLevel, parentType)) {
setParseError(parseError, 0); // Unmatched '{' braces in message.
errorCode=U_UNMATCHED_BRACES;
return 0;
}
addLimitPart(msgStart, UMSGPAT_PART_TYPE_MSG_LIMIT, index, 0, nestingLevel, errorCode);
return index;
}
int32_t
MessagePattern::parseArg(int32_t index, int32_t argStartLength, int32_t nestingLevel,
UParseError *parseError, UErrorCode &errorCode) {
int32_t argStart=partsLength;
UMessagePatternArgType argType=UMSGPAT_ARG_TYPE_NONE;
addPart(UMSGPAT_PART_TYPE_ARG_START, index, argStartLength, argType, errorCode);
if(U_FAILURE(errorCode)) {
return 0;
}
int32_t nameIndex=index=skipWhiteSpace(index+argStartLength);
if(index==msg.length()) {
setParseError(parseError, 0); // Unmatched '{' braces in message.
errorCode=U_UNMATCHED_BRACES;
return 0;
}
// parse argument name or number
index=skipIdentifier(index);
int32_t number=parseArgNumber(nameIndex, index);
if(number>=0) {
int32_t length=index-nameIndex;
if(length>Part::MAX_LENGTH || number>Part::MAX_VALUE) {
setParseError(parseError, nameIndex); // Argument number too large.
errorCode=U_INDEX_OUTOFBOUNDS_ERROR;
return 0;
}
hasArgNumbers=TRUE;
addPart(UMSGPAT_PART_TYPE_ARG_NUMBER, nameIndex, length, number, errorCode);
} else if(number==UMSGPAT_ARG_NAME_NOT_NUMBER) {
int32_t length=index-nameIndex;
if(length>Part::MAX_LENGTH) {
setParseError(parseError, nameIndex); // Argument name too long.
errorCode=U_INDEX_OUTOFBOUNDS_ERROR;
return 0;
}
hasArgNames=TRUE;
addPart(UMSGPAT_PART_TYPE_ARG_NAME, nameIndex, length, 0, errorCode);
} else { // number<-1 (ARG_NAME_NOT_VALID)
setParseError(parseError, nameIndex); // Bad argument syntax.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
index=skipWhiteSpace(index);
if(index==msg.length()) {
setParseError(parseError, 0); // Unmatched '{' braces in message.
errorCode=U_UNMATCHED_BRACES;
return 0;
}
UChar c=msg.charAt(index);
if(c==u_rightCurlyBrace) {
// all done
} else if(c!=u_comma) {
setParseError(parseError, nameIndex); // Bad argument syntax.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
} else /* ',' */ {
// parse argument type: case-sensitive a-zA-Z
int32_t typeIndex=index=skipWhiteSpace(index+1);
while(index<msg.length() && isArgTypeChar(msg.charAt(index))) {
++index;
}
int32_t length=index-typeIndex;
index=skipWhiteSpace(index);
if(index==msg.length()) {
setParseError(parseError, 0); // Unmatched '{' braces in message.
errorCode=U_UNMATCHED_BRACES;
return 0;
}
if(length==0 || ((c=msg.charAt(index))!=u_comma && c!=u_rightCurlyBrace)) {
setParseError(parseError, nameIndex); // Bad argument syntax.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
if(length>Part::MAX_LENGTH) {
setParseError(parseError, nameIndex); // Argument type name too long.
