/******************************************************************** * COPYRIGHT: * Copyright (c) 1997-2001, International Business Machines Corporation and * others. All Rights Reserved. ********************************************************************/ /******************************************************************************** * * File CCONVTST.C * * Modification History: * Name Description * Madhu Katragadda 7/7/2000 Converter Tests for extended code coverage ********************************************************************************* */ #include #include "cmemory.h" #include "unicode/uloc.h" #include "unicode/ucnv.h" #include "cintltst.h" #include "unicode/utypes.h" #include "unicode/ustring.h" #define MAX_LENGTH 999 #define UNICODE_LIMIT 0x10FFFF #define SURROGATE_HIGH_START 0xD800 #define SURROGATE_LOW_END 0xDFFF static int32_t gInBufferSize = 0; static int32_t gOutBufferSize = 0; static char gNuConvTestName[1024]; #define nct_min(x,y) ((x ibm-920 [UCNV_SBCS] not match.\n"); /*LATIN_1*/ if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "LATIN_1", 0, TRUE, U_ZERO_ERROR )) log_err("u-> LATIN_1 not match.\n"); } log_verbose("Testing for DBCS and MBCS\n"); { UChar sampleText[] = {0x00a1, 0xd801, 0xdc01, 0x00a4}; const uint8_t expected[] = {0xa2, 0xae, 0xa1, 0xe0, 0xa2, 0xb4}; int32_t offsets[] = {0x00, 0x00, 0x01, 0x01, 0x03, 0x03 }; /*DBCS*/ if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1362", 0 , TRUE, U_ZERO_ERROR)) log_err("u-> ibm-1362 [UCNV_DBCS] not match.\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1362", offsets , TRUE, U_ZERO_ERROR)) log_err("u-> ibm-1362 [UCNV_DBCS] not match.\n"); /*MBCS*/ if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1363", 0 , TRUE, U_ZERO_ERROR)) log_err("u-> ibm-1363 [UCNV_MBCS] not match.\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1363", offsets, TRUE, U_ZERO_ERROR)) log_err("u-> ibm-1363 [UCNV_MBCS] not match.\n"); } log_verbose("Testing for ISO-2022-jp\n"); { UChar sampleText[] = { 0x4e00, 0x04e01, 0x0031, 0xd801, 0xdc01, 0x0032}; const uint8_t expected[] = {0x1b, 0x24, 0x42,0x30,0x6c,0x43,0x7a,0x1b,0x28,0x42, 0x31,0x1A, 0x32}; int32_t offsets[] = {0,0,0,0,0,1,1,2,2,2,2,3,5 }; /*iso-2022-jp*/ if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-jp", 0 , TRUE, U_ZERO_ERROR)) log_err("u-> not match.\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-jp", offsets , TRUE, U_ZERO_ERROR)) log_err("u-> not match.\n"); } log_verbose("Testing for ISO-2022-cn\n"); { static const UChar sampleText[] = { 0x4e00, 0x04e01, 0x0031, 0xd801, 0xdc01, 0x0032}; static const uint8_t expected[] = { 0x1B, 0x24, 0x29, 0x41, 0x0E, 0x52, 0x3B, 0x36, 0x21, 0x0F, 0x31, 0x1A, 0x0f, 0x32 }; static const int32_t offsets[] = { 0, 0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 3, 5, 5, }; /*iso-2022-CN*/ if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-cn", 0 , TRUE, U_ZERO_ERROR)) log_err("u-> not match.\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-cn", offsets , TRUE, U_ZERO_ERROR)) log_err("u-> not match.\n"); } log_verbose("Testing for ISO-2022-kr\n"); { static const UChar sampleText[] = { 0x4e00,0xd801, 0xdc01, 0x04e01, 0x0031, 0xd801, 0xdc01, 0x0032}; static const uint8_t expected[] = {0x1B, 0x24, 0x29, 0x43, 0x0E, 0x6C, 0x69, 0x0f, 0x1A, 0x0e, 0x6F, 0x4B, 0x0F, 0x31, 0x1A, 0x32 }; static const int32_t offsets[] = {-1, -1, -1, -1, 0, 0, 0, 1, 1, 3, 3, 3, 4, 4, 5, 7, }; /*iso-2022-kr*/ if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-kr", 0 , TRUE, U_ZERO_ERROR)) log_err("u-> iso-2022-kr [UCNV_DBCS] not match.\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-kr", offsets , TRUE, U_ZERO_ERROR)) log_err("u-> iso-2022-kr [UCNV_DBCS] not match.\n"); } log_verbose("Testing for HZ\n"); { static const UChar sampleText[] = { 0x4e00, 0xd801, 0xdc01, 0x04e01, 0x0031, 0xd801, 0xdc01, 0x0032}; static const uint8_t expected[] = {0x7E, 0x7B, 0x52, 0x3B, 0x7E, 0x7D, 0x1A, 0x7E, 0x7B, 0x36, 0x21, 0x7E, 0x7D, 0x31, 0x1A, 0x32 }; static const int32_t offsets[] = {0,0,0,0, 1,1,1, 3,3,3,3, 4,4,4, 5, 7,}; /*hz*/ if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "HZ", 0 , TRUE, U_ZERO_ERROR)) log_err("u-> not match.\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "HZ", offsets , TRUE, U_ZERO_ERROR)) log_err("u-> not match.\n"); } /*UTF-8*/ log_verbose("Testing for UTF8\n"); { static const UChar sampleText[] = { 0x4e00, 0x0701, 0x0031, 0xbfc1, 0xd801, 0xdc01, 0x0032}; static const int32_t offsets[]={0x00, 0x00, 0x00, 0x01, 0x01, 0x02, 0x03, 0x03, 0x03, 0x04, 0x04, 0x04, 0x04, 0x06 }; static const uint8_t expected[] = {0xe4, 0xb8, 0x80, 0xdc, 0x81, 0x31, 0xeb, 0xbf, 0x81, 0xF0, 0x90, 0x90, 0x81, 0x32}; static const int32_t fromOffsets[] = { 0x0000, 0x0003, 0x0005, 0x0006, 0x0009, 0x0009, 0x000D }; /*UTF-8*/ if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "UTF8", offsets, TRUE, U_ZERO_ERROR )) log_err("u-> UTF8 with offsets and flush true did not match.\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "UTF8", 0, TRUE, U_ZERO_ERROR )) log_err("u-> UTF8 with offsets and flush true did not match.\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "UTF8", offsets, FALSE, U_ZERO_ERROR )) log_err("u-> UTF8 with offsets and flush true did not match.