07a2bc0937
FAILURE -> U_FAILURE etc. X-SVN-Rev: 76
599 lines
19 KiB
C
599 lines
19 KiB
C
/*
|
|
*****************************************************************************************
|
|
* *
|
|
* COPYRIGHT: *
|
|
* (C) Copyright Taligent, Inc., 1996 *
|
|
* (C) Copyright International Business Machines Corporation, 1999 *
|
|
* Licensed Material - Program-Property of IBM - All Rights Reserved. *
|
|
* US Government Users Restricted Rights - Use, duplication, or disclosure *
|
|
* restricted by GSA ADP Schedule Contract with IBM Corp. *
|
|
* *
|
|
*****************************************************************************************
|
|
********************************************************************************
|
|
*
|
|
* File CCONVTST.C
|
|
*
|
|
* Modification History:
|
|
* Name Description
|
|
* Steven R. Loomis 7/8/1999 Adding input buffer test
|
|
*********************************************************************************
|
|
*/
|
|
#include <stdio.h>
|
|
#include <stdlib.h>
|
|
#include <string.h>
|
|
#include <ctype.h>
|
|
#include "uloc.h"
|
|
#include "ucnv.h"
|
|
#include "ucnv_err.h"
|
|
#include "cintltst.h"
|
|
#include "utypes.h"
|
|
#include "ustring.h"
|
|
|
|
static void printSeq(const char* a, int len);
|
|
static void printUSeq(const UChar* a, int len);
|
|
|
|
void TestNewConvertWithBufferSizes(int32_t osize, int32_t isize) ;
|
|
void TestConverterTypesAndStarters();
|
|
|
|
#define NEW_MAX_BUFFER 999
|
|
|
|
static int32_t gInBufferSize = 0;
|
|
static int32_t gOutBufferSize = 0;
|
|
static char gNuConvTestName[1024];
|
|
|
|
#define nct_min(x,y) ((x<y) ? x : y)
|
|
|
|
void printSeq(const char* a, int len)
|
|
{
|
|
int i=0;
|
|
log_verbose("\n{");
|
|
while (i<len) log_verbose("%X", a[i++]);
|
|
log_verbose("}\n");
|
|
}
|
|
void printUSeq(const UChar* a, int len)
|
|
{
|
|
int i=0;
|
|
log_verbose("\n{");
|
|
while (i<len) log_verbose("%4X", a[i++]);
|
|
log_verbose("}\n");
|
|
}
|
|
|
|
void printSeqErr(const char* a, int len)
|
|
{
|
|
int i=0;
|
|
fprintf(stderr, "\n{");
|
|
while (i<len) fprintf(stderr, "%X", a[i++]);
|
|
fprintf(stderr, "}\n");
|
|
}
|
|
void printUSeqErr(const UChar* a, int len)
|
|
{
|
|
int i=0;
|
|
fprintf(stderr, "\n{");
|
|
while (i<len) fprintf(stderr, "%4X", a[i++]);
|
|
fprintf(stderr,"}\n");
|
|
}
|
|
|
|
void TestInBufSizes()
|
|
{
|
|
TestNewConvertWithBufferSizes(NEW_MAX_BUFFER,1);
|
|
#if 0
|
|
TestNewConvertWithBufferSizes(NEW_MAX_BUFFER,2);
|
|
TestNewConvertWithBufferSizes(NEW_MAX_BUFFER,3);
|
|
TestNewConvertWithBufferSizes(NEW_MAX_BUFFER,4);
|
|
TestNewConvertWithBufferSizes(NEW_MAX_BUFFER,5);
|
|
TestNewConvertWithBufferSizes(NEW_MAX_BUFFER,6);
|
|
TestNewConvertWithBufferSizes(1,1);
|
|
TestNewConvertWithBufferSizes(2,3);
