Is `rustc -o` supposed to prescribe the *full* filename incl. extensions, or is it only supposed to describe the filename without extensions?

[jieyouxu]

To me, both prescribes-full-filename or filename-without-extension effects make sense, but we should probably be consistent about this. It seems like

• On x86_64-pc-windows-gnu, rustc foo.rs -o foo --target=x86_64-pc-windows-gnu produces foo.exe, but
• On x86_64-pc-windows-msvc, rustc foo.rs -o foo --target=x86_64-pc-windows-msvc produces foo.

I would've expected either both to have the .exe extension, or neither, but not differently? 🤔

I'm on a slightly older nightly toolchain; I don't remember seeing any recent changes surrounding this.

rustc 1.92.0-nightly (595b9a498 2025-10-03)
binary: rustc
commit-hash: 595b9a498bc55fcd30111e430d8e4290ed833b4c
commit-date: 2025-10-03
host: x86_64-pc-windows-msvc
release: 1.92.0-nightly
LLVM version: 21.1.2

[mati865]

Can you show linker invocation for the both cases? I think that would be rustc --print link-args.

[ChrisDenton]

Looking at the code we don't do anything on the rustc side, we just pass "/OUT:foo" (msvc) or "-o" "foo" (gnu) to the linker (or linker driver).

rust/compiler/rustc_codegen_ssa/src/back/linker.rs

Lines 325 to 327 in 4fd3181

	fn output_filename(&mut self, path: &Path) {
	link_or_cc_args(link_or_cc_args(self, &["-o"]), &[path]);
	}

rust/compiler/rustc_codegen_ssa/src/back/linker.rs

Lines 1047 to 1051 in 4fd3181

	fn output_filename(&mut self, path: &Path) {
	let mut arg = OsString::from("/OUT:");
	arg.push(path);
	self.link_arg(&arg);
	}

The .exe is certainly a strong convention and it does ease portability concerns if it's automatically appended (also running an exe via Command appends .exe if no extension is used). So I can see an argument for doing it consistently if no other extension is used.

That said, using the exact name even if it's lacking an extension is also reasonable albeit a very niche use case.

[mati865]

Thanks, that's what I anticipated. Although I find it odd, that link.exe doesn't add the extension.

[ChrisDenton]

msvc link is very old.

In any case we can normalize this. My only concern is that it's technically a breaking change. Albeit one that hopefully won't affect too many people (maybe nobody).

[jieyouxu]

Yeah, I'm not too concerned about changing this behavior, but if we don't want to change it (which is fine by me!) we probably should document this platform difference. At least, it certainly caught me by surprise when trying to figure out a test difference.

[jieyouxu]

In some sense, I think there might be two use cases (and so may deserve separate compiler flags if there's more use cases):

1. Being able to control the name-without-extension part of the output filename. User is totally fine with platform-conventional extensions.
2. Being able to control the exact output filename. User wants absolute control over the filename, explicitly overriding platform conventions.