Sweden-Number/dlls/mlang/tests/mlang.c

731 lines
31 KiB
C

/*
* Unit test suite for MLANG APIs.
*
* Copyright 2004 Dmitry Timoshkov
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
*/
#define COBJMACROS
#include <stdarg.h>
#include "windef.h"
#include "winbase.h"
#include "winerror.h"
#include "mlang.h"
#include "wine/test.h"
#ifndef CP_UNICODE
#define CP_UNICODE 1200
#endif
#if 0
#define DUMP_CP_INFO
#define DUMP_SCRIPT_INFO
#if defined DUMP_CP_INFO || defined DUMP_SCRIPT_INFO
#include "wine/debug.h"
#endif
#endif /* 0 */
#define TRACE_2 OutputDebugStringA
static BOOL (WINAPI *pGetCPInfoExA)(UINT,DWORD,LPCPINFOEXA);
static void test_multibyte_to_unicode_translations(IMultiLanguage2 *iML2)
{
/* these APIs are broken regarding constness of the input buffer */
char stringA[] = "Just a test string\0"; /* double 0 for CP_UNICODE tests */
WCHAR stringW[] = {'J','u','s','t',' ','a',' ','t','e','s','t',' ','s','t','r','i','n','g',0};
char bufA[256];
WCHAR bufW[256];
UINT lenA, lenW, expected_len;
HRESULT ret;
HMODULE hMlang;
FARPROC pConvertINetMultiByteToUnicode;
FARPROC pConvertINetUnicodeToMultiByte;
hMlang = LoadLibraryA("mlang.dll");
ok(hMlang != 0, "couldn't load mlang.dll\n");
pConvertINetMultiByteToUnicode = GetProcAddress(hMlang, "ConvertINetMultiByteToUnicode");
ok(pConvertINetMultiByteToUnicode != NULL, "couldn't resolve ConvertINetMultiByteToUnicode\n");
pConvertINetUnicodeToMultiByte = GetProcAddress(hMlang, "ConvertINetUnicodeToMultiByte");
ok(pConvertINetUnicodeToMultiByte != NULL, "couldn't resolve ConvertINetUnicodeToMultiByte\n");
/* IMultiLanguage2_ConvertStringToUnicode tests */
memset(bufW, 'x', sizeof(bufW));
lenA = 0;
lenW = sizeof(bufW)/sizeof(bufW[0]);
TRACE_2("Call IMultiLanguage2_ConvertStringToUnicode\n");
ret = IMultiLanguage2_ConvertStringToUnicode(iML2, NULL, 1252, stringA, &lenA, bufW, &lenW);
ok(ret == S_OK, "IMultiLanguage2_ConvertStringToUnicode failed: %08x\n", ret);
ok(lenA == 0, "expected lenA 0, got %u\n", lenA);
ok(lenW == 0, "expected lenW 0, got %u\n", lenW);
memset(bufW, 'x', sizeof(bufW));
lenA = -1;
lenW = sizeof(bufW)/sizeof(bufW[0]);
TRACE_2("Call IMultiLanguage2_ConvertStringToUnicode\n");
ret = IMultiLanguage2_ConvertStringToUnicode(iML2, NULL, 1252, stringA, &lenA, bufW, &lenW);
ok(ret == S_OK, "IMultiLanguage2_ConvertStringToUnicode failed: %08x\n", ret);
ok(lenA == lstrlenA(stringA), "expected lenA %u, got %u\n", lstrlenA(stringA), lenA);
ok(lenW == lstrlenW(stringW), "expected lenW %u, got %u\n", lstrlenW(stringW), lenW);
ok(bufW[lenW] != 0, "buf should not be 0 terminated\n");
bufW[lenW] = 0; /* -1 doesn't include 0 terminator */
ok(!lstrcmpW(bufW, stringW), "bufW/stringW mismatch\n");
memset(bufW, 'x', sizeof(bufW));
lenA = -1;
lenW = 5;
TRACE_2("Call IMultiLanguage2_ConvertStringToUnicode\n");
ret = IMultiLanguage2_ConvertStringToUnicode(iML2, NULL, 1252, stringA, &lenA, bufW, &lenW);
ok(ret == E_FAIL, "IMultiLanguage2_ConvertStringToUnicode should fail: %08x\n", ret);
ok(lenW == 0, "expected lenW 0, got %u\n", lenW);
/* still has to do partial conversion */
ok(!memcmp(bufW, stringW, 5 * sizeof(WCHAR)), "bufW/stringW mismatch\n");
memset(bufW, 'x', sizeof(bufW));
lenA = -1;
lenW = sizeof(bufW)/sizeof(bufW[0]);
TRACE_2("Call IMultiLanguage2_ConvertStringToUnicode\n");
ret = IMultiLanguage2_ConvertStringToUnicode(iML2, NULL, CP_UNICODE, stringA, &lenA, bufW, &lenW);
ok(ret == S_OK, "IMultiLanguage2_ConvertStringToUnicode failed: %08x\n", ret);
ok(lenA == lstrlenA(stringA), "expected lenA %u, got %u\n", lstrlenA(stringA), lenA);
ok(lenW == lstrlenW(stringW)/(int)sizeof(WCHAR), "wrong lenW %u\n", lenW);
ok(bufW[lenW] != 0, "buf should not be 0 terminated\n");
bufW[lenW] = 0; /* -1 doesn't include 0 terminator */
