/******************************************************************** * COPYRIGHT: * Copyright (c) 1999-2000, International Business Machines Corporation and * others. All Rights Reserved. ********************************************************************/ /************************************************************************ * Date Name Description * 12/15/99 Madhu Creation. * 01/12/2000 Madhu Updated for changed API and added new tests ************************************************************************/ #include "intltest.h" #include "unicode/brkiter.h" #include "unicode/rbbi.h" #include "unicode/unicode.h" #include #include "unicode/utypes.h" #include "rbbitst.h" #include #include "unicode/schriter.h" //-------------------------------------------------------------------------------------- /** * "Vector" class for holding test tables * (this class is actually a linked list, but we use the name and API of the * java.util.Vector class to keep as much of our test code as possible the same.) */ class Enumeration { // text enumeration public: virtual bool_t hasMoreElements() = 0; virtual UnicodeString nextElement() = 0; }; class Vector { // text vector public: class TextLink { public: TextLink() : fLink(0), fText() {} TextLink(TextLink* link, UnicodeString text) : fLink(link), fText(text) {} TextLink* fLink; UnicodeString fText; }; public: TextLink fBase; TextLink* fEnd; int32_t fSize; public: class VectorEnumeration : public Enumeration { public: VectorEnumeration(Vector* vector) : fVector(vector), fPos(&vector->fBase) {} bool_t hasMoreElements() { return fPos->fLink != &fVector->fBase; } UnicodeString nextElement() { fPos = fPos->fLink; return fPos->fText; } Vector* fVector; TextLink* fPos; }; Vector() : fBase(), fEnd(&fBase), fSize(0) { fBase.fLink = &fBase; } ~Vector() { while (fBase.fLink != &fBase) { TextLink* link = fBase.fLink; fBase.fLink = link->fLink; delete link; } } void addElement(UnicodeString text) { fEnd->fLink = new TextLink(&fBase, text); fEnd = fEnd->fLink; ++fSize; } void insertElementAt(UnicodeString text, int pos) { if(pos >= fSize || pos < 0) ; else if(pos == 0){ TextLink* insert = new TextLink(&fBase, text); insert->fLink=fBase.fLink; ++fSize; fBase.fLink=insert; } else{ TextLink* link = fBase.fLink; while(--pos > 0) link=link->fLink; TextLink* insert = new TextLink(&fBase, text); insert->fLink =link->fLink; link->fLink=insert; ++fSize; } } UnicodeString elementAt(int32_t pos) { if (pos >= fSize) return UnicodeString(); TextLink* link = fBase.fLink; while (pos-- > 0) link = link->fLink; return link->fText; } UnicodeString lastElement() { return fEnd == &fBase ? UnicodeString() : fEnd->fText; } int32_t size() { return fSize; } Enumeration* elements() { return new VectorEnumeration(this); } }; //-------------------------------------------------------------------------------------- /** * RBBITest is medium top level test class RuleBasedBreakIterator */ const UnicodeString halfNA = CharsToUnicodeString("\\u0928\\u094d\\u200d"); //halfform NA = devanigiri NA + virama(supresses inherent vowel)+ zero width joiner const UnicodeString halfSA = CharsToUnicodeString("\\u0938\\u094d\\u200d"); const UnicodeString halfCHA = CharsToUnicodeString("\\u091a\\u094d\\u200d"); const UnicodeString halfKA = CharsToUnicodeString("\\u0915\\u094d\\u200d"); const UnicodeString deadTA = CharsToUnicodeString("\\u0924\\u094d"); //-------------------------------------------------------------------- //tests default rules based character iteration //-------------------------------------------------------------------- void RBBITest::TestDefaultRuleBasedCharacterIteration() { // RuleBasedBreakIterator* rbbi=(RuleBasedBreakIterator*)RuleBasedBreakIterator::createCharacterInstance(); logln((UnicodeString)"Testing the RBBI for character iteration by using default rules"); //fetch the rules used to create the above RuleBasedBreakIterator // UnicodeString defaultRules=rbbi->getRules(); // RuleBasedCharacterIterator charIterDefault = new RuleBasedBreakIterator(defaultRules); UErrorCode status=U_ZERO_ERROR; RuleBasedBreakIterator* charIterDefault=(RuleBasedBreakIterator*)RuleBasedBreakIterator::createCharacterInstance(Locale::getDefault(), status); if(U_FAILURE(status)){ errln("FAIL : in construction"); return; } Vector *chardata = new Vector(); chardata->addElement("H"); chardata->addElement("e"); chardata->addElement("l"); chardata->addElement("l"); chardata->addElement("o"); chardata->addElement(CharsToUnicodeString("e\\u0301")); //acuteE chardata->addElement("&"); chardata->addElement(CharsToUnicodeString("e\\u0303")); //tildaE //devanagiri characters for Hindi support chardata->addElement(CharsToUnicodeString("\\u0906")); //devanagiri AA chardata->addElement(CharsToUnicodeString("\\u093e\\u0901")); //devanagiri vowelsign AA+ chandrabindhu chardata->addElement(CharsToUnicodeString("\\u0916\\u0947")); //devanagiri KHA+vowelsign E chardata->addElement(CharsToUnicodeString("\\u0938\\u0941\\u0902")); //devanagiri SA+vowelsign U + anusvara(bindu) chardata->addElement(CharsToUnicodeString("\\u0926")); //devanagiri consonant DA chardata->addElement(CharsToUnicodeString("\\u0930")); //devanagiri consonant RA chardata->addElement(CharsToUnicodeString("\\u0939\\u094c")); //devanagiri HA+vowel sign AI chardata->addElement(CharsToUnicodeString("\\u0964")); //devanagiri danda //end hindi characters chardata->addElement(CharsToUnicodeString("A\\u0302")); //circumflexA chardata->addElement(CharsToUnicodeString("i\\u0301")); //acuteBelowI // conjoining jamo->.. chardata->addElement(CharsToUnicodeString("\\u1109\\u1161\\u11bc")); chardata->addElement(CharsToUnicodeString("\\u1112\\u1161\\u11bc")); chardata->addElement("\n"); chardata->addElement("\r\n"); //keep CRLF sequences together chardata->addElement(CharsToUnicodeString("S\\u0300")); //graveS chardata->addElement(CharsToUnicodeString("i\\u0301")); //acuteBelowI chardata->addElement("!"); // What follows is a string of Korean characters (I found it in the Yellow Pages // ad for the Korean Presbyterian Church of San Francisco, and I hope I transcribed // it correctly), first as precomposed syllables, and then as conjoining jamo. // Both sequences should be semantically identical and break the same way. // precomposed