Zig build fails if the user folder contains cyrillic characters

[KatPurpy]

Zig Version

0.10.0-dev.139+850b053ea

Steps to Reproduce

The error seems to occur on windows due to Zig toolchain using UTF-8 when Windows uses UTF-16 for paths

1. Have a user folder which name is written in cyrillic characters
2. Run zig init-exe
3. Run zig build
4. Witness the error

Expected Behavior

No errors

Actual Behavior

Just in case: [REDACTED] is a name written in cyrillic characters such as Василий. Edited for privacy reasons.

error(mingw): clang exited with code 1 and stderr: error: unable to open output file 'C:\Users\[REDACTED]\AppData\Local\zig\o\f6eeb75a2cedb4cc58f95d9ea70a607d\kernel32.def': 'Символ Юникода не имеет сопоставления в конечной многобайтовой кодовой странице.'
1 error generated.

error(mingw): clang exited with code 1 and stderr: error: unable to open output file 'C:\Users\[REDACTED]\AppData\Local\zig\o\c6379a4de5888bcd947790c71c974c8e\ntdll.def': 'Символ Юникода не имеет сопоставления в конечной многобайтовой кодовой странице.'
1 error generated.

error(mingw): clang exited with code 1 and stderr: error: unable to open output file 'C:\Users\[REDACTED]\AppData\Local\zig\o\6d7ede8e82a0e42ab5d39e4d7fda3f42\advapi32.def': 'Символ Юникода не имеет сопоставления в конечной многобайтовой кодовой странице.'
1 error generated.

error: unable to generate DLL import .lib file: ClangPreprocessorFailed

[squeek502]

I've tried looking into this a bit and from what I can tell the issue lies with the pub fn main(argc: c_int, argv: [*][*:0]u8) callconv(.C) c_int entrypoint in stage1.zig, as it can't successfully receive non-ASCII characters.

This is reproducible by setting the ZIG_GLOBAL_CACHE_DIR env variable to a path that includes non-ASCII characters, e.g.:

set ZIG_GLOBAL_CACHE_DIR=C:\Users\Ryan\Programming\Zig\tmp\cache\Кириллица

In mingw.zig, def_final_path here:

zig/src/mingw.zig

Lines 331 to 333 in b316c25

	const def_final_path = try comp.global_cache_directory.join(arena, &[_][]const u8{
	"o", &digest, final_def_basename,
	});

is UTF-8 that looks like this when printed with std.zig.fmtEscapes:

"C:\\Users\\Ryan\\Programming\\Zig\\tmp\\cache\\\xd0\x9a\xd0\xb8\xd1\x80\xd0\xb8\xd0\xbb\xd0\xbb\xd0\xb8\xd1\x86\xd0\xb0\\o\\ae484bf4ddcd5e1d90fab86e1a85305e\\kernel32.def"

And then when it's received by main in stage1.zig after mingw.zig calls its zig clang ... incantation via ChildProcess (which correctly converts to UTF-16 before passing to CreateProcessW), it looks like this:

"C:\\Users\\Ryan\\Programming\\Zig\\tmp\\cache\\?????????\\o\\ae484bf4ddcd5e1d90fab86e1a85305e\\kernel32.def"

---

Note: If the non-ASCII characters are within the Windows-1252 range, it causes a different error as the stage1.zig main receives it encoded as Windows-1252:

// before ChildProcess.init in mingw.zig (correct UTF-8 for á)
"L\xc3\xa1tin-1"
// what gets received by stage1.zig's main (Latin-1/Windows-1252 encoded value for á)
"L\xe1tin-1"

And the error looks like this:

error: unable to open output file 'C:\Users\Ryan\Programming\Zig\tmp\cache\L<E1>tin-1\o\b5df4c03d7232331db68bf600c63397a\ntdll.def': 'unspecified system_category error'