On 07/29, Jeff King wrote: > I've noticed for the past couple of weeks that some of my fetches don't > seem to actually update refs, but a follow-up fetch will. I finally > managed to catch it in the act and track it down. It bisects to your > 989b8c4452 (fetch-pack: put shallow info in output parameter, > 2018-06-27). > > A reproduction recipe is below. I can't imagine why this repo in > particular triggers it, but it was the one where I initially saw the > problem (and doing a tiny reproduction does not seem to work). I'm > guessing it has something to do with the refs, since the main change in > the offending commit is that we recompute the refmap. I've noticed this behavior sporadically as well, though I've never been able to reliably reproduce it, so thanks for creating a reproduction recipe. I suspected that it had to do with the ref-in-want series so thanks for tracking that down too. We'll take a look. > > -- >8 -- > # clone the repo as it is today > git clone https://github.com/cmcaine/tridactyl.git > cd tridactyl > > # roll back the refs so that there is something to fetch > for i in refs/heads/master refs/remotes/origin/master; do > git update-ref $i $i^ > done > > # and delete the now-unreferenced objects, pretending we are an earlier > # clone that had not yet fetched > rm -rf .git/logs > git repack -ad > > # now fetch; this will get the objects but fail to update refs > git fetch > > # and fetching again will actually update the refs > git fetch > -- 8< -- > > -Peff -- Brandon Williams
