ICU-12012 Replace irregular sizeof expressions with UPRV_LENGTHOF().

R=markus.icu@gmail.com

Review URL: https://codereview.appspot.com/286430043 .

X-SVN-Rev: 38378
This commit is contained in:
Fredrik Roubert 2016-02-26 11:21:55 +00:00
parent 26f9512326
commit f34de568c4
7 changed files with 158 additions and 157 deletions

View file

@ -113,7 +113,7 @@ static const ILcidPosixElement locmap_ ## id [] =
* @param _posixID the full POSIX ID for this entry.
*/
#define ILCID_POSIX_MAP(_posixID) \
{sizeof(locmap_ ## _posixID)/sizeof(ILcidPosixElement), locmap_ ## _posixID}
{UPRV_LENGTHOF(locmap_ ## _posixID), locmap_ ## _posixID}
/*
////////////////////////////////////////////

View file

@ -1,6 +1,6 @@
/*
******************************************************************************
* Copyright (C) 2003-2013, International Business Machines Corporation
* Copyright (C) 2003-2016, International Business Machines Corporation
* and others. All Rights Reserved.
******************************************************************************
*
@ -17,6 +17,7 @@
#if !UCONFIG_NO_FORMATTING
#include "cmemory.h"
#include "umutex.h"
#include <float.h>
#include "gregoimp.h" // Math
@ -580,7 +581,7 @@ void HebrewCalendar::handleComputeFields(int32_t julianDay, UErrorCode &status)
UBool isLeap = isLeapYear(year);
int32_t month = 0;
int32_t momax = sizeof(MONTH_START) / (3 * sizeof(MONTH_START[0][0]));
int32_t momax = UPRV_LENGTHOF(MONTH_START);
while (month < momax && dayOfYear > ( isLeap ? LEAP_MONTH_START[month][type] : MONTH_START[month][type] ) ) {
month++;
}

