개인공부,기록

Many C++ Windows programmers get confused over what bizarre identifiers like TCHAR, LPCTSTR are. In this article, I would attempt by best to clear out the fog.

In general, a character can be represented in 1 byte or 2 bytes. Let's say 1-byte character is ANSI character - all English characters are represented through this encoding. And let's say a 2-byte character is Unicode, which can represent ALL languages in the world.

The Visual C++ compiler supports char and wchar_t as native data-types for ANSI and Unicode characters, respectively. Though there is more concrete definition of Unicode, but for understanding assume it as two-byte character which Windows OS uses for multiple language support.

What if you want your C/C++ code to be independent of character encoding/mode used?

Suggestion: Use generic data-types and names to represent characters and string.

For example, instead of replacing:

char cResponse; // 'Y' or 'N'
char sUsername[64];
// str* functions

with

wchar_t cResponse; // 'Y' or 'N'
wchar_t sUsername[64];
// wcs* functions

In order to support multi-lingual (i.e., Unicode) in your language, you can simply code it in more generic manner:

#include<TCHAR.H> // Implicit or explicit include
TCHAR cResponse; // 'Y' or 'N'
TCHAR sUsername[64];
// _tcs* functions

The following project setting in General page describes which Character Set is to be used for compilation: (General -> Character Set)

This way, when your project is being compiled as Unicode, the TCHAR would translate to wchar_t. If it is being compiled as ANSI/MBCS, it would be translated to char. You are free to use char and wchar_t, and project settings will not affect any direct use of these keywords.

TCHAR is defined as:

#ifdef _UNICODE
typedef wchar_t TCHAR;
#else
typedef char TCHAR;
#endif

The macro _UNICODE is defined when you set Character Set to "Use Unicode Character Set", and therefore TCHAR would mean wchar_t. When Character Set if set to "Use Multi-Byte Character Set", TCHAR would mean char.

Likewise, to support multiple character-set using single code base, and possibly supporting multi-language, use specific functions (macros). Instead of using strcpy, strlen, strcat (including the secure versions suffixed with _s); or wcscpy, wcslen, wcscat (including secure), you should better use use _tcscpy, _tcslen, _tcscat functions.

As you know strlen is prototyped as:

size_t strlen(const char*);

And, wcslen is prototyped as:

size_t wcslen(const wchar_t* );

You may better use _tcslen, which is logically prototyped as:

size_t _tcslen(const TCHAR* );

WC is for Wide Character. Therefore, wcs turns to be wide-character-string. This way, _tcs would mean _T Character String. And you know _T may be char or what_t, logically.

But, in reality, _tcslen (and other _tcs functions) are actually not functions, but macros. They are defined simply as:

#ifdef _UNICODE
#define _tcslen wcslen 
#else
#define _tcslen strlen
#endif

You should refer TCHAR.H to lookup more macro definitions like this.

You might ask why they are defined as macros, and not implemented as functions instead? The reason is simple: A library or DLL may export a single function, with same name and prototype (Ignore overloading concept of C++). For instance, when you export a function as:

void _TPrintChar(char);

How the client is supposed to call it as?

void _TPrintChar(wchar_t);

_TPrintChar cannot be magically converted into function taking 2-byte character. There has to be two separate functions:

void PrintCharA(char); // A = ANSI 
void PrintCharW(wchar_t); // W = Wide character

And a simple macro, as defined below, would hide the difference:

#ifdef _UNICODE
void _TPrintChar(wchar_t); 
#else 
void _TPrintChar(char);
#endif

The client would simply call it as:

TCHAR cChar;
_TPrintChar(cChar);

Note that both TCHAR and _TPrintChar would map to either Unicode or ANSI, and therefore cChar and the argument to function would be either char or wchar_t.

Macros do avoid these complications, and allows us to use either ANSI or Unicode function for characters and strings. Most of the Windows functions, that take string or a character are implemented this way, and for programmers convenience, only one function (a macro!) is good. SetWindowText is one example:

// WinUser.H
#ifdef UNICODE
#define SetWindowText  SetWindowTextW
#else
#define SetWindowText  SetWindowTextA
#endif // !UNICODE

There are very few functions that do not have macros, and are available only with suffixed W or A. One example is ReadDirectoryChangesW, which doesn't have ANSI equivalent.

You all know that we use double quotation marks to represent strings. The string represented in this manner is ANSI-string, having 1-byte each character. Example:

"This is ANSI String. Each letter takes 1 byte."

