C++程序
|
3836行
|
130.37 KB
/*
********************************************************************************
* Copyright (C) 1999-2011 International Business Machines Corporation and
* others. All Rights Reserved.
********************************************************************************
* Date Name Description
* 10/20/99 alan Creation.
* 03/22/2000 Madhu Added additional tests
********************************************************************************
*/
#include <stdio.h>
#include <string.h>
#include "unicode/utypes.h"
#include "usettest.h"
#include "unicode/ucnv.h"
#include "unicode/uniset.h"
#include "unicode/uchar.h"
#include "unicode/usetiter.h"
#include "unicode/ustring.h"
#include "unicode/parsepos.h"
#include "unicode/symtable.h"
#include "unicode/uversion.h"
#include "hash.h"
#define LENGTHOF(array) (int32_t)(sizeof(array)/sizeof((array)[0]))
#define TEST_ASSERT_SUCCESS(status) {if (U_FAILURE(status)) { \
dataerrln("fail in file \"%s\", line %d: \"%s\"", __FILE__, __LINE__, \
u_errorName(status));}}
#define TEST_ASSERT(expr) {if (!(expr)) { \
dataerrln("fail in file \"%s\", line %d", __FILE__, __LINE__); }}
UnicodeString operator+(const UnicodeString& left, const UnicodeSet& set) {
UnicodeString pat;
set.toPattern(pat);
return left + UnicodeSetTest::escape(pat);
}
#define CASE(id,test) case id: \
name = #test; \
if (exec) { \
logln(#test "---"); \
logln(); \
test(); \
} \
break
UnicodeSetTest::UnicodeSetTest() : utf8Cnv(NULL) {
}
UConverter *UnicodeSetTest::openUTF8Converter() {
if(utf8Cnv==NULL) {
UErrorCode errorCode=U_ZERO_ERROR;
utf8Cnv=ucnv_open("UTF-8", &errorCode);
}
return utf8Cnv;
}
UnicodeSetTest::~UnicodeSetTest() {
ucnv_close(utf8Cnv);
}
void
UnicodeSetTest::runIndexedTest(int32_t index, UBool exec,
const char* &name, char* /*par*/) {
// if (exec) logln((UnicodeString)"TestSuite UnicodeSetTest");
switch (index) {
CASE(0,TestPatterns);
CASE(1,TestAddRemove);
CASE(2,TestCategories);
CASE(3,TestCloneEqualHash);
CASE(4,TestMinimalRep);
CASE(5,TestAPI);
CASE(6,TestScriptSet);
CASE(7,TestPropertySet);
CASE(8,TestClone);
CASE(9,TestExhaustive);
CASE(10,TestToPattern);
CASE(11,TestIndexOf);
CASE(12,TestStrings);
CASE(13,Testj2268);
CASE(14,TestCloseOver);
CASE(15,TestEscapePattern);
CASE(16,TestInvalidCodePoint);
CASE(17,TestSymbolTable);
CASE(18,TestSurrogate);
CASE(19,TestPosixClasses);
CASE(20,TestIteration);
CASE(21,TestFreezable);
CASE(22,TestSpan);
CASE(23,TestStringSpan);
default: name = ""; break;
}
}
static const char NOT[] = "%%%%";
/**
* UVector was improperly copying contents
* This code will crash this is still true
*/
void UnicodeSetTest::Testj2268() {
UnicodeSet t;
t.add(UnicodeString("abc"));
UnicodeSet test(t);
UnicodeString ustrPat;
test.toPattern(ustrPat, TRUE);
}
/**
* Test toPattern().
*/
void UnicodeSetTest::TestToPattern() {
UErrorCode ec = U_ZERO_ERROR;
// Test that toPattern() round trips with syntax characters and
// whitespace.
{
static const char* OTHER_TOPATTERN_TESTS[] = {
"[[:latin:]&[:greek:]]",
"[[:latin:]-[:greek:]]",
"[:nonspacing mark:]",
NULL
};
for (int32_t j=0; OTHER_TOPATTERN_TESTS[j]!=NULL; ++j) {
ec = U_ZERO_ERROR;
UnicodeSet s(OTHER_TOPATTERN_TESTS[j], ec);
if (U_FAILURE(ec)) {
dataerrln((UnicodeString)"FAIL: bad pattern " + OTHER_TOPATTERN_TESTS[j] + " - " + UnicodeString(u_errorName(ec)));
continue;
}
checkPat(OTHER_TOPATTERN_TESTS[j], s);
}
for (UChar32 i = 0; i <= 0x10FFFF; ++i) {
if ((i <= 0xFF && !u_isalpha(i)) || u_isspace(i)) {
// check various combinations to make sure they all work.
if (i != 0 && !toPatternAux(i, i)){
continue;
}
if (!toPatternAux(0, i)){
continue;
}
if (!toPatternAux(i, 0xFFFF)){
continue;
}
}
}
}
// Test pattern behavior of multicharacter strings.
{
ec = U_ZERO_ERROR;
UnicodeSet* s = new UnicodeSet("[a-z {aa} {ab}]", ec);
// This loop isn't a loop. It's here to make the compiler happy.
// If you're curious, try removing it and changing the 'break'
// statements (except for the last) to goto's.
for (;;) {
if (U_FAILURE(ec)) break;
const char* exp1[] = {"aa", "ab", NOT, "ac", NULL};
expectToPattern(*s, "[a-z{aa}{ab}]", exp1);
s->add("ac");
const char* exp2[] = {"aa", "ab", "ac", NOT, "xy", NULL};
expectToPattern(*s, "[a-z{aa}{ab}{ac}]", exp2);
s->applyPattern(UNICODE_STRING_SIMPLE("[a-z {\\{l} {r\\}}]"), ec);
if (U_FAILURE(ec)) break;
const char* exp3[] = {"{l", "r}", NOT, "xy", NULL};
expectToPattern(*s, UNICODE_STRING_SIMPLE("[a-z{r\\}}{\\{l}]"), exp3);
s->add("[]");
const char* exp4[] = {"{l", "r}", "[]", NOT, "xy", NULL};
expectToPattern(*s, UNICODE_STRING_SIMPLE("[a-z{\\[\\]}{r\\}}{\\{l}]"), exp4);
s->applyPattern(UNICODE_STRING_SIMPLE("[a-z {\\u4E01\\u4E02}{\\n\\r}]"), ec);
if (U_FAILURE(ec)) break;
const char* exp5[] = {"\\u4E01\\u4E02", "\n\r", NULL};
expectToPattern(*s, UNICODE_STRING_SIMPLE("[a-z{\\u000A\\u000D}{\\u4E01\\u4E02}]"), exp5);
// j2189
s->clear();
s->add(UnicodeString("abc", ""));
s->add(UnicodeString("abc", ""));
const char* exp6[] = {"abc", NOT, "ab", NULL};
expectToPattern(*s, "[{abc}]", exp6);
break;
}
if (U_FAILURE(ec)) errln("FAIL: pattern parse error");
delete s;
}
// JB#3400: For 2 character ranges prefer [ab] to [a-b]
UnicodeSet s;
s.add((UChar)97, (UChar)98); // 'a', 'b'
expectToPattern(s, "[ab]", NULL);
}
UBool UnicodeSetTest::toPatternAux(UChar32 start, UChar32 end) {
// use Integer.toString because Utility.hex doesn't handle ints
UnicodeString pat = "";
// TODO do these in hex
//String source = "0x" + Integer.toString(start,16).toUpperCase();
//if (start != end) source += "..0x" + Integer.toString(end,16).toUpperCase();
UnicodeString source;
source = source + (uint32_t)start;
if (start != end)
source = source + ".." + (uint32_t)end;
UnicodeSet testSet;
testSet.add(start, end);
return checkPat(source, testSet);
}
UBool UnicodeSetTest::checkPat(const UnicodeString& source,
const UnicodeSet& testSet) {
// What we want to make sure of is that a pattern generated
// by toPattern(), with or without escaped unprintables, can
// be passed back into the UnicodeSet constructor.
UnicodeString pat0;
testSet.toPattern(pat0, TRUE);
if (!checkPat(source + " (escaped)", testSet, pat0)) return FALSE;
//String pat1 = unescapeLeniently(pat0);
//if (!checkPat(source + " (in code)", testSet, pat1)) return false;
UnicodeString pat2;
testSet.toPattern(pat2, FALSE);
if (!checkPat(source, testSet, pat2)) return FALSE;
//String pat3 = unescapeLeniently(pat2);
// if (!checkPat(source + " (in code)", testSet, pat3)) return false;
//logln(source + " => " + pat0 + ", " + pat1 + ", " + pat2 + ", " + pat3);
logln((UnicodeString)source + " => " + pat0 + ", " + pat2);
return TRUE;
}
UBool UnicodeSetTest::checkPat(const UnicodeString& source,
const UnicodeSet& testSet,
const UnicodeString& pat) {
UErrorCode ec = U_ZERO_ERROR;
UnicodeSet testSet2(pat, ec);
if (testSet2 != testSet) {
errln((UnicodeString)"Fail toPattern: " + source + " => " + pat);
return FALSE;
}
return TRUE;
}
void
UnicodeSetTest::TestPatterns(void) {
UnicodeSet set;
expectPattern(set, UnicodeString("[[a-m]&[d-z]&[k-y]]", ""), "km");
expectPattern(set, UnicodeString("[[a-z]-[m-y]-[d-r]]", ""), "aczz");
expectPattern(set, UnicodeString("[a\\-z]", ""), "--aazz");
expectPattern(set, UnicodeString("[-az]", ""), "--aazz");
expectPattern(set, UnicodeString("[az-]", ""), "--aazz");
expectPattern(set, UnicodeString("[[[a-z]-[aeiou]i]]", ""), "bdfnptvz");
// Throw in a test of complement
set.complement();
UnicodeString exp;
exp.append((UChar)0x0000).append("aeeoouu").append((UChar)(0x007a+1)).append((UChar)0xFFFF);
expectPairs(set, exp);
}
void
UnicodeSetTest::TestCategories(void) {
UErrorCode status = U_ZERO_ERROR;
const char* pat = " [:Lu:] "; // Whitespace ok outside [:..:]
UnicodeSet set(pat, status);
if (U_FAILURE(status)) {
dataerrln((UnicodeString)"Fail: Can't construct set with " + pat + " - " + UnicodeString(u_errorName(status)));
return;
} else {
expectContainment(set, pat, "ABC", "abc");
}
UChar32 i;
int32_t failures = 0;
// Make sure generation of L doesn't pollute cached Lu set
// First generate L, then Lu
set.applyPattern("[:L:]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
for (i=0; i<0x200; ++i) {
UBool l = u_isalpha((UChar)i);
if (l != set.contains(i)) {
errln((UnicodeString)"FAIL: L contains " + (unsigned short)i + " = " +
set.contains(i));
if (++failures == 10) break;
}
}
set.applyPattern("[:Lu:]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
for (i=0; i<0x200; ++i) {
UBool lu = (u_charType((UChar)i) == U_UPPERCASE_LETTER);
if (lu != set.contains(i)) {
errln((UnicodeString)"FAIL: Lu contains " + (unsigned short)i + " = " +
set.contains(i));
if (++failures == 20) break;
}
}
}
void
UnicodeSetTest::TestCloneEqualHash(void) {
UErrorCode status = U_ZERO_ERROR;
// set1 and set2 used to be built with the obsolete constructor taking
// UCharCategory values; replaced with pattern constructors
// markus 20030502
UnicodeSet *set1=new UnicodeSet(UNICODE_STRING_SIMPLE("\\p{Lowercase Letter}"), status); // :Ll: Letter, lowercase
UnicodeSet *set1a=new UnicodeSet(UNICODE_STRING_SIMPLE("[:Ll:]"), status); // Letter, lowercase
if (U_FAILURE(status)){
dataerrln((UnicodeString)"FAIL: Can't construst set with category->Ll" + " - " + UnicodeString(u_errorName(status)));
return;
}
UnicodeSet *set2=new UnicodeSet(UNICODE_STRING_SIMPLE("\\p{Decimal Number}"), status); //Number, Decimal digit
UnicodeSet *set2a=new UnicodeSet(UNICODE_STRING_SIMPLE("[:Nd:]"), status); //Number, Decimal digit
if (U_FAILURE(status)){
errln((UnicodeString)"FAIL: Can't construct set with category->Nd");
return;
}
if (*set1 != *set1a) {
errln("FAIL: category constructor for Ll broken");
}
if (*set2 != *set2a) {
errln("FAIL: category constructor for Nd broken");
}
delete set1a;
delete set2a;
logln("Testing copy construction");
UnicodeSet *set1copy=new UnicodeSet(*set1);
if(*set1 != *set1copy || *set1 == *set2 ||
getPairs(*set1) != getPairs(*set1copy) ||
set1->hashCode() != set1copy->hashCode()){
errln("FAIL : Error in copy construction");
return;
}
logln("Testing =operator");
UnicodeSet set1equal=*set1;
UnicodeSet set2equal=*set2;
if(set1equal != *set1 || set1equal != *set1copy || set2equal != *set2 ||
set2equal == *set1 || set2equal == *set1copy || set2equal == set1equal){
errln("FAIL: Error in =operator");
}
logln("Testing clone()");
UnicodeSet *set1clone=(UnicodeSet*)set1->clone();
UnicodeSet *set2clone=(UnicodeSet*)set2->clone();
if(*set1clone != *set1 || *set1clone != *set1copy || *set1clone != set1equal ||
*set2clone != *set2 || *set2clone == *set1copy || *set2clone != set2equal ||
*set2clone == *set1 || *set2clone == set1equal || *set2clone == *set1clone){
errln("FAIL: Error in clone");
}
logln("Testing hashcode");
if(set1->hashCode() != set1equal.hashCode() || set1->hashCode() != set1clone->hashCode() ||
set2->hashCode() != set2equal.hashCode() || set2->hashCode() != set2clone->hashCode() ||
set1copy->hashCode() != set1equal.hashCode() || set1copy->hashCode() != set1clone->hashCode() ||
set1->hashCode() == set2->hashCode() || set1copy->hashCode() == set2->hashCode() ||
set2->hashCode() == set1clone->hashCode() || set2->hashCode() == set1equal.hashCode() ){
errln("FAIL: Error in hashCode()");
}
delete set1;
delete set1copy;
delete set2;
delete set1clone;
delete set2clone;
}
void
UnicodeSetTest::TestAddRemove(void) {
UnicodeSet set; // Construct empty set
doAssert(set.isEmpty() == TRUE, "set should be empty");
doAssert(set.size() == 0, "size should be 0");
set.complement();
doAssert(set.size() == 0x110000, "size should be 0x110000");
set.clear();
set.add(0x0061, 0x007a);
expectPairs(set, "az");
doAssert(set.isEmpty() == FALSE, "set should not be empty");
doAssert(set.size() != 0, "size should not be equal to 0");
doAssert(set.size() == 26, "size should be equal to 26");
set.remove(0x006d, 0x0070);
expectPairs(set, "alqz");
doAssert(set.size() == 22, "size should be equal to 22");
set.remove(0x0065, 0x0067);
expectPairs(set, "adhlqz");
doAssert(set.size() == 19, "size should be equal to 19");
set.remove(0x0064, 0x0069);
expectPairs(set, "acjlqz");
doAssert(set.size() == 16, "size should be equal to 16");
set.remove(0x0063, 0x0072);
expectPairs(set, "absz");
doAssert(set.size() == 10, "size should be equal to 10");
set.add(0x0066, 0x0071);
expectPairs(set, "abfqsz");
doAssert(set.size() == 22, "size should be equal to 22");
set.remove(0x0061, 0x0067);
expectPairs(set, "hqsz");
set.remove(0x0061, 0x007a);
expectPairs(set, "");
doAssert(set.isEmpty() == TRUE, "set should be empty");
doAssert(set.size() == 0, "size should be 0");
set.add(0x0061);
doAssert(set.isEmpty() == FALSE, "set should not be empty");
doAssert(set.size() == 1, "size should not be equal to 1");
set.add(0x0062);
set.add(0x0063);
expectPairs(set, "ac");
doAssert(set.size() == 3, "size should not be equal to 3");
set.add(0x0070);
set.add(0x0071);
expectPairs(set, "acpq");
doAssert(set.size() == 5, "size should not be equal to 5");
set.clear();
expectPairs(set, "");
doAssert(set.isEmpty() == TRUE, "set should be empty");
doAssert(set.size() == 0, "size should be 0");
// Try removing an entire set from another set
expectPattern(set, "[c-x]", "cx");
UnicodeSet set2;
expectPattern(set2, "[f-ky-za-bc[vw]]", "acfkvwyz");
set.removeAll(set2);
expectPairs(set, "deluxx");
// Try adding an entire set to another set
expectPattern(set, "[jackiemclean]", "aacceein");
expectPattern(set2, "[hitoshinamekatajamesanderson]", "aadehkmort");
set.addAll(set2);
expectPairs(set, "aacehort");
doAssert(set.containsAll(set2) == TRUE, "set should contain all the elements in set2");
// Try retaining an set of elements contained in another set (intersection)
UnicodeSet set3;
expectPattern(set3, "[a-c]", "ac");
doAssert(set.containsAll(set3) == FALSE, "set doesn't contain all the elements in set3");
set3.remove(0x0062);
expectPairs(set3, "aacc");
doAssert(set.containsAll(set3) == TRUE, "set should contain all the elements in set3");
set.retainAll(set3);
expectPairs(set, "aacc");
doAssert(set.size() == set3.size(), "set.size() should be set3.size()");
doAssert(set.containsAll(set3) == TRUE, "set should contain all the elements in set3");
set.clear();
doAssert(set.size() != set3.size(), "set.size() != set3.size()");
// Test commutativity
expectPattern(set, "[hitoshinamekatajamesanderson]", "aadehkmort");
expectPattern(set2, "[jackiemclean]", "aacceein");
set.addAll(set2);
expectPairs(set, "aacehort");
doAssert(set.containsAll(set2) == TRUE, "set should contain all the elements in set2");
}
/**
* Make sure minimal representation is maintained.
