/* ********************************************************************** * Copyright (C) 1998-2011, International Business Machines Corporation * and others. All Rights Reserved. ********************************************************************** * * File cstrtest.c * * Modification History: * * Date Name Description * 07/13/2000 Madhu created ******************************************************************************* */ #include "unicode/ustring.h" #include "unicode/ucnv.h" #include "cstring.h" #include "uinvchar.h" #include "cintltst.h" #include "cmemory.h" #define LENGTHOF(array) (int32_t)(sizeof(array)/sizeof((array)[0])) static void TestAPI(void); void addCStringTest(TestNode** root); static void TestInvariant(void); static void TestCompareInvEbcdicAsAscii(void); void addCStringTest(TestNode** root) { addTest(root, &TestAPI, "tsutil/cstrtest/TestAPI"); addTest(root, &TestInvariant, "tsutil/cstrtest/TestInvariant"); addTest(root, &TestCompareInvEbcdicAsAscii, "tsutil/cstrtest/TestCompareInvEbcdicAsAscii"); } static void TestAPI(void) { int32_t intValue=0; char src[30]="HELLO THERE", dest[30]; static const char *const abc="abcdefghijklmnopqrstuvwxyz", *const ABC="ABCDEFGHIJKLMNOPQRSTUVWXYZ"; const char *temp; int32_t i; log_verbose("Testing uprv_tolower() and uprv_toupper()\n"); for(i=0; i<=26; ++i) { dest[i]=uprv_tolower(abc[i]); } if(0!=strcmp(abc, dest)) { log_err("uprv_tolower(abc) failed\n"); } for(i=0; i<=26; ++i) { dest[i]=uprv_tolower(ABC[i]); } if(0!=strcmp(abc, dest)) { log_err("uprv_tolower(ABC) failed\n"); } for(i=0; i<=26; ++i) { dest[i]=uprv_toupper(abc[i]); } if(0!=strcmp(ABC, dest)) { log_err("uprv_toupper(abc) failed\n"); } for(i=0; i<=26; ++i) { dest[i]=uprv_toupper(ABC[i]); } if(0!=strcmp(ABC, dest)) { log_err("uprv_toupper(ABC) failed\n"); } log_verbose("Testing the API in cstring\n"); T_CString_toLowerCase(src); if(uprv_strcmp(src, "hello there") != 0){ log_err("FAIL: *** T_CString_toLowerCase() failed. Expected: \"hello there\", Got: \"%s\"\n", src); } T_CString_toUpperCase(src); if(uprv_strcmp(src, "HELLO THERE") != 0){ log_err("FAIL: *** T_CString_toUpperCase() failed. Expected: \"HELLO THERE\", Got: \"%s\"\n", src); } intValue=T_CString_stringToInteger("34556", 10); if(intValue != 34556){ log_err("FAIL: ****T_CString_stringToInteger(\"34556\", 10) failed. Expected: 34556, Got: %d\n", intValue); } intValue=T_CString_stringToInteger("100", 16); if(intValue != 256){ log_err("FAIL: ****T_CString_stringToInteger(\"100\", 16) failed. Expected: 256, Got: %d\n", intValue); } i = T_CString_integerToString(src, 34556, 10); if(uprv_strcmp(src, "34556") != 0 || i != 5){ log_err("FAIL: ****integerToString(src, 34566, 10); failed. Expected: \"34556\", Got: %s\n", src); } i = T_CString_integerToString(src, 431, 16); if(uprv_stricmp(src, "1AF") != 0 || i != 3){ log_err("FAIL: ****integerToString(src, 431, 16); failed. Expected: \"1AF\", Got: %s\n", src); } i = T_CString_int64ToString(src, U_INT64_MAX, 10); if(uprv_strcmp(src, "9223372036854775807") != 0 || i != 19){ log_err("FAIL: ****integerToString(src, 9223372036854775807, 10); failed. Got: %s\n", src); } i = T_CString_int64ToString(src, U_INT64_MAX, 16); if(uprv_stricmp(src, "7FFFFFFFFFFFFFFF") != 0 || i != 16){ log_err("FAIL: ****integerToString(src, 7FFFFFFFFFFFFFFF, 16); failed. Got: %s\n", src); } uprv_strcpy(src, "this is lower case"); if(uprv_stricmp(src, "THIS is lower CASE") != 0){ log_err("FAIL: *****uprv_stricmp() failed."); } if((intValue=uprv_stricmp(NULL, "first string is null") )!