View file

@ -184,11 +184,11 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
static const int32_t toIBM943Offsskip [] = { 0, 0, 1, 1, 3, 3 };
if(!testConvertFromUnicode(sampleText, UPRV_LENGTHOF(sampleText),
expskipIBM_949, sizeof(expskipIBM_949), "ibm-949",
expskipIBM_949, UPRV_LENGTHOF(expskipIBM_949), "ibm-949",
UCNV_FROM_U_CALLBACK_SKIP, toIBM949Offsskip, NULL, 0 ))
log_err("u-> ibm-949 with skip did not match.\n");
if(!testConvertFromUnicode(sampleText2, UPRV_LENGTHOF(sampleText2),
expskipIBM_943, sizeof(expskipIBM_943), "ibm-943",
expskipIBM_943, UPRV_LENGTHOF(expskipIBM_943), "ibm-943",
UCNV_FROM_U_CALLBACK_SKIP, toIBM943Offsskip, NULL, 0 ))
log_err("u-> ibm-943 with skip did not match.\n");
}
@ -200,7 +200,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
/* test ibm-930 (EBCDIC_STATEFUL) with fallbacks that are not taken to check correct state transitions */
if(!testConvertFromUnicode(fromU, UPRV_LENGTHOF(fromU),
fromUBytes, sizeof(fromUBytes),
fromUBytes, UPRV_LENGTHOF(fromUBytes),
"ibm-930",
UCNV_FROM_U_CALLBACK_SKIP, fromUOffsets,
NULL, 0)
@ -221,7 +221,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
/* US-ASCII */
if(!testConvertFromUnicode(usasciiFromU, UPRV_LENGTHOF(usasciiFromU),
usasciiFromUBytes, sizeof(usasciiFromUBytes),
usasciiFromUBytes, UPRV_LENGTHOF(usasciiFromUBytes),
"US-ASCII",
UCNV_FROM_U_CALLBACK_SKIP, usasciiFromUOffsets,
NULL, 0)
@ -232,7 +232,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
#if !UCONFIG_NO_LEGACY_CONVERSION
/* SBCS NLTC codepage 367 for US-ASCII */
if(!testConvertFromUnicode(usasciiFromU, UPRV_LENGTHOF(usasciiFromU),
usasciiFromUBytes, sizeof(usasciiFromUBytes),
usasciiFromUBytes, UPRV_LENGTHOF(usasciiFromUBytes),
"ibm-367",
UCNV_FROM_U_CALLBACK_SKIP, usasciiFromUOffsets,
NULL, 0)
@ -243,7 +243,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
/* ISO-Latin-1 */
if(!testConvertFromUnicode(latin1FromU, UPRV_LENGTHOF(latin1FromU),
latin1FromUBytes, sizeof(latin1FromUBytes),
latin1FromUBytes, UPRV_LENGTHOF(latin1FromUBytes),
"LATIN_1",
UCNV_FROM_U_CALLBACK_SKIP, latin1FromUOffsets,
NULL, 0)
@ -254,7 +254,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
#if !UCONFIG_NO_LEGACY_CONVERSION
/* windows-1252 */
if(!testConvertFromUnicode(latin1FromU, UPRV_LENGTHOF(latin1FromU),
latin1FromUBytes, sizeof(latin1FromUBytes),
latin1FromUBytes, UPRV_LENGTHOF(latin1FromUBytes),
"windows-1252",
UCNV_FROM_U_CALLBACK_SKIP, latin1FromUOffsets,
NULL, 0)
@ -421,81 +421,81 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
static const int32_t from_isciiOffs1 [] ={0,2};
if(!testConvertFromUnicode(inputTest, UPRV_LENGTHOF(inputTest),
toIBM943, sizeof(toIBM943), "ibm-943",
toIBM943, UPRV_LENGTHOF(toIBM943), "ibm-943",
UCNV_FROM_U_CALLBACK_SKIP, offset, NULL, 0 ))
log_err("u-> ibm-943 with skip did not match.\n");
if(!testConvertFromUnicode(euc_jp_inputText, UPRV_LENGTHOF(euc_jp_inputText),
to_euc_jp, sizeof(to_euc_jp), "IBM-eucJP",
to_euc_jp, UPRV_LENGTHOF(to_euc_jp), "IBM-eucJP",
UCNV_FROM_U_CALLBACK_SKIP, fromEUC_JPOffs, NULL, 0 ))
log_err("u-> euc-jp with skip did not match.\n");
if(!testConvertFromUnicode(euc_tw_inputText, UPRV_LENGTHOF(euc_tw_inputText),
to_euc_tw, sizeof(to_euc_tw), "euc-tw",
to_euc_tw, UPRV_LENGTHOF(to_euc_tw), "euc-tw",
UCNV_FROM_U_CALLBACK_SKIP, from_euc_twOffs, NULL, 0 ))
log_err("u-> euc-tw with skip did not match.\n");
/*iso_2022_jp*/
if(!testConvertFromUnicode(iso_2022_jp_inputText, UPRV_LENGTHOF(iso_2022_jp_inputText),
to_iso_2022_jp, sizeof(to_iso_2022_jp), "iso-2022-jp",
to_iso_2022_jp, UPRV_LENGTHOF(to_iso_2022_jp), "iso-2022-jp",
UCNV_FROM_U_CALLBACK_SKIP, from_iso_2022_jpOffs, NULL, 0 ))
log_err("u-> iso-2022-jp with skip did not match.\n");
/* with context */
if(!testConvertFromUnicodeWithContext(iso_2022_jp_inputText2, UPRV_LENGTHOF(iso_2022_jp_inputText2),
to_iso_2022_jp2, sizeof(to_iso_2022_jp2), "iso-2022-jp",
to_iso_2022_jp2, UPRV_LENGTHOF(to_iso_2022_jp2), "iso-2022-jp",
UCNV_FROM_U_CALLBACK_SKIP, from_iso_2022_jpOffs2, NULL, 0,UCNV_SKIP_STOP_ON_ILLEGAL,U_ILLEGAL_CHAR_FOUND ))
log_err("u-> iso-2022-jp with skip & UCNV_SKIP_STOP_ON_ILLEGAL did not match.\n");
/*iso_2022_cn*/
if(!testConvertFromUnicode(iso_2022_cn_inputText, UPRV_LENGTHOF(iso_2022_cn_inputText),
to_iso_2022_cn, sizeof(to_iso_2022_cn), "iso-2022-cn",
to_iso_2022_cn, UPRV_LENGTHOF(to_iso_2022_cn), "iso-2022-cn",
UCNV_FROM_U_CALLBACK_SKIP, from_iso_2022_cnOffs, NULL, 0 ))
log_err("u-> iso-2022-cn with skip did not match.\n");
/*with context*/
if(!testConvertFromUnicodeWithContext(iso_2022_cn_inputText1, UPRV_LENGTHOF(iso_2022_cn_inputText1),
to_iso_2022_cn1, sizeof(to_iso_2022_cn1), "iso-2022-cn",
to_iso_2022_cn1, UPRV_LENGTHOF(to_iso_2022_cn1), "iso-2022-cn",
UCNV_FROM_U_CALLBACK_SKIP, from_iso_2022_cnOffs1, NULL, 0,UCNV_SKIP_STOP_ON_ILLEGAL,U_ILLEGAL_CHAR_FOUND ))
log_err("u-> iso-2022-cn with skip & UCNV_SKIP_STOP_ON_ILLEGAL did not match.\n");
/*iso_2022_kr*/
if(!testConvertFromUnicode(iso_2022_kr_inputText, UPRV_LENGTHOF(iso_2022_kr_inputText),
to_iso_2022_kr, sizeof(to_iso_2022_kr), "iso-2022-kr",
to_iso_2022_kr, UPRV_LENGTHOF(to_iso_2022_kr), "iso-2022-kr",
UCNV_FROM_U_CALLBACK_SKIP, from_iso_2022_krOffs, NULL, 0 ))
log_err("u-> iso-2022-kr with skip did not match.\n");
/*with context*/
if(!testConvertFromUnicodeWithContext(iso_2022_kr_inputText1, UPRV_LENGTHOF(iso_2022_kr_inputText1),
to_iso_2022_kr1, sizeof(to_iso_2022_kr1), "iso-2022-kr",
to_iso_2022_kr1, UPRV_LENGTHOF(to_iso_2022_kr1), "iso-2022-kr",
UCNV_FROM_U_CALLBACK_SKIP, from_iso_2022_krOffs1, NULL, 0,UCNV_SKIP_STOP_ON_ILLEGAL,U_ILLEGAL_CHAR_FOUND ))
log_err("u-> iso-2022-kr with skip & UCNV_SKIP_STOP_ON_ILLEGAL did not match.\n");
/*hz*/
if(!testConvertFromUnicode(hz_inputText, UPRV_LENGTHOF(hz_inputText),
to_hz, sizeof(to_hz), "HZ",
to_hz, UPRV_LENGTHOF(to_hz), "HZ",
UCNV_FROM_U_CALLBACK_SKIP, from_hzOffs, NULL, 0 ))
log_err("u-> HZ with skip did not match.\n");
/*with context*/
if(!testConvertFromUnicodeWithContext(hz_inputText1, UPRV_LENGTHOF(hz_inputText1),
to_hz1, sizeof(to_hz1), "hz",
to_hz1, UPRV_LENGTHOF(to_hz1), "hz",
UCNV_FROM_U_CALLBACK_SKIP, from_hzOffs1, NULL, 0,UCNV_SKIP_STOP_ON_ILLEGAL,U_ILLEGAL_CHAR_FOUND ))
log_err("u-> hz with skip & UCNV_SKIP_STOP_ON_ILLEGAL did not match.\n");
#endif
/*SCSU*/
if(!testConvertFromUnicode(SCSU_inputText, UPRV_LENGTHOF(SCSU_inputText),
to_SCSU, sizeof(to_SCSU), "SCSU",
to_SCSU, UPRV_LENGTHOF(to_SCSU), "SCSU",
UCNV_FROM_U_CALLBACK_SKIP, from_SCSUOffs, NULL, 0 ))
log_err("u-> SCSU with skip did not match.\n");
#if !UCONFIG_NO_LEGACY_CONVERSION
/*ISCII*/
if(!testConvertFromUnicode(iscii_inputText, UPRV_LENGTHOF(iscii_inputText),
to_iscii, sizeof(to_iscii), "ISCII,version=0",
to_iscii, UPRV_LENGTHOF(to_iscii), "ISCII,version=0",
UCNV_FROM_U_CALLBACK_SKIP, from_isciiOffs, NULL, 0 ))
log_err("u-> iscii with skip did not match.\n");
/*with context*/
if(!testConvertFromUnicodeWithContext(iscii_inputText1, UPRV_LENGTHOF(iscii_inputText1),
to_iscii1, sizeof(to_iscii1), "ISCII,version=0",
to_iscii1, UPRV_LENGTHOF(to_iscii1), "ISCII,version=0",
UCNV_FROM_U_CALLBACK_SKIP, from_isciiOffs1, NULL, 0,UCNV_SKIP_STOP_ON_ILLEGAL,U_ILLEGAL_CHAR_FOUND ))
log_err("u-> iscii with skip & UCNV_SKIP_STOP_ON_ILLEGAL did not match.\n");
#endif
@ -581,7 +581,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
/* BOCU-1 fromUnicode never calls callbacks, so this only tests single-byte and offsets behavior */
if(!testConvertFromUnicode(expected, UPRV_LENGTHOF(expected),
sampleText, sizeof(sampleText),
sampleText, UPRV_LENGTHOF(sampleText),
"BOCU-1",
UCNV_FROM_U_CALLBACK_SKIP, offsets, NULL, 0)
) {
@ -629,7 +629,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
/* without offsets */
if(!testConvertFromUnicode(expected, UPRV_LENGTHOF(expected),
sampleText, sizeof(sampleText),
sampleText, UPRV_LENGTHOF(sampleText),
"CESU-8",
UCNV_FROM_U_CALLBACK_SKIP, NULL, NULL, 0)
) {
@ -638,7 +638,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
/* with offsets */
if(!testConvertFromUnicode(expected, UPRV_LENGTHOF(expected),
sampleText, sizeof(sampleText),
sampleText, UPRV_LENGTHOF(sampleText),
"CESU-8",
UCNV_FROM_U_CALLBACK_SKIP, offsets, NULL, 0)
) {
@ -660,23 +660,23 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
static const int32_t fromIBM943Offs [] = { 0, 2, 4};
static const int32_t fromIBM930Offs [] = { 1, 3, 5};
if(!testConvertToUnicode(expskipIBM_949, sizeof(expskipIBM_949),
IBM_949skiptoUnicode, sizeof(IBM_949skiptoUnicode)/sizeof(IBM_949skiptoUnicode),"ibm-949",
if(!testConvertToUnicode(expskipIBM_949, UPRV_LENGTHOF(expskipIBM_949),
IBM_949skiptoUnicode, UPRV_LENGTHOF(IBM_949skiptoUnicode),"ibm-949",
UCNV_TO_U_CALLBACK_SKIP, fromIBM949Offs, NULL, 0 ))
log_err("ibm-949->u with skip did not match.\n");
