You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

1485 lines
42 KiB

// © 2016 and later: Unicode, Inc. and others.
// License & terms of use: http://www.unicode.org/copyright.html
/********************************************************************
* COPYRIGHT:
* Copyright (c) 1997-2016, International Business Machines Corporation and
* others. All Rights Reserved.
********************************************************************/
#include "unicode/utypes.h"
#if !UCONFIG_NO_COLLATION
#include "unicode/coll.h"
#include "unicode/localpointer.h"
#include "unicode/tblcoll.h"
#include "unicode/unistr.h"
#include "unicode/sortkey.h"
#include "regcoll.h"
#include "sfwdchit.h"
#include "testutil.h"
#include "cmemory.h"
CollationRegressionTest::CollationRegressionTest()
{
UErrorCode status = U_ZERO_ERROR;
en_us = (RuleBasedCollator *)Collator::createInstance(Locale::getUS(), status);
if(U_FAILURE(status)) {
delete en_us;
en_us = 0;
errcheckln(status, "Collator creation failed with %s", u_errorName(status));
return;
}
}
CollationRegressionTest::~CollationRegressionTest()
{
delete en_us;
}
// @bug 4048446
//
// CollationElementIterator.reset() doesn't work
//
void CollationRegressionTest::Test4048446(/* char* par */)
{
const UnicodeString test1 = "XFILE What subset of all possible test cases has the highest probability of detecting the most errors?";
const UnicodeString test2 = "Xf_ile What subset of all possible test cases has the lowest probability of detecting the least errors?";
CollationElementIterator *i1 = en_us->createCollationElementIterator(test1);
CollationElementIterator *i2 = en_us->createCollationElementIterator(test1);
UErrorCode status = U_ZERO_ERROR;
if (i1 == NULL|| i2 == NULL)
{
errln("Could not create CollationElementIterator's");
delete i1;
delete i2;
return;
}
while (i1->next(status) != CollationElementIterator::NULLORDER)
{
if (U_FAILURE(status))
{
errln("error calling next()");
delete i1;
delete i2;
return;
}
}
i1->reset();
assertEqual(*i1, *i2);
delete i1;
delete i2;
}
// @bug 4051866
//
// Collator -> rules -> Collator round-trip broken for expanding characters
//
void CollationRegressionTest::Test4051866(/* char* par */)
{
UnicodeString rules;
UErrorCode status = U_ZERO_ERROR;
rules += "&n < o ";
rules += "& oe ,o";
rules += (UChar)0x3080;
rules += "& oe ,";
rules += (UChar)0x1530;
rules += " ,O";
rules += "& OE ,O";
rules += (UChar)0x3080;
rules += "& OE ,";
rules += (UChar)0x1520;
rules += "< p ,P";
// Build a collator containing expanding characters
LocalPointer<RuleBasedCollator> c1(new RuleBasedCollator(rules, status), status);
if (U_FAILURE(status)) {
errln("RuleBasedCollator(rule string) failed - %s", u_errorName(status));
return;
}
// Build another using the rules from the first
LocalPointer<RuleBasedCollator> c2(new RuleBasedCollator(c1->getRules(), status), status);
if (U_FAILURE(status)) {
errln("RuleBasedCollator(rule string from other RBC) failed - %s", u_errorName(status));
return;
}
// Make sure they're the same
if (!(c1->getRules() == c2->getRules()))
{
errln("Rules are not equal");
}
}
// @bug 4053636
//
// Collator thinks "black-bird" == "black"
//
void CollationRegressionTest::Test4053636(/* char* par */)
{
if (en_us->equals("black_bird", "black"))
{
errln("black-bird == black");
}
}
// @bug 4054238
//
// CollationElementIterator will not work correctly if the associated
// Collator object's mode is changed
//
void CollationRegressionTest::Test4054238(/* char* par */)
{
const UChar chars3[] = {0x61, 0x00FC, 0x62, 0x65, 0x63, 0x6b, 0x20, 0x47, 0x72, 0x00F6, 0x00DF, 0x65, 0x20, 0x4c, 0x00FC, 0x62, 0x63, 0x6b, 0};
const UnicodeString test3(chars3);
RuleBasedCollator *c = en_us->clone();
// NOTE: The Java code uses en_us to create the CollationElementIterators
// but I'm pretty sure that's wrong, so I've changed this to use c.
