On Sun, Apr 09, 2017 at 10:38:42AM +0200, Stefan Haller wrote: > Jeff King <p...@peff.net> wrote: > > > > It might be possible to generate these lease tags prior to operations > > > which modify history and then maybe having a way to list them so you > > > can select which one you meant when you try to use force-with-lease.. > > > > So yeah, I think that is the more interesting direction. I hadn't > > considered resolving the multiple-operation ambiguity at push time. But > > I guess it would be something like "you did a rebase on sha1 X at time > > T, and then one on Y at time T+N", and you pick which one you're > > expecting. > > I think it's wrong to think about these leases as something that you > take before you start a rewindy operation. That's the wrong time to take > the lease; by that time, the remote tracking branch may already contain > new things that you haven't seen yet, so using that as a lease at that > time will overwrite those things later. You have to take the lease at a > time where you know that your local branch and the remote tracking > branch are up to date with each other, which is after pull and push. And > if you do that, there's no multiple-operation ambiguity to deal with at > all.
OK. I was assuming that you'd have just integrated before starting such a rebase, but I guess that doesn't have to be the case. I agree that probably makes the multiple-operation stuff go away, which is nice. It does raise the question of when the integration point happens, and how we handle alternate paths through which commits may land in a local branch (e.g., if both you and upstream do a ff-merge of a particular branch). I think that would probably just end up with extra failures though (so erring on the side of caution). As long as those aren't too common (and this check would kick in only when you're doing --force-with-lease in the first place, so presumably not often), I think it'd be OK. -Peff
