[clang] [llvm] Enable LLDB tests in Linux pre-merge CI (PR #94208)

Aaron Ballman via cfe-commits cfe-commits at lists.llvm.org
Wed Jun 5 11:34:31 PDT 2024


AaronBallman wrote:

I had a nice chat with @ldionne and @nikic today and we touched on the topic of precommit CI for lldb being triggered from Clang, and that resulted in a change of opinion and perhaps a better way for me to phrase what I'm after and how it relates to this PR.

I think that what we ultimately need is improved coordination between projects when a change has wider-reaching impacts. If a change to an implementation detail in Clang breaks an lldb test, we don't want to land the change and disrupt the community. But the developer making the change may not be able to fix the issues outside of Clang (lack of expertise, time constraints, etc) and we don't want them to walk away from the PR as a result. I think what would be a better situation would be for 1) the issue to be noticed during a precommit CI run, 2) the PR author to try to address the issues themselves, 3) the impacted project to be willing to step in and help out if needed (discuss solutions, offer to push changes to the PR to solve the issue, etc). Basically, everyone working together to try to get the patch landed with as little disruption or overhead as possible.

(3) comes with a fair amount of friction though because it's not always obvious *who* to reach out to, and (particularly for new contributors) it can be awkward to reach out to strangers to impose on them. So I wonder if it would be possible (or make sense) to have a precommit CI failure to automatically add the relevant PR subscribers/reviewers group to the PR. e.g., if there's a PR with a `clang` label and it breaks an lldb test, then lldb reviewers get automatically added to the PR, but if there's a PR with a `clang` and an `lldb` label and it breaks an lldb test, no new reviewers are added. This should hopefully reduce the friction for (3) because folks no longer need to figure out who to talk to. It should also hopefully reduce the amount of time spent churning on the review because the newly added reviewers can hopefully give some advice on how to resolve the issue (or just push a change to resolve it for the PR author).

If we had something like that, then I think we could manually exclude the handful of sometimes flaky tests but otherwise add lldb precommit CI to Clang's pipeline.

https://github.com/llvm/llvm-project/pull/94208


More information about the cfe-commits mailing list