UErrorCode status = U_ZERO_ERROR;
c->setAttribute(UCOL_NORMALIZATION_MODE, UCOL_ON, status);
CollationElementIterator *i1 = c->createCollationElementIterator(test3);
delete i1;
delete c;
}
// @bug 4054734
//
// Collator::IDENTICAL documented but not implemented
//
void CollationRegressionTest::Test4054734(/* char* par */)
{
/*
Here's the original Java:
String[] decomp = {
"\u0001", "<", "\u0002",
"\u0001", "=", "\u0001",
"A\u0001", ">", "~\u0002", // Ensure A and ~ are not compared bitwise
"\u00C0", "=", "A\u0300" // Decomp should make these equal
};
String[] nodecomp = {
"\u00C0", ">", "A\u0300" // A-grave vs. A combining-grave
};
*/
static const UChar decomp[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x0001, 0}, {0x3c, 0}, {0x0002, 0},
{0x0001, 0}, {0x3d, 0}, {0x0001, 0},
{0x41, 0x0001, 0}, {0x3e, 0}, {0x7e, 0x0002, 0},
{0x00c0, 0}, {0x3d, 0}, {0x41, 0x0300, 0}
};
UErrorCode status = U_ZERO_ERROR;
RuleBasedCollator *c = en_us->clone();
c->setStrength(Collator::IDENTICAL);
c->setAttribute(UCOL_NORMALIZATION_MODE, UCOL_ON, status);
compareArray(*c, decomp, UPRV_LENGTHOF(decomp));
delete c;
}
// @bug 4054736
//
// Full Decomposition mode not implemented
//
void CollationRegressionTest::Test4054736(/* char* par */)
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedCollator *c = en_us->clone();
c->setStrength(Collator::SECONDARY);
c->setAttribute(UCOL_NORMALIZATION_MODE, UCOL_ON, status);
static const UChar tests[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0xFB4F, 0}, {0x3d, 0}, {0x05D0, 0x05DC} // Alef-Lamed vs. Alef, Lamed
};
compareArray(*c, tests, UPRV_LENGTHOF(tests));
delete c;
}
// @bug 4058613
//
// Collator::createInstance() causes an ArrayIndexOutofBoundsException for Korean
//
void CollationRegressionTest::Test4058613(/* char* par */)
{
// Creating a default collator doesn't work when Korean is the default
// locale
Locale oldDefault = Locale::getDefault();
UErrorCode status = U_ZERO_ERROR;
Locale::setDefault(Locale::getKorean(), status);
if (U_FAILURE(status))
{
errln("Could not set default locale to Locale::KOREAN");
return;
}
Collator *c = NULL;
c = Collator::createInstance("en_US", status);
if (c == NULL || U_FAILURE(status))
{
errln("Could not create a Korean collator");
Locale::setDefault(oldDefault, status);
delete c;
return;
}
// Since the fix to this bug was to turn off decomposition for Korean collators,
// ensure that's what we got
if (c->getAttribute(UCOL_NORMALIZATION_MODE, status) != UCOL_OFF)
{
errln("Decomposition is not set to NO_DECOMPOSITION for Korean collator");
}
delete c;
Locale::setDefault(oldDefault, status);
}
// @bug 4059820
//
// RuleBasedCollator.getRules does not return the exact pattern as input
// for expanding character sequences
//
void CollationRegressionTest::Test4059820(/* char* par */)
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedCollator *c = NULL;
UnicodeString rules = "&9 < a < b , c/a < d < z";
c = new RuleBasedCollator(rules, status);
if (c == NULL || U_FAILURE(status))
{
errln("Failure building a collator.");
delete c;
return;
}
if ( c->getRules().indexOf("c/a") == -1)
{
errln("returned rules do not contain 'c/a'");
}
delete c;
}
// @bug 4060154
//
// MergeCollation::fixEntry broken for "& H < \u0131, \u0130, i, I"
//
void CollationRegressionTest::Test4060154(/* char* par */)
{
UErrorCode status = U_ZERO_ERROR;
UnicodeString rules;
rules += "&f < g, G < h, H < i, I < j, J";
rules += " & H < ";
rules += (UChar)0x0131;
rules += ", ";
rules += (UChar)0x0130;
rules += ", i, I";
RuleBasedCollator *c = NULL;
c = new RuleBasedCollator(rules, status);
if (c == NULL || U_FAILURE(status))
{
errln("failure building collator.");
delete c;
return;
}
c->setAttribute(UCOL_NORMALIZATION_MODE, UCOL_ON, status);
/*
String[] tertiary = {
"A", "<", "B",
"H", "<", "\u0131",
"H", "<", "I",
"\u0131", "<", "\u0130",
"\u0130", "<", "i",
"\u0130", ">", "H",
};
*/
static const UChar tertiary[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x41, 0}, {0x3c, 0}, {0x42, 0},
{0x48, 0}, {0x3c, 0}, {0x0131, 0},
{0x48, 0}, {0x3c, 0}, {0x49, 0},
{0x0131, 0}, {0x3c, 0}, {0x0130, 0},
{0x0130, 0}, {0x3c, 0}, {0x69, 0},
{0x0130, 0}, {0x3e, 0}, {0x48, 0}
};
c->setStrength(Collator::TERTIARY);
compareArray(*c, tertiary, UPRV_LENGTHOF(tertiary));
/*
String[] secondary = {
"H", "<", "I",
"\u0131", "=", "\u0130",
};
*/
static const UChar secondary[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x48, 0}, {0x3c, 0}, {0x49, 0},
{0x0131, 0}, {0x3d, 0}, {0x0130, 0}
};
c->setStrength(Collator::PRIMARY);
compareArray(*c, secondary, UPRV_LENGTHOF(secondary));
delete c;
}
// @bug 4062418
//
// Secondary/Tertiary comparison incorrect in French Secondary
//
void CollationRegressionTest::Test4062418(/* char* par */)
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedCollator *c = NULL;
c = (RuleBasedCollator *) Collator::createInstance(Locale::getCanadaFrench(), status);
if (c == NULL || U_FAILURE(status))
{
errln("Failed to create collator for Locale::getCanadaFrench()");
delete c;
return;
}
c->setStrength(Collator::SECONDARY);
/*
String[] tests = {
"p\u00eache", "<", "p\u00e9ch\u00e9", // Comparing accents from end, p\u00e9ch\u00e9 is greater
};
*/
static const UChar tests[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x70, 0x00EA, 0x63, 0x68, 0x65, 0}, {0x3c, 0}, {0x70, 0x00E9, 0x63, 0x68, 0x00E9, 0}
};
compareArray(*c, tests, UPRV_LENGTHOF(tests));
delete c;
}
// @bug 4065540
//
// Collator::compare() method broken if either string contains spaces
//
void CollationRegressionTest::Test4065540(/* char* par */)
{
if (en_us->compare("abcd e", "abcd f") == 0)
{
errln("'abcd e' == 'abcd f'");
}
}
// @bug 4066189
//
// Unicode characters need to be recursively decomposed to get the
// correct result. For example,
// u1EB1 -> \u0103 + \u0300 -> a + \u0306 + \u0300.
