| 12 Sep 2023 |
 @piegames:matrix.org | Just a random idea, what if the automatic migration skipped all packages with open PRs at first to not make any conflicts? Then do one or two more such runs in the hope of catching some more packages, and then the assumption is that everything left is mostly old stuff that is allowed to conflict | 09:03:19 |
 Rick (Mindavi) | I thought the migration would take care of that by giving git enough hints? Or doesn't that work (well enough)? | 10:27:54 |
| @piegames:matrix.org | Well yes but it still requires rebasing | 10:28:21 |
| @piegames:matrix.org | Not sure how many PRs are even affected, and if it's worth it. Just an idea | 10:28:37 |
 infinisil | piegames: Hmm yeah I discussed this before, my main argument for not doing this is the complexity of doing so | 10:32:42 |
| @piegames:matrix.org | What complexity? (not rhetorical question) | 10:33:14 |
| infinisil | I wonder if it's worth it, because rebase PR's should be fairly quick on an individual basis, whereas implementing this requires searching through all PR's (automated of course), and then adding those to an exception list or something, having to decide when to do it anyways | 10:33:49 |
| @piegames:matrix.org | Though I'd be interested in some numbers here, like how many pull requests and how many packages are affected by this anyways? How many of these pull requests are younger than three months? | 10:34:49 |
| infinisil | And how many PR's will not have to rebase because of this? I guess the turn-over rate is fairly high, just waiting some weeks probably gets a high percentage | 10:34:56 |
 profpatsch | piegames: I had the same idea during nixcon, but honestly I think it’s totally feasible to require PRs to be rebased on top of master | 10:35:42 |
| profpatsch | Since that parallelizes well | 10:35:48 |
 Alyssa Ross | presumably we could just use "Rebase and Merge" in the UI, as long as git can tell what's going on? | 10:36:13 |
| profpatsch | ofc long feature branches might hate that, but they can merge master back into themselves | 10:36:24 |
| Alyssa Ross | makes the history (and especially bisects) easier too | 10:36:26 |
| @piegames:matrix.org | In reply to @qyliss:fairydust.space
presumably we could just use "Rebase and Merge" in the UI, as long as git can tell what's going on? This would be a deviation from us currently having merge commits everywhere. | 10:37:06 |
| infinisil | I think the github UI allows either rebasing or merging | 10:37:29 |
| Alyssa Ross | we don't currently have merge commits everywhere | 10:37:47 |
| infinisil | (by default rebase should be good, but as Profpatsch hints at, merging sometimes makes sense) | 10:37:55 |
| Alyssa Ross | we have a combination of all kinds of merges, based on what each committer thinks is best at the time | 10:38:12 |
| @piegames:matrix.org | What would be needed here IMO is rebase + merge with trivial merge commit. | 10:38:19 |
| Alyssa Ross | yeah we definitely want merge for e.g. staging-next | 10:38:21 |
| Alyssa Ross | what would be the point of the merge commit? | 10:38:35 |
| Alyssa Ross | also, if git can figure out renames on rebasing, I think it should be able to figure them out on merging too/ | 10:38:53 |
| @piegames:matrix.org | In reply to @qyliss:fairydust.space
what would be the point of the merge commit? To associate the individual commits with the PR | 10:39:06 |
| Alyssa Ross | that's not something we do today | 10:39:19 |
| Alyssa Ross | (across the board, at least) | 10:39:31 |
| infinisil | I'm still thinking just ripping the band-aid off with a single commit is probably easiest tbh :)
(and it is what's in the RFC fyi) | 10:39:36 |
| @piegames:matrix.org | hm, didn't know that | 10:39:39 |
| Alyssa Ross | you can look up the PR for a commit using the API / GitHub web anyway | 10:39:53 |
| Alyssa Ross | even if it was "Rebase and Merge"d | 10:39:58 |
