mirror of
https://github.com/reactos/reactos.git
synced 2024-11-18 13:01:40 +00:00
2568 lines
127 KiB
C
2568 lines
127 KiB
C
/* Unit test suite for Rtl string functions
|
|
*
|
|
* Copyright 2002 Robert Shearman
|
|
* Copyright 2003 Thomas Mertes
|
|
*
|
|
* 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
|
|
*
|
|
* NOTES
|
|
* We use function pointers here as there is no import library for NTDLL on
|
|
* windows.
|
|
*/
|
|
|
|
#include <stdlib.h>
|
|
|
|
#define INITGUID
|
|
|
|
#include "ntdll_test.h"
|
|
#include "winnls.h"
|
|
#include "guiddef.h"
|
|
|
|
#define HASH_STRING_ALGORITHM_X65599 1
|
|
#define HASH_STRING_ALGORITHM_INVALID 0xffffffff
|
|
|
|
/* Function ptrs for ntdll calls */
|
|
static HMODULE hntdll = 0;
|
|
static NTSTATUS (WINAPI *pRtlAnsiStringToUnicodeString)(PUNICODE_STRING, PCANSI_STRING, BOOLEAN);
|
|
static NTSTATUS (WINAPI *pRtlAppendAsciizToString)(STRING *, LPCSTR);
|
|
static NTSTATUS (WINAPI *pRtlAppendStringToString)(STRING *, const STRING *);
|
|
static NTSTATUS (WINAPI *pRtlAppendUnicodeStringToString)(UNICODE_STRING *, const UNICODE_STRING *);
|
|
static NTSTATUS (WINAPI *pRtlAppendUnicodeToString)(UNICODE_STRING *, LPCWSTR);
|
|
static NTSTATUS (WINAPI *pRtlCharToInteger)(PCSZ, ULONG, int *);
|
|
static LONG (WINAPI *pRtlCompareUnicodeString)(const UNICODE_STRING*, const UNICODE_STRING*, BOOLEAN);
|
|
static LONG (WINAPI *pRtlCompareUnicodeStrings)(const WCHAR *,SIZE_T,const WCHAR *,SIZE_T,BOOLEAN);
|
|
static VOID (WINAPI *pRtlCopyString)(STRING *, const STRING *);
|
|
static BOOLEAN (WINAPI *pRtlCreateUnicodeString)(PUNICODE_STRING, LPCWSTR);
|
|
static BOOLEAN (WINAPI *pRtlCreateUnicodeStringFromAsciiz)(PUNICODE_STRING, LPCSTR);
|
|
static NTSTATUS (WINAPI *pRtlDowncaseUnicodeString)(UNICODE_STRING *, const UNICODE_STRING *, BOOLEAN);
|
|
static NTSTATUS (WINAPI *pRtlDuplicateUnicodeString)(int, UNICODE_STRING *, UNICODE_STRING *);
|
|
static BOOLEAN (WINAPI *pRtlEqualUnicodeString)(const UNICODE_STRING *, const UNICODE_STRING *, BOOLEAN);
|
|
static NTSTATUS (WINAPI *pRtlFindCharInUnicodeString)(int, const UNICODE_STRING *, const UNICODE_STRING *, USHORT *);
|
|
static VOID (WINAPI *pRtlFreeAnsiString)(PSTRING);
|
|
static VOID (WINAPI *pRtlFreeUnicodeString)(PUNICODE_STRING);
|
|
static VOID (WINAPI *pRtlInitAnsiString)(PSTRING, LPCSTR);
|
|
static VOID (WINAPI *pRtlInitString)(PSTRING, LPCSTR);
|
|
static VOID (WINAPI *pRtlInitUnicodeString)(PUNICODE_STRING, LPCWSTR);
|
|
static NTSTATUS (WINAPI *pRtlInitUnicodeStringEx)(PUNICODE_STRING, LPCWSTR);
|
|
static NTSTATUS (WINAPI *pRtlIntegerToChar)(ULONG, ULONG, ULONG, PCHAR);
|
|
static NTSTATUS (WINAPI *pRtlIntegerToUnicodeString)(ULONG, ULONG, UNICODE_STRING *);
|
|
static NTSTATUS (WINAPI *pRtlMultiAppendUnicodeStringBuffer)(UNICODE_STRING *, LONG, UNICODE_STRING *);
|
|
static NTSTATUS (WINAPI *pRtlUnicodeStringToAnsiString)(STRING *, const UNICODE_STRING *, BOOLEAN);
|
|
static NTSTATUS (WINAPI *pRtlUnicodeStringToInteger)(const UNICODE_STRING *, int, int *);
|
|
static WCHAR (WINAPI *pRtlUpcaseUnicodeChar)(WCHAR);
|
|
static NTSTATUS (WINAPI *pRtlUpcaseUnicodeString)(UNICODE_STRING *, const UNICODE_STRING *, BOOLEAN);
|
|
static CHAR (WINAPI *pRtlUpperChar)(CHAR);
|
|
static NTSTATUS (WINAPI *pRtlUpperString)(STRING *, const STRING *);
|
|
static NTSTATUS (WINAPI *pRtlValidateUnicodeString)(LONG, UNICODE_STRING *);
|
|
static NTSTATUS (WINAPI *pRtlGUIDFromString)(const UNICODE_STRING*,GUID*);
|
|
static NTSTATUS (WINAPI *pRtlStringFromGUID)(const GUID*, UNICODE_STRING*);
|
|
static BOOLEAN (WINAPI *pRtlIsTextUnicode)(LPVOID, INT, INT *);
|
|
static NTSTATUS (WINAPI *pRtlHashUnicodeString)(PCUNICODE_STRING,BOOLEAN,ULONG,ULONG*);
|
|
static NTSTATUS (WINAPI *pRtlUnicodeToUTF8N)(CHAR *, ULONG, ULONG *, const WCHAR *, ULONG);
|
|
static NTSTATUS (WINAPI *pRtlUTF8ToUnicodeN)(WCHAR *, ULONG, ULONG *, const CHAR *, ULONG);
|
|
|
|
/*static VOID (WINAPI *pRtlFreeOemString)(PSTRING);*/
|
|
/*static VOID (WINAPI *pRtlCopyUnicodeString)(UNICODE_STRING *, const UNICODE_STRING *);*/
|
|
/*static VOID (WINAPI *pRtlEraseUnicodeString)(UNICODE_STRING *);*/
|
|
/*static LONG (WINAPI *pRtlCompareString)(const STRING *,const STRING *,BOOLEAN);*/
|
|
/*static BOOLEAN (WINAPI *pRtlEqualString)(const STRING *,const STRING *,BOOLEAN);*/
|
|
/*static BOOLEAN (WINAPI *pRtlPrefixString)(const STRING *, const STRING *, BOOLEAN);*/
|
|
/*static BOOLEAN (WINAPI *pRtlPrefixUnicodeString)(const UNICODE_STRING *, const UNICODE_STRING *, BOOLEAN);*/
|
|
/*static NTSTATUS (WINAPI *pRtlOemStringToUnicodeString)(PUNICODE_STRING, const STRING *, BOOLEAN);*/
|
|
/*static NTSTATUS (WINAPI *pRtlUnicodeStringToOemString)(STRING *, const UNICODE_STRING *, BOOLEAN);*/
|
|
/*static NTSTATUS (WINAPI *pRtlMultiByteToUnicodeN)(LPWSTR, DWORD, LPDWORD, LPCSTR, DWORD);*/
|
|
/*static NTSTATUS (WINAPI *pRtlOemToUnicodeN)(LPWSTR, DWORD, LPDWORD, LPCSTR, DWORD);*/
|
|
/*static NTSTATUS (WINAPI *pRtlUpcaseUnicodeStringToAnsiString)(STRING *, const UNICODE_STRING *, BOOLEAN);*/
|
|
/*static NTSTATUS (WINAPI *pRtlUpcaseUnicodeStringToOemString)(STRING *, const UNICODE_STRING *, BOOLEAN);*/
|
|
/*static NTSTATUS (WINAPI *pRtlUpcaseUnicodeToMultiByteN)(LPSTR, DWORD, LPDWORD, LPCWSTR, DWORD);*/
|
|
/*static NTSTATUS (WINAPI *pRtlUpcaseUnicodeToOemN)(LPSTR, DWORD, LPDWORD, LPCWSTR, DWORD);*/
|
|
/*static UINT (WINAPI *pRtlOemToUnicodeSize)(const STRING *);*/
|
|
/*static DWORD (WINAPI *pRtlAnsiStringToUnicodeSize)(const STRING *);*/
|
|
|
|
|
|
static WCHAR* AtoW( const char* p )
|
|
{
|
|
WCHAR* buffer;
|
|
DWORD len = MultiByteToWideChar( CP_ACP, 0, p, -1, NULL, 0 );
|
|
buffer = HeapAlloc(GetProcessHeap(), 0, len * sizeof(WCHAR) );
|
|
MultiByteToWideChar( CP_ACP, 0, p, -1, buffer, len );
|
|
return buffer;
|
|
}
|
|
|
|
|
|
static void InitFunctionPtrs(void)
|
|
{
|
|
hntdll = LoadLibraryA("ntdll.dll");
|
|
ok(hntdll != 0, "LoadLibrary failed\n");
|
|
if (hntdll) {
|
|
pRtlAnsiStringToUnicodeString = (void *)GetProcAddress(hntdll, "RtlAnsiStringToUnicodeString");
|
|
pRtlAppendAsciizToString = (void *)GetProcAddress(hntdll, "RtlAppendAsciizToString");
|
|
pRtlAppendStringToString = (void *)GetProcAddress(hntdll, "RtlAppendStringToString");
|
|
pRtlAppendUnicodeStringToString = (void *)GetProcAddress(hntdll, "RtlAppendUnicodeStringToString");
|
|
pRtlAppendUnicodeToString = (void *)GetProcAddress(hntdll, "RtlAppendUnicodeToString");
|
|
pRtlCharToInteger = (void *)GetProcAddress(hntdll, "RtlCharToInteger");
|
|
pRtlCompareUnicodeString = (void *)GetProcAddress(hntdll, "RtlCompareUnicodeString");
|
|
pRtlCompareUnicodeStrings = (void *)GetProcAddress(hntdll, "RtlCompareUnicodeStrings");
|
|
pRtlCopyString = (void *)GetProcAddress(hntdll, "RtlCopyString");
|
|
pRtlCreateUnicodeString = (void *)GetProcAddress(hntdll, "RtlCreateUnicodeString");
|
|
pRtlCreateUnicodeStringFromAsciiz = (void *)GetProcAddress(hntdll, "RtlCreateUnicodeStringFromAsciiz");
|
|
pRtlDowncaseUnicodeString = (void *)GetProcAddress(hntdll, "RtlDowncaseUnicodeString");
|
|
pRtlDuplicateUnicodeString = (void *)GetProcAddress(hntdll, "RtlDuplicateUnicodeString");
|
|
pRtlEqualUnicodeString = (void *)GetProcAddress(hntdll, "RtlEqualUnicodeString");
|
|
pRtlFindCharInUnicodeString = (void *)GetProcAddress(hntdll, "RtlFindCharInUnicodeString");
|
|
pRtlFreeAnsiString = (void *)GetProcAddress(hntdll, "RtlFreeAnsiString");
|
|
pRtlFreeUnicodeString = (void *)GetProcAddress(hntdll, "RtlFreeUnicodeString");
|
|
pRtlInitAnsiString = (void *)GetProcAddress(hntdll, "RtlInitAnsiString");
|
|
pRtlInitString = (void *)GetProcAddress(hntdll, "RtlInitString");
|
|
pRtlInitUnicodeString = (void *)GetProcAddress(hntdll, "RtlInitUnicodeString");
|
|
pRtlInitUnicodeStringEx = (void *)GetProcAddress(hntdll, "RtlInitUnicodeStringEx");
|
|
pRtlIntegerToChar = (void *)GetProcAddress(hntdll, "RtlIntegerToChar");
|
|
pRtlIntegerToUnicodeString = (void *)GetProcAddress(hntdll, "RtlIntegerToUnicodeString");
|
|
pRtlMultiAppendUnicodeStringBuffer = (void *)GetProcAddress(hntdll, "RtlMultiAppendUnicodeStringBuffer");
|
|
pRtlUnicodeStringToAnsiString = (void *)GetProcAddress(hntdll, "RtlUnicodeStringToAnsiString");
|
|
pRtlUnicodeStringToInteger = (void *)GetProcAddress(hntdll, "RtlUnicodeStringToInteger");
|
|
pRtlUpcaseUnicodeChar = (void *)GetProcAddress(hntdll, "RtlUpcaseUnicodeChar");
|
|
pRtlUpcaseUnicodeString = (void *)GetProcAddress(hntdll, "RtlUpcaseUnicodeString");
|
|
pRtlUpperChar = (void *)GetProcAddress(hntdll, "RtlUpperChar");
|
|
pRtlUpperString = (void *)GetProcAddress(hntdll, "RtlUpperString");
|
|
pRtlValidateUnicodeString = (void *)GetProcAddress(hntdll, "RtlValidateUnicodeString");
|
|
pRtlGUIDFromString = (void *)GetProcAddress(hntdll, "RtlGUIDFromString");
|
|
pRtlStringFromGUID = (void *)GetProcAddress(hntdll, "RtlStringFromGUID");
|
|
pRtlIsTextUnicode = (void *)GetProcAddress(hntdll, "RtlIsTextUnicode");
|
|
pRtlHashUnicodeString = (void*)GetProcAddress(hntdll, "RtlHashUnicodeString");
|
|
pRtlUnicodeToUTF8N = (void*)GetProcAddress(hntdll, "RtlUnicodeToUTF8N");
|
|
pRtlUTF8ToUnicodeN = (void*)GetProcAddress(hntdll, "RtlUTF8ToUnicodeN");
|
|
}
|
|
}
|
|
|
|
static void test_RtlInitString(void)
|
|
{
|
|
static const char teststring[] = "Some Wild String";
|
|
STRING str;
|
|
|
|
str.Length = 0;
|
|
str.MaximumLength = 0;
|
|
str.Buffer = (void *)0xdeadbeef;
|
|
pRtlInitString(&str, teststring);
|
|
ok(str.Length == sizeof(teststring) - sizeof(char), "Length uninitialized\n");
|
|
ok(str.MaximumLength == sizeof(teststring), "MaximumLength uninitialized\n");
|
|
ok(str.Buffer == teststring, "Buffer not equal to teststring\n");
|
|
ok(strcmp(str.Buffer, "Some Wild String") == 0, "Buffer written to\n");
|
|
pRtlInitString(&str, NULL);
|
|
ok(str.Length == 0, "Length uninitialized\n");
|
|
ok(str.MaximumLength == 0, "MaximumLength uninitialized\n");
|
|
ok(str.Buffer == NULL, "Buffer not equal to NULL\n");
|
|
/* pRtlInitString(NULL, teststring); */
|
|
}
|
|
|
|
|
|
static void test_RtlInitUnicodeString(void)
|
|
{
|
|
#define STRINGW {'S','o','m','e',' ','W','i','l','d',' ','S','t','r','i','n','g',0}
|
|
static const WCHAR teststring[] = STRINGW;
|
|
static const WCHAR originalstring[] = STRINGW;
|
|
#undef STRINGW
|
|
UNICODE_STRING uni;
|
|
|
|
uni.Length = 0;
|
|
uni.MaximumLength = 0;
|
|
uni.Buffer = (void *)0xdeadbeef;
|
|
pRtlInitUnicodeString(&uni, teststring);
|
|
ok(uni.Length == sizeof(teststring) - sizeof(WCHAR), "Length uninitialized\n");
|
|
ok(uni.MaximumLength == sizeof(teststring), "MaximumLength uninitialized\n");
|
|
ok(uni.Buffer == teststring, "Buffer not equal to teststring\n");
|
|
ok(lstrcmpW(uni.Buffer, originalstring) == 0, "Buffer written to\n");
|
|
pRtlInitUnicodeString(&uni, NULL);
|
|
ok(uni.Length == 0, "Length uninitialized\n");
|
|
ok(uni.MaximumLength == 0, "MaximumLength uninitialized\n");
|
|
ok(uni.Buffer == NULL, "Buffer not equal to NULL\n");
|
|
/* pRtlInitUnicodeString(NULL, teststring); */
|
|
}
|
|
|
|
|
|
#define TESTSTRING2_LEN 1000000
|
|
/* #define TESTSTRING2_LEN 32766 */
|
|
|
|
|
|
static void test_RtlInitUnicodeStringEx(void)
|
|
{
|
|
static const WCHAR teststring[] = {'S','o','m','e',' ','W','i','l','d',' ','S','t','r','i','n','g',0};
|
|
WCHAR *teststring2;
|
|
UNICODE_STRING uni;
|
|
NTSTATUS result;
|
|
|
|
if (!pRtlInitUnicodeStringEx)
|
|
{
|
|
win_skip("RtlInitUnicodeStringEx is not available\n");
|
|
return;
|
|
}
|
|
|
|
teststring2 = HeapAlloc(GetProcessHeap(), 0, (TESTSTRING2_LEN + 1) * sizeof(WCHAR));
|
|
memset(teststring2, 'X', TESTSTRING2_LEN * sizeof(WCHAR));
|
|
teststring2[TESTSTRING2_LEN] = '\0';
|
|
|
|
uni.Length = 12345;
|
|
uni.MaximumLength = 12345;
|
|
uni.Buffer = (void *) 0xdeadbeef;
|
|
result = pRtlInitUnicodeStringEx(&uni, teststring);
|
|
ok(result == STATUS_SUCCESS,
|
|
"pRtlInitUnicodeStringEx(&uni, 0) returns %x, expected 0\n",
|
|
result);
|
|
ok(uni.Length == 32,
|
|
"pRtlInitUnicodeStringEx(&uni, 0) sets Length to %u, expected %u\n",
|
|
uni.Length, 32);
|
|
ok(uni.MaximumLength == 34,
|
|
"pRtlInitUnicodeStringEx(&uni, 0) sets MaximumLength to %u, expected %u\n",
|
|
uni.MaximumLength, 34);
|
|
ok(uni.Buffer == teststring,
|
|
"pRtlInitUnicodeStringEx(&uni, 0) sets Buffer to %p, expected %p\n",
|
|
uni.Buffer, teststring);
|
|
|
|
uni.Length = 12345;
|
|
uni.MaximumLength = 12345;
|
|
uni.Buffer = (void *) 0xdeadbeef;
|
|
pRtlInitUnicodeString(&uni, teststring);
|
|
ok(uni.Length == 32,
|
|
"pRtlInitUnicodeString(&uni, 0) sets Length to %u, expected %u\n",
|
|
uni.Length, 32);
|
|
ok(uni.MaximumLength == 34,
|
|
"pRtlInitUnicodeString(&uni, 0) sets MaximumLength to %u, expected %u\n",
|
|
uni.MaximumLength, 34);
|
|
ok(uni.Buffer == teststring,
|
|
"pRtlInitUnicodeString(&uni, 0) sets Buffer to %p, expected %p\n",
|
|
uni.Buffer, teststring);
|
|
|
|
uni.Length = 12345;
|
|
uni.MaximumLength = 12345;
|
|
uni.Buffer = (void *) 0xdeadbeef;
|
|
result = pRtlInitUnicodeStringEx(&uni, teststring2);
|
|
ok(result == STATUS_NAME_TOO_LONG,
|
|
"pRtlInitUnicodeStringEx(&uni, 0) returns %x, expected %x\n",
|
|
result, STATUS_NAME_TOO_LONG);
|
|
ok(uni.Length == 12345 ||
|
|
uni.Length == 0, /* win2k3 */
|
|
"pRtlInitUnicodeStringEx(&uni, 0) sets Length to %u, expected 12345 or 0\n",
|
|
uni.Length);
|
|
ok(uni.MaximumLength == 12345 ||
|
|
uni.MaximumLength == 0, /* win2k3 */
|
|
"pRtlInitUnicodeStringEx(&uni, 0) sets MaximumLength to %u, expected 12345 or 0\n",
|
|
uni.MaximumLength);
|
|
ok(uni.Buffer == (void *) 0xdeadbeef ||
|
|
uni.Buffer == teststring2, /* win2k3 */
|
|
"pRtlInitUnicodeStringEx(&uni, 0) sets Buffer to %p, expected %x or %p\n",
|
|
uni.Buffer, 0xdeadbeef, teststring2);
|
|
|
|
uni.Length = 12345;
|
|
uni.MaximumLength = 12345;
|
|
uni.Buffer = (void *) 0xdeadbeef;
|
|
pRtlInitUnicodeString(&uni, teststring2);
|
|
ok(uni.Length == 33920 /* <= Win2000 */ || uni.Length == 65532 /* >= Win XP */,
|
|
"pRtlInitUnicodeString(&uni, 0) sets Length to %u, expected %u\n",
|
|
uni.Length, 65532);
|
|
ok(uni.MaximumLength == 33922 /* <= Win2000 */ || uni.MaximumLength == 65534 /* >= Win XP */,
|
|
"pRtlInitUnicodeString(&uni, 0) sets MaximumLength to %u, expected %u\n",
|
|
uni.MaximumLength, 65534);
|
|
ok(uni.Buffer == teststring2,
|
|
"pRtlInitUnicodeString(&uni, 0) sets Buffer to %p, expected %p\n",
|
|
uni.Buffer, teststring2);
|
|
ok(memcmp(uni.Buffer, teststring2, (TESTSTRING2_LEN + 1) * sizeof(WCHAR)) == 0,
|
|
"pRtlInitUnicodeString(&uni, 0) changes Buffer\n");
|
|
|
|
uni.Length = 12345;
|
|
uni.MaximumLength = 12345;
|
|
uni.Buffer = (void *) 0xdeadbeef;
|
|
result = pRtlInitUnicodeStringEx(&uni, 0);
|
|
ok(result == STATUS_SUCCESS,
|
|
"pRtlInitUnicodeStringEx(&uni, 0) returns %x, expected 0\n",
|
|
result);
|
|
ok(uni.Length == 0,
|
|
"pRtlInitUnicodeStringEx(&uni, 0) sets Length to %u, expected %u\n",
|
|
uni.Length, 0);
|
|
ok(uni.MaximumLength == 0,
|
|
"pRtlInitUnicodeStringEx(&uni, 0) sets MaximumLength to %u, expected %u\n",
|
|
uni.MaximumLength, 0);
|
|
ok(uni.Buffer == NULL,
|
|
"pRtlInitUnicodeStringEx(&uni, 0) sets Buffer to %p, expected %p\n",
|
|
uni.Buffer, NULL);
|
|
|
|
uni.Length = 12345;
|
|
uni.MaximumLength = 12345;
|
|
uni.Buffer = (void *) 0xdeadbeef;
|
|
pRtlInitUnicodeString(&uni, 0);
|
|
ok(uni.Length == 0,
|
|