if(!testConvertToUnicode(expskipIBM_943, sizeof(expskipIBM_943),
if(!testConvertToUnicode(expskipIBM_943, UPRV_LENGTHOF(expskipIBM_943),
IBM_943skiptoUnicode, UPRV_LENGTHOF(IBM_943skiptoUnicode),"ibm-943",
UCNV_TO_U_CALLBACK_SKIP, fromIBM943Offs, NULL, 0 ))
log_err("ibm-943->u with skip did not match.\n");
if(!testConvertToUnicode(expskipIBM_930, sizeof(expskipIBM_930),
if(!testConvertToUnicode(expskipIBM_930, UPRV_LENGTHOF(expskipIBM_930),
IBM_930skiptoUnicode, UPRV_LENGTHOF(IBM_930skiptoUnicode),"ibm-930",
UCNV_TO_U_CALLBACK_SKIP, fromIBM930Offs, NULL, 0 ))
log_err("ibm-930->u with skip did not match.\n");
if(!testConvertToUnicodeWithContext(expskipIBM_930, sizeof(expskipIBM_930),
if(!testConvertToUnicodeWithContext(expskipIBM_930, UPRV_LENGTHOF(expskipIBM_930),
IBM_930skiptoUnicode, UPRV_LENGTHOF(IBM_930skiptoUnicode),"ibm-930",
UCNV_TO_U_CALLBACK_SKIP, fromIBM930Offs, NULL, 0,"i",U_ILLEGAL_CHAR_FOUND ))
log_err("ibm-930->u with skip did not match.\n");
@ -693,7 +693,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
static const int32_t latin1ToUOffsets[] = { 0, 1, 2 };
/* US-ASCII */
if(!testConvertToUnicode(usasciiToUBytes, sizeof(usasciiToUBytes),
if(!testConvertToUnicode(usasciiToUBytes, UPRV_LENGTHOF(usasciiToUBytes),
usasciiToU, UPRV_LENGTHOF(usasciiToU),
"US-ASCII",
UCNV_TO_U_CALLBACK_SKIP, usasciiToUOffsets,
@ -704,7 +704,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
#if !UCONFIG_NO_LEGACY_CONVERSION
/* SBCS NLTC codepage 367 for US-ASCII */
if(!testConvertToUnicode(usasciiToUBytes, sizeof(usasciiToUBytes),
if(!testConvertToUnicode(usasciiToUBytes, UPRV_LENGTHOF(usasciiToUBytes),
usasciiToU, UPRV_LENGTHOF(usasciiToU),
"ibm-367",
UCNV_TO_U_CALLBACK_SKIP, usasciiToUOffsets,
@ -715,7 +715,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
#endif
/* ISO-Latin-1 */
if(!testConvertToUnicode(latin1ToUBytes, sizeof(latin1ToUBytes),
if(!testConvertToUnicode(latin1ToUBytes, UPRV_LENGTHOF(latin1ToUBytes),
latin1ToU, UPRV_LENGTHOF(latin1ToU),
"LATIN_1",
UCNV_TO_U_CALLBACK_SKIP, latin1ToUOffsets,
@ -726,7 +726,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
#if !UCONFIG_NO_LEGACY_CONVERSION
/* windows-1252 */
if(!testConvertToUnicode(latin1ToUBytes, sizeof(latin1ToUBytes),
if(!testConvertToUnicode(latin1ToUBytes, UPRV_LENGTHOF(latin1ToUBytes),
latin1ToU, UPRV_LENGTHOF(latin1ToU),
"windows-1252",
UCNV_TO_U_CALLBACK_SKIP, latin1ToUOffsets,
@ -846,55 +846,55 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
static const UChar LMBCSToUnicode[]={ 0x4e2e, 0xe5c4};
static const int32_t fromLMBCS[] = {0, 6};
if(!testConvertToUnicode(sampleTxtEBCIDIC_STATEFUL, sizeof(sampleTxtEBCIDIC_STATEFUL),
if(!testConvertToUnicode(sampleTxtEBCIDIC_STATEFUL, UPRV_LENGTHOF(sampleTxtEBCIDIC_STATEFUL),
EBCIDIC_STATEFUL_toUnicode, UPRV_LENGTHOF(EBCIDIC_STATEFUL_toUnicode),"ibm-930",
UCNV_TO_U_CALLBACK_SKIP, from_EBCIDIC_STATEFULOffsets, NULL, 0 ))
log_err("EBCIDIC_STATEFUL->u with skip did not match.\n");
if(!testConvertToUnicodeWithContext(sampleTxtEBCIDIC_STATEFUL, sizeof(sampleTxtEBCIDIC_STATEFUL),
if(!testConvertToUnicodeWithContext(sampleTxtEBCIDIC_STATEFUL, UPRV_LENGTHOF(sampleTxtEBCIDIC_STATEFUL),
EBCIDIC_STATEFUL_toUnicode, UPRV_LENGTHOF(EBCIDIC_STATEFUL_toUnicode),"ibm-930",
UCNV_TO_U_CALLBACK_SKIP, from_EBCIDIC_STATEFULOffsets, NULL, 0,"i",U_ILLEGAL_CHAR_FOUND ))
log_err("EBCIDIC_STATEFUL->u with skip did not match.\n");
if(!testConvertToUnicode(sampleTxt_euc_jp, sizeof(sampleTxt_euc_jp),
if(!testConvertToUnicode(sampleTxt_euc_jp, UPRV_LENGTHOF(sampleTxt_euc_jp),
euc_jptoUnicode, UPRV_LENGTHOF(euc_jptoUnicode),"IBM-eucJP",
UCNV_TO_U_CALLBACK_SKIP, from_euc_jpOffs , NULL, 0))
log_err("euc-jp->u with skip did not match.\n");
if(!testConvertToUnicode(sampleTxt_euc_tw, sizeof(sampleTxt_euc_tw),
if(!testConvertToUnicode(sampleTxt_euc_tw, UPRV_LENGTHOF(sampleTxt_euc_tw),
euc_twtoUnicode, UPRV_LENGTHOF(euc_twtoUnicode),"euc-tw",
UCNV_TO_U_CALLBACK_SKIP, from_euc_twOffs , NULL, 0))
log_err("euc-tw->u with skip did not match.\n");
if(!testConvertToUnicode(sampleTxt_iso_2022_jp, sizeof(sampleTxt_iso_2022_jp),
if(!testConvertToUnicode(sampleTxt_iso_2022_jp, UPRV_LENGTHOF(sampleTxt_iso_2022_jp),
iso_2022_jptoUnicode, UPRV_LENGTHOF(iso_2022_jptoUnicode),"iso-2022-jp",
UCNV_TO_U_CALLBACK_SKIP, from_iso_2022_jpOffs , NULL, 0))
log_err("iso-2022-jp->u with skip did not match.\n");
if(!testConvertToUnicode(sampleTxt_iso_2022_cn, sizeof(sampleTxt_iso_2022_cn),
if(!testConvertToUnicode(sampleTxt_iso_2022_cn, UPRV_LENGTHOF(sampleTxt_iso_2022_cn),
iso_2022_cntoUnicode, UPRV_LENGTHOF(iso_2022_cntoUnicode),"iso-2022-cn",
UCNV_TO_U_CALLBACK_SKIP, from_iso_2022_cnOffs , NULL, 0))
log_err("iso-2022-cn->u with skip did not match.\n");
if(!testConvertToUnicode(sampleTxt_iso_2022_kr, sizeof(sampleTxt_iso_2022_kr),
if(!testConvertToUnicode(sampleTxt_iso_2022_kr, UPRV_LENGTHOF(sampleTxt_iso_2022_kr),
iso_2022_krtoUnicode, UPRV_LENGTHOF(iso_2022_krtoUnicode),"iso-2022-kr",
UCNV_TO_U_CALLBACK_SKIP, from_iso_2022_krOffs , NULL, 0))
log_err("iso-2022-kr->u with skip did not match.\n");
if(!testConvertToUnicode(sampleTxt_hz, sizeof(sampleTxt_hz),
if(!testConvertToUnicode(sampleTxt_hz, UPRV_LENGTHOF(sampleTxt_hz),
hztoUnicode, UPRV_LENGTHOF(hztoUnicode),"HZ",
UCNV_TO_U_CALLBACK_SKIP, from_hzOffs , NULL, 0))
log_err("HZ->u with skip did not match.\n");
if(!testConvertToUnicode(sampleTxt_iscii, sizeof(sampleTxt_iscii),
if(!testConvertToUnicode(sampleTxt_iscii, UPRV_LENGTHOF(sampleTxt_iscii),
isciitoUnicode, UPRV_LENGTHOF(isciitoUnicode),"ISCII,version=0",
UCNV_TO_U_CALLBACK_SKIP, from_isciiOffs , NULL, 0))
log_err("iscii->u with skip did not match.\n");
if(!testConvertToUnicode(sampleTxtLMBCS, sizeof(sampleTxtLMBCS),
if(!testConvertToUnicode(sampleTxtLMBCS, UPRV_LENGTHOF(sampleTxtLMBCS),
LMBCSToUnicode, UPRV_LENGTHOF(LMBCSToUnicode),"LMBCS-1",
UCNV_TO_U_CALLBACK_SKIP, fromLMBCS , NULL, 0))
log_err("LMBCS->u with skip did not match.\n");
@ -909,7 +909,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
UChar expected1[] = { 0x0031, 0x4e8c, 0x0061};
int32_t offsets1[] = { 0x0000, 0x0001, 0x0006};
if(!testConvertToUnicode(sampleText1, sizeof(sampleText1),
if(!testConvertToUnicode(sampleText1, UPRV_LENGTHOF(sampleText1),
expected1, UPRV_LENGTHOF(expected1),"utf8",
UCNV_TO_U_CALLBACK_SKIP, offsets1, NULL, 0 ))
log_err("utf8->u with skip did not match.\n");;
@ -921,7 +921,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
UChar expected1[] = { 0x00ba, 0x008c, 0x00f8, 0x0061,0xfffe,0xfffe};
int32_t offsets1[] = { 0x0000, 0x0001,0x0002,0x0003,4,5};
if(!testConvertToUnicode(sampleText1, sizeof(sampleText1),
if(!testConvertToUnicode(sampleText1, UPRV_LENGTHOF(sampleText1),
expected1, UPRV_LENGTHOF(expected1),"SCSU",
UCNV_TO_U_CALLBACK_SKIP, offsets1, NULL, 0 ))
log_err("scsu->u with skip did not match.\n");
@ -1002,7 +1002,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
71, 72
};
if(!testConvertToUnicode(sampleText, sizeof(sampleText),
if(!testConvertToUnicode(sampleText, UPRV_LENGTHOF(sampleText),
expected, UPRV_LENGTHOF(expected), "BOCU-1",
UCNV_TO_U_CALLBACK_SKIP, offsets, NULL, 0)
) {
@ -1055,7 +1055,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
};
/* without offsets */
if(!testConvertToUnicode(sampleText, sizeof(sampleText),
if(!testConvertToUnicode(sampleText, UPRV_LENGTHOF(sampleText),
expected, UPRV_LENGTHOF(expected), "CESU-8",
UCNV_TO_U_CALLBACK_SKIP, NULL, NULL, 0)
) {
@ -1063,7 +1063,7 @@ static void TestSkip(int32_t inputsize, int32_t outputsize)
}
/* with offsets */
if(!testConvertToUnicode(sampleText, sizeof(sampleText),
if(!testConvertToUnicode(sampleText, UPRV_LENGTHOF(sampleText),
expected, UPRV_LENGTHOF(expected), "CESU-8",
UCNV_TO_U_CALLBACK_SKIP, offsets, NULL, 0)
) {
@ -1106,15 +1106,15 @@ static void TestStop(int32_t inputsize, int32_t outputsize)
#if !UCONFIG_NO_LEGACY_CONVERSION
if(!testConvertFromUnicode(sampleText, UPRV_LENGTHOF(sampleText),
expstopIBM_949, sizeof(expstopIBM_949), "ibm-949",
expstopIBM_949, UPRV_LENGTHOF(expstopIBM_949), "ibm-949",
UCNV_FROM_U_CALLBACK_STOP, toIBM949Offsstop, NULL, 0 ))
log_err("u-> ibm-949 with stop did not match.\n");
if(!testConvertFromUnicode(sampleText2, UPRV_LENGTHOF(sampleText2),
expstopIBM_943, sizeof(expstopIBM_943), "ibm-943",
expstopIBM_943, UPRV_LENGTHOF(expstopIBM_943), "ibm-943",
UCNV_FROM_U_CALLBACK_STOP, toIBM943Offsstop , NULL, 0))
log_err("u-> ibm-943 with stop did not match.\n");
if(!testConvertFromUnicode(sampleText2, UPRV_LENGTHOF(sampleText2),
expstopIBM_930, sizeof(expstopIBM_930), "ibm-930",
expstopIBM_930, UPRV_LENGTHOF(expstopIBM_930), "ibm-930",
UCNV_FROM_U_CALLBACK_STOP, toIBM930Offsstop, NULL, 0 ))
log_err("u-> ibm-930 with stop did not match.\n");
