Separate default push/pull?

If I am collaborating mostly with one other person, I typically want to pull from his publicly-readable repo and push to mine (on which I have write permission). Is there any way to set things up so “git pull” and “git push” without additional arguments will do this by default?

Thanks,

Dave Abrahams           Meet me at BoostCon: http://www.boostcon.com
BoostPro Computing
http://www.boostpro.com

Yes there is a way. I haven't used it myself but search this list and you'll find plenty of references.

(disclaimer: the following is what I think you can do based on some vague recollection and some man pages)

Taking a quick look at the git push --help you'll see the following snippet of configuration.

[remote "<name>"]
url = <url>
pushurl = <pushurl>
push = <refspec>
fetch = <refspec>

so I think if you just add the pushurl to your .git/config should do what you've asked

[remote "origin"]
fetch = +refs/heads/*:refs/remotes/origin/* url = git://git.kernel.org/pub/scm/git/git.git
pushurl = git://git.example.com/yourrepo.git

For your own sanity I suggest doing this by adding your repository as a separate remote.

e.g.

git remote add yourrepo git://git.example.com/yourrepo.git
git push yourrepo master:refs/heads/master # the first time git push yourrepo # subsequent times

There probably is a way to tell push to use something other than "origin" by default but I don't know/can't find it.

That's not quite what David wants, I think. That is a good recipe if you have two ways of accessing the same repo (usually git:// for reading and ssh for pushing). But if the two URLs actually point to _different_ repos, you will get some confusing results. For example, pushing to your private repo will update the tracking branches in refs/remotes/origin/* with values that do not match what is in the actual origin repository.

I don't think there is currently a way to do what he wants. You can set a default remote name by setting branch.*.remote, but that has two problems:

1. It is also used to determine the upstream remote for pulling and for calculating upstream tracking branches. So he probably wants to leave it set as origin.

2. It would have to be set manually for every branch.

I think what he would need is a "push.defaultRemote" config option, which universally overrides branch.*.remote for pushing.

-Peff

Or "branch.*.pushremote".

But does it really make sense to get changes from one place and send changes to somewhere completely unrelated?

That doesn't address my point 2, which is needing to set it up for every branch.

It depends on your workflow. For git.git, your kernel.org repository is my "origin", but I publish my state to a mirror for backup purposes (and I don't publish for others to view, but I could very well do that, too). I type "git push peff.net" and that is not too much trouble. Typing just "git push" would be slightly more convenient, though.

In a distributed setup, I don't think it is that uncommon to not want to push to the place you pull from. You are generally pulling and building on somebody else's work, so if there is no central repo, you will be pushing to somewhere that is not where you pulled it.

-Peff

You are probably right.

It still feels funny to see "git pull" and "git push" goes to different places, but as long as that is what the user explicitly configures, that's fine.

By the way, I am a little iffy on the configuration I suggested. Even though it matches David's workflow, it seems unintuitive to me that a "push.defaultremote" variable would override what's in "branch.*.remote".

-Peff

At Thu, 11 Feb 2010 16:49:26 -0800,

It's not unrelated; it's a publicly-readable clone of the source repo to which I have write permission. After pushing changes there I send a pull request to the owner of the source repo.

Dave Abrahams           Meet me at BoostCon: http://www.boostcon.com
BoostPro Computing
http://www.boostpro.com