reactos/rostests/apitests/kernel32/WideCharToMultiByte.c

176 lines
6.9 KiB
C
Raw Normal View History

/*
* PROJECT: ReactOS API tests
* LICENSE: LGPLv2.1+ - See COPYING.LIB in the top level directory
* PURPOSE: Test for WideCharToMultiByte
* PROGRAMMER: Thomas Faber <thomas.faber@reactos.org>
*/
#include <apitest.h>
#include <winnls.h>
#define ntv6(x) (LOBYTE(LOWORD(GetVersion())) >= 6 ? (x) : 0)
static
VOID
Utf8Convert_(
_In_ PCWSTR WideString,
_In_ PCSTR ExpectedUtf8_2003,
_In_ PCSTR ExpectedUtf8_Vista,
_In_ BOOL IsInvalid,
_In_ PCSTR File,
_In_ INT Line)
{
size_t WideLen;
size_t Utf8Len;
char Buffer[32];
int Ret;
ULONG i;
ULONG Error;
PCSTR ExpectedUtf8;
ExpectedUtf8 = ntv6(1) ? ExpectedUtf8_Vista : ExpectedUtf8_2003;
WideLen = wcslen(WideString);
Utf8Len = strlen(ExpectedUtf8);
/* Get length only */
Ret = WideCharToMultiByte(CP_UTF8, 0, WideString, WideLen, NULL, 0, NULL, NULL);
ok_(File, Line)(Ret == Utf8Len, "Length check: Ret = %d\n", Ret);
/* Get length including nul */
Ret = WideCharToMultiByte(CP_UTF8, 0, WideString, WideLen + 1, NULL, 0, NULL, NULL);
ok_(File, Line)(Ret == Utf8Len + 1, "Length check with null: Ret = %d\n", Ret);
/* Convert, excluding null */
FillMemory(Buffer, sizeof(Buffer), 0x55);
Ret = WideCharToMultiByte(CP_UTF8, 0, WideString, WideLen, Buffer, sizeof(Buffer), NULL, NULL);
ok_(File, Line)(Ret == Utf8Len, "Convert: Ret = %d\n", Ret);
for (i = 0; i < Utf8Len; i++)
{
ok_(File, Line)(Buffer[i] == ExpectedUtf8[i], "Convert: Buffer[%lu] = 0x%x, expected 0x%x\n", i, (unsigned char)Buffer[i], (unsigned char)ExpectedUtf8[i]);
}
/* Convert, including null */
FillMemory(Buffer, sizeof(Buffer), 0x55);
Ret = WideCharToMultiByte(CP_UTF8, 0, WideString, WideLen + 1, Buffer, sizeof(Buffer), NULL, NULL);
ok_(File, Line)(Ret == Utf8Len + 1, "Convert with null: Ret = %d\n", Ret);
for (i = 0; i < Utf8Len + 1; i++)
{
ok_(File, Line)(Buffer[i] == ExpectedUtf8[i], "Convert with null: Buffer[%lu] = 0x%x, expected 0x%x\n", i, (unsigned char)Buffer[i], (unsigned char)ExpectedUtf8[i]);
}
/* Get length, reject invalid */
SetLastError(0xfeedf00d);
Ret = WideCharToMultiByte(CP_UTF8, WC_ERR_INVALID_CHARS, WideString, WideLen + 1, NULL, 0, NULL, NULL);
Error = GetLastError();
if (!ntv6(1))
{
ok_(File, Line)(Ret == 0, "Length check, reject invalid, NT5: Ret = %d\n", Ret);
ok_(File, Line)(Error == ERROR_INVALID_FLAGS, "Length check, reject invalid, NT5: Error = %lu\n", Error);
}
else if (IsInvalid)
{
ok_(File, Line)(Ret == 0, "Length check, reject invalid: Ret = %d\n", Ret);
ok_(File, Line)(Error == ERROR_NO_UNICODE_TRANSLATION, "Length check, reject invalid: Error = %lu\n", Error);
}
else
{
ok_(File, Line)(Ret == Utf8Len + 1, "Length check, reject invalid: Ret = %d\n", Ret);
}
/* Convert, reject invalid */
FillMemory(Buffer, sizeof(Buffer), 0x55);
SetLastError(0xfeedf00d);
Ret = WideCharToMultiByte(CP_UTF8, WC_ERR_INVALID_CHARS, WideString, WideLen + 1, Buffer, sizeof(Buffer), NULL, NULL);
Error = GetLastError();
if (!ntv6(1))
{
ok_(File, Line)(Ret == 0, "Convert, reject invalid, NT5: Ret = %d\n", Ret);