@ -1190,47 +1190,47 @@ static void TestStop(int32_t inputsize, int32_t outputsize)
};
if(!testConvertFromUnicode(inputTest, UPRV_LENGTHOF(inputTest),
toIBM943, sizeof(toIBM943), "ibm-943",
toIBM943, UPRV_LENGTHOF(toIBM943), "ibm-943",
UCNV_FROM_U_CALLBACK_STOP, offset, NULL, 0 ))
log_err("u-> ibm-943 with stop did not match.\n");
if(!testConvertFromUnicode(euc_jp_inputText, UPRV_LENGTHOF(euc_jp_inputText),
to_euc_jp, sizeof(to_euc_jp), "IBM-eucJP",
to_euc_jp, UPRV_LENGTHOF(to_euc_jp), "IBM-eucJP",
UCNV_FROM_U_CALLBACK_STOP, fromEUC_JPOffs, NULL, 0 ))
log_err("u-> euc-jp with stop did not match.\n");
if(!testConvertFromUnicode(euc_tw_inputText, UPRV_LENGTHOF(euc_tw_inputText),
to_euc_tw, sizeof(to_euc_tw), "euc-tw",
to_euc_tw, UPRV_LENGTHOF(to_euc_tw), "euc-tw",
UCNV_FROM_U_CALLBACK_STOP, from_euc_twOffs, NULL, 0 ))
log_err("u-> euc-tw with stop did not match.\n");
if(!testConvertFromUnicode(iso_2022_jp_inputText, UPRV_LENGTHOF(iso_2022_jp_inputText),
to_iso_2022_jp, sizeof(to_iso_2022_jp), "iso-2022-jp",
to_iso_2022_jp, UPRV_LENGTHOF(to_iso_2022_jp), "iso-2022-jp",
UCNV_FROM_U_CALLBACK_STOP, from_iso_2022_jpOffs, NULL, 0 ))
log_err("u-> iso-2022-jp with stop did not match.\n");
if(!testConvertFromUnicode(iso_2022_jp_inputText, UPRV_LENGTHOF(iso_2022_jp_inputText),
to_iso_2022_jp, sizeof(to_iso_2022_jp), "iso-2022-jp",
to_iso_2022_jp, UPRV_LENGTHOF(to_iso_2022_jp), "iso-2022-jp",
UCNV_FROM_U_CALLBACK_STOP, from_iso_2022_jpOffs, NULL, 0 ))
log_err("u-> iso-2022-jp with stop did not match.\n");
if(!testConvertFromUnicode(iso_2022_cn_inputText, UPRV_LENGTHOF(iso_2022_cn_inputText),
to_iso_2022_cn, sizeof(to_iso_2022_cn), "iso-2022-cn",
to_iso_2022_cn, UPRV_LENGTHOF(to_iso_2022_cn), "iso-2022-cn",
UCNV_FROM_U_CALLBACK_STOP, from_iso_2022_cnOffs, NULL, 0 ))
log_err("u-> iso-2022-cn with stop did not match.\n");
if(!testConvertFromUnicode(iso_2022_kr_inputText, UPRV_LENGTHOF(iso_2022_kr_inputText),
to_iso_2022_kr, sizeof(to_iso_2022_kr), "iso-2022-kr",
to_iso_2022_kr, UPRV_LENGTHOF(to_iso_2022_kr), "iso-2022-kr",
UCNV_FROM_U_CALLBACK_STOP, from_iso_2022_krOffs, NULL, 0 ))
log_err("u-> iso-2022-kr with stop did not match.\n");
if(!testConvertFromUnicode(hz_inputText, UPRV_LENGTHOF(hz_inputText),
to_hz, sizeof(to_hz), "HZ",
to_hz, UPRV_LENGTHOF(to_hz), "HZ",
UCNV_FROM_U_CALLBACK_STOP, from_hzOffs, NULL, 0 ))
log_err("u-> HZ with stop did not match.\n");\
if(!testConvertFromUnicode(iscii_inputText, UPRV_LENGTHOF(iscii_inputText),
to_iscii, sizeof(to_iscii), "ISCII,version=0",
to_iscii, UPRV_LENGTHOF(to_iscii), "ISCII,version=0",
UCNV_FROM_U_CALLBACK_STOP, from_isciiOffs, NULL, 0 ))
log_err("u-> iscii with stop did not match.\n");
@ -1251,7 +1251,7 @@ static void TestStop(int32_t inputsize, int32_t outputsize)
};
if(!testConvertFromUnicode(SCSU_inputText, UPRV_LENGTHOF(SCSU_inputText),
to_SCSU, sizeof(to_SCSU), "SCSU",
to_SCSU, UPRV_LENGTHOF(to_SCSU), "SCSU",
UCNV_FROM_U_CALLBACK_STOP, from_SCSUOffs, NULL, 0 ))
log_err("u-> SCSU with skip did not match.\n");
@ -1260,15 +1260,15 @@ static void TestStop(int32_t inputsize, int32_t outputsize)
/*to Unicode*/
#if !UCONFIG_NO_LEGACY_CONVERSION
if(!testConvertToUnicode(expstopIBM_949, sizeof(expstopIBM_949),
if(!testConvertToUnicode(expstopIBM_949, UPRV_LENGTHOF(expstopIBM_949),
IBM_949stoptoUnicode, UPRV_LENGTHOF(IBM_949stoptoUnicode),"ibm-949",
UCNV_TO_U_CALLBACK_STOP, fromIBM949Offs, NULL, 0 ))
log_err("ibm-949->u with stop did not match.\n");
if(!testConvertToUnicode(expstopIBM_943, sizeof(expstopIBM_943),
if(!testConvertToUnicode(expstopIBM_943, UPRV_LENGTHOF(expstopIBM_943),
IBM_943stoptoUnicode, UPRV_LENGTHOF(IBM_943stoptoUnicode),"ibm-943",
UCNV_TO_U_CALLBACK_STOP, fromIBM943Offs, NULL, 0 ))
log_err("ibm-943->u with stop did not match.\n");
if(!testConvertToUnicode(expstopIBM_930, sizeof(expstopIBM_930),
if(!testConvertToUnicode(expstopIBM_930, UPRV_LENGTHOF(expstopIBM_930),
IBM_930stoptoUnicode, UPRV_LENGTHOF(IBM_930stoptoUnicode),"ibm-930",
UCNV_TO_U_CALLBACK_STOP, fromIBM930Offs, NULL, 0 ))
log_err("ibm-930->u with stop did not match.\n");
@ -1301,17 +1301,17 @@ static void TestStop(int32_t inputsize, int32_t outputsize)
if(!testConvertToUnicode(sampleTxtEBCIDIC_STATEFUL, sizeof(sampleTxtEBCIDIC_STATEFUL),
if(!testConvertToUnicode(sampleTxtEBCIDIC_STATEFUL, UPRV_LENGTHOF(sampleTxtEBCIDIC_STATEFUL),
EBCIDIC_STATEFUL_toUnicode, UPRV_LENGTHOF(EBCIDIC_STATEFUL_toUnicode),"ibm-930",
UCNV_TO_U_CALLBACK_STOP, from_EBCIDIC_STATEFULOffsets, NULL, 0 ))
log_err("EBCIDIC_STATEFUL->u with stop did not match.\n");
if(!testConvertToUnicode(sampleTxt_euc_jp, sizeof(sampleTxt_euc_jp),
if(!testConvertToUnicode(sampleTxt_euc_jp, UPRV_LENGTHOF(sampleTxt_euc_jp),
euc_jptoUnicode, UPRV_LENGTHOF(euc_jptoUnicode),"IBM-eucJP",
UCNV_TO_U_CALLBACK_STOP, from_euc_jpOffs , NULL, 0))
log_err("euc-jp->u with stop did not match.\n");
if(!testConvertToUnicode(sampleTxt_euc_tw, sizeof(sampleTxt_euc_tw),
if(!testConvertToUnicode(sampleTxt_euc_tw, UPRV_LENGTHOF(sampleTxt_euc_tw),
euc_twtoUnicode, UPRV_LENGTHOF(euc_twtoUnicode),"euc-tw",
UCNV_TO_U_CALLBACK_STOP, from_euc_twOffs, NULL, 0 ))
log_err("euc-tw->u with stop did not match.\n");
@ -1325,7 +1325,7 @@ static void TestStop(int32_t inputsize, int32_t outputsize)
static const UChar expected1[] = { 0x0031, 0x4e8c,};
static const int32_t offsets1[] = { 0x0000, 0x0001};
if(!testConvertToUnicode(sampleText1, sizeof(sampleText1),
if(!testConvertToUnicode(sampleText1, UPRV_LENGTHOF(sampleText1),
expected1, UPRV_LENGTHOF(expected1),"utf8",
UCNV_TO_U_CALLBACK_STOP, offsets1, NULL, 0 ))
log_err("utf8->u with stop did not match.\n");;
@ -1336,7 +1336,7 @@ static void TestStop(int32_t inputsize, int32_t outputsize)
static const UChar expected1[] = { 0x00ba, 0x008c, 0x00f8, 0x0061};
static const int32_t offsets1[] = { 0x0000, 0x0001,0x0002,0x0003};
if(!testConvertToUnicode(sampleText1, sizeof(sampleText1),
if(!testConvertToUnicode(sampleText1, UPRV_LENGTHOF(sampleText1),
expected1, UPRV_LENGTHOF(expected1),"SCSU",
UCNV_TO_U_CALLBACK_STOP, offsets1, NULL, 0 ))
log_err("scsu->u with stop did not match.\n");;
@ -1377,15 +1377,15 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
#if !UCONFIG_NO_LEGACY_CONVERSION
if(!testConvertFromUnicode(sampleText, UPRV_LENGTHOF(sampleText),
expsubIBM_949, sizeof(expsubIBM_949), "ibm-949",
expsubIBM_949, UPRV_LENGTHOF(expsubIBM_949), "ibm-949",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, toIBM949Offssub, NULL, 0 ))
log_err("u-> ibm-949 with subst did not match.\n");
if(!testConvertFromUnicode(sampleText2, UPRV_LENGTHOF(sampleText2),
expsubIBM_943, sizeof(expsubIBM_943), "ibm-943",
expsubIBM_943, UPRV_LENGTHOF(expsubIBM_943), "ibm-943",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, toIBM943Offssub , NULL, 0))
log_err("u-> ibm-943 with subst did not match.\n");
if(!testConvertFromUnicode(sampleText2, UPRV_LENGTHOF(sampleText2),
expsubIBM_930, sizeof(expsubIBM_930), "ibm-930",
expsubIBM_930, UPRV_LENGTHOF(expsubIBM_930), "ibm-930",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, toIBM930Offssub, NULL, 0 ))
log_err("u-> ibm-930 with subst did not match.\n");
@ -1415,17 +1415,17 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
static const int32_t from_euc_twOffs [] ={ 0, 1, 1, 2, 2, 2, 2, 3, 3, 5, 5, 6, 7, 7, 8,};
if(!testConvertFromUnicode(inputTest, UPRV_LENGTHOF(inputTest),
toIBM943, sizeof(toIBM943), "ibm-943",
toIBM943, UPRV_LENGTHOF(toIBM943), "ibm-943",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, offset, NULL, 0 ))
log_err("u-> ibm-943 with substitute did not match.\n");
if(!testConvertFromUnicode(euc_jp_inputText, UPRV_LENGTHOF(euc_jp_inputText),
to_euc_jp, sizeof(to_euc_jp), "IBM-eucJP",
to_euc_jp, UPRV_LENGTHOF(to_euc_jp), "IBM-eucJP",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, fromEUC_JPOffs, NULL, 0 ))
log_err("u-> euc-jp with substitute did not match.\n");
if(!testConvertFromUnicode(euc_tw_inputText, UPRV_LENGTHOF(euc_tw_inputText),
to_euc_tw, sizeof(to_euc_tw), "euc-tw",
to_euc_tw, UPRV_LENGTHOF(to_euc_tw), "euc-tw",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, from_euc_twOffs, NULL, 0 ))
log_err("u-> euc-tw with substitute did not match.\n");
}
@ -1457,12 +1457,12 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
};
if(!testConvertFromUnicode(SCSU_inputText, UPRV_LENGTHOF(SCSU_inputText),
to_SCSU, sizeof(to_SCSU), "SCSU",
to_SCSU, UPRV_LENGTHOF(to_SCSU), "SCSU",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, from_SCSUOffs, NULL, 0 ))
log_err("u-> SCSU with substitute did not match.\n");
if(!testConvertFromUnicodeWithContext(SCSU_inputText, UPRV_LENGTHOF(SCSU_inputText),
to_SCSU_1, sizeof(to_SCSU_1), "SCSU",
to_SCSU_1, UPRV_LENGTHOF(to_SCSU_1), "SCSU",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, from_SCSUOffs_1, NULL, 0,"i",U_ILLEGAL_CHAR_FOUND ))
