linkers: basic support for the 'zig cc' linker by Akaricchi · Pull Request #1229...

Mind adding a release notes snippet? Just something simple with an example on how to us it would be good enough

I'll add a release snippet but I'm sure if a usage example is necessary. You use it like any other compiler; zig cc/zig c++ are drop-in replacements for clang/clang++ respectively. This already worked with zig 0.10, because it just used the lld linker. Now it just works with zig 0.11 too.

Previously, #11918

This PR seems to be better as it resolves an outstanding review comment from that PR about adding a unique linker class, and even smooths over some incompatibilities. ;)

I would still like to see zig installed in CI and tested, in addition to the release notes. Please also see the comment there about -fuse-ld.

You can squash the commits. Thanks!

Now that I think about it, the linker id should probably be ld.zig, to be consistent with other GNU-style linkers. That way there should be less confusion if e.g. we ever support Zig as a language in Meson — the linker will probably have a different interface there.

Please also see the comment there about -fuse-ld

@eli-schwartz
Zig seems to completely ignore -fuse-ld. What should we do about it?

Some of the comments in the other PR seem inaccurate.

That just means that if users try to configure a linker for zig, meson will error out when trying to pass that linker as -fuse-ld.

There's no error or even a warning; as I said above, -fuse-ld is just ignored.

It depends on the target platform. In linux, it's compatible to clang. It will probably be a different story on MacOS or Windows, which I did not test.

zig cc has the same clang/lld-compatible interface on all platforms. The article I linked even shows this by running it in Wine.
EDIT: or so it seems; I'm actually not sure about the linker part anymore. One thing I know is I did cross-compile a moderately complex project for windows with this patch.

Now that I think about it, the linker id should probably be ld.zig, to be consistent with other GNU-style linkers. That way there should be less confusion if e.g. we ever support Zig as a language in Meson — the linker will probably have a different interface there.

I might be working on that. For the moment Zig looks more like Rust, generating a final output directly. Like Rust it can compile object files but getting Meson to the point it does that will be non-trivial.

I would prefer that the linker be called someting other than just zig, I don't love ld.zig, but I could live with it. what about zigcc ?