*/
void UnicodeSetTest::TestMinimalRep() {
UErrorCode status = U_ZERO_ERROR;
// This is pretty thoroughly tested by checkCanonicalRep()
// run against the exhaustive operation results. Use the code
// here for debugging specific spot problems.
// 1 overlap against 2
UnicodeSet set("[h-km-q]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
UnicodeSet set2("[i-o]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
set.addAll(set2);
expectPairs(set, "hq");
// right
set.applyPattern("[a-m]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
set2.applyPattern("[e-o]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
set.addAll(set2);
expectPairs(set, "ao");
// left
set.applyPattern("[e-o]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
set2.applyPattern("[a-m]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
set.addAll(set2);
expectPairs(set, "ao");
// 1 overlap against 3
set.applyPattern("[a-eg-mo-w]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
set2.applyPattern("[d-q]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
set.addAll(set2);
expectPairs(set, "aw");
}
void UnicodeSetTest::TestAPI() {
UErrorCode status = U_ZERO_ERROR;
// default ct
UnicodeSet set;
if (!set.isEmpty() || set.getRangeCount() != 0) {
errln((UnicodeString)"FAIL, set should be empty but isn't: " +
set);
}
// clear(), isEmpty()
set.add(0x0061);
if (set.isEmpty()) {
errln((UnicodeString)"FAIL, set shouldn't be empty but is: " +
set);
}
set.clear();
if (!set.isEmpty()) {
errln((UnicodeString)"FAIL, set should be empty but isn't: " +
set);
}
// size()
set.clear();
if (set.size() != 0) {
errln((UnicodeString)"FAIL, size should be 0, but is " + set.size() +
": " + set);
}
set.add(0x0061);
if (set.size() != 1) {
errln((UnicodeString)"FAIL, size should be 1, but is " + set.size() +
": " + set);
}
set.add(0x0031, 0x0039);
if (set.size() != 10) {
errln((UnicodeString)"FAIL, size should be 10, but is " + set.size() +
": " + set);
}
// contains(first, last)
set.clear();
set.applyPattern("[A-Y 1-8 b-d l-y]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
for (int32_t i = 0; i<set.getRangeCount(); ++i) {
UChar32 a = set.getRangeStart(i);
UChar32 b = set.getRangeEnd(i);
if (!set.contains(a, b)) {
errln((UnicodeString)"FAIL, should contain " + (unsigned short)a + '-' + (unsigned short)b +
" but doesn't: " + set);
}
if (set.contains((UChar32)(a-1), b)) {
errln((UnicodeString)"FAIL, shouldn't contain " +
(unsigned short)(a-1) + '-' + (unsigned short)b +
" but does: " + set);
}
if (set.contains(a, (UChar32)(b+1))) {
errln((UnicodeString)"FAIL, shouldn't contain " +
(unsigned short)a + '-' + (unsigned short)(b+1) +
" but does: " + set);
}
}
// Ported InversionList test.
UnicodeSet a((UChar32)3,(UChar32)10);
UnicodeSet b((UChar32)7,(UChar32)15);
UnicodeSet c;
logln((UnicodeString)"a [3-10]: " + a);
logln((UnicodeString)"b [7-15]: " + b);
c = a;
c.addAll(b);
UnicodeSet exp((UChar32)3,(UChar32)15);
if (c == exp) {
logln((UnicodeString)"c.set(a).add(b): " + c);
} else {
errln((UnicodeString)"FAIL: c.set(a).add(b) = " + c + ", expect " + exp);
}
c.complement();
exp.set((UChar32)0, (UChar32)2);
exp.add((UChar32)16, UnicodeSet::MAX_VALUE);
if (c == exp) {
logln((UnicodeString)"c.complement(): " + c);
} else {
errln((UnicodeString)"FAIL: c.complement() = " + c + ", expect " + exp);
}
c.complement();
exp.set((UChar32)3, (UChar32)15);
if (c == exp) {
logln((UnicodeString)"c.complement(): " + c);
} else {
errln((UnicodeString)"FAIL: c.complement() = " + c + ", expect " + exp);
}
c = a;
c.complementAll(b);
exp.set((UChar32)3,(UChar32)6);
exp.add((UChar32)11,(UChar32) 15);
if (c == exp) {
logln((UnicodeString)"c.set(a).exclusiveOr(b): " + c);
} else {
errln((UnicodeString)"FAIL: c.set(a).exclusiveOr(b) = " + c + ", expect " + exp);
}
exp = c;
bitsToSet(setToBits(c), c);
if (c == exp) {
logln((UnicodeString)"bitsToSet(setToBits(c)): " + c);
} else {
errln((UnicodeString)"FAIL: bitsToSet(setToBits(c)) = " + c + ", expect " + exp);
}
// Additional tests for coverage JB#2118
//UnicodeSet::complement(class UnicodeString const &)
//UnicodeSet::complementAll(class UnicodeString const &)
//UnicodeSet::containsNone(class UnicodeSet const &)
//UnicodeSet::containsNone(long,long)
//UnicodeSet::containsSome(class UnicodeSet const &)
//UnicodeSet::containsSome(long,long)
//UnicodeSet::removeAll(class UnicodeString const &)
//UnicodeSet::retain(long)
//UnicodeSet::retainAll(class UnicodeString const &)
//UnicodeSet::serialize(unsigned short *,long,enum UErrorCode &)
//UnicodeSetIterator::getString(void)
set.clear();
set.complement("ab");
exp.applyPattern("[{ab}]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
if (set != exp) { errln("FAIL: complement(\"ab\")"); return; }
UnicodeSetIterator iset(set);
if (!iset.next() || !iset.isString()) {
errln("FAIL: UnicodeSetIterator::next/isString");
} else if (iset.getString() != "ab") {
errln("FAIL: UnicodeSetIterator::getString");
}
set.add((UChar32)0x61, (UChar32)0x7A);
set.complementAll("alan");
exp.applyPattern("[{ab}b-kmo-z]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
if (set != exp) { errln("FAIL: complementAll(\"alan\")"); return; }
exp.applyPattern("[a-z]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
if (set.containsNone(exp)) { errln("FAIL: containsNone(UnicodeSet)"); }
if (!set.containsSome(exp)) { errln("FAIL: containsSome(UnicodeSet)"); }
exp.applyPattern("[aln]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
if (!set.containsNone(exp)) { errln("FAIL: containsNone(UnicodeSet)"); }
if (set.containsSome(exp)) { errln("FAIL: containsSome(UnicodeSet)"); }
if (set.containsNone((UChar32)0x61, (UChar32)0x7A)) {
errln("FAIL: containsNone(UChar32, UChar32)");
}
if (!set.containsSome((UChar32)0x61, (UChar32)0x7A)) {
errln("FAIL: containsSome(UChar32, UChar32)");
}
if (!set.containsNone((UChar32)0x41, (UChar32)0x5A)) {
errln("FAIL: containsNone(UChar32, UChar32)");
}
if (set.containsSome((UChar32)0x41, (UChar32)0x5A)) {
errln("FAIL: containsSome(UChar32, UChar32)");
}
set.removeAll("liu");
exp.applyPattern("[{ab}b-hj-kmo-tv-z]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
if (set != exp) { errln("FAIL: removeAll(\"liu\")"); return; }
set.retainAll("star");
exp.applyPattern("[rst]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
if (set != exp) { errln("FAIL: retainAll(\"star\")"); return; }
set.retain((UChar32)0x73);
exp.applyPattern("[s]", status);
if (U_FAILURE(status)) { errln("FAIL"); return; }
if (set != exp) { errln("FAIL: retain('s')"); return; }
uint16_t buf[32];
int32_t slen = set.serialize(buf, sizeof(buf)/sizeof(buf[0]), status);
if (U_FAILURE(status)) { errln("FAIL: serialize"); return; }
if (slen != 3 || buf[0] != 2 || buf[1] != 0x73 || buf[2] != 0x74) {
errln("FAIL: serialize");
return;
}
// Conversions to and from USet
UnicodeSet *uniset = &set;
USet *uset = uniset->toUSet();
TEST_ASSERT((void *)uset == (void *)uniset);
UnicodeSet *setx = UnicodeSet::fromUSet(uset);
TEST_ASSERT((void *)setx == (void *)uset);
const UnicodeSet *constSet = uniset;
const USet *constUSet = constSet->toUSet();
TEST_ASSERT((void *)constUSet == (void *)constSet);
const UnicodeSet *constSetx = UnicodeSet::fromUSet(constUSet);
TEST_ASSERT((void *)constSetx == (void *)constUSet);
// span(UnicodeString) and spanBack(UnicodeString) convenience methods
UnicodeString longString=UNICODE_STRING_SIMPLE("aaaaaaaaaabbbbbbbbbbcccccccccc");
UnicodeSet ac(0x61, 0x63);
ac.remove(0x62).freeze();
if( ac.span(longString, -5, USET_SPAN_CONTAINED)!=10 ||
ac.span(longString, 0, USET_SPAN_CONTAINED)!=10 ||
ac.span(longString, 5, USET_SPAN_CONTAINED)!=10 ||
ac.span(longString, 10, USET_SPAN_CONTAINED)!=10 ||
ac.span(longString, 15, USET_SPAN_CONTAINED)!=15 ||
ac.span(longString, 20, USET_SPAN_CONTAINED)!=30 ||
ac.span(longString, 25, USET_SPAN_CONTAINED)!=30 ||
ac.span(longString, 30, USET_SPAN_CONTAINED)!=30 ||
ac.span(longString, 35, USET_SPAN_CONTAINED)!=30 ||
ac.span(longString, INT32_MAX, USET_SPAN_CONTAINED)!=30
) {
errln("UnicodeSet.span(UnicodeString, ...) returns incorrect end indexes");
}
if( ac.spanBack(longString, -5, USET_SPAN_CONTAINED)!=0 ||
ac.spanBack(longString, 0, USET_SPAN_CONTAINED)!=0 ||
ac.spanBack(longString, 5, USET_SPAN_CONTAINED)!=0 ||
ac.spanBack(longString, 10, USET_SPAN_CONTAINED)!=0 ||
ac.spanBack(longString, 15, USET_SPAN_CONTAINED)!=15 ||
ac.spanBack(longString, 20, USET_SPAN_CONTAINED)!=20 ||
ac.spanBack(longString, 25, USET_SPAN_CONTAINED)!=20 ||
ac.spanBack(longString, 30, USET_SPAN_CONTAINED)!=20 ||
ac.spanBack(longString, 35, USET_SPAN_CONTAINED)!=20 ||
ac.spanBack(longString, INT32_MAX, USET_SPAN_CONTAINED)!=20
) {
errln("UnicodeSet.spanBack(UnicodeString, ...) returns incorrect start indexes");
}
}
void UnicodeSetTest::TestIteration() {
UErrorCode ec = U_ZERO_ERROR;
int i = 0;
int outerLoop;
// 6 code points, 3 ranges, 2 strings, 8 total elements
// Iteration will access them in sorted order - a, b, c, y, z, U0001abcd, "str1", "str2"
UnicodeSet set(UNICODE_STRING_SIMPLE("[zabyc\\U0001abcd{str1}{str2}]"), ec);
TEST_ASSERT_SUCCESS(ec);
UnicodeSetIterator it(set);
for (outerLoop=0; outerLoop<3; outerLoop++) {
// Run the test multiple times, to check that iterator.reset() is working.
for (i=0; i<10; i++) {
UBool nextv = it.next();
UBool isString = it.isString();
int32_t codePoint = it.getCodepoint();
//int32_t codePointEnd = it.getCodepointEnd();
UnicodeString s = it.getString();
switch (i) {
case 0:
TEST_ASSERT(nextv == TRUE);
TEST_ASSERT(isString == FALSE);
TEST_ASSERT(codePoint==0x61);
TEST_ASSERT(s == "a");
break;
case 1:
TEST_ASSERT(nextv == TRUE);
TEST_ASSERT(isString == FALSE);
TEST_ASSERT(codePoint==0x62);
TEST_ASSERT(s == "b");
break;
case 2:
TEST_ASSERT(nextv == TRUE);
TEST_ASSERT(isString == FALSE);
TEST_ASSERT(codePoint==0x63);
TEST_ASSERT(s == "c");
break;
case 3:
TEST_ASSERT(nextv == TRUE);
TEST_ASSERT(isString == FALSE);
TEST_ASSERT(codePoint==0x79);
TEST_ASSERT(s == "y");
break;
case 4:
TEST_ASSERT(nextv == TRUE);
TEST_ASSERT(isString == FALSE);
TEST_ASSERT(codePoint==0x7a);
TEST_ASSERT(s == "z");
break;
case 5:
TEST_ASSERT(nextv == TRUE);
TEST_ASSERT(isString == FALSE);
TEST_ASSERT(codePoint==0x1abcd);
TEST_ASSERT(s == UnicodeString((UChar32)0x1abcd));
break;
case 6:
TEST_ASSERT(nextv == TRUE);
TEST_ASSERT(isString == TRUE);
TEST_ASSERT(s == "str1");
break;
case 7:
TEST_ASSERT(nextv == TRUE);
TEST_ASSERT(isString == TRUE);
TEST_ASSERT(s == "str2");
break;
case 8:
TEST_ASSERT(nextv == FALSE);
break;
case 9:
TEST_ASSERT(nextv == FALSE);
break;
}
}
it.reset(); // prepare to run the iteration again.
}
}
void UnicodeSetTest::TestStrings() {
UErrorCode ec = U_ZERO_ERROR;
UnicodeSet* testList[] = {
UnicodeSet::createFromAll("abc"),
new UnicodeSet("[a-c]", ec),
&(UnicodeSet::createFrom("ch")->add('a','z').add("ll")),
new UnicodeSet("[{ll}{ch}a-z]", ec),
UnicodeSet::createFrom("ab}c"),
new UnicodeSet("[{ab\\}c}]", ec),
&((new UnicodeSet('a','z'))->add('A', 'Z').retain('M','m').complement('X')),
new UnicodeSet("[[a-zA-Z]&[M-m]-[X]]", ec),
NULL
};
if (U_FAILURE(ec)) {
errln("FAIL: couldn't construct test sets");
}
for (int32_t i = 0; testList[i] != NULL; i+=2) {
if (U_SUCCESS(ec)) {
UnicodeString pat0, pat1;
testList[i]->toPattern(pat0, TRUE);
testList[i+1]->toPattern(pat1, TRUE);
if (*testList[i] == *testList[i+1]) {
logln((UnicodeString)"Ok: " + pat0 + " == " + pat1);
} else {
logln((UnicodeString)"FAIL: " + pat0 + " != " + pat1);
}
}
delete testList[i];
delete testList[i+1];
}
}
/**
* Test the [:Latin:] syntax.