errorCode=U_INDEX_OUTOFBOUNDS_ERROR;
return 0;
}
argType=UMSGPAT_ARG_TYPE_SIMPLE;
if(length==6) {
// case-insensitive comparisons for complex-type names
if(isChoice(typeIndex)) {
argType=UMSGPAT_ARG_TYPE_CHOICE;
} else if(isPlural(typeIndex)) {
argType=UMSGPAT_ARG_TYPE_PLURAL;
} else if(isSelect(typeIndex)) {
argType=UMSGPAT_ARG_TYPE_SELECT;
}
} else if(length==13) {
if(isSelect(typeIndex) && isOrdinal(typeIndex+6)) {
argType=UMSGPAT_ARG_TYPE_SELECTORDINAL;
}
}
// change the ARG_START type from NONE to argType
partsList->a[argStart].value=(int16_t)argType;
if(argType==UMSGPAT_ARG_TYPE_SIMPLE) {
addPart(UMSGPAT_PART_TYPE_ARG_TYPE, typeIndex, length, 0, errorCode);
}
// look for an argument style (pattern)
if(c==u_rightCurlyBrace) {
if(argType!=UMSGPAT_ARG_TYPE_SIMPLE) {
setParseError(parseError, nameIndex); // No style field for complex argument.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
} else /* ',' */ {
++index;
if(argType==UMSGPAT_ARG_TYPE_SIMPLE) {
index=parseSimpleStyle(index, parseError, errorCode);
} else if(argType==UMSGPAT_ARG_TYPE_CHOICE) {
index=parseChoiceStyle(index, nestingLevel, parseError, errorCode);
} else {
index=parsePluralOrSelectStyle(argType, index, nestingLevel, parseError, errorCode);
}
}
}
// Argument parsing stopped on the '}'.
addLimitPart(argStart, UMSGPAT_PART_TYPE_ARG_LIMIT, index, 1, argType, errorCode);
return index+1;
}
int32_t
MessagePattern::parseSimpleStyle(int32_t index, UParseError *parseError, UErrorCode &errorCode) {
if(U_FAILURE(errorCode)) {
return 0;
}
int32_t start=index;
int32_t nestedBraces=0;
while(index<msg.length()) {
UChar c=msg.charAt(index++);
if(c==u_apos) {
// Treat apostrophe as quoting but include it in the style part.
// Find the end of the quoted literal text.
index=msg.indexOf(u_apos, index);
if(index<0) {
// Quoted literal argument style text reaches to the end of the message.
setParseError(parseError, start);
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
// skip the quote-ending apostrophe
++index;
} else if(c==u_leftCurlyBrace) {
++nestedBraces;
} else if(c==u_rightCurlyBrace) {
if(nestedBraces>0) {
--nestedBraces;
} else {
int32_t length=--index-start;
if(length>Part::MAX_LENGTH) {
setParseError(parseError, start); // Argument style text too long.
errorCode=U_INDEX_OUTOFBOUNDS_ERROR;
return 0;
}
addPart(UMSGPAT_PART_TYPE_ARG_STYLE, start, length, 0, errorCode);
return index;
}
} // c is part of literal text
}
setParseError(parseError, 0); // Unmatched '{' braces in message.
errorCode=U_UNMATCHED_BRACES;
return 0;
}
int32_t
MessagePattern::parseChoiceStyle(int32_t index, int32_t nestingLevel,
UParseError *parseError, UErrorCode &errorCode) {
if(U_FAILURE(errorCode)) {
return 0;
}
int32_t start=index;
index=skipWhiteSpace(index);
if(index==msg.length() || msg.charAt(index)==u_rightCurlyBrace) {
setParseError(parseError, 0); // Missing choice argument pattern.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
for(;;) {
// The choice argument style contains |-separated (number, separator, message) triples.
// Parse the number.
int32_t numberIndex=index;
index=skipDouble(index);
int32_t length=index-numberIndex;
if(length==0) {
setParseError(parseError, start); // Bad choice pattern syntax.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
if(length>Part::MAX_LENGTH) {
setParseError(parseError, numberIndex); // Choice number too long.
errorCode=U_INDEX_OUTOFBOUNDS_ERROR;
return 0;
}
parseDouble(numberIndex, index, TRUE, parseError, errorCode); // adds ARG_INT or ARG_DOUBLE
if(U_FAILURE(errorCode)) {
return 0;
}
// Parse the separator.
index=skipWhiteSpace(index);
if(index==msg.length()) {
setParseError(parseError, start); // Bad choice pattern syntax.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
UChar c=msg.charAt(index);
if(!(c==u_pound || c==u_lessThan || c==u_lessOrEqual)) { // U+2264 is <=
setParseError(parseError, start); // Expected choice separator (#<\u2264) instead of c.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
addPart(UMSGPAT_PART_TYPE_ARG_SELECTOR, index, 1, 0, errorCode);
// Parse the message fragment.