log_err("u-> SCSU with substitute did not match.\n");
}
@ -1478,7 +1478,7 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
};
static const int32_t offsets[]={ 0, 0, 0, 1, 1, 1, 1, 3, 3, 3, 4, 4, 4, 5, 5, 5, 6 };
if(!testConvertFromUnicode(testinput, UPRV_LENGTHOF(testinput),
expectedUTF8, sizeof(expectedUTF8), "utf8",
expectedUTF8, UPRV_LENGTHOF(expectedUTF8), "utf8",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, offsets, NULL, 0 )) {
log_err("u-> utf8 with stop did not match.\n");
}
@ -1504,7 +1504,7 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
};
if(!testConvertFromUnicode(in, UPRV_LENGTHOF(in),
out, sizeof(out), "UTF-16",
out, UPRV_LENGTHOF(out), "UTF-16",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, offsets, NULL, 0)
) {
log_err("u->UTF-16 with substitute did not match.\n");
@ -1531,7 +1531,7 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
};
if(!testConvertFromUnicode(in, UPRV_LENGTHOF(in),
out, sizeof(out), "UTF-32",
out, UPRV_LENGTHOF(out), "UTF-32",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, offsets, NULL, 0)
) {
log_err("u->UTF-32 with substitute did not match.\n");
@ -1541,15 +1541,15 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
/*to unicode*/
#if !UCONFIG_NO_LEGACY_CONVERSION
if(!testConvertToUnicode(expsubIBM_949, sizeof(expsubIBM_949),
if(!testConvertToUnicode(expsubIBM_949, UPRV_LENGTHOF(expsubIBM_949),
IBM_949subtoUnicode, UPRV_LENGTHOF(IBM_949subtoUnicode),"ibm-949",
UCNV_TO_U_CALLBACK_SUBSTITUTE, fromIBM949Offs, NULL, 0 ))
log_err("ibm-949->u with substitute did not match.\n");
if(!testConvertToUnicode(expsubIBM_943, sizeof(expsubIBM_943),
if(!testConvertToUnicode(expsubIBM_943, UPRV_LENGTHOF(expsubIBM_943),
IBM_943subtoUnicode, UPRV_LENGTHOF(IBM_943subtoUnicode),"ibm-943",
UCNV_TO_U_CALLBACK_SUBSTITUTE, fromIBM943Offs, NULL, 0 ))
log_err("ibm-943->u with substitute did not match.\n");
if(!testConvertToUnicode(expsubIBM_930, sizeof(expsubIBM_930),
if(!testConvertToUnicode(expsubIBM_930, UPRV_LENGTHOF(expsubIBM_930),
IBM_930subtoUnicode, UPRV_LENGTHOF(IBM_930subtoUnicode),"ibm-930",
UCNV_TO_U_CALLBACK_SUBSTITUTE, fromIBM930Offs, NULL, 0 ))
log_err("ibm-930->u with substitute did not match.\n");
@ -1583,25 +1583,25 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
int32_t from_euc_twOffs [] ={ 0, 1, 3, 7, 11, 13};
if(!testConvertToUnicode(sampleTxtEBCIDIC_STATEFUL, sizeof(sampleTxtEBCIDIC_STATEFUL),
if(!testConvertToUnicode(sampleTxtEBCIDIC_STATEFUL, UPRV_LENGTHOF(sampleTxtEBCIDIC_STATEFUL),
EBCIDIC_STATEFUL_toUnicode, UPRV_LENGTHOF(EBCIDIC_STATEFUL_toUnicode),"ibm-930",
UCNV_TO_U_CALLBACK_SUBSTITUTE, from_EBCIDIC_STATEFULOffsets, NULL, 0 ))
log_err("EBCIDIC_STATEFUL->u with substitute did not match.\n");
if(!testConvertToUnicode(sampleTxt_euc_jp, sizeof(sampleTxt_euc_jp),
if(!testConvertToUnicode(sampleTxt_euc_jp, UPRV_LENGTHOF(sampleTxt_euc_jp),
euc_jptoUnicode, UPRV_LENGTHOF(euc_jptoUnicode),"IBM-eucJP",
UCNV_TO_U_CALLBACK_SUBSTITUTE, from_euc_jpOffs, NULL, 0 ))
log_err("euc-jp->u with substitute did not match.\n");
if(!testConvertToUnicode(sampleTxt_euc_tw, sizeof(sampleTxt_euc_tw),
if(!testConvertToUnicode(sampleTxt_euc_tw, UPRV_LENGTHOF(sampleTxt_euc_tw),
euc_twtoUnicode, UPRV_LENGTHOF(euc_twtoUnicode),"euc-tw",
UCNV_TO_U_CALLBACK_SUBSTITUTE, from_euc_twOffs, NULL, 0 ))
log_err("euc-tw->u with substitute did not match.\n");
if(!testConvertToUnicodeWithContext(sampleTxt_euc_jp, sizeof(sampleTxt_euc_jp),
if(!testConvertToUnicodeWithContext(sampleTxt_euc_jp, UPRV_LENGTHOF(sampleTxt_euc_jp),
euc_jptoUnicode, UPRV_LENGTHOF(euc_jptoUnicode),"IBM-eucJP",
UCNV_TO_U_CALLBACK_SUBSTITUTE, from_euc_jpOffs, NULL, 0 ,"i", U_ILLEGAL_CHAR_FOUND))
log_err("euc-jp->u with substitute did not match.\n");
@ -1615,7 +1615,7 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
UChar expected1[] = { 0x0031, 0x4e8c, 0xfffd, 0x0061};
int32_t offsets1[] = { 0x0000, 0x0001, 0x0004, 0x0006};
if(!testConvertToUnicode(sampleText1, sizeof(sampleText1),
if(!testConvertToUnicode(sampleText1, UPRV_LENGTHOF(sampleText1),
expected1, UPRV_LENGTHOF(expected1),"utf8",
UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets1, NULL, 0 ))
log_err("utf8->u with substitute did not match.\n");;
@ -1626,7 +1626,7 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
UChar expected1[] = { 0x00ba, 0x008c, 0x00f8, 0x0061,0xfffd,0xfffd};
int32_t offsets1[] = { 0x0000, 0x0001,0x0002,0x0003,4,5};
if(!testConvertToUnicode(sampleText1, sizeof(sampleText1),
if(!testConvertToUnicode(sampleText1, UPRV_LENGTHOF(sampleText1),
expected1, UPRV_LENGTHOF(expected1),"SCSU",
UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets1, NULL, 0 ))
log_err("scsu->u with stop did not match.\n");;
@ -1643,13 +1643,13 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
static const uint8_t s2[]={ 0x0e, 0x5d, 0x5f, 0x5d, 0x63, 0xfc, 0xfc, 0x46, 0x6b, 0x0f, 0x57 };
static const int32_t offsets2[]={ 1, 3, 5, 7, 10 };
if(!testConvertFromUnicode(u1, UPRV_LENGTHOF(u1), s1, UPRV_LENGTHOF(s1), "ibm-930",
if(!testConvertFromUnicode(u1, UPRV_LENGTHOF(u1), s1, UPRV_LENGTHOF(s1), "ibm-930",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, offsets1, NULL, 0)
) {
log_err("u->ibm-930 subchar/subchar1 did not match.\n");
}
if(!testConvertToUnicode(s2, UPRV_LENGTHOF(s2), u2, UPRV_LENGTHOF(u2), "ibm-930",
if(!testConvertToUnicode(s2, UPRV_LENGTHOF(s2), u2, UPRV_LENGTHOF(u2), "ibm-930",
UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets2, NULL, 0)
) {
log_err("ibm-930->u subchar/subchar1 did not match.\n");
@ -1665,7 +1665,7 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
static const int32_t offsets2[]={
0, 1, 2, 6, 8, 10, 12, 16, 20, 20, 24, 28, 28 };
if(!testConvertToUnicode(gb2, UPRV_LENGTHOF(gb2), u2, UPRV_LENGTHOF(u2), "gb18030",
if(!testConvertToUnicode(gb2, UPRV_LENGTHOF(gb2), u2, UPRV_LENGTHOF(u2), "gb18030",
UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets2, NULL, 0)
) {
log_err("gb18030->u with substitute did not match.\n");
@ -1686,7 +1686,7 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
0, 1, 2, 4, 6, 7, 9, 11, 12, 14, 17, 19, 21, 22, 23, 24
};
if(!testConvertToUnicode(utf7, UPRV_LENGTHOF(utf7), unicode, UPRV_LENGTHOF(unicode), "UTF-7",
if(!testConvertToUnicode(utf7, UPRV_LENGTHOF(utf7), unicode, UPRV_LENGTHOF(unicode), "UTF-7",
UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets, NULL, 0)
) {
log_err("UTF-7->u with substitute did not match.\n");
@ -1710,19 +1710,19 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
offsets2[]={ 2, 4 },
offsets3[]={ 0, 2, 4 };
if(!testConvertToUnicode(in1, UPRV_LENGTHOF(in1), out1, UPRV_LENGTHOF(out1), "UTF-16",
if(!testConvertToUnicode(in1, UPRV_LENGTHOF(in1), out1, UPRV_LENGTHOF(out1), "UTF-16",
UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets1, NULL, 0)
) {
log_err("UTF-16 (BE BOM)->u with substitute did not match.\n");
}
if(!testConvertToUnicode(in2, UPRV_LENGTHOF(in2), out2, UPRV_LENGTHOF(out2), "UTF-16",
if(!testConvertToUnicode(in2, UPRV_LENGTHOF(in2), out2, UPRV_LENGTHOF(out2), "UTF-16",
UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets2, NULL, 0)
) {
log_err("UTF-16 (LE BOM)->u with substitute did not match.\n");
}
if(!testConvertToUnicode(in3, UPRV_LENGTHOF(in3), out3, UPRV_LENGTHOF(out3), "UTF-16",
if(!testConvertToUnicode(in3, UPRV_LENGTHOF(in3), out3, UPRV_LENGTHOF(out3), "UTF-16",
UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets3, NULL, 0)
) {
log_err("UTF-16 (no BOM)->u with substitute did not match.\n");
@ -1749,25 +1749,25 @@ static void TestSub(int32_t inputsize, int32_t outputsize)
offsets3[]={ 0, 4, 4, 8, 12 },
offsets4[]={ 0, 0, 4, 8 };
if(!testConvertToUnicode(in1, UPRV_LENGTHOF(in1), out1, UPRV_LENGTHOF(out1), "UTF-32",
if(!testConvertToUnicode(in1, UPRV_LENGTHOF(in1), out1, UPRV_LENGTHOF(out1), "UTF-32",
UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets1, NULL, 0)
) {
log_err("UTF-32 (BE BOM)->u with substitute did not match.\n");
}
if(!testConvertToUnicode(in2, UPRV_LENGTHOF(in2), out2, UPRV_LENGTHOF(out2), "UTF-32",
if(!testConvertToUnicode(in2, UPRV_LENGTHOF(in2), out2, UPRV_LENGTHOF(out2), "UTF-32",
UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets2, NULL, 0)
) {
log_err("UTF-32 (LE BOM)->u with substitute did not match.\n");
}
if(!testConvertToUnicode(in3, UPRV_LENGTHOF(in3), out3, UPRV_LENGTHOF(out3), "UTF-32",
if(!testConvertToUnicode(in3, UPRV_LENGTHOF(in3), out3, UPRV_LENGTHOF(out3), "UTF-32",
UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets3, NULL, 0)
) {
log_err("UTF-32 (no BOM)->u with substitute did not match.\n");
}
if(!testConvertToUnicode(in4, UPRV_LENGTHOF(in4), out4, UPRV_LENGTHOF(out4), "UTF-32",