*/
void UnicodeSetTest::TestScriptSet() {
expectContainment(UNICODE_STRING_SIMPLE("[:Latin:]"), "aA", CharsToUnicodeString("\\u0391\\u03B1"));
expectContainment(UNICODE_STRING_SIMPLE("[:Greek:]"), CharsToUnicodeString("\\u0391\\u03B1"), "aA");
/* Jitterbug 1423 */
expectContainment(UNICODE_STRING_SIMPLE("[[:Common:][:Inherited:]]"), CharsToUnicodeString("\\U00003099\\U0001D169\\u0000"), "aA");
}
/**
* Test the [:Latin:] syntax.
*/
void UnicodeSetTest::TestPropertySet() {
static const char* const DATA[] = {
// Pattern, Chars IN, Chars NOT in
"[:Latin:]",
"aA",
"\\u0391\\u03B1",
"[\\p{Greek}]",
"\\u0391\\u03B1",
"aA",
"\\P{ GENERAL Category = upper case letter }",
"abc",
"ABC",
#if !UCONFIG_NO_NORMALIZATION
// Combining class: @since ICU 2.2
// Check both symbolic and numeric
"\\p{ccc=Nukta}",
"\\u0ABC",
"abc",
"\\p{Canonical Combining Class = 11}",
"\\u05B1",
"\\u05B2",
"[:c c c = iota subscript :]",
"\\u0345",
"xyz",
#endif
// Bidi class: @since ICU 2.2
"\\p{bidiclass=lefttoright}",
"abc",
"\\u0671\\u0672",
// Binary properties: @since ICU 2.2
"\\p{ideographic}",
"\\u4E0A",
"x",
"[:math=false:]",
"q)*(",
// weiv: )(and * were removed from math in Unicode 4.0.1
//"(*+)",
"+<>^",
// JB#1767 \N{}, \p{ASCII}
"[:Ascii:]",
"abc\\u0000\\u007F",
"\\u0080\\u4E00",
"[\\N{ latin small letter a }[:name= latin small letter z:]]",
"az",
"qrs",
// JB#2015
"[:any:]",
"a\\U0010FFFF",
"",
"[:nv=0.5:]",
"\\u00BD\\u0F2A",
"\\u00BC",
// JB#2653: Age
"[:Age=1.1:]",
"\\u03D6", // 1.1
"\\u03D8\\u03D9", // 3.2
"[:Age=3.1:]",
"\\u1800\\u3400\\U0002f800",
"\\u0220\\u034f\\u30ff\\u33ff\\ufe73\\U00010000\\U00050000",
// JB#2350: Case_Sensitive
"[:Case Sensitive:]",
"A\\u1FFC\\U00010410",
";\\u00B4\\U00010500",
// JB#2832: C99-compatibility props
"[:blank:]",
" \\u0009",
"1-9A-Z",
"[:graph:]",
"19AZ",
" \\u0003\\u0007\\u0009\\u000A\\u000D",
"[:punct:]",
"!@#%&*()[]{}-_\\/;:,.?'\"",
"09azAZ",
"[:xdigit:]",
"09afAF",
"gG!",
// Regex compatibility test
"[-b]", // leading '-' is literal
"-b",
"ac",
"[^-b]", // leading '-' is literal
"ac",
"-b",
"[b-]", // trailing '-' is literal
"-b",
"ac",
"[^b-]", // trailing '-' is literal
"ac",
"-b",
"[a-b-]", // trailing '-' is literal
"ab-",
"c=",
"[[a-q]&[p-z]-]", // trailing '-' is literal
"pq-",
"or=",
"[\\s|\\)|:|$|\\>]", // from regex tests
"s|):$>",
"abc",
"[\\uDC00cd]", // JB#2906: isolated trail at start
"cd\\uDC00",
"ab\\uD800\\U00010000",
"[ab\\uD800]", // JB#2906: isolated trail at start
"ab\\uD800",
"cd\\uDC00\\U00010000",
"[ab\\uD800cd]", // JB#2906: isolated lead in middle
"abcd\\uD800",
"ef\\uDC00\\U00010000",
"[ab\\uDC00cd]", // JB#2906: isolated trail in middle
"abcd\\uDC00",
"ef\\uD800\\U00010000",
#if !UCONFIG_NO_NORMALIZATION
"[:^lccc=0:]", // Lead canonical class
"\\u0300\\u0301",
"abcd\\u00c0\\u00c5",
"[:^tccc=0:]", // Trail canonical class
"\\u0300\\u0301\\u00c0\\u00c5",
"abcd",
"[[:^lccc=0:][:^tccc=0:]]", // Lead and trail canonical class
"\\u0300\\u0301\\u00c0\\u00c5",
"abcd",
"[[:^lccc=0:]-[:^tccc=0:]]", // Stuff that starts with an accent but ends with a base (none right now)
"",
"abcd\\u0300\\u0301\\u00c0\\u00c5",
"[[:ccc=0:]-[:lccc=0:]-[:tccc=0:]]", // Weirdos. Complete canonical class is zero, but both lead and trail are not
"\\u0F73\\u0F75\\u0F81",
"abcd\\u0300\\u0301\\u00c0\\u00c5",
#endif /* !UCONFIG_NO_NORMALIZATION */
"[:Assigned:]",
"A\\uE000\\uF8FF\\uFDC7\\U00010000\\U0010FFFD",
"\\u0888\\uFDD3\\uFFFE\\U00050005",
// Script_Extensions, new in Unicode 6.0
"[:scx=Arab:]",
"\\u061E\\u061F\\u0620\\u0621\\u063F\\u0640\\u0650\\u065E\\uFDF1\\uFDF2\\uFDF3",
"\\u061D\\u065F\\uFDEF\\uFDFE",
// U+FDF2 has Script=Arabic and also Arab in its Script_Extensions,
// so scx-sc is missing U+FDF2.
"[[:Script_Extensions=Arabic:]-[:Arab:]]",
"\\u0640\\u064B\\u0650\\u0655\\uFDFD",
"\\uFDF2"
};
static const int32_t DATA_LEN = sizeof(DATA)/sizeof(DATA[0]);
for (int32_t i=0; i<DATA_LEN; i+=3) {
expectContainment(UnicodeString(DATA[i], -1, US_INV), CharsToUnicodeString(DATA[i+1]),
CharsToUnicodeString(DATA[i+2]));
}
}
/**
* Test that Posix style character classes [:digit:], etc.
* have the Unicode definitions from TR 18.
*/
void UnicodeSetTest::TestPosixClasses() {
{
UErrorCode status = U_ZERO_ERROR;
UnicodeSet s1("[:alpha:]", status);
UnicodeSet s2(UNICODE_STRING_SIMPLE("\\p{Alphabetic}"), status);
TEST_ASSERT_SUCCESS(status);
TEST_ASSERT(s1==s2);
}
{
UErrorCode status = U_ZERO_ERROR;
UnicodeSet s1("[:lower:]", status);
UnicodeSet s2(UNICODE_STRING_SIMPLE("\\p{lowercase}"), status);
TEST_ASSERT_SUCCESS(status);
TEST_ASSERT(s1==s2);
}
{
UErrorCode status = U_ZERO_ERROR;
UnicodeSet s1("[:upper:]", status);
UnicodeSet s2(UNICODE_STRING_SIMPLE("\\p{Uppercase}"), status);
TEST_ASSERT_SUCCESS(status);
TEST_ASSERT(s1==s2);
}
{
UErrorCode status = U_ZERO_ERROR;
UnicodeSet s1("[:punct:]", status);
UnicodeSet s2(UNICODE_STRING_SIMPLE("\\p{gc=Punctuation}"), status);
TEST_ASSERT_SUCCESS(status);
TEST_ASSERT(s1==s2);
}
{
UErrorCode status = U_ZERO_ERROR;
UnicodeSet s1("[:digit:]", status);
UnicodeSet s2(UNICODE_STRING_SIMPLE("\\p{gc=DecimalNumber}"), status);
TEST_ASSERT_SUCCESS(status);
TEST_ASSERT(s1==s2);
}
{
UErrorCode status = U_ZERO_ERROR;
UnicodeSet s1("[:xdigit:]", status);
UnicodeSet s2(UNICODE_STRING_SIMPLE("[\\p{DecimalNumber}\\p{HexDigit}]"), status);
TEST_ASSERT_SUCCESS(status);
TEST_ASSERT(s1==s2);
}
{
UErrorCode status = U_ZERO_ERROR;
UnicodeSet s1("[:alnum:]", status);
UnicodeSet s2(UNICODE_STRING_SIMPLE("[\\p{Alphabetic}\\p{DecimalNumber}]"), status);
TEST_ASSERT_SUCCESS(status);
TEST_ASSERT(s1==s2);
}
{
UErrorCode status = U_ZERO_ERROR;
UnicodeSet s1("[:space:]", status);
UnicodeSet s2(UNICODE_STRING_SIMPLE("\\p{Whitespace}"), status);
TEST_ASSERT_SUCCESS(status);
TEST_ASSERT(s1==s2);
}
{
UErrorCode status = U_ZERO_ERROR;
UnicodeSet s1("[:blank:]", status);
TEST_ASSERT_SUCCESS(status);
UnicodeSet s2(UNICODE_STRING_SIMPLE("[\\p{Whitespace}-[\\u000a\\u000B\\u000c\\u000d\\u0085\\p{LineSeparator}\\p{ParagraphSeparator}]]"),
status);
TEST_ASSERT_SUCCESS(status);
TEST_ASSERT(s1==s2);
}
{
UErrorCode status = U_ZERO_ERROR;
UnicodeSet s1("[:cntrl:]", status);
TEST_ASSERT_SUCCESS(status);
UnicodeSet s2(UNICODE_STRING_SIMPLE("\\p{Control}"), status);
TEST_ASSERT_SUCCESS(status);
TEST_ASSERT(s1==s2);
}
{
UErrorCode status = U_ZERO_ERROR;
UnicodeSet s1("[:graph:]", status);
TEST_ASSERT_SUCCESS(status);
UnicodeSet s2(UNICODE_STRING_SIMPLE("[^\\p{Whitespace}\\p{Control}\\p{Surrogate}\\p{Unassigned}]"), status);
TEST_ASSERT_SUCCESS(status);
TEST_ASSERT(s1==s2);
}
{
UErrorCode status = U_ZERO_ERROR;
UnicodeSet s1("[:print:]", status);
TEST_ASSERT_SUCCESS(status);
UnicodeSet s2(UNICODE_STRING_SIMPLE("[[:graph:][:blank:]-[\\p{Control}]]") ,status);
TEST_ASSERT_SUCCESS(status);
TEST_ASSERT(s1==s2);
}
}
/**
* Test cloning of UnicodeSet. For C++, we test the copy constructor.
*/
void UnicodeSetTest::TestClone() {
UErrorCode ec = U_ZERO_ERROR;
UnicodeSet s("[abcxyz]", ec);
UnicodeSet t(s);
expectContainment(t, "abc", "def");
}
/**
* Test the indexOf() and charAt() methods.
*/
void UnicodeSetTest::TestIndexOf() {
UErrorCode ec = U_ZERO_ERROR;
UnicodeSet set("[a-cx-y3578]", ec);
if (U_FAILURE(ec)) {
errln("FAIL: UnicodeSet constructor");
return;
}
for (int32_t i=0; i<set.size(); ++i) {
UChar32 c = set.charAt(i);
if (set.indexOf(c) != i) {
errln("FAIL: charAt(%d) = %X => indexOf() => %d",
i, c, set.indexOf(c));
}
}
UChar32 c = set.charAt(set.size());
if (c != -1) {
errln("FAIL: charAt(<out of range>) = %X", c);
}
int32_t j = set.indexOf((UChar32)0x71/*'q'*/);
if (j != -1) {
errln((UnicodeString)"FAIL: indexOf('q') = " + j);
}
}
/**
* Test closure API.
*/
void UnicodeSetTest::TestCloseOver() {
UErrorCode ec = U_ZERO_ERROR;
char CASE[] = {(char)USET_CASE_INSENSITIVE};
char CASE_MAPPINGS[] = {(char)USET_ADD_CASE_MAPPINGS};
const char* DATA[] = {
// selector, input, output
CASE,
"[aq\\u00DF{Bc}{bC}{Fi}]",
"[aAqQ\\u00DF\\u1E9E\\uFB01{ss}{bc}{fi}]", // U+1E9E LATIN CAPITAL LETTER SHARP S is new in Unicode 5.1
CASE,
"[\\u01F1]", // 'DZ'
"[\\u01F1\\u01F2\\u01F3]",
CASE,
"[\\u1FB4]",
"[\\u1FB4{\\u03AC\\u03B9}]",
CASE,
"[{F\\uFB01}]",
"[\\uFB03{ffi}]",
CASE, // make sure binary search finds limits
"[a\\uFF3A]",
"[aA\\uFF3A\\uFF5A]",
CASE,
"[a-z]","[A-Za-z\\u017F\\u212A]",
CASE,
"[abc]","[A-Ca-c]",
CASE,
"[ABC]","[A-Ca-c]",
CASE, "[i]", "[iI]",
CASE, "[\\u0130]", "[\\u0130{i\\u0307}]", // dotted I
CASE, "[{i\\u0307}]", "[\\u0130{i\\u0307}]", // i with dot
CASE, "[\\u0131]", "[\\u0131]", // dotless i
CASE, "[\\u0390]", "[\\u0390\\u1FD3{\\u03B9\\u0308\\u0301}]",
CASE, "[\\u03c2]", "[\\u03a3\\u03c2\\u03c3]", // sigmas
CASE, "[\\u03f2]", "[\\u03f2\\u03f9]", // lunate sigmas
CASE, "[\\u03f7]", "[\\u03f7\\u03f8]",
CASE, "[\\u1fe3]", "[\\u03b0\\u1fe3{\\u03c5\\u0308\\u0301}]",
CASE, "[\\ufb05]", "[\\ufb05\\ufb06{st}]",
CASE, "[{st}]", "[\\ufb05\\ufb06{st}]",
CASE, "[\\U0001044F]", "[\\U00010427\\U0001044F]",
CASE, "[{a\\u02BE}]", "[\\u1E9A{a\\u02BE}]", // first in sorted table
CASE, "[{\\u1f7c\\u03b9}]", "[\\u1ff2{\\u1f7c\\u03b9}]", // last in sorted table
#if !UCONFIG_NO_FILE_IO
CASE_MAPPINGS,
"[aq\\u00DF{Bc}{bC}{Fi}]",
"[aAqQ\\u00DF{ss}{Ss}{SS}{Bc}{BC}{bC}{bc}{FI}{Fi}{fi}]",
#endif
CASE_MAPPINGS,
"[\\u01F1]", // 'DZ'
"[\\u01F1\\u01F2\\u01F3]",
CASE_MAPPINGS,
"[a-z]",
"[A-Za-z]",
NULL
};
UnicodeSet s;
UnicodeSet t;
UnicodeString buf;
for (int32_t i=0; DATA[i]!=NULL; i+=3) {
int32_t selector = DATA[i][0];
UnicodeString pat(DATA[i+1], -1, US_INV);
UnicodeString exp(DATA[i+2], -1, US_INV);
s.applyPattern(pat, ec);
s.closeOver(selector);
t.applyPattern(exp, ec);
if (U_FAILURE(ec)) {
errln("FAIL: applyPattern failed");
continue;
}
if (s == t) {
logln((UnicodeString)"Ok: " + pat + ".closeOver(" + selector + ") => " + exp);
} else {
dataerrln((UnicodeString)"FAIL: " + pat + ".closeOver(" + selector + ") => " +
s.toPattern(buf, TRUE) + ", expected " + exp);
}
}
#if 0
/*
* Unused test code.
* This was used to compare the old implementation (using USET_CASE)
* with the new one (using 0x100 temporarily)
* while transitioning from hardcoded case closure tables in uniset.cpp
* (moved to uniset_props.cpp) to building the data by gencase into ucase.icu.
* and using ucase.c functions for closure.
* See Jitterbug 3432 RFE: Move uniset.cpp data to a data file
*
* Note: The old and new implementation never fully matched because
* the old implementation turned out to not map U+0130 and U+0131 correctly
* (dotted I and dotless i) and because the old implementation's data tables
* were outdated compared to Unicode 4.0.1 at the time of the change to the
* new implementation. (So sigmas and some other characters were not handled
* according to the newer Unicode version.)
