/* $Id: tstUtf8.cpp 57358 2015-08-14 15:16:38Z vboxsync $ */ /** @file * IPRT Testcase - UTF-8 and UTF-16 string conversions. */ /* * Copyright (C) 2006-2015 Oracle Corporation * * This file is part of VirtualBox Open Source Edition (OSE), as * available from http://www.virtualbox.org. This file is free software; * you can redistribute it and/or modify it under the terms of the GNU * General Public License (GPL) as published by the Free Software * Foundation, in version 2 as it comes in the "COPYING" file of the * VirtualBox OSE distribution. VirtualBox OSE is distributed in the * hope that it will be useful, but WITHOUT ANY WARRANTY of any kind. * * The contents of this file may alternatively be used under the terms * of the Common Development and Distribution License Version 1.0 * (CDDL) only, as it comes in the "COPYING.CDDL" file of the * VirtualBox OSE distribution, in which case the provisions of the * CDDL are applicable instead of those of the GPL. * * You may elect to license modified versions of this file under the * terms and conditions of either the GPL or the CDDL or both. */ /********************************************************************************************************************************* * Header Files * *********************************************************************************************************************************/ #include #include #include #include #include #include #include #include #include #include #include /** * Generate a random codepoint for simple UTF-16 encoding. */ static RTUTF16 GetRandUtf16(void) { RTUTF16 wc; do { wc = (RTUTF16)RTRandU32Ex(1, 0xfffd); } while (wc >= 0xd800 && wc <= 0xdfff); return wc; } /** * */ static void test1(RTTEST hTest) { static const char s_szBadString1[] = "Bad \xe0\x13\x0"; static const char s_szBadString2[] = "Bad \xef\xbf\xc3"; int rc; char *pszUtf8; char *pszCurrent; PRTUTF16 pwsz; PRTUTF16 pwszRand; /* * Invalid UTF-8 to UCS-2 test. */ RTTestSub(hTest, "Feeding bad UTF-8 to RTStrToUtf16"); rc = RTStrToUtf16(s_szBadString1, &pwsz); RTTEST_CHECK_MSG(hTest, rc == VERR_NO_TRANSLATION || rc == VERR_INVALID_UTF8_ENCODING, (hTest, "Conversion of first bad UTF-8 string to UTF-16 apparently succeeded. It shouldn't. rc=%Rrc\n", rc)); rc = RTStrToUtf16(s_szBadString2, &pwsz); RTTEST_CHECK_MSG(hTest, rc == VERR_NO_TRANSLATION || rc == VERR_INVALID_UTF8_ENCODING, (hTest, "Conversion of second bad UTF-8 strings to UTF-16 apparently succeeded. It shouldn't. rc=%Rrc\n", rc)); /* * Test current CP conversion. */ RTTestSub(hTest, "Rand UTF-16 -> UTF-8 -> CP -> UTF-8"); pwszRand = (PRTUTF16)RTMemAlloc(31 * sizeof(*pwsz)); for (int i = 0; i < 30; i++) pwszRand[i] = GetRandUtf16(); pwszRand[30] = 0; rc = RTUtf16ToUtf8(pwszRand, &pszUtf8); if (rc == VINF_SUCCESS) { rc = RTStrUtf8ToCurrentCP(&pszCurrent, pszUtf8); if (rc == VINF_SUCCESS) { rc = RTStrCurrentCPToUtf8(&pszUtf8, pszCurrent); if (rc == VINF_SUCCESS) RTTestPassed(hTest, "Random UTF-16 -> UTF-8 -> Current -> UTF-8 successful.\n"); else RTTestFailed(hTest, "%d: The third part of random UTF-16 -> UTF-8 -> Current -> UTF-8 failed with return value %Rrc.", __LINE__, rc); } else if (rc == VERR_NO_TRANSLATION) RTTestPassed(hTest, "The second part of random UTF-16 -> UTF-8 -> Current -> UTF-8 returned VERR_NO_TRANSLATION. This is probably as it should be.\n"); else if (rc == VWRN_NO_TRANSLATION) RTTestPassed(hTest, "The second part of random UTF-16 -> UTF-8 -> Current -> UTF-8 returned VWRN_NO_TRANSLATION. This is probably as it should be.\n"); else RTTestFailed(hTest, "%d: The second part of random UTF-16 -> UTF-8 -> Current -> UTF-8 failed with return value %Rrc.", __LINE__, rc); } else RTTestFailed(hTest, "%d: The first part of random UTF-16 -> UTF-8 -> Current -> UTF-8 failed with return value %Rrc.", __LINE__, rc); /* * Generate a new random string. */ RTTestSub(hTest, "Random UTF-16 -> UTF-8 -> UTF-16"); pwszRand = (PRTUTF16)RTMemAlloc(31 * sizeof(*pwsz)); for (int i = 0; i < 30; i++) pwszRand[i] = GetRandUtf16(); pwszRand[30] = 0; rc = RTUtf16ToUtf8(pwszRand, &pszUtf8); if (rc == VINF_SUCCESS) { rc = RTStrToUtf16(pszUtf8, &pwsz); if (rc == VINF_SUCCESS) { int i; for (i = 0; pwszRand[i] == pwsz[i] && pwsz[i] != 0; i++) /* nothing */; if (pwszRand[i] == pwsz[i] && pwsz[i] == 0) RTTestPassed(hTest, "Random UTF-16 -> UTF-8 -> UTF-16 successful.\n"); else { RTTestFailed(hTest, "%d: The second part of random UTF-16 -> UTF-8 -> UTF-16 failed.", __LINE__); RTTestPrintf(hTest, RTTESTLVL_FAILURE, "First differing character is at position %d and has the value %x.\n", i, pwsz[i]); } } else RTTestFailed(hTest, "%d: The second part of random UTF-16 -> UTF-8 -> UTF-16 failed with return value %Rrc.", __LINE__, rc); } else RTTestFailed(hTest, "%d: The first part of random UTF-16 -> UTF-8 -> UTF-16 failed with return value %Rrc.", __LINE__, rc); /* * Generate yet another random string and convert it to a buffer. */ RTTestSub(hTest, "Random RTUtf16ToUtf8Ex + RTStrToUtf16"); pwszRand = (PRTUTF16)RTMemAlloc(31 * sizeof(*pwsz)); for (int i = 0; i < 30; i++) pwszRand[i] = GetRandUtf16(); pwszRand[30] = 0; char szUtf8Array[120]; char *pszUtf8Array = szUtf8Array; rc = RTUtf16ToUtf8Ex(pwszRand, RTSTR_MAX, &pszUtf8Array, 120, NULL); if (rc == 0) { rc = RTStrToUtf16(pszUtf8Array, &pwsz); if (rc == 0) { int i; for (i = 0; pwszRand[i] == pwsz[i] && pwsz[i] != 0; i++) ; if (pwsz[i] == 0 && i >= 8) RTTestPassed(hTest, "Random UTF-16 -> fixed length UTF-8 -> UTF-16 successful.\n"); else { RTTestFailed(hTest, "%d: Incorrect conversion of UTF-16 -> fixed length UTF-8 -> UTF-16.\n", __LINE__); RTTestPrintf(hTest, RTTESTLVL_FAILURE, "First differing character is at position %d and has the value %x.