2000-03-10 00:42:27 +00:00
/********************************************************************
* COPYRIGHT :
* Copyright ( c ) 1999 - 2000 , International Business Machines Corporation and
* others . All Rights Reserved .
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
/************************************************************************
2000-01-17 20:59:08 +00:00
* Date Name Description
* 12 / 14 / 99 Madhu Creation .
* 01 / 12 / 2000 Madhu updated for changed API
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
# include "unicode/utypes.h"
# include "unicode/unicode.h"
# include "intltest.h"
# include "unicode/rbbi.h"
# include "unicode/schriter.h"
# include "rbbiapts.h"
# include "string.h"
# include "stdio.h"
/**
* API Test the RuleBasedBreakIterator class
*/
void RBBIAPITest : : TestCloneEquals ( )
{
UErrorCode status = U_ZERO_ERROR ;
RuleBasedBreakIterator * bi1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createCharacterInstance ( Locale : : getDefault ( ) , status ) ;
RuleBasedBreakIterator * biequal = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createCharacterInstance ( Locale : : getDefault ( ) , status ) ;
RuleBasedBreakIterator * bi3 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createCharacterInstance ( Locale : : getDefault ( ) , status ) ;
RuleBasedBreakIterator * bi2 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createWordInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) ) {
errln ( ( UnicodeString ) " FAIL : in construction " ) ;
return ;
}
UnicodeString testString = " Testing word break iterators's clone() and equals() " ;
bi1 - > setText ( testString ) ;
bi2 - > setText ( testString ) ;
biequal - > setText ( testString ) ;
bi3 - > setText ( " hello " ) ;
logln ( ( UnicodeString ) " Testing equals() " ) ;
logln ( ( UnicodeString ) " Testing == and != " ) ;
if ( * bi1 ! = * biequal | | * bi1 = = * bi2 | | * bi1 = = * bi3 )
errln ( ( UnicodeString ) " ERROR:1 RBBI's == and !- operator failed. " ) ;
if ( * bi2 = = * biequal | | * bi2 = = * bi1 | | * biequal = = * bi3 )
errln ( ( UnicodeString ) " ERROR:2 RBBI's == and != operator failed. " ) ;
logln ( ( UnicodeString ) " Testing clone() " ) ;
RuleBasedBreakIterator * bi1clone = ( RuleBasedBreakIterator * ) bi1 - > clone ( ) ;
RuleBasedBreakIterator * bi2clone = ( RuleBasedBreakIterator * ) bi2 - > clone ( ) ;
if ( * bi1clone ! = * bi1 | | * bi1clone ! = * biequal | |
* bi1clone = = * bi3 | | * bi1clone = = * bi2 )
errln ( ( UnicodeString ) " ERROR:1 RBBI's clone() method failed " ) ;
if ( * bi2clone = = * bi1 | | * bi2clone = = * biequal | |
* bi2clone = = * bi3 | | * bi2clone ! = * bi2 )
errln ( ( UnicodeString ) " ERROR:2 RBBI's clone() method failed " ) ;
if ( bi1 - > getText ( ) ! = bi1clone - > getText ( ) | |
bi2clone - > getText ( ) ! = bi2 - > getText ( ) | |
* bi2clone = = * bi1clone )
errln ( ( UnicodeString ) " ERROR: RBBI's clone() method failed " ) ;
delete bi1 ;
delete bi3 ;
delete bi2 ;
delete biequal ;
}
void RBBIAPITest : : TestgetRules ( )
{
UErrorCode status = U_ZERO_ERROR ;
RuleBasedBreakIterator * bi1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createCharacterInstance ( Locale : : getDefault ( ) , status ) ;
RuleBasedBreakIterator * bi2 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createWordInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) ) {
errln ( ( UnicodeString ) " FAIL: in construction " ) ;
return ;
}
logln ( ( UnicodeString ) " Testing toString() " ) ;