|
|
TestNewConvertWithBufferSizes(3,2);
|
|
#endif
|
|
}
|
|
|
|
void TestOutBufSizes()
|
|
{
|
|
#if 0
|
|
TestNewConvertWithBufferSizes(NEW_MAX_BUFFER,NEW_MAX_BUFFER);
|
|
TestNewConvertWithBufferSizes(1,NEW_MAX_BUFFER);
|
|
TestNewConvertWithBufferSizes(2,NEW_MAX_BUFFER);
|
|
TestNewConvertWithBufferSizes(3,NEW_MAX_BUFFER);
|
|
TestNewConvertWithBufferSizes(4,NEW_MAX_BUFFER);
|
|
TestNewConvertWithBufferSizes(5,NEW_MAX_BUFFER);
|
|
#endif
|
|
}
|
|
|
|
|
|
void addTestNewConvert(TestNode** root)
|
|
{
|
|
addTest(root, &TestInBufSizes, "tsconv/nucnvtst/TestInBufSizes");
|
|
addTest(root, &TestOutBufSizes, "tsconv/nucnvtst/TestOutBufSizes");
|
|
addTest(root, &TestConverterTypesAndStarters, "tsconv/nucnvtst/TestConverterTypesAndStarters");
|
|
|
|
}
|
|
|
|
|
|
/* Note that this test already makes use of statics, so it's not really
|
|
multithread safe.
|
|
This convenience function lets us make the error messages actually useful.
|
|
*/
|
|
|
|
void setNuConvTestName(const char *codepage, const char *direction)
|
|
{
|
|
sprintf(gNuConvTestName, "[Testing %s %s Unicode, InputBufSiz=%d, OutputBufSiz=%d]",
|
|
codepage,
|
|
direction,
|
|
gInBufferSize,
|
|
gOutBufferSize);
|
|
}
|
|
|
|
bool_t testConvertFromU( const UChar *source, int sourceLen, const char *expect, int expectLen,
|
|
const char *codepage, int32_t *expectOffsets)
|
|
{
|
|
UErrorCode status = U_ZERO_ERROR;
|
|
UConverter *conv = 0;
|
|
char junkout[NEW_MAX_BUFFER]; /* FIX */
|
|
int32_t junokout[NEW_MAX_BUFFER]; /* FIX */
|
|
const UChar *src;
|
|
char *end;
|
|
char *targ;
|
|
int32_t *offs;
|
|
int i;
|
|
int32_t realBufferSize;
|
|
char *realBufferEnd;
|
|
const UChar *realSourceEnd;
|
|
const UChar *sourceLimit;
|
|
bool_t checkOffsets = TRUE;
|
|
bool_t doFlush;
|
|
|
|
for(i=0;i<NEW_MAX_BUFFER;i++)
|
|
junkout[i] = 0xF0;
|
|
for(i=0;i<NEW_MAX_BUFFER;i++)
|
|
junokout[i] = 0xFF;
|
|
|
|
setNuConvTestName(codepage, "FROM");
|
|
|
|
log_verbose("\n========= %s\n", gNuConvTestName);
|
|
|
|
conv = ucnv_open(codepage, &status);
|
|
if(U_FAILURE(status))
|
|
{
|
|
log_err("Couldn't open converter %s\n",codepage);
|
|
return FALSE;
|
|
}
|
|
|
|
log_verbose("Converter opened..\n");
|
|
|
|
src = source;
|
|
targ = junkout;
|
|
offs = junokout;
|
|
|
|
realBufferSize = (sizeof(junkout)/sizeof(junkout[0]));
|
|
realBufferEnd = junkout + realBufferSize;
|
|
realSourceEnd = source + sourceLen;
|
|
|
|
if ( gOutBufferSize != realBufferSize )
|
|
checkOffsets = FALSE;
|
|
|
|
if( gInBufferSize != NEW_MAX_BUFFER )
|
|
checkOffsets = FALSE;
|
|
|
|
do
|
|
{
|
|
end = nct_min(targ + gOutBufferSize, realBufferEnd);
|
|
sourceLimit = nct_min(src + gInBufferSize, realSourceEnd);
|
|
|
|
doFlush = (sourceLimit == realSourceEnd);
|
|
|
|