\n", i, pwsz[i]); } } else RTTestFailed(hTest, "%d: The second part of random UTF-16 -> fixed length UTF-8 -> UTF-16 failed with return value %Rrc.\n", __LINE__, rc); } else RTTestFailed(hTest, "%d: The first part of random UTF-16 -> fixed length UTF-8 -> UTF-16 failed with return value %Rrc.\n", __LINE__, rc); /* * And again. */ RTTestSub(hTest, "Random RTUtf16ToUtf8 + RTStrToUtf16Ex"); pwszRand = (PRTUTF16)RTMemAlloc(31 * sizeof(*pwsz)); for (int i = 0; i < 30; i++) pwszRand[i] = GetRandUtf16(); pwszRand[30] = 0; RTUTF16 wszBuf[70]; PRTUTF16 pwsz2Buf = wszBuf; rc = RTUtf16ToUtf8(pwszRand, &pszUtf8); if (rc == 0) { rc = RTStrToUtf16Ex(pszUtf8, RTSTR_MAX, &pwsz2Buf, 70, NULL); if (rc == 0) { int i; for (i = 0; pwszRand[i] == pwsz2Buf[i] && pwsz2Buf[i] != 0; i++) ; if (pwszRand[i] == 0 && pwsz2Buf[i] == 0) RTTestPassed(hTest, "Random UTF-16 -> UTF-8 -> fixed length UTF-16 successful.\n"); else { RTTestFailed(hTest, "%d: Incorrect conversion of random UTF-16 -> UTF-8 -> fixed length UTF-16.\n", __LINE__); RTTestPrintf(hTest, RTTESTLVL_FAILURE, "First differing character is at position %d and has the value %x.\n", i, pwsz2Buf[i]); } } else RTTestFailed(hTest, "%d: The second part of random UTF-16 -> UTF-8 -> fixed length UTF-16 failed with return value %Rrc.\n", __LINE__, rc); } else RTTestFailed(hTest, "%d: The first part of random UTF-16 -> UTF-8 -> fixed length UTF-16 failed with return value %Rrc.\n", __LINE__, rc); pwszRand = (PRTUTF16)RTMemAlloc(31 * sizeof(*pwsz)); for (int i = 0; i < 30; i++) pwszRand[i] = GetRandUtf16(); pwszRand[30] = 0; rc = RTUtf16ToUtf8Ex(pwszRand, RTSTR_MAX, &pszUtf8Array, 20, NULL); if (rc == VERR_BUFFER_OVERFLOW) RTTestPassed(hTest, "Random UTF-16 -> fixed length UTF-8 with too short buffer successfully rejected.\n"); else RTTestFailed(hTest, "%d: Random UTF-16 -> fixed length UTF-8 with too small buffer returned value %d instead of VERR_BUFFER_OVERFLOW.\n", __LINE__, rc); /* * last time... */ RTTestSub(hTest, "Random RTUtf16ToUtf8 + RTStrToUtf16Ex"); pwszRand = (PRTUTF16)RTMemAlloc(31 * sizeof(*pwsz)); for (int i = 0; i < 30; i++) pwszRand[i] = GetRandUtf16(); pwszRand[30] = 0; rc = RTUtf16ToUtf8(pwszRand, &pszUtf8); if (rc == VINF_SUCCESS) { rc = RTStrToUtf16Ex(pszUtf8, RTSTR_MAX, &pwsz2Buf, 20, NULL); if (rc == VERR_BUFFER_OVERFLOW) RTTestPassed(hTest, "Random UTF-16 -> UTF-8 -> fixed length UTF-16 with too short buffer successfully rejected.\n"); else RTTestFailed(hTest, "%d: The second part of random UTF-16 -> UTF-8 -> fixed length UTF-16 with too short buffer returned value %Rrc instead of VERR_BUFFER_OVERFLOW.\n", __LINE__, rc); } else RTTestFailed(hTest, "%d:The first part of random UTF-16 -> UTF-8 -> fixed length UTF-16 failed with return value %Rrc.\n", __LINE__, rc); RTTestSubDone(hTest); } static RTUNICP g_uszAll[0x110000 - 1 - 0x800 - 2 + 1]; static RTUTF16 g_wszAll[0xfffe - (0xe000 - 0xd800) + (0x110000 - 0x10000) * 2]; static char g_szAll[0x7f + (0x800 - 0x80) * 2 + (0xfffe - 0x800 - (0xe000 - 0xd800))* 3 + (0x110000 - 0x10000) * 4 + 1]; static void whereami(int cBits, size_t off) { if (cBits == 8) { if (off < 0x7f) RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "UTF-8 U+%#x\n", off + 1); else if (off < 0xf7f) RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "UTF-8 U+%#x\n", (off - 0x7f) / 2 + 0x80); else if (off < 0x27f7f) RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "UTF-8 U+%#x\n", (off - 0xf7f) / 3 + 0x800); else if (off < 0x2df79) RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "UTF-8 U+%#x\n", (off - 0x27f7f) / 3 + 0xe000); else if (off < 0x42df79) RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "UTF-8 U+%#x\n", (off - 0x2df79) / 4 + 0x10000); else RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "UTF-8 ???\n"); } else if (cBits == 16) { if (off < 0xd7ff*2) RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "UTF-16 U+%#x\n", off / 2 + 1); else if (off < 0xf7fd*2) RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "UTF-16 U+%#x\n", (off - 0xd7ff*2) / 2 + 0xe000); else if (off < 0x20f7fd) RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "UTF-16 U+%#x\n", (off - 0xf7fd*2) / 4 + 0x10000); else RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "UTF-16 ???\n"); } else { if (off < (0xd800 - 1) * sizeof(RTUNICP)) RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "RTUNICP U+%#x\n", off / sizeof(RTUNICP) + 1); else if (off < (0xfffe - 0x800 - 1) * sizeof(RTUNICP)) RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "RTUNICP U+%#x\n", off / sizeof(RTUNICP) + 0x800 + 1); else RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "RTUNICP U+%#x\n", off / sizeof(RTUNICP) + 0x800 + 1 + 2); } } int mymemcmp(const void *pv1, const void *pv2, size_t cb, int cBits) { const uint8_t *pb1 = (const uint8_t *)pv1; const uint8_t *pb2 = (const uint8_t *)pv2; for (size_t off = 0; off < cb; off++) { if (pb1[off] != pb2[off]) { RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "mismatch at %#x: ", off); whereami(cBits, off); if (off > 0) RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, " %#x: %02x != %02x!\n", off-1, pb1[off-1], pb2[off-1]); RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, "*%#x: %02x != %02x!\n", off, pb1[off], pb2[off]); for (size_t i = 1; i < 10; i++) if (off + i < cb) RTTestPrintf(NIL_RTTEST, RTTESTLVL_FAILURE, " %#x: %02x != %02x!