[RFC] pull-request:merge --squash should perform a rebase against base branch

[phansys]

pull-request:merge --squash should perform a rebase against base branch. I've used this command and it sometimes result in regressions for the base branch, since the squashed commits don't take into account the new commits (those that was created after creating the branch and before the merge) at the base branch.

[cordoval]

The rebase is not automatic, it should be done manual because in doing it there could be many conflicts.

[phansys]

In my experience, it must be automatic, otherwise the squashed commits sometimes reverts "silently" changes done at base branch after the PR creation but before its merge, since GH diff isn't showing that situation. So, in short words, a conflicted rebase could warn about issues between the histories that the command is trying to merge, and if it happens, the PR can be marked as not ready to be merged.

[cordoval]

what happens if it fails the rebase? or requires intervention? as often they do. It should just give a warning about, please make sure it is rebased, or the command only should work when it ensures the PR is rebased. Otherwise gives a note, i think git has a way to find out quickly.

[phansys]

Sure, that's the main idea, try to rebase automatically, abort the rebase if it has failed, warn the user about the situation and aborting the PR merge operation. But IMO it is much more painless than let merge PRs that reverts previous changes without any notice of that situation. Last week I found myself reverting PRs for this reason, that only happen with the use of --squash option in PRs with more than one commit as I mentioned before.

[cordoval]

what i mean is there is a way to check already if it is rebased or not via git. That is all, so we can just do an early check for that and abort or even prevent from running the command very early. I personally prefer to do things observing, so i don't often let gush do my rebases, this is in my experience, because i know there are always things I want to do besides skipping any errors. I honestly think the squash thing is only useful in FOSS context, but i do not use squash at work.

[phansys]

I tend to work the same way at proprietary / FOSS environments, but I've updated the PR taking into account your concerns.