"pRtlInitUnicodeString(&uni, 0) sets Length to %u, expected %u\n",
|
|
uni.Length, 0);
|
|
ok(uni.MaximumLength == 0,
|
|
"pRtlInitUnicodeString(&uni, 0) sets MaximumLength to %u, expected %u\n",
|
|
uni.MaximumLength, 0);
|
|
ok(uni.Buffer == NULL,
|
|
"pRtlInitUnicodeString(&uni, 0) sets Buffer to %p, expected %p\n",
|
|
uni.Buffer, NULL);
|
|
|
|
HeapFree(GetProcessHeap(), 0, teststring2);
|
|
}
|
|
|
|
|
|
typedef struct {
|
|
int add_nul;
|
|
int source_Length;
|
|
int source_MaximumLength;
|
|
int source_buf_size;
|
|
const char *source_buf;
|
|
int dest_Length;
|
|
int dest_MaximumLength;
|
|
int dest_buf_size;
|
|
const char *dest_buf;
|
|
int res_Length;
|
|
int res_MaximumLength;
|
|
int res_buf_size;
|
|
const char *res_buf;
|
|
NTSTATUS result;
|
|
} dupl_ustr_t;
|
|
|
|
static const dupl_ustr_t dupl_ustr[] = {
|
|
{ 0, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 32, 32, 32, "This is a string", STATUS_SUCCESS},
|
|
{ 0, 32, 32, 32, "This is a string", 40, 42, 42, "--------------------", 32, 32, 32, "This is a string", STATUS_SUCCESS},
|
|
{ 0, 32, 30, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 0, 32, 34, 34, "This is a string", 40, 42, 42, NULL, 32, 32, 32, "This is a string", STATUS_SUCCESS},
|
|
{ 0, 32, 32, 32, "This is a string", 40, 42, 42, NULL, 32, 32, 32, "This is a string", STATUS_SUCCESS},
|
|
{ 0, 32, 30, 34, "This is a string", 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 1, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 32, 34, 34, "This is a string", STATUS_SUCCESS},
|
|
{ 1, 32, 32, 32, "This is a string", 40, 42, 42, "--------------------", 32, 34, 34, "This is a string", STATUS_SUCCESS},
|
|
{ 1, 32, 30, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 1, 32, 34, 34, "This is a string", 40, 42, 42, NULL, 32, 34, 34, "This is a string", STATUS_SUCCESS},
|
|
{ 1, 32, 32, 32, "This is a string", 40, 42, 42, NULL, 32, 34, 34, "This is a string", STATUS_SUCCESS},
|
|
{ 1, 32, 30, 34, "This is a string", 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 2, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 2, 32, 32, 32, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 2, 32, 30, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 2, 32, 34, 34, "This is a string", 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 2, 32, 32, 32, "This is a string", 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 2, 32, 30, 34, "This is a string", 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 3, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 32, 34, 34, "This is a string", STATUS_SUCCESS},
|
|
{ 3, 32, 32, 32, "This is a string", 40, 42, 42, "--------------------", 32, 34, 34, "This is a string", STATUS_SUCCESS},
|
|
{ 3, 32, 30, 32, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 3, 32, 34, 34, "This is a string", 40, 42, 42, NULL, 32, 34, 34, "This is a string", STATUS_SUCCESS},
|
|
{ 3, 32, 32, 32, "This is a string", 40, 42, 42, NULL, 32, 34, 34, "This is a string", STATUS_SUCCESS},
|
|
{ 3, 32, 30, 32, "This is a string", 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 4, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 5, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 6, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 7, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 8, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 9, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{10, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{11, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{12, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{13, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{14, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{15, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{16, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{-1, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{-5, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{-9, 32, 34, 34, "This is a string", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 0, 0, 2, 2, "", 40, 42, 42, "--------------------", 0, 0, 0, NULL, STATUS_SUCCESS},
|
|
{ 0, 0, 0, 0, "", 40, 42, 42, "--------------------", 0, 0, 0, NULL, STATUS_SUCCESS},
|
|
{ 0, 0, 2, 2, "", 40, 42, 42, NULL, 0, 0, 0, NULL, STATUS_SUCCESS},
|
|
{ 0, 0, 0, 0, "", 40, 42, 42, NULL, 0, 0, 0, NULL, STATUS_SUCCESS},
|
|
{ 0, 0, 2, 2, NULL, 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 0, 0, 0, 0, NULL, 40, 42, 42, "--------------------", 0, 0, 0, NULL, STATUS_SUCCESS},
|
|
{ 0, 0, 2, 2, NULL, 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 0, 0, 0, 0, NULL, 40, 42, 42, NULL, 0, 0, 0, NULL, STATUS_SUCCESS},
|
|
{ 1, 0, 2, 2, "", 40, 42, 42, "--------------------", 0, 0, 0, NULL, STATUS_SUCCESS},
|
|
{ 1, 0, 0, 0, "", 40, 42, 42, "--------------------", 0, 0, 0, NULL, STATUS_SUCCESS},
|
|
{ 1, 0, 2, 2, "", 40, 42, 42, NULL, 0, 0, 0, NULL, STATUS_SUCCESS},
|
|
{ 1, 0, 0, 0, "", 40, 42, 42, NULL, 0, 0, 0, NULL, STATUS_SUCCESS},
|
|
{ 1, 0, 2, 2, NULL, 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 1, 0, 0, 0, NULL, 40, 42, 42, "--------------------", 0, 0, 0, NULL, STATUS_SUCCESS},
|
|
{ 1, 0, 2, 2, NULL, 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 1, 0, 0, 0, NULL, 40, 42, 42, NULL, 0, 0, 0, NULL, STATUS_SUCCESS},
|
|
{ 2, 0, 2, 2, "", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 2, 0, 0, 0, "", 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 2, 0, 2, 2, "", 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 2, 0, 0, 0, "", 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 2, 0, 2, 2, NULL, 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 2, 0, 0, 0, NULL, 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 2, 0, 2, 2, NULL, 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 2, 0, 0, 0, NULL, 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 3, 0, 2, 2, "", 40, 42, 42, "--------------------", 0, 2, 2, "", STATUS_SUCCESS},
|
|
{ 3, 0, 0, 0, "", 40, 42, 42, "--------------------", 0, 2, 2, "", STATUS_SUCCESS},
|
|
{ 3, 0, 2, 2, "", 40, 42, 42, NULL, 0, 2, 2, "", STATUS_SUCCESS},
|
|
{ 3, 0, 0, 0, "", 40, 42, 42, NULL, 0, 2, 2, "", STATUS_SUCCESS},
|
|
{ 3, 0, 2, 2, NULL, 40, 42, 42, "--------------------", 40, 42, 42, "--------------------", STATUS_INVALID_PARAMETER},
|
|
{ 3, 0, 0, 0, NULL, 40, 42, 42, "--------------------", 0, 2, 2, "", STATUS_SUCCESS},
|
|
{ 3, 0, 2, 2, NULL, 40, 42, 42, NULL, 40, 42, 0, NULL, STATUS_INVALID_PARAMETER},
|
|
{ 3, 0, 0, 0, NULL, 40, 42, 42, NULL, 0, 2, 2, "", STATUS_SUCCESS},
|
|
};
|
|
#define NB_DUPL_USTR (sizeof(dupl_ustr)/sizeof(*dupl_ustr))
|
|
|
|
|
|
static void test_RtlDuplicateUnicodeString(void)
|
|
{
|
|
size_t pos;
|
|
WCHAR source_buf[257];
|
|
WCHAR dest_buf[257];
|
|
WCHAR res_buf[257];
|
|
UNICODE_STRING source_str;
|
|
UNICODE_STRING dest_str;
|
|
UNICODE_STRING res_str;
|
|
CHAR dest_ansi_buf[257];
|
|
STRING dest_ansi_str;
|
|
NTSTATUS result;
|
|
unsigned int test_num;
|
|
|
|
if (!pRtlDuplicateUnicodeString)
|
|
{
|
|
win_skip("RtlDuplicateUnicodeString is not available\n");
|
|
return;
|
|
}
|
|
|
|
for (test_num = 0; test_num < NB_DUPL_USTR; test_num++) {
|
|
source_str.Length = dupl_ustr[test_num].source_Length;
|
|
source_str.MaximumLength = dupl_ustr[test_num].source_MaximumLength;
|
|
if (dupl_ustr[test_num].source_buf != NULL) {
|
|
for (pos = 0; pos < dupl_ustr[test_num].source_buf_size/sizeof(WCHAR); pos++) {
|
|
source_buf[pos] = dupl_ustr[test_num].source_buf[pos];
|
|
}
|
|
source_str.Buffer = source_buf;
|
|
} else {
|
|
source_str.Buffer = NULL;
|
|
}
|
|
dest_str.Length = dupl_ustr[test_num].dest_Length;
|
|
dest_str.MaximumLength = dupl_ustr[test_num].dest_MaximumLength;
|
|
if (dupl_ustr[test_num].dest_buf != NULL) {
|
|
for (pos = 0; pos < dupl_ustr[test_num].dest_buf_size/sizeof(WCHAR); pos++) {
|
|
dest_buf[pos] = dupl_ustr[test_num].dest_buf[pos];
|
|
}
|
|
dest_str.Buffer = dest_buf;
|
|
} else {
|
|
dest_str.Buffer = NULL;
|
|
}
|
|
res_str.Length = dupl_ustr[test_num].res_Length;
|
|
res_str.MaximumLength = dupl_ustr[test_num].res_MaximumLength;
|
|
if (dupl_ustr[test_num].res_buf != NULL) {
|
|
for (pos = 0; pos < dupl_ustr[test_num].res_buf_size/sizeof(WCHAR); pos++) {
|
|
res_buf[pos] = dupl_ustr[test_num].res_buf[pos];
|
|
}
|
|
res_str.Buffer = res_buf;
|
|
} else {
|
|
res_str.Buffer = NULL;
|
|
}
|
|
result = pRtlDuplicateUnicodeString(dupl_ustr[test_num].add_nul, &source_str, &dest_str);
|
|
dest_ansi_str.Length = dest_str.Length / sizeof(WCHAR);
|
|
dest_ansi_str.MaximumLength = dest_ansi_str.Length + 1;
|
|
for (pos = 0; pos < dest_ansi_str.Length; pos++) {
|
|
dest_ansi_buf[pos] = (char)dest_buf[pos];
|
|
}
|
|
dest_ansi_buf[dest_ansi_str.Length] = '\0';
|
|
dest_ansi_str.Buffer = dest_ansi_buf;
|
|
ok(result == dupl_ustr[test_num].result,
|
|
"(test %d): RtlDuplicateUnicodeString(%d, source, dest) has result %x, expected %x\n",
|
|
test_num, dupl_ustr[test_num].add_nul, result, dupl_ustr[test_num].result);
|
|
ok(dest_str.Length == dupl_ustr[test_num].res_Length,
|
|
"(test %d): RtlDuplicateUnicodeString(%d, source, dest) destination has Length %d, expected %d\n",
|
|
test_num, dupl_ustr[test_num].add_nul, dest_str.Length, dupl_ustr[test_num].res_Length);
|
|
ok(dest_str.MaximumLength == dupl_ustr[test_num].res_MaximumLength,
|
|
"(test %d): RtlDuplicateUnicodeString(%d, source, dest) destination has MaximumLength %d, expected %d\n",
|
|
test_num, dupl_ustr[test_num].add_nul, dest_str.MaximumLength, dupl_ustr[test_num].res_MaximumLength);
|
|
if (result == STATUS_INVALID_PARAMETER) {
|
|
ok((dest_str.Buffer == NULL && res_str.Buffer == NULL) ||
|
|
dest_str.Buffer == dest_buf,
|
|
"(test %d): RtlDuplicateUnicodeString(%d, source, dest) destination buffer changed %p expected %p\n",
|
|
test_num, dupl_ustr[test_num].add_nul, dest_str.Buffer, dest_buf);
|
|
} else {
|
|
ok(dest_str.Buffer != dest_buf,
|
|
"(test %d): RtlDuplicateUnicodeString(%d, source, dest) has destination buffer unchanged %p\n",
|
|
test_num, dupl_ustr[test_num].add_nul, dest_str.Buffer);
|
|
}
|
|
if (dest_str.Buffer != NULL && dupl_ustr[test_num].res_buf != NULL) {
|
|
ok(memcmp(dest_str.Buffer, res_str.Buffer, dupl_ustr[test_num].res_buf_size) == 0,
|
|
"(test %d): RtlDuplicateUnicodeString(%d, source, dest) has destination \"%s\" expected \"%s\"\n",
|
|
test_num, dupl_ustr[test_num].add_nul, dest_ansi_str.Buffer, dupl_ustr[test_num].res_buf);
|
|
if(result == STATUS_SUCCESS) pRtlFreeUnicodeString(&dest_str);
|
|
} else {
|
|
ok(dest_str.Buffer == NULL && dupl_ustr[test_num].res_buf == NULL,
|
|
"(test %d): RtlDuplicateUnicodeString(%d, source, dest) has destination %p expected %p\n",
|
|
test_num, dupl_ustr[test_num].add_nul, dest_str.Buffer, dupl_ustr[test_num].res_buf);
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
static void test_RtlCopyString(void)
|
|
{
|
|
static const char teststring[] = "Some Wild String";
|
|
char deststring[] = " ";
|
|
STRING str;
|
|
STRING deststr;
|
|
|
|
pRtlInitString(&str, teststring);
|
|
pRtlInitString(&deststr, deststring);
|
|
pRtlCopyString(&deststr, &str);
|
|
ok(strncmp(str.Buffer, deststring, str.Length) == 0, "String not copied\n");
|
|
}
|
|
|
|
|
|
static void test_RtlUpperChar(void)
|
|
{
|
|
int ch;
|
|
int upper_ch;
|
|
int expected_upper_ch;
|
|
int byte_ch;
|
|
|
|
for (ch = -1; ch <= 1024; ch++) {
|
|
upper_ch = pRtlUpperChar(ch);
|
|
byte_ch = ch & 0xff;
|
|
if (byte_ch >= 'a' && byte_ch <= 'z') {
|
|
expected_upper_ch = (CHAR) (byte_ch - 'a' + 'A');
|
|
} else {
|
|
expected_upper_ch = (CHAR) byte_ch;
|
|
}
|
|
ok(upper_ch == expected_upper_ch,
|
|
"RtlUpperChar('%c'[=0x%x]) has result '%c'[=0x%x], expected '%c'[=0x%x]\n",
|
|
ch, ch, upper_ch, upper_ch, expected_upper_ch, expected_upper_ch);
|
|
}
|
|
}
|
|
|
|
|
|
static void test_RtlUpperString(void)
|
|
{
|
|
int i;
|
|
CHAR ch;
|
|
CHAR upper_ch;
|
|
char ascii_buf[257];
|
|
char result_buf[257];
|
|
char upper_buf[257];
|
|
STRING ascii_str;
|
|
STRING result_str;
|
|
STRING upper_str;
|
|
|
|
for (i = 0; i <= 255; i++) {
|
|
ch = (CHAR) i;
|
|
if (ch >= 'a' && ch <= 'z') {
|
|
upper_ch = ch - 'a' + 'A';
|
|
} else {
|
|
upper_ch = ch;
|
|
}
|
|
ascii_buf[i] = ch;
|
|
result_buf[i] = '\0';
|
|
upper_buf[i] = upper_ch;
|
|
}
|
|
ascii_buf[i] = '\0';
|
|
result_buf[i] = '\0';
|
|
upper_buf[i] = '\0';
|
|
ascii_str.Length = 256;
|
|
ascii_str.MaximumLength = 256;
|
|
ascii_str.Buffer = ascii_buf;
|
|
result_str.Length = 256;
|
|
result_str.MaximumLength = 256;
|
|
result_str.Buffer = result_buf;
|
|
upper_str.Length = 256;
|
|
upper_str.MaximumLength = 256;
|
|
upper_str.Buffer = upper_buf;
|
|
|
|
pRtlUpperString(&result_str, &ascii_str);
|
|
ok(memcmp(result_str.Buffer, upper_str.Buffer, 256) == 0,
|
|
"RtlUpperString does not work as expected\n");
|
|
}
|
|
|
|
|
|
static void test_RtlUpcaseUnicodeChar(void)
|
|
{
|
|
int i;
|
|
WCHAR ch;
|
|
WCHAR upper_ch;
|
|
WCHAR expected_upper_ch;
|
|
|
|
for (i = 0; i <= 255; i++) {
|
|
ch = (WCHAR) i;
|
|
upper_ch = pRtlUpcaseUnicodeChar(ch);
|
|
if (ch >= 'a' && ch <= 'z') {
|
|
expected_upper_ch = ch - 'a' + 'A';
|
|
} else if (ch >= 0xe0 && ch <= 0xfe && ch != 0xf7) {
|
|
expected_upper_ch = ch - 0x20;
|
|
} else if (ch == 0xff) {
|
|
expected_upper_ch = 0x178;
|
|
} else {
|
|
expected_upper_ch = ch;
|
|
}
|
|
ok(upper_ch == expected_upper_ch,
|
|
"RtlUpcaseUnicodeChar('%c'[=0x%x]) has result '%c'[=0x%x], expected: '%c'[=0x%x]\n",
|
|
ch, ch, upper_ch, upper_ch, expected_upper_ch, expected_upper_ch);
|
|
}
|
|
}
|
|
|
|
|
|
static void test_RtlUpcaseUnicodeString(void)
|
|
{
|
|
int i;
|
|
WCHAR ch;
|
|
WCHAR upper_ch;
|
|
WCHAR ascii_buf[257];
|
|
WCHAR result_buf[257];
|
|
WCHAR upper_buf[257];
|
|
UNICODE_STRING ascii_str;
|
|
UNICODE_STRING result_str;
|
|
UNICODE_STRING upper_str;
|
|
|
|
for (i = 0; i <= 255; i++) {
|
|
ch = (WCHAR) i;
|
|
if (ch >= 'a' && ch <= 'z') {
|
|
upper_ch = ch - 'a' + 'A';
|
|
} else if (ch >= 0xe0 && ch <= 0xfe && ch != 0xf7) {
|
|
upper_ch = ch - 0x20;
|
|
} else if (ch == 0xff) {
|
|
upper_ch = 0x178;
|
|
} else {
|
|
upper_ch = ch;
|
|
}
|
|
ascii_buf[i] = ch;
|
|
result_buf[i] = '\0';
|
|
upper_buf[i] = upper_ch;
|
|
}
|
|
ascii_buf[i] = '\0';
|
|
result_buf[i] = '\0';
|
|
upper_buf[i] = '\0';
|
|
ascii_str.Length = 512;
|
|
ascii_str.MaximumLength = 512;
|
|
ascii_str.Buffer = ascii_buf;
|
|
result_str.Length = 512;
|
|
result_str.MaximumLength = 512;
|
|
result_str.Buffer = result_buf;
|
|
upper_str.Length = 512;
|
|
upper_str.MaximumLength = 512;
|
|
upper_str.Buffer = upper_buf;
|
|
|
|
pRtlUpcaseUnicodeString(&result_str, &ascii_str, 0);
|
|
for (i = 0; i <= 255; i++) {
|
|
ok(result_str.Buffer[i] == upper_str.Buffer[i],
|
|
"RtlUpcaseUnicodeString works wrong: '%c'[=0x%x] is converted to '%c'[=0x%x], expected: '%c'[=0x%x]\n",
|
|
ascii_str.Buffer[i], ascii_str.Buffer[i],
|
|
result_str.Buffer[i], result_str.Buffer[i],
|
|
upper_str.Buffer[i], upper_str.Buffer[i]);
|
|
}
|
|
}
|
|
|
|
|
|
static void test_RtlDowncaseUnicodeString(void)
|
|
{
|
|
int i;
|
|
WCHAR ch;
|
|
WCHAR lower_ch;
|
|
WCHAR source_buf[1025];
|
|
WCHAR result_buf[1025];
|
|
WCHAR lower_buf[1025];
|
|
UNICODE_STRING source_str;
|
|
UNICODE_STRING result_str;
|
|
UNICODE_STRING lower_str;
|
|
|
|
for (i = 0; i < 1024; i++) {
