+// © 2016 and later: Unicode, Inc. and others.
+// License & terms of use: http://www.unicode.org/copyright.html
/********************************************************************
* COPYRIGHT:
- * Copyright (c) 1997-2003, International Business Machines Corporation and
+ * Copyright (c) 1997-2016, International Business Machines Corporation and
* others. All Rights Reserved.
***************************************************************************/
-/*******************************************************************************
+/*****************************************************************************
*
* File NCNVCBTS
*
* Modification History:
* Name Date Description
* Madhu Katragadda 06/23/2000 Tests for Conveter FallBack API and Functionality
-**********************************************************************************
+******************************************************************************
*/
#include <stdio.h>
#include "unicode/uloc.h"
#include "cmemory.h"
#include "cstring.h"
+#if !UCONFIG_NO_LEGACY_CONVERSION
#define NEW_MAX_BUFFER 999
void addTestConverterFallBack(TestNode** root)
{
+#if !UCONFIG_NO_FILE_IO
addTest(root, &TestConverterFallBack, "tsconv/ncnvfbts/TestConverterFallBack");
+#endif
}
static void setNuConvTestName(const char *codepage, const char *direction)
{
- sprintf(gNuConvTestName, "[Testing %s %s Unicode, InputBufSiz=%d, OutputBufSiz=%d]",
- codepage,
- direction,
- gInBufferSize,
- gOutBufferSize);
+ sprintf(gNuConvTestName, "[Testing %s %s Unicode, InputBufSiz=%d, OutputBufSiz=%d]",
+ codepage,
+ direction,
+ (int)gInBufferSize,
+ (int)gOutBufferSize);
}
UErrorCode status = U_ZERO_ERROR;
UConverter *conv = 0;
- uint8_t junkout[NEW_MAX_BUFFER]; /* FIX */
+ char junkout[NEW_MAX_BUFFER]; /* FIX */
int32_t junokout[NEW_MAX_BUFFER]; /* FIX */
const UChar *src;
- uint8_t *end;
- uint8_t *targ;
+ char *end;
+ char *targ;
int32_t *offs;
int i;
int32_t realBufferSize;
- uint8_t *realBufferEnd;
+ char *realBufferEnd;
const UChar *realSourceEnd;
const UChar *sourceLimit;
UBool checkOffsets = TRUE;
UBool doFlush;
UBool action=FALSE;
- uint8_t *p;
+ char *p;
for(i=0;i<NEW_MAX_BUFFER;i++)
- junkout[i] = 0xF0;
+ junkout[i] = (char)0xF0;
for(i=0;i<NEW_MAX_BUFFER;i++)
junokout[i] = 0xFF;
setNuConvTestName(codepage, "FROM");
targ = junkout;
offs = junokout;
- realBufferSize = (sizeof(junkout)/sizeof(junkout[0]));
+ realBufferSize = UPRV_LENGTHOF(junkout);
realBufferEnd = junkout + realBufferSize;
realSourceEnd = source + sourceLen;
log_verbose("\nConversion done [%d uchars in -> %d chars out]. \nResult :",
sourceLen, targ-junkout);
- if(VERBOSITY)
+ if(getTestOption(VERBOSITY_OPTION))
{
char junk[9999];
char offset_str[9999];
{
log_err("Expected %d chars out, got %d %s\n", expectLen, targ-junkout, gNuConvTestName);
log_verbose("Expected %d chars out, got %d %s\n", expectLen, targ-junkout, gNuConvTestName);
- printSeqErr((const unsigned char*)junkout, targ-junkout);
+ printSeqErr((const unsigned char*)junkout, (int32_t)(targ-junkout));
printSeqErr((const unsigned char*)expect, expectLen);
return FALSE;
}
if(uprv_memcmp(junokout,expectOffsets,(targ-junkout) * sizeof(int32_t) )){