*/
UnicodeSet sens("[:case_sensitive:]", ec), sens2, s2;
UnicodeSetIterator si(sens);
UnicodeString str, buf2;
const UnicodeString *pStr;
UChar32 c;
while(si.next()) {
if(!si.isString()) {
c=si.getCodepoint();
s.clear();
s.add(c);
str.setTo(c);
str.foldCase();
sens2.add(str);
t=s;
s.closeOver(USET_CASE);
t.closeOver(0x100);
if(s!=t) {
errln("FAIL: closeOver(U+%04x) differs: ", c);
errln((UnicodeString)"old "+s.toPattern(buf, TRUE)+" new: "+t.toPattern(buf2, TRUE));
}
}
}
// remove all code points
// should contain all full case folding mapping strings
sens2.remove(0, 0x10ffff);
si.reset(sens2);
while(si.next()) {
if(si.isString()) {
pStr=&si.getString();
s.clear();
s.add(*pStr);
t=s2=s;
s.closeOver(USET_CASE);
t.closeOver(0x100);
if(s!=t) {
errln((UnicodeString)"FAIL: closeOver("+s2.toPattern(buf, TRUE)+") differs: ");
errln((UnicodeString)"old "+s.toPattern(buf, TRUE)+" new: "+t.toPattern(buf2, TRUE));
}
}
}
#endif
// Test the pattern API
s.applyPattern("[abc]", USET_CASE_INSENSITIVE, NULL, ec);
if (U_FAILURE(ec)) {
errln("FAIL: applyPattern failed");
} else {
expectContainment(s, "abcABC", "defDEF");
}
UnicodeSet v("[^abc]", USET_CASE_INSENSITIVE, NULL, ec);
if (U_FAILURE(ec)) {
errln("FAIL: constructor failed");
} else {
expectContainment(v, "defDEF", "abcABC");
}
UnicodeSet cm("[abck]", USET_ADD_CASE_MAPPINGS, NULL, ec);
if (U_FAILURE(ec)) {
errln("FAIL: construct w/case mappings failed");
} else {
expectContainment(cm, "abckABCK", CharsToUnicodeString("defDEF\\u212A"));
}
}
void UnicodeSetTest::TestEscapePattern() {
const char pattern[] =
"[\\uFEFF \\u200A-\\u200E \\U0001D173-\\U0001D17A \\U000F0000-\\U000FFFFD ]";
const char exp[] =
"[\\u200A-\\u200E\\uFEFF\\U0001D173-\\U0001D17A\\U000F0000-\\U000FFFFD]";
// We test this with two passes; in the second pass we
// pre-unescape the pattern. Since U+200E is Pattern_White_Space,
// this fails -- which is what we expect.
for (int32_t pass=1; pass<=2; ++pass) {
UErrorCode ec = U_ZERO_ERROR;
UnicodeString pat(pattern, -1, US_INV);
if (pass==2) {
pat = pat.unescape();
}
// Pattern is only good for pass 1
UBool isPatternValid = (pass==1);
UnicodeSet set(pat, ec);
if (U_SUCCESS(ec) != isPatternValid){
errln((UnicodeString)"FAIL: applyPattern(" +
escape(pat) + ") => " +
u_errorName(ec));
continue;
}
if (U_FAILURE(ec)) {
continue;
}
if (set.contains((UChar)0x0644)){
errln((UnicodeString)"FAIL: " + escape(pat) + " contains(U+0664)");
}
UnicodeString newpat;
set.toPattern(newpat, TRUE);
if (newpat == UnicodeString(exp, -1, US_INV)) {
logln(escape(pat) + " => " + newpat);
} else {
errln((UnicodeString)"FAIL: " + escape(pat) + " => " + newpat);
}
for (int32_t i=0; i<set.getRangeCount(); ++i) {
UnicodeString str("Range ");
str.append((UChar)(0x30 + i))
.append(": ")
.append((UChar32)set.getRangeStart(i))
.append(" - ")
.append((UChar32)set.getRangeEnd(i));
str = str + " (" + set.getRangeStart(i) + " - " +
set.getRangeEnd(i) + ")";
if (set.getRangeStart(i) < 0) {
errln((UnicodeString)"FAIL: " + escape(str));
} else {
logln(escape(str));
}
}
}
}
void UnicodeSetTest::expectRange(const UnicodeString& label,
const UnicodeSet& set,
UChar32 start, UChar32 end) {
UnicodeSet exp(start, end);
UnicodeString pat;
if (set == exp) {
logln(label + " => " + set.toPattern(pat, TRUE));
} else {
UnicodeString xpat;
errln((UnicodeString)"FAIL: " + label + " => " +
set.toPattern(pat, TRUE) +
", expected " + exp.toPattern(xpat, TRUE));
}
}
void UnicodeSetTest::TestInvalidCodePoint() {
const UChar32 DATA[] = {
// Test range Expected range
0, 0x10FFFF, 0, 0x10FFFF,
(UChar32)-1, 8, 0, 8,
8, 0x110000, 8, 0x10FFFF
};
const int32_t DATA_LENGTH = sizeof(DATA)/sizeof(DATA[0]);
UnicodeString pat;
int32_t i;
for (i=0; i<DATA_LENGTH; i+=4) {
UChar32 start = DATA[i];
UChar32 end = DATA[i+1];
UChar32 xstart = DATA[i+2];
UChar32 xend = DATA[i+3];
// Try various API using the test code points
UnicodeSet set(start, end);
expectRange((UnicodeString)"ct(" + start + "," + end + ")",
set, xstart, xend);
set.clear();
set.set(start, end);
expectRange((UnicodeString)"set(" + start + "," + end + ")",
set, xstart, xend);
UBool b = set.contains(start);
b = set.contains(start, end);
b = set.containsNone(start, end);
b = set.containsSome(start, end);
/*int32_t index = set.indexOf(start);*/
set.clear();
set.add(start);
set.add(start, end);
expectRange((UnicodeString)"add(" + start + "," + end + ")",
set, xstart, xend);
set.set(0, 0x10FFFF);
set.retain(start, end);
expectRange((UnicodeString)"retain(" + start + "," + end + ")",
set, xstart, xend);
set.retain(start);
set.set(0, 0x10FFFF);
set.remove(start);
set.remove(start, end);
set.complement();
expectRange((UnicodeString)"!remove(" + start + "," + end + ")",
set, xstart, xend);
set.set(0, 0x10FFFF);
set.complement(start, end);
set.complement();
expectRange((UnicodeString)"!complement(" + start + "," + end + ")",
set, xstart, xend);
set.complement(start);
}
const UChar32 DATA2[] = {
0,
0x10FFFF,
(UChar32)-1,
0x110000
};
const int32_t DATA2_LENGTH = sizeof(DATA2)/sizeof(DATA2[0]);
for (i=0; i<DATA2_LENGTH; ++i) {
UChar32 c = DATA2[i], end = 0x10FFFF;
UBool valid = (c >= 0 && c <= 0x10FFFF);
UnicodeSet set(0, 0x10FFFF);
// For single-codepoint contains, invalid codepoints are NOT contained
UBool b = set.contains(c);
if (b == valid) {
logln((UnicodeString)"[\\u0000-\\U0010FFFF].contains(" + c +
") = " + b);
} else {
errln((UnicodeString)"FAIL: [\\u0000-\\U0010FFFF].contains(" + c +
") = " + b);
}
// For codepoint range contains, containsNone, and containsSome,
// invalid or empty (start > end) ranges have UNDEFINED behavior.
b = set.contains(c, end);
logln((UnicodeString)"* [\\u0000-\\U0010FFFF].contains(" + c +
"," + end + ") = " + b);
b = set.containsNone(c, end);
logln((UnicodeString)"* [\\u0000-\\U0010FFFF].containsNone(" + c +
"," + end + ") = " + b);
b = set.containsSome(c, end);
logln((UnicodeString)"* [\\u0000-\\U0010FFFF].containsSome(" + c +
"," + end + ") = " + b);
int32_t index = set.indexOf(c);
if ((index >= 0) == valid) {
logln((UnicodeString)"[\\u0000-\\U0010FFFF].indexOf(" + c +
") = " + index);
} else {
errln((UnicodeString)"FAIL: [\\u0000-\\U0010FFFF].indexOf(" + c +
") = " + index);
}
}
}
// Used by TestSymbolTable
class TokenSymbolTable : public SymbolTable {
public:
Hashtable contents;
TokenSymbolTable(UErrorCode& ec) : contents(FALSE, ec) {
contents.setValueDeleter(uhash_deleteUnicodeString);
}
~TokenSymbolTable() {}
/**
* (Non-SymbolTable API) Add the given variable and value to
* the table. Variable should NOT contain leading '$'.
*/
void add(const UnicodeString& var, const UnicodeString& value,
UErrorCode& ec) {
if (U_SUCCESS(ec)) {
contents.put(var, new UnicodeString(value), ec);
}
}
/**
* SymbolTable API
*/
virtual const UnicodeString* lookup(const UnicodeString& s) const {
return (const UnicodeString*) contents.get(s);
}
/**
* SymbolTable API
*/
virtual const UnicodeFunctor* lookupMatcher(UChar32 /*ch*/) const {
return NULL;
}
/**
* SymbolTable API
*/
virtual UnicodeString parseReference(const UnicodeString& text,
ParsePosition& pos, int32_t limit) const {
int32_t start = pos.getIndex();
int32_t i = start;
UnicodeString result;
while (i < limit) {
UChar c = text.charAt(i);
if ((i==start && !u_isIDStart(c)) || !u_isIDPart(c)) {
break;
}
++i;
}
if (i == start) { // No valid name chars
return result; // Indicate failure with empty string
}
pos.setIndex(i);
text.extractBetween(start, i, result);
return result;
}
};
void UnicodeSetTest::TestSymbolTable() {
// Multiple test cases can be set up here. Each test case
// is terminated by null:
// var, value, var, value,..., input pat., exp. output pat., null
const char* DATA[] = {
"us", "a-z", "[0-1$us]", "[0-1a-z]", NULL,
"us", "[a-z]", "[0-1$us]", "[0-1[a-z]]", NULL,
"us", "\\[a\\-z\\]", "[0-1$us]", "[-01\\[\\]az]", NULL,
NULL
};
for (int32_t i=0; DATA[i]!=NULL; ++i) {
UErrorCode ec = U_ZERO_ERROR;
TokenSymbolTable sym(ec);
if (U_FAILURE(ec)) {
errln("FAIL: couldn't construct TokenSymbolTable");
continue;
}
// Set up variables
while (DATA[i+2] != NULL) {
sym.add(UnicodeString(DATA[i], -1, US_INV), UnicodeString(DATA[i+1], -1, US_INV), ec);
if (U_FAILURE(ec)) {
errln("FAIL: couldn't add to TokenSymbolTable");
continue;
}
i += 2;
}
// Input pattern and expected output pattern
UnicodeString inpat = UnicodeString(DATA[i], -1, US_INV), exppat = UnicodeString(DATA[i+1], -1, US_INV);
i += 2;
ParsePosition pos(0);
UnicodeSet us(inpat, pos, USET_IGNORE_SPACE, &sym, ec);
if (U_FAILURE(ec)) {
errln("FAIL: couldn't construct UnicodeSet");
continue;
}
// results
if (pos.getIndex() != inpat.length()) {
errln((UnicodeString)"Failed to read to end of string \""
+ inpat + "\": read to "
+ pos.getIndex() + ", length is "
+ inpat.length());
}
UnicodeSet us2(exppat, ec);
if (U_FAILURE(ec)) {
errln("FAIL: couldn't construct expected UnicodeSet");
continue;
}
UnicodeString a, b;
if (us != us2) {
errln((UnicodeString)"Failed, got " + us.toPattern(a, TRUE) +
", expected " + us2.toPattern(b, TRUE));
} else {
logln((UnicodeString)"Ok, got " + us.toPattern(a, TRUE));
}
}
}
void UnicodeSetTest::TestSurrogate() {
const char* DATA[] = {
// These should all behave identically
"[abc\\uD800\\uDC00]",
// "[abc\uD800\uDC00]", // Can't do this on C -- only Java
"[abc\\U00010000]",
0
};
for (int i=0; DATA[i] != 0; ++i) {
UErrorCode ec = U_ZERO_ERROR;
logln((UnicodeString)"Test pattern " + i + " :" + UnicodeString(DATA[i], -1, US_INV));
UnicodeString str = UnicodeString(DATA[i], -1, US_INV);
UnicodeSet set(str, ec);
if (U_FAILURE(ec)) {
errln("FAIL: UnicodeSet constructor");
continue;
}
expectContainment(set,
CharsToUnicodeString("abc\\U00010000"),
CharsToUnicodeString("\\uD800;\\uDC00")); // split apart surrogate-pair
if (set.size() != 4) {
errln((UnicodeString)"FAIL: " + UnicodeString(DATA[i], -1, US_INV) + ".size() == " +
set.size() + ", expected 4");
}
}
}
void UnicodeSetTest::TestExhaustive() {
// exhaustive tests. Simulate UnicodeSets with integers.
// That gives us very solid tests (except for large memory tests).
int32_t limit = 128;
UnicodeSet x, y, z, aa;
for (int32_t i = 0; i < limit; ++i) {
bitsToSet(i, x);
logln((UnicodeString)"Testing " + i + ", " + x);
_testComplement(i, x, y);
// AS LONG AS WE ARE HERE, check roundtrip
checkRoundTrip(bitsToSet(i, aa));
for (int32_t j = 0; j < limit; ++j) {
_testAdd(i,j, x,y,z);
_testXor(i,j, x,y,z);
_testRetain(i,j, x,y,z);
_testRemove(i,j, x,y,z);
}
}
}
void UnicodeSetTest::_testComplement(int32_t a, UnicodeSet& x, UnicodeSet& z) {
bitsToSet(a, x);
z = x;
z.complement();
int32_t c = setToBits(z);
if (c != (~a)) {
errln((UnicodeString)"FAILED: add: ~" + x + " != " + z);
errln((UnicodeString)"FAILED: add: ~" + a + " != " + c);
}
checkCanonicalRep(z, (UnicodeString)"complement " + a);
}
void UnicodeSetTest::_testAdd(int32_t a, int32_t b, UnicodeSet& x, UnicodeSet& y, UnicodeSet& z) {
bitsToSet(a, x);
bitsToSet(b, y);
z = x;
z.addAll(y);
int32_t c = setToBits(z);
if (c != (a | b)) {
errln((UnicodeString)"FAILED: add: " + x + " | " + y + " != " + z);
errln((UnicodeString)"FAILED: add: " + a + " | " + b + " != " + c);
}
checkCanonicalRep(z, (UnicodeString)"add " + a + "," + b);
}
void UnicodeSetTest::_testRetain(int32_t a, int32_t b, UnicodeSet& x, UnicodeSet& y, UnicodeSet& z) {
bitsToSet(a, x);
bitsToSet(b, y);
z = x;
z.retainAll(y);
int32_t c = setToBits(z);
if (c != (a & b)) {
errln((UnicodeString)"FAILED: retain: " + x + " & " + y + " != " + z);
errln((UnicodeString)"FAILED: retain: " + a + " & " + b + " != " + c);
}
checkCanonicalRep(z, (UnicodeString)"retain " + a + "," + b);
}
void UnicodeSetTest::_testRemove(int32_t a, int32_t b, UnicodeSet& x, UnicodeSet& y, UnicodeSet& z) {
bitsToSet(a, x);
bitsToSet(b, y);
z = x;
z.removeAll(y);
int32_t c = setToBits(z);
if (c != (a &~ b)) {
errln((UnicodeString)"FAILED: remove: " + x + " &~ " + y + " != " + z);
errln((UnicodeString)"FAILED: remove: " + a + " &~ " + b + " != " + c);
}
checkCanonicalRep(z, (UnicodeString)"remove " + a + "," + b);
}
void UnicodeSetTest::_testXor(int32_t a, int32_t b, UnicodeSet& x, UnicodeSet& y, UnicodeSet& z) {
bitsToSet(a, x);
bitsToSet(b, y);
z = x;
z.complementAll(y);
int32_t c = setToBits(z);
if (c != (a ^ b)) {
errln((UnicodeString)"FAILED: complement: " + x + " ^ " + y + " != " + z);
errln((UnicodeString)"FAILED: complement: " + a + " ^ " + b + " != " + c);
}
checkCanonicalRep(z, (UnicodeString)"complement " + a + "," + b);
}
/**
* Check that ranges are monotonically increasing and non-
* overlapping.