|
|
ch = (WCHAR) i;
|
|
if (ch >= 'A' && ch <= 'Z') {
|
|
lower_ch = ch - 'A' + 'a';
|
|
} else if (ch >= 0xc0 && ch <= 0xde && ch != 0xd7) {
|
|
lower_ch = ch + 0x20;
|
|
} else if (ch >= 0x391 && ch <= 0x3ab && ch != 0x3a2) {
|
|
lower_ch = ch + 0x20;
|
|
} else {
|
|
switch (ch) {
|
|
case 0x178: lower_ch = 0xff; break;
|
|
case 0x181: lower_ch = 0x253; break;
|
|
case 0x186: lower_ch = 0x254; break;
|
|
case 0x189: lower_ch = 0x256; break;
|
|
case 0x18a: lower_ch = 0x257; break;
|
|
case 0x18e: lower_ch = 0x1dd; break;
|
|
case 0x18f: lower_ch = 0x259; break;
|
|
case 0x190: lower_ch = 0x25b; break;
|
|
case 0x193: lower_ch = 0x260; break;
|
|
case 0x194: lower_ch = 0x263; break;
|
|
case 0x196: lower_ch = 0x269; break;
|
|
case 0x197: lower_ch = 0x268; break;
|
|
case 0x19c: lower_ch = 0x26f; break;
|
|
case 0x19d: lower_ch = 0x272; break;
|
|
case 0x19f: lower_ch = 0x275; break;
|
|
case 0x1a9: lower_ch = 0x283; break;
|
|
case 0x1ae: lower_ch = 0x288; break;
|
|
case 0x1b1: lower_ch = 0x28a; break;
|
|
case 0x1b2: lower_ch = 0x28b; break;
|
|
case 0x1b7: lower_ch = 0x292; break;
|
|
case 0x1c4: lower_ch = 0x1c6; break;
|
|
case 0x1c7: lower_ch = 0x1c9; break;
|
|
case 0x1ca: lower_ch = 0x1cc; break;
|
|
case 0x1f1: lower_ch = 0x1f3; break;
|
|
case 0x386: lower_ch = 0x3ac; break;
|
|
case 0x388: lower_ch = 0x3ad; break;
|
|
case 0x389: lower_ch = 0x3ae; break;
|
|
case 0x38a: lower_ch = 0x3af; break;
|
|
case 0x38c: lower_ch = 0x3cc; break;
|
|
case 0x38e: lower_ch = 0x3cd; break;
|
|
case 0x38f: lower_ch = 0x3ce; break;
|
|
default: lower_ch = ch; break;
|
|
} /* switch */
|
|
}
|
|
source_buf[i] = ch;
|
|
result_buf[i] = '\0';
|
|
lower_buf[i] = lower_ch;
|
|
}
|
|
source_buf[i] = '\0';
|
|
result_buf[i] = '\0';
|
|
lower_buf[i] = '\0';
|
|
source_str.Length = 2048;
|
|
source_str.MaximumLength = 2048;
|
|
source_str.Buffer = source_buf;
|
|
result_str.Length = 2048;
|
|
result_str.MaximumLength = 2048;
|
|
result_str.Buffer = result_buf;
|
|
lower_str.Length = 2048;
|
|
lower_str.MaximumLength = 2048;
|
|
lower_str.Buffer = lower_buf;
|
|
|
|
pRtlDowncaseUnicodeString(&result_str, &source_str, 0);
|
|
for (i = 0; i <= 1024; i++) {
|
|
ok(result_str.Buffer[i] == lower_str.Buffer[i] || result_str.Buffer[i] == source_str.Buffer[i] + 1,
|
|
"RtlDowncaseUnicodeString works wrong: '%c'[=0x%x] is converted to '%c'[=0x%x], expected: '%c'[=0x%x]\n",
|
|
source_str.Buffer[i], source_str.Buffer[i],
|
|
result_str.Buffer[i], result_str.Buffer[i],
|
|
lower_str.Buffer[i], lower_str.Buffer[i]);
|
|
}
|
|
}
|
|
|
|
|
|
typedef struct {
|
|
int ansi_Length;
|
|
int ansi_MaximumLength;
|
|
int ansi_buf_size;
|
|
const char *ansi_buf;
|
|
int uni_Length;
|
|
int uni_MaximumLength;
|
|
int uni_buf_size;
|
|
const char *uni_buf;
|
|
BOOLEAN doalloc;
|
|
int res_Length;
|
|
int res_MaximumLength;
|
|
int res_buf_size;
|
|
const char *res_buf;
|
|
NTSTATUS result;
|
|
} ustr2astr_t;
|
|
|
|
static const ustr2astr_t ustr2astr[] = {
|
|
{ 10, 12, 12, "------------", 0, 0, 0, "", TRUE, 0, 1, 1, "", STATUS_SUCCESS},
|
|
{ 10, 12, 12, "------------", 12, 12, 12, "abcdef", TRUE, 6, 7, 7, "abcdef", STATUS_SUCCESS},
|
|
{ 0, 2, 12, "------------", 12, 12, 12, "abcdef", TRUE, 6, 7, 7, "abcdef", STATUS_SUCCESS},
|
|
{ 10, 12, 12, NULL, 12, 12, 12, "abcdef", TRUE, 6, 7, 7, "abcdef", STATUS_SUCCESS},
|
|
{ 0, 0, 12, "------------", 12, 12, 12, "abcdef", FALSE, 6, 0, 0, "", STATUS_BUFFER_OVERFLOW},
|
|
{ 0, 1, 12, "------------", 12, 12, 12, "abcdef", FALSE, 0, 1, 1, "", STATUS_BUFFER_OVERFLOW},
|
|
{ 0, 2, 12, "------------", 12, 12, 12, "abcdef", FALSE, 1, 2, 2, "a", STATUS_BUFFER_OVERFLOW},
|
|
{ 0, 3, 12, "------------", 12, 12, 12, "abcdef", FALSE, 2, 3, 3, "ab", STATUS_BUFFER_OVERFLOW},
|
|
{ 0, 5, 12, "------------", 12, 12, 12, "abcdef", FALSE, 4, 5, 5, "abcd", STATUS_BUFFER_OVERFLOW},
|
|
{ 8, 5, 12, "------------", 12, 12, 12, "abcdef", FALSE, 4, 5, 5, "abcd", STATUS_BUFFER_OVERFLOW},
|
|
{ 8, 6, 12, "------------", 12, 12, 12, "abcdef", FALSE, 5, 6, 6, "abcde", STATUS_BUFFER_OVERFLOW},
|
|
{ 8, 7, 12, "------------", 12, 12, 12, "abcdef", FALSE, 6, 7, 7, "abcdef", STATUS_SUCCESS},
|
|
{ 8, 7, 12, "------------", 0, 12, 12, NULL, FALSE, 0, 7, 0, "", STATUS_SUCCESS},
|
|
#if 0
|
|
/* crashes on Japanese and Chinese XP */
|
|
{ 0, 0, 12, NULL, 10, 10, 12, NULL, FALSE, 5, 0, 0, NULL, STATUS_BUFFER_OVERFLOW},
|
|
#endif
|
|
};
|
|
#define NB_USTR2ASTR (sizeof(ustr2astr)/sizeof(*ustr2astr))
|
|
|
|
|
|
static void test_RtlUnicodeStringToAnsiString(void)
|
|
{
|
|
size_t pos;
|
|
CHAR ansi_buf[257];
|
|
WCHAR uni_buf[257];
|
|
STRING ansi_str;
|
|
UNICODE_STRING uni_str;
|
|
NTSTATUS result;
|
|
unsigned int test_num;
|
|
|
|
for (test_num = 0; test_num < NB_USTR2ASTR; test_num++) {
|
|
ansi_str.Length = ustr2astr[test_num].ansi_Length;
|
|
ansi_str.MaximumLength = ustr2astr[test_num].ansi_MaximumLength;
|
|
if (ustr2astr[test_num].ansi_buf != NULL) {
|
|
memcpy(ansi_buf, ustr2astr[test_num].ansi_buf, ustr2astr[test_num].ansi_buf_size);
|
|
ansi_buf[ustr2astr[test_num].ansi_buf_size] = '\0';
|
|
ansi_str.Buffer = ansi_buf;
|
|
} else {
|
|
ansi_str.Buffer = NULL;
|
|
}
|
|
uni_str.Length = ustr2astr[test_num].uni_Length;
|
|
uni_str.MaximumLength = ustr2astr[test_num].uni_MaximumLength;
|
|
if (ustr2astr[test_num].uni_buf != NULL) {
|
|
for (pos = 0; pos < ustr2astr[test_num].uni_buf_size/sizeof(WCHAR); pos++) {
|
|
uni_buf[pos] = ustr2astr[test_num].uni_buf[pos];
|
|
}
|
|
uni_str.Buffer = uni_buf;
|
|
} else {
|
|
uni_str.Buffer = NULL;
|
|
}
|
|
result = pRtlUnicodeStringToAnsiString(&ansi_str, &uni_str, ustr2astr[test_num].doalloc);
|
|
ok(result == ustr2astr[test_num].result,
|
|
"(test %d): RtlUnicodeStringToAnsiString(ansi, uni, %d) has result %x, expected %x\n",
|
|
test_num, ustr2astr[test_num].doalloc, result, ustr2astr[test_num].result);
|
|
ok(ansi_str.Length == ustr2astr[test_num].res_Length,
|
|
"(test %d): RtlUnicodeStringToAnsiString(ansi, uni, %d) ansi has Length %d, expected %d\n",
|
|
test_num, ustr2astr[test_num].doalloc, ansi_str.Length, ustr2astr[test_num].res_Length);
|
|
ok(ansi_str.MaximumLength == ustr2astr[test_num].res_MaximumLength,
|
|
"(test %d): RtlUnicodeStringToAnsiString(ansi, uni, %d) ansi has MaximumLength %d, expected %d\n",
|
|
test_num, ustr2astr[test_num].doalloc, ansi_str.MaximumLength, ustr2astr[test_num].res_MaximumLength);
|
|
ok(memcmp(ansi_str.Buffer, ustr2astr[test_num].res_buf, ustr2astr[test_num].res_buf_size) == 0,
|
|
"(test %d): RtlUnicodeStringToAnsiString(ansi, uni, %d) has ansi \"%s\" expected \"%s\"\n",
|
|
test_num, ustr2astr[test_num].doalloc, ansi_str.Buffer, ustr2astr[test_num].res_buf);
|
|
if(result == STATUS_SUCCESS && ustr2astr[test_num].doalloc)
|
|
pRtlFreeAnsiString(&ansi_str);
|
|
}
|
|
}
|
|
|
|
|
|
typedef struct {
|
|
int dest_Length;
|
|
int dest_MaximumLength;
|
|
int dest_buf_size;
|
|
const char *dest_buf;
|
|
const char *src;
|
|
int res_Length;
|
|
int res_MaximumLength;
|
|
int res_buf_size;
|
|
const char *res_buf;
|
|
NTSTATUS result;
|
|
} app_asc2str_t;
|
|
|
|
static const app_asc2str_t app_asc2str[] = {
|
|
{ 5, 12, 15, "TestS01234abcde", "tring", 10, 12, 15, "TestStringabcde", STATUS_SUCCESS},
|
|
{ 5, 11, 15, "TestS01234abcde", "tring", 10, 11, 15, "TestStringabcde", STATUS_SUCCESS},
|
|
{ 5, 10, 15, "TestS01234abcde", "tring", 10, 10, 15, "TestStringabcde", STATUS_SUCCESS},
|
|
{ 5, 9, 15, "TestS01234abcde", "tring", 5, 9, 15, "TestS01234abcde", STATUS_BUFFER_TOO_SMALL},
|
|
{ 5, 0, 15, "TestS01234abcde", "tring", 5, 0, 15, "TestS01234abcde", STATUS_BUFFER_TOO_SMALL},
|
|
{ 5, 14, 15, "TestS01234abcde", "tring", 10, 14, 15, "TestStringabcde", STATUS_SUCCESS},
|
|
{ 5, 14, 15, "TestS01234abcde", NULL, 5, 14, 15, "TestS01234abcde", STATUS_SUCCESS},
|
|
{ 5, 14, 15, NULL, NULL, 5, 14, 15, NULL, STATUS_SUCCESS},
|
|
{ 5, 12, 15, "Tst\0S01234abcde", "tr\0i", 7, 12, 15, "Tst\0Str234abcde", STATUS_SUCCESS},
|
|
};
|
|
#define NB_APP_ASC2STR (sizeof(app_asc2str)/sizeof(*app_asc2str))
|
|
|
|
|
|
static void test_RtlAppendAsciizToString(void)
|
|
{
|
|
CHAR dest_buf[257];
|
|
STRING dest_str;
|
|
NTSTATUS result;
|
|
unsigned int test_num;
|
|
|
|
for (test_num = 0; test_num < NB_APP_ASC2STR; test_num++) {
|
|
dest_str.Length = app_asc2str[test_num].dest_Length;
|
|
dest_str.MaximumLength = app_asc2str[test_num].dest_MaximumLength;
|
|
if (app_asc2str[test_num].dest_buf != NULL) {
|
|
memcpy(dest_buf, app_asc2str[test_num].dest_buf, app_asc2str[test_num].dest_buf_size);
|
|
dest_buf[app_asc2str[test_num].dest_buf_size] = '\0';
|
|
dest_str.Buffer = dest_buf;
|
|
} else {
|
|
dest_str.Buffer = NULL;
|
|
}
|
|
result = pRtlAppendAsciizToString(&dest_str, app_asc2str[test_num].src);
|
|
ok(result == app_asc2str[test_num].result,
|
|
"(test %d): RtlAppendAsciizToString(dest, src) has result %x, expected %x\n",
|
|
test_num, result, app_asc2str[test_num].result);
|
|
ok(dest_str.Length == app_asc2str[test_num].res_Length,
|
|
"(test %d): RtlAppendAsciizToString(dest, src) dest has Length %d, expected %d\n",
|
|
test_num, dest_str.Length, app_asc2str[test_num].res_Length);
|
|
ok(dest_str.MaximumLength == app_asc2str[test_num].res_MaximumLength,
|
|
"(test %d): RtlAppendAsciizToString(dest, src) dest has MaximumLength %d, expected %d\n",
|
|
test_num, dest_str.MaximumLength, app_asc2str[test_num].res_MaximumLength);
|
|
if (dest_str.Buffer == dest_buf) {
|
|
ok(memcmp(dest_buf, app_asc2str[test_num].res_buf, app_asc2str[test_num].res_buf_size) == 0,
|
|
"(test %d): RtlAppendAsciizToString(dest, src) has dest \"%s\" expected \"%s\"\n",
|
|
test_num, dest_buf, app_asc2str[test_num].res_buf);
|
|
} else {
|
|
ok(dest_str.Buffer == app_asc2str[test_num].res_buf,
|
|
"(test %d): RtlAppendAsciizToString(dest, src) dest has Buffer %p expected %p\n",
|
|
test_num, dest_str.Buffer, app_asc2str[test_num].res_buf);
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
typedef struct {
|
|
int dest_Length;
|
|
int dest_MaximumLength;
|
|
int dest_buf_size;
|
|
const char *dest_buf;
|
|
int src_Length;
|
|
int src_MaximumLength;
|
|
int src_buf_size;
|
|
const char *src_buf;
|
|
int res_Length;
|
|
int res_MaximumLength;
|
|
int res_buf_size;
|
|
const char *res_buf;
|
|
NTSTATUS result;
|
|
} app_str2str_t;
|
|
|
|
static const app_str2str_t app_str2str[] = {
|
|
{ 5, 12, 15, "TestS01234abcde", 5, 5, 7, "tringZY", 10, 12, 15, "TestStringabcde", STATUS_SUCCESS},
|
|
{ 5, 11, 15, "TestS01234abcde", 5, 5, 7, "tringZY", 10, 11, 15, "TestStringabcde", STATUS_SUCCESS},
|
|
{ 5, 10, 15, "TestS01234abcde", 5, 5, 7, "tringZY", 10, 10, 15, "TestStringabcde", STATUS_SUCCESS},
|
|
{ 5, 9, 15, "TestS01234abcde", 5, 5, 7, "tringZY", 5, 9, 15, "TestS01234abcde", STATUS_BUFFER_TOO_SMALL},
|
|
{ 5, 0, 15, "TestS01234abcde", 0, 0, 7, "tringZY", 5, 0, 15, "TestS01234abcde", STATUS_SUCCESS},
|
|
{ 5, 14, 15, "TestS01234abcde", 0, 0, 7, "tringZY", 5, 14, 15, "TestS01234abcde", STATUS_SUCCESS},
|
|
{ 5, 14, 15, "TestS01234abcde", 0, 0, 7, NULL, 5, 14, 15, "TestS01234abcde", STATUS_SUCCESS},
|
|
{ 5, 14, 15, NULL, 0, 0, 7, NULL, 5, 14, 15, NULL, STATUS_SUCCESS},
|
|
{ 5, 12, 15, "Tst\0S01234abcde", 4, 4, 7, "tr\0iZY", 9, 12, 15, "Tst\0Str\0i4abcde", STATUS_SUCCESS},
|
|
};
|
|
#define NB_APP_STR2STR (sizeof(app_str2str)/sizeof(*app_str2str))
|
|
|
|
|
|
static void test_RtlAppendStringToString(void)
|
|
{
|
|
CHAR dest_buf[257];
|
|
CHAR src_buf[257];
|
|
STRING dest_str;
|
|
STRING src_str;
|
|
NTSTATUS result;
|
|
unsigned int test_num;
|
|
|
|
for (test_num = 0; test_num < NB_APP_STR2STR; test_num++) {
|
|
dest_str.Length = app_str2str[test_num].dest_Length;
|
|
dest_str.MaximumLength = app_str2str[test_num].dest_MaximumLength;
|
|
if (app_str2str[test_num].dest_buf != NULL) {
|
|
memcpy(dest_buf, app_str2str[test_num].dest_buf, app_str2str[test_num].dest_buf_size);
|
|
dest_buf[app_str2str[test_num].dest_buf_size] = '\0';
|
|
dest_str.Buffer = dest_buf;
|
|
} else {
|
|
dest_str.Buffer = NULL;
|
|
}
|
|
src_str.Length = app_str2str[test_num].src_Length;
|
|
src_str.MaximumLength = app_str2str[test_num].src_MaximumLength;
|
|
if (app_str2str[test_num].src_buf != NULL) {
|
|
memcpy(src_buf, app_str2str[test_num].src_buf, app_str2str[test_num].src_buf_size);
|
|
src_buf[app_str2str[test_num].src_buf_size] = '\0';
|
|
src_str.Buffer = src_buf;
|
|
} else {
|
|
src_str.Buffer = NULL;
|
|
}
|
|
result = pRtlAppendStringToString(&dest_str, &src_str);
|
|
ok(result == app_str2str[test_num].result,
|
|
"(test %d): RtlAppendStringToString(dest, src) has result %x, expected %x\n",
|
|
test_num, result, app_str2str[test_num].result);
|
|
ok(dest_str.Length == app_str2str[test_num].res_Length,
|
|
"(test %d): RtlAppendStringToString(dest, src) dest has Length %d, expected %d\n",
|
|
test_num, dest_str.Length, app_str2str[test_num].res_Length);
|
|
ok(dest_str.MaximumLength == app_str2str[test_num].res_MaximumLength,
|
|
"(test %d): RtlAppendStringToString(dest, src) dest has MaximumLength %d, expected %d\n",
|
|
test_num, dest_str.MaximumLength, app_str2str[test_num].res_MaximumLength);
|
|
if (dest_str.Buffer == dest_buf) {
|
|
ok(memcmp(dest_buf, app_str2str[test_num].res_buf, app_str2str[test_num].res_buf_size) == 0,
|
|
"(test %d): RtlAppendStringToString(dest, src) has dest \"%s\" expected \"%s\"\n",
|
|
test_num, dest_buf, app_str2str[test_num].res_buf);
|
|
} else {
|
|
ok(dest_str.Buffer == app_str2str[test_num].res_buf,
|
|
"(test %d): RtlAppendStringToString(dest, src) dest has Buffer %p expected %p\n",
|
|
test_num, dest_str.Buffer, app_str2str[test_num].res_buf);
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
typedef struct {
|
|
int dest_Length;
|
|
int dest_MaximumLength;
|
|
int dest_buf_size;
|
|
const char *dest_buf;
|
|
const char *src;
|
|
int res_Length;
|
|
int res_MaximumLength;
|
|
int res_buf_size;
|
|
const char *res_buf;
|
|
NTSTATUS result;
|
|
} app_uni2str_t;
|
|
|
|
static const app_uni2str_t app_uni2str[] = {
|
|
{ 4, 12, 14, "Fake0123abcdef", "Ustr\0", 8, 12, 14, "FakeUstr\0\0cdef", STATUS_SUCCESS},
|
|
{ 4, 11, 14, "Fake0123abcdef", "Ustr\0", 8, 11, 14, "FakeUstr\0\0cdef", STATUS_SUCCESS},
|
|
{ 4, 10, 14, "Fake0123abcdef", "Ustr\0", 8, 10, 14, "FakeUstr\0\0cdef", STATUS_SUCCESS},
|
|
/* In the following test the native function writes beyond MaximumLength
|
|
* { 4, 9, 14, "Fake0123abcdef", "Ustr\0", 8, 9, 14, "FakeUstrabcdef", STATUS_SUCCESS},
|
|
*/
|
|
{ 4, 8, 14, "Fake0123abcdef", "Ustr\0", 8, 8, 14, "FakeUstrabcdef", STATUS_SUCCESS},
|
|
{ 4, 7, 14, "Fake0123abcdef", "Ustr\0", 4, 7, 14, "Fake0123abcdef", STATUS_BUFFER_TOO_SMALL},
|
|
{ 4, 0, 14, "Fake0123abcdef", "Ustr\0", 4, 0, 14, "Fake0123abcdef", STATUS_BUFFER_TOO_SMALL},
|
|
{ 4, 14, 14, "Fake0123abcdef", "Ustr\0", 8, 14, 14, "FakeUstr\0\0cdef", STATUS_SUCCESS},
|
|
{ 4, 14, 14, "Fake0123abcdef", NULL, 4, 14, 14, "Fake0123abcdef", STATUS_SUCCESS},
|
|
{ 4, 14, 14, NULL, NULL, 4, 14, 14, NULL, STATUS_SUCCESS},
|
|
{ 4, 14, 14, "Fake0123abcdef", "U\0stri\0", 10, 14, 14, "FakeU\0stri\0\0ef", STATUS_SUCCESS},
|
|
{ 6, 14, 16, "Te\0\0stabcdefghij", "St\0\0ri", 8, 14, 16, "Te\0\0stSt\0\0efghij", STATUS_SUCCESS},
|
|
};
|
|
#define NB_APP_UNI2STR (sizeof(app_uni2str)/sizeof(*app_uni2str))
|
|
|
|
|
|
static void test_RtlAppendUnicodeToString(void)
|