ok(!lstrcmpA((LPCSTR)bufW, stringA), "bufW/stringA mismatch\n");
memset(bufW, 'x', sizeof(bufW));
lenA = lstrlenA(stringA);
lenW = 0;
ret = IMultiLanguage2_ConvertStringToUnicode(iML2, NULL, 1252, stringA, &lenA, NULL, &lenW);
ok(ret == S_OK, "IMultiLanguage2_ConvertStringToUnicode failed: %08x\n", ret);
ok(lenA == lstrlenA(stringA), "expected lenA %u, got %u\n", lstrlenA(stringA), lenA);
expected_len = MultiByteToWideChar(1252, 0, stringA, lenA, NULL, 0);
ok(lenW == expected_len, "expected lenW %u, got %u\n", expected_len, lenW);
memset(bufW, 'x', sizeof(bufW));
lenA = lstrlenA(stringA);
lenW = sizeof(bufW)/sizeof(bufW[0]);
ret = pConvertINetMultiByteToUnicode(NULL, 1252, stringA, &lenA, NULL, &lenW);
ok(ret == S_OK, "ConvertINetMultiByteToUnicode failed: %08x\n", ret);
ok(lenA == lstrlenA(stringA), "expected lenA %u, got %u\n", lstrlenA(stringA), lenA);
expected_len = MultiByteToWideChar(1252, 0, stringA, lenA, NULL, 0);
ok(lenW == expected_len, "expected lenW %u, got %u\n", expected_len, lenW);
memset(bufW, 'x', sizeof(bufW));
lenA = lstrlenA(stringA);
lenW = 0;
ret = pConvertINetMultiByteToUnicode(NULL, 1252, stringA, &lenA, NULL, &lenW);
ok(ret == S_OK, "ConvertINetMultiByteToUnicode failed: %08x\n", ret);
ok(lenA == lstrlenA(stringA), "expected lenA %u, got %u\n", lstrlenA(stringA), lenA);
expected_len = MultiByteToWideChar(1252, 0, stringA, lenA, NULL, 0);
ok(lenW == expected_len, "expected lenW %u, got %u\n", expected_len, lenW);
/* IMultiLanguage2_ConvertStringFromUnicode tests */
memset(bufA, 'x', sizeof(bufA));
lenW = 0;
lenA = sizeof(bufA);
TRACE_2("Call IMultiLanguage2_ConvertStringFromUnicode\n");
ret = IMultiLanguage2_ConvertStringFromUnicode(iML2, NULL, 1252, stringW, &lenW, bufA, &lenA);
ok(ret == S_OK, "IMultiLanguage2_ConvertStringFromUnicode failed: %08x\n", ret);
ok(lenA == 0, "expected lenA 0, got %u\n", lenA);
ok(lenW == 0, "expected lenW 0, got %u\n", lenW);
memset(bufA, 'x', sizeof(bufA));
lenW = -1;
lenA = sizeof(bufA);
TRACE_2("Call IMultiLanguage2_ConvertStringFromUnicode\n");
ret = IMultiLanguage2_ConvertStringFromUnicode(iML2, NULL, 1252, stringW, &lenW, bufA, &lenA);
ok(ret == S_OK, "IMultiLanguage2_ConvertStringFromUnicode failed: %08x\n", ret);
ok(lenA == lstrlenA(stringA), "expected lenA %u, got %u\n", lstrlenA(stringA), lenA);
ok(lenW == lstrlenW(stringW), "expected lenW %u, got %u\n", lstrlenW(stringW), lenW);
ok(bufA[lenA] != 0, "buf should not be 0 terminated\n");
bufA[lenA] = 0; /* -1 doesn't include 0 terminator */
ok(!lstrcmpA(bufA, stringA), "bufA/stringA mismatch\n");
memset(bufA, 'x', sizeof(bufA));
lenW = -1;
lenA = 5;
TRACE_2("Call IMultiLanguage2_ConvertStringFromUnicode\n");
ret = IMultiLanguage2_ConvertStringFromUnicode(iML2, NULL, 1252, stringW, &lenW, bufA, &lenA);
ok(ret == E_FAIL, "IMultiLanguage2_ConvertStringFromUnicode should fail: %08x\n", ret);
ok(lenA == 0, "expected lenA 0, got %u\n", lenA);
/* still has to do partial conversion */
ok(!memcmp(bufA, stringA, 5), "bufW/stringW mismatch\n");
memset(bufA, 'x', sizeof(bufA));
lenW = -1;
lenA = sizeof(bufA);
TRACE_2("Call IMultiLanguage2_ConvertStringFromUnicode\n");
ret = IMultiLanguage2_ConvertStringFromUnicode(iML2, NULL, CP_UNICODE, stringW, &lenW, bufA, &lenA);
ok(ret == S_OK, "IMultiLanguage2_ConvertStringFromUnicode failed: %08x\n", ret);
ok(lenA == lstrlenA(stringA) * (int)sizeof(WCHAR), "wrong lenA %u\n", lenA);
ok(lenW == lstrlenW(stringW), "expected lenW %u, got %u\n", lstrlenW(stringW), lenW);
ok(bufA[lenA] != 0 && bufA[lenA+1] != 0, "buf should not be 0 terminated\n");
bufA[lenA] = 0; /* -1 doesn't include 0 terminator */
bufA[lenA+1] = 0; /* sizeof(WCHAR) */
ok(!lstrcmpW((LPCWSTR)bufA, stringW), "bufA/stringW mismatch\n");