//
void CollationRegressionTest::Test4066189(/* char* par */)
{
static const UChar chars1[] = {0x1EB1, 0};
static const UChar chars2[] = {0x61, 0x0306, 0x0300, 0};
const UnicodeString test1(chars1);
const UnicodeString test2(chars2);
UErrorCode status = U_ZERO_ERROR;
// NOTE: The java code used en_us to create the
// CollationElementIterator's. I'm pretty sure that
// was wrong, so I've change the code to use c1 and c2
RuleBasedCollator *c1 = en_us->clone();
c1->setAttribute(UCOL_NORMALIZATION_MODE, UCOL_ON, status);
CollationElementIterator *i1 = c1->createCollationElementIterator(test1);
RuleBasedCollator *c2 = en_us->clone();
c2->setAttribute(UCOL_NORMALIZATION_MODE, UCOL_OFF, status);
CollationElementIterator *i2 = c2->createCollationElementIterator(test2);
assertEqual(*i1, *i2);
delete i2;
delete c2;
delete i1;
delete c1;
}
// @bug 4066696
//
// French secondary collation checking at the end of compare iteration fails
//
void CollationRegressionTest::Test4066696(/* char* par */)
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedCollator *c = NULL;
c = (RuleBasedCollator *)Collator::createInstance(Locale::getCanadaFrench(), status);
if (c == NULL || U_FAILURE(status))
{
errln("Failure creating collator for Locale::getCanadaFrench()");
delete c;
return;
}
c->setStrength(Collator::SECONDARY);
/*
String[] tests = {
"\u00e0", "<", "\u01fa", // a-grave < A-ring-acute
};
should be:
String[] tests = {
"\u00e0", ">", "\u01fa", // a-grave < A-ring-acute
};
*/
static const UChar tests[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x00E0, 0}, {0x3e, 0}, {0x01FA, 0}
};
compareArray(*c, tests, UPRV_LENGTHOF(tests));
delete c;
}
// @bug 4076676
//
// Bad canonicalization of same-class combining characters
//
void CollationRegressionTest::Test4076676(/* char* par */)
{
// These combining characters are all in the same class, so they should not
// be reordered, and they should compare as unequal.
static const UChar s1[] = {0x41, 0x0301, 0x0302, 0x0300, 0};
static const UChar s2[] = {0x41, 0x0302, 0x0300, 0x0301, 0};
RuleBasedCollator *c = en_us->clone();
c->setStrength(Collator::TERTIARY);
if (c->compare(s1,s2) == 0)
{
errln("Same-class combining chars were reordered");
}
delete c;
}
// @bug 4079231
//
// RuleBasedCollator::operator==(NULL) throws NullPointerException
//
void CollationRegressionTest::Test4079231(/* char* par */)
{
// I don't think there's any way to write this test
// in C++. The following is equivalent to the Java,
// but doesn't compile 'cause NULL can't be converted
// to Collator&
//
// if (en_us->operator==(NULL))
// {
// errln("en_us->operator==(NULL) returned TRUE");
// }
/*
try {
if (en_us->equals(null)) {
errln("en_us->equals(null) returned true");
}
}
catch (Exception e) {
errln("en_us->equals(null) threw " + e.toString());
}
*/
}
// @bug 4078588
//
// RuleBasedCollator breaks on "< a < bb" rule
//
void CollationRegressionTest::Test4078588(/* char *par */)
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedCollator *rbc = new RuleBasedCollator("&9 < a < bb", status);
if (rbc == NULL || U_FAILURE(status))
{
errln("Failed to create RuleBasedCollator.");
delete rbc;
return;
}
Collator::EComparisonResult result = rbc->compare("a","bb");
if (result != Collator::LESS)
{
errln((UnicodeString)"Compare(a,bb) returned " + (int)result
+ (UnicodeString)"; expected -1");
}
delete rbc;
}
// @bug 4081866
//
// Combining characters in different classes not reordered properly.
//
void CollationRegressionTest::Test4081866(/* char* par */)
{
// These combining characters are all in different classes,
// so they should be reordered and the strings should compare as equal.
static const UChar s1[] = {0x41, 0x0300, 0x0316, 0x0327, 0x0315, 0};
static const UChar s2[] = {0x41, 0x0327, 0x0316, 0x0315, 0x0300, 0};
UErrorCode status = U_ZERO_ERROR;
RuleBasedCollator *c = en_us->clone();
c->setStrength(Collator::TERTIARY);
// Now that the default collators are set to NO_DECOMPOSITION
// (as a result of fixing bug 4114077), we must set it explicitly
// when we're testing reordering behavior. -- lwerner, 5/5/98
c->setAttribute(UCOL_NORMALIZATION_MODE, UCOL_ON, status);
if (c->compare(s1,s2) != 0)
{
errln("Combining chars were not reordered");
}
delete c;
}
// @bug 4087241
//
// string comparison errors in Scandinavian collators
//
void CollationRegressionTest::Test4087241(/* char* par */)
{
UErrorCode status = U_ZERO_ERROR;
Locale da_DK("da", "DK");
RuleBasedCollator *c = NULL;