bi1 - > setText ( ( UnicodeString ) " Hello there " ) ;
RuleBasedBreakIterator * bi3 = ( RuleBasedBreakIterator * ) bi1 - > clone ( ) ;
UnicodeString temp = bi1 - > getRules ( ) ;
UnicodeString temp2 = bi2 - > getRules ( ) ;
UnicodeString temp3 = bi3 - > getRules ( ) ;
if ( temp2 . compare ( temp3 ) = = 0 | | temp . compare ( temp2 ) = = 0 | | temp . compare ( temp3 ) ! = 0 )
errln ( ( UnicodeString ) " ERROR: error in getRules() method " ) ;
delete bi1 ;
delete bi2 ;
}
void RBBIAPITest : : TestHashCode ( )
{
UErrorCode status = U_ZERO_ERROR ;
RuleBasedBreakIterator * bi1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createCharacterInstance ( Locale : : getDefault ( ) , status ) ;
RuleBasedBreakIterator * bi3 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createCharacterInstance ( Locale : : getDefault ( ) , status ) ;
RuleBasedBreakIterator * bi2 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createWordInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) ) {
errln ( ( UnicodeString ) " FAIL : in construction " ) ;
return ;
}
logln ( ( UnicodeString ) " Testing hashCode() " ) ;
bi1 - > setText ( ( UnicodeString ) " Hash code " ) ;
bi2 - > setText ( ( UnicodeString ) " Hash code " ) ;
bi3 - > setText ( ( UnicodeString ) " Hash code " ) ;
RuleBasedBreakIterator * bi1clone = ( RuleBasedBreakIterator * ) bi1 - > clone ( ) ;
RuleBasedBreakIterator * bi2clone = ( RuleBasedBreakIterator * ) bi2 - > clone ( ) ;
if ( bi1 - > hashCode ( ) ! = bi1clone - > hashCode ( ) | | bi1 - > hashCode ( ) ! = bi3 - > hashCode ( ) | |
bi1clone - > hashCode ( ) ! = bi3 - > hashCode ( ) | | bi2 - > hashCode ( ) ! = bi2clone - > hashCode ( ) )
errln ( ( UnicodeString ) " ERROR: identical objects have different hasecodes " ) ;
if ( bi1 - > hashCode ( ) = = bi2 - > hashCode ( ) | | bi2 - > hashCode ( ) = = bi3 - > hashCode ( ) | |
bi1clone - > hashCode ( ) = = bi2clone - > hashCode ( ) | | bi1clone - > hashCode ( ) = = bi2 - > hashCode ( ) )
errln ( ( UnicodeString ) " ERROR: different objects have same hasecodes " ) ;
delete bi1 ;
delete bi2 ;
delete bi3 ;
}
void RBBIAPITest : : TestGetSetAdoptText ( )
{
logln ( ( UnicodeString ) " Testing getText setText " ) ;
UErrorCode status = U_ZERO_ERROR ;
UnicodeString str1 = " first string. " ;
UnicodeString str2 = " Second string. " ;
RuleBasedBreakIterator * charIter1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createCharacterInstance ( Locale : : getDefault ( ) , status ) ;
RuleBasedBreakIterator * wordIter1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createWordInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) ) {
errln ( ( UnicodeString ) " FAIL : in construction " ) ;
return ;
}
CharacterIterator * text1 = new StringCharacterIterator ( str1 ) ;
CharacterIterator * text2 = new StringCharacterIterator ( str2 ) ;
wordIter1 - > setText ( str1 ) ;
if ( wordIter1 - > getText ( ) ! = * text1 )
errln ( ( UnicodeString ) " ERROR:1 error in setText or getText " ) ;
if ( wordIter1 - > current ( ) ! = 0 )
errln ( ( UnicodeString ) " ERROR:1 setText did not set the iteration position to the beginning of the text, it is " + wordIter1 - > current ( ) + ( UnicodeString ) " \n " ) ;
wordIter1 - > next ( 2 ) ;
wordIter1 - > setText ( str2 ) ;
if ( wordIter1 - > current ( ) ! = 0 )
errln ( ( UnicodeString ) " ERROR:2 setText did not reset the iteration position to the beginning of the text, it is " + wordIter1 - > current ( ) + ( UnicodeString ) " \n " ) ;
charIter1 - > adoptText ( text1 ) ;
if ( wordIter1 - > getText ( ) = = charIter1 - > getText ( ) | |