memset(bufA, 'x', sizeof(bufA));
lenW = lstrlenW(stringW);
lenA = 0;
ret = IMultiLanguage2_ConvertStringFromUnicode(iML2, NULL, 1252, stringW, &lenW, NULL, &lenA);
ok(ret == S_OK, "IMultiLanguage2_ConvertStringFromUnicode failed: %08x\n", ret);
ok(lenW == lstrlenW(stringW), "expected lenW %u, got %u\n", lstrlenW(stringW), lenW);
expected_len = WideCharToMultiByte(1252, 0, stringW, lenW, NULL, 0, NULL, NULL);
ok(lenA == expected_len, "expected lenA %u, got %u\n", expected_len, lenA);
memset(bufA, 'x', sizeof(bufA));
lenW = lstrlenW(stringW);
lenA = sizeof(bufA);
ret = pConvertINetUnicodeToMultiByte(NULL, 1252, stringW, &lenW, NULL, &lenA);
ok(ret == S_OK, "ConvertINetUnicodeToMultiByte failed: %08x\n", ret);
ok(lenW == lstrlenW(stringW), "expected lenW %u, got %u\n", lstrlenW(stringW), lenW);
expected_len = WideCharToMultiByte(1252, 0, stringW, lenW, NULL, 0, NULL, NULL);
ok(lenA == expected_len, "expected lenA %u, got %u\n", expected_len, lenA);
memset(bufA, 'x', sizeof(bufA));
lenW = lstrlenW(stringW);
lenA = 0;
ret = pConvertINetUnicodeToMultiByte(NULL, 1252, stringW, &lenW, NULL, &lenA);
ok(ret == S_OK, "ConvertINetUnicodeToMultiByte failed: %08x\n", ret);
ok(lenW == lstrlenW(stringW), "expected lenW %u, got %u\n", lstrlenW(stringW), lenW);
expected_len = WideCharToMultiByte(1252, 0, stringW, lenW, NULL, 0, NULL, NULL);
ok(lenA == expected_len, "expected lenA %u, got %u\n", expected_len, lenA);
}
inline static void cpinfo_cmp(MIMECPINFO *cpinfo1, MIMECPINFO *cpinfo2)
{
ok(cpinfo1->dwFlags == cpinfo2->dwFlags, "dwFlags mismatch: %08x != %08x\n", cpinfo1->dwFlags, cpinfo2->dwFlags);
ok(cpinfo1->uiCodePage == cpinfo2->uiCodePage, "uiCodePage mismatch: %u != %u\n", cpinfo1->uiCodePage, cpinfo2->uiCodePage);
ok(cpinfo1->uiFamilyCodePage == cpinfo2->uiFamilyCodePage, "uiFamilyCodePage mismatch: %u != %u\n", cpinfo1->uiFamilyCodePage, cpinfo2->uiFamilyCodePage);
ok(!lstrcmpW(cpinfo1->wszDescription, cpinfo2->wszDescription), "wszDescription mismatch\n");
ok(!lstrcmpW(cpinfo1->wszWebCharset, cpinfo2->wszWebCharset), "wszWebCharset mismatch\n");
ok(!lstrcmpW(cpinfo1->wszHeaderCharset, cpinfo2->wszHeaderCharset), "wszHeaderCharset mismatch\n");
ok(!lstrcmpW(cpinfo1->wszBodyCharset, cpinfo2->wszBodyCharset), "wszBodyCharset mismatch\n");
ok(!lstrcmpW(cpinfo1->wszFixedWidthFont, cpinfo2->wszFixedWidthFont), "wszFixedWidthFont mismatch\n");
ok(!lstrcmpW(cpinfo1->wszProportionalFont, cpinfo2->wszProportionalFont), "wszProportionalFont mismatch\n");
ok(cpinfo1->bGDICharset == cpinfo2->bGDICharset, "bGDICharset mismatch: %d != %d\n", cpinfo1->bGDICharset, cpinfo2->bGDICharset);
}
#ifdef DUMP_CP_INFO
static const char *dump_mime_flags(DWORD flags)
{
static char buf[1024];
buf[0] = 0;
if (flags & MIMECONTF_MAILNEWS) strcat(buf, " MIMECONTF_MAILNEWS");
if (flags & MIMECONTF_BROWSER) strcat(buf, " MIMECONTF_BROWSER");
if (flags & MIMECONTF_MINIMAL) strcat(buf, " MIMECONTF_MINIMAL");
if (flags & MIMECONTF_IMPORT) strcat(buf, " MIMECONTF_IMPORT");
if (flags & MIMECONTF_SAVABLE_MAILNEWS) strcat(buf, " MIMECONTF_SAVABLE_MAILNEWS");
if (flags & MIMECONTF_SAVABLE_BROWSER) strcat(buf, " MIMECONTF_SAVABLE_BROWSER");
if (flags & MIMECONTF_EXPORT) strcat(buf, " MIMECONTF_EXPORT");
if (flags & MIMECONTF_PRIVCONVERTER) strcat(buf, " MIMECONTF_PRIVCONVERTER");
if (flags & MIMECONTF_VALID) strcat(buf, " MIMECONTF_VALID");
if (flags & MIMECONTF_VALID_NLS) strcat(buf, " MIMECONTF_VALID_NLS");
if (flags & MIMECONTF_MIME_IE4) strcat(buf, " MIMECONTF_MIME_IE4");
if (flags & MIMECONTF_MIME_LATEST) strcat(buf, " MIMECONTF_MIME_LATEST");
if (flags & MIMECONTF_MIME_REGISTRY) strcat(buf, " MIMECONTF_MIME_REGISTRY");
return buf;
}
#endif
static void test_EnumCodePages(IMultiLanguage2 *iML2, DWORD flags)
{
