Add NetBSD CI

[nathaniel-bennett]

Description

This PR adds NetBSD to the CI for nightly target and resolves current build/test failures resulting from NetBSD modules.

Fixes #3890
Fixes #3965

[rustbot]

r? @tgross35

rustbot has assigned @tgross35.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[rustbot]

Some changes occurred in OpenBSD module

cc @semarie

Some changes occurred in OpenBSD module

cc @semarie

[nathaniel-bennett]

Looks like this assertion in ctest2 fails during the build process for libc-test:
https://github.com/JohnTitor/ctest2/blob/80f9cc420ed2739631f1604f44744107ad10d9a0/src/lib.rs#L2107

Based on context, it looks like some variadic function in the NetBSD module is tripping up the testing harness. I'm willing to bet it's esetfunc:
https://github.com/rust-lang/libc/blob/fbec9289b33627d03b72701a296095c9e91354fd/src/unix/bsd/netbsdlike/netbsd/mod.rs#L2953C1-L2955C72

...as it's the only function in the entirety of the libc crate that passes a variadic function as a parameter.

Potential solutions:

1. Refactor ctest2 to handle variadic functions as parameters.
2. Configure build.rs to ignore esetfunc until (1.) is accomplished.

I've gone ahead and opted for 2. at the moment, since I have no experience with the ctest2 crate and wouldn't be able to perform that refactor.

[nathaniel-bennett]

It appears that adding an exception to the libc-test/build.rs script doesn't fix this assertion; removing esetfunc from semver/netbsd.txt similarly fails to fix the issue.

I've temporarily opted to remove the esetfunc function entirely to get past that build error and see if the rest of CI works as normal; obviously this is a breaking change, so we may want to fix ctest2 and re-add it before 1.0 is released.

[nathaniel-bennett]

Source for changing sendmmsg/recvmmsg flags parameter from int to unsigned int:

https://github.com/NetBSD/src/blob/da7ff22a4ae89b32ca9e90ca20e7d0d434e8cdd2/sys/sys/socket.h#L651-L654

[nathaniel-bennett]

(Note that these stay as int for OpenBSD):

https://github.com/openbsd/src/blob/22b55b0b9cd9bc475c3f48959939e31a06d92cbe/sys/sys/socket.h#L576
https://github.com/openbsd/src/blob/22b55b0b9cd9bc475c3f48959939e31a06d92cbe/sys/sys/socket.h#L581

[nathaniel-bennett]

Documentation on kevent now using a void pointer for its udata field:

https://github.com/NetBSD/src/blob/da7ff22a4ae89b32ca9e90ca20e7d0d434e8cdd2/sys/sys/event.h#L66

[nathaniel-bennett]

Documentation that uucred using a short type instead of an int for its cr_ngroups field:

https://github.com/NetBSD/src/blob/da7ff22a4ae89b32ca9e90ca20e7d0d434e8cdd2/sys/sys/ucred.h#L48

[nathaniel-bennett]

Documentation that sockaddr_dl uses a separate dl_addr struct with an address field of 24 bytes, not 12:

https://github.com/NetBSD/src/blob/da7ff22a4ae89b32ca9e90ca20e7d0d434e8cdd2/sys/net/if_dl.h#L66-L91

[nathaniel-bennett]

Documentation of correct utmpx fields (thereby removing incomplete type __exit_status):

https://github.com/NetBSD/src/blob/da7ff22a4ae89b32ca9e90ca20e7d0d434e8cdd2/include/utmpx.h#L103-L118

[nathaniel-bennett]

Looks like there are a ton of CI failures scattered across the NetBSD module--too many for me to try to resolve at the moment. I'll try to knock out some more later this week in this PR.

[tgross35]

Thanks for taking a look at this, it would be great to have something here. Feel free to put any of the changes into smaller PRs that can be merged on their own, it would be good to get any problems fixed even if we aren't testing it right away.

Any reason to use GHA rather than Cirrus?

Also I don't think we need to test MSRV, we don't do that for other tier2 targets.

@semarie it would probably be best if you get involved here.

[tgross35]

It would probably be best to just adjust ci/install-rust.sh as needed (if at all) to do what is needed here.

[nathaniel-bennett]

Any reason to use GHA rather than Cirrus?

Honestly, I mostly used it because I knew how to set it up based on past projects (whereas I don't have any familiarity with Cirrus). I would be perfectly fine if someone wanted to move it to a Cirrus setup; based on my preliminary exploring, they seem to run NetBSD VMs on top of Linux just like GHA does (though I could be wrong there).

[tgross35]

There is still some work needed here right?

@rustbot author

[bors]

☔ The latest upstream changes (presumably #4018) made this pull request unmergeable. Please resolve the merge conflicts.

[tgross35]

CI now has a test_tier2_vm job based on vmactions for Solaris

libc/.github/workflows/ci.yaml

Lines 134 to 159 in 93b49b1

	test_tier2_vm:
	name: Test tier2 VM
	needs: [test_tier1, style_check]
	runs-on: ubuntu-latest
	strategy:
	fail-fast: true
	matrix:
	target:
	- x86_64-pc-solaris
	steps:
	- uses: actions/checkout@v4
	- name: test on Solaris
	uses: vmactions/solaris-vm@v1
	with:
	release: "11.4-gcc"
	usesh: true
	mem: 4096
	copyback: false
	prepare: |
	set -x
	source <(curl -s https://raw.githubusercontent.com/psumbera/solaris-rust/refs/heads/main/sh.rust-web-install)
	rustc --version
	uname -a
	run: |
	export PATH=$HOME/.rust_solaris/bin:$PATH
	./ci/run.sh ${{ matrix.target }}

. If the existing prepare script is moved to install-rust.sh then the job could just take a matrix with target and vm to handle both Solaris and NetBSD