I'm reediting the answer in the hope to make it clearer. First of all I'm assuming you are familiar with this: http://www.joelonsoftware.com/articles/Unicode.html. It is required background knowledge when dealing with character encoding.
Now I'm starting with a simple test program I typed on my linux machine test.c
#include <stdio.h>
#include <string.h>
#include <wchar.h>
#define BUF_SZ 255
void test_fwrite_universal(const char *fname)
{
printf("test_fwrite_universal on %s\n", fname);
printf("In memory we have %d bytes: ", strlen(fname));
for (unsigned i=0; i<strlen(fname); ++i) {
printf("%x ", (unsigned char)fname[i]);
}
printf("\n");
FILE* file = fopen(fname, "w");
if (file) {
fwrite((const void*)fname, 1, strlen(fname), file);
fclose(file);
file = NULL;
printf("Wrote to file successfully\n");
}
}
int main()
{
test_fwrite_universal("file_\u00e5.txt");
test_fwrite_universal("file_å.txt");
test_fwrite_universal("file_\u0436.txt");
return 0;
}
the text file is encoded as UTF-8. On my linux machine my locale is en_US.UTF-8 So I compile and run the program like this:
gcc -std=c99 test.c -fexec-charset=UTF-8 -o test
test
test_fwrite_universal on file_å.txt
In memory we have 11 bytes: 66 69 6c 65 5f c3 a5 2e 74 78 74
Wrote to file successfully
test_fwrite_universal on file_å.txt
In memory we have 11 bytes: 66 69 6c 65 5f c3 a5 2e 74 78 74
Wrote to file successfully
test_fwrite_universal on file_ж.txt
In memory we have 11 bytes: 66 69 6c 65 5f d0 b6 2e 74 78 74
Wrote to file successfully
The text file is in UTF-8, my locale is working of of UTF-8 and the execution character set for char is UTF-8. In main I call the function fwrite 3 times with character strings. The function prints the strings byte by byte. Then writes a file with that name and write that string into the file.
We can see that "file_\u00e5.txt" and "file_å.txt" are the same: 66 69 6c 65 5f c3 a5 2e 74 78 74 and sure enough (http://www.fileformat.info/info/unicode/char/e5/index.htm) the UTF-8 representation for code point +00E5 is: c3 a5 In the last example I used \u0436 which is a Russian character ж (UTF-8 d0 b6)
Now lets try the same on my windows machine. Here I use mingw and I execute the same code:
C:\test>gcc -std=c99 test.c -fexec-charset=UTF-8 -o test.exe
C:\test>test
test_fwrite_universal on file_å.txt
In memory we have 11 bytes: 66 69 6c 65 5f c3 a5 2e 74 78 74
Wrote to file successfully
test_fwrite_universal on file_å.txt
In memory we have 11 bytes: 66 69 6c 65 5f c3 a5 2e 74 78 74
Wrote to file successfully
test_fwrite_universal on file_╨╢.txt
In memory we have 11 bytes: 66 69 6c 65 5f d0 b6 2e 74 78 74
Wrote to file successfully
So it looks like something went horribly wrong printf is not writing the characters properly and the files on the disk also look wrong. Two things worth noting: in terms of byte values the file name is the same in both linux and windows. The content of the file is also correct when opened with something like notepad++
The reason for the problem is the C Standard library on windows and the locale. Where on linux the system locale is UTF-8 on windows my default locale is CP-437. And when I call functions such as printf
fopen
it assumes the input is in CP-437 and there c3 a5 are actually two characters.
Before we look at a proper windows solution lets try to explain why you have different results in file_å.txt
vs file_\u00e5.txt
.
I believe the key is the encoding of your text file. If I write the same test.c
in CP-437:
C:\test>iconv -f UTF-8 -t cp437 test.c > test_lcl.c
C:\test>gcc -std=c99 test_lcl.c -fexec-charset=UTF-8 -o test_lcl.exe
C:\test>test_lcl
test_fwrite_universal on file_å.txt
In memory we have 11 bytes: 66 69 6c 65 5f c3 a5 2e 74 78 74
Wrote to file successfully
test_fwrite_universal on file_å.txt
In memory we have 10 bytes: 66 69 6c 65 5f 86 2e 74 78 74
Wrote to file successfully
test_fwrite_universal on file_╨╢.txt
In memory we have 11 bytes: 66 69 6c 65 5f d0 b6 2e 74 78 74
Wrote to file successfully
I now get a difference between file_å and file_\u00e5. The character å in the file is actually encoded as 0x86. Notice that this time the second string is 10 characters long not 11. If we look at the file and tell Notepad++ to use UTF-8 we will see a funny result. Same goes to the actual data written to the file.
Finally how to get the damn thing working on windows. Unfortunately It seems that it is impossible to use the standard library with UTF-8 encoded strings. On windows you can't set the C locale to that. see: What is the Windows equivalent for en_US.UTF-8 locale?.
However we can work around this with wide characters:
#include <stdio.h>
#include <string.h>
#include <windows.h>
#define BUF_SZ 255
void test_fopen_windows(const char *fname)
{
wchar_t buf[BUF_SZ] = {0};
int sz = MultiByteToWideChar(CP_UTF8, 0, fname, strlen(fname), (LPWSTR)buf, BUF_SZ-1);
wprintf(L"converted %d characters\n", sz);
wprintf(L"Converting to wide characters %s\n", buf);
FILE* file =_wfopen(buf, L"w");
if (file) {
fwrite((const void*)fname, 1, strlen(fname), file);
fclose(file);
wprintf(L"Wrote file %s successfully\n", buf);
}
}
int main()
{
test_fopen_windows("file_\u00e5.txt");
return 0;
}
To compile use:
gcc -std=gnu99 -fexec-charset=UTF-8 test_wide.c -o test_wide.exe
_wfopen is not ANSI compliant and -std=c99 actually means STRICT_ANSI so you should use gnu99 to have that function.