IEnumCodePage *iEnumCP = NULL;
MIMECPINFO *cpinfo;
MIMECPINFO cpinfo2;
HRESULT ret;
ULONG i, n;
UINT total;
total = 0;
TRACE_2("Call IMultiLanguage2_GetNumberOfCodePageInfo\n");
ret = IMultiLanguage2_GetNumberOfCodePageInfo(iML2, &total);
ok(ret == S_OK && total != 0, "IMultiLanguage2_GetNumberOfCodePageInfo: expected S_OK/!0, got %08x/%u\n", ret, total);
trace("total mlang supported codepages %u\n", total);
TRACE_2("Call IMultiLanguage2_EnumCodePages\n");
ret = IMultiLanguage2_EnumCodePages(iML2, flags, LANG_NEUTRAL, &iEnumCP);
trace("IMultiLanguage2_EnumCodePages = %08x, iEnumCP = %p\n", ret, iEnumCP);
ok(ret == S_OK && iEnumCP, "IMultiLanguage2_EnumCodePages: expected S_OK/!NULL, got %08x/%p\n", ret, iEnumCP);
TRACE_2("Call IEnumCodePage_Reset\n");
ret = IEnumCodePage_Reset(iEnumCP);
ok(ret == S_OK, "IEnumCodePage_Reset: expected S_OK, got %08x\n", ret);
n = 65536;
TRACE_2("Call IEnumCodePage_Next\n");
ret = IEnumCodePage_Next(iEnumCP, 0, NULL, &n);
ok(n == 0 && ret == S_FALSE, "IEnumCodePage_Next: expected 0/S_FALSE, got %u/%08x\n", n, ret);
TRACE_2("Call IEnumCodePage_Next\n");
ret = IEnumCodePage_Next(iEnumCP, 0, NULL, NULL);
ok(ret == S_FALSE, "IEnumCodePage_Next: expected S_FALSE, got %08x\n", ret);
cpinfo = HeapAlloc(GetProcessHeap(), 0, sizeof(*cpinfo) * total * 2);
n = total * 2;
TRACE_2("Call IEnumCodePage_Next\n");
ret = IEnumCodePage_Next(iEnumCP, 0, cpinfo, &n);
trace("IEnumCodePage_Next = %08x, n = %u\n", ret, n);
ok(ret == S_FALSE && n == 0, "IEnumCodePage_Next: expected S_FALSE/0, got %08x/%u\n", ret, n);
n = total * 2;
TRACE_2("Call IEnumCodePage_Next\n");
ret = IEnumCodePage_Next(iEnumCP, n, cpinfo, &n);
ok(ret == S_OK && n != 0, "IEnumCodePage_Next: expected S_OK/!0, got %08x/%u\n", ret, n);
trace("flags %08x, enumerated codepages %u\n", flags, n);
if (!flags)
{
ok(n == total, "IEnumCodePage_Next: expected %u, got %u\n", total, n);
flags = MIMECONTF_MIME_LATEST;
}
total = n;
for (i = 0; i < n; i++)
{
CPINFOEXA cpinfoex;
CHARSETINFO csi;
MIMECSETINFO mcsi;
static const WCHAR autoW[] = {'_','a','u','t','o',0};
#ifdef DUMP_CP_INFO
trace("MIMECPINFO #%u:\n"
"dwFlags %08x %s\n"
"uiCodePage %u\n"
"uiFamilyCodePage %u\n"
"wszDescription %s\n"
"wszWebCharset %s\n"
"wszHeaderCharset %s\n"
"wszBodyCharset %s\n"
"wszFixedWidthFont %s\n"
"wszProportionalFont %s\n"
"bGDICharset %d\n\n",
i,
cpinfo[i].dwFlags, dump_mime_flags(cpinfo[i].dwFlags),
cpinfo[i].uiCodePage,
cpinfo[i].uiFamilyCodePage,
wine_dbgstr_w(cpinfo[i].wszDescription),
wine_dbgstr_w(cpinfo[i].wszWebCharset),
wine_dbgstr_w(cpinfo[i].wszHeaderCharset),
wine_dbgstr_w(cpinfo[i].wszBodyCharset),
wine_dbgstr_w(cpinfo[i].wszFixedWidthFont),
wine_dbgstr_w(cpinfo[i].wszProportionalFont),
cpinfo[i].bGDICharset);
#endif
ok(cpinfo[i].dwFlags & flags, "enumerated flags %08x do not include requested %08x\n", cpinfo[i].dwFlags, flags);
if (TranslateCharsetInfo((DWORD *)cpinfo[i].uiFamilyCodePage, &csi, TCI_SRCCODEPAGE))
ok(cpinfo[i].bGDICharset == csi.ciCharset, "%d != %d\n", cpinfo[i].bGDICharset, csi.ciCharset);
else
trace("TranslateCharsetInfo failed for cp %u\n", cpinfo[i].uiFamilyCodePage);
if (pGetCPInfoExA)
{
if (pGetCPInfoExA(cpinfo[i].uiCodePage, 0, &cpinfoex))
trace("CodePage %u name: %s\n", cpinfo[i].uiCodePage, cpinfoex.CodePageName);
else
trace("GetCPInfoExA failed for cp %u\n", cpinfo[i].uiCodePage);
if (pGetCPInfoExA(cpinfo[i].uiFamilyCodePage, 0, &cpinfoex))
trace("CodePage %u name: %s\n", cpinfo[i].uiFamilyCodePage, cpinfoex.CodePageName);
else
trace("GetCPInfoExA failed for cp %u\n", cpinfo[i].uiFamilyCodePage);
}
/* Win95 does not support UTF-7 */
if (cpinfo[i].uiCodePage == CP_UTF7) continue;
/* support files for some codepages might be not installed, or
* the codepage is just an alias.