|
{
|
|
WCHAR dest_buf[257];
|
|
UNICODE_STRING dest_str;
|
|
NTSTATUS result;
|
|
unsigned int test_num;
|
|
|
|
for (test_num = 0; test_num < NB_APP_UNI2STR; test_num++) {
|
|
dest_str.Length = app_uni2str[test_num].dest_Length;
|
|
dest_str.MaximumLength = app_uni2str[test_num].dest_MaximumLength;
|
|
if (app_uni2str[test_num].dest_buf != NULL) {
|
|
memcpy(dest_buf, app_uni2str[test_num].dest_buf, app_uni2str[test_num].dest_buf_size);
|
|
dest_buf[app_uni2str[test_num].dest_buf_size/sizeof(WCHAR)] = '\0';
|
|
dest_str.Buffer = dest_buf;
|
|
} else {
|
|
dest_str.Buffer = NULL;
|
|
}
|
|
result = pRtlAppendUnicodeToString(&dest_str, (LPCWSTR) app_uni2str[test_num].src);
|
|
ok(result == app_uni2str[test_num].result,
|
|
"(test %d): RtlAppendUnicodeToString(dest, src) has result %x, expected %x\n",
|
|
test_num, result, app_uni2str[test_num].result);
|
|
ok(dest_str.Length == app_uni2str[test_num].res_Length,
|
|
"(test %d): RtlAppendUnicodeToString(dest, src) dest has Length %d, expected %d\n",
|
|
test_num, dest_str.Length, app_uni2str[test_num].res_Length);
|
|
ok(dest_str.MaximumLength == app_uni2str[test_num].res_MaximumLength,
|
|
"(test %d): RtlAppendUnicodeToString(dest, src) dest has MaximumLength %d, expected %d\n",
|
|
test_num, dest_str.MaximumLength, app_uni2str[test_num].res_MaximumLength);
|
|
if (dest_str.Buffer == dest_buf) {
|
|
ok(memcmp(dest_buf, app_uni2str[test_num].res_buf, app_uni2str[test_num].res_buf_size) == 0,
|
|
"(test %d): RtlAppendUnicodeToString(dest, src) has dest \"%s\" expected \"%s\"\n",
|
|
test_num, (char *) dest_buf, app_uni2str[test_num].res_buf);
|
|
} else {
|
|
ok(dest_str.Buffer == (WCHAR *) app_uni2str[test_num].res_buf,
|
|
"(test %d): RtlAppendUnicodeToString(dest, src) dest has Buffer %p expected %p\n",
|
|
test_num, dest_str.Buffer, app_uni2str[test_num].res_buf);
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
typedef struct {
|
|
int dest_Length;
|
|
int dest_MaximumLength;
|
|
int dest_buf_size;
|
|
const char *dest_buf;
|
|
int src_Length;
|
|
int src_MaximumLength;
|
|
int src_buf_size;
|
|
const char *src_buf;
|
|
int res_Length;
|
|
int res_MaximumLength;
|
|
int res_buf_size;
|
|
const char *res_buf;
|
|
NTSTATUS result;
|
|
} app_ustr2str_t;
|
|
|
|
static const app_ustr2str_t app_ustr2str[] = {
|
|
{ 4, 12, 14, "Fake0123abcdef", 4, 6, 8, "UstrZYXW", 8, 12, 14, "FakeUstr\0\0cdef", STATUS_SUCCESS},
|
|
{ 4, 11, 14, "Fake0123abcdef", 4, 6, 8, "UstrZYXW", 8, 11, 14, "FakeUstr\0\0cdef", STATUS_SUCCESS},
|
|
{ 4, 10, 14, "Fake0123abcdef", 4, 6, 8, "UstrZYXW", 8, 10, 14, "FakeUstr\0\0cdef", STATUS_SUCCESS},
|
|
/* In the following test the native function writes beyond MaximumLength
|
|
* { 4, 9, 14, "Fake0123abcdef", 4, 6, 8, "UstrZYXW", 8, 9, 14, "FakeUstrabcdef", STATUS_SUCCESS},
|
|
*/
|
|
{ 4, 8, 14, "Fake0123abcdef", 4, 6, 8, "UstrZYXW", 8, 8, 14, "FakeUstrabcdef", STATUS_SUCCESS},
|
|
{ 4, 7, 14, "Fake0123abcdef", 4, 6, 8, "UstrZYXW", 4, 7, 14, "Fake0123abcdef", STATUS_BUFFER_TOO_SMALL},
|
|
{ 4, 0, 14, "Fake0123abcdef", 0, 0, 8, "UstrZYXW", 4, 0, 14, "Fake0123abcdef", STATUS_SUCCESS},
|
|
{ 4, 14, 14, "Fake0123abcdef", 0, 0, 8, "UstrZYXW", 4, 14, 14, "Fake0123abcdef", STATUS_SUCCESS},
|
|
{ 4, 14, 14, "Fake0123abcdef", 0, 0, 8, NULL, 4, 14, 14, "Fake0123abcdef", STATUS_SUCCESS},
|
|
{ 4, 14, 14, NULL, 0, 0, 8, NULL, 4, 14, 14, NULL, STATUS_SUCCESS},
|
|
{ 6, 14, 16, "Te\0\0stabcdefghij", 6, 8, 8, "St\0\0riZY", 12, 14, 16, "Te\0\0stSt\0\0ri\0\0ij", STATUS_SUCCESS},
|
|
};
|
|
#define NB_APP_USTR2STR (sizeof(app_ustr2str)/sizeof(*app_ustr2str))
|
|
|
|
|
|
static void test_RtlAppendUnicodeStringToString(void)
|
|
{
|
|
WCHAR dest_buf[257];
|
|
WCHAR src_buf[257];
|
|
UNICODE_STRING dest_str;
|
|
UNICODE_STRING src_str;
|
|
NTSTATUS result;
|
|
unsigned int test_num;
|
|
|
|
for (test_num = 0; test_num < NB_APP_USTR2STR; test_num++) {
|
|
dest_str.Length = app_ustr2str[test_num].dest_Length;
|
|
dest_str.MaximumLength = app_ustr2str[test_num].dest_MaximumLength;
|
|
if (app_ustr2str[test_num].dest_buf != NULL) {
|
|
memcpy(dest_buf, app_ustr2str[test_num].dest_buf, app_ustr2str[test_num].dest_buf_size);
|
|
dest_buf[app_ustr2str[test_num].dest_buf_size/sizeof(WCHAR)] = '\0';
|
|
dest_str.Buffer = dest_buf;
|
|
} else {
|
|
dest_str.Buffer = NULL;
|
|
}
|
|
src_str.Length = app_ustr2str[test_num].src_Length;
|
|
src_str.MaximumLength = app_ustr2str[test_num].src_MaximumLength;
|
|
if (app_ustr2str[test_num].src_buf != NULL) {
|
|
memcpy(src_buf, app_ustr2str[test_num].src_buf, app_ustr2str[test_num].src_buf_size);
|
|
src_buf[app_ustr2str[test_num].src_buf_size/sizeof(WCHAR)] = '\0';
|
|
src_str.Buffer = src_buf;
|
|
} else {
|
|
src_str.Buffer = NULL;
|
|
}
|
|
result = pRtlAppendUnicodeStringToString(&dest_str, &src_str);
|
|
ok(result == app_ustr2str[test_num].result,
|
|
"(test %d): RtlAppendStringToString(dest, src) has result %x, expected %x\n",
|
|
test_num, result, app_ustr2str[test_num].result);
|
|
ok(dest_str.Length == app_ustr2str[test_num].res_Length,
|
|
"(test %d): RtlAppendStringToString(dest, src) dest has Length %d, expected %d\n",
|
|
test_num, dest_str.Length, app_ustr2str[test_num].res_Length);
|
|
ok(dest_str.MaximumLength == app_ustr2str[test_num].res_MaximumLength,
|
|
"(test %d): RtlAppendStringToString(dest, src) dest has MaximumLength %d, expected %d\n",
|
|
test_num, dest_str.MaximumLength, app_ustr2str[test_num].res_MaximumLength);
|
|
if (dest_str.Buffer == dest_buf) {
|
|
ok(memcmp(dest_buf, app_ustr2str[test_num].res_buf, app_ustr2str[test_num].res_buf_size) == 0,
|
|
"(test %d): RtlAppendStringToString(dest, src) has dest \"%s\" expected \"%s\"\n",
|
|
test_num, (char *) dest_buf, app_ustr2str[test_num].res_buf);
|
|
} else {
|
|
ok(dest_str.Buffer == (WCHAR *) app_ustr2str[test_num].res_buf,
|
|
"(test %d): RtlAppendStringToString(dest, src) dest has Buffer %p expected %p\n",
|
|
test_num, dest_str.Buffer, app_ustr2str[test_num].res_buf);
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
typedef struct {
|
|
int flags;
|
|
const char *main_str;
|
|
const char *search_chars;
|
|
USHORT pos;
|
|
NTSTATUS result;
|
|
} find_ch_in_ustr_t;
|
|
|
|
static const find_ch_in_ustr_t find_ch_in_ustr[] = {
|
|
{ 0, "Some Wild String", "S", 2, STATUS_SUCCESS},
|
|
{ 0, "This is a String", "String", 6, STATUS_SUCCESS},
|
|
{ 1, "This is a String", "String", 30, STATUS_SUCCESS},
|
|
{ 2, "This is a String", "String", 2, STATUS_SUCCESS},
|
|
{ 3, "This is a String", "String", 18, STATUS_SUCCESS},
|
|
{ 0, "This is a String", "Wild", 6, STATUS_SUCCESS},
|
|
{ 1, "This is a String", "Wild", 26, STATUS_SUCCESS},
|
|
{ 2, "This is a String", "Wild", 2, STATUS_SUCCESS},
|
|
{ 3, "This is a String", "Wild", 30, STATUS_SUCCESS},
|
|
{ 0, "abcdefghijklmnopqrstuvwxyz", "", 0, STATUS_NOT_FOUND},
|
|
{ 0, "abcdefghijklmnopqrstuvwxyz", "123", 0, STATUS_NOT_FOUND},
|
|
{ 0, "abcdefghijklmnopqrstuvwxyz", "a", 2, STATUS_SUCCESS},
|
|
{ 0, "abcdefghijklmnopqrstuvwxyz", "12a34", 2, STATUS_SUCCESS},
|
|
{ 0, "abcdefghijklmnopqrstuvwxyz", "12b34", 4, STATUS_SUCCESS},
|
|
{ 0, "abcdefghijklmnopqrstuvwxyz", "12y34", 50, STATUS_SUCCESS},
|
|
{ 0, "abcdefghijklmnopqrstuvwxyz", "12z34", 52, STATUS_SUCCESS},
|
|
{ 0, "abcdefghijklmnopqrstuvwxyz", "rvz", 36, STATUS_SUCCESS},
|
|
{ 0, "abcdefghijklmmlkjihgfedcba", "egik", 10, STATUS_SUCCESS},
|
|
{ 1, "abcdefghijklmnopqrstuvwxyz", "", 0, STATUS_NOT_FOUND},
|
|
{ 1, "abcdefghijklmnopqrstuvwxyz", "rvz", 50, STATUS_SUCCESS},
|
|
{ 1, "abcdefghijklmnopqrstuvwxyz", "ravy", 48, STATUS_SUCCESS},
|
|
{ 1, "abcdefghijklmnopqrstuvwxyz", "raxv", 46, STATUS_SUCCESS},
|
|
{ 2, "abcdefghijklmnopqrstuvwxyz", "", 2, STATUS_SUCCESS},
|
|
{ 2, "abcdefghijklmnopqrstuvwxyz", "rvz", 2, STATUS_SUCCESS},
|
|
{ 2, "abcdefghijklmnopqrstuvwxyz", "vaz", 4, STATUS_SUCCESS},
|
|
{ 2, "abcdefghijklmnopqrstuvwxyz", "ravbz", 6, STATUS_SUCCESS},
|
|
{ 3, "abcdefghijklmnopqrstuvwxyz", "", 50, STATUS_SUCCESS},
|
|
{ 3, "abcdefghijklmnopqrstuvwxyz", "123", 50, STATUS_SUCCESS},
|
|
{ 3, "abcdefghijklmnopqrstuvwxyz", "ahp", 50, STATUS_SUCCESS},
|
|
{ 3, "abcdefghijklmnopqrstuvwxyz", "rvz", 48, STATUS_SUCCESS},
|
|
{ 0, NULL, "abc", 0, STATUS_NOT_FOUND},
|
|
{ 1, NULL, "abc", 0, STATUS_NOT_FOUND},
|
|
{ 2, NULL, "abc", 0, STATUS_NOT_FOUND},
|
|
{ 3, NULL, "abc", 0, STATUS_NOT_FOUND},
|
|
{ 0, "abcdefghijklmnopqrstuvwxyz", NULL, 0, STATUS_NOT_FOUND},
|
|
{ 1, "abcdefghijklmnopqrstuvwxyz", NULL, 0, STATUS_NOT_FOUND},
|
|
{ 2, "abcdefghijklmnopqrstuvwxyz", NULL, 2, STATUS_SUCCESS},
|
|
{ 3, "abcdefghijklmnopqrstuvwxyz", NULL, 50, STATUS_SUCCESS},
|
|
{ 0, NULL, NULL, 0, STATUS_NOT_FOUND},
|
|
{ 1, NULL, NULL, 0, STATUS_NOT_FOUND},
|
|
{ 2, NULL, NULL, 0, STATUS_NOT_FOUND},
|
|
{ 3, NULL, NULL, 0, STATUS_NOT_FOUND},
|
|
{ 0, "abcdabcdabcdabcdabcdabcd", "abcd", 2, STATUS_SUCCESS},
|
|
{ 1, "abcdabcdabcdabcdabcdabcd", "abcd", 46, STATUS_SUCCESS},
|
|
{ 2, "abcdabcdabcdabcdabcdabcd", "abcd", 0, STATUS_NOT_FOUND},
|
|
{ 3, "abcdabcdabcdabcdabcdabcd", "abcd", 0, STATUS_NOT_FOUND},
|
|
};
|
|
#define NB_FIND_CH_IN_USTR (sizeof(find_ch_in_ustr)/sizeof(*find_ch_in_ustr))
|
|
|
|
|
|
static void test_RtlFindCharInUnicodeString(void)
|
|
{
|
|
WCHAR main_str_buf[257];
|
|
WCHAR search_chars_buf[257];
|
|
UNICODE_STRING main_str;
|
|
UNICODE_STRING search_chars;
|
|
USHORT pos;
|
|
NTSTATUS result;
|
|
unsigned int idx;
|
|
unsigned int test_num;
|
|
|
|
if (!pRtlFindCharInUnicodeString)
|
|
{
|
|
win_skip("RtlFindCharInUnicodeString is not available\n");
|
|
return;
|
|
}
|
|
|
|
for (test_num = 0; test_num < NB_FIND_CH_IN_USTR; test_num++) {
|
|
if (find_ch_in_ustr[test_num].main_str != NULL) {
|
|
main_str.Length = strlen(find_ch_in_ustr[test_num].main_str) * sizeof(WCHAR);
|
|
main_str.MaximumLength = main_str.Length + sizeof(WCHAR);
|
|
for (idx = 0; idx < main_str.Length / sizeof(WCHAR); idx++) {
|
|
main_str_buf[idx] = find_ch_in_ustr[test_num].main_str[idx];
|
|
}
|
|
main_str.Buffer = main_str_buf;
|
|
} else {
|
|
main_str.Length = 0;
|
|
main_str.MaximumLength = 0;
|
|
main_str.Buffer = NULL;
|
|
}
|
|
if (find_ch_in_ustr[test_num].search_chars != NULL) {
|
|
search_chars.Length = strlen(find_ch_in_ustr[test_num].search_chars) * sizeof(WCHAR);
|
|
search_chars.MaximumLength = search_chars.Length + sizeof(WCHAR);
|
|
for (idx = 0; idx < search_chars.Length / sizeof(WCHAR); idx++) {
|
|
search_chars_buf[idx] = find_ch_in_ustr[test_num].search_chars[idx];
|
|
}
|
|
search_chars.Buffer = search_chars_buf;
|
|
} else {
|
|
search_chars.Length = 0;
|
|
search_chars.MaximumLength = 0;
|
|
search_chars.Buffer = NULL;
|
|
}
|
|
pos = 12345;
|
|
result = pRtlFindCharInUnicodeString(find_ch_in_ustr[test_num].flags, &main_str, &search_chars, &pos);
|
|
ok(result == find_ch_in_ustr[test_num].result,
|
|
"(test %d): RtlFindCharInUnicodeString(%d, %s, %s, [out]) has result %x, expected %x\n",
|
|
test_num, find_ch_in_ustr[test_num].flags,
|
|
find_ch_in_ustr[test_num].main_str, find_ch_in_ustr[test_num].search_chars,
|
|
result, find_ch_in_ustr[test_num].result);
|
|
ok(pos == find_ch_in_ustr[test_num].pos,
|
|
"(test %d): RtlFindCharInUnicodeString(%d, %s, %s, [out]) assigns %d to pos, expected %d\n",
|
|
test_num, find_ch_in_ustr[test_num].flags,
|
|
find_ch_in_ustr[test_num].main_str, find_ch_in_ustr[test_num].search_chars,
|
|
pos, find_ch_in_ustr[test_num].pos);
|
|
}
|
|
}
|
|
|
|
|
|
typedef struct {
|
|
int base;
|
|
const char *str;
|
|
int value;
|
|
NTSTATUS result, alternative;
|
|
} str2int_t;
|
|
|
|
static const str2int_t str2int[] = {
|
|
{ 0, "1011101100", 1011101100, STATUS_SUCCESS},
|
|
{ 0, "1234567", 1234567, STATUS_SUCCESS},
|
|
{ 0, "-214", -214, STATUS_SUCCESS},
|
|
{ 0, "+214", 214, STATUS_SUCCESS}, /* The + sign is allowed also */
|
|
{ 0, "--214", 0, STATUS_SUCCESS}, /* Do not accept more than one sign */
|
|
{ 0, "-+214", 0, STATUS_SUCCESS},
|
|
{ 0, "++214", 0, STATUS_SUCCESS},
|
|
{ 0, "+-214", 0, STATUS_SUCCESS},
|
|
{ 0, "\001\002\003\00411", 11, STATUS_SUCCESS}, /* whitespace char 1 to 4 */
|
|
{ 0, "\005\006\007\01012", 12, STATUS_SUCCESS}, /* whitespace char 5 to 8 */
|
|
{ 0, "\011\012\013\01413", 13, STATUS_SUCCESS}, /* whitespace char 9 to 12 */
|
|
{ 0, "\015\016\017\02014", 14, STATUS_SUCCESS}, /* whitespace char 13 to 16 */
|
|
{ 0, "\021\022\023\02415", 15, STATUS_SUCCESS}, /* whitespace char 17 to 20 */
|
|
{ 0, "\025\026\027\03016", 16, STATUS_SUCCESS}, /* whitespace char 21 to 24 */
|
|
{ 0, "\031\032\033\03417", 17, STATUS_SUCCESS}, /* whitespace char 25 to 28 */
|
|
{ 0, "\035\036\037\04018", 18, STATUS_SUCCESS}, /* whitespace char 29 to 32 */
|
|
{ 0, " \n \r \t214", 214, STATUS_SUCCESS},
|
|
{ 0, " \n \r \t+214", 214, STATUS_SUCCESS}, /* Signs can be used after whitespace */
|
|
{ 0, " \n \r \t-214", -214, STATUS_SUCCESS},
|
|
{ 0, "+214 0", 214, STATUS_SUCCESS}, /* Space terminates the number */
|
|
{ 0, " 214.01", 214, STATUS_SUCCESS}, /* Decimal point not accepted */
|
|
{ 0, " 214,01", 214, STATUS_SUCCESS}, /* Decimal comma not accepted */
|
|
{ 0, "f81", 0, STATUS_SUCCESS},
|
|
{ 0, "0x12345", 0x12345, STATUS_SUCCESS}, /* Hex */
|
|
{ 0, "00x12345", 0, STATUS_SUCCESS},
|
|
{ 0, "0xx12345", 0, STATUS_SUCCESS},
|
|
{ 0, "1x34", 1, STATUS_SUCCESS},
|
|
{ 0, "-9999999999", -1410065407, STATUS_SUCCESS}, /* Big negative integer */
|
|
{ 0, "-2147483649", 2147483647, STATUS_SUCCESS}, /* Too small to fit in 32 Bits */
|
|
{ 0, "-2147483648", 0x80000000L, STATUS_SUCCESS}, /* Smallest negative integer */
|
|
{ 0, "-2147483647", -2147483647, STATUS_SUCCESS},
|
|
{ 0, "-1", -1, STATUS_SUCCESS},
|
|
{ 0, "0", 0, STATUS_SUCCESS},
|
|
{ 0, "1", 1, STATUS_SUCCESS},
|
|
{ 0, "2147483646", 2147483646, STATUS_SUCCESS},
|
|
{ 0, "2147483647", 2147483647, STATUS_SUCCESS}, /* Largest signed positive integer */
|
|
{ 0, "2147483648", 0x80000000L, STATUS_SUCCESS}, /* Positive int equal to smallest negative int */
|
|
{ 0, "2147483649", -2147483647, STATUS_SUCCESS},
|
|
{ 0, "4294967294", -2, STATUS_SUCCESS},
|
|
{ 0, "4294967295", -1, STATUS_SUCCESS}, /* Largest unsigned integer */
|
|
{ 0, "4294967296", 0, STATUS_SUCCESS}, /* Too big to fit in 32 Bits */
|
|
{ 0, "9999999999", 1410065407, STATUS_SUCCESS}, /* Big positive integer */
|
|
{ 0, "056789", 56789, STATUS_SUCCESS}, /* Leading zero and still decimal */
|
|
{ 0, "b1011101100", 0, STATUS_SUCCESS}, /* Binary (b-notation) */
|
|
{ 0, "-b1011101100", 0, STATUS_SUCCESS}, /* Negative Binary (b-notation) */
|
|
{ 0, "b10123456789", 0, STATUS_SUCCESS}, /* Binary with nonbinary digits (2-9) */
|
|
{ 0, "0b1011101100", 748, STATUS_SUCCESS}, /* Binary (0b-notation) */
|
|
{ 0, "-0b1011101100", -748, STATUS_SUCCESS}, /* Negative binary (0b-notation) */
|
|
{ 0, "0b10123456789", 5, STATUS_SUCCESS}, /* Binary with nonbinary digits (2-9) */
|
|
{ 0, "-0b10123456789", -5, STATUS_SUCCESS}, /* Negative binary with nonbinary digits (2-9) */
|
|
{ 0, "0b1", 1, STATUS_SUCCESS}, /* one digit binary */
|
|
{ 0, "0b2", 0, STATUS_SUCCESS}, /* empty binary */
|
|
{ 0, "0b", 0, STATUS_SUCCESS}, /* empty binary */
|
|
{ 0, "o1234567", 0, STATUS_SUCCESS}, /* Octal (o-notation) */
|
|
{ 0, "-o1234567", 0, STATUS_SUCCESS}, /* Negative Octal (o-notation) */
|
|
{ 0, "o56789", 0, STATUS_SUCCESS}, /* Octal with nonoctal digits (8 and 9) */
|
|