wordIter1 - > getText ( ) ! = * text2 | | charIter1 - > getText ( ) ! = * text1 )
errln ( ( UnicodeString ) " ERROR:2 error is getText or setText() " ) ;
RuleBasedBreakIterator * rb = ( RuleBasedBreakIterator * ) wordIter1 - > clone ( ) ;
rb - > adoptText ( text1 ) ;
if ( rb - > getText ( ) ! = * text1 )
errln ( ( UnicodeString ) " ERROR:1 error in adoptText " ) ;
rb - > adoptText ( text2 ) ;
if ( rb - > getText ( ) ! = * text2 )
errln ( ( UnicodeString ) " ERROR:2 error in adoptText " ) ;
delete wordIter1 ;
}
void RBBIAPITest : : TestFirstNextFollowing ( )
{
int32_t p , q ;
UErrorCode status = U_ZERO_ERROR ;
UnicodeString testString = " This is a word break. Isn't it? 2.25 " ;
logln ( ( UnicodeString ) " Testing first() and next(), following() with custom rules " ) ;
logln ( ( UnicodeString ) " testing word iterator - string :- \" " + prettify ( testString ) + ( UnicodeString ) " \" \n " ) ;
RuleBasedBreakIterator * wordIter1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createWordInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) )
errln ( " FAIL : in construction " ) ;
else {
wordIter1 - > setText ( testString ) ;
p = wordIter1 - > first ( ) ;
if ( p ! = 0 )
errln ( ( UnicodeString ) " ERROR: first() returned " + p + ( UnicodeString ) " instead of 0 " ) ;
q = wordIter1 - > next ( 9 ) ;
doTest ( testString , p , q , 20 , " This is a word break " ) ;
p = q ;
q = wordIter1 - > next ( ) ;
doTest ( testString , p , q , 21 , " . " ) ;
p = q ;
q = wordIter1 - > next ( 3 ) ;
doTest ( testString , p , q , 28 , " Isn't " ) ;
p = q ;
q = wordIter1 - > next ( 2 ) ;
doTest ( testString , p , q , 31 , " it? " ) ;
//logln((UnicodeString)"Testing following(int)");
q = wordIter1 - > following ( 2 ) ;
doTest ( testString , 2 , q , 4 , " is " ) ;
q = wordIter1 - > following ( 22 ) ;
doTest ( testString , 22 , q , 27 , " Isn't " ) ;
wordIter1 - > last ( ) ;
p = wordIter1 - > next ( ) ;
q = wordIter1 - > following ( wordIter1 - > last ( ) ) ;
if ( p ! = RuleBasedBreakIterator : : DONE | | q ! = RuleBasedBreakIterator : : DONE )
errln ( ( UnicodeString ) " ERROR: next()/following() at last position returned # " +
p + ( UnicodeString ) " and " + q + ( UnicodeString ) " instead of " + testString . length ( ) + ( UnicodeString ) " \n " ) ;
}
status = U_ZERO_ERROR ;
RuleBasedBreakIterator * charIter1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createCharacterInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) )
errln ( " FAIL : in construction " ) ;
else {
testString = CharsToUnicodeString ( " Write hindi here. \\ u092d \\ u093e \\ u0930 \\ u0924 \\ u0938 \\ u0941 \\ u0902 \\ u0926 \\ u0930 \\ u0939 \\ u094c \\ u0964 " ) ;
logln ( ( UnicodeString ) " testing char iter - string:- \" " + prettify ( testString ) + ( UnicodeString ) " \" " ) ;
charIter1 - > setText ( testString ) ;
p = charIter1 - > first ( ) ;
if ( p ! = 0 )
errln ( ( UnicodeString ) " ERROR: first() returned " + p + ( UnicodeString ) " instead of 0 " ) ;
q = charIter1 - > next ( ) ;
doTest ( testString , p , q , 1 , " W " ) ;
p = q ;
q = charIter1 - > next ( 4 ) ;
doTest ( testString , p , q , 5 , " rite " ) ;
p = q ;
q = charIter1 - > next ( 12 ) ;
doTest ( testString , p , q , 17 , " hindi here. " ) ;
p = q ;
q = charIter1 - > next ( - 6 ) ;
doTest ( testString , p , q , 11 , " here. " ) ;
p = q ;
q = charIter1 - > next ( 6 ) ;
doTest ( testString , p , q , 17 , " here. " ) ;
// hindi starts here
p = q ;
q = charIter1 - > next ( 4 ) ;