syllables... chardata->addElement(CharsToUnicodeString("\\uc0c1")); chardata->addElement(CharsToUnicodeString("\\ud56d")); chardata->addElement(" "); chardata->addElement(CharsToUnicodeString("\\ud55c")); chardata->addElement(CharsToUnicodeString("\\uc778")); chardata->addElement(" "); chardata->addElement(CharsToUnicodeString("\\uc5f0")); chardata->addElement(CharsToUnicodeString("\\ud569")); chardata->addElement(" "); chardata->addElement(CharsToUnicodeString("\\uc7a5")); chardata->addElement(CharsToUnicodeString("\\ub85c")); chardata->addElement(CharsToUnicodeString("\\uad50")); chardata->addElement(CharsToUnicodeString("\\ud68c")); chardata->addElement(" "); // conjoining jamo... chardata->addElement(CharsToUnicodeString("\\u1109\\u1161\\u11bc")); chardata->addElement(CharsToUnicodeString("\\u1112\\u1161\\u11bc")); chardata->addElement(" "); chardata->addElement(CharsToUnicodeString("\\u1112\\u1161\\u11ab")); chardata->addElement(CharsToUnicodeString("\\u110b\\u1175\\u11ab")); chardata->addElement(" "); chardata->addElement(CharsToUnicodeString("\\u110b\\u1167\\u11ab")); chardata->addElement(CharsToUnicodeString("\\u1112\\u1161\\u11b8")); chardata->addElement(" "); chardata->addElement(CharsToUnicodeString("\\u110c\\u1161\\u11bc")); chardata->addElement(CharsToUnicodeString("\\u1105\\u1169")); chardata->addElement(CharsToUnicodeString("\\u1100\\u116d")); chardata->addElement(CharsToUnicodeString("\\u1112\\u116c")); generalIteratorTest(*charIterDefault, chardata); delete charIterDefault; delete chardata; // delete rbbi; } //-------------------------------------------------------------------- //tests default rules based word iteration //-------------------------------------------------------------------- void RBBITest::TestDefaultRuleBasedWordIteration() { logln((UnicodeString)"Testing the RBBI for word iteration using default rules"); // RuleBasedBreakIterator *rbbi=(RuleBasedBreakIterator*)RuleBasedBreakIterator::createWordInstance(); //fetch the rules used to create the above RuleBasedBreakIterator // UnicodeString defaultRules=rbbi->getRules(); // RuleBasedBreakIterator wordIterDefault = new RuleBasedBreakIterator(defaultRules); UErrorCode status=U_ZERO_ERROR; RuleBasedBreakIterator* wordIterDefault=(RuleBasedBreakIterator*)RuleBasedBreakIterator::createWordInstance(Locale::getDefault(), status); if(U_FAILURE(status)){ errln("FAIL : in construction"); return; } Vector *worddata = new Vector(); worddata->addElement ("Write"); worddata->addElement (" "); worddata->addElement ("wordrules"); worddata->addElement ("."); worddata->addElement(" "); worddata->addElement("alpha-beta-gamma"); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u092f\\u0939")); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u0939\\u093f") + halfNA + CharsToUnicodeString("\\u0926\\u0940")); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u0939\\u0948")); //worddata->addElement("\\u0964"); //danda followed by a space "\u0964->danda: hindi phrase seperator" worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u0905\\u093e\\u092a")); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u0938\\u093f\\u0916\\u094b\\u0917\\u0947")); worddata->addElement("?"); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u0968\\u0966.\\u0969\\u096f")); //hindi numbers worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u0967\\u0966\\u0966.\\u0966\\u0966%")); //postnumeric worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u20a8\\u0967,\\u0967\\u0966\\u0966.\\u0966\\u0966")); //pre-number India currency symbol Rs->\\u20aD worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u0905\\u092e\\u091c")); worddata->addElement("\n"); worddata->addElement(halfSA+CharsToUnicodeString("\\u0935\\u0924\\u0902")+deadTA+CharsToUnicodeString("\\u0930")); worddata->addElement("\r"); worddata->addElement("It's"); worddata->addElement(" "); worddata->addElement("$30.10"); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u00A2")); //cent sign worddata->addElement(CharsToUnicodeString("\\u00A3")); //pound sign worddata->addElement(CharsToUnicodeString("\\u00A4")); //currency sign worddata->addElement(CharsToUnicodeString("\\u00A5")); //yen sign worddata->addElement("alpha-beta-gamma"); worddata->addElement(" "); worddata->addElement("Badges"); worddata->addElement("?"); worddata->addElement(" "); worddata->addElement("BADGES"); worddata->addElement("!"); worddata->addElement("1000,233,456.000"); worddata->addElement(" "); worddata->addElement("1,23.322%"); worddata->addElement(" "); worddata->addElement("123.1222"); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u0024123,000.20")); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("179.01\\u0025")); worddata->addElement("X"); worddata->addElement(" "); worddata->addElement("Now"); worddata->addElement("\r"); worddata->addElement("is"); worddata->addElement("\n"); worddata->addElement("the"); worddata->addElement("\r\n"); worddata->addElement("time"); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\uc5f0\\ud569")); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\uc7a5\\ub85c\\uad50\\ud68c")); worddata->addElement(" "); // conjoining jamo... worddata->addElement(CharsToUnicodeString("\\u1109\\u1161\\u11bc\\u1112\\u1161\\u11bc")); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u1112\\u1161\\u11ab\\u110b\\u1175\\u11ab")); worddata->addElement(" "); worddata->addElement("Hello"); worddata->addElement(","); worddata->addElement(" "); worddata->addElement("how"); worddata->addElement(" "); worddata->addElement("are"); worddata->addElement(" "); worddata->addElement("you"); worddata->addElement(" "); generalIteratorTest(*wordIterDefault, worddata); delete wordIterDefault; delete worddata; // delete rbbi; } //-------------------------------------------------------------------- //tests default rules based word iteration //-------------------------------------------------------------------- const UnicodeString kParagraphSeparator = CharsToUnicodeString("\\u2029"); const UnicodeString kLineSeparator = CharsToUnicodeString("\\u2028"); void RBBITest::TestDefaultRuleBasedSentenceIteration() { logln((UnicodeString)"Testing the RBBI for sentence iteration using default rules"); // RuleBasedBreakIterator *rbbi=(RuleBasedBreakIterator*)RuleBasedBreakIterator::createSentenceInstance(); //fetch the rules used to create the above RuleBasedBreakIterator // UnicodeString defaultRules=rbbi->getRules(); // RuleBasedBreakIterator sentIterDefault = new RuleBasedBreakIterator(defaultRules); UErrorCode status=U_ZERO_ERROR; RuleBasedBreakIterator* sentIterDefault=(RuleBasedBreakIterator*)RuleBasedBreakIterator::createSentenceInstance(Locale::getDefault(), status); if(U_FAILURE(status)){ errln("FAIL : in construction"); return; } Vector *sentdata = new Vector(); sentdata->addElement("(This is it.) "); sentdata->addElement("Testing the sentence iterator. "); sentdata->addElement("\"This isn\'t it.