Fixed bug 2786 - "UCS-2-INTERNAL" iconv encoding is not supported everywhere, use UTF-16LE instead

Jonas Kulla

src/main/windows/SDL_windows_main.c:137:
cmdline = SDL_iconv_string("UTF-8", "UCS-2-INTERNAL", (char *)(text), (SDL_wcslen(text)+1)*sizeof(WCHAR));

I'm trying to compile an SDL2 application for windows using the mingw-w64 32bit toolchain provided by my distro (Fedora 19). However, even the simplest test program that does nothing at all fails to startup with a "Fatal error - out of memory" message because the mingw iconv library provided by my distro does not support the "UCS-2-INTERNAL" encoding and the conversion returns null.

From my little bit of research, it turns out that even though this encoding is supported by the external GNU libiconv library, some glibc versions (?) don't support it with their internal iconv routines, and will instead provide the native endian encoding when "UCS-2" is specified.

Nonetheless, I wonder why the native endianness is considered in the first place when Windows doesn't even run on any big endian archs (to my knowledge). And true enough, 'WIN_StringToUTF8' from core/windows/SDL_windows.h is used everywhere else in the windows backend, which is just a macro to iconv with "UTF-16LE" as source. Therefore it would IMO make sense to use this macro here as well, which would solve my problem (patch attached).

src/main/windows/SDL_windows_main.c

@@ -134,7 +134,7 @@ WinMain(HINSTANCE hInst, HINSTANCE hPrev, LPSTR szCmdLine, int sw)
     /* Grab the command line */
     TCHAR *text = GetCommandLine();
 #if UNICODE
-    cmdline = SDL_iconv_string("UTF-8", "UCS-2-INTERNAL", (char *)(text), (SDL_wcslen(text)+1)*sizeof(WCHAR));
+    cmdline = WIN_StringToUTF8(text);
 #else
     cmdline = SDL_strdup(text);
 #endif