mingw: support -municode

[ypsvlq]

fixes #18621

also rename mingwex to mingw32, which feels more appropriate as it contains the basic C runtime code

[squeek502]

Nice work, thanks for getting this fixed! Tested and confirmed this fixes the minimal reproduction in https://github.com/ziglang/zig/issues/18621#issuecomment-1989673384.

wWinMain also works when using zig cc:

#include <windows.h>

int APIENTRY wWinMain(HINSTANCE hInst, HINSTANCE hInstPrev, PWSTR cmdline, int cmdshow) {
    return 0;
}

zig cc -o test-wwinmain-cc.exe test-wwinmain.c -lc -municode -Wl,--subsystem,windows

> dumpbin /headers test-wwinmain-cc.exe
...
            1180 entry point (0000000000401180) WinMainCRTStartup
...
               2 subsystem (Windows GUI)
...

But is there an intended way to provide -municode when using zig build-exe?

zig build-exe -cflags -municode -- test-wwinmain.c -lc --subsystem windows

doesn't seem to work

[ypsvlq]

fixed

[squeek502]

Great! One last thing to make this maximally usable is integration with build.zig. I've added that + a standalone test and submitted it as a PR to your branch here: https://github.com/ypsvlq/zig/pull/1

---

For future reference, after this PR, here's how to build a C file with all possible entry points using zig build-exe (add -target x86_64-windows-gnu if cross compiling):

zig build-exe main.c -lc

zig build-exe wmain.c -lc -municode

zig build-exe winmain.c -lc

zig build-exe wwinmain.c -lc -municode

(can add --subsystem windows to set the subsystem for the WinMain/wWinMain versions but it's not strictly necessary)

See the added standalone test for the build.zig equivalent.

[squeek502]

I'm confused why https://github.com/ziglang/zig/issues/18621 is marked as a regression when the fix requires adding a new feature (the -municode flag). How did it used to work?

There are two parts to that issue (and this PR):

• The regression is that if WinMain is defined, it isn't detected/recognized when targeting MinGW and the linker fails with undefined symbol: main. In particular, this comment has a reproduction of only the regression
• The -municode part of this PR is to support wmain/wWinMain which I don't think has ever been fully supported before when targeting MinGW (at least I couldn't get wWinMain to work with 0.11.0)

@ypsvlq can better clarify which parts of this PR address one or the other.

---

One thing I'd be interested in is if it's at all feasible to make MinGW recognize wWinMain/wmain without needing an explicit CLI flag (e.g. this is the case when targeting MSVC--it will use wWinMain without any CLI flags needed and without needing UNICODE defined AFAICT). Currently with this PR, these are functionally equivalent:

> zig build-exe wwinmain.c -target x86_64-windows-msvc -lc
> dumpbin /headers /nologo wwinmain.exe
           A9AB0 entry point (00000000004A9AB0) wWinMainCRTStartup
               2 subsystem (Windows GUI)

(both the entry point and the subsystem are inferred)

> zig build-exe wwinmain.c -target x86_64-windows-gnu -lc -municode --subsystem windows
> dumpbin /headers /nologo wwinmain.exe
            11A0 entry point (00000000004011A0) WinMainCRTStartup
               2 subsystem (Windows GUI)

(without -municode, the linker errors with undefined symbol: WinMain; without --subsystem windows, the subsystem is 3 subsystem (Windows CUI))

[ypsvlq]

How did it used to work?

The version of mingw used in 0.11 includes the WinMain startup code in crtexe.c, in later versions it has been moved to a separate file which wasn't included in Zig. As far as I can tell wWinMain was never supported.

[chawyehsu]

This is great! Thanks @ypsvlq for working on it! And I believe it will also close #9924 , in which unicode entry points support against mingw was reported missing the first time I guess.