c = (RuleBasedCollator *) Collator::createInstance(da_DK, status);
if (c == NULL || U_FAILURE(status))
{
errln("Failed to create collator for da_DK locale");
delete c;
return;
}
c->setStrength(Collator::SECONDARY);
static const UChar tests[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x7a, 0}, {0x3c, 0}, {0x00E6, 0}, // z < ae
{0x61, 0x0308, 0}, {0x3c, 0}, {0x61, 0x030A, 0}, // a-umlaut < a-ring
{0x59, 0}, {0x3c, 0}, {0x75, 0x0308, 0}, // Y < u-umlaut
};
compareArray(*c, tests, UPRV_LENGTHOF(tests));
delete c;
}
// @bug 4087243
//
// CollationKey takes ignorable strings into account when it shouldn't
//
void CollationRegressionTest::Test4087243(/* char* par */)
{
RuleBasedCollator *c = en_us->clone();
c->setStrength(Collator::TERTIARY);
static const UChar tests[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x31, 0x32, 0x33, 0}, {0x3d, 0}, {0x31, 0x32, 0x33, 0x0001, 0} // 1 2 3 = 1 2 3 ctrl-A
};
compareArray(*c, tests, UPRV_LENGTHOF(tests));
delete c;
}
// @bug 4092260
//
// Mu/micro conflict
// Micro symbol and greek lowercase letter Mu should sort identically
//
void CollationRegressionTest::Test4092260(/* char* par */)
{
UErrorCode status = U_ZERO_ERROR;
Locale el("el", "");
Collator *c = NULL;
c = Collator::createInstance(el, status);
if (c == NULL || U_FAILURE(status))
{
errln("Failed to create collator for el locale.");
delete c;
return;
}
// These now have tertiary differences in UCA
c->setAttribute(UCOL_STRENGTH, UCOL_SECONDARY, status);
static const UChar tests[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x00B5, 0}, {0x3d, 0}, {0x03BC, 0}
};
compareArray(*c, tests, UPRV_LENGTHOF(tests));
delete c;
}
// @bug 4095316
//
void CollationRegressionTest::Test4095316(/* char* par */)
{
UErrorCode status = U_ZERO_ERROR;
Locale el_GR("el", "GR");
Collator *c = Collator::createInstance(el_GR, status);
if (c == NULL || U_FAILURE(status))
{
errln("Failed to create collator for el_GR locale");
delete c;
return;
}
// These now have tertiary differences in UCA
//c->setStrength(Collator::TERTIARY);
c->setAttribute(UCOL_STRENGTH, UCOL_SECONDARY, status);
static const UChar tests[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x03D4, 0}, {0x3d, 0}, {0x03AB, 0}
};
compareArray(*c, tests, UPRV_LENGTHOF(tests));
delete c;
}
// @bug 4101940
//
void CollationRegressionTest::Test4101940(/* char* par */)
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedCollator *c = NULL;
UnicodeString rules = "&9 < a < b";
UnicodeString nothing = "";
c = new RuleBasedCollator(rules, status);
if (c == NULL || U_FAILURE(status))
{
errln("Failed to create RuleBasedCollator");
delete c;
return;
}
CollationElementIterator *i = c->createCollationElementIterator(nothing);
i->reset();
if (i->next(status) != CollationElementIterator::NULLORDER)
{
errln("next did not return NULLORDER");
}
delete i;
delete c;
}
// @bug 4103436
//
// Collator::compare not handling spaces properly
//
void CollationRegressionTest::Test4103436(/* char* par */)
{
RuleBasedCollator *c = en_us->clone();
c->setStrength(Collator::TERTIARY);
static const UChar tests[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x66, 0x69, 0x6c, 0x65, 0}, {0x3c, 0}, {0x66, 0x69, 0x6c, 0x65, 0x20, 0x61, 0x63, 0x63, 0x65, 0x73, 0x73, 0},
{0x66, 0x69, 0x6c, 0x65, 0}, {0x3c, 0}, {0x66, 0x69, 0x6c, 0x65, 0x61, 0x63, 0x63, 0x65, 0x73, 0x73, 0}
};
compareArray(*c, tests, UPRV_LENGTHOF(tests));
delete c;
}
// @bug 4114076
//
// Collation not Unicode conformant with Hangul syllables
//
void CollationRegressionTest::Test4114076(/* char* par */)
{
UErrorCode status = U_ZERO_ERROR;
RuleBasedCollator *c = en_us->clone();
c->setStrength(Collator::TERTIARY);
//
// With Canonical decomposition, Hangul syllables should get decomposed
// into Jamo, but Jamo characters should not be decomposed into
// conjoining Jamo
//
static const UChar test1[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0xd4db, 0}, {0x3d, 0}, {0x1111, 0x1171, 0x11b6, 0}
};
c->setAttribute(UCOL_NORMALIZATION_MODE, UCOL_ON, status);
compareArray(*c, test1, UPRV_LENGTHOF(test1));
// From UTR #15:
// *In earlier versions of Unicode, jamo characters like ksf
// had compatibility mappings to kf + sf. These mappings were
// removed in Unicode 2.1.9 to ensure that Hangul syllables are maintained.)
// That is, the following test is obsolete as of 2.1.9
//obsolete- // With Full decomposition, it should go all the way down to
//obsolete- // conjoining Jamo characters.
