Reading GHC MRs related with structured error messages, I realized that there are seemingly inconsistent parts. Some MRs are marked close while the progress says “completed”. Other closed MRs are noted as in progress. What is happening here, and what is the process of MRs getting merged?
Are there some examples you have in mind? It’s easier to explain in the context of some examples.
Sure!
5569 is closed, but it does not seem to mean that the plan is cancelled.
It’s because we use Marge bot which creates a new MR combining the commits of several other MRs into one batch. That’s what get merged. E.g. !5569 is closed because it has been merged via !5800.
Oh. Though, there is no mention that 5800 was merged.
And in fact !5800 says "The changes were not merged into master". However, I don’t think we needed to look there because !5569 says “Merged in aac87bd3”.
And in fact !5800 says "The changes were not merged into
master".
Yes because Marge bot merges into the master branch independently of Gitlab, so for Gitlab the MR is just closed and not merged, hence the message. Perhaps Marge bot could merge the MR instead. But Marge bot is just a temporary solution until the merge train feature works correctly (cf Move to Merge Trains for merging branches (#19046) · Issues · Glasgow Haskell Compiler / GHC · GitLab) so there is some reluctance to modify it.
1 Like
I see. Perhaps in that case it would be helpful if Marge Bot left a message saying "Actually I did merge this into master" – but I understand the reluctance to spend effort modifying Marge if she is going to be replaced in due course.
I think the structural error handling MR is nearly finished, judging by what little I know of the progress. Is it the case, or does it need extra batch of work?