On Wed, Feb 18, 2015 at 4:43 PM, Shawn Pearce <spearce@xxxxxxxxxxx> wrote: > On Wed, Feb 18, 2015 at 4:25 PM, Junio C Hamano <gitster@xxxxxxxxx> wrote: >> Checking out a random branch is absolutely the worst thing you can >> do. Personally, I would think that the best thing you can do is to >> educate your users not to clone from a void. Create some history >> that is worth sharing, and then push into an empty repository. > > Users are dumb. > > Consider a website like GitHub where you create a repository on a a > remote server. It is initialized as an empty repo. They now want to > clone that empty repo to work with it: > > git clone https://example.com/foo > cd foo > .... hack hack > git commit -a -m "Foo is awesome!" > git push > > This works from git-core servers because the ref advertisement tells > the client its an empty repo and its OK to checkout HEAD from nothing > and its all fine. > > This fails from a JGit server because the ref advertisement for an > empty repository differs and disagrees with the format used by > git-core. The result is a confused client spewing an odd checkout > message. The difference in JGit is intentional. Our cross data center consistency system at $DAY_JOB depends on being able to fetch from one empty repository to another empty repository using the want any object capability, and side-band-64k. The JGit client and server code that speaks the wire protocol needs to send and see the capability lines for that to work, but with no refs there are no lines to send. So we send capabilities^{} like receive-pack does. But git-core hates all over that during clone of an empty repo. -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html