Refresh git configuration before looking for the tracking branch.

[alanjrogers]

After publishing a new branch I noticed that -trackingBranchWithError: success: was still returning a nil, after a bit of investigation it turns out libgit2 was using a stale version of the repo's config when it was looking up the branch.*.remote and branch.*.merge in git_branch_upstream_name.

So I've added a [configuration refresh:] call before we call git_branch_upstream.

[carlosmn]

This shouldn't be needed. If it does fix it, it would point out a bug in libgit2's snapshotting.

[alanjrogers]

@carlosmn it is needed. In my tests, the config is stale after first publishing a local branch (via git.git. I'll look into the libgit2 confit snapshotting code tomorrow.

[jspahrsummers]

@alanjrogers This looks good to me. Do you want me to wait to merge it until you've had a chance to look at libgit2?

[alanjrogers]

@jspahrsummers yup, thanks.

[alanjrogers]

Ok, I've attempted to make this fix in libgit2 instead. This PR is now dependent on libgit2/libgit2#2620.

[joshaber]

🤘

[alanjrogers]

Now depends on libgit2/libgit2#2622 .

[alanjrogers]

Marking this as [WIP] until we've resolved the issues we're encountering using this branch in GHfM.

[alanjrogers]

libgit2 master seems to resolve the issue, once the libgit2 dependency is merged I'll bump the submodule to libgit2 master and we should be good to go.

[alanjrogers]

This should be good to go. The tests are failing because the CI machine is now running Xcode 6.1 😢

[jspahrsummers]

I don't think this __block is necessary, since it'll only be captured in @onExit after receiving a value.

[alanjrogers]

Compiler says no. Looks like it captures a const reference in the block. Although I could cast that error away, but at that point I'd prefer to use the __block qualifier.

[jspahrsummers]

Ah, right. 👍

[jspahrsummers]