audit all contributions by @devnexen for correctness and delete unnecessary libc ABI bits

[devnexen]

No description provided.

[andrewrk]

@mikdusan is there anything here that you want to keep? If so, would you be willing to audit it for correctness, and then add a commit to this branch to keep it?

I will do the same.

[rofrol]

What happend here?

[andrewrk]

Context: #16333 (comment)

[lin72h]

@devnexen Thanks for working on these freebsd integration bits, I've been following all your PR and learned a lots about freebsd internal.

[likern]

Why there isnt any integration tests written to run under BSD running in QEMU? I think that would solve the problem with correctness and trust, isn't it?

[mikdusan]

Why there isnt any integration tests written to run under BSD running in QEMU? I think that would solve the problem with correctness and trust, isn't it?

• bsd-user is not supported on a linux host
• full disclosure, I have never tried bsd-user on any system, but it looks like it needs to run on a bsd host
• emulating a bsd system is really the only way we are going to get a bsd host and is something we'd like to do and there appears to be a path to that goal but it just hasn't happened yet

IMO, we have a few issues about libc on BSD. It is not very well organized and lacks direct tests. Basically libc only gets test coverage to the extent zig's std depends on it. Some further thoughts; feel free to critique:

1. A direct libc test suite. eg: create a test for freebsd mmap() special MAP_ALIGNED(n) flag and verify alignment. Should be a simple test and yet it verifies that we are using the correct binary flags. It also doesn't depend on zig std somehow requiring to use it.

2. It is possible that a direct libc test suite could get "out of hand" to be bundled with zig. If that is or becomes the case, there is nothing preventing us from launching a libc testsuite repo for zig.

3. libc tests optioned to indicate host capabilities for testing; eg: -Dnetwork-ipv6,-Dnetwork-ipv4, -Dnetwork-internet, -Dnetwork-dns, -Dnetwork-icmp. If a host doesn't have the capability, skip.

[semarie]

I added lines comments regarding OpenBSD contributions for what I am sure it is need or important to have. I didn't comment for others BSDs.

[likern]

I thought of more like running OpenBSD over full virtualization using kvm, which seems supports it according to this, at least having it run and failed on PR and commits.

To run it on GitHub Actions seems it need to have nested virtualization, kvm inside GA. It's not supported as is *BSD natively.

BUT macos machines does support this (can run Virtualbox). There is GitHub Action which supports it using this method.

Another way is to run self-hosted runners on native BSD system and run GA on VM (containers?) using bhyve, as shown here https://github.com/myb-project/guide/blob/main/en/gh_runners.md. It allows to run DragonflyBSD, FreeBSD, NetBSD, OpenBSD.

Similar approach is to run self-hosted runner on Linux with kvm support on it and run VMs on it.

[andrewrk]

I re-did the work myself over here, in order to avoid unnecessary trust: #16636

Thank you for your comments @semarie - I will go over them one by one now.

[andrewrk]

To all: testing is not a solution to sloppy contributions. Testing is certainly an important component of quality assurance, but when it comes to external ABIs, there is no substitution for expertise and carefulness. There could easily be extern functions that pass automated tests, yet expose a subtle flaw that could be exploited or triggered under certain conditions, and all of these commits are exactly the kind that are in danger of creating that exact scenario.