if(!testConvertToUnicode(in4, UPRV_LENGTHOF(in4), out4, UPRV_LENGTHOF(out4), "UTF-32",
UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets4, NULL, 0)
) {
log_err("UTF-32 (no BOM, with error)->u with substitute did not match.\n");
@ -1802,17 +1802,17 @@ static void TestSubWithValue(int32_t inputsize, int32_t outputsize)
#if !UCONFIG_NO_LEGACY_CONVERSION
if(!testConvertFromUnicode(sampleText, UPRV_LENGTHOF(sampleText),
expsubwvalIBM_949, sizeof(expsubwvalIBM_949), "ibm-949",
expsubwvalIBM_949, UPRV_LENGTHOF(expsubwvalIBM_949), "ibm-949",
UCNV_FROM_U_CALLBACK_ESCAPE, toIBM949Offs, NULL, 0 ))
log_err("u-> ibm-949 with subst with value did not match.\n");
if(!testConvertFromUnicode(sampleText2, UPRV_LENGTHOF(sampleText2),
expsubwvalIBM_943, sizeof(expsubwvalIBM_943), "ibm-943",
expsubwvalIBM_943, UPRV_LENGTHOF(expsubwvalIBM_943), "ibm-943",
UCNV_FROM_U_CALLBACK_ESCAPE, toIBM943Offs, NULL, 0 ))
log_err("u-> ibm-943 with sub with value did not match.\n");
if(!testConvertFromUnicode(sampleText2, UPRV_LENGTHOF(sampleText2),
expsubwvalIBM_930, sizeof(expsubwvalIBM_930), "ibm-930",
expsubwvalIBM_930, UPRV_LENGTHOF(expsubwvalIBM_930), "ibm-930",
UCNV_FROM_U_CALLBACK_ESCAPE, toIBM930Offs, NULL, 0 ))
log_err("u-> ibm-930 with subst with value did not match.\n");
@ -2046,32 +2046,32 @@ static void TestSubWithValue(int32_t inputsize, int32_t outputsize)
};
if(!testConvertFromUnicode(inputTest, UPRV_LENGTHOF(inputTest),
toIBM943, sizeof(toIBM943), "ibm-943",
toIBM943, UPRV_LENGTHOF(toIBM943), "ibm-943",
UCNV_FROM_U_CALLBACK_ESCAPE, offset, NULL, 0 ))
log_err("u-> ibm-943 with subst with value did not match.\n");
if(!testConvertFromUnicode(euc_jp_inputText, UPRV_LENGTHOF(euc_jp_inputText),
to_euc_jp, sizeof(to_euc_jp), "IBM-eucJP",
to_euc_jp, UPRV_LENGTHOF(to_euc_jp), "IBM-eucJP",
UCNV_FROM_U_CALLBACK_ESCAPE, fromEUC_JPOffs, NULL, 0 ))
log_err("u-> euc-jp with subst with value did not match.\n");
if(!testConvertFromUnicode(euc_tw_inputText, UPRV_LENGTHOF(euc_tw_inputText),
to_euc_tw, sizeof(to_euc_tw), "euc-tw",
to_euc_tw, UPRV_LENGTHOF(to_euc_tw), "euc-tw",
UCNV_FROM_U_CALLBACK_ESCAPE, from_euc_twOffs, NULL, 0 ))
log_err("u-> euc-tw with subst with value did not match.\n");
if(!testConvertFromUnicode(iso_2022_jp_inputText1, UPRV_LENGTHOF(iso_2022_jp_inputText1),
to_iso_2022_jp1, sizeof(to_iso_2022_jp1), "iso-2022-jp",
to_iso_2022_jp1, UPRV_LENGTHOF(to_iso_2022_jp1), "iso-2022-jp",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_jpOffs1, NULL, 0 ))
log_err("u-> iso_2022_jp with subst with value did not match.\n");
if(!testConvertFromUnicode(iso_2022_jp_inputText1, UPRV_LENGTHOF(iso_2022_jp_inputText1),
to_iso_2022_jp1, sizeof(to_iso_2022_jp1), "iso-2022-jp",
to_iso_2022_jp1, UPRV_LENGTHOF(to_iso_2022_jp1), "iso-2022-jp",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_jpOffs1, NULL, 0 ))
log_err("u-> iso_2022_jp with subst with value did not match.\n");
if(!testConvertFromUnicode(iso_2022_jp_inputText2, UPRV_LENGTHOF(iso_2022_jp_inputText2),
to_iso_2022_jp2, sizeof(to_iso_2022_jp2), "iso-2022-jp",
to_iso_2022_jp2, UPRV_LENGTHOF(to_iso_2022_jp2), "iso-2022-jp",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_jpOffs2, NULL, 0 ))
log_err("u-> iso_2022_jp with subst with value did not match.\n");
/*ESCAPE OPTIONS*/
@ -2101,7 +2101,7 @@ static void TestSubWithValue(int32_t inputsize, int32_t outputsize)
};
if(!testConvertFromUnicodeWithContext(iso_2022_jp_inputText3, UPRV_LENGTHOF(iso_2022_jp_inputText3),
to_iso_2022_jp3_v2, sizeof(to_iso_2022_jp3_v2), "iso-2022-jp",
to_iso_2022_jp3_v2, UPRV_LENGTHOF(to_iso_2022_jp3_v2), "iso-2022-jp",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_jpOffs3_v2, NULL, 0,UCNV_ESCAPE_XML_DEC,U_ZERO_ERROR ))
log_err("u-> iso-2022-jp with sub & UCNV_ESCAPE_XML_DEC did not match.\n");
}
@ -2128,7 +2128,7 @@ static void TestSubWithValue(int32_t inputsize, int32_t outputsize)
7,7,7,7,7,7
};
if(!testConvertFromUnicodeWithContext(iso_2022_cn_inputText5, UPRV_LENGTHOF(iso_2022_cn_inputText5),
to_iso_2022_cn5_v2, sizeof(to_iso_2022_cn5_v2), "iso-2022-cn",
to_iso_2022_cn5_v2, UPRV_LENGTHOF(to_iso_2022_cn5_v2), "iso-2022-cn",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_cnOffs5_v2, NULL, 0,UCNV_ESCAPE_JAVA,U_ZERO_ERROR ))
log_err("u-> iso-2022-cn with sub & UCNV_ESCAPE_JAVA did not match.\n");
@ -2152,7 +2152,7 @@ static void TestSubWithValue(int32_t inputsize, int32_t outputsize)
7, 7, 7, 7, 7, 7, 7, 7,
};
if(!testConvertFromUnicodeWithContext(iso_2022_cn_inputText6, UPRV_LENGTHOF(iso_2022_cn_inputText6),
to_iso_2022_cn6_v2, sizeof(to_iso_2022_cn6_v2), "iso-2022-cn",
to_iso_2022_cn6_v2, UPRV_LENGTHOF(to_iso_2022_cn6_v2), "iso-2022-cn",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_cnOffs6_v2, NULL, 0,UCNV_ESCAPE_UNICODE,U_ZERO_ERROR ))
log_err("u-> iso-2022-cn with sub & UCNV_ESCAPE_UNICODE did not match.\n");
@ -2175,7 +2175,7 @@ static void TestSubWithValue(int32_t inputsize, int32_t outputsize)
7, 7, 7, 7, 7, 7,
};
if(!testConvertFromUnicodeWithContext(iso_2022_cn_inputText7, UPRV_LENGTHOF(iso_2022_cn_inputText7),
to_iso_2022_cn7_v2, sizeof(to_iso_2022_cn7_v2), "iso-2022-cn",
to_iso_2022_cn7_v2, UPRV_LENGTHOF(to_iso_2022_cn7_v2), "iso-2022-cn",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_cnOffs7_v2, NULL, 0,"K" ,U_ZERO_ERROR ))
log_err("u-> iso-2022-cn with sub & K did not match.\n");
@ -2208,7 +2208,7 @@ static void TestSubWithValue(int32_t inputsize, int32_t outputsize)
9, 9, 9, 9, 9
};
if(!testConvertFromUnicodeWithContext(iso_2022_cn_inputText8, UPRV_LENGTHOF(iso_2022_cn_inputText8),
to_iso_2022_cn8_v2, sizeof(to_iso_2022_cn8_v2), "iso-2022-cn",
to_iso_2022_cn8_v2, UPRV_LENGTHOF(to_iso_2022_cn8_v2), "iso-2022-cn",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_cnOffs8_v2, NULL, 0,UCNV_ESCAPE_CSS2,U_ZERO_ERROR ))
log_err("u-> iso-2022-cn with sub & UCNV_ESCAPE_CSS2 did not match.\n");
@ -2234,40 +2234,40 @@ static void TestSubWithValue(int32_t inputsize, int32_t outputsize)
};
if(!testConvertFromUnicodeWithContext(iso_2022_cn_inputText4, UPRV_LENGTHOF(iso_2022_cn_inputText4),
to_iso_2022_cn4_v3, sizeof(to_iso_2022_cn4_v3), "iso-2022-cn",
to_iso_2022_cn4_v3, UPRV_LENGTHOF(to_iso_2022_cn4_v3), "iso-2022-cn",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_cnOffs4_v3, NULL, 0,UCNV_ESCAPE_C,U_ZERO_ERROR ))
{
log_err("u-> iso-2022-cn with skip & UCNV_ESCAPE_C did not match.\n");
}
}
if(!testConvertFromUnicode(iso_2022_cn_inputText, UPRV_LENGTHOF(iso_2022_cn_inputText),
to_iso_2022_cn, sizeof(to_iso_2022_cn), "iso-2022-cn",
to_iso_2022_cn, UPRV_LENGTHOF(to_iso_2022_cn), "iso-2022-cn",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_cnOffs, NULL, 0 ))
log_err("u-> iso_2022_cn with subst with value did not match.\n");
if(!testConvertFromUnicode(iso_2022_cn_inputText4, UPRV_LENGTHOF(iso_2022_cn_inputText4),
to_iso_2022_cn4, sizeof(to_iso_2022_cn4), "iso-2022-cn",
to_iso_2022_cn4, UPRV_LENGTHOF(to_iso_2022_cn4), "iso-2022-cn",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_cnOffs4, NULL, 0 ))
log_err("u-> iso_2022_cn with subst with value did not match.\n");
if(!testConvertFromUnicode(iso_2022_kr_inputText, UPRV_LENGTHOF(iso_2022_kr_inputText),
to_iso_2022_kr, sizeof(to_iso_2022_kr), "iso-2022-kr",
to_iso_2022_kr, UPRV_LENGTHOF(to_iso_2022_kr), "iso-2022-kr",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_krOffs, NULL, 0 ))
log_err("u-> iso_2022_kr with subst with value did not match.\n");
if(!testConvertFromUnicode(iso_2022_kr_inputText2, UPRV_LENGTHOF(iso_2022_kr_inputText2),
to_iso_2022_kr2, sizeof(to_iso_2022_kr2), "iso-2022-kr",
to_iso_2022_kr2, UPRV_LENGTHOF(to_iso_2022_kr2), "iso-2022-kr",
UCNV_FROM_U_CALLBACK_ESCAPE, from_iso_2022_krOffs2, NULL, 0 ))
log_err("u-> iso_2022_kr2 with subst with value did not match.\n");
if(!testConvertFromUnicode(hz_inputText, UPRV_LENGTHOF(hz_inputText),
to_hz, sizeof(to_hz), "HZ",
to_hz, UPRV_LENGTHOF(to_hz), "HZ",
UCNV_FROM_U_CALLBACK_ESCAPE, from_hzOffs, NULL, 0 ))
log_err("u-> hz with subst with value did not match.\n");
if(!testConvertFromUnicode(hz_inputText2, UPRV_LENGTHOF(hz_inputText2),
to_hz2, sizeof(to_hz2), "HZ",
to_hz2, UPRV_LENGTHOF(to_hz2), "HZ",
UCNV_FROM_U_CALLBACK_ESCAPE, from_hzOffs2, NULL, 0 ))
log_err("u-> hz with subst with value did not match.\n");
if(!testConvertFromUnicode(iscii_inputText, UPRV_LENGTHOF(iscii_inputText),
to_iscii, sizeof(to_iscii), "ISCII,version=0",
to_iscii, UPRV_LENGTHOF(to_iscii), "ISCII,version=0",
UCNV_FROM_U_CALLBACK_ESCAPE, from_isciiOffs, NULL, 0 ))
log_err("u-> iscii with subst with value did not match.\n");
}
@ -2429,27 +2429,27 @@ static void TestSubWithValue(int32_t inputsize, int32_t outputsize)
#if !UCONFIG_NO_LEGACY_CONVERSION
if(!testConvertToUnicode(sampleTxtToU, sizeof(sampleTxtToU),
