Append a commit or rewrite all commits ?

[ewjoachim]

Append a commit or rewrite all commits ?

[webknjaz]

Maybe inject black fixup commits in between user-created ones? This way the user could attempt to squash related things fairly easily if they want it...

P.S. I also do not like wholesale commits like this

[ewjoachim]

I like this idea, this would allow a clear PR where you can check the diff, and then the repo owner would be resposible for the (madness of) rewriting the whole history.

But is there a way to fixup that wouldn't create rebase conflicts ? Not sure.

I was already surprised that git filter-branch would manage to blacken all commits without getting lost in applying the diffs. I don't know enough of the git internal specifics, but I guess git filter-branch is able to work on the state of the tree at each commit, whereas my experience with rebase is that it really much more works on the diff each commit introduces. But, again, I'm no git expert.

[webknjaz]

You can try injecting two commits, not one. Fixup + its revert. This would probably simplify the interactive rebase part. Then, you could squash those reverts with the following commits...

But yes, you are right. It's not exactly trivial. I'm currently helping out with packaging of pygit2 wheels so hopefully it will be possible to get those easily installable soon. It's interface should simplify plumbing level Git interactions...

[ewjoachim]

Hmm, it's interesting but the --autosquash flag on rebase only allows moving commits to right after existing ones, and the 2 only commit merge actions, fixup and squash, only merge with the previous commit.

I'll need an example:
there's A ---- B, you're blackouting the repo, so you'll get 3 commits: A-Black, A-Black-Revert and B-Black : A --- A-Black --- A-Black-Revert --- B --- B-Black

There's 2 possible todo here:

pick A
fixup A-Black
reword A-Black-Revert
fixup B
fixup B-Black

but you'd have to manually enter B's commit message at the reword step, or

pick A
fixup A-Black
pick A-Black-Revert
squash B
fixup B-Black

But, again, you'd have to enter B's commit message at the squash step.

And non of those solutions can be achieved using fixup! or squash! commits only wich kinda defeats the purpose...

Except if I'm mistaken ?