Ali Caglayan (Nov 25 2022 at 12:39): Ali Caglayan (Nov 25 2022 at 12:39):Why can't we apply overlays in our CI and stop applying them when they are merged?
Ali Caglayan (Nov 25 2022 at 12:39):I never understood why it only applied to the PR being merged.
Théo Zimmermann (Nov 25 2022 at 13:33):Do you have a way of auto-detecting whether the overlay was merged?
Théo Zimmermann (Nov 25 2022 at 13:34):Or would that require additional PRs (possibly opened automatically) to remove the overlays once they are not needed? (If that's the latter, then that would be close to a proposal I made years ago but which was blocked.)
Ali Caglayan (Nov 25 2022 at 13:53):gh CLI can detect if a PR is merged or not I believe. https://cli.github.com/manual/gh_pr_status
Ali Caglayan (Nov 25 2022 at 13:54):And when we detect the overlay has been merged we can emit a warning or do occasional batch clean ups.
Ali Caglayan (Nov 25 2022 at 13:56):We do this check when we are about to apply the overlay.
Ali Caglayan (Nov 25 2022 at 13:57):I'm sure there is a way to tell coqbot about it too.
Ali Caglayan (Nov 25 2022 at 13:57):Then coqbot can open a PR automatically cleaning up the overlays.
Théo Zimmermann (Nov 25 2022 at 13:58):What coqbot could do would be to do a routine check every once in a while.
Ali Caglayan (Nov 25 2022 at 13:58):Every week would be sufficient IMO.
Ali Caglayan (Nov 25 2022 at 13:59):Ali Caglayan said:
ghCLI can detect if a PR is merged or not I believe. https://cli.github.com/manual/gh_pr_status
This obviously doesn't work for Gitlab however.
Pierre-Marie Pédrot (Nov 25 2022 at 13:59):I don't think this is a good idea, because overlays can become very quickly obsolete.
Ali Caglayan (Nov 25 2022 at 14:00):@Pierre-Marie Pédrot What do you mean?
Théo Zimmermann (Nov 25 2022 at 14:00):What do you fear @Pierre-Marie Pédrot?
Pierre-Marie Pédrot (Nov 25 2022 at 14:00):Especially when there are competing PRs with overlays on the same dev, this means essentially adding a one week-delay for any progress
Ali Caglayan (Nov 25 2022 at 14:00):How is that any different to the current situation?
Théo Zimmermann (Nov 25 2022 at 14:00):Hum? What are you talking about?
Pierre-Marie Pédrot (Nov 25 2022 at 14:01):An overlay only makes sense for a given PR.
Pierre-Marie Pédrot (Nov 25 2022 at 14:01):Once you start applying overlays for several branches you have to consider the possibility of conflict.
Ali Caglayan (Nov 25 2022 at 14:02):and the job will fail right?
Ali Caglayan (Nov 25 2022 at 14:02):So like it does today then
Pierre-Marie Pédrot (Nov 25 2022 at 14:02):Yes, for everybody.
Ali Caglayan (Nov 25 2022 at 14:02):If an overlay has not been merged it will fail for everybody, today.
Pierre-Marie Pédrot (Nov 25 2022 at 14:02):Today the lock is not on our side, which is the problem.
Pierre-Marie Pédrot (Nov 25 2022 at 14:03):But making the lock time-based with a week delay is probably worse than a few hours, or days at best, of missing downstream dev overlays.
Ali Caglayan (Nov 25 2022 at 14:03):We are not making anything time based?
Ali Caglayan (Nov 25 2022 at 14:03):We are suggesting that the CI for everybody tries to apply the overlay.
Pierre-Marie Pédrot (Nov 25 2022 at 14:03):That means a merge, then?
Ali Caglayan (Nov 25 2022 at 14:03):And doesn't apply it once it is merged.
Ali Caglayan (Nov 25 2022 at 14:04):Pierre-Marie Pédrot said:
That means a merge, then?
Yes we will merge the PR into the CI sources.
Théo Zimmermann (Nov 25 2022 at 14:04):Pierre-Marie Pédrot said:
But making the lock time-based with a week delay is probably worse than a few hours, or days at best, of missing downstream dev overlays.
You've completely misunderstood Ali's proposal.
Pierre-Marie Pédrot (Nov 25 2022 at 14:04):What I fear is the criterion to detect that an overlay was merged.
Ali Caglayan (Nov 25 2022 at 14:05):Right, that is what gh does. You can query GitHub directly about a PR.
Ali Caglayan (Nov 25 2022 at 14:05):And come up with a curl request for GL probably.
Ali Caglayan (Nov 25 2022 at 14:05):We can quite easily understand if a PR/MR has been merged and act accordingly.
Ali Caglayan (Nov 25 2022 at 14:05):What we were suggesting with the week clean up cycle is to remove the overlays we know have been merged.
Théo Zimmermann (Nov 25 2022 at 14:06):Another solution: if the overlay does not apply cleanly on the base branch, you just test the base branch (instead of failing).
Pierre-Marie Pédrot (Nov 25 2022 at 14:06):We still have to care for semantic conflict of overlays, though.
Ali Caglayan (Nov 25 2022 at 14:06):That's fine, we fail if overlays conflict.
Ali Caglayan (Nov 25 2022 at 14:06):That is strictly better than what we are doing today.
Ali Caglayan (Nov 25 2022 at 14:06):Which is fail everywhere when unmerged.
Pierre-Marie Pédrot (Nov 25 2022 at 14:07):Indeed. I guess I'll to see it in action.
Notification Bot (Nov 25 2022 at 14:08):43 messages were moved here from #Coq devs & plugin devs > compcert / vst waiting for overlay by Ali Caglayan.
Pierre-Marie Pédrot (Nov 25 2022 at 14:09):(this also assumes that downstream devs merge PR overlays correctly, which is not always the case. I've seen several people rebasing the commits locally and closing the PR.)
Ali Caglayan (Nov 25 2022 at 14:09):Yes that is a problem
Ali Caglayan (Nov 25 2022 at 14:10):Anyway just an idea how we might improve things rather than be at the whim of every CI entry.
Théo Zimmermann (Nov 25 2022 at 14:12):Pierre-Marie Pédrot said:
(this also assumes that downstream devs merge PR overlays correctly, which is not always the case. I've seen several people rebasing the commits locally and closing the PR.)
That's fine if we do not use the gh trick proposed by Ali but instead just check if the overlay merges cleanly with the base and otherwise skip it.
Théo Zimmermann (Nov 25 2022 at 14:12):As for the auto-removal PR that coqbot could open, it could be very stupid: just remove overlays and let CI tell you if that works or not.
Last updated: May 31 2023 at 17:01 UTC