\" "); sentdata->addElement("Hi! "); sentdata = new Vector(); sentdata->addElement("This is a simple sample sentence. "); sentdata->addElement("(This is it.) "); sentdata->addElement("This is a simple sample sentence. "); sentdata->addElement("\"This isn\'t it.\" "); sentdata->addElement("Hi! "); sentdata->addElement("This is a simple sample sentence. "); sentdata->addElement("It does not have to make any sense as you can see. "); sentdata->addElement("Nel mezzo del cammin di nostra vita, mi ritrovai in una selva oscura. "); sentdata->addElement("Che la dritta via aveo smarrita. "); sentdata->addElement("He said, that I said, that you said!! "); sentdata->addElement("Don't rock the boat." + kParagraphSeparator); sentdata->addElement("Because I am the daddy, that is why. "); sentdata->addElement("Not on my time (el timo.)! "); sentdata->addElement("So what!!" + kParagraphSeparator); sentdata->addElement("\"But now,\" he said, \"I know!\" "); sentdata->addElement("Harris thumbed down several, including \"Away We Go\" (which became the huge success Oklahoma!). "); sentdata->addElement("One species, B. anthracis, is highly virulent.\n"); sentdata->addElement("Wolf said about Sounder:\"Beautifully thought-out and directed.\" "); sentdata->addElement("Have you ever said, \"This is where\tI shall live\"? "); sentdata->addElement("He answered, \"You may not!\" "); sentdata->addElement("Another popular saying is: \"How do you do?\". "); sentdata->addElement("Yet another popular saying is: \'I\'m fine thanks.\' "); sentdata->addElement("What is the proper use of the abbreviation pp.? "); sentdata->addElement("Yes, I am definatelly 12\" tall!!"); // test for bug #4113835: \n and \r count as spaces, not as paragraph breaks sentdata->addElement(CharsToUnicodeString("Now\ris\nthe\r\ntime\n\rfor\r\rall\\u2029")); // test that it doesn't break sentences at the boundary between CJK // and other letters sentdata->addElement(CharsToUnicodeString("\\u5487\\u67ff\\ue591\\u5017\\u61b3\\u60a1\\u9510\\u8165:\"JAVA\\u821c") + CharsToUnicodeString("\\u8165\\u7fc8\\u51ce\\u306d,\\u2494\\u56d8\\u4ec0\\u60b1\\u8560\\u51ba") + CharsToUnicodeString("\\u611d\\u57b6\\u2510\\u5d46\".\\u2029")); sentdata->addElement(CharsToUnicodeString("\\u5487\\u67ff\\ue591\\u5017\\u61b3\\u60a1\\u9510\\u8165\\u9de8") + CharsToUnicodeString("\\u97e4JAVA\\u821c\\u8165\\u7fc8\\u51ce\\u306d\\ue30b\\u2494\\u56d8\\u4ec0") + CharsToUnicodeString("\\u60b1\\u8560\\u51ba\\u611d\\u57b6\\u2510\\u5d46\\u97e5\\u7751\\u2029")); sentdata->addElement(CharsToUnicodeString("\\u5487\\u67ff\\ue591\\u5017\\u61b3\\u60a1\\u9510\\u8165\\u9de8\\u97e4") + CharsToUnicodeString("\\u6470\\u8790JAVA\\u821c\\u8165\\u7fc8\\u51ce\\u306d\\ue30b\\u2494\\u56d8") + CharsToUnicodeString("\\u4ec0\\u60b1\\u8560\\u51ba\\u611d\\u57b6\\u2510\\u5d46\\u97e5\\u7751\\u2029")); sentdata->addElement(CharsToUnicodeString("He said, \"I can go there.\"\\u2029")); // Treat fullwidth variants of .!? the same as their // normal counterparts sentdata->addElement(CharsToUnicodeString("I know I'm right\\uff0e ")); sentdata->addElement(CharsToUnicodeString("Right\\uff1f ")); sentdata->addElement(CharsToUnicodeString("Right\\uff01 ")); // Don't break sentences at boundary between CJK and digits sentdata->addElement(CharsToUnicodeString("\\u5487\\u67ff\\ue591\\u5017\\u61b3\\u60a1\\u9510\\u8165\\u9de8") + CharsToUnicodeString("\\u97e48888\\u821c\\u8165\\u7fc8\\u51ce\\u306d\\ue30b\\u2494\\u56d8\\u4ec0") + CharsToUnicodeString("\\u60b1\\u8560\\u51ba\\u611d\\u57b6\\u2510\\u5d46\\u97e5\\u7751\\u2029")); // Break sentence between a sentence terminator and // opening punctuation sentdata->addElement("How do you do?"); sentdata->addElement("(fine)."); //sentence breaks for hindi which used Devanagari script //make sure there is sentence break after ?,danda(hindi phrase separator),fullstop followed by space and no break after \n \r sentdata->addElement(CharsToUnicodeString("\\u0928\\u092e")+halfSA + CharsToUnicodeString("\\u0924\\u0947 ") + CharsToUnicodeString("\\u0930\\u092e\\u0947\\u0936, ") + CharsToUnicodeString("\\u0905\\u093e\\u092a") + CharsToUnicodeString("\\u0915\\u0948\\u0938\\u0947 ") + CharsToUnicodeString("\\u0939\\u0948?")); sentdata->addElement(CharsToUnicodeString("\\u092e\\u0948 \\u0905") + halfCHA +CharsToUnicodeString("\\u091b\\u093e \\u0939\\u0942\\u0901\\u0964 ")); sentdata->addElement(CharsToUnicodeString("\\u0905\\u093e\\u092a\r\n \\u0915\\u0948\\u0938\\u0947 \\u0939\\u0948?")); sentdata->addElement(CharsToUnicodeString("\\u0935\\u0939 ") + halfKA +CharsToUnicodeString("\\u092f\\u093e\n \\u0939\\u0948?")); sentdata->addElement(CharsToUnicodeString("\\u092f\\u0939 \\u0905\\u093e\\u092e \\u0939\\u0948. ")); sentdata->addElement(CharsToUnicodeString("\\u092f\\u0939 means \"this\". ")); sentdata->addElement(CharsToUnicodeString("\"\\u092a\\u095d\\u093e\\u0908\" meaning \"education\" or \"studies\". ")); sentdata->addElement(CharsToUnicodeString("\\u0905\\u093e\\u091c") + CharsToUnicodeString("(") + halfSA + CharsToUnicodeString("\\u0935\\u0924\\u0902") + deadTA+ CharsToUnicodeString("\\u0930 ") + CharsToUnicodeString("\\u0926\\u093f\\u0935\\u093e\\u0938) ") + CharsToUnicodeString("\\u0939\\u0948\\u0964 ")); sentdata->addElement("Let's end here. "); generalIteratorTest(*sentIterDefault, sentdata); delete sentIterDefault; delete sentdata; // delete rbbi; } //-------------------------------------------------------------------- //tests default rules based line iteration //-------------------------------------------------------------------- void RBBITest::TestDefaultRuleBasedLineIteration() { UErrorCode status= U_ZERO_ERROR; RuleBasedBreakIterator* lineIterDefault=(RuleBasedBreakIterator*)RuleBasedBreakIterator::createLineInstance(Locale::getDefault(), status); if(U_FAILURE(status)){ errln("FAIL : in construction"); return; } Vector *linedata = new Vector(); linedata->addElement("Multi-"); linedata->addElement("Level "); linedata->addElement("example "); linedata->addElement("of "); linedata->addElement("a "); linedata->addElement("semi-"); linedata->addElement("idiotic "); linedata->addElement("non-"); linedata->addElement("sensical "); linedata->addElement("(non-"); linedata->addElement("important) "); linedata->addElement("sentence. "); linedata->addElement("Hi "); linedata->addElement("Hello "); linedata->addElement("How\n"); linedata->addElement("are\r"); linedata->addElement("you" + kLineSeparator); linedata->addElement("fine.