index=parseMessage(++index, 0, nestingLevel+1, UMSGPAT_ARG_TYPE_CHOICE, parseError, errorCode);
if(U_FAILURE(errorCode)) {
return 0;
}
// parseMessage(..., CHOICE) returns the index of the terminator, or msg.length().
if(index==msg.length()) {
return index;
}
if(msg.charAt(index)==u_rightCurlyBrace) {
if(!inMessageFormatPattern(nestingLevel)) {
setParseError(parseError, start); // Bad choice pattern syntax.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
return index;
} // else the terminator is '|'
index=skipWhiteSpace(index+1);
}
}
int32_t
MessagePattern::parsePluralOrSelectStyle(UMessagePatternArgType argType,
int32_t index, int32_t nestingLevel,
UParseError *parseError, UErrorCode &errorCode) {
if(U_FAILURE(errorCode)) {
return 0;
}
int32_t start=index;
UBool isEmpty=TRUE;
UBool hasOther=FALSE;
for(;;) {
// First, collect the selector looking for a small set of terminators.
// It would be a little faster to consider the syntax of each possible
// token right here, but that makes the code too complicated.
index=skipWhiteSpace(index);
UBool eos=index==msg.length();
if(eos || msg.charAt(index)==u_rightCurlyBrace) {
if(eos==inMessageFormatPattern(nestingLevel)) {
setParseError(parseError, start); // Bad plural/select pattern syntax.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
if(!hasOther) {
setParseError(parseError, 0); // Missing 'other' keyword in plural/select pattern.
errorCode=U_DEFAULT_KEYWORD_MISSING;
return 0;
}
return index;
}
int32_t selectorIndex=index;
if(UMSGPAT_ARG_TYPE_HAS_PLURAL_STYLE(argType) && msg.charAt(selectorIndex)==u_equal) {
// explicit-value plural selector: =double
index=skipDouble(index+1);
int32_t length=index-selectorIndex;
if(length==1) {
setParseError(parseError, start); // Bad plural/select pattern syntax.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
if(length>Part::MAX_LENGTH) {
setParseError(parseError, selectorIndex); // Argument selector too long.
errorCode=U_INDEX_OUTOFBOUNDS_ERROR;
return 0;
}
addPart(UMSGPAT_PART_TYPE_ARG_SELECTOR, selectorIndex, length, 0, errorCode);
parseDouble(selectorIndex+1, index, FALSE,
parseError, errorCode); // adds ARG_INT or ARG_DOUBLE
} else {
index=skipIdentifier(index);
int32_t length=index-selectorIndex;
if(length==0) {
setParseError(parseError, start); // Bad plural/select pattern syntax.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
// Note: The ':' in "offset:" is just beyond the skipIdentifier() range.
if( UMSGPAT_ARG_TYPE_HAS_PLURAL_STYLE(argType) && length==6 && index<msg.length() &&
0==msg.compare(selectorIndex, 7, kOffsetColon, 0, 7)
) {
// plural offset, not a selector
if(!isEmpty) {
// Plural argument 'offset:' (if present) must precede key-message pairs.
setParseError(parseError, start);
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
// allow whitespace between offset: and its value
int32_t valueIndex=skipWhiteSpace(index+1); // The ':' is at index.
index=skipDouble(valueIndex);
if(index==valueIndex) {
setParseError(parseError, start); // Missing value for plural 'offset:'.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
if((index-valueIndex)>Part::MAX_LENGTH) {
setParseError(parseError, valueIndex); // Plural offset value too long.
errorCode=U_INDEX_OUTOFBOUNDS_ERROR;
return 0;
}
parseDouble(valueIndex, index, FALSE,
parseError, errorCode); // adds ARG_INT or ARG_DOUBLE
if(U_FAILURE(errorCode)) {
return 0;
}
isEmpty=FALSE;
continue; // no message fragment after the offset
} else {
// normal selector word
if(length>Part::MAX_LENGTH) {
setParseError(parseError, selectorIndex); // Argument selector too long.