if(!testConvertToUnicode(sampleTxtToU, UPRV_LENGTHOF(sampleTxtToU),
IBM_943toUnicode, UPRV_LENGTHOF(IBM_943toUnicode),"ibm-943",
UCNV_TO_U_CALLBACK_ESCAPE, fromIBM943Offs, NULL, 0 ))
log_err("ibm-943->u with substitute with value did not match.\n");
if(!testConvertToUnicode(sampleTxt_EUC_JP, sizeof(sampleTxt_EUC_JP),
if(!testConvertToUnicode(sampleTxt_EUC_JP, UPRV_LENGTHOF(sampleTxt_EUC_JP),
EUC_JPtoUnicode, UPRV_LENGTHOF(EUC_JPtoUnicode),"IBM-eucJP",
UCNV_TO_U_CALLBACK_ESCAPE, fromEUC_JPOffs, NULL, 0))
log_err("euc-jp->u with substitute with value did not match.\n");
if(!testConvertToUnicode(sampleTxt_euc_tw, sizeof(sampleTxt_euc_tw),
if(!testConvertToUnicode(sampleTxt_euc_tw, UPRV_LENGTHOF(sampleTxt_euc_tw),
euc_twtoUnicode, UPRV_LENGTHOF(euc_twtoUnicode),"euc-tw",
UCNV_TO_U_CALLBACK_ESCAPE, from_euc_twOffs, NULL, 0))
log_err("euc-tw->u with substitute with value did not match.\n");
if(!testConvertToUnicode(sampleTxt_iso_2022_jp, sizeof(sampleTxt_iso_2022_jp),
if(!testConvertToUnicode(sampleTxt_iso_2022_jp, UPRV_LENGTHOF(sampleTxt_iso_2022_jp),
iso_2022_jptoUnicode, UPRV_LENGTHOF(iso_2022_jptoUnicode),"iso-2022-jp",
UCNV_TO_U_CALLBACK_ESCAPE, from_iso_2022_jpOffs, NULL, 0))
log_err("iso-2022-jp->u with substitute with value did not match.\n");
if(!testConvertToUnicodeWithContext(sampleTxt_iso_2022_jp, sizeof(sampleTxt_iso_2022_jp),
if(!testConvertToUnicodeWithContext(sampleTxt_iso_2022_jp, UPRV_LENGTHOF(sampleTxt_iso_2022_jp),
iso_2022_jptoUnicode, UPRV_LENGTHOF(iso_2022_jptoUnicode),"iso-2022-jp",
UCNV_TO_U_CALLBACK_ESCAPE, from_iso_2022_jpOffs, NULL, 0,"K",U_ZERO_ERROR))
log_err("iso-2022-jp->u with substitute with value did not match.\n");
@ -2463,8 +2463,8 @@ static void TestSubWithValue(int32_t inputsize, int32_t outputsize)
0x0026, 0x0023, 0x0032, 0x0036, 0x003b,
0x0042 };
static const int32_t from_iso_2022_jpOffsDec [] ={ 3,7,7,7,7,7,7,7,7,7,7,12, };
if(!testConvertToUnicodeWithContext(sampleTxt_iso_2022_jp, sizeof(sampleTxt_iso_2022_jp),
iso_2022_jptoUnicodeDec, sizeof(iso_2022_jptoUnicodeDec)/sizeof(iso_2022_jptoUnicode[0]),"iso-2022-jp",
if(!testConvertToUnicodeWithContext(sampleTxt_iso_2022_jp, UPRV_LENGTHOF(sampleTxt_iso_2022_jp),
iso_2022_jptoUnicodeDec, UPRV_LENGTHOF(iso_2022_jptoUnicodeDec),"iso-2022-jp",
UCNV_TO_U_CALLBACK_ESCAPE, from_iso_2022_jpOffsDec, NULL, 0,UCNV_ESCAPE_XML_DEC,U_ZERO_ERROR ))
log_err("iso-2022-jp->u with substitute with value and UCNV_ESCAPE_XML_DEC did not match.\n");
}
@ -2476,8 +2476,8 @@ static void TestSubWithValue(int32_t inputsize, int32_t outputsize)
0x0026, 0x0023, 0x0078, 0x0031, 0x0041, 0x003b,
0x0042 };
static const int32_t from_iso_2022_jpOffsHex [] ={ 3,7,7,7,7,7,7,7,7,7,7,7,7,12 };
if(!testConvertToUnicodeWithContext(sampleTxt_iso_2022_jp, sizeof(sampleTxt_iso_2022_jp),
iso_2022_jptoUnicodeHex, sizeof(iso_2022_jptoUnicodeHex)/sizeof(iso_2022_jptoUnicode[0]),"iso-2022-jp",
if(!testConvertToUnicodeWithContext(sampleTxt_iso_2022_jp, UPRV_LENGTHOF(sampleTxt_iso_2022_jp),
iso_2022_jptoUnicodeHex, UPRV_LENGTHOF(iso_2022_jptoUnicodeHex),"iso-2022-jp",
UCNV_TO_U_CALLBACK_ESCAPE, from_iso_2022_jpOffsHex, NULL, 0,UCNV_ESCAPE_XML_HEX,U_ZERO_ERROR ))
log_err("iso-2022-jp->u with substitute with value and UCNV_ESCAPE_XML_HEX did not match.\n");
}
@ -2488,38 +2488,38 @@ static void TestSubWithValue(int32_t inputsize, int32_t outputsize)
0x005C, 0x0078, 0x0031, 0x0041, /* \x1A */
0x0042 };
int32_t from_iso_2022_jpOffsC [] ={ 3,7,7,7,7,7,7,7,7,12 };
if(!testConvertToUnicodeWithContext(sampleTxt_iso_2022_jp, sizeof(sampleTxt_iso_2022_jp),
iso_2022_jptoUnicodeC, sizeof(iso_2022_jptoUnicodeC)/sizeof(iso_2022_jptoUnicode[0]),"iso-2022-jp",
if(!testConvertToUnicodeWithContext(sampleTxt_iso_2022_jp, UPRV_LENGTHOF(sampleTxt_iso_2022_jp),
iso_2022_jptoUnicodeC, UPRV_LENGTHOF(iso_2022_jptoUnicodeC),"iso-2022-jp",
UCNV_TO_U_CALLBACK_ESCAPE, from_iso_2022_jpOffsC, NULL, 0,UCNV_ESCAPE_C,U_ZERO_ERROR ))
log_err("iso-2022-jp->u with substitute with value and UCNV_ESCAPE_C did not match.\n");
}
}
if(!testConvertToUnicode(sampleTxt_iso_2022_cn, sizeof(sampleTxt_iso_2022_cn),
if(!testConvertToUnicode(sampleTxt_iso_2022_cn, UPRV_LENGTHOF(sampleTxt_iso_2022_cn),
iso_2022_cntoUnicode, UPRV_LENGTHOF(iso_2022_cntoUnicode),"iso-2022-cn",
UCNV_TO_U_CALLBACK_ESCAPE, from_iso_2022_cnOffs, NULL, 0))
log_err("iso-2022-cn->u with substitute with value did not match.\n");
if(!testConvertToUnicode(sampleTxt_iso_2022_kr, sizeof(sampleTxt_iso_2022_kr),
if(!testConvertToUnicode(sampleTxt_iso_2022_kr, UPRV_LENGTHOF(sampleTxt_iso_2022_kr),
iso_2022_krtoUnicode, UPRV_LENGTHOF(iso_2022_krtoUnicode),"iso-2022-kr",
UCNV_TO_U_CALLBACK_ESCAPE, from_iso_2022_krOffs, NULL, 0))
log_err("iso-2022-kr->u with substitute with value did not match.\n");
if(!testConvertToUnicode(sampleTxt_hz, sizeof(sampleTxt_hz),
if(!testConvertToUnicode(sampleTxt_hz, UPRV_LENGTHOF(sampleTxt_hz),
hztoUnicode, UPRV_LENGTHOF(hztoUnicode),"HZ",
UCNV_TO_U_CALLBACK_ESCAPE, from_hzOffs, NULL, 0))
log_err("hz->u with substitute with value did not match.\n");
if(!testConvertToUnicode(sampleTxt_iscii, sizeof(sampleTxt_iscii),
if(!testConvertToUnicode(sampleTxt_iscii, UPRV_LENGTHOF(sampleTxt_iscii),
isciitoUnicode, UPRV_LENGTHOF(isciitoUnicode),"ISCII,version=0",
UCNV_TO_U_CALLBACK_ESCAPE, from_isciiOffs, NULL, 0))
log_err("ISCII ->u with substitute with value did not match.\n");
#endif
if(!testConvertToUnicode(sampleTxtUTF8, sizeof(sampleTxtUTF8),
if(!testConvertToUnicode(sampleTxtUTF8, UPRV_LENGTHOF(sampleTxtUTF8),
UTF8ToUnicode, UPRV_LENGTHOF(UTF8ToUnicode),"UTF-8",
UCNV_TO_U_CALLBACK_ESCAPE, fromUTF8, NULL, 0))
log_err("UTF8->u with UCNV_TO_U_CALLBACK_ESCAPE with value did not match.\n");
if(!testConvertToUnicodeWithContext(sampleTxtUTF8, sizeof(sampleTxtUTF8),
if(!testConvertToUnicodeWithContext(sampleTxtUTF8, UPRV_LENGTHOF(sampleTxtUTF8),
UTF8ToUnicodeXML_DEC, UPRV_LENGTHOF(UTF8ToUnicodeXML_DEC),"UTF-8",
UCNV_TO_U_CALLBACK_ESCAPE, fromUTF8XML_DEC, NULL, 0, UCNV_ESCAPE_XML_DEC, U_ZERO_ERROR))
log_err("UTF8->u with UCNV_TO_U_CALLBACK_ESCAPE with value did not match.\n");
@ -2548,23 +2548,23 @@ static void TestLegalAndOthers(int32_t inputsize, int32_t outputsize)
gOutBufferSize = outputsize;
/*checking with a legal value*/
if(!testConvertFromUnicode(legalText, UPRV_LENGTHOF(legalText),
templegal949, sizeof(templegal949), "ibm-949",
templegal949, UPRV_LENGTHOF(templegal949), "ibm-949",
UCNV_FROM_U_CALLBACK_SKIP, to949legal, NULL, 0 ))
log_err("u-> ibm-949 with skip did not match.\n");
/*checking illegal value for ibm-943 with substitute*/
if(!testConvertToUnicode(text943, sizeof(text943),
if(!testConvertToUnicode(text943, UPRV_LENGTHOF(text943),
toUnicode943sub, UPRV_LENGTHOF(toUnicode943sub),"ibm-943",
UCNV_TO_U_CALLBACK_SUBSTITUTE, fromIBM943Offssub, NULL, 0 ))
log_err("ibm-943->u with subst did not match.\n");
/*checking illegal value for ibm-943 with skip */
if(!testConvertToUnicode(text943, sizeof(text943),
if(!testConvertToUnicode(text943, UPRV_LENGTHOF(text943),
toUnicode943skip, UPRV_LENGTHOF(toUnicode943skip),"ibm-943",
UCNV_TO_U_CALLBACK_SKIP, fromIBM943Offsskip, NULL, 0 ))
log_err("ibm-943->u with skip did not match.\n");
/*checking illegal value for ibm-943 with stop */
if(!testConvertToUnicode(text943, sizeof(text943),
if(!testConvertToUnicode(text943, UPRV_LENGTHOF(text943),
toUnicode943stop, UPRV_LENGTHOF(toUnicode943stop),"ibm-943",
UCNV_TO_U_CALLBACK_STOP, fromIBM943Offsstop, NULL, 0 ))
log_err("ibm-943->u with stop did not match.\n");
@ -2582,7 +2582,7 @@ static void TestSingleByte(int32_t inputsize, int32_t outputsize)
gInBufferSize = inputsize;
gOutBufferSize = outputsize;
if(!testConvertToUnicode(sampleText, sizeof(sampleText),
if(!testConvertToUnicode(sampleText, UPRV_LENGTHOF(sampleText),
toUnicode943sub, UPRV_LENGTHOF(toUnicode943sub),"ibm-943",
UCNV_TO_U_CALLBACK_SUBSTITUTE, fromIBM943Offssub, NULL, 0 ))
log_err("ibm-943->u with subst did not match.\n");
@ -2605,12 +2605,12 @@ static void TestEBCDIC_STATEFUL_Sub(int32_t inputsize, int32_t outputsize)
gOutBufferSize = outputsize;
if(!testConvertFromUnicode(ebcdic_inputTest, UPRV_LENGTHOF(ebcdic_inputTest),
toIBM930, sizeof(toIBM930), "ibm-930",
toIBM930, UPRV_LENGTHOF(toIBM930), "ibm-930",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, offset_930, NULL, 0 ))
log_err("u-> ibm-930(EBCDIC_STATEFUL) with subst did not match.\n");
if(!testConvertFromUnicode(ebcdic_inputTest, UPRV_LENGTHOF(ebcdic_inputTest),
toIBM930_subvaried, sizeof(toIBM930_subvaried), "ibm-930",
toIBM930_subvaried, UPRV_LENGTHOF(toIBM930_subvaried), "ibm-930",
UCNV_FROM_U_CALLBACK_SUBSTITUTE, offset_930_subvaried, mySubChar, 1 ))
log_err("u-> ibm-930(EBCDIC_STATEFUL) with subst(setSubChar=0x3f) did not match.\n");
}