\n", off+i, pb1[off+i], pb2[off+i]); return 1; } } return 0; } void InitStrings() { /* * Generate unicode string containing all the legal UTF-16 codepoints, both UTF-16 and UTF-8 version. */ /* the simple code point array first */ unsigned i = 0; RTUNICP uc = 1; while (uc < 0xd800) g_uszAll[i++] = uc++; uc = 0xe000; while (uc < 0xfffe) g_uszAll[i++] = uc++; uc = 0x10000; while (uc < 0x110000) g_uszAll[i++] = uc++; g_uszAll[i++] = 0; Assert(RT_ELEMENTS(g_uszAll) == i); /* the utf-16 one */ i = 0; uc = 1; //RTPrintf("tstUtf8: %#x=%#x", i, uc); while (uc < 0xd800) g_wszAll[i++] = uc++; uc = 0xe000; //RTPrintf(" %#x=%#x", i, uc); while (uc < 0xfffe) g_wszAll[i++] = uc++; uc = 0x10000; //RTPrintf(" %#x=%#x", i, uc); while (uc < 0x110000) { g_wszAll[i++] = 0xd800 | ((uc - 0x10000) >> 10); g_wszAll[i++] = 0xdc00 | ((uc - 0x10000) & 0x3ff); uc++; } //RTPrintf(" %#x=%#x\n", i, uc); g_wszAll[i++] = '\0'; Assert(RT_ELEMENTS(g_wszAll) == i); /* * The utf-8 one */ i = 0; uc = 1; //RTPrintf("tstUtf8: %#x=%#x", i, uc); while (uc < 0x80) g_szAll[i++] = uc++; //RTPrintf(" %#x=%#x", i, uc); while (uc < 0x800) { g_szAll[i++] = 0xc0 | (uc >> 6); g_szAll[i++] = 0x80 | (uc & 0x3f); Assert(!((uc >> 6) & ~0x1f)); uc++; } //RTPrintf(" %#x=%#x", i, uc); while (uc < 0xd800) { g_szAll[i++] = 0xe0 | (uc >> 12); g_szAll[i++] = 0x80 | ((uc >> 6) & 0x3f); g_szAll[i++] = 0x80 | (uc & 0x3f); Assert(!((uc >> 12) & ~0xf)); uc++; } uc = 0xe000; //RTPrintf(" %#x=%#x", i, uc); while (uc < 0xfffe) { g_szAll[i++] = 0xe0 | (uc >> 12); g_szAll[i++] = 0x80 | ((uc >> 6) & 0x3f); g_szAll[i++] = 0x80 | (uc & 0x3f); Assert(!((uc >> 12) & ~0xf)); uc++; } uc = 0x10000; //RTPrintf(" %#x=%#x", i, uc); while (uc < 0x110000) { g_szAll[i++] = 0xf0 | (uc >> 18); g_szAll[i++] = 0x80 | ((uc >> 12) & 0x3f); g_szAll[i++] = 0x80 | ((uc >> 6) & 0x3f); g_szAll[i++] = 0x80 | (uc & 0x3f); Assert(!((uc >> 18) & ~0x7)); uc++; } //RTPrintf(" %#x=%#x\n", i, uc); g_szAll[i++] = '\0'; Assert(RT_ELEMENTS(g_szAll) == i); } void test2(RTTEST hTest) { /* * Convert to UTF-8 and back. */ RTTestSub(hTest, "UTF-16 -> UTF-8 -> UTF-16"); char *pszUtf8; int rc = RTUtf16ToUtf8(&g_wszAll[0], &pszUtf8); if (rc == VINF_SUCCESS) { pszUtf8[0] = 1; if (mymemcmp(pszUtf8, g_szAll, sizeof(g_szAll), 8)) RTTestFailed(hTest, "UTF-16 -> UTF-8 mismatch!"); PRTUTF16 pwszUtf16; rc = RTStrToUtf16(pszUtf8, &pwszUtf16); if (rc == VINF_SUCCESS) { if (mymemcmp(pwszUtf16, g_wszAll, sizeof(g_wszAll), 16)) RTTestFailed(hTest, "UTF-8 -> UTF-16 failed compare!"); RTUtf16Free(pwszUtf16); } else RTTestFailed(hTest, "UTF-8 -> UTF-16 failed, rc=%Rrc.", rc); RTStrFree(pszUtf8); } else RTTestFailed(hTest, "UTF-16 -> UTF-8 failed, rc=%Rrc.", rc); /* * Convert to UTF-16 and back. (just in case the above test fails) */ RTTestSub(hTest, "UTF-8 -> UTF-16 -> UTF-8"); PRTUTF16 pwszUtf16; rc = RTStrToUtf16(&g_szAll[0], &pwszUtf16); if (rc == VINF_SUCCESS) { if (mymemcmp(pwszUtf16, g_wszAll, sizeof(g_wszAll), 16)) RTTestFailed(hTest, "UTF-8 -> UTF-16 failed compare!"); rc = RTUtf16ToUtf8(pwszUtf16, &pszUtf8); if (rc == VINF_SUCCESS) { if (mymemcmp(pszUtf8, g_szAll, sizeof(g_szAll), 8)) RTTestFailed(hTest, "UTF-16 -> UTF-8 failed compare!"); RTStrFree(pszUtf8); } else RTTestFailed(hTest, "UTF-16 -> UTF-8 failed, rc=%Rrc.", rc); RTUtf16Free(pwszUtf16); } else RTTestFailed(hTest, "UTF-8 -> UTF-16 failed, rc=%Rrc.", rc); /* * Convert UTF-8 to CPs. */ RTTestSub(hTest, "UTF-8 -> UNI -> UTF-8"); PRTUNICP paCps; rc = RTStrToUni(g_szAll, &paCps); if (rc == VINF_SUCCESS) { if (mymemcmp(paCps, g_uszAll, sizeof(g_uszAll), 32)) RTTestFailed(hTest, "UTF-8 -> UTF-16 failed, rc=%Rrc.", rc); size_t cCps; rc = RTStrToUniEx(g_szAll, RTSTR_MAX, &paCps, RT_ELEMENTS(g_uszAll), &cCps); if (rc == VINF_SUCCESS) { if (cCps != RT_ELEMENTS(g_uszAll) - 1) RTTestFailed(hTest, "wrong Code Point count %zu, expected %zu\n", cCps, RT_ELEMENTS(g_uszAll) - 1); } else RTTestFailed(hTest, "UTF-8 -> Code Points failed, rc=%Rrc.\n", rc); /** @todo RTCpsToUtf8 or something. */ } else RTTestFailed(hTest, "UTF-8 -> Code Points failed, rc=%Rrc.\n", rc); /* * Check the various string lengths. */ RTTestSub(hTest, "Lengths"); size_t cuc1 = RTStrCalcUtf16Len(g_szAll); size_t cuc2 = RTUtf16Len(g_wszAll); if (cuc1 != cuc2) RTTestFailed(hTest, "cuc1=%zu != cuc2=%zu\n", cuc1, cuc2); //size_t cuc3 = RTUniLen(g_uszAll); /* * Enumerate the strings. */ RTTestSub(hTest, "Code Point Getters and Putters"); char *pszPut1Base = (char *)RTMemAlloc(sizeof(g_szAll)); AssertRelease(pszPut1Base); char *pszPut1 = pszPut1Base; PRTUTF16 pwszPut2Base = (PRTUTF16)RTMemAlloc(sizeof(g_wszAll)); AssertRelease(pwszPut2Base); PRTUTF16 pwszPut2 = pwszPut2Base; const char *psz1 = g_szAll; const char *psz2 = g_szAll; PCRTUTF16 pwsz3 = g_wszAll; PCRTUTF16 pwsz4 = g_wszAll; for (;;) { /* * getters */ RTUNICP uc1; rc = RTStrGetCpEx(&psz1, &uc1); if (RT_FAILURE(rc)) { RTTestFailed(hTest, "RTStrGetCpEx failed with rc=%Rrc at %.10Rhxs", rc, psz2); whereami(8, psz2 - &g_szAll[0]); break; } char *pszPrev1 = RTStrPrevCp(g_szAll, psz1); if (pszPrev1 != psz2) { RTTestFailed(hTest, "RTStrPrevCp returned %p expected %p!", pszPrev1, psz2); whereami(8, psz2 - &g_szAll[0]); break; } RTUNICP uc2 = RTStrGetCp(psz2); if (uc2 != uc1) { RTTestFailed(hTest, "RTStrGetCpEx and RTStrGetCp returned different CPs: %RTunicp != %RTunicp", uc2, uc1); whereami(8, psz2 - &g_szAll[0]); break; } psz2 = RTStrNextCp(psz2); if (psz2 != psz1) { RTTestFailed(hTest, "RTStrGetCpEx and RTStrGetNext returned different next pointer!"); whereami(8, psz2 - &g_szAll[0]); break; } RTUNICP uc3; rc = RTUtf16GetCpEx(&pwsz3, &uc3); if (RT_FAILURE(rc)) { RTTestFailed(hTest, "RTUtf16GetCpEx failed with rc=%Rrc at %.10Rhxs", rc, pwsz4); whereami(16, pwsz4 - &g_wszAll[0]); break; } if (uc3 != uc2) { RTTestFailed(hTest, "RTUtf16GetCpEx and RTStrGetCp returned different CPs: %RTunicp != %RTunicp", uc3, uc2); whereami(16, pwsz4 - &g_wszAll[0]); break; } RTUNICP uc4 = RTUtf16GetCp(pwsz4); if (uc3 != uc4) { RTTestFailed(hTest, "RTUtf16GetCpEx and RTUtf16GetCp returned different CPs: %RTunicp != %RTunicp", uc3, uc4); whereami(16, pwsz4 - &g_wszAll[0]); break; } pwsz4 = RTUtf16NextCp(pwsz4); if (pwsz4 != pwsz3) { RTTestFailed(hTest, "RTUtf16GetCpEx and RTUtf16GetNext returned different next pointer!"); whereami(8, pwsz4 - &g_wszAll[0]); break; } /* * putters */ pszPut1 = RTStrPutCp(pszPut1, uc1); if (pszPut1 - pszPut1Base != psz1 - &g_szAll[0]) { RTTestFailed(hTest, "RTStrPutCp is not at the same offset! %p != %p", pszPut1 - pszPut1Base, psz1 - &g_szAll[0]); whereami(8, psz2 - &g_szAll[0]); break; } pwszPut2 = RTUtf16PutCp(pwszPut2, uc3); if (pwszPut2 - pwszPut2Base != pwsz3 - &g_wszAll[0]) { RTTestFailed(hTest, "RTStrPutCp is not at the same offset! %p != %p", pwszPut2 - pwszPut2Base, pwsz3 - &g_wszAll[0]); whereami(8, pwsz4 - &g_wszAll[0]); break; } /* the end? */ if (!uc1) break; } /* check output if we seems to have made it thru it all. */ if (psz2 == &g_szAll[sizeof(g_szAll)]) { if (mymemcmp(pszPut1Base, g_szAll, sizeof(g_szAll), 8)) RTTestFailed(hTest, "RTStrPutCp encoded the string incorrectly."); if (mymemcmp(pwszPut2Base, g_wszAll, sizeof(g_wszAll), 16)) RTTestFailed(hTest, "RTUtf16PutCp encoded the string incorrectly."); } RTMemFree(pszPut1Base); RTMemFree(pwszPut2Base); RTTestSubDone(hTest); } /** * Check case insensitivity. */ void test3(RTTEST hTest) { RTTestSub(hTest, "Case Sensitivity"); if ( RTUniCpToLower('a') != 'a' || RTUniCpToLower('A') != 'a' || RTUniCpToLower('b') != 'b' || RTUniCpToLower('B') != 'b' || RTUniCpToLower('Z') != 'z' || RTUniCpToLower('z') != 'z' || RTUniCpToUpper('c') != 'C' || RTUniCpToUpper('C') != 'C' || RTUniCpToUpper('z') != 'Z' || RTUniCpToUpper('Z') != 'Z') RTTestFailed(hTest, "RTUniToUpper/Lower failed basic tests.\n"); if (RTUtf16ICmp(g_wszAll, g_wszAll)) RTTestFailed(hTest, "RTUtf16ICmp failed the basic test.\n"); if (RTUtf16Cmp(g_wszAll, g_wszAll)) RTTestFailed(hTest, "RTUtf16Cmp failed the basic test.\n"); static RTUTF16 s_wszTst1a[] = { 'a', 'B', 'c', 'D', 'E', 'f', 'g', 'h', 'i', 'j', 'K', 'L', 'm', 'N', 'o', 'P', 'q', 'r', 'S', 't', 'u', 'V', 'w', 'x', 'Y', 'Z', 0xc5, 0xc6, 0xf8, 0 }; static RTUTF16 s_wszTst1b[] = { 'A', 'B', 'c', 'd', 'e', 'F', 'G', 'h', 'i', 'J', 'k', 'l', 'M', 'n', 'O', 'p', 'Q', 'R', 's', 't', 'U', 'v', 'w', 'X', 'y', 'z', 0xe5, 0xe6, 0xd8, 0 }; if ( RTUtf16ICmp(s_wszTst1b, s_wszTst1b) || RTUtf16ICmp(s_wszTst1a, s_wszTst1a) || RTUtf16ICmp(s_wszTst1a, s_wszTst1b) || RTUtf16ICmp(s_wszTst1b, s_wszTst1a) ) RTTestFailed(hTest, "RTUtf16ICmp failed the alphabet test.\n"); if ( RTUtf16Cmp(s_wszTst1b, s_wszTst1b) || RTUtf16Cmp(s_wszTst1a, s_wszTst1a) || !RTUtf16Cmp(s_wszTst1a, s_wszTst1b) || !RTUtf16Cmp(s_wszTst1b, s_wszTst1a) ) RTTestFailed(hTest, "RTUtf16Cmp failed the alphabet test.\n"); RTTestSubDone(hTest); } /** * Test the RTStr*Cmp functions. */ void TstRTStrXCmp(RTTEST hTest) { #define CHECK_DIFF(expr, op) \ do \ { \ int iDiff = expr; \ if (!(iDiff op 0)) \ RTTestFailed(hTest, "%d: %d " #op " 0: %s\n", __LINE__, iDiff, #expr); \ } while (0) /** @todo test the non-ascii bits. */ RTTestSub(hTest, "RTStrCmp"); CHECK_DIFF(RTStrCmp(NULL, NULL), == ); CHECK_DIFF(RTStrCmp(NULL, ""), < ); CHECK_DIFF(RTStrCmp("", NULL), > ); CHECK_DIFF(RTStrCmp("", ""), == ); CHECK_DIFF(RTStrCmp("abcdef", "abcdef"), == ); CHECK_DIFF(RTStrCmp("abcdef", "abcde"), > ); CHECK_DIFF(RTStrCmp("abcde", "abcdef"), < ); CHECK_DIFF(RTStrCmp("abcdeg", "abcdef"), > ); CHECK_DIFF(RTStrCmp("abcdef", "abcdeg"), < ); CHECK_DIFF(RTStrCmp("abcdeF", "abcdef"), < ); CHECK_DIFF(RTStrCmp("abcdef", "abcdeF"), > ); RTTestSub(hTest, "RTStrNCmp"); CHECK_DIFF(RTStrNCmp(NULL, NULL, RTSTR_MAX), == ); CHECK_DIFF(RTStrNCmp(NULL, "", RTSTR_MAX), < ); CHECK_DIFF(RTStrNCmp("", NULL, RTSTR_MAX), > ); CHECK_DIFF(RTStrNCmp("", "", RTSTR_MAX), == ); CHECK_DIFF(RTStrNCmp("abcdef", "abcdef", RTSTR_MAX), == ); CHECK_DIFF(RTStrNCmp("abcdef", "abcde", RTSTR_MAX), > ); CHECK_DIFF(RTStrNCmp("abcde", "abcdef", RTSTR_MAX), < ); CHECK_DIFF(RTStrNCmp("abcdeg", "abcdef", RTSTR_MAX), > ); CHECK_DIFF(RTStrNCmp("abcdef", "abcdeg", RTSTR_MAX), < ); CHECK_DIFF(RTStrNCmp("abcdeF", "abcdef", RTSTR_MAX), < ); CHECK_DIFF(RTStrNCmp("abcdef", "abcdeF", RTSTR_MAX), > ); CHECK_DIFF(RTStrNCmp("abcdef", "fedcba", 0), ==); CHECK_DIFF(RTStrNCmp("abcdef", "abcdeF", 5), ==); CHECK_DIFF(RTStrNCmp("abcdef", "abcdeF", 6), > ); RTTestSub(hTest, "RTStrICmp"); CHECK_DIFF(RTStrICmp(NULL, NULL), == ); CHECK_DIFF(RTStrICmp(NULL, ""), < ); CHECK_DIFF(RTStrICmp("", NULL), > ); CHECK_DIFF(RTStrICmp("", ""), == ); CHECK_DIFF(RTStrICmp("abcdef", "abcdef"), == ); CHECK_DIFF(RTStrICmp("abcdef", "abcde"), > ); CHECK_DIFF(RTStrICmp("abcde", "abcdef"), < ); CHECK_DIFF(RTStrICmp("abcdeg", "abcdef"), > ); CHECK_DIFF(RTStrICmp("abcdef", "abcdeg"), < ); CHECK_DIFF(RTStrICmp("abcdeF", "abcdef"), ==); CHECK_DIFF(RTStrICmp("abcdef", "abcdeF"), ==); CHECK_DIFF(RTStrICmp("ABCDEF", "abcdef"), ==); CHECK_DIFF(RTStrICmp("abcdef", "ABCDEF"), ==); CHECK_DIFF(RTStrICmp("AbCdEf", "aBcDeF"), ==); CHECK_DIFF(RTStrICmp("AbCdEg", "aBcDeF"), > ); CHECK_DIFF(RTStrICmp("AbCdEG", "aBcDef"), > ); /* diff performed on the lower case cp. */ RTTestSub(hTest, "RTStrNICmp"); CHECK_DIFF(RTStrNICmp(NULL, NULL, RTSTR_MAX), == ); CHECK_DIFF(RTStrNICmp(NULL, "", RTSTR_MAX), < ); CHECK_DIFF(RTStrNICmp("", NULL, RTSTR_MAX), > ); CHECK_DIFF(RTStrNICmp("", "", RTSTR_MAX), == ); CHECK_DIFF(RTStrNICmp(NULL, NULL, 0), == ); CHECK_DIFF(RTStrNICmp(NULL, "", 0), == ); CHECK_DIFF(RTStrNICmp("", NULL, 0), == ); CHECK_DIFF(RTStrNICmp("", "", 0), == ); CHECK_DIFF(RTStrNICmp("abcdef", "abcdef", RTSTR_MAX), == ); CHECK_DIFF(RTStrNICmp("abcdef", "abcde", RTSTR_MAX), > ); CHECK_DIFF(RTStrNICmp("abcde", "abcdef", RTSTR_MAX), < ); CHECK_DIFF(RTStrNICmp("abcdeg", "abcdef", RTSTR_MAX), > ); CHECK_DIFF(RTStrNICmp("abcdef", "abcdeg", RTSTR_MAX), < ); CHECK_DIFF(RTStrNICmp("abcdeF", "abcdef", RTSTR_MAX), ==); CHECK_DIFF(RTStrNICmp("abcdef", "abcdeF", RTSTR_MAX), ==); CHECK_DIFF(RTStrNICmp("ABCDEF", "abcdef", RTSTR_MAX), ==); CHECK_DIFF(RTStrNICmp("abcdef", "ABCDEF", RTSTR_MAX), ==); CHECK_DIFF(RTStrNICmp("AbCdEf", "aBcDeF", RTSTR_MAX), ==); CHECK_DIFF(RTStrNICmp("AbCdEg", "aBcDeF", RTSTR_MAX), > ); CHECK_DIFF(RTStrNICmp("AbCdEG", "aBcDef", RTSTR_MAX), > ); /* diff