\t"); linedata->addElement("good. "); linedata->addElement("Now\r"); linedata->addElement("is\n"); linedata->addElement("the\r\n"); linedata->addElement("time\n"); linedata->addElement("\r"); linedata->addElement("for\r"); linedata->addElement("\r"); linedata->addElement("all"); // to test for bug #4068133 linedata->addElement(CharsToUnicodeString("\\u96f6")); linedata->addElement(CharsToUnicodeString("\\u4e00\\u3002")); linedata->addElement(CharsToUnicodeString("\\u4e8c\\u3001")); linedata->addElement(CharsToUnicodeString("\\u4e09\\u3002\\u3001")); linedata->addElement(CharsToUnicodeString("\\u56db\\u3001\\u3002\\u3001")); linedata->addElement(CharsToUnicodeString("\\u4e94,")); linedata->addElement(CharsToUnicodeString("\\u516d.")); linedata->addElement(CharsToUnicodeString("\\u4e03.\\u3001,\\u3002")); linedata->addElement(CharsToUnicodeString("\\u516b")); // to test for bug #4086052 linedata->addElement(CharsToUnicodeString("foo\\u00a0bar ")); // linedata->addElement("foo\\ufeffbar"); // to test for bug #4097920 linedata->addElement("dog,"); linedata->addElement("cat,"); linedata->addElement("mouse "); linedata->addElement("(one)"); linedata->addElement("(two)\n"); // to test for bug #4035266 linedata->addElement("The "); linedata->addElement("balance "); linedata->addElement("is "); linedata->addElement("$-23,456.78, "); linedata->addElement("not "); linedata->addElement("-$32,456.78!\n"); // to test for bug #4098467 // What follows is a string of Korean characters (I found it in the Yellow Pages // ad for the Korean Presbyterian Church of San Francisco, and I hope I transcribed // it correctly), first as precomposed syllables, and then as conjoining jamo. // Both sequences should be semantically identical and break the same way. // precomposed syllables... linedata->addElement(CharsToUnicodeString("\\uc0c1\\ud56d ")); linedata->addElement(CharsToUnicodeString("\\ud55c\\uc778 ")); linedata->addElement(CharsToUnicodeString("\\uc5f0\\ud569 ")); linedata->addElement(CharsToUnicodeString("\\uc7a5\\ub85c\\uad50\\ud68c ")); // conjoining jamo... linedata->addElement(CharsToUnicodeString("\\u1109\\u1161\\u11bc\\u1112\\u1161\\u11bc ")); linedata->addElement(CharsToUnicodeString("\\u1112\\u1161\\u11ab\\u110b\\u1175\\u11ab ")); linedata->addElement(CharsToUnicodeString("\\u110b\\u1167\\u11ab\\u1112\\u1161\\u11b8 ")); linedata->addElement(CharsToUnicodeString("\\u110c\\u1161\\u11bc\\u1105\\u1169\\u1100\\u116d\\u1112\\u116c")); // to test for bug #4117554: Fullwidth .!? should be treated as postJwrd linedata->addElement(CharsToUnicodeString("\\u4e01\\uff0e")); linedata->addElement(CharsToUnicodeString("\\u4e02\\uff01")); linedata->addElement(CharsToUnicodeString("\\u4e03\\uff1f")); generalIteratorTest(*lineIterDefault, linedata); delete lineIterDefault; delete linedata; } //-------------------------------------------------------------------- //Testing the BreakIterator for devanagari script //-------------------------------------------------------------------- const UnicodeString deadRA = CharsToUnicodeString("\\u0930\\u094d"); /*deadform RA = devanagari RA + virama*/ const UnicodeString deadPHA = CharsToUnicodeString("\\u092b\\u094d"); /*deadform PHA = devanagari PHA + virama*/ const UnicodeString deadTTHA= CharsToUnicodeString("\\u0920\\u094d"); const UnicodeString deadPA = CharsToUnicodeString("\\u092a\\u094d"); const UnicodeString deadSA = CharsToUnicodeString("\\u0938\\u094d"); const UnicodeString visarga = CharsToUnicodeString("\\u0903"); /*devanagari visarga looks like a english colon*/ void RBBITest::TestHindiCharacterBreak() { Vector *hindicharData = new Vector(); //devanagari characters for Hindi support hindicharData->addElement(CharsToUnicodeString("\\u0906")); //devanagari AA //hindi character break should make sure that it // doesn't break in-between a vowelsign and a chandrabindu hindicharData->addElement(CharsToUnicodeString("\\u093e\\u0901")); //devanagari vowelsign AA+ chandrabindu hindicharData->addElement(CharsToUnicodeString("\\u0916\\u0947")); //devanagari KHA+vowelsign E hindicharData->addElement(CharsToUnicodeString("\\u0938\\u0941\\u0902")); //devanagari SA+vowelsign U + anusvara(bindu) hindicharData->addElement(CharsToUnicodeString("\\u0926")); //devanagari consonant DA hindicharData->addElement(CharsToUnicodeString("\\u0930")); //devanagari consonant RA hindicharData->addElement(CharsToUnicodeString("\\u0939\\u094c")); //devanagari consonant HA+dependent vowel sign AI hindicharData->addElement(CharsToUnicodeString("\\u0964")); //devanagari danda hindicharData->addElement(CharsToUnicodeString("\\u0950")); //devanagari OM hindicharData->addElement(CharsToUnicodeString("\\u0915\\u0943")); //devanagari KA+dependent vowel RI->KRI //dependent half-forms hindicharData->addElement(halfSA+CharsToUnicodeString("\\u0924")); //halfSA+base consonant TA->STA hindicharData->addElement(halfSA+CharsToUnicodeString("\\u0925")); //halfSA+base consonant THA->STHA hindicharData->addElement(halfSA+CharsToUnicodeString("\\u092e")); //halfSA+base consonant MA->SMA hindicharData->addElement(halfCHA+CharsToUnicodeString("\\u091b")); //halfCHA+base consonant CHHA->CHHHA hindicharData->addElement(halfNA+CharsToUnicodeString("\\u0917")); //halfNA+base consonant GA->NGA hindicharData->addElement(CharsToUnicodeString("\\u092a\\u094d\\u200d\\u092f")); //halfPA(PA+virama+zerowidthjoiner+base consonant YA->PYA //consonant RA rules ---------- //if the dead consonant RA precedes either a consonant or an independent vowel, //then it is replaced by its superscript non-spacing mark hindicharData->addElement(deadRA+ CharsToUnicodeString("\\u0915")); //deadRA+devanagari consonant KA->KA+superRA hindicharData->addElement(deadRA+ CharsToUnicodeString("\\u0923")); //deadRA+devanagari consonant NNA->NNA+superRA hindicharData->addElement(deadRA+ CharsToUnicodeString("\\u0917")); //deadRA+devanagari consonant GA->GA+superRA // hindicharData->addElement(deadRA+ CharsToUnicodeString("\\u0960")); //deadRA+devanagari cosonant RRI->RRI+superRA //if any dead consonant(other than dead RA)precedes the consonant RA, then //it is replaced with its nominal forma nd RA is replaced by the subscript non-spacing mark. hindicharData->addElement(deadPHA+ CharsToUnicodeString("\\u0930")); //deadPHA+devanagari consonant RA->PHA+subRA hindicharData->addElement(deadPA+ CharsToUnicodeString("\\u0930")); //deadPA+devanagari consonant RA->PA+subRA hindicharData->addElement(deadTTHA+ CharsToUnicodeString("\\u0930")); //deadTTHA+devanagari consonant RA->TTHA+subRA hindicharData->addElement(deadTA+ CharsToUnicodeString("\\u0930")); //deadTA+RA->TRA hindicharData->addElement(CharsToUnicodeString("\\u0936\\u094d\\u0930")); //deadSHA(SHA+virama)+RA->SHRA //conjuct ligatures hindicharData->addElement(CharsToUnicodeString("\\u0915\\u094d\\u0937")); //deadKA(KA+virama) followed by SSHA wraps up into a single character KSSHA hindicharData->addElement(deadTA+CharsToUnicodeString("\\u0924")); //deadTA+TA wraps up into glyph TTHA hindicharData->addElement(CharsToUnicodeString("\\u0926\\u094d\\u0935")); //deadDA(DA+virama)+VA wraps up into DVA hindicharData->addElement(CharsToUnicodeString("\\u091c\\u094d\\u091e")); //deadJA(JA+virama)+NYA wraps up into JNYA UErrorCode status= U_ZERO_ERROR; RuleBasedBreakIterator *e=(RuleBasedBreakIterator*)RuleBasedBreakIterator::createCharacterInstance(Locale::getDefault(), status); if(U_FAILURE(status)){ errln("FAIL : in construction"); return; } generalIteratorTest(*e, hindicharData); delete e; delete hindicharData; } void RBBITest::TestHindiWordBreak() { Vector *hindiWordData = new Vector(); //hindi hindiWordData->addElement(CharsToUnicodeString("\\u0917\\u092a-\\u0936\\u092a")); hindiWordData->addElement("!"); hindiWordData->addElement(CharsToUnicodeString("\\u092f\\u0939")); hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u0939\\u093f") + halfNA + CharsToUnicodeString("\\u0926\\u0940")); hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u0939\\u0948")); //danda is similar to full stop. danda is a hindi phrase seperator //Make sure it breaks before danda and after danda when it is followed by a space //hindiWordData->addElement(CharsToUnicodeString("\\u0964")); //fails here doesn't break at danda hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u0905\\u093e\\u092a")); hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u0938\\u093f\\u0916\\u094b\\u0917\\u0947")); hindiWordData->addElement("?"); hindiWordData->addElement("\n"); hindiWordData->addElement(":"); hindiWordData->addElement(deadPA+CharsToUnicodeString("\\u0930\\u093e\\u092f")+visarga); //no break before visarga hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u0935") + deadRA+ CharsToUnicodeString("\\u0937\\u093e")); hindiWordData->addElement("\r\n"); hindiWordData->addElement(deadPA+ CharsToUnicodeString("\\u0930\\u0915\\u093e\\u0936")); //deadPA+RA+KA+vowel AA+SHA -> prakash hindiWordData->addElement(","); hindiWordData->addElement(CharsToUnicodeString("\\u0924\\u0941\\u092e\\u093e\\u0930\\u094b")); hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u092e\\u093f")+ deadTA+ CharsToUnicodeString("\\u0930")); //MA+vowel I+ deadTA + RA hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u0915\\u093e")); hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u092a")+ deadTA + CharsToUnicodeString("\\u0930")); //PA + deadTA + RA hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u092a\\u095d\\u094b")); // hindiWordData->addElement(CharsToUnicodeString("\\u0964")); //fails here doesn't break at danda hindiWordData->addElement(" "); hindiWordData->addElement(deadSA + deadTA + CharsToUnicodeString("\\u0930\\u093f")); //deadSA+deadTA+RA+vowel I->sthri hindiWordData->addElement("."); hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u0968\\u0966.\\u0969\\u096f")); //hindi numbers hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u0967\\u0966\\u0966.\\u0966\\u0966%")); //postnumeric hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u20a8\\u0967,\\u0967\\u0966\\u0966.\\u0966\\u0966")); //pre-number India currency symbol Rs.\\u20aD hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u0905\\u092e\\u091c")); hindiWordData->addElement("\n"); hindiWordData->addElement(halfSA+CharsToUnicodeString("\\u0935\\u0924\\u0902")+deadTA+CharsToUnicodeString("\\u0930")); hindiWordData->addElement("\r"); UErrorCode status=U_ZERO_ERROR; RuleBasedBreakIterator *e=(RuleBasedBreakIterator*)RuleBasedBreakIterator::createWordInstance(Locale::getDefault(), status); if(U_FAILURE(status)){ errln("FAIL : in construction"); return; } generalIteratorTest(*e, hindiWordData); delete e; delete hindiWordData; } /* //Bug: if there is no word break before and after danda when it is followed by a space void RBBITest::TestDanda() { Vector *hindiWordData = new Vector(); //hindi hindiWordData->addElement(CharsToUnicodeString("\\u092f\\u0939")); hindiWordData->addElement(" "); //Danda is similar to full stop, danda is a hindi phrase seperator. //Make sure there is a word break before and after danda when it is followed by a space //following fail---- hindiWordData->addElement(CharsToUnicodeString("\\u0939\\u0948")); // hindiWordData->addElement(CharsToUnicodeString("\\u0964")); // devanagari danda hindiWordData->addElement(" "); hindiWordData->addElement(CharsToUnicodeString("\\u092f\\u0939")); // hindiWordData->addElement(CharsToUnicodeString("\\u0965")); //devanagari double danda hindiWordData->addElement(" "); RuleBasedBreakIterator* e=(RuleBasedBreakIterator*)RuleBasedBreakIterator::createWordInstance(); generalIteratorTest(*e, hindiWordData); delete e; delete hindiWordData; } //Make sure the character wrapping is done correctly void RBBITest::TestHindiCharacterWrapping() { Vector *hindicharData = new Vector(); //if the dead consonant RA precedes either a consonant or an independent vowel, //then it is replaced by its superscript non-spacing mark hindicharData->addElement(deadRA+ CharsToUnicodeString("\\u0917")); //deadRA+devanagari consonant GA->GA+superRA //following fail---- // hindicharData->addElement(deadRA+ CharsToUnicodeString("\\u0960")); //deadRA+devanagari RRI->RRI+superRA RuleBasedBreakIterator* e=(RuleBasedBreakIterator*)RuleBasedBreakIterator::createCharacterInstance(); generalIteratorTest(*e, hindicharData); delete