//obsolete- //
//obsolete- static const UChar test2[][CollationRegressionTest::MAX_TOKEN_LEN] =
//obsolete- {
//obsolete- {0xd4db, 0}, {0x3d, 0}, {0x1111, 0x116e, 0x1175, 0x11af, 0x11c2, 0}
//obsolete- };
//obsolete-
//obsolete- c->setDecomposition(Normalizer::DECOMP_COMPAT);
//obsolete- compareArray(*c, test2, UPRV_LENGTHOF(test2));
delete c;
}
// @bug 4124632
//
// Collator::getCollationKey was hanging on certain character sequences
//
void CollationRegressionTest::Test4124632(/* char* par */)
{
UErrorCode status = U_ZERO_ERROR;
Collator *coll = NULL;
coll = Collator::createInstance(Locale::getJapan(), status);
if (coll == NULL || U_FAILURE(status))
{
errln("Failed to create collator for Locale::JAPAN");
delete coll;
return;
}
static const UChar test[] = {0x41, 0x0308, 0x62, 0x63, 0};
CollationKey key;
coll->getCollationKey(test, key, status);
if (key.isBogus() || U_FAILURE(status))
{
errln("CollationKey creation failed.");
}
delete coll;
}
// @bug 4132736
//
// sort order of french words with multiple accents has errors
//
void CollationRegressionTest::Test4132736(/* char* par */)
{
UErrorCode status = U_ZERO_ERROR;
Collator *c = NULL;
c = Collator::createInstance(Locale::getCanadaFrench(), status);
c->setStrength(Collator::TERTIARY);
if (c == NULL || U_FAILURE(status))
{
errln("Failed to create a collator for Locale::getCanadaFrench()");
delete c;
return;
}
static const UChar test1[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x65, 0x0300, 0x65, 0x0301, 0}, {0x3c, 0}, {0x65, 0x0301, 0x65, 0x0300, 0},
{0x65, 0x0300, 0x0301, 0}, {0x3c, 0}, {0x65, 0x0301, 0x0300, 0}
};
compareArray(*c, test1, UPRV_LENGTHOF(test1));
delete c;
}
// @bug 4133509
//
// The sorting using java.text.CollationKey is not in the exact order
//
void CollationRegressionTest::Test4133509(/* char* par */)
{
static const UChar test1[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x45, 0x78, 0x63, 0x65, 0x70, 0x74, 0x69, 0x6f, 0x6e, 0}, {0x3c, 0}, {0x45, 0x78, 0x63, 0x65, 0x70, 0x74, 0x69, 0x6f, 0x6e, 0x49, 0x6e, 0x49, 0x6e, 0x69, 0x74, 0x69, 0x61, 0x6c, 0x69, 0x7a, 0x65, 0x72, 0x45, 0x72, 0x72, 0x6f, 0x72, 0},
{0x47, 0x72, 0x61, 0x70, 0x68, 0x69, 0x63, 0x73, 0}, {0x3c, 0}, {0x47, 0x72, 0x61, 0x70, 0x68, 0x69, 0x63, 0x73, 0x45, 0x6e, 0x76, 0x69, 0x72, 0x6f, 0x6e, 0x6d, 0x65, 0x6e, 0x74, 0},
{0x53, 0x74, 0x72, 0x69, 0x6e, 0x67, 0}, {0x3c, 0}, {0x53, 0x74, 0x72, 0x69, 0x6e, 0x67, 0x42, 0x75, 0x66, 0x66, 0x65, 0x72, 0}
};
compareArray(*en_us, test1, UPRV_LENGTHOF(test1));
}
// @bug 4114077
//
// Collation with decomposition off doesn't work for Europe
//
void CollationRegressionTest::Test4114077(/* char* par */)
{
// Ensure that we get the same results with decomposition off
// as we do with it on....
UErrorCode status = U_ZERO_ERROR;
RuleBasedCollator *c = en_us->clone();
c->setStrength(Collator::TERTIARY);
static const UChar test1[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x00C0, 0}, {0x3d, 0}, {0x41, 0x0300, 0}, // Should be equivalent
{0x70, 0x00ea, 0x63, 0x68, 0x65, 0}, {0x3e, 0}, {0x70, 0x00e9, 0x63, 0x68, 0x00e9, 0},
{0x0204, 0}, {0x3d, 0}, {0x45, 0x030F, 0},
{0x01fa, 0}, {0x3d, 0}, {0x41, 0x030a, 0x0301, 0}, // a-ring-acute -> a-ring, acute
// -> a, ring, acute
{0x41, 0x0300, 0x0316, 0}, {0x3c, 0}, {0x41, 0x0316, 0x0300, 0} // No reordering --> unequal
};
c->setAttribute(UCOL_NORMALIZATION_MODE, UCOL_OFF, status);
compareArray(*c, test1, UPRV_LENGTHOF(test1));
static const UChar test2[][CollationRegressionTest::MAX_TOKEN_LEN] =
{
{0x41, 0x0300, 0x0316, 0}, {0x3d, 0}, {0x41, 0x0316, 0x0300, 0} // Reordering --> equal
};
c->setAttribute(UCOL_NORMALIZATION_MODE, UCOL_ON, status);
compareArray(*c, test2, UPRV_LENGTHOF(test2));
delete c;
}
// @bug 4141640
//
// Support for Swedish gone in 1.1.6 (Can't create Swedish collator)
//
void CollationRegressionTest::Test4141640(/* char* par */)
{
//
// Rather than just creating a Swedish collator, we might as well
// try to instantiate one for every locale available on the system
// in order to prevent this sort of bug from cropping up in the future
//
UErrorCode status = U_ZERO_ERROR;
int32_t i, localeCount;
const Locale *locales = Locale::getAvailableLocales(localeCount);
for (i = 0; i < localeCount; i += 1)
{
Collator *c = NULL;
status = U_ZERO_ERROR;
c = Collator::createInstance(locales[i], status);
if (c == NULL || U_FAILURE(status))
{
UnicodeString msg, localeName;
msg += "Could not create collator for locale ";
msg += locales[i].getName();
errln(msg);
}
delete c;
}
}
// @bug 4139572
//
// getCollationKey throws exception for spanish text
// Cannot reproduce this bug on 1.2, however it DOES fail on 1.1.6
//
void CollationRegressionTest::Test4139572(/* char* par */)
{
//
// Code pasted straight from the bug report
// (and then translated to C++ ;-)
//
// create spanish locale and collator
UErrorCode status = U_ZERO_ERROR;
Locale l("es", "es");
Collator *col = NULL;
col = Collator::createInstance(l, status);
if (col == NULL || U_FAILURE(status))
{
errln("Failed to create a collator for es_es locale.");
delete col;
return;
}
CollationKey key;
// this spanish phrase kills it!