*/
if (IsValidCodePage(cpinfo[i].uiCodePage))
{
TRACE_2("Call IMultiLanguage2_IsConvertible\n");
ret = IMultiLanguage2_IsConvertible(iML2, cpinfo[i].uiCodePage, CP_UNICODE);
ok(ret == S_OK, "IMultiLanguage2_IsConvertible(%u -> CP_UNICODE) = %08x\n", cpinfo[i].uiCodePage, ret);
TRACE_2("Call IMultiLanguage2_IsConvertible\n");
ret = IMultiLanguage2_IsConvertible(iML2, CP_UNICODE, cpinfo[i].uiCodePage);
ok(ret == S_OK, "IMultiLanguage2_IsConvertible(CP_UNICODE -> %u) = %08x\n", cpinfo[i].uiCodePage, ret);
TRACE_2("Call IMultiLanguage2_IsConvertible\n");
ret = IMultiLanguage2_IsConvertible(iML2, cpinfo[i].uiCodePage, CP_UTF8);
ok(ret == S_OK, "IMultiLanguage2_IsConvertible(%u -> CP_UTF8) = %08x\n", cpinfo[i].uiCodePage, ret);
TRACE_2("Call IMultiLanguage2_IsConvertible\n");
ret = IMultiLanguage2_IsConvertible(iML2, CP_UTF8, cpinfo[i].uiCodePage);
ok(ret == S_OK, "IMultiLanguage2_IsConvertible(CP_UTF8 -> %u) = %08x\n", cpinfo[i].uiCodePage, ret);
}
else
trace("IsValidCodePage failed for cp %u\n", cpinfo[i].uiCodePage);
ret = IMultiLanguage2_GetCharsetInfo(iML2, cpinfo[i].wszWebCharset, &mcsi);
/* _autoxxx charsets are a fake and GetCharsetInfo fails for them */
if (memcmp(cpinfo[i].wszWebCharset, autoW, 5 * sizeof(WCHAR)))
{
ok (ret == S_OK, "IMultiLanguage2_GetCharsetInfo failed: %08x\n", ret);
#ifdef DUMP_CP_INFO
trace("%s: %u %u %s\n", wine_dbgstr_w(cpinfo[i].wszWebCharset), mcsi.uiCodePage, mcsi.uiInternetEncoding, wine_dbgstr_w(mcsi.wszCharset));
#endif
ok(!lstrcmpiW(cpinfo[i].wszWebCharset, mcsi.wszCharset),
#ifdef DUMP_CP_INFO
"%s != %s\n",
wine_dbgstr_w(cpinfo[i].wszWebCharset), wine_dbgstr_w(mcsi.wszCharset));
#else
"wszWebCharset mismatch");
#endif
if (0)
{
/* native mlang returns completely messed up encodings in some cases */
ok(mcsi.uiInternetEncoding == cpinfo[i].uiCodePage || mcsi.uiInternetEncoding == cpinfo[i].uiFamilyCodePage,
"%u != %u || %u\n", mcsi.uiInternetEncoding, cpinfo[i].uiCodePage, cpinfo[i].uiFamilyCodePage);
ok(mcsi.uiCodePage == cpinfo[i].uiCodePage || mcsi.uiCodePage == cpinfo[i].uiFamilyCodePage,
"%u != %u || %u\n", mcsi.uiCodePage, cpinfo[i].uiCodePage, cpinfo[i].uiFamilyCodePage);
}
}
ret = IMultiLanguage2_GetCharsetInfo(iML2, cpinfo[i].wszHeaderCharset, &mcsi);
/* _autoxxx charsets are a fake and GetCharsetInfo fails for them */
if (memcmp(cpinfo[i].wszHeaderCharset, autoW, 5 * sizeof(WCHAR)))
{
ok (ret == S_OK, "IMultiLanguage2_GetCharsetInfo failed: %08x\n", ret);
#ifdef DUMP_CP_INFO
trace("%s: %u %u %s\n", wine_dbgstr_w(cpinfo[i].wszHeaderCharset), mcsi.uiCodePage, mcsi.uiInternetEncoding, wine_dbgstr_w(mcsi.wszCharset));
#endif
ok(!lstrcmpiW(cpinfo[i].wszHeaderCharset, mcsi.wszCharset),
#ifdef DUMP_CP_INFO
"%s != %s\n",
wine_dbgstr_w(cpinfo[i].wszHeaderCharset), wine_dbgstr_w(mcsi.wszCharset));
#else
"wszHeaderCharset mismatch");
#endif
if (0)
{
/* native mlang returns completely messed up encodings in some cases */
ok(mcsi.uiInternetEncoding == cpinfo[i].uiCodePage || mcsi.uiInternetEncoding == cpinfo[i].uiFamilyCodePage,
"%u != %u || %u\n", mcsi.uiInternetEncoding, cpinfo[i].uiCodePage, cpinfo[i].uiFamilyCodePage);