*/
void UnicodeSetTest::checkCanonicalRep(const UnicodeSet& set, const UnicodeString& msg) {
int32_t n = set.getRangeCount();
if (n < 0) {
errln((UnicodeString)"FAIL result of " + msg +
": range count should be >= 0 but is " +
n /*+ " for " + set.toPattern())*/);
return;
}
UChar32 last = 0;
for (int32_t i=0; i<n; ++i) {
UChar32 start = set.getRangeStart(i);
UChar32 end = set.getRangeEnd(i);
if (start > end) {
errln((UnicodeString)"FAIL result of " + msg +
": range " + (i+1) +
" start > end: " + (int)start + ", " + (int)end +
" for " + set);
}
if (i > 0 && start <= last) {
errln((UnicodeString)"FAIL result of " + msg +
": range " + (i+1) +
" overlaps previous range: " + (int)start + ", " + (int)end +
" for " + set);
}
last = end;
}
}
/**
* Convert a bitmask to a UnicodeSet.
*/
UnicodeSet& UnicodeSetTest::bitsToSet(int32_t a, UnicodeSet& result) {
result.clear();
for (UChar32 i = 0; i < 32; ++i) {
if ((a & (1<<i)) != 0) {
result.add(i);
}
}
return result;
}
/**
* Convert a UnicodeSet to a bitmask. Only the characters
* U+0000 to U+0020 are represented in the bitmask.
*/
int32_t UnicodeSetTest::setToBits(const UnicodeSet& x) {
int32_t result = 0;
for (int32_t i = 0; i < 32; ++i) {
if (x.contains((UChar32)i)) {
result |= (1<<i);
}
}
return result;
}
/**
* Return the representation of an inversion list based UnicodeSet
* as a pairs list. Ranges are listed in ascending Unicode order.
* For example, the set [a-zA-M3] is represented as "33AMaz".
*/
UnicodeString UnicodeSetTest::getPairs(const UnicodeSet& set) {
UnicodeString pairs;
for (int32_t i=0; i<set.getRangeCount(); ++i) {
UChar32 start = set.getRangeStart(i);
UChar32 end = set.getRangeEnd(i);
if (end > 0xFFFF) {
end = 0xFFFF;
i = set.getRangeCount(); // Should be unnecessary
}
pairs.append((UChar)start).append((UChar)end);
}
return pairs;
}
/**
* Basic consistency check for a few items.
* That the iterator works, and that we can create a pattern and
* get the same thing back
*/
void UnicodeSetTest::checkRoundTrip(const UnicodeSet& s) {
UErrorCode ec = U_ZERO_ERROR;
UnicodeSet t(s);
checkEqual(s, t, "copy ct");
t = s;
checkEqual(s, t, "operator=");
copyWithIterator(t, s, FALSE);
checkEqual(s, t, "iterator roundtrip");
copyWithIterator(t, s, TRUE); // try range
checkEqual(s, t, "iterator roundtrip");
UnicodeString pat; s.toPattern(pat, FALSE);
t.applyPattern(pat, ec);
if (U_FAILURE(ec)) {
errln("FAIL: applyPattern");
return;
} else {
checkEqual(s, t, "toPattern(false)");
}
s.toPattern(pat, TRUE);
t.applyPattern(pat, ec);
if (U_FAILURE(ec)) {
errln("FAIL: applyPattern");
return;
} else {
checkEqual(s, t, "toPattern(true)");
}
}
void UnicodeSetTest::copyWithIterator(UnicodeSet& t, const UnicodeSet& s, UBool withRange) {
t.clear();
UnicodeSetIterator it(s);
if (withRange) {
while (it.nextRange()) {
if (it.isString()) {
t.add(it.getString());
} else {
t.add(it.getCodepoint(), it.getCodepointEnd());
}
}
} else {
while (it.next()) {
if (it.isString()) {
t.add(it.getString());
} else {
t.add(it.getCodepoint());
}
}
}
}
UBool UnicodeSetTest::checkEqual(const UnicodeSet& s, const UnicodeSet& t, const char* message) {
UnicodeString source; s.toPattern(source, TRUE);
UnicodeString result; t.toPattern(result, TRUE);
if (s != t) {
errln((UnicodeString)"FAIL: " + message
+ "; source = " + source
+ "; result = " + result
);
return FALSE;
} else {
logln((UnicodeString)"Ok: " + message
+ "; source = " + source
+ "; result = " + result
);
}
return TRUE;
}
void
UnicodeSetTest::expectContainment(const UnicodeString& pat,
const UnicodeString& charsIn,
const UnicodeString& charsOut) {
UErrorCode ec = U_ZERO_ERROR;
UnicodeSet set(pat, ec);
if (U_FAILURE(ec)) {
dataerrln((UnicodeString)"FAIL: pattern \"" +
pat + "\" => " + u_errorName(ec));
return;
}
expectContainment(set, pat, charsIn, charsOut);
}
void
UnicodeSetTest::expectContainment(const UnicodeSet& set,
const UnicodeString& charsIn,
const UnicodeString& charsOut) {
UnicodeString pat;
set.toPattern(pat);
expectContainment(set, pat, charsIn, charsOut);
}
void
UnicodeSetTest::expectContainment(const UnicodeSet& set,
const UnicodeString& setName,
const UnicodeString& charsIn,
const UnicodeString& charsOut) {
UnicodeString bad;
UChar32 c;
int32_t i;
for (i=0; i<charsIn.length(); i+=U16_LENGTH(c)) {
c = charsIn.char32At(i);
if (!set.contains(c)) {
bad.append(c);
}
}
if (bad.length() > 0) {
errln((UnicodeString)"Fail: set " + setName + " does not contain " + prettify(bad) +
", expected containment of " + prettify(charsIn));
} else {
logln((UnicodeString)"Ok: set " + setName + " contains " + prettify(charsIn));
}
bad.truncate(0);
for (i=0; i<charsOut.length(); i+=U16_LENGTH(c)) {
c = charsOut.char32At(i);
if (set.contains(c)) {
bad.append(c);
}
}
if (bad.length() > 0) {
errln((UnicodeString)"Fail: set " + setName + " contains " + prettify(bad) +
", expected non-containment of " + prettify(charsOut));
} else {
logln((UnicodeString)"Ok: set " + setName + " does not contain " + prettify(charsOut));
}
}
void
UnicodeSetTest::expectPattern(UnicodeSet& set,
const UnicodeString& pattern,
const UnicodeString& expectedPairs){
UErrorCode status = U_ZERO_ERROR;
set.applyPattern(pattern, status);
if (U_FAILURE(status)) {
errln(UnicodeString("FAIL: applyPattern(\"") + pattern +
"\") failed");
return;
} else {
if (getPairs(set) != expectedPairs ) {
errln(UnicodeString("FAIL: applyPattern(\"") + pattern +
"\") => pairs \"" +
escape(getPairs(set)) + "\", expected \"" +
escape(expectedPairs) + "\"");
} else {
logln(UnicodeString("Ok: applyPattern(\"") + pattern +
"\") => pairs \"" +
escape(getPairs(set)) + "\"");
}
}
// the result of calling set.toPattern(), which is the string representation of
// this set(set), is passed to a UnicodeSet constructor, and tested that it
// will produce another set that is equal to this one.
UnicodeString temppattern;
set.toPattern(temppattern);
UnicodeSet *tempset=new UnicodeSet(temppattern, status);
if (U_FAILURE(status)) {
errln(UnicodeString("FAIL: applyPattern(\""+ pattern + "\").toPattern() => " + temppattern + " => invalid pattern"));
return;
}
if(*tempset != set || getPairs(*tempset) != getPairs(set)){
errln(UnicodeString("FAIL: applyPattern(\""+ pattern + "\").toPattern() => " + temppattern + " => pairs \""+ escape(getPairs(*tempset)) + "\", expected pairs \"" +
escape(getPairs(set)) + "\""));
} else{
logln(UnicodeString("Ok: applyPattern(\""+ pattern + "\").toPattern() => " + temppattern + " => pairs \"" + escape(getPairs(*tempset)) + "\""));
}
delete tempset;
}
void
UnicodeSetTest::expectPairs(const UnicodeSet& set, const UnicodeString& expectedPairs) {
if (getPairs(set) != expectedPairs) {
errln(UnicodeString("FAIL: Expected pair list \"") +
escape(expectedPairs) + "\", got \"" +
escape(getPairs(set)) + "\"");
}
}
void UnicodeSetTest::expectToPattern(const UnicodeSet& set,
const UnicodeString& expPat,
const char** expStrings) {
UnicodeString pat;
set.toPattern(pat, TRUE);
if (pat == expPat) {
logln((UnicodeString)"Ok: toPattern() => \"" + pat + "\"");
} else {
errln((UnicodeString)"FAIL: toPattern() => \"" + pat + "\", expected \"" + expPat + "\"");
return;
}
if (expStrings == NULL) {
return;
}
UBool in = TRUE;
for (int32_t i=0; expStrings[i] != NULL; ++i) {
if (expStrings[i] == NOT) { // sic; pointer comparison
in = FALSE;
continue;
}
UnicodeString s = CharsToUnicodeString(expStrings[i]);
UBool contained = set.contains(s);
if (contained == in) {
logln((UnicodeString)"Ok: " + expPat +
(contained ? " contains {" : " does not contain {") +
escape(expStrings[i]) + "}");
} else {
errln((UnicodeString)"FAIL: " + expPat +
(contained ? " contains {" : " does not contain {") +
escape(expStrings[i]) + "}");
}
}
}
static UChar toHexString(int32_t i) { return (UChar)(i + (i < 10 ? 0x30 : (0x41 - 10))); }
void
UnicodeSetTest::doAssert(UBool condition, const char *message)
{
if (!condition) {
errln(UnicodeString("ERROR : ") + message);
}
}
UnicodeString
UnicodeSetTest::escape(const UnicodeString& s) {
UnicodeString buf;
for (int32_t i=0; i<s.length(); )
{
UChar32 c = s.char32At(i);
if (0x0020 <= c && c <= 0x007F) {
buf += c;
} else {
if (c <= 0xFFFF) {
buf += (UChar)0x5c; buf += (UChar)0x75;
} else {
buf += (UChar)0x5c; buf += (UChar)0x55;
buf += toHexString((c & 0xF0000000) >> 28);
buf += toHexString((c & 0x0F000000) >> 24);
buf += toHexString((c & 0x00F00000) >> 20);
buf += toHexString((c & 0x000F0000) >> 16);
}
buf += toHexString((c & 0xF000) >> 12);
buf += toHexString((c & 0x0F00) >> 8);
buf += toHexString((c & 0x00F0) >> 4);
buf += toHexString(c & 0x000F);
}
i += U16_LENGTH(c);
}
return buf;
}
void UnicodeSetTest::TestFreezable() {
UErrorCode errorCode=U_ZERO_ERROR;
UnicodeString idPattern=UNICODE_STRING("[:ID_Continue:]", 15);
UnicodeSet idSet(idPattern, errorCode);
if(U_FAILURE(errorCode)) {
dataerrln("FAIL: unable to create UnicodeSet([:ID_Continue:]) - %s", u_errorName(errorCode));
return;
}
UnicodeString wsPattern=UNICODE_STRING("[:White_Space:]", 15);
UnicodeSet wsSet(wsPattern, errorCode);
if(U_FAILURE(errorCode)) {
dataerrln("FAIL: unable to create UnicodeSet([:White_Space:]) - %s", u_errorName(errorCode));
return;
}
idSet.add(idPattern);
UnicodeSet frozen(idSet);
frozen.freeze();
if(idSet.isFrozen() || !frozen.isFrozen()) {
errln("FAIL: isFrozen() is wrong");
}
if(frozen!=idSet || !(frozen==idSet)) {
errln("FAIL: a copy-constructed frozen set differs from its original");
}
frozen=wsSet;
if(frozen!=idSet || !(frozen==idSet)) {
errln("FAIL: a frozen set was modified by operator=");
}
UnicodeSet frozen2(frozen);
if(frozen2!=frozen || frozen2!=idSet) {
errln("FAIL: a copied frozen set differs from its frozen original");
}
if(!frozen2.isFrozen()) {
errln("FAIL: copy-constructing a frozen set results in a thawed one");
}
UnicodeSet frozen3(5, 55); // Set to some values to really test assignment below, not copy construction.
if(frozen3.contains(0, 4) || !frozen3.contains(5, 55) || frozen3.contains(56, 0x10ffff)) {
errln("FAIL: UnicodeSet(5, 55) failed");
}
frozen3=frozen;
if(!frozen3.isFrozen()) {
errln("FAIL: copying a frozen set results in a thawed one");
}
UnicodeSet *cloned=(UnicodeSet *)frozen.clone();
if(!cloned->isFrozen() || *cloned!=frozen || cloned->containsSome(0xd802, 0xd805)) {
errln("FAIL: clone() failed");
}
cloned->add(0xd802, 0xd805);
if(cloned->containsSome(0xd802, 0xd805)) {
errln("FAIL: unable to modify clone");
}
delete cloned;
UnicodeSet *thawed=(UnicodeSet *)frozen.cloneAsThawed();
if(thawed->isFrozen() || *thawed!=frozen || thawed->containsSome(0xd802, 0xd805)) {
errln("FAIL: cloneAsThawed() failed");
}
thawed->add(0xd802, 0xd805);
if(!thawed->contains(0xd802, 0xd805)) {
errln("FAIL: unable to modify thawed clone");
}
delete thawed;
frozen.set(5, 55);
if(frozen!=idSet || !(frozen==idSet)) {
errln("FAIL: UnicodeSet::set() modified a frozen set");
}
frozen.clear();
if(frozen!=idSet || !(frozen==idSet)) {
errln("FAIL: UnicodeSet::clear() modified a frozen set");
}
frozen.closeOver(USET_CASE_INSENSITIVE);
if(frozen!=idSet || !(frozen==idSet)) {
errln("FAIL: UnicodeSet::closeOver() modified a frozen set");
}
frozen.compact();
if(frozen!=idSet || !(frozen==idSet)) {
errln("FAIL: UnicodeSet::compact() modified a frozen set");
}
ParsePosition pos;
frozen.
applyPattern(wsPattern, errorCode).
applyPattern(wsPattern, USET_IGNORE_SPACE, NULL, errorCode).
applyPattern(wsPattern, pos, USET_IGNORE_SPACE, NULL, errorCode).
applyIntPropertyValue(UCHAR_CANONICAL_COMBINING_CLASS, 230, errorCode).
applyPropertyAlias(UNICODE_STRING_SIMPLE("Assigned"), UnicodeString(), errorCode);
if(frozen!=idSet || !(frozen==idSet)) {
errln("FAIL: UnicodeSet::applyXYZ() modified a frozen set");
}
frozen.
add(0xd800).
add(0xd802, 0xd805).
add(wsPattern).
addAll(idPattern).
addAll(wsSet);
if(frozen!=idSet || !(frozen==idSet)) {
errln("FAIL: UnicodeSet::addXYZ() modified a frozen set");
}
frozen.
retain(0x62).
retain(0x64, 0x69).
retainAll(wsPattern).
retainAll(wsSet);
if(frozen!=idSet || !(frozen==idSet)) {
errln("FAIL: UnicodeSet::retainXYZ() modified a frozen set");
}
frozen.
remove(0x62).
remove(0x64, 0x69).
remove(idPattern).
removeAll(idPattern).
removeAll(idSet);
if(frozen!=idSet || !(frozen==idSet)) {
errln("FAIL: UnicodeSet::removeXYZ() modified a frozen set");
}
frozen.
complement().
complement(0x62).
complement(0x64, 0x69).
complement(idPattern).
complementAll(idPattern).
complementAll(idSet);
if(frozen!=idSet || !(frozen==idSet)) {
errln("FAIL: UnicodeSet::complementXYZ() modified a frozen set");
}
}
// Test span() etc. -------------------------------------------------------- ***
// Append the UTF-8 version of the string to t and return the appended UTF-8 length.
static int32_t
appendUTF8(const UChar *s, int32_t length, char *t, int32_t capacity) {
UErrorCode errorCode=U_ZERO_ERROR;
int32_t length8=0;
u_strToUTF8(t, capacity, &length8, s, length, &errorCode);
if(U_SUCCESS(errorCode)) {
return length8;
} else {
// The string contains an unpaired surrogate.
// Ignore this string.
return 0;
}
}
class UnicodeSetWithStringsIterator;
// Make the strings in a UnicodeSet easily accessible.
class UnicodeSetWithStrings {
public:
UnicodeSetWithStrings(const UnicodeSet &normalSet) :
set(normalSet), stringsLength(0), hasSurrogates(FALSE) {
int32_t size=set.size();
if(size>0 && set.charAt(size-1)<0) {
// If a set's last element is not a code point, then it must contain strings.
