fopen file name with UTF8 string in windows - c++

Question

When I used opencv's API cvLoadImage(const char *filename, int iscolor) It accepts const char * as file name. When the file name is not ASCII-character, I tried to convert it to UTF8 string. It fails because fopen() called in cvLoadImage() can not interpret the characters of the file name literally as ASCII string. I may used _wfopen() if tried to open file names, but if fopen() is called in the third-party library, is there any method to handle this problem?
Thank you.

Answer 1

Use GetShortPathName. It will return an old (8.3) name for the file, which you should be able to convert to char*, as it should not contain any non ASCII characters.
I've just tested it with some language specific characters and it worked as I described. I've successfully opened a file from C:\łęłęł\ąóąóą.tsttgbb using fopen.

Answer 2

setlocale(LC_ALL, ".65001");
fopen(u8"中文路径.txt", "rb"); //window7(中文) vs2017 ok

Answer 3

A quick search came up with nothing but people saying it can't be done. If you can't change cvLoadImage (which is reasonable, you don't want to mess with that), you can try to trick it.
You can create a link to the file, using the CreateSymbolicLink. I'm not sure it'll work, though, because the MKLINK command line utility requires administrative privileges.
If you can't create a symbolic link, you can always copy the file to a different location with an ASCII-only name.
If you really don't want to copy the file and symlinks don't work, you can create a file-proxy - created a named pipe with an ASCII only name, and translate each read from the pipe to a read from the file.
I would go with options 1 or 2, though - a lot simpler.

Answer 4

Here's a late contribution to this problem. I looked through the source of the runtime library (which Microsoft kindly supply) and found that I could replace the routine used by fopen to map an ANSI string with the following code (just link this into your exe and it will replace the routine in the runtime library).
The version listed works for Visual Studio 2017 using the v141_xp toolkit. I haven't tested it for other versions but I imagine some minor changes (such as the name of the routine itself) might be needed. It won't work of course if the offending library is a DLL. Make of it what you will.
#ifdef _DEBUG
#define _NORMAL_BLOCK 1
#define _CRT_BLOCK 2
#define _malloc_crt(s) (_malloc_dbg (s, _CRT_BLOCK, __FILE__, __LINE__))
#else
#define _malloc_crt _malloc_base
#endif
// A hack to make fopen et al accept UTF8 strings (as at Visual Studio 2017), see:
// D:\Program Files (x86)\Windows Kits\10\Source\10.0.10240.0\ucrt\internal\string_utilities.cpp
// D:\Program Files (x86)\Windows Kits\10\Source\10.0.10240.0\ucrt\inc\corecrt_internal_traits.h
extern "C" BOOL __cdecl __acrt_copy_path_to_wide_string (char const* const path, wchar_t** const result)
{
#if _MSC_VER != 1910
#define STRINGIZE_HELPER(x) #x
#define STRINGIZE(x) STRINGIZE_HELPER(x)
__pragma (message (__FILE__ "(" STRINGIZE (__LINE__) ") : Error: Code not tested for this version of Visual Studio"));
#endif
assert (path);
assert (result);
// Compute the required size of the wide character buffer:
int length = MultiByteToWideChar (CP_UTF8, 0, path, -1, nullptr, 0);
assert (length > 0);
*result = (wchar_t *) _malloc_crt (T2B (length));
// Do the conversion:
length = MultiByteToWideChar (CP_UTF8, 0, path, -1, *result, length);
assert (length);
return TRUE;
}