doTest ( testString , p , q , 22 , " \\ u092d \\ u093e \\ u0930 \\ u0924 " ) ;
p = q ;
q = charIter1 - > next ( 2 ) ;
doTest ( testString , p , q , 26 , " \\ u0938 \\ u0941 \\ u0902 " ) ;
q = charIter1 - > following ( 24 ) ;
doTest ( testString , 24 , q , 26 , " \\ u0941 \\ u0902 " ) ;
q = charIter1 - > following ( 20 ) ;
doTest ( testString , 20 , q , 21 , " \\ u0930 " ) ;
p = charIter1 - > following ( charIter1 - > last ( ) ) ;
q = charIter1 - > next ( charIter1 - > last ( ) ) ;
if ( p ! = RuleBasedBreakIterator : : DONE | | q ! = RuleBasedBreakIterator : : DONE )
errln ( ( UnicodeString ) " ERROR: following()/next() at last position returned # " +
p + ( UnicodeString ) " and " + q + ( UnicodeString ) " instead of " + testString . length ( ) ) ;
}
status = U_ZERO_ERROR ;
testString = " Hello! how are you? I'am fine. Thankyou. How are you doing? This \n costs $20,00,000. " ;
RuleBasedBreakIterator * sentIter1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createSentenceInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) )
errln ( " FAIL : in construction " ) ;
else {
logln ( ( UnicodeString ) " testing sentence iter - String:- \" " + prettify ( testString ) + ( UnicodeString ) " \" " ) ;
sentIter1 - > setText ( testString ) ;
p = sentIter1 - > first ( ) ;
if ( p ! = 0 )
errln ( ( UnicodeString ) " ERROR: first() returned " + p + ( UnicodeString ) " instead of 0 " ) ;
q = sentIter1 - > next ( ) ;
doTest ( testString , p , q , 7 , " Hello! " ) ;
p = q ;
q = sentIter1 - > next ( 2 ) ;
doTest ( testString , p , q , 31 , " how are you? I'am fine. " ) ;
p = q ;
q = sentIter1 - > next ( - 2 ) ;
doTest ( testString , p , q , 7 , " how are you? I'am fine. " ) ;
p = q ;
q = sentIter1 - > next ( 4 ) ;
doTest ( testString , p , q , 60 , " how are you? I'am fine. Thankyou. How are you doing? " ) ;
p = q ;
q = sentIter1 - > next ( ) ;
doTest ( testString , p , q , 83 , " This \n costs $20,00,000. " ) ;
q = sentIter1 - > following ( 1 ) ;
doTest ( testString , 1 , q , 7 , " ello! " ) ;
q = sentIter1 - > following ( 10 ) ;
doTest ( testString , 10 , q , 20 , " are you? " ) ;
q = sentIter1 - > following ( 20 ) ;
doTest ( testString , 20 , q , 31 , " I'am fine. " ) ;
p = sentIter1 - > following ( sentIter1 - > last ( ) ) ;
q = sentIter1 - > next ( sentIter1 - > last ( ) ) ;
if ( p ! = RuleBasedBreakIterator : : DONE | | q ! = RuleBasedBreakIterator : : DONE )
errln ( ( UnicodeString ) " ERROR: following()/next() at last position returned # " +
p + ( UnicodeString ) " and " + q + ( UnicodeString ) " instead of " + testString . length ( ) ) ;
}
status = U_ZERO_ERROR ;
testString = CharsToUnicodeString ( " Hello! how \r \n (are) \r you? I'am fine- Thankyou. foo \\ u00a0bar How, are, you? This, costs $20,00,000. " ) ;
logln ( " (UnicodeString)testing line iter - String:- \" " + prettify ( testString ) + ( UnicodeString ) " \" " ) ;
RuleBasedBreakIterator * lineIter1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createLineInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) )
errln ( " FAIL : in construction " ) ;
else {
lineIter1 - > setText ( testString ) ;
p = lineIter1 - > first ( ) ;
if ( p ! = 0 )
errln ( ( UnicodeString ) " ERROR: first() returned " + p + ( UnicodeString ) " instead of 0 " ) ;
q = lineIter1 - > next ( ) ;
doTest ( testString , p , q , 7 , " Hello! " ) ;
p = q ;
p = q ;
q = lineIter1 - > next ( 4 ) ;
doTest ( testString , p , q , 20 , " how \r \n (are) \r " ) ;
p = q ;
q = lineIter1 - > next ( - 4 ) ;
doTest ( testString , p , q , 7 , " how \r \n (are) \r " ) ;