ok(mcsi.uiCodePage == cpinfo[i].uiCodePage || mcsi.uiCodePage == cpinfo[i].uiFamilyCodePage,
"%u != %u || %u\n", mcsi.uiCodePage, cpinfo[i].uiCodePage, cpinfo[i].uiFamilyCodePage);
}
}
ret = IMultiLanguage2_GetCharsetInfo(iML2, cpinfo[i].wszBodyCharset, &mcsi);
/* _autoxxx charsets are a fake and GetCharsetInfo fails for them */
if (memcmp(cpinfo[i].wszBodyCharset, autoW, 5 * sizeof(WCHAR)))
{
ok (ret == S_OK, "IMultiLanguage2_GetCharsetInfo failed: %08x\n", ret);
#ifdef DUMP_CP_INFO
trace("%s: %u %u %s\n", wine_dbgstr_w(cpinfo[i].wszBodyCharset), mcsi.uiCodePage, mcsi.uiInternetEncoding, wine_dbgstr_w(mcsi.wszCharset));
#endif
ok(!lstrcmpiW(cpinfo[i].wszBodyCharset, mcsi.wszCharset),
#ifdef DUMP_CP_INFO
"%s != %s\n",
wine_dbgstr_w(cpinfo[i].wszBodyCharset), wine_dbgstr_w(mcsi.wszCharset));
#else
"wszBodyCharset mismatch");
#endif
if (0)
{
/* native mlang returns completely messed up encodings in some cases */
ok(mcsi.uiInternetEncoding == cpinfo[i].uiCodePage || mcsi.uiInternetEncoding == cpinfo[i].uiFamilyCodePage,
"%u != %u || %u\n", mcsi.uiInternetEncoding, cpinfo[i].uiCodePage, cpinfo[i].uiFamilyCodePage);
ok(mcsi.uiCodePage == cpinfo[i].uiCodePage || mcsi.uiCodePage == cpinfo[i].uiFamilyCodePage,
"%u != %u || %u\n", mcsi.uiCodePage, cpinfo[i].uiCodePage, cpinfo[i].uiFamilyCodePage);
}
}
trace("---\n");
}
/* now IEnumCodePage_Next should fail, since pointer is at the end */
n = 1;
ret = IEnumCodePage_Next(iEnumCP, 1, &cpinfo2, &n);
ok(ret == S_FALSE && n == 0, "IEnumCodePage_Next: expected S_FALSE/0, got %08x/%u\n", ret, n);
ret = IEnumCodePage_Reset(iEnumCP);
ok(ret == S_OK, "IEnumCodePage_Reset: expected S_OK, got %08x\n", ret);
n = 0;
ret = IEnumCodePage_Next(iEnumCP, 1, &cpinfo2, &n);
ok(n == 1 && ret == S_OK, "IEnumCodePage_Next: expected 1/S_OK, got %u/%08x\n", n, ret);
cpinfo_cmp(&cpinfo[0], &cpinfo2);
if (0)
{
/* Due to a bug in MS' implementation of IEnumCodePage_Skip
* it's not used here.
*/
ret = IEnumCodePage_Skip(iEnumCP, 1);
ok(ret == S_OK, "IEnumCodePage_Skip: expected S_OK, got %08x\n", ret);
}
for (i = 0; i < total - 1; i++)
{
n = 0;
ret = IEnumCodePage_Next(iEnumCP, 1, &cpinfo2, &n);
ok(n == 1 && ret == S_OK, "IEnumCodePage_Next: expected 1/S_OK, got %u/%08x\n", n, ret);
cpinfo_cmp(&cpinfo[i + 1], &cpinfo2);
}
HeapFree(GetProcessHeap(), 0, cpinfo);
IEnumCodePage_Release(iEnumCP);
}
inline static void scriptinfo_cmp(SCRIPTINFO *sinfo1, SCRIPTINFO *sinfo2)
{
ok(sinfo1->ScriptId == sinfo2->ScriptId, "ScriptId mismatch: %d != %d\n", sinfo1->ScriptId, sinfo2->ScriptId);
ok(sinfo1->uiCodePage == sinfo2->uiCodePage, "uiCodePage mismatch: %u != %u\n", sinfo1->uiCodePage, sinfo2->uiCodePage);
ok(!lstrcmpW(sinfo1->wszDescription, sinfo2->wszDescription), "wszDescription mismatch\n");
ok(!lstrcmpW(sinfo1->wszFixedWidthFont, sinfo2->wszFixedWidthFont), "wszFixedWidthFont mismatch\n");
ok(!lstrcmpW(sinfo1->wszProportionalFont, sinfo2->wszProportionalFont), "wszProportionalFont mismatch\n");
}
static void test_EnumScripts(IMultiLanguage2 *iML2, DWORD flags)
{
IEnumScript *iEnumScript = NULL;
SCRIPTINFO *sinfo;