if(targ == realBufferEnd)
|
|
{
|
|
log_err("Error, overflowed the real buffer while about to call fromUnicode! targ=%08lx %s", targ, gNuConvTestName);
|
|
return FALSE;
|
|
}
|
|
log_verbose("calling fromUnicode @ SOURCE:%08lx to %08lx TARGET: %08lx to %08lx, flush=%s\n", src,sourceLimit, targ,end, doFlush?"TRUE":"FALSE");
|
|
|
|
|
|
status = U_ZERO_ERROR;
|
|
|
|
ucnv_fromUnicode (conv,
|
|
&targ,
|
|
end,
|
|
&src,
|
|
sourceLimit,
|
|
checkOffsets ? offs : NULL,
|
|
doFlush, /* flush if we're at the end of the input data */
|
|
&status);
|
|
|
|
} while ( (status == U_INDEX_OUTOFBOUNDS_ERROR) || (sourceLimit < realSourceEnd) );
|
|
|
|
if(U_FAILURE(status))
|
|
{
|
|
log_err("Problem tdoing fromUnicode, errcode %d %s\n", codepage, status, gNuConvTestName);
|
|
return FALSE;
|
|
}
|
|
|
|
log_verbose("\nConversion done [%d uchars in -> %d chars out]. \nResult :",
|
|
sourceLen, targ-junkout);
|
|
if(VERBOSITY)
|
|
{
|
|
char junk[9999];
|
|
char offset_str[9999];
|
|
char *p;
|
|
|
|
junk[0] = 0;
|
|
offset_str[0] = 0;
|
|
for(p = junkout;p<targ;p++)
|
|
{
|
|
sprintf(junk + strlen(junk), "0x%02x, ", (0xFF) & (unsigned int)*p);
|
|
sprintf(offset_str + strlen(offset_str), "0x%02x, ", (0xFF) & (unsigned int)junokout[p-junkout]);
|
|
}
|
|
|
|
log_verbose(junk);
|
|
printSeq(expect, expectLen);
|
|
if ( checkOffsets )
|
|
{
|
|
log_verbose("\nOffsets:");
|
|
log_verbose(offset_str);
|
|
}
|
|
log_verbose("\n");
|
|
}
|
|
ucnv_close(conv);
|
|
|
|
|
|
if(expectLen != targ-junkout)
|
|
{
|
|
log_err("Expected %d chars out, got %d %s\n", expectLen, targ-junkout, gNuConvTestName);
|
|
return FALSE;
|
|
}
|
|
|
|
if (checkOffsets && (expectOffsets != 0) )
|
|
{
|
|
log_verbose("comparing %d offsets..\n", targ-junkout);
|
|
if(memcmp(junokout,expectOffsets,(targ-junkout) * sizeof(int32_t) ))
|
|
log_err("did not get the expected offsets. %s", gNuConvTestName);
|
|
}
|
|
|
|
log_verbose("comparing..\n");
|
|
if(!memcmp(junkout, expect, expectLen))
|
|
{
|
|
log_verbose("Matches!\n");
|
|
return TRUE;
|
|
}
|
|
else
|
|
{
|
|
log_err("String does not match. %s\n", gNuConvTestName);
|
|
printSeqErr(junkout, expectLen);
|
|
printSeqErr(expect, expectLen);
|
|
|
|
return FALSE;
|
|
}
|
|
}
|
|
|
|
bool_t testConvertToU( const char *source, int sourcelen, const UChar *expect, int expectlen,
|
|
const char *codepage, int32_t *expectOffsets)
|
|
{
|
|
UErrorCode status = U_ZERO_ERROR;
|
|
UConverter *conv = 0;
|
|
UChar junkout[NEW_MAX_BUFFER]; /* FIX */
|
|
int32_t junokout[NEW_MAX_BUFFER]; /* FIX */
|
|
const char *src;
|
|
const char *realSourceEnd;
|
|
const char *srcLimit;
|
|
UChar *targ;
|
|
UChar *end;
|
|
int32_t *offs;
|
|
int i;
|
|
bool_t checkOffsets = TRUE;
|
|
|
|
int32_t realBufferSize;
|
|
UChar *realBufferEnd;