\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "UTF8", 0, FALSE, U_ZERO_ERROR )) log_err("u-> UTF8 with offsets and flush true did not match.\n"); if(!convertToU(expected, sizeof(expected), sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "UTF8", 0, TRUE, U_ZERO_ERROR )) log_err("UTF8 -> did not match.\n"); if(!convertToU(expected, sizeof(expected), sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "UTF8", 0, FALSE, U_ZERO_ERROR )) log_err("UTF8 -> did not match.\n"); if(!convertToU(expected, sizeof(expected), sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "UTF8", fromOffsets, TRUE, U_ZERO_ERROR )) log_err("UTF8 -> did not match.\n"); if(!convertToU(expected, sizeof(expected), sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "UTF8", fromOffsets, FALSE, U_ZERO_ERROR )) log_err("UTF8 -> did not match.\n"); } } /*test various error behaviours*/ static void TestErrorBehaviour(){ log_verbose("Testing for SBCS and LATIN_1\n"); { static const UChar sampleText[] = { 0x0031, 0xd801}; static const UChar sampleText2[] = { 0x0031, 0xd801, 0x0032}; static const uint8_t expected[] = { 0x31}; static const uint8_t expected2[] = { 0x31, 0x1a, 0x32}; /*SBCS*/ if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-920", 0, TRUE, U_TRUNCATED_CHAR_FOUND)) log_err("u-> ibm-920 [UCNV_SBCS] \n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-920", 0, FALSE, U_ZERO_ERROR)) log_err("u-> ibm-920 [UCNV_SBCS] \n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "ibm-920", 0, TRUE, U_ZERO_ERROR)) log_err("u-> ibm-920 [UCNV_SBCS] did not match\n"); /*LATIN_1*/ if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "LATIN_1", 0, TRUE, U_TRUNCATED_CHAR_FOUND)) log_err("u-> LATIN_1 is supposed to fail\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "LATIN_1", 0, FALSE, U_ZERO_ERROR)) log_err("u-> LATIN_1 is supposed to fail\n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "LATIN_1", 0, TRUE, U_ZERO_ERROR)) log_err("u-> LATIN_1 did not match\n"); } log_verbose("Testing for DBCS and MBCS\n"); { static const UChar sampleText[] = { 0x00a1, 0xd801}; static const uint8_t expected[] = { 0xa2, 0xae}; static const int32_t offsets[] = { 0x00, 0x00, 0x01, 0x01}; static const UChar sampleText2[] = { 0x00a1, 0xd801, 0x00a4}; static const uint8_t expected2[] = { 0xa2, 0xae, 0xa1, 0xe0, 0xa2, 0xb4}; static const int32_t offsets2[] = { 0x00, 0x00, 0x01, 0x01, 0x02, 0x02}; static const UChar sampleText3MBCS[] = { 0x0001, 0x00a4, 0xdc01}; static const uint8_t expected3MBCS[] = { 0x01, 0xa2, 0xb4, 0xa1, 0xe0}; static const int32_t offsets3MBCS[] = { 0x00, 0x01, 0x01, 0x02, 0x02}; static const UChar sampleText4MBCS[] = { 0x0061, 0x00a6, 0xdc01}; static const uint8_t expected4MBCS[] = { 0x61, 0x8f, 0xa2, 0xc3, 0xf4, 0xfe}; static const int32_t offsets4MBCS[] = { 0x00, 0x01, 0x01, 0x01, 0x02, 0x02 }; /*DBCS*/ if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1362", 0, TRUE, U_TRUNCATED_CHAR_FOUND)) log_err("u-> ibm-1362 [UCNV_DBCS] is supposed to fail\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1362", 0, FALSE, U_ZERO_ERROR)) log_err("u-> ibm-1362 [UCNV_DBCS] is supposed to fail\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1362", offsets, TRUE, U_TRUNCATED_CHAR_FOUND)) log_err("u-> ibm-1362 [UCNV_DBCS] is supposed to fail\n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1362", offsets, FALSE, U_ZERO_ERROR)) log_err("u-> ibm-1362 [UCNV_DBCS] is supposed to fail\n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "ibm-1362", 0, TRUE, U_ZERO_ERROR)) log_err("u-> ibm-1362 [UCNV_DBCS] did not match \n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "ibm-1362", offsets2, TRUE, U_ZERO_ERROR)) log_err("u-> ibm-1362 [UCNV_DBCS] did not match \n"); /*MBCS*/ if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1363", 0, TRUE, U_TRUNCATED_CHAR_FOUND)) log_err("u-> ibm-1363 [UCNV_MBCS] \n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1363", 0, FALSE, U_ZERO_ERROR)) log_err("u-> ibm-1363 [UCNV_MBCS] \n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "ibm-1363", 0, TRUE, U_ZERO_ERROR)) log_err("u-> ibm-1363 [UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "ibm-1363", 0, FALSE, U_ZERO_ERROR)) log_err("u-> ibm-1363 [UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "ibm-1363", offsets2, FALSE, U_ZERO_ERROR)) log_err("u-> ibm-1363 [UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText3MBCS, sizeof(sampleText3MBCS)/sizeof(sampleText3MBCS[0]), expected3MBCS, sizeof(expected3MBCS), "ibm-1363", offsets3MBCS, TRUE, U_ZERO_ERROR)) log_err("u-> ibm-1363 [UCNV_MBCS] \n"); if(!convertFromU(sampleText3MBCS, sizeof(sampleText3MBCS)/sizeof(sampleText3MBCS[0]), expected3MBCS, sizeof(expected3MBCS), "ibm-1363", offsets3MBCS, FALSE, U_ZERO_ERROR)) log_err("u-> ibm-1363 [UCNV_MBCS] \n"); if(!convertFromU(sampleText4MBCS, sizeof(sampleText4MBCS)/sizeof(sampleText4MBCS[0]), expected4MBCS, sizeof(expected4MBCS), "euc-jp", offsets4MBCS, TRUE, U_ZERO_ERROR)) log_err("u-> euc-jp [UCNV_MBCS] \n"); if(!convertFromU(sampleText4MBCS, sizeof(sampleText4MBCS)/sizeof(sampleText4MBCS[0]), expected4MBCS, sizeof(expected4MBCS), "euc-jp", offsets4MBCS, FALSE, U_ZERO_ERROR)) log_err("u-> euc-jp [UCNV_MBCS] \n"); } /*iso-2022-jp*/ log_verbose("Testing for iso-2022-jp\n"); { static const UChar sampleText[] = { 0x0031, 0xd801}; static const uint8_t expected[] = { 0x31}; static const int32_t offsets[] = { 0x00}; static const UChar sampleText2[] = { 0x0031, 0xd801, 0x0032}; static const uint8_t expected2[] = { 0x31,0x1A,0x32}; static const int32_t offsets2[] = { 0x00,0x01,0x02}; static const UChar sampleText3MBCS[] = { 0x3000, 0x0050, 0xdc01,0x3001}; static const uint8_t expected3MBCS[] = { 0x1B, 0x24, 0x42, 0x21, 0x21, 0x1B, 0x28, 0x42, 0x50, 0x1A, 0x1B, 0x24, 0x42, 0x21, 0x22,}; static const int32_t offsets3MBCS[] = { 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01, 0x01, 0x01, 0x02, 0x03, 0x03, 0x03, 0x03, 0x03,}; static const UChar sampleText4MBCS[] = { 0x0061, 0x4e00, 0xdc01}; static const uint8_t expected4MBCS[] = { 0x61, 0x1b, 0x24, 0x42, 0x30, 0x6c,0x1b,0x28,0x42,0x1a}; static const int32_t offsets4MBCS[] = { 0x00, 0x01, 0x01 ,0x01, 0x01, 0x01,0x02,0x02,0x02,0x02 }; if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-jp", offsets, TRUE, U_TRUNCATED_CHAR_FOUND)) log_err("u-> iso-2022-jp [UCNV_MBCS] \n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-jp", offsets, FALSE, U_ZERO_ERROR)) log_err("u-> ibm-1363 [UCNV_MBCS] \n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "iso-2022-jp", offsets2, TRUE, U_ZERO_ERROR)) log_err("u->iso-2022-jp[UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "iso-2022-jp", offsets2, FALSE, U_ZERO_ERROR)) log_err("u-> iso-2022-jp [UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "iso-2022-jp", offsets2, FALSE, U_ZERO_ERROR)) log_err("u-> iso-2022-jp [UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText3MBCS, sizeof(sampleText3MBCS)/sizeof(sampleText3MBCS[0]), expected3MBCS, sizeof(expected3MBCS), "iso-2022-jp", offsets3MBCS, TRUE, U_ZERO_ERROR)) log_err("u->iso-2022-jp [UCNV_MBCS] \n"); if(!convertFromU(sampleText3MBCS, sizeof(sampleText3MBCS)/sizeof(sampleText3MBCS[0]), expected3MBCS, sizeof(expected3MBCS), "iso-2022-jp", offsets3MBCS, FALSE, U_ZERO_ERROR)) log_err("u-> iso-2022-jp[UCNV_MBCS] \n"); if(!convertFromU(sampleText4MBCS, sizeof(sampleText4MBCS)/sizeof(sampleText4MBCS[0]), expected4MBCS, sizeof(expected4MBCS), "iso-2022-jp", offsets4MBCS, TRUE, U_ZERO_ERROR)) log_err("u-> iso-2022-jp [UCNV_MBCS] \n"); if(!convertFromU(sampleText4MBCS, sizeof(sampleText4MBCS)/sizeof(sampleText4MBCS[0]), expected4MBCS, sizeof(expected4MBCS), "iso-2022-jp", offsets4MBCS, FALSE, U_ZERO_ERROR)) log_err("u-> iso-2022-jp [UCNV_MBCS] \n"); } /*iso-2022-cn*/ log_verbose("Testing for iso-2022-cn\n"); { static const UChar sampleText[] = { 0x0031, 0xd801}; static const uint8_t expected[] = { 0x0f, 0x31}; static const int32_t offsets[] = { 0x00, 0x00}; static const UChar sampleText2[] = { 0x0031, 0xd801, 0x0032}; static const uint8_t expected2[] = { 0x0f, 0x31, 0x1A,0x32}; static const int32_t offsets2[] = { 0x00, 0x00, 0x01,0x02}; static const UChar sampleText3MBCS[] = { 0x0051, 0x0050, 0xdc01}; static const uint8_t expected3MBCS[] = {0x0f, 0x51, 0x50, 0x1A}; static const int32_t offsets3MBCS[] = { 0x00, 0x00, 0x01, 0x02 }; static const UChar sampleText4MBCS[] = { 0x0061, 0x4e00, 0xdc01}; static const uint8_t expected4MBCS[] = { 0x0f, 0x61, 0x1b, 0x24, 0x29, 0x41, 0x0e, 0x52, 0x3b, 0x0f, 0x1a }; static const int32_t offsets4MBCS[] = { 0x00, 0x00, 0x01, 0x01 ,0x01, 0x01, 0x01, 0x01, 0x01, 0x02, 0x02 }; if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-cn", offsets, TRUE, U_TRUNCATED_CHAR_FOUND)) log_err("u-> iso-2022-cn [UCNV_MBCS] \n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-cn", offsets, FALSE, U_ZERO_ERROR)) log_err("u-> ibm-1363 [UCNV_MBCS] \n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "iso-2022-cn", offsets2, TRUE, U_ZERO_ERROR)) log_err("u->iso-2022-cn[UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "iso-2022-cn", offsets2, FALSE, U_ZERO_ERROR)) log_err("u-> iso-2022-cn [UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "iso-2022-cn", offsets2, FALSE, U_ZERO_ERROR)) log_err("u-> iso-2022-cn [UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText3MBCS, sizeof(sampleText3MBCS)/sizeof(sampleText3MBCS[0]), expected3MBCS, sizeof(expected3MBCS), "iso-2022-cn", offsets3MBCS, TRUE, U_ZERO_ERROR)) log_err("u->iso-2022-cn [UCNV_MBCS] \n"); if(!convertFromU(sampleText3MBCS, sizeof(sampleText3MBCS)/sizeof(sampleText3MBCS[0]), expected3MBCS, sizeof(expected3MBCS), "iso-2022-cn", offsets3MBCS, FALSE, U_ZERO_ERROR)) log_err("u-> iso-2022-cn[UCNV_MBCS] \n"); if(!convertFromU(sampleText4MBCS, sizeof(sampleText4MBCS)/sizeof(sampleText4MBCS[0]), expected4MBCS, sizeof(expected4MBCS), "iso-2022-cn", offsets4MBCS, TRUE, U_ZERO_ERROR)) log_err("u-> iso-2022-cn [UCNV_MBCS] \n"); if(!convertFromU(sampleText4MBCS, sizeof(sampleText4MBCS)/sizeof(sampleText4MBCS[0]), expected4MBCS, sizeof(expected4MBCS), "iso-2022-cn", offsets4MBCS, FALSE, U_ZERO_ERROR)) log_err("u-> iso-2022-cn [UCNV_MBCS] \n"); } /*iso-2022-kr*/ log_verbose("Testing for iso-2022-kr\n"); { static const UChar sampleText[] = { 0x0031, 0xd801}; static const uint8_t expected[] = { 