col->getCollationKey("Nombre De Objeto", key, status);
if (key.isBogus() || U_FAILURE(status))
{
errln("Error creating CollationKey for \"Nombre De Ojbeto\"");
}
delete col;
}
/* HSYS : RuleBasedCollator::compare() performance enhancements
compare() does not create CollationElementIterator() anymore.*/
class My4146160Collator : public RuleBasedCollator
{
public:
My4146160Collator(RuleBasedCollator &rbc, UErrorCode &status);
~My4146160Collator();
CollationElementIterator *createCollationElementIterator(const UnicodeString &text) const;
CollationElementIterator *createCollationElementIterator(const CharacterIterator &text) const;
static int32_t count;
};
int32_t My4146160Collator::count = 0;
My4146160Collator::My4146160Collator(RuleBasedCollator &rbc, UErrorCode &status)
: RuleBasedCollator(rbc.getRules(), status)
{
}
My4146160Collator::~My4146160Collator()
{
}
CollationElementIterator *My4146160Collator::createCollationElementIterator(const UnicodeString &text) const
{
count += 1;
return RuleBasedCollator::createCollationElementIterator(text);
}
CollationElementIterator *My4146160Collator::createCollationElementIterator(const CharacterIterator &text) const
{
count += 1;
return RuleBasedCollator::createCollationElementIterator(text);
}
// @bug 4146160
//
// RuleBasedCollator doesn't use createCollationElementIterator internally
//
void CollationRegressionTest::Test4146160(/* char* par */)
{
#if 0
//
// Use a custom collator class whose createCollationElementIterator
// methods increment a count....
//
UErrorCode status = U_ZERO_ERROR;
CollationKey key;
My4146160Collator::count = 0;
My4146160Collator *mc = NULL;
mc = new My4146160Collator(*en_us, status);
if (mc == NULL || U_FAILURE(status))
{
errln("Failed to create a My4146160Collator.");
delete mc;
return;
}
mc->getCollationKey("1", key, status);
if (key.isBogus() || U_FAILURE(status))
{
errln("Failure to get a CollationKey from a My4146160Collator.");
delete mc;
return;
}
if (My4146160Collator::count < 1)
{
errln("My4146160Collator::createCollationElementIterator not called for getCollationKey");
}
My4146160Collator::count = 0;
mc->compare("1", "2");
if (My4146160Collator::count < 1)
{
errln("My4146160Collator::createtCollationElementIterator not called for compare");
}
delete mc;
#endif
}
void CollationRegressionTest::Test4179216() {
// you can position a CollationElementIterator in the middle of
// a contracting character sequence, yielding a bogus collation
// element
IcuTestErrorCode errorCode(*this, "Test4179216");
RuleBasedCollator coll(en_us->getRules() + " & C < ch , cH , Ch , CH < cat < crunchy", errorCode);
UnicodeString testText = "church church catcatcher runcrunchynchy";
CollationElementIterator *iter = coll.createCollationElementIterator(testText);
// test that the "ch" combination works properly
iter->setOffset(4, errorCode);
int32_t elt4 = CollationElementIterator::primaryOrder(iter->next(errorCode));
iter->reset();
int32_t elt0 = CollationElementIterator::primaryOrder(iter->next(errorCode));
iter->setOffset(5, errorCode);
int32_t elt5 = CollationElementIterator::primaryOrder(iter->next(errorCode));
// Compares and prints only 16-bit primary weights.
if (elt4 != elt0 || elt5 != elt0) {
errln("The collation elements at positions 0 (0x%04x), "
"4 (0x%04x), and 5 (0x%04x) don't match.",
elt0, elt4, elt5);
}
// test that the "cat" combination works properly
iter->setOffset(14, errorCode);
int32_t elt14 = CollationElementIterator::primaryOrder(iter->next(errorCode));
iter->setOffset(15, errorCode);
int32_t elt15 = CollationElementIterator::primaryOrder(iter->next(errorCode));
iter->setOffset(16, errorCode);
int32_t elt16 = CollationElementIterator::primaryOrder(iter->next(errorCode));
iter->setOffset(17, errorCode);
int32_t elt17 = CollationElementIterator::primaryOrder(iter->next(errorCode));
iter->setOffset(18, errorCode);
int32_t elt18 = CollationElementIterator::primaryOrder(iter->next(errorCode));
iter->setOffset(19, errorCode);
int32_t elt19 = CollationElementIterator::primaryOrder(iter->next(errorCode));
// Compares and prints only 16-bit primary weights.