log_err("\ndid not get the expected offsets while %s \n", gNuConvTestName);
log_err("Got : ");
- printSeqErr((const unsigned char*)junkout, targ-junkout);
+ printSeqErr((const unsigned char*)junkout, (int32_t)(targ-junkout));
for(p=junkout;p<targ;p++)
log_err("%d, ", junokout[p-junkout]);
log_err("\nExpected: ");
UConverter *conv = 0;
UChar junkout[NEW_MAX_BUFFER]; /* FIX */
int32_t junokout[NEW_MAX_BUFFER]; /* FIX */
- const uint8_t *src;
- const uint8_t *realSourceEnd;
- const uint8_t *srcLimit;
+ const char *src;
+ const char *realSourceEnd;
+ const char *srcLimit;
UChar *targ;
UChar *end;
int32_t *offs;
log_verbose("Converter opened..\n");
- src = source;
+ src = (const char *)source;
targ = junkout;
offs = junokout;
- realBufferSize = (sizeof(junkout)/sizeof(junkout[0]));
+ realBufferSize = UPRV_LENGTHOF(junkout);
realBufferEnd = junkout + realBufferSize;
realSourceEnd = src + sourcelen;
/*----setting the fallback routine----*/
log_verbose("\nConversion done. %d bytes -> %d chars.\nResult :",
sourcelen, targ-junkout);
- if(VERBOSITY)
+ if(getTestOption(VERBOSITY_OPTION))
{
junk[0] = 0;
for(i=0; i<(targ-junkout); i++)
log_err("0x%04X,", junkout[i]);
log_err("");
- for(i=0; i<(src-source); i++)
+ for(i=0; i<(src-(const char *)source); i++)
log_err("0x%04X,", (unsigned char)source[i]);
}
}
static const char* nativeCodePage[]={
/*NLCS Mapping*/
- "ibm-367",
+ "ibm-437",
+ "ibm-850",
+ "ibm-878",
+ "ibm-923",
"ibm-1051",
"ibm-1089",
"ibm-1250",
"ibm-1256",
"ibm-1257",
"ibm-1258",
- "ibm-1275",
"ibm-1276"
};
gInBufferSize = insize;
gOutBufferSize = outsize;
- for(i=0; i<sizeof(nativeCodePage)/sizeof(nativeCodePage[0]); i++){
+ for(i=0; i<UPRV_LENGTHOF(nativeCodePage); i++){
log_verbose("Testing %s\n", nativeCodePage[i]);
- if(!testConvertFromUnicode(SBCSText, sizeof(SBCSText)/sizeof(SBCSText[0]),
+ if(!testConvertFromUnicode(SBCSText, UPRV_LENGTHOF(SBCSText),
expectedNative, sizeof(expectedNative), nativeCodePage[i], TRUE, toNativeOffs ))
log_err("u-> %s(SBCS) with FallBack did not match.\n", nativeCodePage[i]);
if(!testConvertToUnicode(expectedNative, sizeof(expectedNative),
- retrievedSBCSText, sizeof(retrievedSBCSText)/sizeof(retrievedSBCSText[0]), nativeCodePage[i], TRUE, fromNativeoffs ))
+ retrievedSBCSText, UPRV_LENGTHOF(retrievedSBCSText), nativeCodePage[i], TRUE, fromNativeoffs ))
log_err("%s->u(SBCS) with Fallback did not match.\n", nativeCodePage[i]);
}
/*DBCS*/
- if(!testConvertFromUnicode(DBCSText, sizeof(DBCSText)/sizeof(DBCSText[0]),
+ if(!testConvertFromUnicode(DBCSText, UPRV_LENGTHOF(DBCSText),
expectedIBM1363_DBCS, sizeof(expectedIBM1363_DBCS), "ibm-1363", TRUE, toIBM1363Offs_DBCS ))
log_err("u-> ibm-1363(DBCS portion) with FallBack did not match.\n");
if(!testConvertToUnicode(expectedIBM1363_DBCS, sizeof(expectedIBM1363_DBCS),
- retrievedDBCSText, sizeof(retrievedDBCSText)/sizeof(retrievedDBCSText[0]),"ibm-1363", TRUE, fromIBM1363offs_DBCS ))
+ retrievedDBCSText, UPRV_LENGTHOF(retrievedDBCSText),"ibm-1363", TRUE, fromIBM1363offs_DBCS ))
log_err("ibm-1363->u(DBCS portion) with Fallback did not match.\n");
/*MBCS*/
- if(!testConvertFromUnicode(MBCSText, sizeof(MBCSText)/sizeof(MBCSText[0]),
+ if(!testConvertFromUnicode(MBCSText, UPRV_LENGTHOF(MBCSText),