// Iterate over the set, skip all code point ranges, and cache the strings.
// Convert them to UTF-8 for spanUTF8().
UnicodeSetIterator iter(set);
const UnicodeString *s;
char *s8=utf8;
int32_t length8, utf8Count=0;
while(iter.nextRange() && stringsLength<LENGTHOF(strings)) {
if(iter.isString()) {
// Store the pointer to the set's string element
// which we happen to know is a stable pointer.
strings[stringsLength]=s=&iter.getString();
utf8Count+=
utf8Lengths[stringsLength]=length8=
appendUTF8(s->getBuffer(), s->length(),
s8, (int32_t)(sizeof(utf8)-utf8Count));
if(length8==0) {
hasSurrogates=TRUE; // Contains unpaired surrogates.
}
s8+=length8;
++stringsLength;
}
}
}
}
const UnicodeSet &getSet() const {
return set;
}
UBool hasStrings() const {
return (UBool)(stringsLength>0);
}
UBool hasStringsWithSurrogates() const {
return hasSurrogates;
}
private:
friend class UnicodeSetWithStringsIterator;
const UnicodeSet &set;
const UnicodeString *strings[20];
int32_t stringsLength;
UBool hasSurrogates;
char utf8[1024];
int32_t utf8Lengths[20];
int32_t nextStringIndex;
int32_t nextUTF8Start;
};
class UnicodeSetWithStringsIterator {
public:
UnicodeSetWithStringsIterator(const UnicodeSetWithStrings &set) :
fSet(set), nextStringIndex(0), nextUTF8Start(0) {
}
void reset() {
nextStringIndex=nextUTF8Start=0;
}
const UnicodeString *nextString() {
if(nextStringIndex<fSet.stringsLength) {
return fSet.strings[nextStringIndex++];
} else {
return NULL;
}
}
// Do not mix with calls to nextString().
const char *nextUTF8(int32_t &length) {
if(nextStringIndex<fSet.stringsLength) {
const char *s8=fSet.utf8+nextUTF8Start;
nextUTF8Start+=length=fSet.utf8Lengths[nextStringIndex++];
return s8;
} else {
length=0;
return NULL;
}
}
private:
const UnicodeSetWithStrings &fSet;
int32_t nextStringIndex;
int32_t nextUTF8Start;
};
// Compare 16-bit Unicode strings (which may be malformed UTF-16)
// at code point boundaries.
// That is, each edge of a match must not be in the middle of a surrogate pair.
static inline UBool
matches16CPB(const UChar *s, int32_t start, int32_t limit, const UnicodeString &t) {
s+=start;
limit-=start;
int32_t length=t.length();
return 0==t.compare(s, length) &&
!(0<start && U16_IS_LEAD(s[-1]) && U16_IS_TRAIL(s[0])) &&
!(length<limit && U16_IS_LEAD(s[length-1]) && U16_IS_TRAIL(s[length]));
}
// Implement span() with contains() for comparison.
static int32_t containsSpanUTF16(const UnicodeSetWithStrings &set, const UChar *s, int32_t length,
USetSpanCondition spanCondition) {
const UnicodeSet &realSet(set.getSet());
if(!set.hasStrings()) {
if(spanCondition!=USET_SPAN_NOT_CONTAINED) {
spanCondition=USET_SPAN_CONTAINED; // Pin to 0/1 values.
}
UChar32 c;
int32_t start=0, prev;
while((prev=start)<length) {
U16_NEXT(s, start, length, c);
if(realSet.contains(c)!=spanCondition) {
break;
}
}
return prev;
} else if(spanCondition==USET_SPAN_NOT_CONTAINED) {
UnicodeSetWithStringsIterator iter(set);
UChar32 c;
int32_t start, next;
for(start=next=0; start<length;) {
U16_NEXT(s, next, length, c);
if(realSet.contains(c)) {
break;
}
const UnicodeString *str;
iter.reset();
while((str=iter.nextString())!=NULL) {
if(str->length()<=(length-start) && matches16CPB(s, start, length, *str)) {
// spanNeedsStrings=TRUE;
return start;
}
}
start=next;
}
return start;
} else /* USET_SPAN_CONTAINED or USET_SPAN_SIMPLE */ {
UnicodeSetWithStringsIterator iter(set);
UChar32 c;
int32_t start, next, maxSpanLimit=0;
for(start=next=0; start<length;) {
U16_NEXT(s, next, length, c);
if(!realSet.contains(c)) {
next=start; // Do not span this single, not-contained code point.
}
const UnicodeString *str;
iter.reset();
while((str=iter.nextString())!=NULL) {
if(str->length()<=(length-start) && matches16CPB(s, start, length, *str)) {
// spanNeedsStrings=TRUE;
int32_t matchLimit=start+str->length();
if(matchLimit==length) {
return length;
}
if(spanCondition==USET_SPAN_CONTAINED) {
// Iterate for the shortest match at each position.
// Recurse for each but the shortest match.
if(next==start) {
next=matchLimit; // First match from start.
} else {
if(matchLimit<next) {
// Remember shortest match from start for iteration.
int32_t temp=next;
next=matchLimit;
matchLimit=temp;
}
// Recurse for non-shortest match from start.
int32_t spanLength=containsSpanUTF16(set, s+matchLimit, length-matchLimit,
USET_SPAN_CONTAINED);
if((matchLimit+spanLength)>maxSpanLimit) {
maxSpanLimit=matchLimit+spanLength;
if(maxSpanLimit==length) {
return length;
}
}
}
} else /* spanCondition==USET_SPAN_SIMPLE */ {
if(matchLimit>next) {
// Remember longest match from start.
next=matchLimit;
}
}
}
}
if(next==start) {
break; // No match from start.
}
start=next;
}
if(start>maxSpanLimit) {
return start;
} else {
return maxSpanLimit;
}
}
}
static int32_t containsSpanBackUTF16(const UnicodeSetWithStrings &set, const UChar *s, int32_t length,
USetSpanCondition spanCondition) {
if(length==0) {
return 0;
}
const UnicodeSet &realSet(set.getSet());
if(!set.hasStrings()) {
if(spanCondition!=USET_SPAN_NOT_CONTAINED) {
spanCondition=USET_SPAN_CONTAINED; // Pin to 0/1 values.
}
UChar32 c;
int32_t prev=length;
do {
U16_PREV(s, 0, length, c);
if(realSet.contains(c)!=spanCondition) {
break;
}
} while((prev=length)>0);
return prev;
} else if(spanCondition==USET_SPAN_NOT_CONTAINED) {
UnicodeSetWithStringsIterator iter(set);
UChar32 c;
int32_t prev=length, length0=length;
do {
U16_PREV(s, 0, length, c);
if(realSet.contains(c)) {
break;
}
const UnicodeString *str;
iter.reset();
while((str=iter.nextString())!=NULL) {
if(str->length()<=prev && matches16CPB(s, prev-str->length(), length0, *str)) {
// spanNeedsStrings=TRUE;
return prev;
}
}
} while((prev=length)>0);
return prev;
} else /* USET_SPAN_CONTAINED or USET_SPAN_SIMPLE */ {
UnicodeSetWithStringsIterator iter(set);
UChar32 c;
int32_t prev=length, minSpanStart=length, length0=length;
do {
U16_PREV(s, 0, length, c);
if(!realSet.contains(c)) {
length=prev; // Do not span this single, not-contained code point.
}
const UnicodeString *str;
iter.reset();
while((str=iter.nextString())!=NULL) {
if(str->length()<=prev && matches16CPB(s, prev-str->length(), length0, *str)) {
// spanNeedsStrings=TRUE;
int32_t matchStart=prev-str->length();
if(matchStart==0) {
return 0;
}
if(spanCondition==USET_SPAN_CONTAINED) {
// Iterate for the shortest match at each position.
// Recurse for each but the shortest match.
if(length==prev) {
length=matchStart; // First match from prev.
} else {
if(matchStart>length) {
// Remember shortest match from prev for iteration.
int32_t temp=length;
length=matchStart;
matchStart=temp;
}
// Recurse for non-shortest match from prev.
int32_t spanStart=containsSpanBackUTF16(set, s, matchStart,
USET_SPAN_CONTAINED);
if(spanStart<minSpanStart) {
minSpanStart=spanStart;
if(minSpanStart==0) {
return 0;
}
}
}
} else /* spanCondition==USET_SPAN_SIMPLE */ {
if(matchStart<length) {
// Remember longest match from prev.
length=matchStart;
}
}
}
}
if(length==prev) {
break; // No match from prev.
}
} while((prev=length)>0);
if(prev<minSpanStart) {
return prev;
} else {
return minSpanStart;
}
}
}
static int32_t containsSpanUTF8(const UnicodeSetWithStrings &set, const char *s, int32_t length,
USetSpanCondition spanCondition) {
const UnicodeSet &realSet(set.getSet());
if(!set.hasStrings()) {
if(spanCondition!=USET_SPAN_NOT_CONTAINED) {
spanCondition=USET_SPAN_CONTAINED; // Pin to 0/1 values.
}
UChar32 c;
int32_t start=0, prev;
while((prev=start)<length) {
U8_NEXT(s, start, length, c);
if(c<0) {
c=0xfffd;
}
if(realSet.contains(c)!=spanCondition) {
break;
}
}
return prev;
} else if(spanCondition==USET_SPAN_NOT_CONTAINED) {
UnicodeSetWithStringsIterator iter(set);
UChar32 c;
int32_t start, next;
for(start=next=0; start<length;) {
U8_NEXT(s, next, length, c);
if(c<0) {
c=0xfffd;
}
if(realSet.contains(c)) {
break;
}
const char *s8;
int32_t length8;
iter.reset();
while((s8=iter.nextUTF8(length8))!=NULL) {
if(length8!=0 && length8<=(length-start) && 0==memcmp(s+start, s8, length8)) {
// spanNeedsStrings=TRUE;
return start;
}
}
start=next;
}
return start;
} else /* USET_SPAN_CONTAINED or USET_SPAN_SIMPLE */ {
UnicodeSetWithStringsIterator iter(set);
UChar32 c;
int32_t start, next, maxSpanLimit=0;
for(start=next=0; start<length;) {
U8_NEXT(s, next, length, c);
if(c<0) {
c=0xfffd;
}
if(!realSet.contains(c)) {
next=start; // Do not span this single, not-contained code point.
}
const char *s8;
int32_t length8;
iter.reset();
while((s8=iter.nextUTF8(length8))!=NULL) {
if(length8!=0 && length8<=(length-start) && 0==memcmp(s+start, s8, length8)) {
// spanNeedsStrings=TRUE;
int32_t matchLimit=start+length8;
if(matchLimit==length) {
return length;
}
if(spanCondition==USET_SPAN_CONTAINED) {
// Iterate for the shortest match at each position.
// Recurse for each but the shortest match.
if(next==start) {
next=matchLimit; // First match from start.
} else {
if(matchLimit<next) {
// Remember shortest match from start for iteration.
int32_t temp=next;
next=matchLimit;
matchLimit=temp;
}
// Recurse for non-shortest match from start.
int32_t spanLength=containsSpanUTF8(set, s+matchLimit, length-matchLimit,
USET_SPAN_CONTAINED);
if((matchLimit+spanLength)>maxSpanLimit) {
maxSpanLimit=matchLimit+spanLength;
if(maxSpanLimit==length) {
return length;
}
}
}
} else /* spanCondition==USET_SPAN_SIMPLE */ {
if(matchLimit>next) {
// Remember longest match from start.
next=matchLimit;
}
}
}
}
if(next==start) {
break; // No match from start.
}
start=next;
}
if(start>maxSpanLimit) {
return start;
} else {
return maxSpanLimit;
}
}
}
static int32_t containsSpanBackUTF8(const UnicodeSetWithStrings &set, const char *s, int32_t length,
USetSpanCondition spanCondition) {
if(length==0) {
return 0;
}
const UnicodeSet &realSet(set.getSet());
if(!set.hasStrings()) {
if(spanCondition!=USET_SPAN_NOT_CONTAINED) {
spanCondition=USET_SPAN_CONTAINED; // Pin to 0/1 values.
}
UChar32 c;
int32_t prev=length;
do {
U8_PREV(s, 0, length, c);
if(c<0) {
c=0xfffd;
}
if(realSet.contains(c)!=spanCondition) {
break;
}
} while((prev=length)>0);
return prev;
} else if(spanCondition==USET_SPAN_NOT_CONTAINED) {
UnicodeSetWithStringsIterator iter(set);
UChar32 c;
int32_t prev=length;
do {
U8_PREV(s, 0, length, c);
if(c<0) {
c=0xfffd;
}
if(realSet.contains(c)) {
break;
}
const char *s8;
int32_t length8;
iter.reset();
while((s8=iter.nextUTF8(length8))!=NULL) {
if(length8!=0 && length8<=prev && 0==memcmp(s+prev-length8, s8, length8)) {
// spanNeedsStrings=TRUE;
return prev;
}
}
} while((prev=length)>0);
return prev;
} else /* USET_SPAN_CONTAINED or USET_SPAN_SIMPLE */ {
UnicodeSetWithStringsIterator iter(set);
UChar32 c;
int32_t prev=length, minSpanStart=length;
do {
U8_PREV(s, 0, length, c);
if(c<0) {
c=0xfffd;
}
if(!realSet.contains(c)) {
length=prev; // Do not span this single, not-contained code point.
}
const char *s8;
int32_t length8;
iter.reset();
while((s8=iter.nextUTF8(length8))!=NULL) {
if(length8!=0 && length8<=prev && 0==memcmp(s+prev-length8, s8, length8)) {
// spanNeedsStrings=TRUE;
int32_t matchStart=prev-length8;
if(matchStart==0) {
return 0;
}
if(spanCondition==USET_SPAN_CONTAINED) {
// Iterate for the shortest match at each position.
// Recurse for each but the shortest match.
if(length==prev) {
length=matchStart; // First match from prev.
} else {
if(matchStart>length) {
// Remember shortest match from prev for iteration.
int32_t temp=length;
length=matchStart;
matchStart=temp;
}
// Recurse for non-shortest match from prev.
int32_t spanStart=containsSpanBackUTF8(set, s, matchStart,
USET_SPAN_CONTAINED);
if(spanStart<minSpanStart) {
minSpanStart=spanStart;
if(minSpanStart==0) {
return 0;
}
}
}
} else /* spanCondition==USET_SPAN_SIMPLE */ {
if(matchStart<length) {
// Remember longest match from prev.
length=matchStart;
}
}
}
}
if(length==prev) {
break; // No match from prev.
}
} while((prev=length)>0);
if(prev<minSpanStart) {
return prev;
} else {
return minSpanStart;
}
}
}
// spans to be performed and compared
enum {
SPAN_UTF16 =1,
SPAN_UTF8 =2,
SPAN_UTFS =3,
SPAN_SET =4,
SPAN_COMPLEMENT =8,
SPAN_POLARITY =0xc,
SPAN_FWD =0x10,
SPAN_BACK =0x20,
SPAN_DIRS =0x30,
SPAN_CONTAINED =0x100,
SPAN_SIMPLE =0x200,
SPAN_CONDITION =0x300,
SPAN_ALL =0x33f
};
static inline USetSpanCondition invertSpanCondition(USetSpanCondition spanCondition, USetSpanCondition contained) {
return spanCondition == USET_SPAN_NOT_CONTAINED ? contained : USET_SPAN_NOT_CONTAINED;
}
static inline int32_t slen(const void *s, UBool isUTF16) {
return isUTF16 ? u_strlen((const UChar *)s) : strlen((const char *)s);
}
/*
* Count spans on a string with the method according to type and set the span limits.
* The set may be the complement of the original.
* When using spanBack() and comparing with span(), use a span condition for the first spanBack()
* according to the expected number of spans.
* Sets typeName to an empty string if there is no such type.
* Returns -1 if the span option is filtered out.