|
|
|
|
|
|
for(i=0;i<NEW_MAX_BUFFER;i++)
|
|
junkout[i] = 0xFFFE;
|
|
|
|
for(i=0;i<NEW_MAX_BUFFER;i++)
|
|
junokout[i] = -1;
|
|
|
|
setNuConvTestName(codepage, "TO");
|
|
|
|
log_verbose("\n========= %s\n", gNuConvTestName);
|
|
|
|
conv = ucnv_open(codepage, &status);
|
|
if(U_FAILURE(status))
|
|
{
|
|
log_err("Couldn't open converter %s\n",gNuConvTestName);
|
|
return FALSE;
|
|
}
|
|
|
|
log_verbose("Converter opened..\n");
|
|
|
|
src = source;
|
|
targ = junkout;
|
|
offs = junokout;
|
|
|
|
realBufferSize = (sizeof(junkout)/sizeof(junkout[0]));
|
|
realBufferEnd = junkout + realBufferSize;
|
|
realSourceEnd = src + sourcelen;
|
|
|
|
if ( gOutBufferSize != realBufferSize )
|
|
checkOffsets = FALSE;
|
|
|
|
if( gInBufferSize != NEW_MAX_BUFFER )
|
|
checkOffsets = FALSE;
|
|
|
|
do
|
|
{
|
|
end = nct_min( targ + gOutBufferSize, realBufferEnd);
|
|
srcLimit = nct_min(realSourceEnd, src + gInBufferSize);
|
|
|
|
if(targ == realBufferEnd)
|
|
{
|
|
log_err("Error, the end would overflow the real output buffer while about to call toUnicode! tarjey=%08lx %s",targ,gNuConvTestName);
|
|
return FALSE;
|
|
}
|
|
log_verbose("calling toUnicode @ %08lx to %08lx\n", targ,end);
|
|
|
|
/* oldTarg = targ; */
|
|
|
|
status = U_ZERO_ERROR;
|
|
|
|
ucnv_toUnicode (conv,
|
|
&targ,
|
|
end,
|
|
&src,
|
|
srcLimit,
|
|
checkOffsets ? offs : NULL,
|
|
(srcLimit == realSourceEnd), /* flush if we're at the end of hte source data */
|
|
&status);
|
|
|
|
/* offs += (targ-oldTarg); */
|
|
|
|
} while ( (status == U_INDEX_OUTOFBOUNDS_ERROR) || (srcLimit < realSourceEnd) ); /* while we just need another buffer */
|
|
|
|
if(U_FAILURE(status))
|
|
{
|
|
log_err("Problem doing toUnicode, errcode %d %s\n", status, gNuConvTestName);
|
|
return FALSE;
|
|
}
|
|
|
|
log_verbose("\nConversion done. %d bytes -> %d chars.\nResult :",
|
|
sourcelen, targ-junkout);
|
|
if(VERBOSITY)
|
|
{
|
|
char junk[9999];
|
|
char offset_str[9999];
|
|
|
|
UChar *p;
|
|
|
|
junk[0] = 0;
|
|
offset_str[0] = 0;
|
|
|
|
for(p = junkout;p<targ;p++)
|
|
{
|
|
sprintf(junk + strlen(junk), "0x%04x, ", (0xFFFF) & (unsigned int)*p);
|
|
sprintf(offset_str + strlen(offset_str), "0x%04x, ", (0xFFFF) & (unsigned int)junokout[p-junkout]);
|
|
}
|
|
|
|
log_verbose(junk);
|
|
|
|
if ( checkOffsets )
|
|
{
|
|
log_verbose("\nOffsets:");
|
|
log_verbose(offset_str);
|
|
}
|
|
log_verbose("\n");
|
|
}
|
|
ucnv_close(conv);
|
|
|
|
log_verbose("comparing %d uchars (%d bytes)..\n",expectlen,expectlen*2);
|
|
|
|
if (checkOffsets && (expectOffsets != 0))
|
|
{
|
|
if(memcmp(junokout,expectOffsets,(targ-junkout) * sizeof(int32_t)))
|
|
log_err("did not get the expected offsets. %s",gNuConvTestName);
|
|
}
|
|
|
|
if(!memcmp(junkout, expect, expectlen*2))
|
|
{
|
|
