On Thu, Aug 01, 2019 at 02:35:43AM +0800, 16657101987@xxxxxxx wrote: > > So I actually think the best path forward is just always refreshing when > > we take the lock, something like: > > > > Ultimately the best solution there is to move to a better format (like > > the reftables proposal). > > I do not know if we could get the new reftables in the next few versions, > So I commit the changes as you suggested, which is also the same as > another way I metioned in `PATCH v1`: > > **force `update-ref -d` to update the snapshot before rewrite packed-refs.** > > But if the reftables is comeing soon, please just ignore my PATCH :) I'm undecided on this. I think reftables are still a while off, and even once they are here, many people will still be using the older format. So it makes sense to still apply fixes to the old code. What I wonder, though, is whether always refreshing will cause a noticeable performance impact (and that's why I was so slow in responding -- I had hoped to try to come up with some numbers, but I just hadn't gotten around to it). My gut says it's _probably_ not an issue, but it would be nice to have some data to back it up. > **And thank a lot for your reply, it's great to me, because it's my first > PATCh to git myself :)** You're welcome. Thanks for diagnosing a rather tricky case. :) -Peff