= -1){ log_err("FAIL: uprv_stricmp() where the first string is null failed. Expected: -1, returned %d\n", intValue); } if((intValue=uprv_stricmp("second string is null", NULL)) != 1){ log_err("FAIL: uprv_stricmp() where the second string is null failed. Expected: 1, returned %d\n", intValue); } if((intValue=uprv_stricmp(NULL, NULL)) != 0){ log_err("FAIL: uprv_stricmp(NULL, NULL) failed. Expected: 0, returned %d\n", intValue);; } if((intValue=uprv_stricmp("", "")) != 0){ log_err("FAIL: uprv_stricmp(\"\", \"\") failed. Expected: 0, returned %d\n", intValue);; } if((intValue=uprv_stricmp("", "abc")) != -1){ log_err("FAIL: uprv_stricmp(\"\", \"abc\") failed. Expected: -1, returned %d\n", intValue); } if((intValue=uprv_stricmp("abc", "")) != 1){ log_err("FAIL: uprv_stricmp(\"abc\", \"\") failed. Expected: 1, returned %d\n", intValue); } temp=uprv_strdup("strdup"); if(uprv_strcmp(temp, "strdup") !=0 ){ log_err("FAIL: uprv_strdup() failed. Expected: \"strdup\", Got: %s\n", temp); } uprv_free((char *)temp); uprv_strcpy(src, "this is lower case"); if(uprv_strnicmp(src, "THIS", 4 ) != 0){ log_err("FAIL: *****uprv_strnicmp() failed."); } if((intValue=uprv_strnicmp(NULL, "first string is null", 10) )!= -1){ log_err("FAIL: uprv_strnicmp() where the first string is null failed. Expected: -1, returned %d\n", intValue); } if((intValue=uprv_strnicmp("second string is null", NULL, 10)) != 1){ log_err("FAIL: uprv_strnicmp() where the second string is null failed. Expected: 1, returned %d\n", intValue); } if((intValue=uprv_strnicmp(NULL, NULL, 10)) != 0){ log_err("FAIL: uprv_strnicmp(NULL, NULL, 10) failed. Expected: 0, returned %d\n", intValue);; } if((intValue=uprv_strnicmp("", "", 10)) != 0){ log_err("FAIL: uprv_strnicmp(\"\", \"\") failed. Expected: 0, returned %d\n", intValue);; } if((intValue=uprv_strnicmp("", "abc", 10)) != -1){ log_err("FAIL: uprv_stricmp(\"\", \"abc\", 10) failed. Expected: -1, returned %d\n", intValue); } if((intValue=uprv_strnicmp("abc", "", 10)) != 1){ log_err("FAIL: uprv_strnicmp(\"abc\", \"\", 10) failed. Expected: 1, returned %d\n", intValue); } } /* test invariant-character handling */ static void TestInvariant() { /* all invariant graphic chars and some control codes (not \n!) */ const char invariantChars[]= "\t\r \"%&'()*+,-./" "0123456789:;<=>?" "ABCDEFGHIJKLMNOPQRSTUVWXYZ_" "abcdefghijklmnopqrstuvwxyz"; const UChar invariantUChars[]={ 9, 0xd, 0x20, 0x22, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c, 0x2d, 0x2e, 0x2f, 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, 0x3a, 0x3b, 0x3c, 0x3d, 0x3e, 0x3f, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f, 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 0x5a, 0x5f, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, 0x6a, 0x6b, 0x6c, 0x6d, 0x6e, 0x6f, 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, 0x7a, 0 }; const char variantChars[]="\n!#$@[\\]^`{|}~"; const UChar variantUChars[]={ 0x0a, 0x21, 0x23, 0x24, 0x40, 0x5b, 0x5c, 0x5d, 0x5e, 0x60, 0x7b, 0x7c, 0x7d, 0x7e, 0 }; const UChar nonASCIIUChars[]={ 0x80, 0xa0, 0x900, 0xff51 }; UChar us[120]; char cs[120]; int32_t i, length; /* make sure that all invariant characters convert both ways */ length=sizeof(invariantChars); u_charsToUChars(invariantChars, us, length); if(u_strcmp(us, invariantUChars)!=0) { log_err("u_charsToUChars(invariantChars) failed\n"); } u_UCharsToChars(invariantUChars, cs, length); if(strcmp(cs, invariantChars)!=0) { log_err("u_UCharsToChars(invariantUChars) failed\n"); } /* * make sure that variant characters convert from source code literals to Unicode * but not back to char * */ length=sizeof(variantChars); u_charsToUChars(variantChars, us, length); if(u_strcmp(us, variantUChars)!