e; delete hindicharData; }*/ //---------------------------------------------------------------------------------- //adds rules for telugu support and tests the behaviour of chracterIterator of RBBI //---------------------------------------------------------------------------------- /*void RBBITest::TestTeluguRuleBasedCharacterIteration() { logln((UnicodeString)"Testing the RBBI by adding rules for Telugu(Indian Language) Support"); //get the default rules RuleBasedBreakIterator *rb= (RuleBasedBreakIterator*)BreakIterator::createCharacterInstance(); //additional rules for Telugu(Indian Language) support UnicodeString crules1 = rb->getRules() + //default rules + "=[\\u0c4d];" + //telugu virama "=[\\u0c3e-\\u0c44\\u0c46\\u0c47\\u0c48\\u0c4a\\u0c4b\\u0c4c];" + //telugu dependent vowel signs "=[\\u0c15-\\u0c28\\u0c2a-\\u0c33\\u0c35-\\u0c39];" + //telugu consonants "=[\\u0c02\\u0c03\\u0c55\\u0c56];" + //to create half forms and dead forms "=({{}});" + "{}{};"; RuleBasedBreakIterator charIter=null; charIter = new RuleBasedBreakIterator(crules1); Vector *chardata = new Vector(); //behaviour of telugu characters from specified rules chardata->addElement(CharsToUnicodeString("\\u0c15")); //telugu consonant KA chardata->addElement(CharsToUnicodeString("\\u0c30\\u0c47")); //telugu consonant RA+telugu dependent vowel EE chardata->addElement(CharsToUnicodeString("\\u0c1b\\u0c3e")); //telugu consonant CHA+telegu depenednt vowel AA chardata->addElement(CharsToUnicodeString("\\u0c17\\u0c48")); //telegu consonant GA+teleugu dependent vowel AI chardata->addElement(CharsToUnicodeString("\\u0c17\\u0c46\\u0c56")); //telugu consonant GA+telugu dependent vowel sign E+telugu AI length mark chardata->addElement(CharsToUnicodeString("\\u0c28\\u0c4d\\u200d\\u0c28")); //telugu consonant NA+telugu virama+zwj=>halfNA+NA->NNA(dependent half-form) chardata->addElement(CharsToUnicodeString("\\u0c17\\u0c4d\\u0c30")); //GA+deadRA(RA+telvirama)->GA+subRA->GRA chardata->addElement(CharsToUnicodeString("\\u0c66")); //telugu digit chardata->addElement(CharsToUnicodeString("\\u0c37\\u0c4d\\u0c15")); //deadSSA(SSA+telvirama)+KA+subSSA->KSHA //behaviour of other characters from default rules chardata->addElement("h"); chardata->addElement(CharsToUnicodeString("A\\u0302")); // circumflexA chardata->addElement(CharsToUnicodeString("i\\u0301")); // acuteBelowI chardata->addElement(CharsToUnicodeString("\\u1109\\u1161\\u11bc")); chardata->addElement(CharsToUnicodeString("\\u1112\\u1161\\u11bc")); chardata->addElement("\n"); chardata->addElement("\r\n"); generalIteratorTest(charIter, chardata); delete charIter; delete charData; delete rb; } //-------------------------------------------------------------------- //tests the behaviour of character iteration of RBBI with custom rules //-------------------------------------------------------------------- void RBBITest::TestCustomRuleBasedCharacterIteration() { logln((UnicodeString)"Testing the RBBI by using custom rules for character iteration"); UnicodeString crules2="=[e];" + //ignore the character "e" ".;" + "=[\\u093e-\\u094c\\u0962\\u0963];" + //devanagiri vowel = \\u093e tp \\u094c and \\u0962.\\u0963 "=[\\u0915-\\u0939];" + //devanagiri consonant = \\u0915 to \\u0939 "{};" ; //break at all places except the following //devanagiri consonants+ devanagiri vowelsign RuleBasedCharacterIterator charIterCustom = new RuleBasedBreakIterator(crules2); Vector *chardata = new Vector(); chardata->addElement("He"); //ignores 'e' chardata->addElement("l"); chardata->addElement("l"); chardata->addElement("oe"); //ignores 'e' hence wraps it into 'o' instead of wrapping with chardata->addElement(CharsToUnicodeString("\\u0301")); //'\\u0301' to form 'acuteE ' chardata->addElement("&e"); //ignores 'e' hence wraps it into '&' instead of wrapping with chardata->addElement(CharsToUnicodeString("\\u0303")); //'\\u0303 to form 'tildaE' //devanagiri characters chardata->addElement(CharsToUnicodeString("\\u0906")); //devanagiri AA chardata->addElement(CharsToUnicodeString("\\u093e")); //devanagiri vowelsign AA:--breaks at \\u0901 which is devanagiri chardata->addElement(CharsToUnicodeString("\\u0901")); //chandra bindhu since it is not mentioned in the rules chardata->addElement(CharsToUnicodeString("\\u0916\\u0947")); //devanagiri KHA+vowelsign E chardata->addElement(CharsToUnicodeString("\\u0938\\u0941")); //devanagiri SA+vowelsign U : - breaks at chardata->addElement(CharsToUnicodeString("\\u0902")); //\\u0902 devanagiri anusvara since it is not mentioned in the rules chardata->addElement(CharsToUnicodeString("\\u0926")); //devanagiri consonant DA chardata->addElement(CharsToUnicodeString("\\u0930")); //devanagiri consonant RA chardata->addElement(CharsToUnicodeString("\\u0939\\u094c")); //devanagiri HA+vowel sign AI chardata->addElement(CharsToUnicodeString("\\u0964")); //devanagiri danda // devanagiri chracters end chardata->addElement("A"); //breaks in between since it is not mentioned in the rules chardata->addElement(CharsToUnicodeString("\\u0302")); // circumflexA chardata->addElement("i"); //breaks in between since not mentioned in the rules chardata->addElement(CharsToUnicodeString("\\u0301")); // acuteBelowI //Rules don't support conjoining jamo->->.. chardata->addElement(CharsToUnicodeString("\\u1109")); //break at every character since rules chardata->addElement(CharsToUnicodeString("\\u1161")); //don't support conjoining jamo chardata->addElement(CharsToUnicodeString("\\u11bc")); chardata->addElement(CharsToUnicodeString("\\u1112")); chardata->addElement(CharsToUnicodeString("\\u1161")); chardata->addElement(CharsToUnicodeString("\\u11bc")); chardata->addElement("\n"); chardata->addElement("\r"); //doesn't keep CRLGF together since rules do not mention it chardata->addElement("\n"); chardata->addElement("S"); //graveS chardata->addElement(CharsToUnicodeString("\\u0300")); //breaks in between since it is not mentioned in the rules chardata->addElement("i"); //acuteBelowI chardata->addElement(CharsToUnicodeString("\\u0301")); //breaks in between since it is not mentioned in the rules generalIteratorTest(charIterCustom, chardata); delete charIterCustom; delete chardata; }*/ /*//-------------------------------------------------------------------- //tests custom rules based word iteration //-------------------------------------------------------------------- void