errorCode=U_INDEX_OUTOFBOUNDS_ERROR;
return 0;
}
addPart(UMSGPAT_PART_TYPE_ARG_SELECTOR, selectorIndex, length, 0, errorCode);
if(0==msg.compare(selectorIndex, length, kOther, 0, 5)) {
hasOther=TRUE;
}
}
}
if(U_FAILURE(errorCode)) {
return 0;
}
// parse the message fragment following the selector
index=skipWhiteSpace(index);
if(index==msg.length() || msg.charAt(index)!=u_leftCurlyBrace) {
setParseError(parseError, selectorIndex); // No message fragment after plural/select selector.
errorCode=U_PATTERN_SYNTAX_ERROR;
return 0;
}
index=parseMessage(index, 1, nestingLevel+1, argType, parseError, errorCode);
if(U_FAILURE(errorCode)) {
return 0;
}
isEmpty=FALSE;
}
}
int32_t
MessagePattern::parseArgNumber(const UnicodeString &s, int32_t start, int32_t limit) {
// If the identifier contains only ASCII digits, then it is an argument _number_
// and must not have leading zeros (except "0" itself).
// Otherwise it is an argument _name_.
if(start>=limit) {
return UMSGPAT_ARG_NAME_NOT_VALID;
}
int32_t number;
// Defer numeric errors until we know there are only digits.
UBool badNumber;
UChar c=s.charAt(start++);
if(c==0x30) {
if(start==limit) {
return 0;
} else {
number=0;
badNumber=TRUE; // leading zero
}
} else if(0x31<=c && c<=0x39) {
number=c-0x30;
badNumber=FALSE;
} else {
return UMSGPAT_ARG_NAME_NOT_NUMBER;
}
while(start<limit) {
c=s.charAt(start++);
if(0x30<=c && c<=0x39) {
if(number>=INT32_MAX/10) {
badNumber=TRUE; // overflow
}
number=number*10+(c-0x30);
} else {
return UMSGPAT_ARG_NAME_NOT_NUMBER;
}
}
// There are only ASCII digits.
if(badNumber) {
return UMSGPAT_ARG_NAME_NOT_VALID;
} else {
return number;
}
}
void
MessagePattern::parseDouble(int32_t start, int32_t limit, UBool allowInfinity,
UParseError *parseError, UErrorCode &errorCode) {
if(U_FAILURE(errorCode)) {
return;
}
U_ASSERT(start<limit);
// fake loop for easy exit and single throw statement
for(;;) { /*loop doesn't iterate*/
// fast path for small integers and infinity
int32_t value=0;
int32_t isNegative=0; // not boolean so that we can easily add it to value
int32_t index=start;
UChar c=msg.charAt(index++);
if(c==u_minus) {
isNegative=1;
if(index==limit) {
break; // no number
}
c=msg.charAt(index++);
} else if(c==u_plus) {
if(index==limit) {
break; // no number
}
c=msg.charAt(index++);
}
if(c==0x221e) { // infinity
if(allowInfinity && index==limit) {
double infinity=uprv_getInfinity();
addArgDoublePart(
isNegative!=0 ? -infinity : infinity,
start, limit-start, errorCode);
return;
} else {
break;
}
}
// try to parse the number as a small integer but fall back to a double
while('0'<=c && c<='9') {
value=value*10+(c-'0');
if(value>(Part::MAX_VALUE+isNegative)) {
break; // not a small-enough integer
}
if(index==limit) {
addPart(UMSGPAT_PART_TYPE_ARG_INT, start, limit-start,
isNegative!=0 ? -value : value, errorCode);
return;
}
c=msg.charAt(index++);
}
// Let Double.parseDouble() throw a NumberFormatException.
char numberChars[128];
int32_t capacity=(int32_t)sizeof(numberChars);
int32_t length=limit-start;
if(length>=capacity) {
break; // number too long
}
msg.extract(start, length, numberChars, capacity, US_INV);
if((int32_t)uprv_strlen(numberChars)<length) {
break; // contains non-invariant character that was turned into NUL
}
char *end;
double numericValue=uprv_strtod(numberChars, &end);
if(end!=(numberChars+length)) {
break; // parsing error
}
addArgDoublePart(numericValue, start, length, errorCode);
return;
}
setParseError(parseError, start /*, limit*/); // Bad syntax for numeric value.