if (elt14 != elt15 || elt14 != elt16 || elt14 != elt17
|| elt14 != elt18 || elt14 != elt19) {
errln("\"cat\" elements don't match: elt14 = 0x%04x, "
"elt15 = 0x%04x, elt16 = 0x%04x, elt17 = 0x%04x, "
"elt18 = 0x%04x, elt19 = 0x%04x",
elt14, elt15, elt16, elt17, elt18, elt19);
}
// now generate a complete list of the collation elements,
// first using next() and then using setOffset(), and
// make sure both interfaces return the same set of elements
iter->reset();
int32_t elt = iter->next(errorCode);
int32_t count = 0;
while (elt != CollationElementIterator::NULLORDER) {
++count;
elt = iter->next(errorCode);
}
LocalArray<UnicodeString> nextElements(new UnicodeString[count]);
LocalArray<UnicodeString> setOffsetElements(new UnicodeString[count]);
int32_t lastPos = 0;
iter->reset();
elt = iter->next(errorCode);
count = 0;
while (elt != CollationElementIterator::NULLORDER) {
nextElements[count++] = testText.tempSubStringBetween(lastPos, iter->getOffset());
lastPos = iter->getOffset();
elt = iter->next(errorCode);
}
int32_t nextElementsLength = count;
count = 0;
for (int32_t i = 0; i < testText.length(); ) {
iter->setOffset(i, errorCode);
lastPos = iter->getOffset();
elt = iter->next(errorCode);
setOffsetElements[count++] = testText.tempSubStringBetween(lastPos, iter->getOffset());
i = iter->getOffset();
}
for (int32_t i = 0; i < nextElementsLength; i++) {
if (nextElements[i] == setOffsetElements[i]) {
logln(nextElements[i]);
} else {
errln(UnicodeString("Error: next() yielded ") + nextElements[i] +
", but setOffset() yielded " + setOffsetElements[i]);
}
}
delete iter;
}
// Ticket 7189
//
// nextSortKeyPart incorrect for EO_S1 collation
static int32_t calcKeyIncremental(UCollator *coll, const UChar* text, int32_t len, uint8_t *keyBuf, int32_t /*keyBufLen*/, UErrorCode& status) {
UCharIterator uiter;
uint32_t state[2] = { 0, 0 };
int32_t keyLen;
int32_t count = 8;
uiter_setString(&uiter, text, len);
keyLen = 0;
while (TRUE) {
int32_t keyPartLen = ucol_nextSortKeyPart(coll, &uiter, state, &keyBuf[keyLen], count, &status);
if (U_FAILURE(status)) {
return -1;
}
if (keyPartLen == 0) {
break;
}
keyLen += keyPartLen;
}
return keyLen;
}
void CollationRegressionTest::TestT7189() {
UErrorCode status = U_ZERO_ERROR;
UCollator *coll;
uint32_t i;
static const UChar text1[][CollationRegressionTest::MAX_TOKEN_LEN] = {
// "Achter De Hoven"
{ 0x41, 0x63, 0x68, 0x74, 0x65, 0x72, 0x20, 0x44, 0x65, 0x20, 0x48, 0x6F, 0x76, 0x65, 0x6E, 0x00 },
// "ABC"
{ 0x41, 0x42, 0x43, 0x00 },
// "HELLO world!"
{ 0x48, 0x45, 0x4C, 0x4C, 0x4F, 0x20, 0x77, 0x6F, 0x72, 0x6C, 0x64, 0x21, 0x00 }
};
static const UChar text2[][CollationRegressionTest::MAX_TOKEN_LEN] = {
// "Achter de Hoven"
{ 0x41, 0x63, 0x68, 0x74, 0x65, 0x72, 0x20, 0x64, 0x65, 0x20, 0x48, 0x6F, 0x76, 0x65, 0x6E, 0x00 },
// "abc"
{ 0x61, 0x62, 0x63, 0x00 },
// "hello world!"
{ 0x68, 0x65, 0x6C, 0x6C, 0x6F, 0x20, 0x77, 0x6F, 0x72, 0x6C, 0x64, 0x21, 0x00 }
};
// Open the collator
coll = ucol_openFromShortString("EO_S1", FALSE, NULL, &status);
if (U_FAILURE(status)) {
errln("Failed to create a collator for short string EO_S1");
return;
}
for (i = 0; i < UPRV_LENGTHOF(text1); i++) {
uint8_t key1[100], key2[100];
int32_t len1, len2;
len1 = calcKeyIncremental(coll, text1[i], -1, key1, sizeof(key1), status);
if (U_FAILURE(status)) {
errln(UnicodeString("Failed to get a partial collation key for ") + text1[i]);
break;
}
len2 = calcKeyIncremental(coll, text2[i], -1, key2, sizeof(key2), status);
if (U_FAILURE(status)) {
errln(UnicodeString("Failed to get a partial collation key for ") + text2[i]);
break;
}
if (len1 == len2 && uprv_memcmp(key1, key2, len1) == 0) {
errln(UnicodeString("Failed: Identical key\n") + " text1: " + text1[i] + "\n" + " text2: " + text2[i] + "\n" + " key : " + TestUtility::hex(key1, len1));
} else {
logln(UnicodeString("Keys produced -\n") + " text1: " + text1[i] + "\n" + " key1 : " + TestUtility::hex(key1, len1) + "\n" + " text2: " + text2[i] + "\n" + " key2 : "
+ TestUtility::hex(key2, len2));
}
}
ucol_close(coll);
}
void CollationRegressionTest::TestCaseFirstCompression() {
RuleBasedCollator *col = en_us->clone();
UErrorCode status = U_ZERO_ERROR;
// default
caseFirstCompressionSub(col, "default");
// Upper first
col->setAttribute(UCOL_CASE_FIRST, UCOL_UPPER_FIRST, status);
if (U_FAILURE(status)) {
errln("Failed to set UCOL_UPPER_FIRST");
return;
}
caseFirstCompressionSub(col, "upper first");
// Lower first
col->setAttribute(UCOL_CASE_FIRST, UCOL_LOWER_FIRST, status);
if (U_FAILURE(status)) {
errln("Failed to set UCOL_LOWER_FIRST");
return;
}
caseFirstCompressionSub(col, "lower first");
delete col;
}
void CollationRegressionTest::caseFirstCompressionSub(Collator *col, UnicodeString opt) {
const int32_t maxLength = 50;
UChar str1[maxLength];
UChar str2[maxLength];
CollationKey key1, key2;
for (int32_t len = 1; len <= maxLength; len++) {
int32_t i = 0;
for (; i < len - 1; i++) {
str1[i] = str2[i] = (UChar)0x61; // 'a'
}
str1[i] = (UChar)0x41; // 'A'
str2[i] = (UChar)0x61; // 'a'
UErrorCode status = U_ZERO_ERROR;
col->getCollationKey(str1, len, key1, status);
col->getCollationKey(str2, len, key2, status);
UCollationResult cmpKey = key1.compareTo(key2, status);
UCollationResult cmpCol = col->compare(str1, len, str2, len, status);
if (U_FAILURE(status)) {
errln("Error in caseFirstCompressionSub");
} else if (cmpKey != cmpCol) {
errln((UnicodeString)"Inconsistent comparison(" + opt
+ "): str1=" + UnicodeString(str1, len) + ", str2=" + UnicodeString(str2, len)
+ ", cmpKey=" + cmpKey + ", cmpCol=" + cmpCol);
}
}
}
void CollationRegressionTest::TestTrailingComment() {
// ICU ticket #8070:
// Check that the rule parser handles a comment without terminating end-of-line.