SCRIPTINFO sinfo2;
HRESULT ret;
ULONG i, n;
UINT total;
total = 0;
TRACE_2("Call IMultiLanguage2_GetNumberOfScripts\n");
ret = IMultiLanguage2_GetNumberOfScripts(iML2, &total);
ok(ret == S_OK && total != 0, "IMultiLanguage2_GetNumberOfScripts: expected S_OK/!0, got %08x/%u\n", ret, total);
trace("total mlang supported scripts %u\n", total);
TRACE_2("Call IMultiLanguage2_EnumScripts\n");
ret = IMultiLanguage2_EnumScripts(iML2, flags, LANG_NEUTRAL, &iEnumScript);
trace("IMultiLanguage2_EnumScripts = %08x, iEnumScript = %p\n", ret, iEnumScript);
ok(ret == S_OK && iEnumScript, "IMultiLanguage2_EnumScripts: expected S_OK/!NULL, got %08x/%p\n", ret, iEnumScript);
TRACE_2("Call IEnumScript_Reset\n");
ret = IEnumScript_Reset(iEnumScript);
ok(ret == S_OK, "IEnumScript_Reset: expected S_OK, got %08x\n", ret);
n = 65536;
TRACE_2("Call IEnumScript_Next\n");
ret = IEnumScript_Next(iEnumScript, 0, NULL, &n);
ok(n == 65536 && ret == E_FAIL, "IEnumScript_Next: expected 65536/E_FAIL, got %u/%08x\n", n, ret);
TRACE_2("Call IEnumScript_Next\n");
ret = IEnumScript_Next(iEnumScript, 0, NULL, NULL);
ok(ret == E_FAIL, "IEnumScript_Next: expected E_FAIL, got %08x\n", ret);
sinfo = HeapAlloc(GetProcessHeap(), 0, sizeof(*sinfo) * total * 2);
n = total * 2;
TRACE_2("Call IEnumScript_Next\n");
ret = IEnumScript_Next(iEnumScript, 0, sinfo, &n);
ok(ret == S_FALSE && n == 0, "IEnumScript_Next: expected S_FALSE/0, got %08x/%u\n", ret, n);
n = total * 2;
TRACE_2("Call IEnumScript_Next\n");
ret = IEnumScript_Next(iEnumScript, n, sinfo, &n);
ok(ret == S_OK && n != 0, "IEnumScript_Next: expected S_OK, got %08x/%u\n", ret, n);
trace("flags %08x, enumerated scripts %u\n", flags, n);
if (!flags)
{
ok(n == total, "IEnumScript_Next: expected %u, got %u\n", total, n);
flags = SCRIPTCONTF_SCRIPT_USER | SCRIPTCONTF_SCRIPT_HIDE | SCRIPTCONTF_SCRIPT_SYSTEM;
}
total = n;
for (i = 0; pGetCPInfoExA && i < n; i++)
{
CPINFOEXA cpinfoex;
#ifdef DUMP_SCRIPT_INFO
trace("SCRIPTINFO #%u:\n"
"ScriptId %08x\n"
"uiCodePage %u\n"
"wszDescription %s\n"
"wszFixedWidthFont %s\n"
"wszProportionalFont %s\n\n",
i,
sinfo[i].ScriptId,
sinfo[i].uiCodePage,
wine_dbgstr_w(sinfo[i].wszDescription),
wine_dbgstr_w(sinfo[i].wszFixedWidthFont),
wine_dbgstr_w(sinfo[i].wszProportionalFont));
#endif
if (pGetCPInfoExA(sinfo[i].uiCodePage, 0, &cpinfoex))
trace("CodePage %u name: %s\n", sinfo[i].uiCodePage, cpinfoex.CodePageName);
else
trace("GetCPInfoExA failed for cp %u\n", sinfo[i].uiCodePage);
trace("---\n");
}
/* now IEnumScript_Next should fail, since pointer is at the end */
n = 1;
ret = IEnumScript_Next(iEnumScript, 1, &sinfo2, &n);
ok(ret == S_FALSE && n == 0, "IEnumScript_Next: expected S_FALSE/0, got %08x/%u\n", ret, n);
ret = IEnumScript_Reset(iEnumScript);
ok(ret == S_OK, "IEnumScript_Reset: expected S_OK, got %08x\n", ret);
n = 0;
ret = IEnumScript_Next(iEnumScript, 1, &sinfo2, &n);
ok(n == 1 && ret == S_OK, "IEnumScript_Next: expected 1/S_OK, got %u/%08x\n", n, ret);
scriptinfo_cmp(&sinfo[0], &sinfo2);
if (0)
{
/* Due to a bug in MS' implementation of IEnumScript_Skip
* it's not used here.