0x1b, 0x24, 0x29, 0x43, 0x31}; static const int32_t offsets[] = { -1, -1, -1, -1, 0x00}; static const UChar sampleText2[] = { 0x0031, 0xd801, 0x0032}; static const uint8_t expected2[] = { 0x1b, 0x24, 0x29, 0x43, 0x31, 0x1A, 0x32}; static const int32_t offsets2[] = { -1, -1, -1, -1, 0x00, 0x01, 0x02}; static const UChar sampleText3MBCS[] = { 0x0051, 0x0050, 0xdc01}; static const uint8_t expected3MBCS[] = { 0x1b, 0x24, 0x29, 0x43, 0x51, 0x50, 0x1A }; static const int32_t offsets3MBCS[] = { -1, -1, -1, -1, 0x00, 0x01, 0x02, 0x02 }; static const UChar sampleText4MBCS[] = { 0x0061, 0x4e00, 0xdc01,0x4e00}; static const uint8_t expected4MBCS[] = { 0x1b, 0x24, 0x29, 0x43, 0x61, 0x0e, 0x6c, 0x69, 0x0f, 0x1a, 0x0e, 0x6c, 0x69,}; static const int32_t offsets4MBCS[] = { -1, -1, -1, -1, 0x00, 0x01 ,0x01, 0x01, 0x02, 0x02, 0x03, 0x03, 0x03 }; if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-kr", offsets, TRUE, U_TRUNCATED_CHAR_FOUND)) log_err("u-> iso-2022-kr [UCNV_MBCS] \n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-kr", offsets, FALSE, U_ZERO_ERROR)) log_err("u-> ibm-1363 [UCNV_MBCS] \n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "iso-2022-kr", offsets2, TRUE, U_ZERO_ERROR)) log_err("u->iso-2022-kr[UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "iso-2022-kr", offsets2, FALSE, U_ZERO_ERROR)) log_err("u-> iso-2022-kr [UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "iso-2022-kr", offsets2, FALSE, U_ZERO_ERROR)) log_err("u-> iso-2022-kr [UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText3MBCS, sizeof(sampleText3MBCS)/sizeof(sampleText3MBCS[0]), expected3MBCS, sizeof(expected3MBCS), "iso-2022-kr", offsets3MBCS, TRUE, U_ZERO_ERROR)) log_err("u->iso-2022-kr [UCNV_MBCS] \n"); if(!convertFromU(sampleText3MBCS, sizeof(sampleText3MBCS)/sizeof(sampleText3MBCS[0]), expected3MBCS, sizeof(expected3MBCS), "iso-2022-kr", offsets3MBCS, FALSE, U_ZERO_ERROR)) log_err("u-> iso-2022-kr[UCNV_MBCS] \n"); if(!convertFromU(sampleText4MBCS, sizeof(sampleText4MBCS)/sizeof(sampleText4MBCS[0]), expected4MBCS, sizeof(expected4MBCS), "iso-2022-kr", offsets4MBCS, TRUE, U_ZERO_ERROR)) log_err("u-> iso-2022-kr [UCNV_MBCS] \n"); if(!convertFromU(sampleText4MBCS, sizeof(sampleText4MBCS)/sizeof(sampleText4MBCS[0]), expected4MBCS, sizeof(expected4MBCS), "iso-2022-kr", offsets4MBCS, FALSE, U_ZERO_ERROR)) log_err("u-> iso-2022-kr [UCNV_MBCS] \n"); } /*HZ*/ log_verbose("Testing for HZ\n"); { static const UChar sampleText[] = { 0x0031, 0xd801}; static const uint8_t expected[] = { 0x7e, 0x7d, 0x31}; static const int32_t offsets[] = { 0x00, 0x00, 0x00}; static const UChar sampleText2[] = { 0x0031, 0xd801, 0x0032}; static const uint8_t expected2[] = { 0x7e, 0x7d, 0x31, 0x1A, 0x32 }; static const int32_t offsets2[] = { 0x00, 0x00, 0x00, 0x01, 0x02 }; static const UChar sampleText3MBCS[] = { 0x0051, 0x0050, 0xdc01}; static const uint8_t expected3MBCS[] = { 0x7e, 0x7d, 0x51, 0x50, 0x1A }; static const int32_t offsets3MBCS[] = { 0x00, 0x00, 0x00, 0x01, 0x02}; static const UChar sampleText4MBCS[] = { 0x0061, 0x4e00, 0xdc01}; static const uint8_t expected4MBCS[] = { 0x7e, 0x7d, 0x61, 0x7e, 0x7b, 0x52, 0x3b, 0x7e, 0x7d, 0x1a }; static const int32_t offsets4MBCS[] = { 0x00, 0x00, 0x00, 0x01, 0x01, 0x01 ,0x01, 0x02, 0x02, 0x02 }; if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "HZ", offsets, TRUE, U_TRUNCATED_CHAR_FOUND)) log_err("u-> HZ [UCNV_MBCS] \n"); if(!convertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "HZ", offsets, FALSE, U_ZERO_ERROR)) log_err("u-> ibm-1363 [UCNV_MBCS] \n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "HZ", offsets2, TRUE, U_ZERO_ERROR)) log_err("u->HZ[UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "HZ", offsets2, FALSE, U_ZERO_ERROR)) log_err("u-> HZ [UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText2, sizeof(sampleText2)/sizeof(sampleText2[0]), expected2, sizeof(expected2), "HZ", offsets2, FALSE, U_ZERO_ERROR)) log_err("u-> HZ [UCNV_DBCS] did not match\n"); if(!convertFromU(sampleText3MBCS, sizeof(sampleText3MBCS)/sizeof(sampleText3MBCS[0]), expected3MBCS, sizeof(expected3MBCS), "HZ", offsets3MBCS, TRUE, U_ZERO_ERROR)) log_err("u->HZ [UCNV_MBCS] \n"); if(!convertFromU(sampleText3MBCS, sizeof(sampleText3MBCS)/sizeof(sampleText3MBCS[0]), expected3MBCS, sizeof(expected3MBCS), "HZ", offsets3MBCS, FALSE, U_ZERO_ERROR)) log_err("u-> HZ[UCNV_MBCS] \n"); if(!convertFromU(sampleText4MBCS, sizeof(sampleText4MBCS)/sizeof(sampleText4MBCS[0]), expected4MBCS, sizeof(expected4MBCS), "HZ", offsets4MBCS, TRUE, U_ZERO_ERROR)) log_err("u-> HZ [UCNV_MBCS] \n"); if(!convertFromU(sampleText4MBCS, sizeof(sampleText4MBCS)/sizeof(sampleText4MBCS[0]), expected4MBCS, sizeof(expected4MBCS), "HZ", offsets4MBCS, FALSE, U_ZERO_ERROR)) log_err("u-> HZ [UCNV_MBCS] \n"); } } /*test different convertToUnicode error behaviours*/ static void TestToUnicodeErrorBehaviour() { log_verbose("Testing error conditions for DBCS\n"); { uint8_t sampleText[] = { 0xa2, 0xae, 0x03, 0x04}; const UChar expected[] = { 0x00a1 }; uint8_t sampleText2[] = { 0xa2, 0xae, 0xa2}; const UChar expected2[] = { 0x00a1 }; if(!convertToU(sampleText, sizeof(sampleText), expected, sizeof(expected)/sizeof(expected[0]), "ibm-1362", 0, TRUE, U_ZERO_ERROR )) log_err("DBCS (ibm-1362)->Unicode did not match.