The string text given above is not Unicode, and would be quantifiable for multi-language support. To represent Unicode string, you need to use prefix L. An example:

L"This is Unicode string. Each letter would take 2 bytes, including spaces."

Note the L at the beginning of string, which makes it a Unicode string. All characters (I repeat all characters) would take two bytes, including all English letters, spaces, digits, and the null character. Therefore, length of Unicode string would always be in multiple of 2-bytes. A Unicode string of length 7 characters would need 14 bytes, and so on. Unicode string taking 15 bytes, for example, would not be valid in any context.

In general, string would be in multiple of sizeof(TCHAR) bytes!

When you need to express hard-coded string, you can use:

"ANSI String"; // ANSI
L"Unicode String"; // Unicode

_T("Either string, depending on compilation"); // ANSI or Unicode
// or use TEXT macro, if you need more readability

The non-prefixed string is ANSI string, the L prefixed string is Unicode, and string specified in _T or TEXT would be either, depending on compilation. Again, _T and TEXT are nothing but macros, and are defined as:

// SIMPLIFIED
#ifdef _UNICODE 
 #define _T(c) L##c
 #define TEXT(c) L##c
#else 
 #define _T(c) c
 #define TEXT(c) c
#endif

The ## symbol is token pasting operator, which would turn _T("Unicode") into L"Unicode", where the string passed is argument to macro - If _UNICODE is defined. If _UNICODE is not defined, _T("Unicode") would simply mean "Unicode". The token pasting operator did exist even in C language, and is not specific about VC++ or character encoding.

Note that these macros can be used for strings as well as characters. _T('R') would turn into L'R' or simple 'R' - former is Unicode character, latter is ANSI character.

No, you cannot use these macros to convert variables (string or character) into Unicode/non-Unicode text. Following is not valid:

char c = 'C';
char str[16] = "CodeProject";

_T(c);
_T(str);

The bold lines would get successfully compiled in ANSI (Multi-Byte) build, since _T(x) would simply be x, and therefore _T(c) and _T(str) would come out to be c and str, respectively. But, when you build it with Unicode character set, it would fail to compile:

error C2065: 'Lc' : undeclared identifier
error C2065: 'Lstr' : undeclared identifier

I would not like to insult your intelligence by describing why and what those errors are.

There exist set of conversion routine to convert MBCS to Unicode and vice versa, which I would explain soon.

It is important to note that almost all functions that take string (or character), primarily in Windows API, would have generalized prototype in MSDN and elsewhere. The function SetWindowTextA/W, for instance, be classified as:

BOOL SetWindowText(HWND, const TCHAR*);

But, as you know, SetWindowText is just a macro, and depending on your build settings, it would mean either of following:

BOOL SetWindowTextA(HWND, const char*);
BOOL SetWindowTextW(HWND, const wchar_t*);

Therefore, don't be puzzled if following call fails to get address of this function!

HMODULE hDLLHandle;
FARPROC pFuncPtr;

hDLLHandle = LoadLibrary(L"user32.dll");

pFuncPtr = GetProcAddress(hDLLHandle, "SetWindowText");
//pFuncPtr will be null, since there doesn't exist any function with name SetWindowText !

From User32.DLL, the two functions SetWindowTextA and SetWindowTextW are exported, not the function with generalized name.

Interestingly, .NET Framework is smart enough to locate function from DLL with generalized name:

[DllImport("user32.dll")]
extern public static int SetWindowText(IntPtr hWnd, string lpString);

No rocket science, just bunch of ifs and else around GetProcAddress!

All of the functions that have ANSI and Unicode versions, would have actual implementation only in Unicode version. That means, when you call SetWindowTextA from your code, passing an ANSI string - it would convert the ANSI string to Unicode text and then would call SetWindowTextW. The actual work (setting the window text/title/caption) will be performed by Unicode version only!

Take another example, which would retrieve the window text, using GetWindowText. You call GetWindowTextA, passing ANSI buffer as target buffer. GetWindowTextA would first call GetWindowTextW, probably allocating a Unicode string (a wchar_t array) for it. Then it would convert that Unicode stuff, for you, into ANSI string.

