Crash resilience on push

[riastradh]

While experimenting with #335, I discovered that after a push is accepted by the remote hg repo (adding new changesets), git cinnabar may not have yet updated refs/cinnabar/*.

Does that pose any issues for crash resilience? Can anything be lost by a successful push without updating refs/cinnabar/*?

(If not, one wonders why refs/cinnabar/* needs to exist at all! But I'm not familiar with the data structures stored here; maybe there's an obvious answer about refs/cinnabar/* being a cache that's necessary for performance in large repositories but not a big deal for a few changesets at a time.)

[glandium]

git-cinnabar is idempotent, so pushing the same things in the same way will do the same thing. You're not really losing anything except time if the metadata behind refs/cinnabar is not updated.

[riastradh]

Thanks! Will anything be potentially lost or confused if, instead of pushing again, I fetch from the hg repo -- getting the hg view of the new commit -- and then try to push again?

[glandium]

If the new commits come from git-cinnabar in the first place, there isn't going to be a problem.