\n"); if(!convertToU(sampleText, sizeof(sampleText), expected, sizeof(expected)/sizeof(expected[0]), "ibm-1362", 0, FALSE, U_ZERO_ERROR )) log_err("DBCS (ibm-1362)->Unicode with flush = false did not match.\n"); if(!convertToU(sampleText2, sizeof(sampleText2), expected2, sizeof(expected2)/sizeof(expected2[0]), "ibm-1362", 0, TRUE, U_TRUNCATED_CHAR_FOUND )) log_err("DBCS (ibm-1362)->Unicode with TRUNCATED CHARACTER did not match.\n"); } log_verbose("Testing error conditions for SBCS\n"); { uint8_t sampleText[] = { 0xa2, 0xFF}; const UChar expected[] = { 0x00c2 }; /* uint8_t sampleText2[] = { 0xa2, 0x70 }; const UChar expected2[] = { 0x0073 };*/ if(!convertToU(sampleText, sizeof(sampleText), expected, sizeof(expected)/sizeof(expected[0]), "ibm-1051", 0, TRUE, U_ZERO_ERROR )) log_err("SBCS (ibm-1051)->Unicode did not match.\n"); if(!convertToU(sampleText, sizeof(sampleText), expected, sizeof(expected)/sizeof(expected[0]), "ibm-1051", 0, FALSE, U_ZERO_ERROR )) log_err("SBCS (ibm-1051)->Unicode with flush = false did not match.\n"); } log_verbose("Testing error conditions for UTF8\n"); { const uint8_t sampleText[] = { 0x31, 0xe4, 0xba, 0x8c, 0xe4, 0xb8 }; UChar expectedUTF8[] = { 0x0031, 0x4e8c}; int32_t offsets[] = { 0x0000, 0x0001}; const uint8_t sampleText2[] = { 0x31, 0xff, 0xe4, 0xba, 0x8c, 0xe0, 0x80, 0x61}; UChar expected2UTF8[] = { 0x0031, 0xfffd, 0x4e8c, 0xfffd, 0x0061}; int32_t offsets2[] = { 0x0000, 0x0001, 0x0002, 0x0005, 0x0007}; const uint8_t sampleText3[] = { 0x31, 0xfb, 0xbf, 0xbf, 0xbf, 0xbf, 0x61}; UChar expected3UTF8[] = { 0x0031, 0xfffd, 0x0061}; int32_t offsets3[] = { 0x0000, 0x0001, 0x0006}; if(!convertToU(sampleText, sizeof(sampleText), expectedUTF8, sizeof(expectedUTF8)/sizeof(expectedUTF8[0]), "utf-8", 0, TRUE, U_TRUNCATED_CHAR_FOUND )) log_err("utf-8->Unicode did not match.\n"); if(!convertToU(sampleText, sizeof(sampleText), expectedUTF8, sizeof(expectedUTF8)/sizeof(expectedUTF8[0]), "utf-8", 0, FALSE, U_ZERO_ERROR )) log_err("utf-8->Unicode did not match.\n"); if(!convertToU(sampleText, sizeof(sampleText), expectedUTF8, sizeof(expectedUTF8)/sizeof(expectedUTF8[0]), "utf-8", offsets, TRUE, U_TRUNCATED_CHAR_FOUND )) log_err("utf-8->Unicode did not match.\n"); if(!convertToU(sampleText, sizeof(sampleText), expectedUTF8, sizeof(expectedUTF8)/sizeof(expectedUTF8[0]), "utf-8", offsets, FALSE, U_ZERO_ERROR )) log_err("utf-8->Unicode did not match.\n"); if(!convertToU(sampleText2, sizeof(sampleText2), expected2UTF8, sizeof(expected2UTF8)/sizeof(expected2UTF8[0]), "utf-8", 0, TRUE, U_ZERO_ERROR )) log_err("utf-8->Unicode did not match.\n"); if(!convertToU(sampleText2, sizeof(sampleText2), expected2UTF8, sizeof(expected2UTF8)/sizeof(expected2UTF8[0]), "utf-8", 0, FALSE, U_ZERO_ERROR )) log_err("utf-8->Unicode did not match.\n"); if(!convertToU(sampleText2, sizeof(sampleText2), expected2UTF8, sizeof(expected2UTF8)/sizeof(expected2UTF8[0]), "utf-8", offsets2, TRUE, U_ZERO_ERROR )) log_err("utf-8->Unicode did not match.\n"); if(!convertToU(sampleText2, sizeof(sampleText2), expected2UTF8, sizeof(expected2UTF8)/sizeof(expected2UTF8[0]), "utf-8", offsets2, FALSE, U_ZERO_ERROR )) log_err("utf-8->Unicode did not match.\n"); if(!convertToU(sampleText3, sizeof(sampleText3), expected3UTF8, sizeof(expected3UTF8)/sizeof(expected3UTF8[0]), "utf-8", offsets3, TRUE, U_ZERO_ERROR )) log_err("utf-8->Unicode did not match.\n"); if(!convertToU(sampleText3, sizeof(sampleText3), expected3UTF8, sizeof(expected3UTF8)/sizeof(expected3UTF8[0]), "utf-8", offsets3, FALSE, U_ZERO_ERROR )) log_err("utf-8->Unicode did not match with flush false.\n"); } } static void TestGetNextErrorBehaviour(){ /*Test for unassigned character*/ #define INPUT_SIZE 1 static const char input1[INPUT_SIZE]={ 0x70 }; const char* source=(const char*)input1; UErrorCode err=U_ZERO_ERROR; UChar32 c=0; UConverter *cnv=ucnv_open("ibm-1159", &err); if(U_FAILURE(err)) { log_err("Unable to open a SBCS(ibm-1159) converter: %s\n", u_errorName(err)); return; } c=ucnv_getNextUChar(cnv, &source, source + INPUT_SIZE, &err); if(err != U_INVALID_CHAR_FOUND && c!=0xfffd){ log_err("FAIL in TestGetNextErrorBehaviour(unassigned): Expected: U_INVALID_CHAR_ERROR or 0xfffd ----Got:%s and 0x%lx\n", myErrorName(err), c); } ucnv_close(cnv); } #define MAX_UTF16_LEN 2 #define MAX_UTF8_LEN 4 /*Regression test for utf8 converter*/ static void TestRegressionUTF8(){ UChar32 currCh = 0; int32_t offset8; int32_t offset16; UChar *standardForm = (UChar*)uprv_malloc(MAX_LENGTH*sizeof(UChar)); uint8_t *utf8 = (uint8_t*)uprv_malloc(MAX_LENGTH); while (currCh <= UNICODE_LIMIT) { offset16 = 0; offset8 = 0; while(currCh <= UNICODE_LIMIT && offset16 < (MAX_LENGTH/sizeof(UChar) - MAX_UTF16_LEN) && offset8 < (MAX_LENGTH - MAX_UTF8_LEN)) { if (currCh == SURROGATE_HIGH_START) { currCh = SURROGATE_LOW_END + 1; /* Skip surrogate range */ } UTF16_APPEND_CHAR_SAFE(standardForm, offset16, MAX_LENGTH, currCh); UTF8_APPEND_CHAR_SAFE(utf8, offset8, MAX_LENGTH, currCh); currCh++; } if(!convertFromU(standardForm, offset16, utf8, offset8, "UTF8", 0, TRUE, U_ZERO_ERROR )) { log_err("Unicode->UTF8 did not match.