Only add CFGuard on windows-msvc targets

[ajpaverd]

As @ollie27 pointed out in #73893, the cfguard module flag causes incorrect behavior on windows-gnu targets. This patch restricts rustc to only add this flag for windows-msvc targets (this may need to be changed if other linkers gain support for CFGuard).

[rust-highfive]

r? @ecstatic-morse

(rust_highfive has picked a reviewer for you, use r? to override)

[ollie27]

I think this could do with a couple of ui tests like:

// run-pass
// compile-flags: -Z control-flow-guard
// ignore-msvc

fn main() {
    println!("Hello, world!");
}

and

// run-pass
// compile-flags: -Z control-flow-guard
// only-msvc

fn main() {
    println!("Hello, world!");
}

to make sure we're actually producing working binaries and the warning is printed when it should be.

[ajpaverd]

Thanks @ollie27, do the new ui tests cover the cases you had in mind?

[ollie27]

Thanks @ollie27, do the new ui tests cover the cases you had in mind?

Yep, looks good.

[nikomatsakis]

@bors r+

[bors]

📌 Commit 04a78c6a02db8a70da0fd71f7ad3e878e20c16e7 has been approved by nikomatsakis

[nikomatsakis]

@bors r- -- wait, CI looks unhappy?

[ajpaverd]

Yes, I'm wondering if this is the correct way to test compiler warnings? In the CI, the --emit metadata flag seems to suppress this warning.

[nikomatsakis]

@ajpaverd where is that warning emitted, anyway?

I think that with a run-pass test, the stderr file might not include compiler output, I'm not sure.

[ajpaverd]

@nikomatsakis these warnings are emitted from librustc_codegen_ssa/back/linker.rs. So it looks like the test passes when run normally (L1509), but fails when run with --pass=check (L2765) because there's no codegen. The CI seems to run ui tests in both modes, so I'm not sure what to do here. Should the warnings go somewhere else (or be removed)?

[nikomatsakis]

@ajpaverd ah, I see, that's annoying. I guess the question then is whether there is some kind of test mode that only runs normally

[ajpaverd]

So I think changing this to a run-make test gets us some of the way. We can test that this compiler flag produces runnable binaries on all targets (the issue reported by @ollie27). We don't have an explicit check that the warning is emitted on unsupported targets, but I'm not sure that's strictly necessary (and I'm not sure how we would test it with the current CI config).

[petrochenkov]

You can use // ignore-pass to make the test pass in --pass=check mode.

Emitting hard-coded warnings is never a good idea though, they've been gradually removed from the compiler in favor of lints or hard errors.
There are many -C options that don't have any effect for some targets or linkers, sometimes we don't even decide whether they have an effect or not - LLVM does that for us, usually such options are just silently accepted without warnings.

[ajpaverd]

@petrochenkov thanks! I'm actually in favor of removing these warnings to make this cleaner and more in-line with other -C options. I'll update this PR.

[nikomatsakis]

@bors r+

[bors]

📌 Commit 1ca7bfe has been approved by nikomatsakis

[Manishearth]

@bors rollup=iffy

[ognevny]

did the situation change for non-msvc targets?