=0) { log_err("u_charsToUChars(variantChars) failed\n"); } #ifdef NDEBUG /* * Test u_UCharsToChars(variantUChars) only in release mode because it will * cause an assertion failure in debug builds. */ u_UCharsToChars(variantUChars, cs, length); for(i=0; i<length; ++i) { if(cs[i]!=0) { log_err("u_UCharsToChars(variantUChars) converted the %d-th character to %02x instead of 00\n", i, cs[i]); } } #endif /* * Verify that invariant characters roundtrip from Unicode to the * default converter and back. */ { UConverter *cnv; UErrorCode errorCode; errorCode=U_ZERO_ERROR; cnv=ucnv_open(NULL, &errorCode); if(U_FAILURE(errorCode)) { log_err("unable to open the default converter\n"); } else { length=ucnv_fromUChars(cnv, cs, sizeof(cs), invariantUChars, -1, &errorCode); if(U_FAILURE(errorCode)) { log_err("ucnv_fromUChars(invariantUChars) failed - %s\n", u_errorName(errorCode)); } else if(length!=sizeof(invariantChars)-1 || strcmp(cs, invariantChars)!=0) { log_err("ucnv_fromUChars(invariantUChars) failed\n"); } errorCode=U_ZERO_ERROR; length=ucnv_toUChars(cnv, us, LENGTHOF(us), invariantChars, -1, &errorCode); if(U_FAILURE(errorCode)) { log_err("ucnv_toUChars(invariantChars) failed - %s\n", u_errorName(errorCode)); } else if(length!=LENGTHOF(invariantUChars)-1 || u_strcmp(us, invariantUChars)!=0) { log_err("ucnv_toUChars(invariantChars) failed\n"); } ucnv_close(cnv); } } /* API tests */ if(!uprv_isInvariantString(invariantChars, -1)) { log_err("uprv_isInvariantString(invariantChars) failed\n"); } if(!uprv_isInvariantUString(invariantUChars, -1)) { log_err("uprv_isInvariantUString(invariantUChars) failed\n"); } if(!uprv_isInvariantString(invariantChars+strlen(invariantChars), 1)) { log_err("uprv_isInvariantString(\"\\0\") failed\n"); } for(i=0; i<(sizeof(variantChars)-1); ++i) { if(uprv_isInvariantString(variantChars+i, 1)) { log_err("uprv_isInvariantString(variantChars[%d]) failed\n", i); } if(uprv_isInvariantUString(variantUChars+i, 1)) { log_err("uprv_isInvariantUString(variantUChars[%d]) failed\n", i); } } for(i=0; i<LENGTHOF(nonASCIIUChars); ++i) { if(uprv_isInvariantUString(nonASCIIUChars+i, 1)) { log_err("uprv_isInvariantUString(nonASCIIUChars[%d]) failed\n", i); } } } static int32_t getSign(int32_t n) { if(n<0) { return -1; } else if(n==0) { return 0; } else { return 1; } } static void TestCompareInvEbcdicAsAscii() { static const char *const invStrings[][2]={ /* invariant-character strings in ascending ASCII order */ /* EBCDIC native */ { "", "" }, { "\x6c", "%" }, { "\xf0", "0" }, { "\xf0\xf0", "00" }, { "\xf0\xf0\x81", "00a" }, { "\x7e", "=" }, { "\xc1", "A" }, { "\xc1\xf0\xf0", "A00" }, { "\xc1\xf0\xf0", "A00" }, { "\xc1\xc1", "AA" }, { "\xc1\xc1\xf0", "AA0" }, { "\x6d", "_" }, { "\x81", "a" }, { "\x81\xf0\xf0", "a00" }, { "\x81\xf0\xf0", "a00" }, { "\x81\x81", "aa" }, { "\x81\x81\xf0", "aa0" }, { "\x81\x81\x81", "aaa" }, { "\x81\x81\x82", "aab" } }; int32_t i; for(i=1; i<LENGTHOF(invStrings); ++i) { int32_t diff1, diff2; /* compare previous vs. current */ diff1=getSign(uprv_compareInvEbcdicAsAscii(invStrings[i-1][0], invStrings[i][0])); if(diff1>0 || (diff1==0 && 0!=uprv_strcmp(invStrings[i-1][0], invStrings[i][0]))) { log_err("uprv_compareInvEbcdicAsAscii(%s, %s)=%hd is wrong\n", invStrings[i-1][1], invStrings[i][1], (short)diff1); } /* compare current vs. previous, should be inverse diff */ diff2=getSign(uprv_compareInvEbcdicAsAscii(invStrings[i][0], invStrings[i-1][0])); if(diff2!=-diff1) { log_err("uprv_compareInvEbcdicAsAscii(%s, %s)=%hd is wrong\n", invStrings[i][1], invStrings[i-1][1], (short)diff2); } } }