*/
ret = IEnumScript_Skip(iEnumScript, 1);
ok(ret == S_OK, "IEnumScript_Skip: expected S_OK, got %08x\n", ret);
}
for (i = 0; i < total - 1; i++)
{
n = 0;
ret = IEnumScript_Next(iEnumScript, 1, &sinfo2, &n);
ok(n == 1 && ret == S_OK, "IEnumScript_Next: expected 1/S_OK, got %u/%08x\n", n, ret);
scriptinfo_cmp(&sinfo[i + 1], &sinfo2);
}
HeapFree(GetProcessHeap(), 0, sinfo);
IEnumScript_Release(iEnumScript);
}
static void IMLangFontLink_Test(IMLangFontLink* iMLFL)
{
DWORD dwCodePages = 0;
DWORD dwManyCodePages = 0;
UINT CodePage = 0;
ok(IMLangFontLink_CodePageToCodePages(iMLFL, 932, &dwCodePages)==S_OK,
"IMLangFontLink_CodePageToCodePages failed\n");
ok (dwCodePages != 0, "No CodePages returned\n");
ok(IMLangFontLink_CodePagesToCodePage(iMLFL, dwCodePages, 1035,
&CodePage)==S_OK,
"IMLangFontLink_CodePagesToCodePage failed\n");
ok(CodePage == 932, "Incorrect CodePage Returned (%i)\n",CodePage);
ok(IMLangFontLink_CodePageToCodePages(iMLFL, 1252, &dwCodePages)==S_OK,
"IMLangFontLink_CodePageToCodePages failed\n");
dwManyCodePages = dwManyCodePages | dwCodePages;
ok(IMLangFontLink_CodePageToCodePages(iMLFL, 1256, &dwCodePages)==S_OK,
"IMLangFontLink_CodePageToCodePages failed\n");
dwManyCodePages = dwManyCodePages | dwCodePages;
ok(IMLangFontLink_CodePageToCodePages(iMLFL, 874, &dwCodePages)==S_OK,
"IMLangFontLink_CodePageToCodePages failed\n");
dwManyCodePages = dwManyCodePages | dwCodePages;
ok(IMLangFontLink_CodePagesToCodePage(iMLFL, dwManyCodePages, 1256,
&CodePage)==S_OK,
"IMLangFontLink_CodePagesToCodePage failed\n");
ok(CodePage == 1256, "Incorrect CodePage Returned (%i)\n",CodePage);
ok(IMLangFontLink_CodePagesToCodePage(iMLFL, dwManyCodePages, 936,
&CodePage)==S_OK,
"IMLangFontLink_CodePagesToCodePage failed\n");
ok(CodePage == 1252, "Incorrect CodePage Returned (%i)\n",CodePage);
}
static void test_rfc1766(IMultiLanguage2 *iML2)
{
IEnumRfc1766 *pEnumRfc1766;
RFC1766INFO info;
ULONG n;
HRESULT ret;
ret = IMultiLanguage2_EnumRfc1766(iML2, LANG_NEUTRAL, &pEnumRfc1766);
ok(ret == S_OK, "IMultiLanguage2_EnumRfc1766 error %08x\n", ret);
while (1)
{
ret = IEnumRfc1766_Next(pEnumRfc1766, 1, &info, &n);
if (ret != S_OK) break;
#ifdef DUMP_CP_INFO
trace("lcid %04x rfc_name %s locale_name %s\n",
info.lcid, wine_dbgstr_w(info.wszRfc1766), wine_dbgstr_w(info.wszLocaleName));
#endif
ok(n == 1, "couldn't fetch 1 RFC1766INFO structure\n");
ok(IsValidLocale(info.lcid, LCID_SUPPORTED), "invalid lcid %04x\n", info.lcid);
}
}
START_TEST(mlang)
{
IMultiLanguage2 *iML2 = NULL;
IMLangFontLink *iMLFL = NULL;
HRESULT ret;
pGetCPInfoExA = (void *)GetProcAddress(GetModuleHandleA("kernel32.dll"), "GetCPInfoExA");
CoInitialize(NULL);
TRACE_2("Call CoCreateInstance\n");
ret = CoCreateInstance(&CLSID_CMultiLanguage, NULL, CLSCTX_INPROC_SERVER,
&IID_IMultiLanguage2, (void **)&iML2);
trace("ret = %08x, MultiLanguage2 iML2 = %p\n", ret, iML2);
if (ret != S_OK || !iML2) return;
test_rfc1766(iML2);
test_EnumCodePages(iML2, 0);
test_EnumCodePages(iML2, MIMECONTF_MIME_LATEST);
test_EnumCodePages(iML2, MIMECONTF_BROWSER);
test_EnumCodePages(iML2, MIMECONTF_MINIMAL);
test_EnumCodePages(iML2, MIMECONTF_VALID);
/* FIXME: why MIMECONTF_MIME_REGISTRY returns 0 of supported codepages? */
/*test_EnumCodePages(iML2, MIMECONTF_MIME_REGISTRY);*/
test_EnumScripts(iML2, 0);
test_EnumScripts(iML2, SCRIPTCONTF_SCRIPT_USER);
test_EnumScripts(iML2, SCRIPTCONTF_SCRIPT_USER | SCRIPTCONTF_SCRIPT_HIDE | SCRIPTCONTF_SCRIPT_SYSTEM);
TRACE_2("Call IMultiLanguage2_IsConvertible\n");
ret = IMultiLanguage2_IsConvertible(iML2, CP_UTF8, CP_UNICODE);
ok(ret == S_OK, "IMultiLanguage2_IsConvertible(CP_UTF8 -> CP_UNICODE) = %08x\n", ret);
TRACE_2("Call IMultiLanguage2_IsConvertible\n");
ret = IMultiLanguage2_IsConvertible(iML2, CP_UNICODE, CP_UTF8);
ok(ret == S_OK, "IMultiLanguage2_IsConvertible(CP_UNICODE -> CP_UTF8) = %08x\n", ret);
test_multibyte_to_unicode_translations(iML2);
IMultiLanguage2_Release(iML2);
ret = CoCreateInstance(&CLSID_CMultiLanguage, NULL, CLSCTX_INPROC_SERVER,
&IID_IMLangFontLink, (void **)&iMLFL);
trace("ret = %08x, IMLangFontLink iMLFL = %p\n", ret, iMLFL);
if (ret != S_OK || !iML2) return;
IMLangFontLink_Test(iMLFL);
IMLangFontLink_Release(iMLFL);
CoUninitialize();
}