p = q ;
q = lineIter1 - > next ( 6 ) ;
doTest ( testString , p , q , 30 , " how \r \n (are) \r you? I'am " ) ;
p = q ;
q = lineIter1 - > next ( ) ;
doTest ( testString , p , q , 36 , " fine- " ) ;
p = q ;
q = lineIter1 - > next ( 2 ) ;
doTest ( testString , p , q , 54 , " Thankyou. foo \\ u00a0bar " ) ;
q = lineIter1 - > following ( 60 ) ;
doTest ( testString , 60 , q , 64 , " re, " ) ;
q = lineIter1 - > following ( 1 ) ;
doTest ( testString , 1 , q , 7 , " ello! " ) ;
q = lineIter1 - > following ( 10 ) ;
doTest ( testString , 10 , q , 12 , " \r \n " ) ;
q = lineIter1 - > following ( 20 ) ;
doTest ( testString , 20 , q , 25 , " you? " ) ;
p = lineIter1 - > following ( lineIter1 - > last ( ) ) ;
q = lineIter1 - > next ( lineIter1 - > last ( ) ) ;
if ( p ! = RuleBasedBreakIterator : : DONE | | q ! = RuleBasedBreakIterator : : DONE )
errln ( ( UnicodeString ) " ERROR: following()/next() at last position returned # " +
p + ( UnicodeString ) " and " + q + ( UnicodeString ) " instead of " + testString . length ( ) ) ;
}
delete wordIter1 ;
delete charIter1 ;
delete sentIter1 ;
delete lineIter1 ;
}
void RBBIAPITest : : TestLastPreviousPreceding ( )
{
int32_t p , q ;
UErrorCode status = U_ZERO_ERROR ;
UnicodeString testString = " This is a word break. Isn't it? 2.25 dollars " ;
logln ( ( UnicodeString ) " Testing last(),previous(), preceding() with custom rules " ) ;
logln ( ( UnicodeString ) " testing word iteration for string \" " + prettify ( testString ) + ( UnicodeString ) " \" " ) ;
RuleBasedBreakIterator * wordIter1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createWordInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) )
errln ( " FAIL : in construction " ) ;
else {
wordIter1 - > setText ( testString ) ;
p = wordIter1 - > last ( ) ;
if ( p ! = testString . length ( ) ) {
errln ( ( UnicodeString ) " ERROR: first() returned " + p + ( UnicodeString ) " instead of " + testString . length ( ) ) ;
}
q = wordIter1 - > previous ( ) ;
doTest ( testString , p , q , 37 , " dollars " ) ;
p = q ;
q = wordIter1 - > previous ( ) ;
doTest ( testString , p , q , 36 , " " ) ;
q = wordIter1 - > preceding ( 25 ) ;
doTest ( testString , 25 , q , 22 , " Isn " ) ;
p = q ;
q = wordIter1 - > previous ( ) ;
doTest ( testString , p , q , 21 , " " ) ;
q = wordIter1 - > preceding ( 20 ) ;
doTest ( testString , 20 , q , 15 , " break " ) ;
p = wordIter1 - > preceding ( wordIter1 - > first ( ) ) ;
if ( p ! = RuleBasedBreakIterator : : DONE )
errln ( ( UnicodeString ) " ERROR: preceding() at starting position returned # " + p + ( UnicodeString ) " instead of 0 " ) ;
}
status = U_ZERO_ERROR ;
testString = CharsToUnicodeString ( " Write hindi here. \\ u092d \\ u093e \\ u0930 \\ u0924 \\ u0938 \\ u0941 \\ u0902 \\ u0926 \\ u0930 \\ u0939 \\ u094c \\ u0964 " ) ;
logln ( ( UnicodeString ) " testing character iteration for string \" " + prettify ( testString ) + ( UnicodeString ) " \" \n " ) ;
RuleBasedBreakIterator * charIter1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createCharacterInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) )
errln ( " FAIL : in construction " ) ;
else {
charIter1 - > setText ( testString ) ;
p = charIter1 - > last ( ) ;
if ( p ! = testString . length ( ) )
errln ( ( UnicodeString ) " ERROR: first() returned " + p + ( UnicodeString ) " instead of " + testString . length ( ) ) ;
q = charIter1 - > previous ( ) ;
doTest ( testString , p , q , 31 , " \\ u0964 " ) ;
p = q ;
q = charIter1 - > previous ( ) ;