RBBITest::TestCustomRuleBasedWordIteration(){ logln("(UnicodeString)Testing the RBBI by using custom rules for word iteration"); UnicodeString wrules1="=[:Mn::Me::Cf:];" + //ignore non-spacing marks, enclosing marks, and format characters, "=[\\u0964\\u0965];" + //Hindi Phrase seperator "=[:L::Mc:];" + //uppercase(Lu), lowercase(Ll), titlecase(Lt), modifier(Lm) letters, Mc-combining space mark "=[:Pd:\\\"\\\'\\.];" + //dashes, quotation, apostraphes, period "=[\\n\\u000c\\u2028\\u2029];" + //line separators: LF, FF, PS, and LS "=[:Zs:\\t];" + //all space separators and the tab character "=((*(*)*));" + ".;" + //break after every character, with the following exceptions "{};" + "*{\\r}{}{};" ; RuleBasedBreakIterator wordIterCustom = new RuleBasedBreakIterator(wrules1); Vector *worddata = new Vector(); worddata->addElement("Write"); worddata->addElement(" "); worddata->addElement("wordrules"); worddata->addElement("."); worddata->addElement(" "); //play with hindi worddata->addElement(CharsToUnicodeString("\\u092f\\u0939")); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u0939\\u093f") + halfNA + CharsToUnicodeString("\\u0926\\u0940")); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u0939\\u0948")); worddata->addElement(CharsToUnicodeString("\\u0964")); //Danda is similar to full stop-> Danda followed by a space worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u0905\\u093e\\u092a")); worddata->addElement(" "); worddata->addElement(CharsToUnicodeString("\\u0938\\u093f\\u0916\\u094b\\u0917\\u0947")); worddata->addElement("?"); worddata->addElement(" "); worddata->addElement("It's"); worddata->addElement(" "); worddata->addElement("$"); worddata->addElement("3"); worddata->addElement("0"); worddata->addElement("."); worddata->addElement("1"); worddata->addElement("0"); worddata->addElement(" "); // worddata->addElement(" "); generalIteratorTest(wordIterCustom, worddata); delete wordIterCustom; delete worddata; } //------------------------------------------------------------------------------- //adds extra rules to deal with abbrevations(limited) and test the word Iteration //------------------------------------------------------------------------------- void RBBITest::TestAbbrRuleBasedWordIteration() { logln((UnicodeString)"Testing the RBBI for word iteration by adding rules to support abbreviation"); RuleBasedBreakIterator *rb =(RuleBasedBreakIterator*)BreakIterator::createWordInstance(); UnicodeString wrules2="=((Mr.)|(Mrs.)|(Ms.)|(Dr.)|(U.S.));" + // abbreviations. rb->getRules() + "{()*};"; RuleBasedBreakIterator wordIter=null; //try{ wordIter = new RuleBasedBreakIterator(wrules2); // }catch(IllegalArgumentException iae){ // errln("ERROR: failed construction illegal rules"); // } Vector *worddata = new Vector(); worddata->addElement("Mr. George"); worddata->addElement(" "); worddata->addElement("is"); worddata->addElement(" "); worddata->addElement("from"); worddata->addElement(" "); worddata->addElement("U.S. Navy"); worddata->addElement("."); worddata->addElement(" "); worddata->addElement("His"); worddata->addElement("\n"); worddata->addElement("friend"); worddata->addElement("\t"); worddata->addElement("Dr. Steven"); worddata->addElement(" "); worddata->addElement("married"); worddata->addElement(" "); worddata->addElement("Ms. Benneth"); worddata->addElement("!"); worddata->addElement(" "); worddata->addElement("Mrs. Johnson"); worddata->addElement("\r\n"); worddata->addElement("paid"); worddata->addElement(" "); worddata->addElement("$2,400.00"); generalIteratorTest(wordIter, worddata); delete wordIter; delete worddata; delete rb; } */ //--------------------------------------------- // runIndexedTest //--------------------------------------------- void RBBITest::runIndexedTest( int32_t index, bool_t exec, char* &name, char* par ) { if (exec) logln("TestSuite RuleBasedBreakIterator: "); switch (index) { case 0: name = "TestDefaultRuleBasedCharacterIteration"; if(exec) TestDefaultRuleBasedCharacterIteration(); break; case 1: name = "TestDefaultRuleBasedWordIteration"; if(exec) TestDefaultRuleBasedWordIteration(); break; case 2: name = "TestDefaultRuleBasedSentenceIteration"; if(exec) TestDefaultRuleBasedSentenceIteration(); break; case 3: name = "TestDefaulRuleBasedLineIteration()"; if(exec) TestDefaultRuleBasedLineIteration(); break; case 4: name = "TestHindiCharacterBreak()"; if(exec) TestHindiCharacterBreak(); break; case 5: name = "TestHindiWordBreak()"; if(exec) TestHindiWordBreak(); break; case 6: name = "TestHindiWordBreak()"; if(exec) TestHindiWordBreak(); break; // case 7: name = "TestDanda()"; // if(exec) TestDanda(); break; // case 8: name = "TestHindiCharacterWrapping()"; // if(exec) TestHindiCharacterWrapping(); break; // case 9: name = "TestCustomRuleBasedWordIteration"; // if(exec) TestCustomRuleBasedWordIteration(); break; // case 10: name = "TestAbbrRuleBasedWordIteration"; // if(exec) TestAbbrRuleBasedWordIteration(); break; // case 11: name = "TestTeluguRuleBasedCharacterIteration"; // if(exec) TestTeluguRuleBasedCharacterIteration(); break; // case 12: name = "TestCustomRuleBasedCharacterIteration"; // if(exec) TestCustomRuleBasedCharacterIteration(); break; default: name = ""; break; //needed to end loop } } //--------------------------------------------- // Test implementation routines //--------------------------------------------- UnicodeString RBBITest::createTestData(Enumeration* e) { UnicodeString result = ""; while (e->hasMoreElements()) { result += e->nextElement(); } return result; } // general test Implementation subroutines void RBBITest::generalIteratorTest(RuleBasedBreakIterator& bi, Vector* expectedResult) { Enumeration *elems = expectedResult->elements(); UnicodeString text = createTestData(elems); delete elems; bi.setText(&text); Vector *nextResults = testFirstAndNext(bi, text); Vector *previousResults = testLastAndPrevious(bi, text); logln("comparing forward and backward..."); int errs = getErrors(); UnicodeString str1="forward iteration"; UnicodeString str2="backward iteration"; compareFragmentLists(str1, str2, nextResults, previousResults); if (getErrors() == errs) { logln("comparing expected and actual..."); str1="expected result"; str2="actual result"; compareFragmentLists(str1, str2, expectedResult, nextResults); } int32_t *boundaries = new int32_t[expectedResult->size() + 3]; boundaries[0] = RuleBasedBreakIterator::DONE; boundaries[1] = 