{ 0, "0o1234567", 01234567, STATUS_SUCCESS}, /* Octal (0o-notation) */
|
|
{ 0, "-0o1234567", -01234567, STATUS_SUCCESS}, /* Negative octal (0o-notation) */
|
|
{ 0, "0o56789", 0567, STATUS_SUCCESS}, /* Octal with nonoctal digits (8 and 9) */
|
|
{ 0, "-0o56789", -0567, STATUS_SUCCESS}, /* Negative octal with nonoctal digits (8 and 9) */
|
|
{ 0, "0o7", 7, STATUS_SUCCESS}, /* one digit octal */
|
|
{ 0, "0o8", 0, STATUS_SUCCESS}, /* empty octal */
|
|
{ 0, "0o", 0, STATUS_SUCCESS}, /* empty octal */
|
|
{ 0, "0d1011101100", 0, STATUS_SUCCESS}, /* explicit decimal with 0d */
|
|
{ 0, "x89abcdef", 0, STATUS_SUCCESS}, /* Hex with lower case digits a-f (x-notation) */
|
|
{ 0, "xFEDCBA00", 0, STATUS_SUCCESS}, /* Hex with upper case digits A-F (x-notation) */
|
|
{ 0, "-xFEDCBA00", 0, STATUS_SUCCESS}, /* Negative Hexadecimal (x-notation) */
|
|
{ 0, "0x89abcdef", 0x89abcdef, STATUS_SUCCESS}, /* Hex with lower case digits a-f (0x-notation) */
|
|
{ 0, "0xFEDCBA00", 0xFEDCBA00, STATUS_SUCCESS}, /* Hex with upper case digits A-F (0x-notation) */
|
|
{ 0, "-0xFEDCBA00", 19088896, STATUS_SUCCESS}, /* Negative Hexadecimal (0x-notation) */
|
|
{ 0, "0xabcdefgh", 0xabcdef, STATUS_SUCCESS}, /* Hex with illegal lower case digits (g-z) */
|
|
{ 0, "0xABCDEFGH", 0xABCDEF, STATUS_SUCCESS}, /* Hex with illegal upper case digits (G-Z) */
|
|
{ 0, "0xF", 0xf, STATUS_SUCCESS}, /* one digit hexadecimal */
|
|
{ 0, "0xG", 0, STATUS_SUCCESS}, /* empty hexadecimal */
|
|
{ 0, "0x", 0, STATUS_SUCCESS}, /* empty hexadecimal */
|
|
{ 0, "", 0, STATUS_SUCCESS, STATUS_INVALID_PARAMETER}, /* empty string */
|
|
{ 2, "1011101100", 748, STATUS_SUCCESS},
|
|
{ 2, "-1011101100", -748, STATUS_SUCCESS},
|
|
{ 2, "2", 0, STATUS_SUCCESS},
|
|
{ 2, "0b1011101100", 0, STATUS_SUCCESS},
|
|
{ 2, "0o1011101100", 0, STATUS_SUCCESS},
|
|
{ 2, "0d1011101100", 0, STATUS_SUCCESS},
|
|
{ 2, "0x1011101100", 0, STATUS_SUCCESS},
|
|
{ 2, "", 0, STATUS_SUCCESS, STATUS_INVALID_PARAMETER}, /* empty string */
|
|
{ 8, "1011101100", 136610368, STATUS_SUCCESS},
|
|
{ 8, "-1011101100", -136610368, STATUS_SUCCESS},
|
|
{ 8, "8", 0, STATUS_SUCCESS},
|
|
{ 8, "0b1011101100", 0, STATUS_SUCCESS},
|
|
{ 8, "0o1011101100", 0, STATUS_SUCCESS},
|
|
{ 8, "0d1011101100", 0, STATUS_SUCCESS},
|
|
{ 8, "0x1011101100", 0, STATUS_SUCCESS},
|
|
{ 8, "", 0, STATUS_SUCCESS, STATUS_INVALID_PARAMETER}, /* empty string */
|
|
{10, "1011101100", 1011101100, STATUS_SUCCESS},
|
|
{10, "-1011101100", -1011101100, STATUS_SUCCESS},
|
|
{10, "0b1011101100", 0, STATUS_SUCCESS},
|
|
{10, "0o1011101100", 0, STATUS_SUCCESS},
|
|
{10, "0d1011101100", 0, STATUS_SUCCESS},
|
|
{10, "0x1011101100", 0, STATUS_SUCCESS},
|
|
{10, "o12345", 0, STATUS_SUCCESS}, /* Octal although base is 10 */
|
|
{10, "", 0, STATUS_SUCCESS, STATUS_INVALID_PARAMETER}, /* empty string */
|
|
{16, "1011101100", 286265600, STATUS_SUCCESS},
|
|
{16, "-1011101100", -286265600, STATUS_SUCCESS},
|
|
{16, "G", 0, STATUS_SUCCESS},
|
|
{16, "g", 0, STATUS_SUCCESS},
|
|
{16, "0b1011101100", 286265600, STATUS_SUCCESS},
|
|
{16, "0o1011101100", 0, STATUS_SUCCESS},
|
|
{16, "0d1011101100", 286265600, STATUS_SUCCESS},
|
|
{16, "0x1011101100", 0, STATUS_SUCCESS},
|
|
{16, "", 0, STATUS_SUCCESS, STATUS_INVALID_PARAMETER}, /* empty string */
|
|
{20, "0", 0, STATUS_INVALID_PARAMETER}, /* illegal base */
|
|
{-8, "0", 0, STATUS_INVALID_PARAMETER}, /* Negative base */
|
|
/* { 0, NULL, 0, STATUS_SUCCESS}, */ /* NULL as string */
|
|
};
|
|
#define NB_STR2INT (sizeof(str2int)/sizeof(*str2int))
|
|
|
|
|
|
static void test_RtlUnicodeStringToInteger(void)
|
|
{
|
|
unsigned int test_num;
|
|
int value;
|
|
NTSTATUS result;
|
|
WCHAR *wstr;
|
|
UNICODE_STRING uni;
|
|
|
|
for (test_num = 0; test_num < NB_STR2INT; test_num++) {
|
|
wstr = AtoW(str2int[test_num].str);
|
|
value = 0xdeadbeef;
|
|
pRtlInitUnicodeString(&uni, wstr);
|
|
result = pRtlUnicodeStringToInteger(&uni, str2int[test_num].base, &value);
|
|
ok(result == str2int[test_num].result ||
|
|
(str2int[test_num].alternative && result == str2int[test_num].alternative),
|
|
"(test %d): RtlUnicodeStringToInteger(\"%s\", %d, [out]) has result %x, expected: %x (%x)\n",
|
|
test_num, str2int[test_num].str, str2int[test_num].base, result,
|
|
str2int[test_num].result, str2int[test_num].alternative);
|
|
if (result == STATUS_SUCCESS)
|
|
ok(value == str2int[test_num].value ||
|
|
broken(str2int[test_num].str[0] == '\0' && str2int[test_num].base == 16), /* nt4 */
|
|
"(test %d): RtlUnicodeStringToInteger(\"%s\", %d, [out]) assigns value %d, expected: %d\n",
|
|
test_num, str2int[test_num].str, str2int[test_num].base, value, str2int[test_num].value);
|
|
else
|
|
ok(value == 0xdeadbeef || value == 0 /* vista */,
|
|
"(test %d): RtlUnicodeStringToInteger(\"%s\", %d, [out]) assigns value %d, expected 0 or deadbeef\n",
|
|
test_num, str2int[test_num].str, str2int[test_num].base, value);
|
|
HeapFree(GetProcessHeap(), 0, wstr);
|
|
}
|
|
|
|
wstr = AtoW(str2int[1].str);
|
|
pRtlInitUnicodeString(&uni, wstr);
|
|
result = pRtlUnicodeStringToInteger(&uni, str2int[1].base, NULL);
|
|
ok(result == STATUS_ACCESS_VIOLATION,
|
|
"call failed: RtlUnicodeStringToInteger(\"%s\", %d, NULL) has result %x\n",
|
|
str2int[1].str, str2int[1].base, result);
|
|
result = pRtlUnicodeStringToInteger(&uni, 20, NULL);
|
|
ok(result == STATUS_INVALID_PARAMETER || result == STATUS_ACCESS_VIOLATION,
|
|
"call failed: RtlUnicodeStringToInteger(\"%s\", 20, NULL) has result %x\n",
|
|
str2int[1].str, result);
|
|
|
|
uni.Length = 10; /* Make Length shorter (5 WCHARS instead of 7) */
|
|
result = pRtlUnicodeStringToInteger(&uni, str2int[1].base, &value);
|
|
ok(result == STATUS_SUCCESS,
|
|
"call failed: RtlUnicodeStringToInteger(\"12345\", %d, [out]) has result %x\n",
|
|
str2int[1].base, result);
|
|
ok(value == 12345,
|
|
"didn't return expected value (test a): expected: %d, got: %d\n",
|
|
12345, value);
|
|
|
|
uni.Length = 5; /* Use odd Length (2.5 WCHARS) */
|
|
result = pRtlUnicodeStringToInteger(&uni, str2int[1].base, &value);
|
|
ok(result == STATUS_SUCCESS || result == STATUS_INVALID_PARAMETER /* vista */,
|
|
"call failed: RtlUnicodeStringToInteger(\"12\", %d, [out]) has result %x\n",
|
|
str2int[1].base, result);
|
|
if (result == STATUS_SUCCESS)
|
|
ok(value == 12, "didn't return expected value (test b): expected: %d, got: %d\n", 12, value);
|
|
|
|
uni.Length = 2;
|
|
result = pRtlUnicodeStringToInteger(&uni, str2int[1].base, &value);
|
|
ok(result == STATUS_SUCCESS,
|
|
"call failed: RtlUnicodeStringToInteger(\"1\", %d, [out]) has result %x\n",
|
|
str2int[1].base, result);
|
|
ok(value == 1,
|
|
"didn't return expected value (test c): expected: %d, got: %d\n",
|
|
1, value);
|
|
/* w2k: uni.Length = 0 returns value 11234567 instead of 0 */
|
|
HeapFree(GetProcessHeap(), 0, wstr);
|
|
}
|
|
|
|
|
|
static void test_RtlCharToInteger(void)
|
|
{
|
|
unsigned int test_num;
|
|
int value;
|
|
NTSTATUS result;
|
|
|
|
for (test_num = 0; test_num < NB_STR2INT; test_num++) {
|
|
/* w2k skips a leading '\0' and processes the string after */
|
|
if (str2int[test_num].str[0] != '\0') {
|
|
value = 0xdeadbeef;
|
|
result = pRtlCharToInteger(str2int[test_num].str, str2int[test_num].base, &value);
|
|
ok(result == str2int[test_num].result ||
|
|
(str2int[test_num].alternative && result == str2int[test_num].alternative),
|
|
"(test %d): call failed: RtlCharToInteger(\"%s\", %d, [out]) has result %x, expected: %x (%x)\n",
|
|
test_num, str2int[test_num].str, str2int[test_num].base, result,
|
|
str2int[test_num].result, str2int[test_num].alternative);
|
|
if (result == STATUS_SUCCESS)
|
|
ok(value == str2int[test_num].value,
|
|
"(test %d): call failed: RtlCharToInteger(\"%s\", %d, [out]) assigns value %d, expected: %d\n",
|
|
test_num, str2int[test_num].str, str2int[test_num].base, value, str2int[test_num].value);
|
|
else
|
|
ok(value == 0 || value == 0xdeadbeef,
|
|
"(test %d): call failed: RtlCharToInteger(\"%s\", %d, [out]) assigns value %d, expected 0 or deadbeef\n",
|
|
test_num, str2int[test_num].str, str2int[test_num].base, value);
|
|
}
|
|
}
|
|
|
|
result = pRtlCharToInteger(str2int[1].str, str2int[1].base, NULL);
|
|
ok(result == STATUS_ACCESS_VIOLATION,
|
|
"call failed: RtlCharToInteger(\"%s\", %d, NULL) has result %x\n",
|
|
str2int[1].str, str2int[1].base, result);
|
|
|
|
result = pRtlCharToInteger(str2int[1].str, 20, NULL);
|
|
ok(result == STATUS_INVALID_PARAMETER,
|
|
"call failed: RtlCharToInteger(\"%s\", 20, NULL) has result %x\n",
|
|
str2int[1].str, result);
|
|
}
|
|
|
|
|
|
#define STRI_BUFFER_LENGTH 35
|
|
|
|
typedef struct {
|
|
int base;
|
|
ULONG value;
|
|
USHORT Length;
|
|
USHORT MaximumLength;
|
|
const char *Buffer;
|
|
NTSTATUS result;
|
|
} int2str_t;
|
|
|
|
static const int2str_t int2str[] = {
|
|
{10, 123, 3, 11, "123\0-------------------------------", STATUS_SUCCESS},
|
|
|
|
{ 0, 0x80000000U, 10, 11, "2147483648\0------------------------", STATUS_SUCCESS}, /* min signed int */
|
|
{ 0, -2147483647, 10, 11, "2147483649\0------------------------", STATUS_SUCCESS},
|
|
{ 0, -2, 10, 11, "4294967294\0------------------------", STATUS_SUCCESS},
|
|
{ 0, -1, 10, 11, "4294967295\0------------------------", STATUS_SUCCESS},
|
|
{ 0, 0, 1, 11, "0\0---------------------------------", STATUS_SUCCESS},
|
|
{ 0, 1, 1, 11, "1\0---------------------------------", STATUS_SUCCESS},
|
|
{ 0, 12, 2, 11, "12\0--------------------------------", STATUS_SUCCESS},
|
|
{ 0, 123, 3, 11, "123\0-------------------------------", STATUS_SUCCESS},
|
|
{ 0, 1234, 4, 11, "1234\0------------------------------", STATUS_SUCCESS},
|
|
{ 0, 12345, 5, 11, "12345\0-----------------------------", STATUS_SUCCESS},
|
|
{ 0, 123456, 6, 11, "123456\0----------------------------", STATUS_SUCCESS},
|
|
{ 0, 1234567, 7, 11, "1234567\0---------------------------", STATUS_SUCCESS},
|
|
{ 0, 12345678, 8, 11, "12345678\0--------------------------", STATUS_SUCCESS},
|
|
{ 0, 123456789, 9, 11, "123456789\0-------------------------", STATUS_SUCCESS},
|
|
{ 0, 2147483646, 10, 11, "2147483646\0------------------------", STATUS_SUCCESS},
|
|
{ 0, 2147483647, 10, 11, "2147483647\0------------------------", STATUS_SUCCESS}, /* max signed int */
|
|
{ 0, 2147483648U, 10, 11, "2147483648\0------------------------", STATUS_SUCCESS}, /* uint = -max int */
|
|
{ 0, 2147483649U, 10, 11, "2147483649\0------------------------", STATUS_SUCCESS},
|
|
{ 0, 4294967294U, 10, 11, "4294967294\0------------------------", STATUS_SUCCESS},
|
|
{ 0, 4294967295U, 10, 11, "4294967295\0------------------------", STATUS_SUCCESS}, /* max unsigned int */
|
|
|
|
{ 2, 0x80000000U, 32, 33, "10000000000000000000000000000000\0--", STATUS_SUCCESS}, /* min signed int */
|
|
{ 2, -2147483647, 32, 33, "10000000000000000000000000000001\0--", STATUS_SUCCESS},
|
|
{ 2, -2, 32, 33, "11111111111111111111111111111110\0--", STATUS_SUCCESS},
|
|
{ 2, -1, 32, 33, "11111111111111111111111111111111\0--", STATUS_SUCCESS},
|
|
{ 2, 0, 1, 33, "0\0---------------------------------", STATUS_SUCCESS},
|
|
{ 2, 1, 1, 33, "1\0---------------------------------", STATUS_SUCCESS},
|
|
{ 2, 10, 4, 33, "1010\0------------------------------", STATUS_SUCCESS},
|
|
{ 2, 100, 7, 33, "1100100\0---------------------------", STATUS_SUCCESS},
|
|
{ 2, 1000, 10, 33, "1111101000\0------------------------", STATUS_SUCCESS},
|
|
{ 2, 10000, 14, 33, "10011100010000\0--------------------", STATUS_SUCCESS},
|
|
{ 2, 32767, 15, 33, "111111111111111\0-------------------", STATUS_SUCCESS},
|
|
/* { 2, 32768, 16, 33, "1000000000000000\0------------------", STATUS_SUCCESS}, broken on windows */
|
|
/* { 2, 65535, 16, 33, "1111111111111111\0------------------", STATUS_SUCCESS}, broken on windows */
|
|
{ 2, 65536, 17, 33, "10000000000000000\0-----------------", STATUS_SUCCESS},
|
|
{ 2, 100000, 17, 33, "11000011010100000\0-----------------", STATUS_SUCCESS},
|
|
{ 2, 1000000, 20, 33, "11110100001001000000\0--------------", STATUS_SUCCESS},
|
|
{ 2, 10000000, 24, 33, "100110001001011010000000\0----------", STATUS_SUCCESS},
|
|
{ 2, 100000000, 27, 33, "101111101011110000100000000\0-------", STATUS_SUCCESS},
|
|
{ 2, 1000000000, 30, 33, "111011100110101100101000000000\0----", STATUS_SUCCESS},
|
|
{ 2, 1073741823, 30, 33, "111111111111111111111111111111\0----", STATUS_SUCCESS},
|
|
{ 2, 2147483646, 31, 33, "1111111111111111111111111111110\0---", STATUS_SUCCESS},
|
|
{ 2, 2147483647, 31, 33, "1111111111111111111111111111111\0---", STATUS_SUCCESS}, /* max signed int */
|
|
{ 2, 2147483648U, 32, 33, "10000000000000000000000000000000\0--", STATUS_SUCCESS}, /* uint = -max int */
|
|
{ 2, 2147483649U, 32, 33, "10000000000000000000000000000001\0--", STATUS_SUCCESS},
|
|
{ 2, 4294967294U, 32, 33, "11111111111111111111111111111110\0--", STATUS_SUCCESS},
|
|
{ 2, 4294967295U, 32, 33, "11111111111111111111111111111111\0--", STATUS_SUCCESS}, /* max unsigned int */
|
|
|
|
{ 8, 0x80000000U, 11, 12, "20000000000\0-----------------------", STATUS_SUCCESS}, /* min signed int */
|
|
{ 8, -2147483647, 11, 12, "20000000001\0-----------------------", STATUS_SUCCESS},
|
|
{ 8, -2, 11, 12, "37777777776\0-----------------------", STATUS_SUCCESS},
|
|
{ 8, -1, 11, 12, "37777777777\0-----------------------", STATUS_SUCCESS},
|
|
{ 8, 0, 1, 12, "0\0---------------------------------", STATUS_SUCCESS},
|
|
{ 8, 1, 1, 12, "1\0---------------------------------", STATUS_SUCCESS},
|
|
{ 8, 2147483646, 11, 12, "17777777776\0-----------------------", STATUS_SUCCESS},
|
|
{ 8, 2147483647, 11, 12, "17777777777\0-----------------------", STATUS_SUCCESS}, /* max signed int */
|
|
{ 8, 2147483648U, 11, 12, "20000000000\0-----------------------", STATUS_SUCCESS}, /* uint = -max int */
|
|
{ 8, 2147483649U, 11, 12, "20000000001\0-----------------------", STATUS_SUCCESS},
|
|
{ 8, 4294967294U, 11, 12, "37777777776\0-----------------------", STATUS_SUCCESS},
|
|
{ 8, 4294967295U, 11, 12, "37777777777\0-----------------------", STATUS_SUCCESS}, /* max unsigned int */
|
|
|
|
{10, 0x80000000U, 10, 11, "2147483648\0------------------------", STATUS_SUCCESS}, /* min signed int */
|
|
{10, -2147483647, 10, 11, "2147483649\0------------------------", STATUS_SUCCESS},
|
|
{10, -2, 10, 11, "4294967294\0------------------------", STATUS_SUCCESS},
|
|
{10, -1, 10, 11, "4294967295\0------------------------", STATUS_SUCCESS},
|
|
{10, 0, 1, 11, "0\0---------------------------------", STATUS_SUCCESS},
|
|
{10, 1, 1, 11, "1\0---------------------------------", STATUS_SUCCESS},
|
|
{10, 2147483646, 10, 11, "2147483646\0------------------------", STATUS_SUCCESS},
|
|
{10, 2147483647, 10, 11, "2147483647\0------------------------", STATUS_SUCCESS}, /* max signed int */
|
|
{10, 2147483648U, 10, 11, "2147483648\0------------------------", STATUS_SUCCESS}, /* uint = -max int */
|
|
{10, 2147483649U, 10, 11, "2147483649\0------------------------", STATUS_SUCCESS},
|
|
{10, 4294967294U, 10, 11, "4294967294\0------------------------", STATUS_SUCCESS},
|
|
{10, 4294967295U, 10, 11, "4294967295\0------------------------", STATUS_SUCCESS}, /* max unsigned int */
|
|
|
|
{16, 0x80000000U, 8, 9, "80000000\0--------------------------", STATUS_SUCCESS}, /* min signed int */
|
|
{16, -2147483647, 8, 9, "80000001\0--------------------------", STATUS_SUCCESS},
|
|