log_verbose("Matches!\n");
|
|
return TRUE;
|
|
}
|
|
else
|
|
{
|
|
log_err("String does not match. %s\n", gNuConvTestName);
|
|
printUSeq(expect, expectlen);
|
|
return FALSE;
|
|
}
|
|
}
|
|
|
|
|
|
void TestNewConvertWithBufferSizes(int32_t outsize, int32_t insize )
|
|
{
|
|
/** test chars #1 */
|
|
/* 1 2 3 1Han 2Han 3Han . */
|
|
UChar sampleText[] =
|
|
{ 0x0031, 0x0032, 0x0033, 0x4e00, 0x4e8c, 0x4e09, 0x002E };
|
|
|
|
int32_t fmUTF8Offs[] =
|
|
{ 0x0000, 0x0001, 0x0002, 0x0003, 0x0006, 0x0009, 0x000c };
|
|
|
|
int32_t fmISO2022Offs[] =
|
|
{ 0x03, 0x04, 0x05, 0x06, 0x09, 0x0c, 0x0f }; /* is this right? */
|
|
|
|
int32_t fmIBM930Offs[] =
|
|
{ 0x0000, 0x0001, 0x0002, 0x0004, 0x0006, 0x0008, 0x000b, };
|
|
|
|
int32_t fmIBM943Offs[] =
|
|
{ 0x0000, 0x0001, 0x0002, 0x0003, 0x0005, 0x0007, 0x0009, };
|
|
|
|
int32_t fmUTF16LEOffs[] =
|
|
{ 0x0000, 0x0002, 0x0004, 0x0006, 0x0008, 0x000a, 0x000c, };
|
|
|
|
const char expectedUTF8[] =
|
|
{ (char)0x31, (char)0x32, (char)0x33, (char)0xe4, (char)0xb8, (char)0x80, (char)0xe4, (char)0xba, (char)0x8c, (char)0xe4, (char)0xb8, (char)0x89, (char)0x2E };
|
|
|
|
int32_t toUTF8Offs[] =
|
|
{ (char)0x00, (char)0x01, (char)0x02, (char)0x03, (char)0x03, (char)0x03, (char)0x04, (char)0x04, (char)0x04, (char)0x05, (char)0x05, (char)0x05, (char)0x06 };
|
|
|
|
/* Same as UTF8, but with ^[%B preceeding */
|
|
const char expectedISO2022[] =
|
|
{ (char)0x1b, (char)0x25, (char)0x42, (char)0x31, (char)0x32, (char)0x33, (char)0xe4, (char)0xb8, (char)0x80, (char)0xe4, (char)0xba, (char)0x8c, (char)0xe4, (char)0xb8, (char)0x89, (char)0x2E };
|
|
|
|
int32_t toISO2022Offs[] =
|
|
{ (char)0xff, (char)0xff, (char)0xff, (char)0x00, (char)0x01, (char)0x02, (char)0x03, (char)0x03, (char)0x03,
|
|
(char)0x04, (char)0x04, (char)0x04, (char)0x05, (char)0x05, (char)0x05, (char)0x06 }; /* right? */
|
|
|
|
/* 1 2 3 <SO> h1 h2 h3 <SI> . */
|
|
const char expectedIBM930[] =
|
|
{ (char)0xF1, (char)0xF2, (char)0xF3, (char)0x0E, (char)0x45, (char)0x41, (char)0x45, (char)0x42, (char)0x45, (char)0x43, (char)0x0F, (char)0x4B };
|
|
|
|
int32_t toIBM930Offs[] =
|
|
{ (char)0x00, (char)0x01, (char)0x02, (char)0x03, (char)0x03, (char)0x03, (char)0x04, (char)0x04, (char)0x05, (char)0x05, (char)0x06, (char)0x06, };
|
|
|
|
/* 1 2 3 <?> <?> <?> . */
|
|
const char expectedISO88593[] =
|
|
{ (char)0x31, (char)0x32, (char)0x33, (char)0x1a, (char)0x1a, (char)0x1a, (char)0x2E };
|
|
|
|
int32_t toISO88593Offs[] =
|
|
{(char) 0x00, (char)0x01, (char)0x02, (char)0x03, (char)0x04, (char)0x05, (char)0x06, };
|
|
|
|
/* 1 2 3 h1 h2 h3 . */
|
|
const char expectedIBM943[] =
|
|
{ (char)0x31, (char)0x32, (char)0x33, (char)0x88, (char)0xea, (char)0x93, (char)0xf1, (char)0x8e, (char)0x4f, (char)0x2e };
|
|
