CommandLineToArgvW() always returns a null pointer on Windows 9x

[Aerocatia]

Take the example here

#include <windows.h>

int WINAPI WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance, LPSTR lpCmdLine, int nCmdShow) {
    int argc;
    wchar_t **argv = CommandLineToArgvW(GetCommandLineW(), &argc);
    if(!argv) {
        MessageBoxA(NULL, "returned nullptr", "test", MB_OK | MB_ICONERROR);
    }
    else {
        MessageBoxA(NULL, "success", "test", MB_OK);
    }

    return 0;
}

It always fails on 9x. I don't think it could be a real allocation problem because regular allocations with malloc() work just fine.
GetCommandLineW() seems to work correctly.
Tested on Windows 98 SE and Windows ME. I am using mingw32_686-msvcrt_win98-15.2.0-r7

[CyanoHao]

Background:

• Windows 9x: most -W APIs are implemented as stubs (exist but always fail), with several exceptions.
• MinGW Lite: missing APIs required by C++ standard library are replaced by thunks (translate wide strings to narrow strings and call -A APIs).

The problem:
CommandLineToArgvW is a stub (and not replaced by thunk), and GetCommandLineW is one of the exceptions.

Workaround:
There is a thunk used by the toolchain itself. You can copy it to your project:

extern "C"
LPWSTR *my_CommandLineToArgvW(LPCWSTR lpCmdLine, int *pNumArgs)
{
    if (*lpCmdLine == 0) {
      *pNumArgs = 1;

    // ......
    // (replace namespace stl with std, and remove namespace libc)
}

Further plan:
Since “using -W version of Win32 API” is considered “best practice”, I think it’s good to fill the gap by providing more thunks for -W APIs.

[Aerocatia]

Ah, I see. Thank you for explaining. I was under the impression those thunks were silently exposed to programs compiled by this version of mingw.
This is fine anyway, I only came across this porting other code. I am personally not too worried about using the -A APIs and using the UTF-8 codepage on Windows 10 and 11, although this would leave 2000 through 8.1 in an awkward spot so copying them is not a bad idea either.

As an aside, do you think silently linking to thunks using a system similar to the __USE_MINGW_ANSI_STDIO define would be feasible in the future?