Hi George, On Sat, 21 Mar 2020, George Spelvin wrote: > On Sat, Mar 21, 2020 at 11:51:10AM +0100, Johannes Schindelin wrote: > > My biggest caveat is that I had to force-exit the rebase at some stage > > due to reasons I only vaguely remember. It had something to do with the > > replacement cache not being updated when an `exec` is executed that adds a > > replacement object via `git replace` [*1*]. This issue might have > > _nothing_ to do with nested rebases, but as I said, my recollection is > > vague. > > This is the sort of internal implementation gotcha I worry about. There's plenty more gotchas like that ;-) > > There are a couple more concerns, of course, such as: what to do if the > > user deletes the entire todo list (which is traditionally the only way to > > abort a rebase)? My gut feeling is that it should go back to the > > _previous_ version of the todo list. > > My assumption has been that, for simplicity, there would only be one > commit in progress, and aborting it aborts everything. But that does not necessarily make sense. Imagine that you rebase the latest three commits, interactively. Then a merge conflict in the third makes you realize that the first commit is no longer needed. Enter the nested rebase. You manually re-schedule the failed `pick` via `git rebase --edit-todo` and then run the nested rebase: `git reset --hard && git rebase -i --nested HEAD~2`. Except that you made a typo and said `HEAD~3` instead of `HEAD~2`. You delete the entire todo list to get a chance to restart the nested rebase. But now the entire rebase gets aborted? If that would happen to me, I would unleash a whole slew of rarely used words in the vague direction of whoever implemented the nested rebase feature... > > Another big concern is what to do about `rebase.missingCommitsCheck`: with > > nested rebases, this will get increasingly tricky. Like, imagine you are > > rebasing 5 commits, the third of them results in merge conflicts, you > > realize that it is obsolete and so is now the first, already rebased > > commit. You do a nested rebase of the latest two commits to drop them, but > > they don't have their original commit hashes any longer. So it gets a bit > > finicky to keep track of what commit has been dropped on purpose and what > > was forgotten to pick instead. > > This doesn't *seem* difficult, but I don't know how the current mechanism > works. The implementation details do not matter at this stage. You have to get the design of the feature right. I am unfamiliar with the design of the feature as it is implemented right now, but I imagine that it needs to be adjusted for nested rebases, as we no longer have a single original todo list to roll back to. > It just checks that all commits that were on the to-do list when the > editor started are still listed (possibly marked "drop") when it exits. > > When you do a nested commit, additional commits are prepended to the to-do > list, you're dropped into the editor, and the same check is made when the > editor returns. > > If rebase.missingCommitsCheck = error is triggered, you end up with the > <upstream> tree state with nothing applied and may either --continue to > ignore the error or --edit-todo to put back the missing commits. > > > Let me give an example. Suppose I have commits a-b-c-d-e, and start > with rebase -i b. > > My to-do list starts out as c-d-e, but suppose I decide to cherry-pick > z and add it to the list, so it's now z-c-d-e. > > So I start rebasing, and it turns out that d creates a merge conflict > because I forgot a prerequisite patch y. And I really want y and z before > b, anyway. > > So the tree state is currently a-b-z'-c', with d in progress and e yet to > do. In my simple model, I have to resolve and commit d, so the tree > state is a-b-z'-c'-d'. Then I can rebase -i a, and am presented with > a to-do list of b-z'-c'-d'-e. > > If I delete any of those five commits, then rebase.missingCommitsCheck > will trigger. If I put y in the list, save it, then change my mind and > --edit-todo and delete y, it will also trigger. As I said, I am not using that feature myself, so I do not even know what "trigger" means in this context. It might totally be okay to use the existing code as-is in the context of a nested rebase. That remains to be verified, though, I think. > Now, it sould be nice if there were a way to say "screw this mess; just > check out HEAD and put d back on the to-do list", but that could > create a bit of a mess if I've split d and already committed half of > it. If I used that after doung that, I'd have a to-do list of > b-z'-c'-d'-d-e which might be awkward, but maybe it wouldn't be too bad. There is all kind of opportunity for messes, all right. Ciao, Johannes
