Re: interactive rebase not rebasing
- From
Andreas Ericsson <ae@op5.se>
- Date
- Oct 1, 2008, 07:50 UTC
- Message-ID
- <48E32BD4.1050107@op5.se>
- In-Reply-To
- <20081001010306.01cc34eb.stephen@exigencecorp.com>
Stephen Haberman wrote:
Show 40 quoted lines
>>> # A --C------ <-- origin/stable >>> # \ | \ >>> # B -- D -- E -- F <-- origin/topic2 >>> # \| >>> # g -- h <-- topic2 >>> >>> Nothing has changed. g & h haven't moved...I can keep executing this >>> operation and the commits never make it on top of origin/topic2's F. >>> >>> Frustratingly, if I run non-interactive rebase, it works perfectly. >> I can imagine. Since you don't want to preserve the merges in this >> case, you shouldn't be using the -p flag. > > No, I do want to preserve most merges. This "most" qualification is > because the merge "g", if rebased, would have been a no-op, so `rebase > -i -p` correctly kept it out of the TODO file. > > Which is cool, except that later on, when rewriting the other TODO > commits, some of which were children of "g", it did not remember that > "g" had gone away, so did nothing to take "g" out of the rewritten > children's parent list. > >> In fact, for this particular scenario (assuming "h" is really the only >> commit on topic2), you probably want to just cherry-pick that commit >> into origin/topic2: >> >> git checkout topic2 >> git reset --hard origin/topic2 >> git cherry-pick ORIG_HEAD > > Agreed. This makes a lot of sense for me, who has been hacking around in > git-rebase--interactive fixing things, but I'd really like the other > people on my team to just have to run `git rebase -i -p`. > >> I don't think you can have a single command that does all the things >> you want, because the possible differences in input makes it very >> nearly impossible to always do "the right thing". > > Ah, you are too pessimistic. :-) >
Perhaps, although I think you're being overly optimistic if you think rebase can do all of this intelligently while still retaining clear semantics. I'd start with writing a separate tool for it, probably based on git sequencer. When that works out ok, get it to do what rebase does today and then incorporate the new tool as an option to rebase and get ready to answer complex questions for the cases where it doesn't do what the user wanted it to do.
Git is stupid very, very much by design. Linus constantly says that he prefers tools that he can figure out why they did something stupid over tools that try really hard to get it right, and I agree with him 100%, as do most of the core contributors (insofar as I've understood it). What you're suggesting is that a git command should try hard to dwim a lot of complexity about choosing tools away, and that goes right against the KISS principle which serves git so well. I'm not saying you shouldn't do it. Merely that you should think hard about it and then make sure it doesn't break anything people are already doing today with the current toolset.
-- Andreas Ericsson andreas.ericsson@op5.se OP5 AB www.op5.se Tel: +46 8-230225 Fax: +46 8-230231