Bedevere is a little trigger-happy closing PRs

[StanFromIreland]

For example, it closed python/cpython#155197 today, because "PRs attempting to merge a maintenance branch into the main branch are deemed to be spam and automatically closed." However, this PR was not attempting to do that, it had a branch named 3.16 with a commit with a fix. I'm not familiar with the logic here behind this, but I think a little sanity check would be good to avoid spurious closures in the future.

[Mariatta]

The logic checks for the branch name only, not the content. So if people made a branch that matches a release version name, it will be closed. I feel this is easily solvable by the contributor by choosing a different branch name. Perhaps the message can be updated to mention that they shouldn't create branch that matches a release branch name?

[StanFromIreland]

Sounds good to me. It's rare enough that a note should be sufficient. I just don't want to ruin a contributor's day with a blunt "closed as spam" message.

[zware]

For some history, see #70 (the first version of the auto PR closure) and #137/#138 (which added the current message). We added this back at a time when we were getting rather a lot of these PRs, possibly at least in part due to not having had issues enabled at the time. It might be possible that we don't actually need it anymore; checking how many of this kind of thing we've had in the past N months could be good.

[Mariatta]

Yeah back then GitHub UI made it super easy for people to create this kind of PR so we were getting many of these.
Is the UI different now?

[zware]

I'm not sure if the UI is different, if branch protections have caught up to our needs, or if most of the old spam was actually just due to not having issues turned on (and now we just get exciting new issue spam). It does seem like the problem has pretty much evaporated, though: a search shows exceedingly few true positive matches for this since 3.11 in 2022. In fact, I don't see any for 3.12, 3.13¹, 3.14, or 3.15. There were quite a few in the 3.7 time period.

Unless those searches are failing me, I think we might actually be able to drop this feature.

Footnotes

1. There are a couple of results here, but they appear to have been well-meaning mistakes rather than the spam we were combating. ↩

[hugovk]

For some history, see #70 (the first version of the auto PR closure) and #137/#138 (which added the current message).

23 results for the message since it was added in 2018.

How many were triggered on PRs like python/cpython#155197 which shouldn't have?

[zware]

There are at least 32 correct closures for just 3.7 in 2019 (see my 3.7 link above), so I don't think we can trust GitHub search at all for this :)