\n"); } if(!convertToU(utf8, offset8, standardForm, offset16, "UTF8", 0, TRUE, U_ZERO_ERROR )) { log_err("UTF8->Unicode did not match.\n"); } } uprv_free(standardForm); uprv_free(utf8); } #define MAX_UTF32_LEN 1 static void TestRegressionUTF32(){ UChar32 currCh = 0; int32_t offset32; int32_t offset16; UChar *standardForm = (UChar*)uprv_malloc(MAX_LENGTH*sizeof(UChar)); UChar32 *utf32 = (UChar32*)uprv_malloc(MAX_LENGTH*sizeof(UChar32)); while (currCh <= UNICODE_LIMIT) { offset16 = 0; offset32 = 0; while(currCh <= UNICODE_LIMIT && offset16 < (MAX_LENGTH/sizeof(UChar) - MAX_UTF16_LEN) && offset32 < (MAX_LENGTH/sizeof(UChar32) - MAX_UTF32_LEN)) { if (currCh == SURROGATE_HIGH_START) { currCh = SURROGATE_LOW_END + 1; /* Skip surrogate range */ } UTF16_APPEND_CHAR_SAFE(standardForm, offset16, MAX_LENGTH, currCh); UTF32_APPEND_CHAR_SAFE(utf32, offset32, MAX_LENGTH, currCh); currCh++; } if(!convertFromU(standardForm, offset16, (const uint8_t *)utf32, offset32*sizeof(UChar32), "UTF32", 0, TRUE, U_ZERO_ERROR )) { log_err("Unicode->UTF8 did not match.\n"); } if(!convertToU((const uint8_t *)utf32, offset32*sizeof(UChar32), standardForm, offset16, "UTF32", 0, TRUE, U_ZERO_ERROR )) { log_err("UTF32->Unicode did not match.\n"); } } uprv_free(standardForm); uprv_free(utf32); } /*Walk through the available converters*/ static void TestAvailableConverters(){ UErrorCode status=U_ZERO_ERROR; UConverter *conv=NULL; int32_t i=0; for(i=0; i < ucnv_countAvailable(); i++){ status=U_ZERO_ERROR; conv=ucnv_open(ucnv_getAvailableName(i), &status); if(U_FAILURE(status)){ log_err("ERROR: converter creation failed. Failure in alias table or the data table for \n converter=%s. Error=%s\n", ucnv_getAvailableName(i), myErrorName(status)); continue; } ucnv_close(conv); } } static void TestFlushInternalBuffer(){ TestWithBufferSize(MAX_LENGTH, 1); TestWithBufferSize(1, 1); TestWithBufferSize(1, MAX_LENGTH); TestWithBufferSize(MAX_LENGTH, MAX_LENGTH); } static void TestWithBufferSize(int32_t insize, int32_t outsize){ gInBufferSize =insize; gOutBufferSize = outsize; log_verbose("Testing fromUnicode for UTF-8 with UCNV_TO_U_CALLBACK_SUBSTITUTE \n"); { UChar sampleText[] = { 0x0031, 0x0032, 0x0033, 0x0000, 0x4e00, 0x4e8c, 0x4e09, 0x002E }; const uint8_t expectedUTF8[] = { 0x31, 0x32, 0x33, 0x00, 0xe4, 0xb8, 0x80, 0xe4, 0xba, 0x8c, 0xe4, 0xb8, 0x89, 0x2E }; int32_t toUTF8Offs[] = { 0x00, 0x01, 0x02, 0x03, 0x04, 0x04, 0x04, 0x05, 0x05, 0x05, 0x06, 0x06, 0x06, 0x07}; /* int32_t fmUTF8Offs[] = { 0x0000, 0x0001, 0x0002, 0x0003, 0x0004, 0x0007, 0x000a, 0x000d };*/ /*UTF-8*/ if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expectedUTF8, sizeof(expectedUTF8), "UTF8", UCNV_FROM_U_CALLBACK_SUBSTITUTE, toUTF8Offs ,FALSE)) log_err("u-> UTF8 did not match.\n"); } log_verbose("Testing fromUnicode with UCNV_FROM_U_CALLBACK_ESCAPE \n"); { UChar inputTest[] = { 0x0061, 0xd801, 0xdc01, 0xd801, 0x0061 }; const uint8_t toIBM943[]= { 0x61, 0x25, 0x55, 0x44, 0x38, 0x30, 0x31, 0x25, 0x55, 0x44, 0x43, 0x30, 0x31, 0x25, 0x55, 0x44, 0x38, 0x30, 0x31, 0x61 }; int32_t offset[]= {0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 3, 3, 3, 3, 3, 4}; if(!testConvertFromU(inputTest, sizeof(inputTest)/sizeof(inputTest[0]), toIBM943, sizeof(toIBM943), "ibm-943", (UConverterFromUCallback)UCNV_FROM_U_CALLBACK_ESCAPE, offset,FALSE)) log_err("u-> ibm-943 with subst with value did not match.\n"); } log_verbose("Testing fromUnicode for UTF-8 with UCNV_TO_U_CALLBACK_SUBSTITUTE \n"); { const uint8_t sampleText1[] = { 0x31, 0xe4, 0xba, 0x8c, 0xe0, 0x80, 0x61}; UChar expected1[] = { 0x0031, 0x4e8c, 0xfffd, 0x0061}; int32_t offsets1[] = { 0x0000, 0x0001, 0x0004, 0x0006}; if(!testConvertToU(sampleText1, sizeof(sampleText1), expected1, sizeof(expected1)/sizeof(expected1[0]),"utf8", UCNV_TO_U_CALLBACK_SUBSTITUTE, offsets1,FALSE)) log_err("utf8->u with substitute did not match.\n");; } log_verbose("Testing toUnicode with UCNV_TO_U_CALLBACK_ESCAPE \n"); /*to Unicode*/ { const uint8_t sampleTxtToU[]= { 0x00, 0x9f, 0xaf, 0x81, 0xad, /*unassigned*/ 0x89, 0xd3 }; UChar IBM_943toUnicode[] = { 0x0000, 0x6D63, 0x25, 0x58, 0x38, 0x31, 0x25, 0x58, 0x41, 0x44, 0x7B87}; int32_t fromIBM943Offs [] = { 0, 1, 3, 3, 3, 3, 3, 3, 3, 3, 5}; if(!testConvertToU(sampleTxtToU, sizeof(sampleTxtToU), IBM_943toUnicode, sizeof(IBM_943toUnicode)/sizeof(IBM_943toUnicode[0]),"ibm-943", (UConverterToUCallback)UCNV_TO_U_CALLBACK_ESCAPE, fromIBM943Offs,FALSE)) log_err("ibm-943->u with substitute with value did not match.\n"); } } static UBool convertFromU( const UChar *source, int sourceLen, const uint8_t *expect, int expectLen, const char *codepage, const int32_t *expectOffsets, UBool doFlush, UErrorCode expectedStatus) { int32_t i=0; uint8_t *p=0; const UChar *src; uint8_t buffer[MAX_LENGTH]; int32_t offsetBuffer[MAX_LENGTH]; int32_t *offs=0; uint8_t *targ; uint8_t *targetLimit; UChar *sourceLimit=0; UErrorCode status = U_ZERO_ERROR; UConverter *conv = 0; conv = ucnv_open(codepage, &status); if(U_FAILURE(status)) { log_err("Couldn't open converter %s\n",codepage); return FALSE; } log_verbose("Converter %s opened..