Close more old stage1 issues

[wooster0]

How can I help closing old stage1 issues?

The current policy is that before an old stage1 issue can be closed, test coverage for it has to be added.

1. Go to https://github.com/ziglang/zig/issues?q=is%3Aopen+is%3Aissue+label%3Astage1

2. Close your eyes and click on an issue

3. Determine whether the code in the issue is supposed to compile-error or simply pass.

• If the code from the issue is supposed to just pass, maybe clean it up a bit and put the code in test/behavior/bugs/github_issue_number.zig and then reference that file in test/behavior.zig at the right spot (there's an order). Use zig test x.zig to check locally if it does actually pass.
  Example and its reference in test/behavior.zig
• If it's supposed to compile-error, create an explicitly-named file for how the code is supposed to compile-error in test/cases/compile_errors/. You can check the code locally with zig test x.zig to check what errors you get and copy-paste them into the file at the end as comments like in the other compile error case files.
  Example

4. In your commits' description you can list all issues you provided test coverage for like Closes #1 each on a separate line, like in the commits of this PR.

[wooster0]

The two failures only in linux-debug look kind of sporadic.

Edit: and just after I sent this comment the linux-debug CIs are magically re-running? Same failure though so I'm going to rebase.

[wooster0]

All 118 tests passed.
thread 776410 panic: integer overflow
/home/ci/actions-runner11/_work/zig/zig/src/type.zig:3579:55: 0x36e66c7 in intAbiAlignment (zig)
            std.math.ceilPowerOfTwoPromote(u16, (bits + 7) / 8),
                                                      ^

These two CI failures are a mystery to me. If I just zig test (no optimizations and I'm on x86_64-linux so it'll be equal to x86_64-linux-debug) every new file in test/behavior/bugs/ myself, it all passes. I also checked the CBE with -ofmt=c.

[Vexu]

These two CI failures are a mystery to me. If I just zig test (no optimizations and I'm on x86_64-linux so it'll be equal to x86_64-linux-debug) every new file in test/behavior/bugs/ myself, it all passes. I also checked the CBE with -ofmt=c.

The failure depends on whether your compiler was built with safety checks and is caused by bugs/12116.zig

[andrewrk]

need to actually test that the code runs correctly; it's not enough to just check that the compiler doesn't crash.

[wooster0]

Alright, I tried to come up with some things I could do there. You can find it in my latest force-push.

[andrewrk]

Note that test/behavior/bugs/*.zig is also kind of a chore list for me. Ideally these would be properly categorized like this: #13997

It's a slight improvement to get from github issues to test/behavior/bugs/*.zig, so I still appreciate the work you're doing here, but it would be even more helpful if the behavior was categorized.

[wooster0]

I fully agree on that and was going to bring up this kind of dependence on GitHub because I don't think we should really reference GitHub issues in our codebase. The issues only exist on Microsoft's GitHub and I think it's best to stay flexible in terms of the code hosting platform. Maybe someday Zig uses a different forge like sourcehut or something.

[wooster0]

I was considering removing the newly added issue files but as you said it's still a slight improvement over what we have currently so.

[andrewrk]

Behavior tests need to verify that the logic ran correctly, not only verify that the compiler did not crash.

[wooster0]

Right, that makes sense. Something like this?

[andrewrk]

Wonderful, thank you!