expectedIBM950, sizeof(expectedIBM950), "ibm-950", TRUE, toIBM950Offs ))
log_err("u-> ibm-950(MBCS) with FallBack did not match.\n");
if(!testConvertToUnicode(expectedIBM950, sizeof(expectedIBM950),
- retrievedMBCSText, sizeof(retrievedMBCSText)/sizeof(retrievedMBCSText[0]),"ibm-950", TRUE, fromIBM950offs ))
+ retrievedMBCSText, UPRV_LENGTHOF(retrievedMBCSText),"ibm-950", TRUE, fromIBM950offs ))
log_err("ibm-950->u(MBCS) with Fallback did not match.\n");
/*commented untill data table is available*/
UChar expectedFallbackFalse[]= { 0x5165, 0x5165, 0x516b, 0x516b, 0x9ef9, 0x9ef9};
if(!testConvertToUnicode(IBM950input, sizeof(IBM950input),
- expectedUnicodeText, sizeof(expectedUnicodeText)/sizeof(expectedUnicodeText[0]),"ibm-950", TRUE, fromIBM950inputOffs ))
+ expectedUnicodeText, UPRV_LENGTHOF(expectedUnicodeText),"ibm-950", TRUE, fromIBM950inputOffs ))
log_err("ibm-950->u(MBCS) with Fallback did not match.\n");
if(!testConvertToUnicode(IBM950input, sizeof(IBM950input),
- expectedFallbackFalse, sizeof(expectedFallbackFalse)/sizeof(expectedFallbackFalse[0]),"ibm-950", FALSE, fromIBM950inputOffs ))
+ expectedFallbackFalse, UPRV_LENGTHOF(expectedFallbackFalse),"ibm-950", FALSE, fromIBM950inputOffs ))
log_err("ibm-950->u(MBCS) with Fallback did not match.\n");
}
UChar expectedFallbackFalse[]= { 0x5C6E, 0x5C6E, 0x81FC, 0x81FC, 0x8278, 0x8278};
if(!testConvertToUnicode(euc_tw_input, sizeof(euc_tw_input),
- expectedUnicodeText, sizeof(expectedUnicodeText)/sizeof(expectedUnicodeText[0]),"euc-tw", TRUE, from_euc_tw_offs ))
+ expectedUnicodeText, UPRV_LENGTHOF(expectedUnicodeText),"euc-tw", TRUE, from_euc_tw_offs ))
log_err("from euc-tw->u with Fallback did not match.\n");
if(!testConvertToUnicode(euc_tw_input, sizeof(euc_tw_input),
- expectedFallbackFalse, sizeof(expectedFallbackFalse)/sizeof(expectedFallbackFalse[0]),"euc-tw", FALSE, from_euc_tw_offs ))
+ expectedFallbackFalse, UPRV_LENGTHOF(expectedFallbackFalse),"euc-tw", FALSE, from_euc_tw_offs ))
log_err("from euc-tw->u with Fallback false did not match.\n");
int32_t to_euc_tw_offs [] = { 0, 1, 2, 2, 3, 3, 4, 4, 5, 5, 5, 5, 6, 6,
6, 6, 7, 7, 7, 7, 8, 8, 9, 9, 10, 10, 10, 10, 11, 11, 11, 11, 12, 12, 12, 12};
- if(!testConvertFromUnicode(inputText, sizeof(inputText)/sizeof(inputText[0]),
+ if(!testConvertFromUnicode(inputText, UPRV_LENGTHOF(inputText),
expected_euc_tw, sizeof(expected_euc_tw), "euc-tw", TRUE, to_euc_tw_offs ))
log_err("u-> euc-tw with FallBack did not match.\n");
}
/*MBCS 1363*/
- if(!testConvertFromUnicode(MBCSText1363, sizeof(MBCSText1363)/sizeof(MBCSText1363[0]),
+ if(!testConvertFromUnicode(MBCSText1363, UPRV_LENGTHOF(MBCSText1363),
expectedIBM1363, sizeof(expectedIBM1363), "ibm-1363", TRUE, toIBM1363Offs ))
log_err("u-> ibm-1363(MBCS) with FallBack did not match.\n");
if(!testConvertToUnicode(expectedIBM1363, sizeof(expectedIBM1363),
- retrievedMBCSText1363, sizeof(retrievedMBCSText1363)/sizeof(retrievedMBCSText1363[0]),"ibm-1363", TRUE, fromIBM1363offs ))
+ retrievedMBCSText1363, UPRV_LENGTHOF(retrievedMBCSText1363),"ibm-1363", TRUE, fromIBM1363offs ))
log_err("ibm-1363->u(MBCS) with Fallback did not match.\n");
int32_t fromtest1Offs[] = { 0, 1, 2, 3, 3, 4,5};
/*from Unicode*/