ok_(File, Line)(Error == ERROR_INVALID_FLAGS, "Convert, reject invalid, NT5: Error = %lu\n", Error);
}
else if (IsInvalid)
{
ok_(File, Line)(Ret == 0, "Convert, reject invalid: Ret = %d\n", Ret);
ok_(File, Line)(Error == ERROR_NO_UNICODE_TRANSLATION, "Convert, reject invalid: Error = %lu\n", Error);
for (i = 0; i < Utf8Len + 1; i++)
{
ok_(File, Line)(Buffer[i] == ExpectedUtf8[i], "Convert, reject invalid: Buffer[%lu] = 0x%x, expected 0x%x\n", i, (unsigned char)Buffer[i], (unsigned char)ExpectedUtf8[i]);
}
}
else
{
ok_(File, Line)(Ret == Utf8Len + 1, "Convert, reject invalid: Ret = %d\n", Ret);
for (i = 0; i < Utf8Len + 1; i++)
{
ok_(File, Line)(Buffer[i] == ExpectedUtf8[i], "Convert, reject invalid: Buffer[%lu] = 0x%x, expected 0x%x\n", i, (unsigned char)Buffer[i], (unsigned char)ExpectedUtf8[i]);
}
}
}
#define Utf8Convert(w, e, i) Utf8Convert_(w, e, e, i, __FILE__, __LINE__)
#define Utf8Convert_Vista(w, e, i, e2) Utf8Convert_(w, e, e2, i, __FILE__, __LINE__)
static
VOID
TestUtf8(VOID)
{
Utf8Convert(L"", "", FALSE);
/* Various character ranges */
Utf8Convert(L"A", "A", FALSE);
Utf8Convert(L"\x007f", "\x7f", FALSE);
Utf8Convert(L"\x0080", "\xc2\x80", FALSE);
Utf8Convert(L"\x00ff", "\xc3\xbf", FALSE);
Utf8Convert(L"\x0100", "\xc4\x80", FALSE);
Utf8Convert(L"\x07ff", "\xdf\xbf", FALSE);
Utf8Convert(L"\x0800", "\xe0\xa0\x80", FALSE);
Utf8Convert(L"\xd7ff", "\xed\x9f\xbf", FALSE);
Utf8Convert(L"\xe000", "\xee\x80\x80", FALSE);
Utf8Convert(L"\xffff", "\xef\xbf\xbf", FALSE);
/* surrogate pairs */
Utf8Convert(L"\xd800\xdc00", "\xf0\x90\x80\x80", FALSE); /* U+10000 */
Utf8Convert(L"\xd800\xdfff", "\xf0\x90\x8f\xbf", FALSE); /* U+103ff */
Utf8Convert(L"\xd801\xdc00", "\xf0\x90\x90\x80", FALSE); /* U+10400 */
Utf8Convert(L"\xdbff\xdfff", "\xf4\x8f\xbf\xbf", FALSE); /* U+10ffff */
/* standalone lead surrogate becomes 0xfffd on Vista, goes through verbatim on 2003 */
Utf8Convert_Vista(L"\xd800", "\xed\xa0\x80", TRUE,
"\xef\xbf\xbd");
Utf8Convert_Vista(L"\xd800-", "\xed\xa0\x80-", TRUE,
"\xef\xbf\xbd-");
Utf8Convert_Vista(L"\xdbff", "\xed\xaf\xbf", TRUE,
"\xef\xbf\xbd");
Utf8Convert_Vista(L"\xdbff-", "\xed\xaf\xbf-", TRUE,
"\xef\xbf\xbd-");
/* standalone trail surrogate becomes 0xfffd */
Utf8Convert_Vista(L"\xdc00", "\xed\xb0\x80", TRUE,
"\xef\xbf\xbd");
Utf8Convert_Vista(L"\xdc00-", "\xed\xb0\x80-", TRUE,
"\xef\xbf\xbd-");
Utf8Convert_Vista(L"\xdfff", "\xed\xbf\xbf", TRUE,
"\xef\xbf\xbd");
Utf8Convert_Vista(L"\xdfff-", "\xed\xbf\xbf-", TRUE,
"\xef\xbf\xbd-");
/* Reverse surrogate pair */
Utf8Convert_Vista(L"\xdfff\xdbff", "\xed\xbf\xbf\xed\xaf\xbf", TRUE,
"\xef\xbf\xbd\xef\xbf\xbd");
/* Byte order marks */
Utf8Convert(L"\xfeff", "\xef\xbb\xbf", FALSE);
Utf8Convert(L"\xfffe", "\xef\xbf\xbe", FALSE);
/* canonically equivalent representations -- no normalization should happen */
Utf8Convert(L"\x1e09", "\xe1\xb8\x89", FALSE);
Utf8Convert(L"\x0107\x0327", "\xc4\x87\xcc\xa7", FALSE);
Utf8Convert(L"\x00e7\x0301", "\xc3\xa7\xcc\x81", FALSE);
Utf8Convert(L"\x0063\x0327\x0301", "\x63\xcc\xa7\xcc\x81", FALSE);
Utf8Convert(L"\x0063\x0301\x0327", "\x63\xcc\x81\xcc\xa7", FALSE);
}
START_TEST(WideCharToMultiByte)
{
TestUtf8();
}