{16, -2, 8, 9, "FFFFFFFE\0--------------------------", STATUS_SUCCESS},
|
|
{16, -1, 8, 9, "FFFFFFFF\0--------------------------", STATUS_SUCCESS},
|
|
{16, 0, 1, 9, "0\0---------------------------------", STATUS_SUCCESS},
|
|
{16, 1, 1, 9, "1\0---------------------------------", STATUS_SUCCESS},
|
|
{16, 2147483646, 8, 9, "7FFFFFFE\0--------------------------", STATUS_SUCCESS},
|
|
{16, 2147483647, 8, 9, "7FFFFFFF\0--------------------------", STATUS_SUCCESS}, /* max signed int */
|
|
{16, 2147483648U, 8, 9, "80000000\0--------------------------", STATUS_SUCCESS}, /* uint = -max int */
|
|
{16, 2147483649U, 8, 9, "80000001\0--------------------------", STATUS_SUCCESS},
|
|
{16, 4294967294U, 8, 9, "FFFFFFFE\0--------------------------", STATUS_SUCCESS},
|
|
{16, 4294967295U, 8, 9, "FFFFFFFF\0--------------------------", STATUS_SUCCESS}, /* max unsigned int */
|
|
|
|
/* { 2, 32768, 16, 17, "1000000000000000\0------------------", STATUS_SUCCESS}, broken on windows */
|
|
/* { 2, 32768, 16, 16, "1000000000000000-------------------", STATUS_SUCCESS}, broken on windows */
|
|
{ 2, 65536, 17, 18, "10000000000000000\0-----------------", STATUS_SUCCESS},
|
|
{ 2, 65536, 17, 17, "10000000000000000------------------", STATUS_SUCCESS},
|
|
{ 2, 131072, 18, 19, "100000000000000000\0----------------", STATUS_SUCCESS},
|
|
{ 2, 131072, 18, 18, "100000000000000000-----------------", STATUS_SUCCESS},
|
|
{16, 0xffffffff, 8, 9, "FFFFFFFF\0--------------------------", STATUS_SUCCESS},
|
|
{16, 0xffffffff, 8, 8, "FFFFFFFF---------------------------", STATUS_SUCCESS}, /* No \0 term */
|
|
{16, 0xffffffff, 8, 7, "-----------------------------------", STATUS_BUFFER_OVERFLOW}, /* Too short */
|
|
{16, 0xa, 1, 2, "A\0---------------------------------", STATUS_SUCCESS},
|
|
{16, 0xa, 1, 1, "A----------------------------------", STATUS_SUCCESS}, /* No \0 term */
|
|
{16, 0, 1, 0, "-----------------------------------", STATUS_BUFFER_OVERFLOW},
|
|
{20, 0xdeadbeef, 0, 9, "-----------------------------------", STATUS_INVALID_PARAMETER}, /* ill. base */
|
|
{-8, 07654321, 0, 12, "-----------------------------------", STATUS_INVALID_PARAMETER}, /* neg. base */
|
|
};
|
|
#define NB_INT2STR (sizeof(int2str)/sizeof(*int2str))
|
|
|
|
|
|
static void one_RtlIntegerToUnicodeString_test(int test_num, const int2str_t *int2str)
|
|
{
|
|
int pos;
|
|
WCHAR expected_str_Buffer[STRI_BUFFER_LENGTH + 1];
|
|
UNICODE_STRING expected_unicode_string;
|
|
STRING expected_ansi_str;
|
|
WCHAR str_Buffer[STRI_BUFFER_LENGTH + 1];
|
|
UNICODE_STRING unicode_string;
|
|
STRING ansi_str;
|
|
NTSTATUS result;
|
|
|
|
for (pos = 0; pos < STRI_BUFFER_LENGTH; pos++) {
|
|
expected_str_Buffer[pos] = int2str->Buffer[pos];
|
|
}
|
|
expected_unicode_string.Length = int2str->Length * sizeof(WCHAR);
|
|
expected_unicode_string.MaximumLength = int2str->MaximumLength * sizeof(WCHAR);
|
|
expected_unicode_string.Buffer = expected_str_Buffer;
|
|
pRtlUnicodeStringToAnsiString(&expected_ansi_str, &expected_unicode_string, 1);
|
|
|
|
for (pos = 0; pos < STRI_BUFFER_LENGTH; pos++) {
|
|
str_Buffer[pos] = '-';
|
|
}
|
|
unicode_string.Length = 0;
|
|
unicode_string.MaximumLength = int2str->MaximumLength * sizeof(WCHAR);
|
|
unicode_string.Buffer = str_Buffer;
|
|
|
|
result = pRtlIntegerToUnicodeString(int2str->value, int2str->base, &unicode_string);
|
|
pRtlUnicodeStringToAnsiString(&ansi_str, &unicode_string, 1);
|
|
if (result == STATUS_BUFFER_OVERFLOW) {
|
|
/* On BUFFER_OVERFLOW the string Buffer should be unchanged */
|
|
for (pos = 0; pos < STRI_BUFFER_LENGTH; pos++) {
|
|
expected_str_Buffer[pos] = '-';
|
|
}
|
|
/* w2k: The native function has two reasons for BUFFER_OVERFLOW: */
|
|
/* If the value is too large to convert: The Length is unchanged */
|
|
/* If str is too small to hold the string: Set str->Length to the length */
|
|
/* the string would have (which can be larger than the MaximumLength). */
|
|
/* To allow all this in the tests we do the following: */
|
|
if (expected_unicode_string.Length > 32 && unicode_string.Length == 0) {
|
|
/* The value is too large to convert only triggered when testing native */
|
|
expected_unicode_string.Length = 0;
|
|
}
|
|
} else {
|
|
ok(result == int2str->result,
|
|
"(test %d): RtlIntegerToUnicodeString(%u, %d, [out]) has result %x, expected: %x\n",
|
|
test_num, int2str->value, int2str->base, result, int2str->result);
|
|
if (result == STATUS_SUCCESS) {
|
|
ok(unicode_string.Buffer[unicode_string.Length/sizeof(WCHAR)] == '\0',
|
|
"(test %d): RtlIntegerToUnicodeString(%u, %d, [out]) string \"%s\" is not NULL terminated\n",
|
|
test_num, int2str->value, int2str->base, ansi_str.Buffer);
|
|
}
|
|
}
|
|
ok(memcmp(unicode_string.Buffer, expected_unicode_string.Buffer, STRI_BUFFER_LENGTH * sizeof(WCHAR)) == 0,
|
|
"(test %d): RtlIntegerToUnicodeString(%u, %d, [out]) assigns string \"%s\", expected: \"%s\"\n",
|
|
test_num, int2str->value, int2str->base, ansi_str.Buffer, expected_ansi_str.Buffer);
|
|
ok(unicode_string.Length == expected_unicode_string.Length,
|
|
"(test %d): RtlIntegerToUnicodeString(%u, %d, [out]) string has Length %d, expected: %d\n",
|
|
test_num, int2str->value, int2str->base, unicode_string.Length, expected_unicode_string.Length);
|
|
ok(unicode_string.MaximumLength == expected_unicode_string.MaximumLength,
|
|
"(test %d): RtlIntegerToUnicodeString(%u, %d, [out]) string has MaximumLength %d, expected: %d\n",
|
|
test_num, int2str->value, int2str->base, unicode_string.MaximumLength, expected_unicode_string.MaximumLength);
|
|
pRtlFreeAnsiString(&expected_ansi_str);
|
|
pRtlFreeAnsiString(&ansi_str);
|
|
}
|
|
|
|
|
|
static void test_RtlIntegerToUnicodeString(void)
|
|
{
|
|
size_t test_num;
|
|
|
|
for (test_num = 0; test_num < NB_INT2STR; test_num++)
|
|
one_RtlIntegerToUnicodeString_test(test_num, &int2str[test_num]);
|
|
}
|
|
|
|
|
|
static void one_RtlIntegerToChar_test(int test_num, const int2str_t *int2str)
|
|
{
|
|
NTSTATUS result;
|
|
char dest_str[STRI_BUFFER_LENGTH + 1];
|
|
|
|
memset(dest_str, '-', STRI_BUFFER_LENGTH);
|
|
dest_str[STRI_BUFFER_LENGTH] = '\0';
|
|
result = pRtlIntegerToChar(int2str->value, int2str->base, int2str->MaximumLength, dest_str);
|
|
ok(result == int2str->result,
|
|
"(test %d): RtlIntegerToChar(%u, %d, %d, [out]) has result %x, expected: %x\n",
|
|
test_num, int2str->value, int2str->base, int2str->MaximumLength, result, int2str->result);
|
|
ok(memcmp(dest_str, int2str->Buffer, STRI_BUFFER_LENGTH) == 0,
|
|
"(test %d): RtlIntegerToChar(%u, %d, %d, [out]) assigns string \"%s\", expected: \"%s\"\n",
|
|
test_num, int2str->value, int2str->base, int2str->MaximumLength, dest_str, int2str->Buffer);
|
|
}
|
|
|
|
|
|
static void test_RtlIntegerToChar(void)
|
|
{
|
|
NTSTATUS result;
|
|
size_t test_num;
|
|
|
|
for (test_num = 0; test_num < NB_INT2STR; test_num++)
|
|
one_RtlIntegerToChar_test(test_num, &int2str[test_num]);
|
|
|
|
result = pRtlIntegerToChar(int2str[0].value, 20, int2str[0].MaximumLength, NULL);
|
|
ok(result == STATUS_INVALID_PARAMETER,
|
|
"(test a): RtlIntegerToChar(%u, %d, %d, NULL) has result %x, expected: %x\n",
|
|
int2str[0].value, 20, int2str[0].MaximumLength, result, STATUS_INVALID_PARAMETER);
|
|
|
|
result = pRtlIntegerToChar(int2str[0].value, 20, 0, NULL);
|
|
ok(result == STATUS_INVALID_PARAMETER,
|
|
"(test b): RtlIntegerToChar(%u, %d, %d, NULL) has result %x, expected: %x\n",
|
|
int2str[0].value, 20, 0, result, STATUS_INVALID_PARAMETER);
|
|
|
|
result = pRtlIntegerToChar(int2str[0].value, int2str[0].base, 0, NULL);
|
|
ok(result == STATUS_BUFFER_OVERFLOW,
|
|
"(test c): RtlIntegerToChar(%u, %d, %d, NULL) has result %x, expected: %x\n",
|
|
int2str[0].value, int2str[0].base, 0, result, STATUS_BUFFER_OVERFLOW);
|
|
|
|
result = pRtlIntegerToChar(int2str[0].value, int2str[0].base, int2str[0].MaximumLength, NULL);
|
|
ok(result == STATUS_ACCESS_VIOLATION,
|
|
"(test d): RtlIntegerToChar(%u, %d, %d, NULL) has result %x, expected: %x\n",
|
|
int2str[0].value, int2str[0].base, int2str[0].MaximumLength, result, STATUS_ACCESS_VIOLATION);
|
|
}
|
|
|
|
static void test_RtlIsTextUnicode(void)
|
|
{
|
|
char ascii[] = "A simple string";
|
|
char false_positive[] = {0x41, 0x0a, 0x0d, 0x1d};
|
|
WCHAR false_negative = 0x0d0a;
|
|
WCHAR unicode[] = {'A',' ','U','n','i','c','o','d','e',' ','s','t','r','i','n','g',0};
|
|
WCHAR unicode_no_controls[] = {'A','U','n','i','c','o','d','e','s','t','r','i','n','g',0};
|
|
/* String with both byte-reversed and standard Unicode control characters. */
|
|
WCHAR mixed_controls[] = {'\t',0x9000,0x0d00,'\n',0};
|
|
WCHAR *be_unicode;
|
|
WCHAR *be_unicode_no_controls;
|
|
BOOLEAN res;
|
|
int flags;
|
|
int i;
|
|
|
|
if (!pRtlIsTextUnicode)
|
|
{
|
|
win_skip("RtlIsTextUnicode is not available\n");
|
|
return;
|
|
}
|
|
|
|
ok(!pRtlIsTextUnicode(ascii, sizeof(ascii), NULL), "ASCII text detected as Unicode\n");
|
|
|
|
res = pRtlIsTextUnicode(unicode, sizeof(unicode), NULL);
|
|
ok(res ||
|
|
broken(res == FALSE), /* NT4 */
|
|
"Text should be Unicode\n");
|
|
|
|
ok(!pRtlIsTextUnicode(unicode, sizeof(unicode) - 1, NULL), "Text should be Unicode\n");
|
|
|
|
flags = IS_TEXT_UNICODE_UNICODE_MASK;
|
|
ok(pRtlIsTextUnicode(unicode, sizeof(unicode), &flags), "Text should not pass a Unicode\n");
|
|
ok(flags == (IS_TEXT_UNICODE_STATISTICS | IS_TEXT_UNICODE_CONTROLS),
|
|
"Expected flags 0x6, obtained %x\n", flags);
|
|
|
|
flags = IS_TEXT_UNICODE_REVERSE_MASK;
|
|
ok(!pRtlIsTextUnicode(unicode, sizeof(unicode), &flags), "Text should not pass reverse Unicode tests\n");
|
|
ok(flags == 0, "Expected flags 0, obtained %x\n", flags);
|
|
|
|
flags = IS_TEXT_UNICODE_ODD_LENGTH;
|
|
ok(!pRtlIsTextUnicode(unicode, sizeof(unicode) - 1, &flags), "Odd length test should have passed\n");
|
|
ok(flags == IS_TEXT_UNICODE_ODD_LENGTH, "Expected flags 0x200, obtained %x\n", flags);
|
|
|
|
be_unicode = HeapAlloc(GetProcessHeap(), 0, sizeof(unicode) + sizeof(WCHAR));
|
|
be_unicode[0] = 0xfffe;
|
|
for (i = 0; i < sizeof(unicode)/sizeof(unicode[0]); i++)
|
|
{
|
|
be_unicode[i + 1] = (unicode[i] >> 8) | ((unicode[i] & 0xff) << 8);
|
|
}
|
|
ok(!pRtlIsTextUnicode(be_unicode, sizeof(unicode) + 2, NULL), "Reverse endian should not be Unicode\n");
|
|
ok(!pRtlIsTextUnicode(&be_unicode[1], sizeof(unicode), NULL), "Reverse endian should not be Unicode\n");
|
|
|
|
flags = IS_TEXT_UNICODE_REVERSE_MASK;
|
|
ok(!pRtlIsTextUnicode(&be_unicode[1], sizeof(unicode), &flags), "Reverse endian should be Unicode\n");
|
|
todo_wine
|
|
ok(flags == (IS_TEXT_UNICODE_REVERSE_ASCII16 | IS_TEXT_UNICODE_REVERSE_STATISTICS | IS_TEXT_UNICODE_REVERSE_CONTROLS),
|
|
"Expected flags 0x70, obtained %x\n", flags);
|
|
|
|
flags = IS_TEXT_UNICODE_REVERSE_MASK;
|
|
ok(!pRtlIsTextUnicode(be_unicode, sizeof(unicode) + 2, &flags), "Reverse endian should be Unicode\n");
|
|
ok(flags == (IS_TEXT_UNICODE_REVERSE_CONTROLS | IS_TEXT_UNICODE_REVERSE_SIGNATURE),
|
|
"Expected flags 0xc0, obtained %x\n", flags);
|
|
|
|
/* build byte reversed unicode string with no control chars */
|
|
be_unicode_no_controls = HeapAlloc(GetProcessHeap(), 0, sizeof(unicode) + sizeof(WCHAR));
|
|
ok(be_unicode_no_controls != NULL, "Expected HeapAlloc to succeed.\n");
|
|
be_unicode_no_controls[0] = 0xfffe;
|
|
for (i = 0; i < sizeof(unicode_no_controls)/sizeof(unicode_no_controls[0]); i++)
|
|
be_unicode_no_controls[i + 1] = (unicode_no_controls[i] >> 8) | ((unicode_no_controls[i] & 0xff) << 8);
|
|
|
|
|
|
/* The following tests verify that the tests for */
|
|
/* IS_TEXT_UNICODE_CONTROLS and IS_TEXT_UNICODE_REVERSE_CONTROLS */
|
|
/* are not mutually exclusive. Regardless of whether the strings */
|
|
/* contain an indication of endianness, the tests are still */
|
|
/* run if the flag is passed to (Rtl)IsTextUnicode. */
|
|
|
|
/* Test IS_TEXT_UNICODE_CONTROLS flag */
|
|
flags = IS_TEXT_UNICODE_CONTROLS;
|
|
ok(!pRtlIsTextUnicode(unicode_no_controls, sizeof(unicode_no_controls), &flags), "Test should not pass on Unicode string lacking control characters.\n");
|
|
ok(flags == 0, "Expected flags 0x0, obtained %x\n", flags);
|
|
|
|
flags = IS_TEXT_UNICODE_CONTROLS;
|
|
ok(!pRtlIsTextUnicode(be_unicode_no_controls, sizeof(unicode_no_controls), &flags), "Test should not pass on byte-reversed Unicode string lacking control characters.\n");
|
|
ok(flags == 0, "Expected flags 0x0, obtained %x\n", flags);
|
|
|
|
flags = IS_TEXT_UNICODE_CONTROLS;
|
|
ok(pRtlIsTextUnicode(unicode, sizeof(unicode), &flags), "Test should pass on Unicode string lacking control characters.\n");
|
|
ok(flags == IS_TEXT_UNICODE_CONTROLS, "Expected flags 0x04, obtained %x\n", flags);
|
|
|
|
flags = IS_TEXT_UNICODE_CONTROLS;
|
|
ok(!pRtlIsTextUnicode(be_unicode_no_controls, sizeof(unicode_no_controls) + 2, &flags),
|
|
"Test should not pass with standard Unicode string.\n");
|
|
ok(flags == 0, "Expected flags 0x0, obtained %x\n", flags);
|
|
|
|
flags = IS_TEXT_UNICODE_CONTROLS;
|
|
ok(pRtlIsTextUnicode(mixed_controls, sizeof(mixed_controls), &flags), "Test should pass on a string containing control characters.\n");
|
|
ok(flags == IS_TEXT_UNICODE_CONTROLS, "Expected flags 0x04, obtained %x\n", flags);
|
|
|
|
/* Test IS_TEXT_UNICODE_REVERSE_CONTROLS flag */
|
|
flags = IS_TEXT_UNICODE_REVERSE_CONTROLS;
|
|
ok(!pRtlIsTextUnicode(be_unicode_no_controls, sizeof(unicode_no_controls), &flags), "Test should not pass on Unicode string lacking control characters.\n");
|
|
ok(flags == 0, "Expected flags 0x0, obtained %x\n", flags);
|
|
|
|
flags = IS_TEXT_UNICODE_REVERSE_CONTROLS;
|
|
ok(!pRtlIsTextUnicode(unicode_no_controls, sizeof(unicode_no_controls), &flags), "Test should not pass on Unicode string lacking control characters.\n");
|
|
ok(flags == 0, "Expected flags 0x0, obtained %x\n", flags);
|
|
|
|
flags = IS_TEXT_UNICODE_REVERSE_CONTROLS;
|
|
ok(!pRtlIsTextUnicode(unicode, sizeof(unicode), &flags), "Test should not pass on Unicode string lacking control characters.\n");
|
|
ok(flags == 0, "Expected flags 0x0, obtained %x\n", flags);
|
|
|
|
flags = IS_TEXT_UNICODE_REVERSE_CONTROLS;
|
|
ok(!pRtlIsTextUnicode(be_unicode, sizeof(unicode) + 2, &flags),
|
|
"Test should pass with byte-reversed Unicode string containing control characters.\n");
|
|
ok(flags == IS_TEXT_UNICODE_REVERSE_CONTROLS, "Expected flags 0x40, obtained %x\n", flags);
|
|
|
|
flags = IS_TEXT_UNICODE_REVERSE_CONTROLS;
|
|
ok(!pRtlIsTextUnicode(mixed_controls, sizeof(mixed_controls), &flags), "Test should pass on a string containing byte-reversed control characters.\n");
|
|
ok(flags == IS_TEXT_UNICODE_REVERSE_CONTROLS, "Expected flags 0x40, obtained %x\n", flags);
|
|
|
|
/* Test with flags for both byte-reverse and standard Unicode characters */
|
|
flags = IS_TEXT_UNICODE_CONTROLS | IS_TEXT_UNICODE_REVERSE_CONTROLS;
|
|
ok(!pRtlIsTextUnicode(mixed_controls, sizeof(mixed_controls), &flags), "Test should pass on string containing both byte-reversed and standard control characters.\n");
|
|
ok(flags == (IS_TEXT_UNICODE_CONTROLS | IS_TEXT_UNICODE_REVERSE_CONTROLS), "Expected flags 0x44, obtained %x\n", flags);
|
|
|
|
flags = IS_TEXT_UNICODE_STATISTICS;
|
|
todo_wine ok(pRtlIsTextUnicode(false_positive, sizeof(false_positive), &flags), "Test should pass on false positive.\n");
|
|
|
|