errorCode=U_PATTERN_SYNTAX_ERROR;
return;
}
int32_t
MessagePattern::skipWhiteSpace(int32_t index) {
const UChar *s=msg.getBuffer();
int32_t msgLength=msg.length();
const UChar *t=PatternProps::skipWhiteSpace(s+index, msgLength-index);
return (int32_t)(t-s);
}
int32_t
MessagePattern::skipIdentifier(int32_t index) {
const UChar *s=msg.getBuffer();
int32_t msgLength=msg.length();
const UChar *t=PatternProps::skipIdentifier(s+index, msgLength-index);
return (int32_t)(t-s);
}
int32_t
MessagePattern::skipDouble(int32_t index) {
int32_t msgLength=msg.length();
while(index<msgLength) {
UChar c=msg.charAt(index);
// U+221E: Allow the infinity symbol, for ChoiceFormat patterns.
if((c<0x30 && c!=u_plus && c!=u_minus && c!=u_dot) || (c>0x39 && c!=u_e && c!=u_E && c!=0x221e)) {
break;
}
++index;
}
return index;
}
UBool
MessagePattern::isArgTypeChar(UChar32 c) {
return (u_a<=c && c<=u_z) || (u_A<=c && c<=u_Z);
}
UBool
MessagePattern::isChoice(int32_t index) {
UChar c;
return
((c=msg.charAt(index++))==u_c || c==u_C) &&
((c=msg.charAt(index++))==u_h || c==u_H) &&
((c=msg.charAt(index++))==u_o || c==u_O) &&
((c=msg.charAt(index++))==u_i || c==u_I) &&
((c=msg.charAt(index++))==u_c || c==u_C) &&
((c=msg.charAt(index))==u_e || c==u_E);
}
UBool
MessagePattern::isPlural(int32_t index) {
UChar c;
return
((c=msg.charAt(index++))==u_p || c==u_P) &&
((c=msg.charAt(index++))==u_l || c==u_L) &&
((c=msg.charAt(index++))==u_u || c==u_U) &&
((c=msg.charAt(index++))==u_r || c==u_R) &&
((c=msg.charAt(index++))==u_a || c==u_A) &&
((c=msg.charAt(index))==u_l || c==u_L);
}
UBool
MessagePattern::isSelect(int32_t index) {
UChar c;
return
((c=msg.charAt(index++))==u_s || c==u_S) &&
((c=msg.charAt(index++))==u_e || c==u_E) &&
((c=msg.charAt(index++))==u_l || c==u_L) &&
((c=msg.charAt(index++))==u_e || c==u_E) &&
((c=msg.charAt(index++))==u_c || c==u_C) &&
((c=msg.charAt(index))==u_t || c==u_T);
}
UBool
MessagePattern::isOrdinal(int32_t index) {
UChar c;
return
((c=msg.charAt(index++))==u_o || c==u_O) &&
((c=msg.charAt(index++))==u_r || c==u_R) &&
((c=msg.charAt(index++))==u_d || c==u_D) &&
((c=msg.charAt(index++))==u_i || c==u_I) &&
((c=msg.charAt(index++))==u_n || c==u_N) &&
((c=msg.charAt(index++))==u_a || c==u_A) &&
((c=msg.charAt(index))==u_l || c==u_L);
}
UBool
MessagePattern::inMessageFormatPattern(int32_t nestingLevel) {
return nestingLevel>0 || partsList->a[0].type==UMSGPAT_PART_TYPE_MSG_START;
}
UBool
MessagePattern::inTopLevelChoiceMessage(int32_t nestingLevel, UMessagePatternArgType parentType) {
return
nestingLevel==1 &&
parentType==UMSGPAT_ARG_TYPE_CHOICE &&
partsList->a[0].type!=UMSGPAT_PART_TYPE_MSG_START;
}
void
MessagePattern::addPart(UMessagePatternPartType type, int32_t index, int32_t length,
int32_t value, UErrorCode &errorCode) {
if(partsList->ensureCapacityForOneMore(partsLength, errorCode)) {
Part &part=partsList->a[partsLength++];
part.type=type;
part.index=index;
part.length=(uint16_t)length;
part.value=(int16_t)value;
part.limitPartIndex=0;
}
}
void
MessagePattern::addLimitPart(int32_t start,
UMessagePatternPartType type, int32_t index, int32_t length,
int32_t value, UErrorCode &errorCode) {