- if(!testConvertFromUnicode(unicodeInput, sizeof(unicodeInput)/sizeof(unicodeInput[0]),
+ if(!testConvertFromUnicode(unicodeInput, UPRV_LENGTHOF(unicodeInput),
expectedtest1, sizeof(expectedtest1), "@test1", TRUE, totest1Offs ))
log_err("u-> test1(MBCS conversion with single-byte) did not match.\n");
/*to Unicode*/
if(!testConvertToUnicode(test1input, sizeof(test1input),
- expectedUnicode, sizeof(expectedUnicode)/sizeof(expectedUnicode[0]), "@test1", TRUE, fromtest1Offs ))
+ expectedUnicode, UPRV_LENGTHOF(expectedUnicode), "@test1", TRUE, fromtest1Offs ))
log_err("test1(MBCS conversion with single-byte) -> u did not match.\n");
}
int32_t fromtest3Offs[] = { 0, 1, 2, 3, 6, 6, 7, 7, 10, 13, 13, 16, 17};
/*from Unicode*/
- if(!testConvertFromUnicode(unicodeInput, sizeof(unicodeInput)/sizeof(unicodeInput[0]),
+ if(!testConvertFromUnicode(unicodeInput, UPRV_LENGTHOF(unicodeInput),
expectedtest3, sizeof(expectedtest3), "@test3", TRUE, totest3Offs ))
log_err("u-> test3(MBCS conversion with three-byte) did not match.\n");
/*to Unicode*/
if(!testConvertToUnicode(test3input, sizeof(test3input),
- expectedUnicode, sizeof(expectedUnicode)/sizeof(expectedUnicode[0]), "@test3", TRUE, fromtest3Offs ))
+ expectedUnicode, UPRV_LENGTHOF(expectedUnicode), "@test3", TRUE, fromtest3Offs ))
log_err("test3(MBCS conversion with three-byte) -> u did not match.\n");
}
0x01, 0x02, 0x03, 0x0a, 0x01, 0x02, 0x03, 0x0e, 0x01, 0x02, 0x03, 0x0d, 0x03, 0x01, 0x02, 0x03, 0x0c,};
const UChar expectedUnicode[] =
{ 0x20ac, 0x0005, 0x0006, 0x000b, 0xdbc4, 0xde34, 0xdbba, 0xdfcd,
- 0xd84d, 0xdc56, 0x000e, 0xd891, 0xdd67, 0xfffd, 0xfffd};
+ 0xd84d, 0xdc56, 0x000e, 0xd891, 0xdd67, 0x1a, 0xfffd};
int32_t fromtest4Offs[] =
{ 0, 1, 2, 3, 7, 7, 8, 8, 9, 9, 13, 17, 17, 21, 22,};
/*from Unicode*/
- if(!testConvertFromUnicode(unicodeInput, sizeof(unicodeInput)/sizeof(unicodeInput[0]),
+ if(!testConvertFromUnicode(unicodeInput, UPRV_LENGTHOF(unicodeInput),
expectedtest4, sizeof(expectedtest4), "@test4", TRUE, totest4Offs ))
log_err("u-> test4(MBCS conversion with four-byte) did not match.\n");
/*to Unicode*/
if(!testConvertToUnicode(test4input, sizeof(test4input),
- expectedUnicode, sizeof(expectedUnicode)/sizeof(expectedUnicode[0]), "@test4", TRUE, fromtest4Offs ))
+ expectedUnicode, UPRV_LENGTHOF(expectedUnicode), "@test4", TRUE, fromtest4Offs ))
log_err("test4(MBCS conversion with four-byte) -> u did not match.\n");
}
const UChar expectedUnicode[] = {0x203e, 0x2014, 0xff5c, 0x004c, 0x5f5e, 0x223c };
int32_t fromtest1Offs[] = {1, 3, 5, 8, 10, 12 };
/*from Unicode*/
- if(!testConvertFromUnicode(unicodeInput, sizeof(unicodeInput)/sizeof(unicodeInput[0]),
+ if(!testConvertFromUnicode(unicodeInput, UPRV_LENGTHOF(unicodeInput),
expectedtest1, sizeof(expectedtest1), "ibm-1371", TRUE, totest1Offs ))
log_err("u-> ibm-1371(MBCS conversion with single-byte) did not match.,\n");
/*to Unicode*/
if(!testConvertToUnicode(test1input, sizeof(test1input),
- expectedUnicode, sizeof(expectedUnicode)/sizeof(expectedUnicode[0]), "ibm-1371", TRUE, fromtest1Offs ))
+ expectedUnicode, UPRV_LENGTHOF(expectedUnicode), "ibm-1371", TRUE, fromtest1Offs ))
log_err("ibm-1371(MBCS conversion with single-byte) -> u did not match.,\n");
}
}
+#endif