This ANSI to Unicode and vice-versa conversion is not limited to GUI functions, but entire set of Windows API, which do take strings and have two variants. Few examples could be:

• CreateProcess
• GetUserName
• OpenDesktop
• DeleteFile
• etc

It is therefore very much recommended to call the Unicode version directly. In turn, it means you should always target for Unicode builds, and not ANSI builds - just because you are accustomed to using ANSI string for years. Yes, you may save and retrieve ANSI strings, for example in file, or send as chat message in your messenger application. The conversion routines do exist for such needs.

Note: There exists another typedef: WCHAR, which is equivalent to wchar_t.

The TCHAR macro is for a single character. You can definitely declare an array of TCHAR. What if you would like to express a character-pointer, or a const-character-pointer - Which one of the following?

// ANSI characters 
foo_ansi(char*); 
foo_ansi(const char*); 
/*const*/ char* pString; 

// Unicode/wide-string 
foo_uni(WCHAR*); 
wchar_t* foo_uni(const WCHAR*); 
/*const*/ WCHAR* pString; 

// Independent 
foo_char(TCHAR*); 
foo_char(const TCHAR*); 
/*const*/ TCHAR* pString;

After reading about TCHAR stuff, you would definitely select the last one as your choice. There are better alternatives available to represent strings. For that, you just need to include Windows.h. Note: If your project implicitly or explicitly includes Windows.h, you need not include TCHAR.H

First, revisit old string functions for better understanding. You know strlen:

size_t strlen(const char*);

Which may be represented as:

size_t strlen(LPCSTR);

Where symbol LPCSTR is typedef'ed as:

// Simplified
typedef const char* LPCSTR;  

The meaning goes like:

• LP - Long Pointer
• C - Constant
• STR - String

Essentially, LPCSTR would mean (Long) Pointer to a Constant String.

Let's represent strcpy using new style type-names:

LPSTR strcpy(LPSTR szTarget, LPCSTR szSource);

The type of szTarget is LPSTR, without C in the type-name. It is defined as:

typedef char* LPSTR;

Note that the szSource is LPCSTR, since strcpy function will not modify the source buffer, hence the const attribute. The return type is non-constant-string: LPSTR.

Alright, these str-functions are for ANSI string manipulation. But we want routines for 2-byte Unicode strings. For the same, the equivalent wide-character str-functions are provided. For example, to calculate length of wide-character (Unicode string), you would use wcslen:

size_t nLength;
nLength = wcslen(L"Unicode");

The prototype of wcslen is:

size_t wcslen(const wchar_t* szString); // Or WCHAR*

And that can be represented as:

size_t wcslen(LPCWSTR szString);

Where the symbol LPCWSTR is defined as:

typedef const WCHAR* LPCWSTR;
// const wchar_t*

Which can be broken down as:

• LP - Pointer
• C - Constant
• WSTR - Wide character String

Similarly, strcpy equivalent is wcscpy, for Unicode strings:

wchar_t* wcscpy(wchar_t* szTarget, const wchar_t* szSource)

Which can be represented as:

LPWSTR wcscpy(LPWSTR szTarget, LPWCSTR szSource);

Where the target is non-constant wide-string (LPWSTR), and source is constant-wide-string.

There exist set of equivalent wcs-functions for str-functions. The str-functions would be used for plain ANSI strings, and wcs-functions would be used for Unicode strings.

Though, I already advised to use Unicode native functions, instead of ANSI-only or TCHAR-synthesized functions. The reason was simple - your application must only be Unicode, and you should not even care about code portability for ANSI builds. But for the sake of completeness, I am mentioning these generic mappings.

To calculate length of string, you may use _tcslen function (a macro). In general, it is prototyped as:

size_t _tcslen(const TCHAR* szString); 

Or, as:

size_t _tcslen(LPCTSTR szString);

Where the type-name LPCTSTR can be classified as:

• LP - Pointer
• C - Constant
• T = TCHAR
• STR = String

Depending on the project settings, LPCTSTR would be mapped to either LPCSTR (ANSI) or LPCWSTR (Unicode).

Note: strlen, wcslen or _tcslen will return number of characters in string, not the number of bytes.

The generalized string-copy routine _tcscpy is defined as:

size_t _tcscpy(TCHAR* pTarget, const TCHAR* pSource);

Or, in more generalized form, as:

size_t _tcscpy(LPTSTR pTarget, LPCTSTR pSource);

You can deduce the meaning of LPTSTR!

Usage Examples

First, a broken code:

int main()
{
    TCHAR name[] = "Saturn";
    int nLen; // Or size_t

    lLen = strlen(name);
}

On ANSI build, this code will successfully compile since TCHAR would be char, and hence name would be an array of char. Calling strlen against name variable would also work flawlessly.