\n", ucnv_getName(conv, &status)); for(i=0; i %d chars out]. \nResult :", sourceLen, targ-buffer); if(expectLen != targ-buffer) { log_err("Expected %d chars out, got %d FROM Unicode to %s\n", expectLen, targ-buffer, codepage); log_verbose("Expected %d chars out, got %d FROM Unicode to %s\n", expectLen, targ-buffer, codepage); printSeqErr((const unsigned char *)buffer, targ-buffer); printSeqErr((const unsigned char*)expect, expectLen); return FALSE; } if(memcmp(buffer, expect, expectLen)){ log_err("String does not match. FROM Unicode to codePage%s\n", codepage); printf("\nGot:"); printSeqErr((const unsigned char *)buffer, expectLen); printf("\nExpected:"); printSeqErr((const unsigned char *)expect, expectLen); return FALSE; } else { log_verbose("Matches!\n"); } if (expectOffsets != 0){ log_verbose("comparing %d offsets..\n", targ-buffer); if(memcmp(offsetBuffer,expectOffsets,(targ-buffer) * sizeof(int32_t) )){ log_err("did not get the expected offsets. for FROM Unicode to %s\n", codepage); printf("\nGot : "); printSeqErr((const unsigned char*)buffer, targ-buffer); for(p=buffer;p %d chars out]. \nResult :", sourceLen, targ-buffer); log_verbose("comparing %d uchars (%d bytes)..\n",expectLen,expectLen*2); if (expectOffsets != 0) { if(memcmp(offsetBuffer, expectOffsets, (targ-buffer) * sizeof(int32_t))){ log_err("did not get the expected offsets from %s To UNICODE\n", codepage); printf("\nGot : "); for(p=buffer;p %d chars out]. \nResult :", sourceLen, targ-junkout); if(VERBOSITY) { char junk[999]; char offset_str[999]; uint8_t *ptr; junk[0] = 0; offset_str[0] = 0; for(ptr = junkout;ptr %d chars.\nResult :", sourcelen, targ-junkout); if(VERBOSITY) { char junk[999]; char offset_str[999]; UChar *ptr; junk[0] = 0; offset_str[0] = 0; for(ptr = junkout;ptr ibm-920 [UCNV_SBCS] not match.\n"); if(!testConvertToU(expected1, sizeof(expected1), sampleText1, sizeof(sampleText1)/sizeof(sampleText1[0]), "ibm-920",UCNV_TO_U_CALLBACK_SUBSTITUTE , offsets1, TRUE)) log_err("ibm -920 -> did not match.\n"); /*LATIN_1*/ if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "LATIN1", UCNV_FROM_U_CALLBACK_SUBSTITUTE , offsets, TRUE)) log_err("u-> LATIN_1 not match.\n"); if(!testConvertToU(expected1, sizeof(expected1), sampleText1, sizeof(sampleText1)/sizeof(sampleText1[0]), "LATIN1",UCNV_TO_U_CALLBACK_SUBSTITUTE , offsets1, TRUE)) log_err("LATIN1 -> did not match.\n"); } log_verbose("Testing Reset for DBCS and MBCS\n"); { static const UChar sampleText[] = {0x00a1, 0xd801, 0xdc01, 0x00a4}; static const uint8_t expected[] = {0xa2, 0xae, 0xa1, 0xe0, 0xa2, 0xb4}; static const int32_t offsets[] = {0x00, 0x00, 0x01, 0x01, 0x03, 0x03 }; static const UChar sampleText1[] = {0x00a1, 0x00a4, 0x00a7, 0x00a8}; static const uint8_t expected1[] = {0xa2, 0xae,0xA2,0xB4,0xA1,0xD7,0xA1,0xA7}; static const int32_t offsets1[] = { 0,2,4,6}; /*DBCS*/ if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1362", UCNV_FROM_U_CALLBACK_SUBSTITUTE , NULL, TRUE)) log_err("u-> ibm-1362 [UCNV_DBCS] not match.\n"); if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1362", UCNV_FROM_U_CALLBACK_SUBSTITUTE,offsets , TRUE)) log_err("u-> ibm-1362 [UCNV_DBCS] not match.\n"); if(!testConvertToU(expected1, sizeof(expected1), sampleText1, sizeof(sampleText1)/sizeof(sampleText1[0]), "ibm-1362",UCNV_TO_U_CALLBACK_SUBSTITUTE , offsets1, TRUE)) log_err("ibm-1362 -> did not match.\n"); /*MBCS*/ if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1363", UCNV_FROM_U_CALLBACK_SUBSTITUTE , NULL, TRUE)) log_err("u-> ibm-1363 [UCNV_MBCS] not match.\n"); if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "ibm-1363", UCNV_FROM_U_CALLBACK_SUBSTITUTE,offsets , TRUE)) log_err("u-> ibm-1363 [UCNV_MBCS] not match.\n"); if(!testConvertToU(expected1, sizeof(expected1), sampleText1, sizeof(sampleText1)/sizeof(sampleText1[0]), "ibm-1363",UCNV_TO_U_CALLBACK_SUBSTITUTE , offsets1, TRUE)) log_err("ibm-1363 -> did not match.\n"); } log_verbose("Testing Reset for ISO-2022-jp\n"); { static const UChar sampleText[] = { 0x4e00, 0x04e01, 0x0031, 0xd801, 0xdc01, 0x0032}; static const uint8_t expected[] = {0x1b, 0x24, 0x42,0x30,0x6c,0x43,0x7a,0x1b,0x28,0x42, 0x31,0x1A, 0x32}; static const int32_t offsets[] = {0,0,0,0,0,1,1,2,2,2,2,3,5 }; static const UChar sampleText1[] = {0x4e00, 0x04e01, 0x0031,0x001A, 0x0032}; static const uint8_t expected1[] = {0x1b, 0x24, 0x42,0x30,0x6c,0x43,0x7a,0x1b,0x28,0x42, 0x31,0x1A, 0x32}; static const int32_t offsets1[] = { 3,5,10,11,12}; /*iso-2022-jp*/ if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-jp", UCNV_FROM_U_CALLBACK_SUBSTITUTE , NULL, TRUE)) log_err("u-> not match.\n"); if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-jp", UCNV_FROM_U_CALLBACK_SUBSTITUTE,offsets , TRUE)) log_err("u-> not match.\n"); if(!testConvertToU(expected1, sizeof(expected1), sampleText1, sizeof(sampleText1)/sizeof(sampleText1[0]), "iso-2022-jp",UCNV_TO_U_CALLBACK_SUBSTITUTE , offsets1, TRUE)) log_err("iso-2022-jp -> did not match.