performed on the lower case cp. */ CHECK_DIFF(RTStrNICmp("ABCDEF", "fedcba", 0), ==); CHECK_DIFF(RTStrNICmp("AbCdEg", "aBcDeF", 5), ==); CHECK_DIFF(RTStrNICmp("AbCdEf", "aBcDeF", 5), ==); CHECK_DIFF(RTStrNICmp("AbCdE", "aBcDe", 5), ==); CHECK_DIFF(RTStrNICmp("AbCdE", "aBcDeF", 5), ==); CHECK_DIFF(RTStrNICmp("AbCdEf", "aBcDe", 5), ==); CHECK_DIFF(RTStrNICmp("AbCdEg", "aBcDeF", 6), > ); CHECK_DIFF(RTStrNICmp("AbCdEG", "aBcDef", 6), > ); /* diff performed on the lower case cp. */ /* We should continue using byte comparison when we hit the invalid CP. Will assert in debug builds. */ // CHECK_DIFF(RTStrNICmp("AbCd\xff""eg", "aBcD\xff""eF", 6), ==); RTTestSubDone(hTest); } /** * Check UTF-8 encoding purging. */ void TstRTStrPurgeEncoding(RTTEST hTest) { RTTestSub(hTest, "RTStrPurgeEncoding"); /* * Test some good strings. */ char sz1[] = "1234567890wertyuiopsdfghjklzxcvbnm"; char sz1Copy[sizeof(sz1)]; memcpy(sz1Copy, sz1, sizeof(sz1)); RTTESTI_CHECK_RETV(RTStrPurgeEncoding(sz1) == 0); RTTESTI_CHECK_RETV(!memcmp(sz1, sz1Copy, sizeof(sz1))); char *pszAll = RTStrDup(g_szAll); if (pszAll) { RTTESTI_CHECK(RTStrPurgeEncoding(pszAll) == 0); RTTESTI_CHECK(!memcmp(pszAll, g_szAll, sizeof(g_szAll))); RTStrFree(pszAll); } /* * Test some bad stuff. */ struct { size_t cErrors; unsigned char szIn[5]; const char *pszExpect; } aTests[] = { { 0, { '1', '2', '3', '4', '\0' }, "1234" }, { 1, { 0x80, '2', '3', '4', '\0' }, "?234" }, { 1, { '1', 0x80, '3', '4', '\0' }, "1?34" }, { 1, { '1', '2', 0x80, '4', '\0' }, "12?4" }, { 1, { '1', '2', '3', 0x80, '\0' }, "123?" }, { 2, { 0x80, 0x81, '3', '4', '\0' }, "??34" }, { 2, { '1', 0x80, 0x81, '4', '\0' }, "1??4" }, { 2, { '1', '2', 0x80, 0x81, '\0' }, "12??" }, }; for (size_t i = 0; i < RT_ELEMENTS(aTests); i++) { size_t cErrors = RTStrPurgeEncoding((char *)aTests[i].szIn); if (cErrors != aTests[i].cErrors) RTTestFailed(hTest, "#%u: cErrors=%u expected %u\n", i, cErrors, aTests[i].cErrors); else if (strcmp((char *)aTests[i].szIn, aTests[i].pszExpect)) RTTestFailed(hTest, "#%u: %.5Rhxs expected %.5Rhxs (%s)\n", i, aTests[i].szIn, aTests[i].pszExpect, aTests[i].pszExpect); } RTTestSubDone(hTest); } /** * Check string sanitising. */ void TstRTStrPurgeComplementSet(RTTEST hTest) { RTTestSub(hTest, "RTStrPurgeComplementSet"); RTUNICP aCpSet[] = { '1', '5', 'w', 'w', 'r', 'r', 'e', 'f', 't', 't', '\0' }; RTUNICP aCpBadSet[] = { '1', '5', 'w', 'w', 'r', 'r', 'e', 'f', 't', 't', '7', '\0' }; /* Contains an incomplete pair. */ struct { const char *pcszIn; const char *pcszOut; PCRTUNICP pcCpSet; char chReplacement; ssize_t cExpected; } aTests[] = { { "1234werttrew4321", "1234werttrew4321", aCpSet, '_', 0 }, { "123654wert\xc2\xa2trew\xe2\x82\xac""4321", "123_54wert__trew___4321", aCpSet, '_', 3 }, { "hjhj8766", "????????", aCpSet, '?', 8 }, { "123\xf0\xa4\xad\xa2""4", "123____4", aCpSet, '_', 1 }, { "\xff", "\xff", aCpSet, '_', -1 }, { "____", "____", aCpBadSet, '_', -1 } }; enum { MAX_IN_STRING = 256 }; for (unsigned i = 0; i < RT_ELEMENTS(aTests); ++i) { char szCopy[MAX_IN_STRING]; ssize_t cReplacements; AssertRC(RTStrCopy(szCopy, RT_ELEMENTS(szCopy), aTests[i].pcszIn)); cReplacements = RTStrPurgeComplementSet(szCopy, aTests[i].pcCpSet, aTests[i].chReplacement); if (cReplacements != aTests[i].cExpected) RTTestFailed(hTest, "#%u: expected %lld, actual %lld\n", i, (long long) aTests[i].cExpected, (long long) cReplacements); if (strcmp(aTests[i].pcszOut, szCopy)) RTTestFailed(hTest, "#%u: expected %s, actual %s\n", i, aTests[i].pcszOut, szCopy); } } /** * Check string sanitising. */ void TstRTUtf16PurgeComplementSet(RTTEST hTest) { RTTestSub(hTest, "RTUtf16PurgeComplementSet"); RTUNICP aCpSet[] = { '1', '5', 'w', 'w', 'r', 'r', 'e', 'f', 't', 't', '\0' }; RTUNICP aCpBadSet[] = { '1', '5', 'w', 'w', 'r', 'r', 'e', 'f', 't', 't', '7', '\0' }; /* Contains an incomplete pair. */ struct { const char *pcszIn; const char *pcszOut; size_t cwc; /* Zero means the strings are Utf-8. */ PCRTUNICP pcCpSet; char chReplacement; ssize_t cExpected; } aTests[] = { { "1234werttrew4321", "1234werttrew4321", 0, aCpSet, '_', 0 }, { "123654wert\xc2\xa2trew\xe2\x82\xac""4321", "123_54wert_trew_4321", 0, aCpSet, '_', 3 }, { "hjhj8766", "????????", 0, aCpSet, '?', 8 }, { "123\xf0\xa4\xad\xa2""4", "123__4", 0, aCpSet, '_', 1 }, { "\xff\xff\0", "\xff\xff\0", 2, aCpSet, '_', -1 }, { "\xff\xff\0", "\xff\xff\0", 2, aCpSet, '_', -1 }, { "____", "____", 0, aCpBadSet, '_', -1 } }; enum { MAX_IN_STRING = 256 }; for (unsigned i = 0; i < RT_ELEMENTS(aTests); ++i) { RTUTF16 wszInCopy[MAX_IN_STRING], *pwszInCopy = wszInCopy; RTUTF16 wszOutCopy[MAX_IN_STRING], *pwszOutCopy = wszOutCopy; ssize_t cReplacements; if (!aTests[i].cwc) { AssertRC(RTStrToUtf16Ex(aTests[i].pcszIn, RTSTR_MAX, &pwszInCopy, RT_ELEMENTS(wszInCopy), NULL)); AssertRC(RTStrToUtf16Ex(aTests[i].pcszOut, RTSTR_MAX, &pwszOutCopy, RT_ELEMENTS(wszOutCopy), NULL)); } else { Assert(aTests[i].cwc <= RT_ELEMENTS(wszInCopy)); memcpy(wszInCopy, aTests[i].pcszIn, aTests[i].cwc * 2); memcpy(wszOutCopy, aTests[i].pcszOut, aTests[i].cwc * 2); } cReplacements = RTUtf16PurgeComplementSet(wszInCopy, aTests[i].pcCpSet, aTests[i].chReplacement); if (cReplacements != aTests[i].cExpected) RTTestFailed(hTest, "#%u: expected %lld, actual %lld\n", i, (long long) aTests[i].cExpected, (long long) cReplacements); if (RTUtf16Cmp(wszInCopy, wszOutCopy)) RTTestFailed(hTest, "#%u: expected %ls, actual %ls\n", i, wszOutCopy, wszInCopy); } } /** * Benchmark stuff. */ void Benchmarks(RTTEST hTest) { static union { RTUTF16 wszBuf[sizeof(g_wszAll)]; char szBuf[sizeof(g_szAll)]; } s_Buf; RTTestSub(hTest, "Benchmarks"); /** @todo add RTTest* methods for reporting benchmark results. */ RTTestPrintf(hTest, RTTESTLVL_ALWAYS, "Benchmarking RTStrToUtf16Ex: "); /** @todo figure this stuff into the test framework. */ PRTUTF16 pwsz = &s_Buf.wszBuf[0]; int rc = RTStrToUtf16Ex(&g_szAll[0], RTSTR_MAX, &pwsz, RT_ELEMENTS(s_Buf.wszBuf), NULL); if (RT_SUCCESS(rc)) { int i; uint64_t u64Start = RTTimeNanoTS(); for (i = 0; i < 100; i++) { rc = RTStrToUtf16Ex(&g_szAll[0], RTSTR_MAX, &pwsz, RT_ELEMENTS(s_Buf.wszBuf), NULL); if (RT_FAILURE(rc)) { RTTestFailed(hTest, "UTF-8 -> UTF-16 benchmark failed at i=%d, rc=%Rrc\n", i, rc); break; } } uint64_t u64Elapsed = RTTimeNanoTS() - u64Start; RTTestPrintf(hTest, RTTESTLVL_ALWAYS, "%d in %'RI64 ns\n", i, u64Elapsed); } RTTestPrintf(hTest, RTTESTLVL_ALWAYS, "Benchmarking RTUtf16ToUtf8Ex: "); char *psz = &s_Buf.szBuf[0]; rc = RTUtf16ToUtf8Ex(&g_wszAll[0], RTSTR_MAX, &psz, RT_ELEMENTS(s_Buf.szBuf), NULL); if (RT_SUCCESS(rc)) { int i; uint64_t u64Start = RTTimeNanoTS(); for (i = 0; i < 100; i++) { rc = RTUtf16ToUtf8Ex(&g_wszAll[0], RTSTR_MAX, &psz, RT_ELEMENTS(s_Buf.szBuf), NULL); if (RT_FAILURE(rc)) { RTTestFailed(hTest, "UTF-16 -> UTF-8 benchmark failed at i=%d, rc=%Rrc\n", i, rc); break; } } uint64_t u64Elapsed = RTTimeNanoTS() - u64Start; RTTestPrintf(hTest, RTTESTLVL_ALWAYS, "%d in %'RI64 ns\n", i, u64Elapsed); } RTTestSubDone(hTest); } /** * Tests RTStrEnd */ static void testStrEnd(RTTEST hTest) { RTTestSub(hTest, "RTStrEnd"); static char const s_szEmpty[1] = ""; RTTESTI_CHECK(RTStrEnd(s_szEmpty, 0) == NULL); RTTESTI_CHECK(RTStrEnd(s_szEmpty, 1) == &s_szEmpty[0]); for (size_t i = 0; i < _1M; i++) RTTESTI_CHECK(RTStrEnd(s_szEmpty, ~i) == &s_szEmpty[0]); } /** * Tests RTStrStr and RTStrIStr. */ static void testStrStr(RTTEST hTest) { #define CHECK_NULL(expr) \ do { \ const char *pszRet = expr; \ if (pszRet != NULL) \ RTTestFailed(hTest, "%d: %#x -> %s expected NULL", __LINE__, #expr, pszRet); \ } while (0) #define CHECK(expr, expect) \ do { \ const char *pszRet = expr; \ if ( (pszRet != NULL && (expect) == NULL) \ || (pszRet == NULL && (expect) != NULL) \ || strcmp(pszRet, (expect)) \ ) \ RTTestFailed(hTest, "%d: %#x -> %s expected %s", __LINE__, #expr, pszRet, (expect)); \ } while (0) RTTestSub(hTest, "RTStrStr"); CHECK(RTStrStr("abcdef", ""), "abcdef"); CHECK_NULL(RTStrStr("abcdef", NULL)); CHECK_NULL(RTStrStr(NULL, "")); CHECK_NULL(RTStrStr(NULL, NULL)); CHECK(RTStrStr("abcdef", "abcdef"), "abcdef"); CHECK(RTStrStr("abcdef", "b"), "bcdef"); CHECK(RTStrStr("abcdef", "bcdef"), "bcdef"); CHECK(RTStrStr("abcdef", "cdef"), "cdef"); CHECK(RTStrStr("abcdef", "cde"), "cdef"); CHECK(RTStrStr("abcdef", "cd"), "cdef"); CHECK(RTStrStr("abcdef", "c"), "cdef"); CHECK(RTStrStr("abcdef", "f"), "f"); CHECK(RTStrStr("abcdef", "ef"), "ef"); CHECK(RTStrStr("abcdef", "e"), "ef"); CHECK_NULL(RTStrStr("abcdef", "z")); CHECK_NULL(RTStrStr("abcdef", "A")); CHECK_NULL(RTStrStr("abcdef", "F")); RTTestSub(hTest, "RTStrIStr"); CHECK(RTStrIStr("abcdef", ""), "abcdef"); CHECK_NULL(RTStrIStr("abcdef", NULL)); CHECK_NULL(RTStrIStr(NULL, "")); CHECK_NULL(RTStrIStr(NULL, NULL)); CHECK(RTStrIStr("abcdef", "abcdef"), "abcdef"); CHECK(RTStrIStr("abcdef", "Abcdef"), "abcdef"); CHECK(RTStrIStr("abcdef", "ABcDeF"), "abcdef"); CHECK(RTStrIStr("abcdef", "b"), "bcdef"); CHECK(RTStrIStr("abcdef", "B"), "bcdef"); CHECK(RTStrIStr("abcdef", "bcdef"), "bcdef"); CHECK(RTStrIStr("abcdef", "BCdEf"), "bcdef"); CHECK(RTStrIStr("abcdef", "bCdEf"), "bcdef"); CHECK(RTStrIStr("abcdef", "bcdEf"), "bcdef"); CHECK(RTStrIStr("abcdef", "BcdEf"), "bcdef"); CHECK(RTStrIStr("abcdef", "cdef"), "cdef"); CHECK(RTStrIStr("abcdef", "cde"), "cdef"); CHECK(RTStrIStr("abcdef", "cd"), "cdef"); CHECK(RTStrIStr("abcdef", "c"), "cdef"); CHECK(RTStrIStr("abcdef", "f"), "f"); CHECK(RTStrIStr("abcdeF", "F"), "F"); CHECK(RTStrIStr("abcdef", "F"), "f"); CHECK(RTStrIStr("abcdef", "ef"), "ef"); CHECK(RTStrIStr("EeEef", "e"), "EeEef"); CHECK(RTStrIStr("EeEef", "E"), "EeEef"); CHECK(RTStrIStr("EeEef", "EE"), "EeEef"); CHECK(RTStrIStr("EeEef", "EEE"), "EeEef"); CHECK(RTStrIStr("EeEef", "EEEF"), "eEef"); CHECK_NULL(RTStrIStr("EeEef", "z")); #undef CHECK #undef CHECK_NULL RTTestSubDone(hTest); } void testUtf8Latin1(RTTEST hTest) { RTTestSub(hTest, "Latin-1 <-> Utf-8 conversion functions"); /* Test Utf8 -> Latin1 */ size_t cch_szAll = 0; size_t cbShort = RTStrCalcLatin1Len(g_szAll); RTTEST_CHECK(hTest, cbShort == 0); int rc = RTStrCalcLatin1LenEx(g_szAll, 383, &cch_szAll); RTTEST_CHECK(hTest, (cch_szAll == 255)); rc = RTStrCalcLatin1LenEx(g_szAll, RTSTR_MAX, &cch_szAll); RTTEST_CHECK_RC(hTest, rc, VERR_NO_TRANSLATION); char *psz = NULL; char szShort[256] = { 0 }; memcpy(szShort, g_szAll, 255); cbShort = RTStrCalcLatin1Len(szShort); RTTEST_CHECK(hTest, cbShort == 191); rc = RTStrToLatin1(szShort, &psz); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) { RTTEST_CHECK(hTest, (strlen(psz) == 191)); for (unsigned i = 0, j = 1; psz[i] != '\0'; ++i, ++j) if (psz[i] != (char) j) { RTTestFailed(hTest, "conversion of g_szAll to Latin1 failed at position %u\n", i); break; } } RTStrFree(psz); rc = RTStrToLatin1(g_szAll, &psz); RTTEST_CHECK_RC(hTest, rc, VERR_NO_TRANSLATION); char sz[512]; char *psz2 = &sz[0]; size_t cchActual = 0; rc = RTStrToLatin1Ex(g_szAll, sizeof(sz) - 1, &psz2, sizeof(sz), &cchActual); RTTEST_CHECK_RC(hTest, rc, VERR_NO_TRANSLATION); RTTEST_CHECK_MSG(hTest, cchActual == 0, (hTest, "cchActual=%lu\n", cchActual)); rc = RTStrToLatin1Ex(g_szAll, 383, &psz2, sizeof(sz), &cchActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) { RTTEST_CHECK(hTest, (cchActual == 255)); RTTEST_CHECK(hTest, (cchActual == strlen(sz))); for (unsigned i = 0, j = 1; psz2[i] != '\0'; ++i, ++j) if (psz2[i] != (char) j) { RTTestFailed(hTest, "second conversion of g_szAll to Latin1 failed at position %u\n", i); break; } } rc = RTStrToLatin1Ex(g_szAll, 129, &psz2, 128, &cchActual); RTTEST_CHECK_RC(hTest, rc, VERR_BUFFER_OVERFLOW); RTTEST_CHECK_MSG(hTest, cchActual == 128, (hTest, "cchActual=%lu\n", cchActual)); rc = RTStrToLatin1Ex(g_szAll, 383, &psz, 0, &cchActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) { RTTEST_CHECK(hTest, (cchActual == 255)); RTTEST_CHECK(hTest, (cchActual == strlen(psz))); for (unsigned i = 0, j = 1; psz[i] != '\0'; ++i, ++j) if ( ((j < 0x100) && (psz[i] != (char) j)) || ((j > 0xff) && psz[i] != '?')) { RTTestFailed(hTest, "third conversion of g_szAll to Latin1 failed at position %u\n", i); break; } } const char *pszBad = "Hello\xDC\xD8"; rc = RTStrToLatin1Ex(pszBad, RTSTR_MAX, &psz2, sizeof(sz), &cchActual); RTTEST_CHECK_RC(hTest, rc, VERR_INVALID_UTF8_ENCODING); RTStrFree(psz); /* Test Latin1 -> Utf8 */ const char *pszLat1 = "\x01\x20\x40\x80\x81"; RTTEST_CHECK(hTest, RTLatin1CalcUtf8Len(pszLat1) == 7); rc = RTLatin1CalcUtf8LenEx(pszLat1, 3, &cchActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) RTTEST_CHECK(hTest, cchActual == 3); rc = RTLatin1CalcUtf8LenEx(pszLat1, RTSTR_MAX, &cchActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) RTTEST_CHECK(hTest, cchActual == 7); char *pch = NULL; char ch[8]; char *pch2 = &ch[0]; cchActual = 0; rc = RTLatin1ToUtf8(pszLat1, &pch); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) RTTEST_CHECK(hTest, !strcmp(pch, "\x01\x20\x40\xC2\x80\xC2\x81")); RTStrFree(pch); rc = RTLatin1ToUtf8Ex(pszLat1, RTSTR_MAX, &pch, 0, &cchActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) { RTTEST_CHECK(hTest, (cchActual == 7)); RTTEST_CHECK(hTest, !strcmp(pch, "\x01\x20\x40\xC2\x80\xC2\x81")); } RTStrFree(pch); rc = RTLatin1ToUtf8Ex(pszLat1, RTSTR_MAX, &pch, 0, NULL); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) RTTEST_CHECK(hTest, !strcmp(pch, "\x01\x20\x40\xC2\x80\xC2\x81")); RTStrFree(pch); rc = RTLatin1ToUtf8Ex(pszLat1, RTSTR_MAX, &pch2, RT_ELEMENTS(ch), &cchActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) { RTTEST_CHECK(hTest, (cchActual == 7)); RTTEST_CHECK(hTest, !strcmp(pch2, "\x01\x20\x40\xC2\x80\xC2\x81")); } rc = RTLatin1ToUtf8Ex(pszLat1, 3, &pch2, RT_ELEMENTS(ch), &cchActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) { RTTEST_CHECK(hTest, (cchActual == 3)); RTTEST_CHECK(hTest, !strcmp(pch2, "\x01\x20\x40")); } rc = RTLatin1ToUtf8Ex(pszLat1, RTSTR_MAX, &pch2, RT_ELEMENTS(ch) - 1, &cchActual); RTTEST_CHECK_RC(hTest, rc, VERR_BUFFER_OVERFLOW); RTTEST_CHECK(hTest, (cchActual == 7)); RTTestSubDone(hTest); } void testUtf16Latin1(RTTEST hTest) { RTTestSub(hTest, "Latin-1 <-> Utf-16 conversion functions"); /* Test Utf16 -> Latin1 */ size_t cch_szAll = 0; size_t cbShort = RTUtf16CalcLatin1Len(g_wszAll); RTTEST_CHECK(hTest, cbShort == 0); int rc = RTUtf16CalcLatin1LenEx(g_wszAll, 255, &cch_szAll); RTTEST_CHECK(hTest, (cch_szAll == 255)); rc = RTUtf16CalcLatin1LenEx(g_wszAll, RTSTR_MAX, &cch_szAll); RTTEST_CHECK_RC(hTest, rc, VERR_NO_TRANSLATION); char *psz = NULL; RTUTF16 wszShort[256] = { 0 }; for (unsigned i = 0; i < 255; ++i) wszShort[i] = i + 1; cbShort = RTUtf16CalcLatin1Len(wszShort); RTTEST_CHECK(hTest, cbShort == 255); rc = RTUtf16ToLatin1(wszShort, &psz); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) { RTTEST_CHECK(hTest, (strlen(psz) == 255)); for (unsigned i = 0, j = 1; psz[i] != '\0'; ++i, ++j) if (psz[i] != (char) j) { RTTestFailed(hTest, "conversion of g_wszAll to Latin1 failed at position %u\n", i); break; } } RTStrFree(psz); rc = RTUtf16ToLatin1(g_wszAll, &psz); RTTEST_CHECK_RC(hTest, rc, VERR_NO_TRANSLATION); char sz[512]; char *psz2 = &sz[0]; size_t cchActual = 0; rc = RTUtf16ToLatin1Ex(g_wszAll, sizeof(sz) - 1, &psz2, sizeof(sz), &cchActual); RTTEST_CHECK_RC(hTest, rc, VERR_NO_TRANSLATION); RTTEST_CHECK_MSG(hTest, cchActual == 0, (hTest, "cchActual=%lu\n", cchActual)); rc = RTUtf16ToLatin1Ex(g_wszAll, 255, &psz2, sizeof(sz), &cchActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) { RTTEST_CHECK(hTest, (cchActual == 255)); RTTEST_CHECK(hTest, (cchActual == strlen(sz))); for (unsigned i = 0, j = 1; psz2[i] != '\0'; ++i, ++j) if (psz2[i] != (char) j) { RTTestFailed(hTest, "second conversion of g_wszAll to Latin1 failed at position %u\n", i); break; } } rc = RTUtf16ToLatin1Ex(g_wszAll, 128, &psz2, 128, &cchActual); RTTEST_CHECK_RC(hTest, rc, VERR_BUFFER_OVERFLOW); RTTEST_CHECK_MSG(hTest, cchActual == 128, (hTest, "cchActual=%lu\n", cchActual)); rc = RTUtf16ToLatin1Ex(g_wszAll, 255, &psz, 0, &cchActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) { RTTEST_CHECK(hTest, (cchActual == 255)); RTTEST_CHECK(hTest, (cchActual == strlen(psz))); for (unsigned i = 0, j = 1; psz[i] != '\0'; ++i, ++j) if ( ((j < 0x100) && (psz[i] != (char) j)) || ((j > 0xff) && psz[i] != '?')) { RTTestFailed(hTest, "third conversion of g_wszAll to Latin1 failed at position %u\n", i); break; } } const char *pszBad = "H\0e\0l\0l\0o\0\0\xDC\0\xD8\0"; rc = RTUtf16ToLatin1Ex((RTUTF16 *) pszBad, RTSTR_MAX, &psz2, sizeof(sz), &cchActual); RTTEST_CHECK_RC(hTest, rc, VERR_INVALID_UTF16_ENCODING); RTStrFree(psz); /* Test Latin1 -> Utf16 */ const char *pszLat1 = "\x01\x20\x40\x80\x81"; RTTEST_CHECK(hTest, RTLatin1CalcUtf16Len(pszLat1) == 5); rc = RTLatin1CalcUtf16LenEx(pszLat1, 3, &cchActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) RTTEST_CHECK(hTest, cchActual == 3); rc = RTLatin1CalcUtf16LenEx(pszLat1, RTSTR_MAX, &cchActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) RTTEST_CHECK(hTest, cchActual == 5); RTUTF16 *pwc = NULL; RTUTF16 wc[6]; RTUTF16 *pwc2 = &wc[0]; size_t cwActual = 0; rc = RTLatin1ToUtf16(pszLat1, &pwc); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) RTTEST_CHECK(hTest, (pwc[0] == 1) && (pwc[1] == 0x20) && (pwc[2] == 0x40) && (pwc[3] == 0x80) && (pwc[4] == 0x81) && (pwc[5] == '\0')); RTUtf16Free(pwc); rc = RTLatin1ToUtf16Ex(pszLat1, RTSTR_MAX, &pwc, 0, &cwActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) { RTTEST_CHECK(hTest, (cwActual == 5)); RTTEST_CHECK(hTest, (pwc[0] == 1) && (pwc[1] == 0x20) && (pwc[2] == 0x40) && (pwc[3] == 0x80) && (pwc[4] == 0x81) && (pwc[5] == '\0')); } RTUtf16Free(pwc); rc = RTLatin1ToUtf16Ex(pszLat1, RTSTR_MAX, &pwc, 0, NULL); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) RTTEST_CHECK(hTest, (pwc[0] == 1) && (pwc[1] == 0x20) && (pwc[2] == 0x40) && (pwc[3] == 0x80) && (pwc[4] == 0x81) && (pwc[5] == '\0')); RTUtf16Free(pwc); rc = RTLatin1ToUtf16Ex(pszLat1, RTSTR_MAX, &pwc2, RT_ELEMENTS(wc), &cwActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) { RTTEST_CHECK(hTest, (cwActual == 5)); RTTEST_CHECK(hTest, (wc[0] == 1) && (wc[1] == 0x20) && (wc[2] == 0x40) && (wc[3] == 0x80) && (wc[4] == 0x81) && (wc[5] == '\0')); } rc = RTLatin1ToUtf16Ex(pszLat1, 3, &pwc2, RT_ELEMENTS(wc), &cwActual); RTTEST_CHECK_RC_OK(hTest, rc); if (RT_SUCCESS(rc)) { RTTEST_CHECK(hTest, (cwActual == 3)); RTTEST_CHECK(hTest, (wc[0] == 1) && (wc[1] == 0x20) && (wc[2] == 0x40) && (wc[3] == '\0')); } rc = RTLatin1ToUtf16Ex(pszLat1, RTSTR_MAX, &pwc2, RT_ELEMENTS(wc) - 1, &cwActual); RTTEST_CHECK_RC(hTest, rc, VERR_BUFFER_OVERFLOW); RTTEST_CHECK(hTest, (cwActual == 5)); RTTestSubDone(hTest); } static void testNoTransation(RTTEST hTest) { /* * Try trigger a VERR_NO_TRANSLATION error in convert to * current CP to latin-1. */ const RTUTF16 s_swzTest1[] = { 0x2358, 0x2242, 0x2357, 0x2359, 0x22f9, 0x2c4e, 0x0030, 0x0060, 0x0092, 0x00c1, 0x00f2, 0x1f80, 0x0088, 0x2c38, 0x2c30, 0x0000 }; char *pszTest1; int rc = RTUtf16ToUtf8(s_swzTest1, &pszTest1); RTTESTI_CHECK_RC_RETV(rc, VINF_SUCCESS); RTTestSub(hTest, "VERR_NO_TRANSLATION/RTStrUtf8ToCurrentCP"); char *pszOut; rc = RTStrUtf8ToCurrentCP(&pszOut, pszTest1); if (rc == VINF_SUCCESS) { RTTESTI_CHECK(!strcmp(pszOut, pszTest1)); RTTestIPrintf(RTTESTLVL_ALWAYS, "CurrentCP is UTF-8 or similar (LC_ALL=%s LANG=%s LC_CTYPE=%s)\n", RTEnvGet("LC_ALL"), RTEnvGet("LANG"), RTEnvGet("LC_CTYPE")); RTStrFree(pszOut); } else RTTESTI_CHECK_MSG(rc == VWRN_NO_TRANSLATION || rc == VERR_NO_TRANSLATION, ("rc=%Rrc\n", rc)); RTTestSub(hTest, "VERR_NO_TRANSLATION/RTUtf16ToLatin1"); rc = RTUtf16ToLatin1(s_swzTest1, &pszOut); RTTESTI_CHECK_RC(rc, VERR_NO_TRANSLATION); if (RT_SUCCESS(rc)) RTStrFree(pszOut); RTStrFree(pszTest1); RTTestSubDone(hTest); } static void testGetPut(RTTEST hTest) { /* * Test RTStrPutCp, RTStrGetCp and RTStrGetCpEx. */ RTTestSub(hTest, "RTStrPutCp, RTStrGetCp and RTStrGetCpEx"); RTUNICP uc = 0; while (uc <= 0x10fffd) { /* Figure the range - skip illegal ranges. */ RTUNICP ucFirst = uc; if (ucFirst - UINT32_C(0xd800) <= 0x7ff) ucFirst = 0xe000; else if (ucFirst == UINT32_C(0xfffe) || ucFirst == UINT32_C(0xffff)) ucFirst = 0x10000; RTUNICP ucLast = ucFirst + 1023; if (ucLast - UINT32_C(0xd800) <= 0x7ff) ucLast = 0xd7ff; else if (ucLast == UINT32_C(0xfffe) || ucLast == UINT32_C(0xffff)) ucLast = 0xfffd; /* Encode the range into a string, decode each code point as we go along. */ char sz1[8192]; char *pszDst = sz1; for (uc = ucFirst; uc <= ucLast; uc++) { char *pszBefore = pszDst; pszDst = RTStrPutCp(pszDst, uc); RTTESTI_CHECK(pszBefore - pszDst < 6); RTUNICP uc2 = RTStrGetCp(pszBefore); RTTESTI_CHECK_MSG(uc2 == uc, ("uc2=%#x uc=%#x\n", uc2, uc)); const char *pszSrc = pszBefore; RTUNICP uc3 = 42; RTTESTI_CHECK_RC(RTStrGetCpEx(&pszSrc, &uc3), VINF_SUCCESS); RTTESTI_CHECK_MSG(uc3 == uc, ("uc3=%#x uc=%#x\n", uc3, uc)); RTTESTI_CHECK_MSG(pszSrc == pszDst, ("pszSrc=%p pszDst=%p\n", pszSrc, pszDst)); } /* Decode and re-encode it. */ const char *pszSrc = pszDst = sz1; for (uc = ucFirst; uc <= ucLast; uc++) { RTUNICP uc2 = RTStrGetCp(pszSrc); RTTESTI_CHECK_MSG(uc2 == uc, ("uc2=%#x uc=%#x\n", uc2, uc)); RTUNICP uc3 = 42; RTTESTI_CHECK_RC(RTStrGetCpEx(&pszSrc, &uc3), VINF_SUCCESS); RTTESTI_CHECK_MSG(uc3 == uc, ("uc3=%#x uc=%#x\n", uc3, uc)); pszDst = RTStrPutCp(pszDst, uc); RTTESTI_CHECK_MSG(pszSrc == pszDst, ("pszSrc=%p pszDst=%p\n", pszSrc, pszDst)); pszSrc = pszDst; } /* Decode and wipe it (checking compiler optimizations). */ pszSrc = pszDst = sz1; for (uc = ucFirst; uc <= ucLast; uc++) { RTUNICP uc2 = RTStrGetCp(pszSrc); RTTESTI_CHECK_MSG(uc2 == uc, ("uc2=%#x uc=%#x\n", uc2, uc)); RTUNICP uc3 = 42; RTTESTI_CHECK_RC(RTStrGetCpEx(&pszSrc, &uc3), VINF_SUCCESS); RTTESTI_CHECK_MSG(uc3 == uc, ("uc3=%#x uc=%#x\n", uc3, uc)); pszDst = RTStrPutCp(pszDst, 0); } /* advance */ uc = ucLast + 1; } } int main() { /* * Init the runtime, test and say hello. */ RTTEST hTest; RTEXITCODE rcExit = RTTestInitAndCreate("tstUtf8", &hTest); if (rcExit != RTEXITCODE_SUCCESS) return rcExit; RTTestBanner(hTest); /* * Run the tests. */ InitStrings(); test1(hTest); test2(hTest); test3(hTest); TstRTStrXCmp(hTest); TstRTStrPurgeEncoding(hTest); /* TstRT*PurgeComplementSet test conditions which assert. */ bool fAreQuiet = RTAssertAreQuiet(), fMayPanic = RTAssertMayPanic(); RTAssertSetQuiet(true); RTAssertSetMayPanic(false); TstRTStrPurgeComplementSet(hTest); TstRTUtf16PurgeComplementSet(hTest); RTAssertSetQuiet(fAreQuiet); RTAssertSetMayPanic(fMayPanic); testStrEnd(hTest); testStrStr(hTest); testUtf8Latin1(hTest); testUtf16Latin1(hTest); testNoTransation(hTest); testGetPut(hTest); Benchmarks(hTest); /* * Summary */ return RTTestSummaryAndDestroy(hTest); }