ok(!pRtlIsTextUnicode(&false_negative, sizeof(false_negative), NULL), "Test should fail on 0x0d0a (MALAYALAM LETTER UU).\n");
|
|
|
|
HeapFree(GetProcessHeap(), 0, be_unicode);
|
|
HeapFree(GetProcessHeap(), 0, be_unicode_no_controls);
|
|
}
|
|
|
|
static void test_RtlCompareUnicodeString(void)
|
|
{
|
|
WCHAR ch1, ch2;
|
|
UNICODE_STRING str1, str2;
|
|
|
|
str1.Buffer = &ch1;
|
|
str1.Length = str1.MaximumLength = sizeof(WCHAR);
|
|
str2.Buffer = &ch2;
|
|
str2.Length = str2.MaximumLength = sizeof(WCHAR);
|
|
for (ch1 = 0; ch1 < 512; ch1++)
|
|
{
|
|
for (ch2 = 0; ch2 < 1024; ch2++)
|
|
{
|
|
LONG res = pRtlCompareUnicodeString( &str1, &str2, FALSE );
|
|
ok( res == (ch1 - ch2), "wrong result %d %04x %04x\n", res, ch1, ch2 );
|
|
res = pRtlCompareUnicodeString( &str1, &str2, TRUE );
|
|
ok( res == (pRtlUpcaseUnicodeChar(ch1) - pRtlUpcaseUnicodeChar(ch2)),
|
|
"wrong result %d %04x %04x\n", res, ch1, ch2 );
|
|
if (pRtlCompareUnicodeStrings)
|
|
{
|
|
res = pRtlCompareUnicodeStrings( &ch1, 1, &ch2, 1, FALSE );
|
|
ok( res == (ch1 - ch2), "wrong result %d %04x %04x\n", res, ch1, ch2 );
|
|
res = pRtlCompareUnicodeStrings( &ch1, 1, &ch2, 1, TRUE );
|
|
ok( res == (pRtlUpcaseUnicodeChar(ch1) - pRtlUpcaseUnicodeChar(ch2)),
|
|
"wrong result %d %04x %04x\n", res, ch1, ch2 );
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
static const WCHAR szGuid[] = { '{','0','1','0','2','0','3','0','4','-',
|
|
'0','5','0','6','-' ,'0','7','0','8','-','0','9','0','A','-',
|
|
'0','B','0','C','0','D','0','E','0','F','0','A','}','\0' };
|
|
static const WCHAR szGuid2[] = { '{','0','1','0','2','0','3','0','4','-',
|
|
'0','5','0','6','-' ,'0','7','0','8','-','0','9','0','A','-',
|
|
'0','B','0','C','0','D','0','E','0','F','0','A',']','\0' };
|
|
DEFINE_GUID(IID_Endianness, 0x01020304, 0x0506, 0x0708, 0x09, 0x0A, 0x0B,
|
|
0x0C, 0x0D, 0x0E, 0x0F, 0x0A);
|
|
|
|
static void test_RtlGUIDFromString(void)
|
|
{
|
|
GUID guid;
|
|
UNICODE_STRING str;
|
|
NTSTATUS ret;
|
|
|
|
if (!pRtlGUIDFromString)
|
|
{
|
|
win_skip("RtlGUIDFromString is not available\n");
|
|
return;
|
|
}
|
|
|
|
str.Length = str.MaximumLength = sizeof(szGuid) - sizeof(WCHAR);
|
|
str.Buffer = (LPWSTR)szGuid;
|
|
|
|
ret = pRtlGUIDFromString(&str, &guid);
|
|
ok(ret == 0, "expected ret=0, got 0x%0x\n", ret);
|
|
ok(IsEqualGUID(&guid, &IID_Endianness), "Endianness broken\n");
|
|
|
|
str.Length = str.MaximumLength = sizeof(szGuid2) - sizeof(WCHAR);
|
|
str.Buffer = (LPWSTR)szGuid2;
|
|
|
|
ret = pRtlGUIDFromString(&str, &guid);
|
|
ok(ret, "expected ret!=0\n");
|
|
}
|
|
|
|
static void test_RtlStringFromGUID(void)
|
|
{
|
|
UNICODE_STRING str;
|
|
NTSTATUS ret;
|
|
|
|
if (!pRtlStringFromGUID)
|
|
{
|
|
win_skip("RtlStringFromGUID is not available\n");
|
|
return;
|
|
}
|
|
|
|
str.Length = str.MaximumLength = 0;
|
|
str.Buffer = NULL;
|
|
|
|
ret = pRtlStringFromGUID(&IID_Endianness, &str);
|
|
ok(ret == 0, "expected ret=0, got 0x%0x\n", ret);
|
|
ok(str.Buffer && !lstrcmpiW(str.Buffer, szGuid), "Endianness broken\n");
|
|
pRtlFreeUnicodeString(&str);
|
|
}
|
|
|
|
struct hash_unicodestring_test {
|
|
WCHAR str[50];
|
|
BOOLEAN case_insensitive;
|
|
ULONG hash;
|
|
};
|
|
|
|
static const struct hash_unicodestring_test hash_test[] = {
|
|
{ {'T',0}, FALSE, 0x00000054 },
|
|
{ {'T','e','s','t',0}, FALSE, 0x766bb952 },
|
|
{ {'T','e','S','t',0}, FALSE, 0x764bb172 },
|
|
{ {'t','e','s','t',0}, FALSE, 0x4745d132 },
|
|
{ {'t','e','s','t',0}, TRUE, 0x6689c132 },
|
|
{ {'T','E','S','T',0}, TRUE, 0x6689c132 },
|
|
{ {'T','E','S','T',0}, FALSE, 0x6689c132 },
|
|
{ {'a','b','c','d','e','f',0}, FALSE, 0x971318c3 },
|
|
{ { 0 } }
|
|
};
|
|
|
|
static void test_RtlHashUnicodeString(void)
|
|
{
|
|
static const WCHAR strW[] = {'T','e','s','t',0,'1',0};
|
|
const struct hash_unicodestring_test *ptr;
|
|
UNICODE_STRING str;
|
|
NTSTATUS status;
|
|
ULONG hash;
|
|
|
|
if (!pRtlHashUnicodeString)
|
|
{
|
|
win_skip("RtlHashUnicodeString is not available\n");
|
|
return;
|
|
}
|
|
|
|
status = pRtlHashUnicodeString(NULL, FALSE, HASH_STRING_ALGORITHM_X65599, &hash);
|
|
ok(status == STATUS_INVALID_PARAMETER, "got status 0x%08x\n", status);
|
|
|
|
RtlInitUnicodeString(&str, strW);
|
|
status = pRtlHashUnicodeString(&str, FALSE, HASH_STRING_ALGORITHM_X65599, NULL);
|
|
ok(status == STATUS_INVALID_PARAMETER, "got status 0x%08x\n", status);
|
|
|
|
status = pRtlHashUnicodeString(&str, FALSE, HASH_STRING_ALGORITHM_INVALID, &hash);
|
|
ok(status == STATUS_INVALID_PARAMETER, "got status 0x%08x\n", status);
|
|
|
|
/* embedded null */
|
|
str.Buffer = (PWSTR)strW;
|
|
str.Length = sizeof(strW) - sizeof(WCHAR);
|
|
str.MaximumLength = sizeof(strW);
|
|
status = pRtlHashUnicodeString(&str, FALSE, HASH_STRING_ALGORITHM_X65599, &hash);
|
|
ok(status == STATUS_SUCCESS, "got status 0x%08x\n", status);
|
|
ok(hash == 0x32803083, "got 0x%08x\n", hash);
|
|
|
|
ptr = hash_test;
|
|
while (*ptr->str)
|
|
{
|
|
RtlInitUnicodeString(&str, ptr->str);
|
|
hash = 0;
|
|
status = pRtlHashUnicodeString(&str, ptr->case_insensitive, HASH_STRING_ALGORITHM_X65599, &hash);
|
|
ok(status == STATUS_SUCCESS, "got status 0x%08x for %s\n", status, wine_dbgstr_w(ptr->str));
|
|
ok(hash == ptr->hash, "got wrong hash 0x%08x, expected 0x%08x, for %s, mode %d\n", hash, ptr->hash,
|
|
wine_dbgstr_w(ptr->str), ptr->case_insensitive);
|
|
|
|
ptr++;
|
|
}
|
|
}
|
|
|
|
struct unicode_to_utf8_test {
|
|
WCHAR unicode[128];
|
|
const char *expected;
|
|
NTSTATUS status;
|
|
};
|
|
|
|
static const struct unicode_to_utf8_test unicode_to_utf8[] = {
|
|
{ { 0 }, "", STATUS_SUCCESS },
|
|
{ { '-',0 }, "-", STATUS_SUCCESS },
|
|
{ { 'h','e','l','l','o',0 }, "hello", STATUS_SUCCESS },
|
|
{ { '-',0x7f,'-',0x80,'-',0xff,'-',0x100,'-',0 }, "-\x7F-\xC2\x80-\xC3\xBF-\xC4\x80-", STATUS_SUCCESS },
|
|
{ { '-',0x7ff,'-',0x800,'-',0 }, "-\xDF\xBF-\xE0\xA0\x80-", STATUS_SUCCESS },
|
|
{ { '-',0xd7ff,'-',0xe000,'-',0 }, "-\xED\x9F\xBF-\xEE\x80\x80-", STATUS_SUCCESS },
|
|
/* 0x10000 */
|
|
{ { '-',0xffff,'-',0xd800,0xdc00,'-',0 }, "-\xEF\xBF\xBF-\xF0\x90\x80\x80-", STATUS_SUCCESS },
|
|
/* 0x103ff */ /* 0x10400 */
|
|
{ { '-',0xd800,0xdfff,'-',0xd801,0xdc00,'-',0 }, "-\xF0\x90\x8F\xBF-\xF0\x90\x90\x80-", STATUS_SUCCESS },
|
|
/* 0x10ffff */
|
|
{ { '-',0xdbff,0xdfff,'-',0 }, "-\xF4\x8F\xBF\xBF-", STATUS_SUCCESS },
|
|
/* standalone lead surrogates become 0xFFFD */
|
|
{ { '-',0xd800,'-',0xdbff,'-',0 }, "-\xEF\xBF\xBD-\xEF\xBF\xBD-", STATUS_SOME_NOT_MAPPED },
|
|
/* standalone trail surrogates become 0xFFFD */
|
|
{ { '-',0xdc00,'-',0xdfff,'-',0 }, "-\xEF\xBF\xBD-\xEF\xBF\xBD-", STATUS_SOME_NOT_MAPPED },
|
|
/* reverse surrogate pair */
|
|
{ { '-',0xdfff,0xdbff,'-',0 }, "-\xEF\xBF\xBD\xEF\xBF\xBD-", STATUS_SOME_NOT_MAPPED },
|
|
/* byte order marks */
|
|
{ { '-',0xfeff,'-',0xfffe,'-',0 }, "-\xEF\xBB\xBF-\xEF\xBF\xBE-", STATUS_SUCCESS },
|
|
{ { 0xfeff,'-',0 }, "\xEF\xBB\xBF-", STATUS_SUCCESS },
|
|
{ { 0xfffe,'-',0 }, "\xEF\xBF\xBE-", STATUS_SUCCESS },
|
|
/* invalid code point */
|
|
{ { 0xffff,'-',0 }, "\xEF\xBF\xBF-", STATUS_SUCCESS },
|
|
/* canonically equivalent representations -- no normalization should happen */
|
|
{ { '-',0x1e09,'-',0 }, "-\xE1\xB8\x89-", STATUS_SUCCESS },
|
|
{ { '-',0x0107,0x0327,'-',0 }, "-\xC4\x87\xCC\xA7-", STATUS_SUCCESS },
|
|
{ { '-',0x00e7,0x0301,'-',0 }, "-\xC3\xA7\xCC\x81-", STATUS_SUCCESS },
|
|
{ { '-',0x0063,0x0327,0x0301,'-',0 }, "-\x63\xCC\xA7\xCC\x81-", STATUS_SUCCESS },
|
|
{ { '-',0x0063,0x0301,0x0327,'-',0 }, "-\x63\xCC\x81\xCC\xA7-", STATUS_SUCCESS },
|
|
};
|
|
|
|
static void utf8_expect_(const unsigned char *out_string, ULONG buflen, ULONG out_bytes,
|
|
const WCHAR *in_string, ULONG in_bytes,
|
|
NTSTATUS expect_status, int line)
|
|
{
|
|
NTSTATUS status;
|
|
ULONG bytes_out;
|
|
char buffer[128];
|
|
unsigned char *buf = (unsigned char *)buffer;
|
|
unsigned int i;
|
|
|
|
if (buflen == (ULONG)-1)
|
|
buflen = sizeof(buffer);
|
|
bytes_out = 0x55555555;
|
|
memset(buffer, 0x55, sizeof(buffer));
|
|
status = pRtlUnicodeToUTF8N(
|
|
out_string ? buffer : NULL, buflen, &bytes_out,
|
|
in_string, in_bytes);
|
|
ok_(__FILE__, line)(status == expect_status, "status = 0x%x\n", status);
|
|
ok_(__FILE__, line)(bytes_out == out_bytes, "bytes_out = %u\n", bytes_out);
|
|
if (out_string)
|
|
{
|
|
for (i = 0; i < bytes_out; i++)
|
|
ok_(__FILE__, line)(buf[i] == out_string[i],
|
|
"buffer[%d] = 0x%x, expected 0x%x\n",
|
|
i, buf[i], out_string[i]);
|
|
for (; i < sizeof(buffer); i++)
|
|
ok_(__FILE__, line)(buf[i] == 0x55,
|
|
"buffer[%d] = 0x%x, expected 0x55\n",
|
|
i, buf[i]);
|
|
}
|
|
}
|
|
#define utf8_expect(out_string, buflen, out_bytes, in_string, in_bytes, expect_status) \
|
|
utf8_expect_(out_string, buflen, out_bytes, in_string, in_bytes, expect_status, __LINE__)
|
|
|
|
static void test_RtlUnicodeToUTF8N(void)
|
|
{
|
|
NTSTATUS status;
|
|
ULONG bytes_out;
|
|
ULONG bytes_out_array[2];
|
|
void * const invalid_pointer = (void *)0x8;
|
|
char buffer[128];
|
|
const WCHAR empty_string[] = { 0 };
|
|
const WCHAR test_string[] = { 'A',0,'a','b','c','d','e','f','g',0 };
|
|
const WCHAR special_string[] = { 'X',0x80,0xd800,0 };
|
|
const unsigned char special_expected[] = { 'X',0xc2,0x80,0xef,0xbf,0xbd,0 };
|
|
unsigned int input_len;
|
|
const unsigned int test_count = sizeof(unicode_to_utf8) / sizeof(unicode_to_utf8[0]);
|
|
unsigned int i;
|
|
|
|
if (!pRtlUnicodeToUTF8N)
|
|
{
|
|
skip("RtlUnicodeToUTF8N unavailable\n");
|
|
return;
|
|
}
|
|
|
|
/* show that bytes_out is really ULONG */
|
|
memset(bytes_out_array, 0x55, sizeof(bytes_out_array));
|
|
status = pRtlUnicodeToUTF8N(NULL, 0, bytes_out_array, empty_string, 0);
|
|
ok(status == STATUS_SUCCESS, "status = 0x%x\n", status);
|
|
ok(bytes_out_array[0] == 0x00000000, "Got 0x%x\n", bytes_out_array[0]);
|
|
ok(bytes_out_array[1] == 0x55555555, "Got 0x%x\n", bytes_out_array[1]);
|
|
|
|
/* parameter checks */
|
|
status = pRtlUnicodeToUTF8N(NULL, 0, NULL, NULL, 0);
|
|
ok(status == STATUS_INVALID_PARAMETER_4, "status = 0x%x\n", status);
|
|
|
|
status = pRtlUnicodeToUTF8N(NULL, 0, NULL, empty_string, 0);
|
|
ok(status == STATUS_INVALID_PARAMETER, "status = 0x%x\n", status);
|
|
|
|
bytes_out = 0x55555555;
|
|
status = pRtlUnicodeToUTF8N(NULL, 0, &bytes_out, NULL, 0);
|
|
ok(status == STATUS_INVALID_PARAMETER_4, "status = 0x%x\n", status);
|
|
ok(bytes_out == 0x55555555, "bytes_out = 0x%x\n", bytes_out);
|
|
|
|
bytes_out = 0x55555555;
|
|
status = pRtlUnicodeToUTF8N(NULL, 0, &bytes_out, invalid_pointer, 0);
|
|
ok(status == STATUS_SUCCESS, "status = 0x%x\n", status);
|
|
ok(bytes_out == 0, "bytes_out = 0x%x\n", bytes_out);
|
|
|
|
bytes_out = 0x55555555;
|
|
status = pRtlUnicodeToUTF8N(NULL, 0, &bytes_out, empty_string, 0);
|
|
ok(status == STATUS_SUCCESS, "status = 0x%x\n", status);
|
|
ok(bytes_out == 0, "bytes_out = 0x%x\n", bytes_out);
|
|
|
|
bytes_out = 0x55555555;
|
|
status = pRtlUnicodeToUTF8N(NULL, 0, &bytes_out, test_string, 0);
|
|
ok(status == STATUS_SUCCESS, "status = 0x%x\n", status);
|
|
ok(bytes_out == 0, "bytes_out = 0x%x\n", bytes_out);
|
|
|
|
bytes_out = 0x55555555;
|
|
status = pRtlUnicodeToUTF8N(NULL, 0, &bytes_out, empty_string, 1);
|
|
ok(status == STATUS_SUCCESS, "status = 0x%x\n", status);
|
|
ok(bytes_out == 0, "bytes_out = 0x%x\n", bytes_out);
|
|
|
|
bytes_out = 0x55555555;
|
|
status = pRtlUnicodeToUTF8N(invalid_pointer, 0, &bytes_out, empty_string, 1);
|
|
ok(status == STATUS_INVALID_PARAMETER_5, "status = 0x%x\n", status);
|
|
ok(bytes_out == 0x55555555, "bytes_out = 0x%x\n", bytes_out);
|
|
|
|
bytes_out = 0x55555555;
|
|
status = pRtlUnicodeToUTF8N(invalid_pointer, 8, &bytes_out, empty_string, 1);
|
|
ok(status == STATUS_INVALID_PARAMETER_5, "status = 0x%x\n", status);
|
|
ok(bytes_out == 0x55555555, "bytes_out = 0x%x\n", bytes_out);
|
|
|
|
/* length output with special chars */
|
|
#define length_expect(in_chars, out_bytes, expect_status) \
|
|
utf8_expect_(NULL, 0, out_bytes, \
|
|
special_string, in_chars * sizeof(WCHAR), \
|
|
expect_status, __LINE__)
|
|
|
|
length_expect(0, 0, STATUS_SUCCESS);
|
|
length_expect(1, 1, STATUS_SUCCESS);
|
|
length_expect(2, 3, STATUS_SUCCESS);
|
|
length_expect(3, 6, STATUS_SOME_NOT_MAPPED);
|
|
length_expect(4, 7, STATUS_SOME_NOT_MAPPED);
|
|
#undef length_expect
|
|
|
|
/* output truncation */
|
|
#define truncate_expect(buflen, out_bytes, expect_status) \
|
|
utf8_expect_(special_expected, buflen, out_bytes, \
|
|
special_string, sizeof(special_string), \
|
|
expect_status, __LINE__)
|
|
|
|
truncate_expect(0, 0, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect(1, 1, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect(2, 1, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect(3, 3, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect(4, 3, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect(5, 3, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect(6, 6, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect(7, 7, STATUS_SOME_NOT_MAPPED);
|
|
#undef truncate_expect
|
|
|
|
/* conversion behavior with varying input length */
|
|
for (input_len = 0; input_len <= sizeof(test_string); input_len++) {
|
|
/* no output buffer, just length */
|
|
utf8_expect(NULL, 0, input_len / sizeof(WCHAR),
|
|
test_string, input_len, STATUS_SUCCESS);
|
|
|
|
/* write output */
|
|
bytes_out = 0x55555555;
|
|
memset(buffer, 0x55, sizeof(buffer));
|
|
status = pRtlUnicodeToUTF8N(
|
|
buffer, sizeof(buffer), &bytes_out,
|
|
test_string, input_len);
|
|
if (input_len % sizeof(WCHAR) == 0) {
|
|
ok(status == STATUS_SUCCESS,
|
|
"(len %u): status = 0x%x\n", input_len, status);
|
|
ok(bytes_out == input_len / sizeof(WCHAR),
|
|
"(len %u): bytes_out = 0x%x\n", input_len, bytes_out);
|
|
for (i = 0; i < bytes_out; i++) {
|
|
ok(buffer[i] == test_string[i],
|
|
"(len %u): buffer[%d] = 0x%x, expected 0x%x\n",
|
|
input_len, i, buffer[i], test_string[i]);
|
|
}
|
|
for (; i < sizeof(buffer); i++) {
|
|
ok(buffer[i] == 0x55,
|
|
"(len %u): buffer[%d] = 0x%x\n", input_len, i, buffer[i]);
|
|
}
|
|
} else {
|
|
ok(status == STATUS_INVALID_PARAMETER_5,
|
|
"(len %u): status = 0x%x\n", input_len, status);
|
|
ok(bytes_out == 0x55555555,
|
|
"(len %u): bytes_out = 0x%x\n", input_len, bytes_out);
|
|
for (i = 0; i < sizeof(buffer); i++) {
|
|
ok(buffer[i] == 0x55,
|
|
"(len %u): buffer[%d] = 0x%x\n", input_len, i, buffer[i]);
|
|
}
|
|
}
|
|
}
|
|
|
|
/* test cases for special characters */
|
|
for (i = 0; i < test_count; i++) {
|
|
bytes_out = 0x55555555;
|
|
memset(buffer, 0x55, sizeof(buffer));
|
|
status = pRtlUnicodeToUTF8N(
|
|
buffer, sizeof(buffer), &bytes_out,
|
|
unicode_to_utf8[i].unicode, lstrlenW(unicode_to_utf8[i].unicode) * sizeof(WCHAR));
|
|
ok(status == unicode_to_utf8[i].status,
|
|
"(test %d): status is 0x%x, expected 0x%x\n",
|
|
i, status, unicode_to_utf8[i].status);
|
|
ok(bytes_out == strlen(unicode_to_utf8[i].expected),
|
|
"(test %d): bytes_out is %u, expected %u\n",
|
|
i, bytes_out, lstrlenA(unicode_to_utf8[i].expected));
|
|
ok(!memcmp(buffer, unicode_to_utf8[i].expected, bytes_out),
|
|
"(test %d): got \"%.*s\", expected \"%s\"\n",
|
|
i, bytes_out, buffer, unicode_to_utf8[i].expected);
|
|
ok(buffer[bytes_out] == 0x55,
|
|
"(test %d): behind string: 0x%x\n", i, buffer[bytes_out]);
|
|
|
|
/* same test but include the null terminator */