*/
static int32_t getSpans(const UnicodeSetWithStrings &set, UBool isComplement,
const void *s, int32_t length, UBool isUTF16,
uint32_t whichSpans,
int type, const char *&typeName,
int32_t limits[], int32_t limitsCapacity,
int32_t expectCount) {
const UnicodeSet &realSet(set.getSet());
int32_t start, count;
USetSpanCondition spanCondition, firstSpanCondition, contained;
UBool isForward;
if(type<0 || 7<type) {
typeName="";
return 0;
}
static const char *const typeNames16[]={
"contains", "contains(LM)",
"span", "span(LM)",
"containsBack", "containsBack(LM)",
"spanBack", "spanBack(LM)"
};
static const char *const typeNames8[]={
"containsUTF8", "containsUTF8(LM)",
"spanUTF8", "spanUTF8(LM)",
"containsBackUTF8", "containsBackUTF8(LM)", // not implemented
"spanBackUTF8", "spanBackUTF8(LM)"
};
typeName= isUTF16 ? typeNames16[type] : typeNames8[type];
// filter span options
if(type<=3) {
// span forward
if((whichSpans&SPAN_FWD)==0) {
return -1;
}
isForward=TRUE;
} else {
// span backward
if((whichSpans&SPAN_BACK)==0) {
return -1;
}
isForward=FALSE;
}
if((type&1)==0) {
// use USET_SPAN_CONTAINED
if((whichSpans&SPAN_CONTAINED)==0) {
return -1;
}
contained=USET_SPAN_CONTAINED;
} else {
// use USET_SPAN_SIMPLE
if((whichSpans&SPAN_SIMPLE)==0) {
return -1;
}
contained=USET_SPAN_SIMPLE;
}
// Default first span condition for going forward with an uncomplemented set.
spanCondition=USET_SPAN_NOT_CONTAINED;
if(isComplement) {
spanCondition=invertSpanCondition(spanCondition, contained);
}
// First span condition for span(), used to terminate the spanBack() iteration.
firstSpanCondition=spanCondition;
// spanBack(): Its initial span condition is span()'s last span condition,
// which is the opposite of span()'s first span condition
// if we expect an even number of spans.
// (The loop inverts spanCondition (expectCount-1) times
// before the expectCount'th span() call.)
// If we do not compare forward and backward directions, then we do not have an
// expectCount and just start with firstSpanCondition.
if(!isForward && (whichSpans&SPAN_FWD)!=0 && (expectCount&1)==0) {
spanCondition=invertSpanCondition(spanCondition, contained);
}
count=0;
switch(type) {
case 0:
case 1:
start=0;
if(length<0) {
length=slen(s, isUTF16);
}
for(;;) {
start+= isUTF16 ? containsSpanUTF16(set, (const UChar *)s+start, length-start, spanCondition) :
containsSpanUTF8(set, (const char *)s+start, length-start, spanCondition);
if(count<limitsCapacity) {
limits[count]=start;
}
++count;
if(start>=length) {
break;
}
spanCondition=invertSpanCondition(spanCondition, contained);
}
break;
case 2:
case 3:
start=0;
for(;;) {
start+= isUTF16 ? realSet.span((const UChar *)s+start, length>=0 ? length-start : length, spanCondition) :
realSet.spanUTF8((const char *)s+start, length>=0 ? length-start : length, spanCondition);
if(count<limitsCapacity) {
limits[count]=start;
}
++count;
if(length>=0 ? start>=length :
isUTF16 ? ((const UChar *)s)[start]==0 :
((const char *)s)[start]==0
) {
break;
}
spanCondition=invertSpanCondition(spanCondition, contained);
}
break;
case 4:
case 5:
if(length<0) {
length=slen(s, isUTF16);
}
for(;;) {
++count;
if(count<=limitsCapacity) {
limits[limitsCapacity-count]=length;
}
length= isUTF16 ? containsSpanBackUTF16(set, (const UChar *)s, length, spanCondition) :
containsSpanBackUTF8(set, (const char *)s, length, spanCondition);
if(length==0 && spanCondition==firstSpanCondition) {
break;
}
spanCondition=invertSpanCondition(spanCondition, contained);
}
if(count<limitsCapacity) {
memmove(limits, limits+(limitsCapacity-count), count*4);
}
break;
case 6:
case 7:
for(;;) {
++count;
if(count<=limitsCapacity) {
limits[limitsCapacity-count]= length >=0 ? length : slen(s, isUTF16);
}
// Note: Length<0 is tested only for the first spanBack().
// If we wanted to keep length<0 for all spanBack()s, we would have to
// temporarily modify the string by placing a NUL where the previous spanBack() stopped.
length= isUTF16 ? realSet.spanBack((const UChar *)s, length, spanCondition) :
realSet.spanBackUTF8((const char *)s, length, spanCondition);
if(length==0 && spanCondition==firstSpanCondition) {
break;
}
spanCondition=invertSpanCondition(spanCondition, contained);
}
if(count<limitsCapacity) {
memmove(limits, limits+(limitsCapacity-count), count*4);
}
break;
default:
typeName="";
return -1;
}
return count;
}
// sets to be tested; odd index=isComplement
enum {
SLOW,
SLOW_NOT,
FAST,
FAST_NOT,
SET_COUNT
};
static const char *const setNames[SET_COUNT]={
"slow",
"slow.not",
"fast",
"fast.not"
};
/*
* Verify that we get the same results whether we look at text with contains(),
* span() or spanBack(), using unfrozen or frozen versions of the set,
* and using the set or its complement (switching the spanConditions accordingly).
* The latter verifies that
* set.span(spanCondition) == set.complement().span(!spanCondition).
*
* The expectLimits[] are either provided by the caller (with expectCount>=0)
* or returned to the caller (with an input expectCount<0).
*/
void UnicodeSetTest::testSpan(const UnicodeSetWithStrings *sets[4],
const void *s, int32_t length, UBool isUTF16,
uint32_t whichSpans,
int32_t expectLimits[], int32_t &expectCount,
const char *testName, int32_t index) {
int32_t limits[500];
int32_t limitsCount;
int i, j;
const char *typeName;
int type;
for(i=0; i<SET_COUNT; ++i) {
if((i&1)==0) {
// Even-numbered sets are original, uncomplemented sets.
if((whichSpans&SPAN_SET)==0) {
continue;
}
} else {
// Odd-numbered sets are complemented.
if((whichSpans&SPAN_COMPLEMENT)==0) {
continue;
}
}
for(type=0;; ++type) {
limitsCount=getSpans(*sets[i], (UBool)(i&1),
s, length, isUTF16,
whichSpans,
type, typeName,
limits, LENGTHOF(limits), expectCount);
if(typeName[0]==0) {
break; // All types tried.
}
if(limitsCount<0) {
continue; // Span option filtered out.
}
if(expectCount<0) {
expectCount=limitsCount;
if(limitsCount>LENGTHOF(limits)) {
errln("FAIL: %s[0x%lx].%s.%s span count=%ld > %ld capacity - too many spans",
testName, (long)index, setNames[i], typeName, (long)limitsCount, (long)LENGTHOF(limits));
return;
}
memcpy(expectLimits, limits, limitsCount*4);
} else if(limitsCount!=expectCount) {
errln("FAIL: %s[0x%lx].%s.%s span count=%ld != %ld",
testName, (long)index, setNames[i], typeName, (long)limitsCount, (long)expectCount);
} else {
for(j=0; j<limitsCount; ++j) {
if(limits[j]!=expectLimits[j]) {
errln("FAIL: %s[0x%lx].%s.%s span count=%ld limits[%d]=%ld != %ld",
testName, (long)index, setNames[i], typeName, (long)limitsCount,
j, (long)limits[j], (long)expectLimits[j]);
break;
}
}
}
}
}
// Compare span() with containsAll()/containsNone(),
// but only if we have expectLimits[] from the uncomplemented set.
if(isUTF16 && (whichSpans&SPAN_SET)!=0) {
const UChar *s16=(const UChar *)s;
UnicodeString string;
int32_t prev=0, limit, length;
for(i=0; i<expectCount; ++i) {
limit=expectLimits[i];
length=limit-prev;
if(length>0) {
string.setTo(FALSE, s16+prev, length); // read-only alias
if(i&1) {
if(!sets[SLOW]->getSet().containsAll(string)) {
errln("FAIL: %s[0x%lx].%s.containsAll(%ld..%ld)==FALSE contradicts span()",
testName, (long)index, setNames[SLOW], (long)prev, (long)limit);
return;
}
if(!sets[FAST]->getSet().containsAll(string)) {
errln("FAIL: %s[0x%lx].%s.containsAll(%ld..%ld)==FALSE contradicts span()",
testName, (long)index, setNames[FAST], (long)prev, (long)limit);
return;
}
} else {
if(!sets[SLOW]->getSet().containsNone(string)) {
errln("FAIL: %s[0x%lx].%s.containsNone(%ld..%ld)==FALSE contradicts span()",
testName, (long)index, setNames[SLOW], (long)prev, (long)limit);
return;
}
if(!sets[FAST]->getSet().containsNone(string)) {
errln("FAIL: %s[0x%lx].%s.containsNone(%ld..%ld)==FALSE contradicts span()",
testName, (long)index, setNames[FAST], (long)prev, (long)limit);
return;
}
}
}
prev=limit;
}
}
}
// Specifically test either UTF-16 or UTF-8.
void UnicodeSetTest::testSpan(const UnicodeSetWithStrings *sets[4],
const void *s, int32_t length, UBool isUTF16,
uint32_t whichSpans,
const char *testName, int32_t index) {
int32_t expectLimits[500];
int32_t expectCount=-1;
testSpan(sets, s, length, isUTF16, whichSpans, expectLimits, expectCount, testName, index);
}
UBool stringContainsUnpairedSurrogate(const UChar *s, int32_t length) {
UChar c, c2;
if(length>=0) {
while(length>0) {
c=*s++;
--length;
if(0xd800<=c && c<0xe000) {
if(c>=0xdc00 || length==0 || !U16_IS_TRAIL(c2=*s++)) {
return TRUE;
}
--length;
}
}
} else {
while((c=*s++)!=0) {
if(0xd800<=c && c<0xe000) {
if(c>=0xdc00 || !U16_IS_TRAIL(c2=*s++)) {
return TRUE;
}
}
}
}
return FALSE;
}
// Test both UTF-16 and UTF-8 versions of span() etc. on the same sets and text,
// unless either UTF is turned off in whichSpans.
// Testing UTF-16 and UTF-8 together requires that surrogate code points
// have the same contains(c) value as U+FFFD.
void UnicodeSetTest::testSpanBothUTFs(const UnicodeSetWithStrings *sets[4],
const UChar *s16, int32_t length16,
uint32_t whichSpans,
const char *testName, int32_t index) {
int32_t expectLimits[500];
int32_t expectCount;
expectCount=-1; // Get expectLimits[] from testSpan().
if((whichSpans&SPAN_UTF16)!=0) {
testSpan(sets, s16, length16, TRUE, whichSpans, expectLimits, expectCount, testName, index);
}
if((whichSpans&SPAN_UTF8)==0) {
return;
}
// Convert s16[] and expectLimits[] to UTF-8.
uint8_t s8[3000];
int32_t offsets[3000];
const UChar *s16Limit=s16+length16;
char *t=(char *)s8;
char *tLimit=t+sizeof(s8);
int32_t *o=offsets;
UErrorCode errorCode=U_ZERO_ERROR;
// Convert with substitution: Turn unpaired surrogates into U+FFFD.
ucnv_fromUnicode(openUTF8Converter(), &t, tLimit, &s16, s16Limit, o, TRUE, &errorCode);
if(U_FAILURE(errorCode)) {
errln("FAIL: %s[0x%lx] ucnv_fromUnicode(to UTF-8) fails with %s",
testName, (long)index, u_errorName(errorCode));
ucnv_resetFromUnicode(utf8Cnv);
return;
}
int32_t length8=(int32_t)(t-(char *)s8);
// Convert expectLimits[].
int32_t i, j, expect;
for(i=j=0; i<expectCount; ++i) {
expect=expectLimits[i];
if(expect==length16) {
expectLimits[i]=length8;
} else {
while(offsets[j]<expect) {
++j;
}
expectLimits[i]=j;
}
}
testSpan(sets, s8, length8, FALSE, whichSpans, expectLimits, expectCount, testName, index);
}
static UChar32 nextCodePoint(UChar32 c) {
// Skip some large and boring ranges.
switch(c) {
case 0x3441:
return 0x4d7f;
case 0x5100:
return 0x9f00;
case 0xb040:
return 0xd780;
case 0xe041:
return 0xf8fe;
case 0x10100:
return 0x20000;
case 0x20041:
return 0xe0000;
case 0xe0101:
return 0x10fffd;
default:
return c+1;
}
}
// Verify that all implementations represent the same set.
void UnicodeSetTest::testSpanContents(const UnicodeSetWithStrings *sets[4], uint32_t whichSpans, const char *testName) {
// contains(U+FFFD) is inconsistent with contains(some surrogates),
// or the set contains strings with unpaired surrogates which don't translate to valid UTF-8:
// Skip the UTF-8 part of the test - if the string contains surrogates -
// because it is likely to produce a different result.
UBool inconsistentSurrogates=
(!(sets[0]->getSet().contains(0xfffd) ?
sets[0]->getSet().contains(0xd800, 0xdfff) :
sets[0]->getSet().containsNone(0xd800, 0xdfff)) ||
sets[0]->hasStringsWithSurrogates());
UChar s[1000];
int32_t length=0;
uint32_t localWhichSpans;
UChar32 c, first;
for(first=c=0;; c=nextCodePoint(c)) {
if(c>0x10ffff || length>(LENGTHOF(s)-U16_MAX_LENGTH)) {
localWhichSpans=whichSpans;
if(stringContainsUnpairedSurrogate(s, length) && inconsistentSurrogates) {
localWhichSpans&=~SPAN_UTF8;
}
testSpanBothUTFs(sets, s, length, localWhichSpans, testName, first);
if(c>0x10ffff) {
break;
}
length=0;
first=c;
}
U16_APPEND_UNSAFE(s, length, c);
}
}
// Test with a particular, interesting string.
// Specify length and try NUL-termination.
void UnicodeSetTest::testSpanUTF16String(const UnicodeSetWithStrings *sets[4], uint32_t whichSpans, const char *testName) {
static const UChar s[]={
0x61, 0x62, 0x20, // Latin, space
0x3b1, 0x3b2, 0x3b3, // Greek
0xd900, // lead surrogate
0x3000, 0x30ab, 0x30ad, // wide space, Katakana
0xdc05, // trail surrogate
0xa0, 0xac00, 0xd7a3, // nbsp, Hangul
0xd900, 0xdc05, // unassigned supplementary
0xd840, 0xdfff, 0xd860, 0xdffe, // Han supplementary
0xd7a4, 0xdc05, 0xd900, 0x2028, // unassigned, surrogates in wrong order, LS
0 // NUL
};
if((whichSpans&SPAN_UTF16)==0) {
return;
}
testSpan(sets, s, -1, TRUE, (whichSpans&~SPAN_UTF8), testName, 0);
testSpan(sets, s, LENGTHOF(s)-1, TRUE, (whichSpans&~SPAN_UTF8), testName, 1);
}
void UnicodeSetTest::testSpanUTF8String(const UnicodeSetWithStrings *sets[4], uint32_t whichSpans, const char *testName) {
static const char s[]={
"abc" // Latin
/* trail byte in lead position */
"\x80"
" " // space
/* truncated multi-byte sequences */
"\xd0"
"\xe0"
"\xe1"
"\xed"
"\xee"
"\xf0"
"\xf1"
"\xf4"
"\xf8"
"\xfc"
"\xCE\xB1\xCE\xB2\xCE\xB3" // Greek
/* trail byte in lead position */
"\x80"
"\xe0\x80"
"\xe0\xa0"
"\xe1\x80"
"\xed\x80"
"\xed\xa0"
"\xee\x80"
"\xf0\x80"
"\xf0\x90"
"\xf1\x80"
"\xf4\x80"
"\xf4\x90"
"\xf8\x80"
"\xfc\x80"
"\xE3\x80\x80\xE3\x82\xAB\xE3\x82\xAD" // wide space, Katakana
/* trail byte in lead position */
"\x80"
"\xf0\x80\x80"
"\xf0\x90\x80"
"\xf1\x80\x80"
"\xf4\x80\x80"
"\xf4\x90\x80"
"\xf8\x80\x80"
"\xfc\x80\x80"
"\xC2\xA0\xEA\xB0\x80\xED\x9E\xA3" // nbsp, Hangul
/* trail byte in lead position */
"\x80"
"\xf8\x80\x80\x80"
"\xfc\x80\x80\x80"
"\xF1\x90\x80\x85" // unassigned supplementary
/* trail byte in lead position */
"\x80"
"\xfc\x80\x80\x80\x80"
"\xF0\xA0\x8F\xBF\xF0\xA8\x8F\xBE" // Han supplementary
/* trail byte in lead position */
"\x80"
/* complete sequences but non-shortest forms or out of range etc. */
"\xc0\x80"
"\xe0\x80\x80"
"\xed\xa0\x80"
"\xf0\x80\x80\x80"
"\xf4\x90\x80\x80"
"\xf8\x80\x80\x80\x80"
"\xfc\x80\x80\x80\x80\x80"
"\xfe"
"\xff"
/* trail byte in lead position */
"\x80"
"\xED\x9E\xA4\xE2\x80\xA8" // unassigned, LS, NUL-terminated
};
if((whichSpans&SPAN_UTF8)==0) {
return;
}
testSpan(sets, s, -1, FALSE, (whichSpans&~SPAN_UTF16), testName, 0);
testSpan(sets, s, LENGTHOF(s)-1, FALSE, (whichSpans&~SPAN_UTF16), testName, 1);
}
// Take a set of span options and multiply them so that
// each portion only has one of the options a, b and c.