partsList->a[start].limitPartIndex=partsLength;
addPart(type, index, length, value, errorCode);
}
void
MessagePattern::addArgDoublePart(double numericValue, int32_t start, int32_t length,
UErrorCode &errorCode) {
if(U_FAILURE(errorCode)) {
return;
}
int32_t numericIndex=numericValuesLength;
if(numericValuesList==NULL) {
numericValuesList=new MessagePatternDoubleList();
if(numericValuesList==NULL) {
errorCode=U_MEMORY_ALLOCATION_ERROR;
return;
}
} else if(!numericValuesList->ensureCapacityForOneMore(numericValuesLength, errorCode)) {
return;
} else {
if(numericIndex>Part::MAX_VALUE) {
errorCode=U_INDEX_OUTOFBOUNDS_ERROR;
return;
}
}
numericValuesList->a[numericValuesLength++]=numericValue;
addPart(UMSGPAT_PART_TYPE_ARG_DOUBLE, start, length, numericIndex, errorCode);
}
void
MessagePattern::setParseError(UParseError *parseError, int32_t index) {
if(parseError==NULL) {
return;
}
parseError->offset=index;
// Set preContext to some of msg before index.
// Avoid splitting a surrogate pair.
int32_t length=index;
if(length>=U_PARSE_CONTEXT_LEN) {
length=U_PARSE_CONTEXT_LEN-1;
if(length>0 && U16_IS_TRAIL(msg[index-length])) {
--length;
}
}
msg.extract(index-length, length, parseError->preContext);
parseError->preContext[length]=0;
// Set postContext to some of msg starting at index.
length=msg.length()-index;
if(length>=U_PARSE_CONTEXT_LEN) {
length=U_PARSE_CONTEXT_LEN-1;
if(length>0 && U16_IS_LEAD(msg[index+length-1])) {
--length;
}
}
msg.extract(index, length, parseError->postContext);
parseError->postContext[length]=0;
}
// MessageImpl ------------------------------------------------------------- ***
void
MessageImpl::appendReducedApostrophes(const UnicodeString &s, int32_t start, int32_t limit,
UnicodeString &sb) {
int32_t doubleApos=-1;
for(;;) {
int32_t i=s.indexOf(u_apos, start);
if(i<0 || i>=limit) {
sb.append(s, start, limit-start);
break;
}
if(i==doubleApos) {
// Double apostrophe at start-1 and start==i, append one.
sb.append(u_apos);
++start;
doubleApos=-1;
} else {
// Append text between apostrophes and skip this one.
sb.append(s, start, i-start);
doubleApos=start=i+1;
}
}
}
// Ported from second half of ICU4J SelectFormat.format(String).
UnicodeString &
MessageImpl::appendSubMessageWithoutSkipSyntax(const MessagePattern &msgPattern,
int32_t msgStart,
UnicodeString &result) {
const UnicodeString &msgString=msgPattern.getPatternString();
int32_t prevIndex=msgPattern.getPart(msgStart).getLimit();
for(int32_t i=msgStart;;) {
const MessagePattern::Part &part=msgPattern.getPart(++i);
UMessagePatternPartType type=part.getType();
int32_t index=part.getIndex();
if(type==UMSGPAT_PART_TYPE_MSG_LIMIT) {
return result.append(msgString, prevIndex, index-prevIndex);
} else if(type==UMSGPAT_PART_TYPE_SKIP_SYNTAX) {
result.append(msgString, prevIndex, index-prevIndex);
prevIndex=part.getLimit();
} else if(type==UMSGPAT_PART_TYPE_ARG_START) {
result.append(msgString, prevIndex, index-prevIndex);
prevIndex=index;
i=msgPattern.getLimitPartIndex(i);
index=msgPattern.getPart(i).getLimit();
appendReducedApostrophes(msgString, prevIndex, index, result);
prevIndex=index;
}
}
}
U_NAMESPACE_END
#endif // !UCONFIG_NO_FORMATTING