View file

@ -1209,7 +1209,7 @@ NumberFormatTest::TestCurrency(void)
if (U_FAILURE(status))
errln((UnicodeString)"FAIL: Status " + (int32_t)status);
for(int i=0; i < (int)(sizeof(testCases)/sizeof(testCases[i])); i++){
for(int i=0; i < UPRV_LENGTHOF(testCases); i++){
status = U_ZERO_ERROR;
const char *localeID = testCases[i][0];
UnicodeString expected(testCases[i][1], -1, US_INV);
@ -6738,7 +6738,7 @@ void NumberFormatTest::TestFormatAttributes() {
UNUM_DECIMAL_SEPARATOR_FIELD, 7, 8,
UNUM_FRACTION_FIELD, 8, 10,
};
int32_t tupleCount = sizeof(expected)/(3 * sizeof(*expected));
int32_t tupleCount = UPRV_LENGTHOF(expected)/3;
FieldPositionIterator posIter;
UnicodeString result;
@ -6771,7 +6771,7 @@ void NumberFormatTest::TestFormatAttributes() {
UNUM_EXPONENT_SIGN_FIELD, 6, 7,
UNUM_EXPONENT_FIELD, 7, 8
};
int32_t tupleCount = sizeof(expected)/(3 * sizeof(*expected));
int32_t tupleCount = UPRV_LENGTHOF(expected)/3;
FieldPositionIterator posIter;
UnicodeString result;
@ -7048,7 +7048,7 @@ void NumberFormatTest::TestExplicitParents() {
UnicodeString s;
for(int i=0; i < (int)(sizeof(parentLocaleTests)/sizeof(parentLocaleTests[i])); i++){
for(int i=0; i < UPRV_LENGTHOF(parentLocaleTests); i++){
UErrorCode status = U_ZERO_ERROR;
const char *localeID = parentLocaleTests[i][0];
UnicodeString expected(parentLocaleTests[i][1], -1, US_INV);

View file

@ -1225,7 +1225,7 @@ void CollationRegressionTest::TestT7189() {
return;
}
for (i = 0; i < sizeof(text1) / (CollationRegressionTest::MAX_TOKEN_LEN * sizeof(UChar)); i++) {
for (i = 0; i < UPRV_LENGTHOF(text1); i++) {
uint8_t key1[100], key2[100];
int32_t len1, len2;

View file

@ -1,6 +1,6 @@
/*
**********************************************************************
* Copyright (C) 2000-2014, International Business Machines
* Copyright (C) 2000-2016, International Business Machines
* Corporation and others. All Rights Reserved.
**********************************************************************
* Date Name Description
@ -1619,7 +1619,7 @@ void TransliteratorRoundTripTest::TestDebug(const char* name,const char fromSet[
void TransliteratorRoundTripTest::TestInterIndic() {
//TestDebug("Latin-Gurmukhi", latinForIndic, "[:Gurmukhi:]","[\\u0965\\u0a02\\u0a72\\u0a73\\u0a74]",TRUE);
int32_t num = (int32_t)(sizeof(interIndicArray)/(INTER_INDIC_ARRAY_WIDTH*sizeof(char*)));
int32_t num = UPRV_LENGTHOF(interIndicArray)/INTER_INDIC_ARRAY_WIDTH;
if(quick){
logln("Testing only 5 of %i. Skipping rest (use -e for exhaustive)",num);
num = 5;

View file

@ -2104,7 +2104,7 @@ TimeZoneRuleTest::TestT6216(void) {
int32_t i, j;
// Get test times
UDate times[sizeof(TestDates) / (3 * sizeof(int32_t))];
UDate times[UPRV_LENGTHOF(TestDates)];
int32_t numTimes;
UErrorCode status = U_ZERO_ERROR;