Alright. Let's compile the same with with UNICODE/_UNICODE defined (i.e. "Use Unicode Character Set" in project settings). Now, the compiler would report set of errors:

• error C2440: 'initializing' : cannot convert from 'const char [7]' to 'TCHAR []'
• error C2664: 'strlen' : cannot convert parameter 1 from 'TCHAR []' to 'const char *'

And the programmers would start committing mistakes by correcting it this way (first error):

TCHAR name[] = (TCHAR*)"Saturn";

Which will not pacify the compiler, since the conversion is not possible from TCHAR* to TCHAR[7]. The same error would also come when native ANSI string is passed to a Unicode function:

nLen = wcslen("Saturn");
// ERROR: cannot convert parameter 1 from 'const char [7]' to 'const wchar_t *'

Unfortunately (or fortunately), this error can be incorrectly corrected by simple C-style typecast:

nLen = wcslen((const wchar_t*)"Saturn");

And you'd think you've attained one more experience level in pointers! You are wrong - the code would give incorrect result, and in most cases would simply cause Access Violation. Typecasting this way is like passing a float variable where a structure of 80 bytes is expected (logically).

The string "Saturn" is sequence of 7 bytes:

But when you pass same set of bytes to wcslen, it treats each 2-byte as a single character. Therefore first two bytes [97, 83] would be treated as one character having value: 24915 (97<<8 | 83). It is Unicode character: ?. And the next character is represented by [117, 116] and so on.

For sure, you didn't pass those set of Chinese characters, but improper typecasting has done it! Therefore it is very essential to know that type-casting will not work! So, for the first line of initialization, you must do:

TCHAR name[] = _T("Saturn");

Which would translate to 7-bytes or 14-bytes, depending on compilation. The call to wcslen should be:

wcslen(L"Saturn");

In the sample program code given above, I used strlen, which causes error when building in Unicode. The non-working solution is C-sytle typecast:

lLen = strlen ((const char*)name);

On Unicode build, name would be of 14-bytes (7 Unicode characters, including null). Since string "Saturn" contains only English letters, which can be represented using original ASCII, the Unicode letter 'S' would be represented as [83, 0]. Other ASCII characters would be represented with a zero next to them. Note that 'S' is now represented as 2-byte value 83. The end of string would be represented by two bytes having value 0.

So, when you pass such string to strlen, the first character (i.e. first byte) would be correct ('S' in case of "Saturn"). But the second character/byte would indicate end of string. Therefore, strlen would return incorrect value 1 as the length of string.

As you know, Unicode string may contain non-English characters, the result of strlen would be more undefined.

In short, typecasting will not work. You either need to represent strings in correct form itself, or use ANSI to Unicode, and vice-versa, routines for conversions.

(There is more to add from this location, stay tuned!)

Now, I hope you understand the following signatures:

BOOL SetCurrentDirectory( LPCTSTR lpPathName );
DWORD GetCurrentDirectory(DWORD nBufferLength,LPTSTR lpBuffer);

Continuing. You must have seen some functions/methods asking you to pass number of characters, or returning the number of characters. Well, like GetCurrentDirectory, you need to pass number of characters, and not number of bytes. For example:

TCHAR sCurrentDir[255];
 
// Pass 255 and not 255*2 
GetCurrentDirectory(sCurrentDir, 255);

On the other side, if you need to allocate number or characters, you must allocate proper number of bytes. In C++, you can simply use new:

LPTSTR pBuffer; // TCHAR* 

pBuffer = new TCHAR[128]; // Allocates 128 or 256 BYTES, depending on compilation.

But if you use memory allocation functions like malloc, LocalAlloc, GlobalAlloc, etc; you must specify the number of bytes!

pBuffer = (TCHAR*) malloc (128 * sizeof(TCHAR) );

Typecasting the return value is required, as you know. The expression in malloc's argument ensures that it allocates desired number of bytes - and makes up room for desired number of characters.

결론 : 영어는 너무 어렵다..

가아니고. TCHAR을 이용하면 유니코드,멀티바이트 상관없이 쓸수 있다는것. 많은 설명을 봤지만, 요기나온 개념이 가장 이해하기 쉬웠던것 같습니다.

영어로된 설명을 많이 봐버릇 해야겠습니다. 외국사이트에 좋은게 많네요