\n"); } log_verbose("Testing Reset for ISO-2022-cn\n"); { static const UChar sampleText[] = { 0x4e00, 0x04e01, 0x0031, 0xd801, 0xdc01, 0x0032}; static const uint8_t expected[] = { 0x1B, 0x24, 0x29, 0x41, 0x0E, 0x52, 0x3B, 0x36, 0x21, 0x0f, 0x31, 0x1A, 0x0f, 0x32 }; static const int32_t offsets[] = { 0, 0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 3, 5, 5, }; UChar sampleText1[] = {0x4e00, 0x04e01, 0x0031,0x001A, 0x0032}; static const uint8_t expected1[] = { 0x1B, 0x24, 0x29, 0x41, 0x0E, 0x52, 0x3B, 0x36, 0x21, 0x1B, 0x24, 0x29, 0x47, 0x1B, 0x4E, 0x24, 0x22, 0x0f, 0x1A, 0x32 }; static const int32_t offsets1[] = { 5,7,15,18,19}; /*iso-2022-CN*/ if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-cn", UCNV_FROM_U_CALLBACK_SUBSTITUTE , NULL, TRUE)) log_err("u-> not match.\n"); if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-cn", UCNV_FROM_U_CALLBACK_SUBSTITUTE,offsets , TRUE)) log_err("u-> not match.\n"); if(!testConvertToU(expected1, sizeof(expected1), sampleText1, sizeof(sampleText1)/sizeof(sampleText1[0]), "iso-2022-cn",UCNV_TO_U_CALLBACK_SUBSTITUTE , offsets1, TRUE)) log_err("iso-2022-cn -> did not match.\n"); } log_verbose("Testing Reset for ISO-2022-kr\n"); { UChar sampleText[] = { 0x4e00,0xd801, 0xdc01, 0x04e01, 0x0031, 0xd801, 0xdc01, 0x0032}; static const uint8_t expected[] = {0x1B, 0x24, 0x29, 0x43, 0x0E, 0x6C, 0x69, 0x0f, 0x1A, 0x0e, 0x6F, 0x4B, 0x0F, 0x31, 0x1A, 0x32 }; static const int32_t offsets[] = {-1, -1, -1, -1, 0, 0, 0, 1, 1, 3, 3, 3, 4, 4, 5, 7, }; static const UChar sampleText1[] = { 0x4e00,0x0041, 0x04e01, 0x0031, 0x0042, 0x0032}; static const uint8_t expected1[] = {0x1B, 0x24, 0x29, 0x43, 0x0E, 0x6C, 0x69, 0x0f, 0x41, 0x0e, 0x6F, 0x4B, 0x0F, 0x31, 0x42, 0x32 }; static const int32_t offsets1[] = { 5, 8, 10, 13, 14, 15 }; /*iso-2022-kr*/ if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-kr", UCNV_FROM_U_CALLBACK_SUBSTITUTE , NULL, TRUE)) log_err("u-> iso-2022-kr [UCNV_DBCS] not match.\n"); if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "iso-2022-kr", UCNV_FROM_U_CALLBACK_SUBSTITUTE,offsets , TRUE)) log_err("u-> iso-2022-kr [UCNV_DBCS] not match.\n"); if(!testConvertToU(expected1, sizeof(expected1), sampleText1, sizeof(sampleText1)/sizeof(sampleText1[0]), "iso-2022-kr",UCNV_TO_U_CALLBACK_SUBSTITUTE , offsets1, TRUE)) log_err("iso-2022-kr -> did not match.\n"); } log_verbose("Testing Reset for HZ\n"); { static const UChar sampleText[] = { 0x4e00, 0xd801, 0xdc01, 0x04e01, 0x0031, 0xd801, 0xdc01, 0x0032}; static const uint8_t expected[] = {0x7E, 0x7B, 0x52, 0x3B, 0x7E, 0x7D, 0x1A, 0x7E, 0x7B, 0x36, 0x21, 0x7E, 0x7D, 0x31, 0x1A, 0x32 }; static const int32_t offsets[] = {0,0,0,0, 1,1,1, 3,3,3,3, 4,4,4, 5, 7,}; static const UChar sampleText1[] = { 0x4e00, 0x0035, 0x04e01, 0x0031, 0x0041, 0x0032}; static const uint8_t expected1[] = {0x7E, 0x7B, 0x52, 0x3B, 0x7E, 0x7D, 0x35, 0x7E, 0x7B, 0x36, 0x21, 0x7E, 0x7D, 0x31, 0x41, 0x32 }; static const int32_t offsets1[] = {2,6,9,13,14,15 }; /*hz*/ if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "HZ", UCNV_FROM_U_CALLBACK_SUBSTITUTE,NULL , TRUE)) log_err("u-> not match.\n"); if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "HZ", UCNV_FROM_U_CALLBACK_SUBSTITUTE,offsets , TRUE)) log_err("u-> not match.\n"); if(!testConvertToU(expected1, sizeof(expected1), sampleText1, sizeof(sampleText1)/sizeof(sampleText1[0]), "hz",UCNV_TO_U_CALLBACK_SUBSTITUTE , offsets1, TRUE)) log_err("hz -> did not match.\n"); } /*UTF-8*/ log_verbose("Testing for UTF8\n"); { static const UChar sampleText[] = { 0x4e00, 0x0701, 0x0031, 0xbfc1, 0xd801, 0xdc01, 0x0032}; int32_t offsets[]={0x00, 0x00, 0x00, 0x01, 0x01, 0x02, 0x03, 0x03, 0x03, 0x04, 0x04, 0x04, 0x04, 0x06 }; static const uint8_t expected[] = {0xe4, 0xb8, 0x80, 0xdc, 0x81, 0x31, 0xeb, 0xbf, 0x81, 0xF0, 0x90, 0x90, 0x81, 0x32}; static const int32_t fromOffsets[] = { 0x0000, 0x0003, 0x0005, 0x0006, 0x0009, 0x0009, 0x000D }; /*UTF-8*/ if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "UTF8", UCNV_FROM_U_CALLBACK_SUBSTITUTE,offsets , TRUE)) log_err("u-> UTF8 with offsets and flush true did not match.\n"); if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "UTF8", UCNV_FROM_U_CALLBACK_SUBSTITUTE,NULL , TRUE)) log_err("u-> UTF8 with offsets and flush true did not match.\n"); if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "UTF8", UCNV_FROM_U_CALLBACK_SUBSTITUTE,offsets , TRUE)) log_err("u-> UTF8 with offsets and flush true did not match.\n"); if(!testConvertFromU(sampleText, sizeof(sampleText)/sizeof(sampleText[0]), expected, sizeof(expected), "UTF8", UCNV_FROM_U_CALLBACK_SUBSTITUTE,NULL , TRUE)) log_err("u-> UTF8 with offsets and flush true did not match.\n"); if(!testConvertToU(expected, sizeof(expected), sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "UTF8",UCNV_TO_U_CALLBACK_SUBSTITUTE , NULL, TRUE)) log_err("UTF8 -> did not match.\n"); if(!testConvertToU(expected, sizeof(expected), sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "UTF8", UCNV_TO_U_CALLBACK_SUBSTITUTE , NULL, TRUE)) log_err("UTF8 -> did not match.\n"); if(!testConvertToU(expected, sizeof(expected), sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "UTF8",UCNV_TO_U_CALLBACK_SUBSTITUTE , fromOffsets, TRUE)) log_err("UTF8 -> did not match.\n"); if(!testConvertToU(expected, sizeof(expected), sampleText, sizeof(sampleText)/sizeof(sampleText[0]), "UTF8", UCNV_TO_U_CALLBACK_SUBSTITUTE , fromOffsets, TRUE)) log_err("UTF8 -> did not match.\n"); } }