|
|
bytes_out = 0x55555555;
|
|
memset(buffer, 0x55, sizeof(buffer));
|
|
status = pRtlUnicodeToUTF8N(
|
|
buffer, sizeof(buffer), &bytes_out,
|
|
unicode_to_utf8[i].unicode, (lstrlenW(unicode_to_utf8[i].unicode) + 1) * sizeof(WCHAR));
|
|
ok(status == unicode_to_utf8[i].status,
|
|
"(test %d): status is 0x%x, expected 0x%x\n",
|
|
i, status, unicode_to_utf8[i].status);
|
|
ok(bytes_out == strlen(unicode_to_utf8[i].expected) + 1,
|
|
"(test %d): bytes_out is %u, expected %u\n",
|
|
i, bytes_out, lstrlenA(unicode_to_utf8[i].expected) + 1);
|
|
ok(!memcmp(buffer, unicode_to_utf8[i].expected, bytes_out),
|
|
"(test %d): got \"%.*s\", expected \"%s\"\n",
|
|
i, bytes_out, buffer, unicode_to_utf8[i].expected);
|
|
ok(buffer[bytes_out] == 0x55,
|
|
"(test %d): behind string: 0x%x\n", i, buffer[bytes_out]);
|
|
}
|
|
}
|
|
|
|
struct utf8_to_unicode_test {
|
|
const char *utf8;
|
|
WCHAR expected[128];
|
|
NTSTATUS status;
|
|
};
|
|
|
|
static const struct utf8_to_unicode_test utf8_to_unicode[] = {
|
|
{ "", { 0 }, STATUS_SUCCESS },
|
|
{ "-", { '-',0 }, STATUS_SUCCESS },
|
|
{ "hello", { 'h','e','l','l','o',0 }, STATUS_SUCCESS },
|
|
/* first and last of each range */
|
|
{ "-\x7F-\xC2\x80-\xC3\xBF-\xC4\x80-", { '-',0x7f,'-',0x80,'-',0xff,'-',0x100,'-',0 }, STATUS_SUCCESS },
|
|
{ "-\xDF\xBF-\xE0\xA0\x80-", { '-',0x7ff,'-',0x800,'-',0 }, STATUS_SUCCESS },
|
|
{ "-\xED\x9F\xBF-\xEE\x80\x80-", { '-',0xd7ff,'-',0xe000,'-',0 }, STATUS_SUCCESS },
|
|
/* 0x10000 */
|
|
{ "-\xEF\xBF\xBF-\xF0\x90\x80\x80-", { '-',0xffff,'-',0xd800,0xdc00,'-',0 }, STATUS_SUCCESS },
|
|
/* 0x103ff */ /* 0x10400 */
|
|
{ "-\xF0\x90\x8F\xBF-\xF0\x90\x90\x80-", { '-',0xd800,0xdfff,'-',0xd801,0xdc00,'-',0 }, STATUS_SUCCESS },
|
|
/* 0x10ffff */
|
|
{ "-\xF4\x8F\xBF\xBF-", { '-',0xdbff,0xdfff,'-',0 }, STATUS_SUCCESS },
|
|
/* standalone surrogate code points */
|
|
/* 0xd800 */ /* 0xdbff */
|
|
{ "-\xED\xA0\x80-\xED\xAF\xBF-", { '-',0xfffd,0xfffd,'-',0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* 0xdc00 */ /* 0xdfff */
|
|
{ "-\xED\xB0\x80-\xED\xBF\xBF-", { '-',0xfffd,0xfffd,'-',0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* UTF-8 encoded surrogate pair */
|
|
/* 0xdbff *//* 0xdfff */
|
|
{ "-\xED\xAF\xBF\xED\xBF\xBF-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* reverse surrogate pair */
|
|
/* 0xdfff *//* 0xdbff */
|
|
{ "-\xED\xBF\xBF\xED\xAF\xBF-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* code points outside the UTF-16 range */
|
|
/* 0x110000 */
|
|
{ "-\xF4\x90\x80\x80-", { '-',0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* 0x1fffff */
|
|
{ "-\xF7\xBF\xBF\xBF-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* 0x200000 */
|
|
{ "-\xFA\x80\x80\x80\x80-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* 0x3ffffff */
|
|
{ "-\xFB\xBF\xBF\xBF\xBF-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* 0x4000000 */
|
|
{ "-\xFC\x84\x80\x80\x80\x80-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* 0x7fffffff */
|
|
{ "-\xFD\xBF\xBF\xBF\xBF\xBF-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* overlong encodings of each length for -, NUL, and the highest possible value */
|
|
{ "-\xC0\xAD-\xC0\x80-\xC1\xBF-", { '-',0xfffd,0xfffd,'-',0xfffd,0xfffd,'-',0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "-\xE0\x80\xAD-\xE0\x80\x80-\xE0\x9F\xBF-", { '-',0xfffd,0xfffd,'-',0xfffd,0xfffd,'-',0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "-\xF0\x80\x80\xAD-", { '-',0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "-\xF0\x80\x80\x80-", { '-',0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "-\xF0\x8F\xBF\xBF-", { '-',0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "-\xF8\x80\x80\x80\xAD-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "-\xF8\x80\x80\x80\x80-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "-\xF8\x87\xBF\xBF\xBF-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "-\xFC\x80\x80\x80\x80\xAD-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "-\xFC\x80\x80\x80\x80\x80-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "-\xFC\x83\xBF\xBF\xBF\xBF-", { '-',0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* invalid bytes */
|
|
{ "\xFE", { 0xfffd,0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\xFF", { 0xfffd,0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\xFE\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF", { 0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\xFF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF", { 0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\xFF\x80\x80\x80\x80\x80\x80\x80\x80", { 0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\xFF\x40\x80\x80\x80\x80\x80\x80\x80", { 0xfffd,0x40,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* lone continuation bytes */
|
|
{ "\x80", { 0xfffd,0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\x80\x80", { 0xfffd,0xfffd,0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\xBF", { 0xfffd,0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\xBF\xBF", { 0xfffd,0xfffd,0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* incomplete sequences */
|
|
{ "\xC2-", { 0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\xE0\xA0-", { 0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\xF0\x90\x80-", { 0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\xF4\x8F\xBF-", { 0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\xFA\x80\x80\x80-", { 0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
{ "\xFC\x84\x80\x80\x80-", { 0xfffd,0xfffd,0xfffd,0xfffd,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* multibyte sequence followed by lone continuation byte */
|
|
{ "\xE0\xA0\x80\x80-", { 0x800,0xfffd,'-',0 }, STATUS_SOME_NOT_MAPPED },
|
|
/* byte order marks */
|
|
{ "-\xEF\xBB\xBF-\xEF\xBF\xBE-", { '-',0xfeff,'-',0xfffe,'-',0 }, STATUS_SUCCESS },
|
|
{ "\xEF\xBB\xBF-", { 0xfeff,'-',0 }, STATUS_SUCCESS },
|
|
{ "\xEF\xBF\xBE-", { 0xfffe,'-',0 }, STATUS_SUCCESS },
|
|
/* invalid code point */
|
|
/* 0xffff */
|
|
{ "\xEF\xBF\xBF-", { 0xffff,'-',0 }, STATUS_SUCCESS },
|
|
/* canonically equivalent representations -- no normalization should happen */
|
|
{ "-\xE1\xB8\x89-", { '-',0x1e09,'-',0 }, STATUS_SUCCESS },
|
|
{ "-\xC4\x87\xCC\xA7-", { '-',0x0107,0x0327,'-',0 }, STATUS_SUCCESS },
|
|
{ "-\xC3\xA7\xCC\x81-", { '-',0x00e7,0x0301,'-',0 }, STATUS_SUCCESS },
|
|
{ "-\x63\xCC\xA7\xCC\x81-", { '-',0x0063,0x0327,0x0301,'-',0 }, STATUS_SUCCESS },
|
|
{ "-\x63\xCC\x81\xCC\xA7-", { '-',0x0063,0x0301,0x0327,'-',0 }, STATUS_SUCCESS },
|
|
};
|
|
|
|
static void unicode_expect_(const WCHAR *out_string, ULONG buflen, ULONG out_chars,
|
|
const char *in_string, ULONG in_chars,
|
|
NTSTATUS expect_status, int line)
|
|
{
|
|
NTSTATUS status;
|
|
ULONG bytes_out;
|
|
WCHAR buffer[128];
|
|
unsigned int i;
|
|
|
|
if (buflen == (ULONG)-1)
|
|
buflen = sizeof(buffer);
|
|
bytes_out = 0x55555555;
|
|
memset(buffer, 0x55, sizeof(buffer));
|
|
status = pRtlUTF8ToUnicodeN(
|
|
out_string ? buffer : NULL, buflen, &bytes_out,
|
|
in_string, in_chars);
|
|
ok_(__FILE__, line)(status == expect_status, "status = 0x%x\n", status);
|
|
ok_(__FILE__, line)(bytes_out == out_chars * sizeof(WCHAR),
|
|
"bytes_out = %u, expected %u\n", bytes_out, out_chars * (ULONG)sizeof(WCHAR));
|
|
if (out_string)
|
|
{
|
|
for (i = 0; i < bytes_out / sizeof(WCHAR); i++)
|
|
ok_(__FILE__, line)(buffer[i] == out_string[i],
|
|
"buffer[%d] = 0x%x, expected 0x%x\n",
|
|
i, buffer[i], out_string[i]);
|
|
for (; i < sizeof(buffer) / sizeof(WCHAR); i++)
|
|
ok_(__FILE__, line)(buffer[i] == 0x5555,
|
|
"buffer[%d] = 0x%x, expected 0x5555\n",
|
|
i, buffer[i]);
|
|
}
|
|
}
|
|
#define unicode_expect(out_string, buflen, out_chars, in_string, in_chars, expect_status) \
|
|
unicode_expect_(out_string, buflen, out_chars, in_string, in_chars, expect_status, __LINE__)
|
|
|
|
static void test_RtlUTF8ToUnicodeN(void)
|
|
{
|
|
NTSTATUS status;
|
|
ULONG bytes_out;
|
|
ULONG bytes_out_array[2];
|
|
void * const invalid_pointer = (void *)0x8;
|
|
WCHAR buffer[128];
|
|
const char empty_string[] = "";
|
|
const char test_string[] = "A\0abcdefg";
|
|
const WCHAR test_stringW[] = {'A',0,'a','b','c','d','e','f','g',0 };
|
|
const char special_string[] = { 'X',0xc2,0x80,0xF0,0x90,0x80,0x80,0 };
|
|
const WCHAR special_expected[] = { 'X',0x80,0xd800,0xdc00,0 };
|
|
unsigned int input_len;
|
|
const unsigned int test_count = sizeof(utf8_to_unicode) / sizeof(utf8_to_unicode[0]);
|
|
unsigned int i;
|
|
|
|
if (!pRtlUTF8ToUnicodeN)
|
|
{
|
|
skip("RtlUTF8ToUnicodeN unavailable\n");
|
|
return;
|
|
}
|
|
|
|
/* show that bytes_out is really ULONG */
|
|
memset(bytes_out_array, 0x55, sizeof(bytes_out_array));
|
|
status = pRtlUTF8ToUnicodeN(NULL, 0, bytes_out_array, empty_string, 0);
|
|
ok(status == STATUS_SUCCESS, "status = 0x%x\n", status);
|
|
ok(bytes_out_array[0] == 0x00000000, "Got 0x%x\n", bytes_out_array[0]);
|
|
ok(bytes_out_array[1] == 0x55555555, "Got 0x%x\n", bytes_out_array[1]);
|
|
|
|
/* parameter checks */
|
|
status = pRtlUTF8ToUnicodeN(NULL, 0, NULL, NULL, 0);
|
|
ok(status == STATUS_INVALID_PARAMETER_4, "status = 0x%x\n", status);
|
|
|
|
status = pRtlUTF8ToUnicodeN(NULL, 0, NULL, empty_string, 0);
|
|
ok(status == STATUS_INVALID_PARAMETER, "status = 0x%x\n", status);
|
|
|
|
bytes_out = 0x55555555;
|
|
status = pRtlUTF8ToUnicodeN(NULL, 0, &bytes_out, NULL, 0);
|
|
ok(status == STATUS_INVALID_PARAMETER_4, "status = 0x%x\n", status);
|
|
ok(bytes_out == 0x55555555, "bytes_out = 0x%x\n", bytes_out);
|
|
|
|
bytes_out = 0x55555555;
|
|
status = pRtlUTF8ToUnicodeN(NULL, 0, &bytes_out, invalid_pointer, 0);
|
|
ok(status == STATUS_SUCCESS, "status = 0x%x\n", status);
|
|
ok(bytes_out == 0, "bytes_out = 0x%x\n", bytes_out);
|
|
|
|
bytes_out = 0x55555555;
|
|
status = pRtlUTF8ToUnicodeN(NULL, 0, &bytes_out, empty_string, 0);
|
|
ok(status == STATUS_SUCCESS, "status = 0x%x\n", status);
|
|
ok(bytes_out == 0, "bytes_out = 0x%x\n", bytes_out);
|
|
|
|
bytes_out = 0x55555555;
|
|
status = pRtlUTF8ToUnicodeN(NULL, 0, &bytes_out, test_string, 0);
|
|
ok(status == STATUS_SUCCESS, "status = 0x%x\n", status);
|
|
ok(bytes_out == 0, "bytes_out = 0x%x\n", bytes_out);
|
|
|
|
bytes_out = 0x55555555;
|
|
status = pRtlUTF8ToUnicodeN(NULL, 0, &bytes_out, empty_string, 1);
|
|
ok(status == STATUS_SUCCESS, "status = 0x%x\n", status);
|
|
ok(bytes_out == sizeof(WCHAR), "bytes_out = 0x%x\n", bytes_out);
|
|
|
|
/* length output with special chars */
|
|
#define length_expect(in_chars, out_chars, expect_status) \
|
|
unicode_expect_(NULL, 0, out_chars, special_string, in_chars, \
|
|
expect_status, __LINE__)
|
|
|
|
length_expect(0, 0, STATUS_SUCCESS);
|
|
length_expect(1, 1, STATUS_SUCCESS);
|
|
length_expect(2, 2, STATUS_SOME_NOT_MAPPED);
|
|
length_expect(3, 2, STATUS_SUCCESS);
|
|
length_expect(4, 3, STATUS_SOME_NOT_MAPPED);
|
|
length_expect(5, 3, STATUS_SOME_NOT_MAPPED);
|
|
length_expect(6, 3, STATUS_SOME_NOT_MAPPED);
|
|
length_expect(7, 4, STATUS_SUCCESS);
|
|
length_expect(8, 5, STATUS_SUCCESS);
|
|
#undef length_expect
|
|
|
|
/* output truncation */
|
|
#define truncate_expect(buflen, out_chars, expect_status) \
|
|
unicode_expect_(special_expected, buflen, out_chars, \
|
|
special_string, sizeof(special_string), \
|
|
expect_status, __LINE__)
|
|
|
|
truncate_expect( 0, 0, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect( 1, 0, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect( 2, 1, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect( 3, 1, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect( 4, 2, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect( 5, 2, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect( 6, 3, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect( 7, 3, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect( 8, 4, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect( 9, 4, STATUS_BUFFER_TOO_SMALL);
|
|
truncate_expect(10, 5, STATUS_SUCCESS);
|
|
#undef truncate_expect
|
|
|
|
/* conversion behavior with varying input length */
|
|
for (input_len = 0; input_len <= sizeof(test_string); input_len++) {
|
|
/* no output buffer, just length */
|
|
unicode_expect(NULL, 0, input_len,
|
|
test_string, input_len, STATUS_SUCCESS);
|
|
|
|
/* write output */
|
|
unicode_expect(test_stringW, -1, input_len,
|
|
test_string, input_len, STATUS_SUCCESS);
|
|
}
|
|
|
|
/* test cases for special characters */
|
|
for (i = 0; i < test_count; i++) {
|
|
bytes_out = 0x55555555;
|
|
memset(buffer, 0x55, sizeof(buffer));
|
|
status = pRtlUTF8ToUnicodeN(
|
|
buffer, sizeof(buffer), &bytes_out,
|
|
utf8_to_unicode[i].utf8, strlen(utf8_to_unicode[i].utf8));
|
|
ok(status == utf8_to_unicode[i].status,
|
|
"(test %d): status is 0x%x, expected 0x%x\n",
|
|
i, status, utf8_to_unicode[i].status);
|
|
ok(bytes_out == lstrlenW(utf8_to_unicode[i].expected) * sizeof(WCHAR),
|
|
"(test %d): bytes_out is %u, expected %u\n",
|
|
i, bytes_out, lstrlenW(utf8_to_unicode[i].expected) * (ULONG)sizeof(WCHAR));
|
|
ok(!memcmp(buffer, utf8_to_unicode[i].expected, bytes_out),
|
|
"(test %d): got %s, expected %s\n",
|
|
i, wine_dbgstr_wn(buffer, bytes_out / sizeof(WCHAR)), wine_dbgstr_w(utf8_to_unicode[i].expected));
|
|
ok(buffer[bytes_out] == 0x5555,
|
|
"(test %d): behind string: 0x%x\n", i, buffer[bytes_out]);
|
|
|
|
/* same test but include the null terminator */
|
|
bytes_out = 0x55555555;
|
|
memset(buffer, 0x55, sizeof(buffer));
|
|
status = pRtlUTF8ToUnicodeN(
|
|
buffer, sizeof(buffer), &bytes_out,
|
|
utf8_to_unicode[i].utf8, strlen(utf8_to_unicode[i].utf8) + 1);
|
|
ok(status == utf8_to_unicode[i].status,
|
|
"(test %d): status is 0x%x, expected 0x%x\n",
|
|
i, status, utf8_to_unicode[i].status);
|
|
ok(bytes_out == (lstrlenW(utf8_to_unicode[i].expected) + 1) * sizeof(WCHAR),
|
|
"(test %d): bytes_out is %u, expected %u\n",
|
|
i, bytes_out, (lstrlenW(utf8_to_unicode[i].expected) + 1) * (ULONG)sizeof(WCHAR));
|
|
ok(!memcmp(buffer, utf8_to_unicode[i].expected, bytes_out),
|
|
"(test %d): got %s, expected %s\n",
|
|
i, wine_dbgstr_wn(buffer, bytes_out / sizeof(WCHAR)), wine_dbgstr_w(utf8_to_unicode[i].expected));
|
|
ok(buffer[bytes_out] == 0x5555,
|
|
"(test %d): behind string: 0x%x\n", i, buffer[bytes_out]);
|
|
}
|
|
}
|
|
|
|
START_TEST(rtlstr)
|
|
{
|
|
InitFunctionPtrs();
|
|
if (pRtlInitAnsiString) {
|
|
test_RtlInitString();
|
|
test_RtlInitUnicodeString();
|
|
test_RtlCopyString();
|
|
test_RtlUnicodeStringToInteger();
|
|
test_RtlCharToInteger();
|
|
test_RtlIntegerToUnicodeString();
|
|
test_RtlIntegerToChar();
|
|
test_RtlUpperChar();
|
|
test_RtlUpperString();
|
|
test_RtlUnicodeStringToAnsiString();
|
|
test_RtlAppendAsciizToString();
|
|
test_RtlAppendStringToString();
|
|
test_RtlAppendUnicodeToString();
|
|
test_RtlAppendUnicodeStringToString();
|
|
}
|
|
|
|
test_RtlInitUnicodeStringEx();
|
|
test_RtlDuplicateUnicodeString();
|
|
test_RtlFindCharInUnicodeString();
|
|
test_RtlGUIDFromString();
|
|
test_RtlStringFromGUID();
|
|
test_RtlIsTextUnicode();
|
|
test_RtlCompareUnicodeString();
|
|
if(0)
|
|
{
|
|
test_RtlUpcaseUnicodeChar();
|
|
test_RtlUpcaseUnicodeString();
|
|
test_RtlDowncaseUnicodeString();
|
|
}
|
|
test_RtlHashUnicodeString();
|
|
test_RtlUnicodeToUTF8N();
|
|
test_RtlUTF8ToUnicodeN();
|
|
}
|