doTest ( testString , p , q , 29 , " \\ u0939 \\ u094c " ) ;
q = charIter1 - > preceding ( 26 ) ;
doTest ( testString , 26 , q , 23 , " \\ u0938 \\ u0941 \\ u0902 " ) ;
q = charIter1 - > preceding ( 16 ) ;
doTest ( testString , 16 , q , 15 , " e " ) ;
p = q ;
q = charIter1 - > previous ( ) ;
doTest ( testString , p , q , 14 , " r " ) ;
charIter1 - > first ( ) ;
p = charIter1 - > previous ( ) ;
q = charIter1 - > preceding ( charIter1 - > first ( ) ) ;
if ( p ! = RuleBasedBreakIterator : : DONE | | q ! = RuleBasedBreakIterator : : DONE )
errln ( ( UnicodeString ) " ERROR: previous()/preceding() at starting position returned # " +
p + ( UnicodeString ) " and " + q + ( UnicodeString ) " instead of 0 \n " ) ;
}
status = U_ZERO_ERROR ;
testString = " Hello! how are you? I'am fine. Thankyou. How are you doing? This \n costs $20,00,000. " ;
logln ( ( UnicodeString ) " testing sentence iter - String:- \" " + prettify ( testString ) + ( UnicodeString ) " \" " ) ;
RuleBasedBreakIterator * sentIter1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createSentenceInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) )
errln ( " FAIL : in construction " ) ;
else {
sentIter1 - > setText ( testString ) ;
p = sentIter1 - > last ( ) ;
if ( p ! = testString . length ( ) )
errln ( ( UnicodeString ) " ERROR: last() returned " + p + ( UnicodeString ) " instead of " + testString . length ( ) ) ;
q = sentIter1 - > previous ( ) ;
doTest ( testString , p , q , 60 , " This \n costs $20,00,000. " ) ;
p = q ;
q = sentIter1 - > previous ( ) ;
doTest ( testString , p , q , 41 , " How are you doing? " ) ;
q = sentIter1 - > preceding ( 40 ) ;
doTest ( testString , 40 , q , 31 , " Thankyou. " ) ;
q = sentIter1 - > preceding ( 25 ) ;
doTest ( testString , 25 , q , 20 , " I'am " ) ;
sentIter1 - > first ( ) ;
p = sentIter1 - > previous ( ) ;
q = sentIter1 - > preceding ( sentIter1 - > first ( ) ) ;
if ( p ! = RuleBasedBreakIterator : : DONE | | q ! = RuleBasedBreakIterator : : DONE )
errln ( ( UnicodeString ) " ERROR: previous()/preceding() at starting position returned # " +
p + ( UnicodeString ) " and " + q + ( UnicodeString ) " instead of 0 \n " ) ;
}
status = U_ZERO_ERROR ;
testString = " Hello! how are you? I'am fine. Thankyou. How are you doing? This \n costs $20,00,000. " ;
logln ( ( UnicodeString ) " testing line iter - String:- \" " + prettify ( testString ) + ( UnicodeString ) " \" " ) ;
RuleBasedBreakIterator * lineIter1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createLineInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) )
errln ( " FAIL : in construction " ) ;
else {
lineIter1 - > setText ( testString ) ;
p = lineIter1 - > last ( ) ;
if ( p ! = testString . length ( ) )
errln ( ( UnicodeString ) " ERROR: last() returned " + p + ( UnicodeString ) " instead of " + testString . length ( ) ) ;
q = lineIter1 - > previous ( ) ;
doTest ( testString , p , q , 72 , " $20,00,000. " ) ;
p = q ;
q = lineIter1 - > previous ( ) ;
doTest ( testString , p , q , 66 , " costs " ) ;
q = lineIter1 - > preceding ( 40 ) ;
doTest ( testString , 40 , q , 31 , " Thankyou. " ) ;
q = lineIter1 - > preceding ( 25 ) ;
doTest ( testString , 25 , q , 20 , " I'am " ) ;
lineIter1 - > first ( ) ;
p = lineIter1 - > previous ( ) ;
q = lineIter1 - > preceding ( sentIter1 - > first ( ) ) ;
if ( p ! = RuleBasedBreakIterator : : DONE | | q ! = RuleBasedBreakIterator : : DONE )
errln ( ( UnicodeString ) " ERROR: previous()/preceding() at starting position returned # " +