|
|
int32_t toIBM943Offs [] =
|
|
{ (char)0x00, (char)0x01, (char)0x02, (char)0x03, (char)0x03, (char)0x04, (char)0x04, (char)0x05, (char)0x05, (char)0x06, };
|
|
|
|
/* etc */
|
|
const char expectedUTF16LE[] =
|
|
{ (char)0x31, (char)0x00, (char)0x32, (char)0x00, (char)0x33, (char)0x00, (char)0x00, (char)0x4e, (char)0x8c, (char)0x4e, (char)0x09, (char)0x4e, (char)0x2e, (char)0x00 };
|
|
int32_t toUTF16LEOffs[]=
|
|
{ (char)0x00, (char)0x00, (char)0x02, (char)0x02, (char)0x04, (char)0x04, (char)0x06, (char)0x06, (char)0x08, (char)0x08, (char)0x0a, (char)0x0a, (char)0x0c, (char)0x0c, };
|
|
|
|
|
|
/** Test chars #2 NOT USED YET**/
|
|
|
|
/* Sahha [health], slashed h's */
|
|
const UChar malteseUChars[] = { 0x0053, 0x0061, 0x0127, 0x0127, 0x0061 };
|
|
const char maltesechars[] = { '\0' };
|
|
/*********************************** START OF CODE finally *************/
|
|
|
|
gInBufferSize = insize;
|
|
gOutBufferSize = outsize;
|
|
|
|
log_verbose("\n\n\nTesting conversions with InputBufferSize = %d, OutputBufferSize = %d\n", gInBufferSize, gOutBufferSize);
|
|
|
|
|
|
#if 0
|
|
if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]),
|
|
expectedUTF8, sizeof(expectedUTF8), "UTF8", toUTF8Offs ))
|
|
log_err("u-> UTF8 did not match.\n");
|
|
|
|
if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]),
|
|
expectedISO2022, sizeof(expectedISO2022), "iso-2022", toISO2022Offs ))
|
|
log_err("u-> iso-2022 did not match.\n");
|
|
|
|
if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]),
|
|
expectedIBM930, sizeof(expectedIBM930), "ibm-930", toIBM930Offs ))
|
|
log_err("u-> ibm-930 did not match.\n");
|
|
|
|
if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]),
|
|
expectedISO88593, sizeof(expectedISO88593), "iso-8859-3", toISO88593Offs ))
|
|
log_err("u-> iso-8859-3 did not match.\n");
|
|
|
|
if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]),
|
|
expectedIBM943, sizeof(expectedIBM943), "ibm-943", toIBM943Offs ))
|
|
log_err("u-> ibm-943 [UCNV_MBCS] not match.\n");
|
|
|
|
if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]),
|
|
expectedUTF16LE, sizeof(expectedUTF16LE), "utf-16le", toUTF16LEOffs ))
|
|
log_err("u-> utf-16le did not match.\n");
|
|
|
|
/****/
|
|
#endif
|
|
if(!testConvertToU(expectedUTF8, sizeof(expectedUTF8),
|
|
sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "utf8", fmUTF8Offs ))
|
|
log_err("utf8 -> u did not match\n");
|
|
|
|
if(!testConvertToU(expectedISO2022, sizeof(expectedISO2022),
|
|
sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "iso-2022", fmISO2022Offs ))
|
|
|
|
log_err("iso-2022 -> u did not match");
|
|
|
|
#if 0
|
|
if(!testConvertToU(expectedIBM930, sizeof(expectedIBM930),
|
|
sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "ibm-930", fmIBM930Offs ))