// If b==0, then the set of options is just modified with mask and a.
// If b!=0 and c==0, then the set of options is just modified with mask, a and b.
static int32_t
addAlternative(uint32_t whichSpans[], int32_t whichSpansCount,
uint32_t mask, uint32_t a, uint32_t b, uint32_t c) {
uint32_t s;
int32_t i;
for(i=0; i<whichSpansCount; ++i) {
s=whichSpans[i]&mask;
whichSpans[i]=s|a;
if(b!=0) {
whichSpans[whichSpansCount+i]=s|b;
if(c!=0) {
whichSpans[2*whichSpansCount+i]=s|c;
}
}
}
return b==0 ? whichSpansCount : c==0 ? 2*whichSpansCount : 3*whichSpansCount;
}
#define _63_a "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"
#define _64_a "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"
#define _63_b "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb"
#define _64_b "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb"
void UnicodeSetTest::TestSpan() {
// "[...]" is a UnicodeSet pattern.
// "*" performs tests on all Unicode code points and on a selection of
// malformed UTF-8/16 strings.
// "-options" limits the scope of testing for the current set.
// By default, the test verifies that equivalent boundaries are found
// for UTF-16 and UTF-8, going forward and backward,
// alternating USET_SPAN_NOT_CONTAINED with
// either USET_SPAN_CONTAINED or USET_SPAN_SIMPLE.
// Single-character options:
// 8 -- UTF-16 and UTF-8 boundaries may differ.
// Cause: contains(U+FFFD) is inconsistent with contains(some surrogates),
// or the set contains strings with unpaired surrogates
// which do not translate to valid UTF-8.
// c -- set.span() and set.complement().span() boundaries may differ.
// Cause: Set strings are not complemented.
// b -- span() and spanBack() boundaries may differ.
// Cause: Strings in the set overlap, and spanBack(USET_SPAN_CONTAINED)
// and spanBack(USET_SPAN_SIMPLE) are defined to
// match with non-overlapping substrings.
// For example, with a set containing "ab" and "ba",
// span() of "aba" yields boundaries { 0, 2, 3 }
// because the initial "ab" matches from 0 to 2,
// while spanBack() yields boundaries { 0, 1, 3 }
// because the final "ba" matches from 1 to 3.
// l -- USET_SPAN_CONTAINED and USET_SPAN_SIMPLE boundaries may differ.
// Cause: Strings in the set overlap, and a longer match may
// require a sequence including non-longest substrings.
// For example, with a set containing "ab", "abc" and "cd",
// span(contained) of "abcd" spans the entire string
// but span(longest match) only spans the first 3 characters.
// Each "-options" first resets all options and then applies the specified options.
// A "-" without options resets the options.
// The options are also reset for each new set.
// Other strings will be spanned.
static const char *const testdata[]={
"[:ID_Continue:]",
"*",
"[:White_Space:]",
"*",
"[]",
"*",
"[\\u0000-\\U0010FFFF]",
"*",
"[\\u0000\\u0080\\u0800\\U00010000]",
"*",
"[\\u007F\\u07FF\\uFFFF\\U0010FFFF]",
"*",
"[[[:ID_Continue:]-[\\u30ab\\u30ad]]{\\u3000\\u30ab}{\\u3000\\u30ab\\u30ad}]",
"-c",
"*",
"[[[:ID_Continue:]-[\\u30ab\\u30ad]]{\\u30ab\\u30ad}{\\u3000\\u30ab\\u30ad}]",
"-c",
"*",
// Overlapping strings cause overlapping attempts to match.
"[x{xy}{xya}{axy}{ax}]",
"-cl",
// More repetitions of "xya" would take too long with the recursive
// reference implementation.
// containsAll()=FALSE
// test_string 0x14
"xx"
"xyaxyaxyaxya" // set.complement().span(longest match) will stop here.
"xx" // set.complement().span(contained) will stop between the two 'x'es.
"xyaxyaxyaxya"
"xx"
"xyaxyaxyaxya" // span() ends here.
"aaa",
// containsAll()=TRUE
// test_string 0x15
"xx"
"xyaxyaxyaxya"
"xx"
"xyaxyaxyaxya"
"xx"
"xyaxyaxyaxy",
"-bc",
// test_string 0x17
"byayaxya", // span() -> { 4, 7, 8 } spanBack() -> { 5, 8 }
"-c",
"byayaxy", // span() -> { 4, 7 } complement.span() -> { 7 }
"byayax", // span() -> { 4, 6 } complement.span() -> { 6 }
"-",
"byaya", // span() -> { 5 }
"byay", // span() -> { 4 }
"bya", // span() -> { 3 }
// span(longest match) will not span the whole string.
"[a{ab}{bc}]",
"-cl",
// test_string 0x21
"abc",
"[a{ab}{abc}{cd}]",
"-cl",
"acdabcdabccd",
// spanBack(longest match) will not span the whole string.
"[c{ab}{bc}]",
"-cl",
"abc",
"[d{cd}{bcd}{ab}]",
"-cl",
"abbcdabcdabd",
// Test with non-ASCII set strings - test proper handling of surrogate pairs
// and UTF-8 trail bytes.
// Copies of above test sets and strings, but transliterated to have
// different code points with similar trail units.
// Previous: a b c d
// Unicode: 042B 30AB 200AB 204AB
// UTF-16: 042B 30AB D840 DCAB D841 DCAB
// UTF-8: D0 AB E3 82 AB F0 A0 82 AB F0 A0 92 AB
"[\\u042B{\\u042B\\u30AB}{\\u042B\\u30AB\\U000200AB}{\\U000200AB\\U000204AB}]",
"-cl",
"\\u042B\\U000200AB\\U000204AB\\u042B\\u30AB\\U000200AB\\U000204AB\\u042B\\u30AB\\U000200AB\\U000200AB\\U000204AB",
"[\\U000204AB{\\U000200AB\\U000204AB}{\\u30AB\\U000200AB\\U000204AB}{\\u042B\\u30AB}]",
"-cl",
"\\u042B\\u30AB\\u30AB\\U000200AB\\U000204AB\\u042B\\u30AB\\U000200AB\\U000204AB\\u042B\\u30AB\\U000204AB",
// Stress bookkeeping and recursion.
// The following strings are barely doable with the recursive
// reference implementation.
// The not-contained character at the end prevents an early exit from the span().
"[b{bb}]",
"-c",
// test_string 0x33
"bbbbbbbbbbbbbbbbbbbbbbbb-",
// On complement sets, span() and spanBack() get different results
// because b is not in the complement set and there is an odd number of b's
// in the test string.
"-bc",
"bbbbbbbbbbbbbbbbbbbbbbbbb-",
// Test with set strings with an initial or final code point span
// longer than 254.
"[a{" _64_a _64_a _64_a _64_a "b}"
"{a" _64_b _64_b _64_b _64_b "}]",
"-c",
_64_a _64_a _64_a _63_a "b",
_64_a _64_a _64_a _64_a "b",
_64_a _64_a _64_a _64_a "aaaabbbb",
"a" _64_b _64_b _64_b _63_b,
"a" _64_b _64_b _64_b _64_b,
"aaaabbbb" _64_b _64_b _64_b _64_b,
// Test with strings containing unpaired surrogates.
// They are not representable in UTF-8, and a leading trail surrogate
// and a trailing lead surrogate must not match in the middle of a proper surrogate pair.
// U+20001 == \\uD840\\uDC01
// U+20400 == \\uD841\\uDC00
"[a\\U00020001\\U00020400{ab}{b\\uD840}{\\uDC00a}]",
"-8cl",
"aaab\\U00020001ba\\U00020400aba\\uD840ab\\uD840\\U00020000b\\U00020000a\\U00020000\\uDC00a\\uDC00babbb"
};
uint32_t whichSpans[96]={ SPAN_ALL };
int32_t whichSpansCount=1;
UnicodeSet *sets[SET_COUNT]={ NULL };
const UnicodeSetWithStrings *sets_with_str[SET_COUNT]={ NULL };
char testName[1024];
char *testNameLimit=testName;
int32_t i, j;
for(i=0; i<LENGTHOF(testdata); ++i) {
const char *s=testdata[i];
if(s[0]=='[') {
// Create new test sets from this pattern.
for(j=0; j<SET_COUNT; ++j) {
delete sets_with_str[j];
delete sets[j];
}
UErrorCode errorCode=U_ZERO_ERROR;
sets[SLOW]=new UnicodeSet(UnicodeString(s, -1, US_INV).unescape(), errorCode);
if(U_FAILURE(errorCode)) {
dataerrln("FAIL: Unable to create UnicodeSet(%s) - %s", s, u_errorName(errorCode));
break;
}
sets[SLOW_NOT]=new UnicodeSet(*sets[SLOW]);
sets[SLOW_NOT]->complement();
// Intermediate set: Test cloning of a frozen set.
UnicodeSet *fast=new UnicodeSet(*sets[SLOW]);
fast->freeze();
sets[FAST]=(UnicodeSet *)fast->clone();
delete fast;
UnicodeSet *fastNot=new UnicodeSet(*sets[SLOW_NOT]);
fastNot->freeze();
sets[FAST_NOT]=(UnicodeSet *)fastNot->clone();
delete fastNot;
for(j=0; j<SET_COUNT; ++j) {
sets_with_str[j]=new UnicodeSetWithStrings(*sets[j]);
}
strcpy(testName, s);
testNameLimit=strchr(testName, 0);
*testNameLimit++=':';
*testNameLimit=0;
whichSpans[0]=SPAN_ALL;
whichSpansCount=1;
} else if(s[0]=='-') {
whichSpans[0]=SPAN_ALL;
whichSpansCount=1;
while(*++s!=0) {
switch(*s) {
case 'c':
whichSpansCount=addAlternative(whichSpans, whichSpansCount,
~SPAN_POLARITY,
SPAN_SET,
SPAN_COMPLEMENT,
0);
break;
case 'b':
whichSpansCount=addAlternative(whichSpans, whichSpansCount,
~SPAN_DIRS,
SPAN_FWD,
SPAN_BACK,
0);
break;
case 'l':
// test USET_SPAN_CONTAINED FWD & BACK, and separately
// USET_SPAN_SIMPLE only FWD, and separately
// USET_SPAN_SIMPLE only BACK
whichSpansCount=addAlternative(whichSpans, whichSpansCount,
~(SPAN_DIRS|SPAN_CONDITION),
SPAN_DIRS|SPAN_CONTAINED,
SPAN_FWD|SPAN_SIMPLE,
SPAN_BACK|SPAN_SIMPLE);
break;
case '8':
whichSpansCount=addAlternative(whichSpans, whichSpansCount,
~SPAN_UTFS,
SPAN_UTF16,
SPAN_UTF8,
0);
break;
default:
errln("FAIL: unrecognized span set option in \"%s\"", testdata[i]);
break;
}
}
} else if(0==strcmp(s, "*")) {
strcpy(testNameLimit, "bad_string");
for(j=0; j<whichSpansCount; ++j) {
if(whichSpansCount>1) {
sprintf(testNameLimit+10 /* strlen("bad_string") */,
"%%0x%3x",
whichSpans[j]);
}
testSpanUTF16String(sets_with_str, whichSpans[j], testName);
testSpanUTF8String(sets_with_str, whichSpans[j], testName);
}
strcpy(testNameLimit, "contents");
for(j=0; j<whichSpansCount; ++j) {
if(whichSpansCount>1) {
sprintf(testNameLimit+8 /* strlen("contents") */,
"%%0x%3x",
whichSpans[j]);
}
testSpanContents(sets_with_str, whichSpans[j], testName);
}
} else {
UnicodeString string=UnicodeString(s, -1, US_INV).unescape();
strcpy(testNameLimit, "test_string");
for(j=0; j<whichSpansCount; ++j) {
if(whichSpansCount>1) {
sprintf(testNameLimit+11 /* strlen("test_string") */,
"%%0x%3x",
whichSpans[j]);
}
testSpanBothUTFs(sets_with_str, string.getBuffer(), string.length(), whichSpans[j], testName, i);
}
}
}
for(j=0; j<SET_COUNT; ++j) {
delete sets_with_str[j];
delete sets[j];
}
}
// Test select patterns and strings, and test USET_SPAN_SIMPLE.
void UnicodeSetTest::TestStringSpan() {
static const char *pattern="[x{xy}{xya}{axy}{ax}]";
static const char *const string=
"xx"
"xyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxya"
"xx"
"xyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxya"
"xx"
"xyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxyaxy"
"aaaa";
UErrorCode errorCode=U_ZERO_ERROR;
UnicodeString pattern16=UnicodeString(pattern, -1, US_INV);
UnicodeSet set(pattern16, errorCode);
if(U_FAILURE(errorCode)) {
errln("FAIL: Unable to create UnicodeSet(%s) - %s", pattern, u_errorName(errorCode));
return;
}
UnicodeString string16=UnicodeString(string, -1, US_INV).unescape();
if(set.containsAll(string16)) {
errln("FAIL: UnicodeSet(%s).containsAll(%s) should be FALSE", pattern, string);
}
// Remove trailing "aaaa".
string16.truncate(string16.length()-4);
if(!set.containsAll(string16)) {
errln("FAIL: UnicodeSet(%s).containsAll(%s[:-4]) should be TRUE", pattern, string);
}
string16=UNICODE_STRING_SIMPLE("byayaxya");
const UChar *s16=string16.getBuffer();
int32_t length16=string16.length();
if( set.span(s16, 8, USET_SPAN_NOT_CONTAINED)!=4 ||
set.span(s16, 7, USET_SPAN_NOT_CONTAINED)!=4 ||
set.span(s16, 6, USET_SPAN_NOT_CONTAINED)!=4 ||
set.span(s16, 5, USET_SPAN_NOT_CONTAINED)!=5 ||
set.span(s16, 4, USET_SPAN_NOT_CONTAINED)!=4 ||
set.span(s16, 3, USET_SPAN_NOT_CONTAINED)!=3
) {
errln("FAIL: UnicodeSet(%s).span(while not) returns the wrong value", pattern);
}
pattern="[a{ab}{abc}{cd}]";
pattern16=UnicodeString(pattern, -1, US_INV);
set.applyPattern(pattern16, errorCode);
if(U_FAILURE(errorCode)) {
errln("FAIL: Unable to create UnicodeSet(%s) - %s", pattern, u_errorName(errorCode));
return;
}
string16=UNICODE_STRING_SIMPLE("acdabcdabccd");
s16=string16.getBuffer();
length16=string16.length();
if( set.span(s16, 12, USET_SPAN_CONTAINED)!=12 ||
set.span(s16, 12, USET_SPAN_SIMPLE)!=6 ||
set.span(s16+7, 5, USET_SPAN_SIMPLE)!=5
) {
errln("FAIL: UnicodeSet(%s).span(while longest match) returns the wrong value", pattern);
}
pattern="[d{cd}{bcd}{ab}]";
pattern16=UnicodeString(pattern, -1, US_INV);
set.applyPattern(pattern16, errorCode).freeze();
if(U_FAILURE(errorCode)) {
errln("FAIL: Unable to create UnicodeSet(%s) - %s", pattern, u_errorName(errorCode));
return;
}
string16=UNICODE_STRING_SIMPLE("abbcdabcdabd");
s16=string16.getBuffer();
length16=string16.length();
if( set.spanBack(s16, 12, USET_SPAN_CONTAINED)!=0 ||
set.spanBack(s16, 12, USET_SPAN_SIMPLE)!=6 ||
set.spanBack(s16, 5, USET_SPAN_SIMPLE)!=0
) {
errln("FAIL: UnicodeSet(%s).spanBack(while longest match) returns the wrong value", pattern);
}
}