0; for (int i = 0; i < expectedResult->size(); i++) boundaries[i + 2] = boundaries[i + 1] + ((UnicodeString)expectedResult->elementAt(i)). length(); int len = expectedResult->size() + 3 -1; boundaries[len] = RuleBasedBreakIterator::DONE; testFollowing(bi, text, boundaries); testPreceding(bi, text, boundaries); testIsBoundary(bi, text, boundaries); doMultipleSelectionTest(bi, text); } Vector* RBBITest::testFirstAndNext(RuleBasedBreakIterator& bi, UnicodeString& text) { int32_t p = bi.first(); int32_t lastP = p; Vector *result = new Vector(); UnicodeString selection; if (p != 0) errln((UnicodeString)"first() returned " + p + (UnicodeString)" instead of 0"); while (p != RuleBasedBreakIterator::DONE) { p = bi.next(); if (p != RuleBasedBreakIterator::DONE) { if (p <= lastP) errln((UnicodeString)"next() failed to move forward: next() on position " + lastP + (UnicodeString)" yielded " + p); text.extractBetween(lastP, p, selection); result->addElement(selection); } else { if (lastP != text.length()) errln((UnicodeString)"next() returned DONE prematurely: offset was " + lastP + (UnicodeString)" instead of " + text.length()); } lastP = p; } return result; } Vector* RBBITest::testLastAndPrevious(RuleBasedBreakIterator& bi, UnicodeString& text) { int32_t p = bi.last(); int32_t lastP = p; Vector *result = new Vector(); UnicodeString selection; if (p != text.length()) errln((UnicodeString)"last() returned " + p + (UnicodeString)" instead of " + text.length()); while (p != RuleBasedBreakIterator::DONE) { p = bi.previous(); if (p != RuleBasedBreakIterator::DONE) { if (p >= lastP) errln((UnicodeString)"previous() failed to move backward: previous() on position " + lastP + (UnicodeString)" yielded " + p); text.extractBetween(p, lastP, selection); result->insertElementAt(selection, 0); } else { if (lastP != 0) errln((UnicodeString)"previous() returned DONE prematurely: offset was " + lastP + (UnicodeString)" instead of 0"); } lastP = p; } return result; } void RBBITest::compareFragmentLists(UnicodeString& f1Name, UnicodeString& f2Name, Vector* f1, Vector* f2) { int32_t p1 = 0; int32_t p2 = 0; UnicodeString s1; UnicodeString s2; int32_t t1 = 0; int32_t t2 = 0; while (p1 < f1->size() && p2 < f2->size()) { s1 = (UnicodeString)f1->elementAt(p1); s2 = (UnicodeString)f2->elementAt(p2); t1 += s1.length(); t2 += s2.length(); if (s1.compare(s2) == 0) { logln(prettify((UnicodeString)" >" + s1 + (UnicodeString)"<")); ++p1; ++p2; } else { int32_t tempT1 = t1; int32_t tempT2 = t2; int32_t tempP1 = p1; int32_t tempP2 = p2; while (tempT1 != tempT2 && tempP1 < f1->size() && tempP2 < f2->size()) { while (tempT1 < tempT2 && tempP1 < f1->size()) { tempT1 += ((UnicodeString)f1->elementAt(tempP1)).length(); ++tempP1; } while (tempT2 < tempT1 && tempP2 < f2->size()) { tempT2 += ((UnicodeString)f2->elementAt(tempP2)).length(); ++tempP2; } } logln((UnicodeString)"*** " + f1Name + (UnicodeString)" has:"); while (p1 <= tempP1 && p1 < f1->size()) { s1 = (UnicodeString)f1->elementAt(p1); t1 += s1.length(); logln(prettify((UnicodeString)" *** >" + s1 + (UnicodeString)"<")); ++p1; } logln("***** " + f2Name + " has:"); while (p2 <= tempP2 && p2 < f2->size()) { s2 = (UnicodeString)f2->elementAt(p2); t2 += s2.length(); logln(prettify(" ***** >" + s2 + "<")); ++p2; } errln((UnicodeString)"Discrepancy between " + f1Name + (UnicodeString)" and " + f2Name); } } } void RBBITest::testFollowing(RuleBasedBreakIterator& bi, UnicodeString& text, int32_t *boundaries) { logln("testFollowing():"); int p = 2; for (int i = 0; i <= text.length(); i++) { if (i == boundaries[p]) ++p; int32_t b = bi.following(i); logln((UnicodeString)"bi.following(" + i + ") -> " + b); if (b != boundaries[p]) errln((UnicodeString)"Wrong result from following() for " + i + (UnicodeString)": expected " + boundaries[p] + (UnicodeString)", got " + b); } } void RBBITest::testPreceding(RuleBasedBreakIterator& bi, UnicodeString& text, int32_t *boundaries) { logln("testPreceding():"); int p = 0; for (int i = 0; i <= text.length(); i++) { int32_t b = bi.preceding(i); logln((UnicodeString)"bi.preceding(" + i + ") -> " + b); if (b != boundaries[p]) errln((UnicodeString)"Wrong result from preceding() for " + i + (UnicodeString)": expected " + boundaries[p] + (UnicodeString)", got " + b); if (i == boundaries[p + 1]) ++p; } } void RBBITest::testIsBoundary(RuleBasedBreakIterator& bi, UnicodeString& text, int32_t *boundaries) { logln("testIsBoundary():"); int p = 1; bool_t isB; for (int i = 0; i < text.length(); i++) { isB = bi.isBoundary(i); logln((UnicodeString)"bi.isBoundary(" + i + ") -> " + isB); if (i == boundaries[p]) { if (!isB) errln((UnicodeString)"Wrong result from isBoundary() for " + i + (UnicodeString)": expected true, got false"); p++; } else { if (isB) errln((UnicodeString)"Wrong result from isBoundary() for " + i + (UnicodeString)": expected false, got true"); } } } void RBBITest::doMultipleSelectionTest(RuleBasedBreakIterator& iterator, UnicodeString& testText) { iterator.setText(&testText); RuleBasedBreakIterator* testIterator =(RuleBasedBreakIterator*)iterator.clone(); int32_t offset = iterator.first(); int32_t testOffset; int32_t count = 0; logln("doMultipleSelectionTest text of length: "+testText.length()); if (*testIterator != iterator) errln("clone() or operator!= failed: two clones compared unequal"); do { testOffset = testIterator->first(); testOffset = testIterator->next(count); if (offset != testOffset) errln(UnicodeString("next(n) and next() not returning consistent results: for step ") + count + ", next(n) returned " + testOffset + " and next() had " + offset); if (offset != RuleBasedBreakIterator::DONE) { count++; offset = iterator.next(); if (offset != RuleBasedBreakIterator::DONE && *testIterator == iterator) errln("operator== failed: Two unequal iterators compared equal."); } } while (offset != RuleBasedBreakIterator::DONE); // now do it backwards... offset = iterator.last(); count = 0; do { testOffset = testIterator->last(); testOffset = testIterator->next(count); if (offset != testOffset) errln(UnicodeString("next(n) and next() not returning consistent results: for step ") + count + ", next(n) returned " + testOffset + " and next() had " + offset); if (offset != RuleBasedBreakIterator::DONE) { count--; offset = iterator.previous(); } } while (offset != RuleBasedBreakIterator::DONE); //delete testIterator; }