|
|
log_err("ibm-930 -> u did not match");
|
|
|
|
if(!testConvertToU(expectedIBM943, sizeof(expectedIBM943),
|
|
sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "ibm-943", fmIBM943Offs ))
|
|
log_err("ibm-943 -> u did not match");
|
|
|
|
if(!testConvertToU(expectedUTF16LE, sizeof(expectedUTF16LE),
|
|
sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "utf-16le", fmUTF16LEOffs ))
|
|
log_err("utf-16le -> u did not match");
|
|
#endif
|
|
}
|
|
|
|
void TestConverterTypesAndStarters()
|
|
{
|
|
UConverter* myConverter[3];
|
|
UErrorCode err = U_ZERO_ERROR;
|
|
bool_t mystarters[256];
|
|
int i;
|
|
|
|
const bool_t expectedKSCstarters[256] = {
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
|
|
FALSE, FALSE, FALSE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE,
|
|
TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE,
|
|
TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE,
|
|
TRUE, TRUE, TRUE, FALSE, FALSE, TRUE, TRUE, TRUE, TRUE, TRUE,
|
|
TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE,
|
|
TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE,
|
|
TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE,
|
|
TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE,
|
|
TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE,
|
|
TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE,
|
|
TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE,
|
|
TRUE, TRUE, TRUE, TRUE, TRUE, TRUE};
|
|
|
|
|
|
log_verbose("Testing KSC, ibm-930, ibm-878 for starters and their conversion types.");
|
|
|
|
myConverter[0] = ucnv_open("ksc", &err);
|
|
if (U_FAILURE(err)) log_err("Failed to create an ibm-949 converter\n");
|
|
myConverter[1] = ucnv_open("ibm-930", &err);
|
|
if (U_FAILURE(err)) log_err("Failed to create an ibm-930 converter\n");
|
|
myConverter[2] = ucnv_open("ibm-878", &err);
|
|
if (U_FAILURE(err)) log_err("Failed to create an ibm-815 converter\n");
|
|
|
|
if (ucnv_getType(myConverter[0])!=UCNV_MBCS) log_err("ucnv_getType Failed for ibm-949\n");
|
|
else log_verbose("ucnv_getType ibm-949 ok\n");
|
|
if (ucnv_getType(myConverter[1])!=UCNV_EBCDIC_STATEFUL) log_err("ucnv_getType Failed for ibm-930\n");
|
|
else log_verbose("ucnv_getType ibm-930 ok\n");
|
|
if (ucnv_getType(myConverter[2])!=UCNV_SBCS) log_err("ucnv_getType Failed for ibm-815\n");
|
|
else log_verbose("ucnv_getType ibm-815 ok\n");
|
|
|
|
|
|
ucnv_getStarters(myConverter[0], mystarters, &err);
|
|
/*if (memcmp(expectedKSCstarters, mystarters, sizeof(expectedKSCstarters)))
|
|
log_err("Failed ucnv_getStarters for ksc\n");
|
|
else
|
|
log_verbose("ucnv_getStarters ok\n");*/
|
|
|
|
ucnv_close(myConverter[0]);
|
|
ucnv_close(myConverter[1]);
|
|
ucnv_close(myConverter[2]);
|
|
}
|