Re: Pingback/Trackback for Git?

From: Thomas Perl <th.perl@gmail.com>
Date: Jul 4, 2007, 09:59 UTC
Message-ID: <372f54da0707040259kd034b79ofa41b46611052bee@mail.gmail.com>
In-Reply-To: <7vbqeslpnz.fsf@assigned-by-dhcp.cox.net>

Hello, Junio!

2007/7/4, Junio C Hamano <gitster@pobox.com>:

Show 8 quoted lines

> It was unclear to me who does what and when in the overall
> picture you envision after reading that page.  You seem to
> assume there is a prominent ping/trackback repository that
> everybody who is interested looks at.  Who updates it?  When?
> With what?  How does the namespace inside that repository
> managed, if managed in any way?  Given a project name, how does
> one learn which pick/grackback repository (and perhaps which
> branch) to go to to learn the current status about it?

Every repository keeps track of the pingbacks it receives. The simplest way of storing all pingbacks would be a simple text file that lists one repository per line in a URL, i.e.

http://git.server.com/repos/game.git git://mirrors.git-hosting.com/ga/game.git http://johnny.org/my-projects/game.git http://universityx.edu/~bob/src/game.git

If a user clones a repository, it can copy this trackback list. Git can also optionally use this list to fetch objects from multiple servers to increase performance (imagine an upstream git:// repository with a forked http:// repository that has packs, the client could use the git:// repository for fetching as much single object as possible and only using the forked repository for all objects not yet found in the upstream repository).

Sending a pingback should append the pinging repository URL to the pinged repository's list (see above). Before appending the URL, the pinged repository should check if the URL exists and if there is a Git repository at the given URL and if the pinging repository is based on the pinged repository.

There is not a single repository with pingbacks/trackbacks, but every repository has its own list of "related" repositories. So when I'm interested in the repository from project at URL "X", I can fetch its repository or only its mirror list and get a list of URLs that also carry (derived) versions of the project at URL "X".

Of course, your idea with having a centralized repository (that only stores URLs) would be nice when an upstream Git repository dies to find some other repository where the data is still available:

1) Clone repository from URL "X"
2) 404 -- not found
3) Get list of mirrors from URL repository
4) Download repository from other URL

But that defeats some gains from a truly distributed vcs, so just keeping track of a list of URLs that cloned from this repository should be enough.