p + ( UnicodeString ) " and " + q + ( UnicodeString ) " instead of 0 \n " ) ;
}
delete sentIter1 ;
delete charIter1 ;
delete wordIter1 ;
delete lineIter1 ;
}
void RBBIAPITest : : TestIsBoundary ( ) {
UErrorCode status = U_ZERO_ERROR ;
UnicodeString testString1 = CharsToUnicodeString ( " Write here. \\ u092d \\ u093e \\ u0930 \\ u0924 \\ u0938 \\ u0941 \\ u0902 \\ u0926 \\ u0930 \\ u0939 \\ u094c \\ u0964 " ) ;
RuleBasedBreakIterator * charIter1 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createCharacterInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) )
errln ( " FAIL: in construction " ) ;
else {
charIter1 - > setText ( testString1 ) ;
int32_t bounds1 [ ] = { 0 , 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 14 , 15 , 16 , 17 , 20 , 21 , 22 , 23 , 25 , 26 } ;
doBoundaryTest ( * charIter1 , testString1 , bounds1 ) ;
}
RuleBasedBreakIterator * wordIter2 = ( RuleBasedBreakIterator * ) RuleBasedBreakIterator : : createWordInstance ( Locale : : getDefault ( ) , status ) ;
if ( U_FAILURE ( status ) )
errln ( " FAIL : in construction " ) ;
else {
wordIter2 - > setText ( testString1 ) ;
int32_t bounds2 [ ] = { 0 , 5 , 6 , 10 , 11 , 12 , 16 , 17 , 22 , 23 , 26 } ;
doBoundaryTest ( * wordIter2 , testString1 , bounds2 ) ;
}
delete wordIter2 ;
delete charIter1 ;
}
//---------------------------------------------
// runIndexedTest
//---------------------------------------------
void RBBIAPITest : : runIndexedTest ( int32_t index , bool_t exec , char * & name , char * par )
{
if ( exec ) logln ( ( UnicodeString ) " TestSuite RuleBasedBreakIterator API " ) ;
switch ( index ) {
// case 0: name = "TestConstruction"; if (exec) TestConstruction(); break;
case 0 : name = " TestCloneEquals " ; if ( exec ) TestCloneEquals ( ) ; break ;
case 1 : name = " TestgetRules " ; if ( exec ) TestgetRules ( ) ; break ;
case 2 : name = " TestHashCode " ; if ( exec ) TestHashCode ( ) ; break ;
case 3 : name = " TestGetSetAdoptText " ; if ( exec ) TestGetSetAdoptText ( ) ; break ;
case 4 : name = " TestFirstNextFollowing " ; if ( exec ) TestFirstNextFollowing ( ) ; break ;
case 5 : name = " TestLastPreviousPreceding " ; if ( exec ) TestLastPreviousPreceding ( ) ; break ;
case 6 : name = " TestIsBoundary " ; if ( exec ) TestIsBoundary ( ) ; break ;
default : name = " " ; break ; /*needed to end loop*/
}
}
//---------------------------------------------
//Internal subroutines
//---------------------------------------------
void RBBIAPITest : : doBoundaryTest ( RuleBasedBreakIterator & bi , UnicodeString & text , int32_t * boundaries ) {
logln ( ( UnicodeString ) " testIsBoundary(): " ) ;
int32_t p = 0 ;
bool_t isB ;
for ( int32_t 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 RBBIAPITest : : doTest ( UnicodeString & testString , int32_t start , int32_t gotoffset , int32_t expectedOffset , const char * expectedString ) {
UnicodeString selected ;
UnicodeString expected = CharsToUnicodeString ( expectedString ) ;
if ( gotoffset ! = expectedOffset )
errln ( ( UnicodeString ) " ERROR:****returned # " + gotoffset + ( UnicodeString ) " instead of # " + expectedOffset ) ;
if ( start < = gotoffset ) {
testString . extractBetween ( start , gotoffset , selected ) ;
}
else {
testString . extractBetween ( gotoffset , start , selected ) ;
}
if ( selected . compare ( expected ) ! = 0 )
errln ( prettify ( ( UnicodeString ) " ERROR:****selected \" " + selected + " \" instead of \" " + expected + " \" " ) ) ;
else
logln ( prettify ( " ****selected \" " + selected + " \" " ) ) ;
}
2000-01-20 06:48:37 +00:00