Git repository mesh?

Hi,

I work on a few machines so I have repositories on all of them. One repository can pull from or push to any other repositories (in case of push, it pushes on remote branches). I avoid a central repository because it's quite inconvenient when you just need to push some changes to a machine, you have to push it to the central repository then pull from that (and if the central repository is on WAN, double inconvenient). Maybe this model is just plain wrong, but it'd be fun to see if Git can work with this model.

The first thing that annoys me is remote repository management. Everytime I add a new repository to the mesh, I need to update .config of all repositories. Putting part of $GIT_DIR/config into working directory may help (though there will be security concerns, perhaps only remote.* should be allowed in in-tree config file). Next thing is sometimes I want to see how many commits ahead/behind other repositories, from a repository. "git status" does not help much. Never used Mercurial but the idea of "git in" and "git out" to see how many commits ahead/behind would be nice.

Has anyone done something like this? Any suggestions? -- Duy

Nguyen Thai Ngoc Duy <pclouds@gmail.com> writes:

In Git, you'd run "git fetch" to get everything locally, and then

git log ..origin/master => what origin/master has that you don't git log origin/master.. => what you have that origin/master doesn't

(and you can define aliases for that)

-- Matthieu Moy http://www-verimag.imag.fr/~moy/

On Wed, Nov 25, 2009 at 2:55 PM, Matthieu Moy <Matthieu.Moy@grenoble-inp.fr> wrote:

But I have many origins (any repository could be origin). "git status" also have this kind of information, but only for "origin".

Nguyen Thai Ngoc Duy <pclouds@gmail.com> writes:

Then, fetch from all of them and:

  git log ^HEAD repo1/master repo2/master repo3/master

-- Matthieu Moy http://www-verimag.imag.fr/~moy/

On 11/25/09, Matthieu Moy <Matthieu.Moy@grenoble-inp.fr> wrote:

Very nice. Thanks I did not know about "^HEAD". Now I need to think of how to avoid typing all "repo*/master" stuff..

Nguyen Thai Ngoc Duy <pclouds@gmail.com> writes:

Hmmm... shouldn't 'repo1' expand to 'repo1/HEAD', which is 'repo1/master'? I mean shouldn't simply

  git log ^HEAD repo1 repo2 repo3

work?

Also, git aliases (so you would be able to just say "git in").

Nguyen Thai Ngoc Duy <pclouds@gmail.com> writes:

Read "not HEAD", this means "remove the ancestry of HEAD in the output".

See "man git-rev-parse" for details.

Nguyen Thai Ngoc Duy <pclouds <at> gmail.com> writes:

If I remember correctly, there was in the past the idea of allowing some limited shell variable substitution in the config file.

This would allow to have a single config for all the machines with remote entries. Something like

[remote "host1"]
        url = ssh://host1/path
        fetch = +refs/heads/*:refs/remotes/host1/*
        push = +refs/heads/*:refs/remotes/$hostname/*
[remote "host2"]
        url = ssh://host2/path
        fetch = +refs/heads/*:refs/remotes/host2/*
        push = +refs/heads/*:refs/remotes/$hostname/*
...

In other terms, admitting that all the candidate machines are known at the very start, you could start a repo on any of the host, placing in the config file all the possible hosts. Moreover, not needing to have a different config for each host may make it easier to manage (propagate by simple scp) or even to version control.

To the best of my knowledge the idea was not implemented in the end, but I think that a patch for variable substitution in the config file might still be found on the ML.

Sergio

On Wed, Nov 25, 2009 at 05:07, Nguyen Thai Ngoc Duy <pclouds@gmail.com> wrote:

You can always use fetch (and pull) without specifying a remote:

  $ git fetch host:/path/repo some-branch
  $ git log ..FETCH_HEAD # git merge FETCH_HEAD

Maybe fetch should be extended to record this operation in .git/config?

  $ git fetch ---save=host-some host:/path/repo some-branch some-branch
  $ git remote -v
  ...
  host-some	host:/path/repo (fetch)
  $