IcuTestErrorCode errorCode(*this, "TestTrailingComment");
RuleBasedCollator coll(UNICODE_STRING_SIMPLE("&c<b#comment1\n<a#comment2"), errorCode);
UnicodeString a((UChar)0x61), b((UChar)0x62), c((UChar)0x63);
assertTrue("c<b", coll.compare(c, b) < 0);
assertTrue("b<a", coll.compare(b, a) < 0);
}
void CollationRegressionTest::TestBeforeWithTooStrongAfter() {
// ICU ticket #9959:
// Forbid rules with a before-reset followed by a stronger relation.
IcuTestErrorCode errorCode(*this, "TestBeforeWithTooStrongAfter");
RuleBasedCollator before2(UNICODE_STRING_SIMPLE("&[before 2]x<<q<p"), errorCode);
if(errorCode.isSuccess()) {
errln("should forbid before-2-reset followed by primary relation");
} else {
errorCode.reset();
}
RuleBasedCollator before3(UNICODE_STRING_SIMPLE("&[before 3]x<<<q<<s<p"), errorCode);
if(errorCode.isSuccess()) {
errln("should forbid before-3-reset followed by primary or secondary relation");
} else {
errorCode.reset();
}
}
void CollationRegressionTest::compareArray(Collator &c,
const UChar tests[][CollationRegressionTest::MAX_TOKEN_LEN],
int32_t testCount)
{
int32_t i;
Collator::EComparisonResult expectedResult = Collator::EQUAL;
for (i = 0; i < testCount; i += 3)
{
UnicodeString source(tests[i]);
UnicodeString comparison(tests[i + 1]);
UnicodeString target(tests[i + 2]);
if (comparison == "<")
{
expectedResult = Collator::LESS;
}
else if (comparison == ">")
{
expectedResult = Collator::GREATER;
}
else if (comparison == "=")
{
expectedResult = Collator::EQUAL;
}
else
{
UnicodeString bogus1("Bogus comparison string \"");
UnicodeString bogus2("\"");
errln(bogus1 + comparison + bogus2);
}
Collator::EComparisonResult compareResult = c.compare(source, target);
CollationKey sourceKey, targetKey;
UErrorCode status = U_ZERO_ERROR;
c.getCollationKey(source, sourceKey, status);
if (U_FAILURE(status))
{
errln("Couldn't get collationKey for source");
continue;
}
c.getCollationKey(target, targetKey, status);
if (U_FAILURE(status))
{
errln("Couldn't get collationKey for target");
continue;
}
Collator::EComparisonResult keyResult = sourceKey.compareTo(targetKey);
reportCResult( source, target, sourceKey, targetKey, compareResult, keyResult, compareResult, expectedResult );
}
}
void CollationRegressionTest::assertEqual(CollationElementIterator &i1, CollationElementIterator &i2)
{
int32_t c1, c2, count = 0;
UErrorCode status = U_ZERO_ERROR;
do
{
c1 = i1.next(status);
c2 = i2.next(status);
if (c1 != c2)
{
UnicodeString msg, msg1(" ");
msg += msg1 + count;
msg += ": strength(0x";
appendHex(c1, 8, msg);
msg += ") != strength(0x";
appendHex(c2, 8, msg);
msg += ")";
errln(msg);
break;
}
count += 1;
}
while (c1 != CollationElementIterator::NULLORDER);
}
void CollationRegressionTest::runIndexedTest(int32_t index, UBool exec, const char* &name, char* /* par */)
{
if (exec)
{
logln("Collation Regression Tests: ");
}
if(en_us == NULL) {
dataerrln("Class collator not instantiated");
name = "";
return;
}
TESTCASE_AUTO_BEGIN;
TESTCASE_AUTO(Test4048446);
TESTCASE_AUTO(Test4051866);
TESTCASE_AUTO(Test4053636);
TESTCASE_AUTO(Test4054238);
TESTCASE_AUTO(Test4054734);
TESTCASE_AUTO(Test4054736);
TESTCASE_AUTO(Test4058613);
TESTCASE_AUTO(Test4059820);
TESTCASE_AUTO(Test4060154);
TESTCASE_AUTO(Test4062418);
TESTCASE_AUTO(Test4065540);
TESTCASE_AUTO(Test4066189);
TESTCASE_AUTO(Test4066696);
TESTCASE_AUTO(Test4076676);
TESTCASE_AUTO(Test4078588);
TESTCASE_AUTO(Test4079231);
TESTCASE_AUTO(Test4081866);
TESTCASE_AUTO(Test4087241);
TESTCASE_AUTO(Test4087243);
TESTCASE_AUTO(Test4092260);
TESTCASE_AUTO(Test4095316);
TESTCASE_AUTO(Test4101940);
TESTCASE_AUTO(Test4103436);
TESTCASE_AUTO(Test4114076);
TESTCASE_AUTO(Test4114077);
TESTCASE_AUTO(Test4124632);
TESTCASE_AUTO(Test4132736);
TESTCASE_AUTO(Test4133509);
TESTCASE_AUTO(Test4139572);
TESTCASE_AUTO(Test4141640);
TESTCASE_AUTO(Test4146160);
TESTCASE_AUTO(Test4179216);
TESTCASE_AUTO(TestT7189);
TESTCASE_AUTO(TestCaseFirstCompression);
TESTCASE_AUTO(TestTrailingComment);
TESTCASE_AUTO(TestBeforeWithTooStrongAfter);
TESTCASE_AUTO_END;
}
#endif /* #if !UCONFIG_NO_COLLATION */