[lkml]   [2005]   [Sep]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectCurrent state of GIT fetch/pull clients
After Linus suggested to use objects/info/alternates to point at
his linux-2.6 tree so that the maintainer trees can borrow from
it, there may be a bit of confusion. Here is my attempt to
clarify the current state of affairs.

If you are a subsystem maintainer with an account $u at and have your repository $tree derived from
Linus' linux-2.6.git reopsitory, it would be at:


Of course, you may have more than one such $tree. The
suggestion by Linus was to do (please do not do this yet -- that
is what this message is about):

$ cd /pub/scm/linux/kernel/git/$u/$tree
$ cat /pub/scm/linux/kernel/git/torvalds/linux-2.6/objects \
$ GIT_DIR=. git prune

What this does is:

* A process on and its mirrors running in
your repository (i.e. /pub/scm/linux/kernel/git/$u/$tree)
can read objects not found in your repository from Linus'
repository. This means that $u/$tree/objects/?? would
contain only your changes and practically 99% of the
objects are coming from Linus' repository for such
process. You do not have to have any objects Linus'
repository has in your repository.

Sounds very nice, doesn't it?

However, there are currently certain issues around almost all
git clients, and Cogito cg-pull shares this problem.

Neither http nor rsync transports know about the 'alternates'
mechanism yet, so if a downloader does:

$ git pull$u/$tree
$ git pull rsync://$u/$tree

unless the downloader has already fetched from Linus'
repository, this will not work.

* In the case of http transport, it would start from your branch
head, walking commits backwards, and as soon as it hits a commit
(or a tree/blob) that your repository is borrowing from Linus.

* In the case of rsync transport, it would slurp all objects
your repository has, but does not get objects from Linus'
repository. Also, rsync will overwrite the
objects/info/alternates file the downloader has in his
repository with what you have in your repository, which is
not what we want.

The only transport that works is what Linus uses himself. If
the downloader has an account on

$ git pull$u/$tree

would work, because this transport runs git in your repository
on the side, and that knows how to use
objects/info/alternates file you set up as Linus suggested.
Another transport that _could_ work is git-daemon:

$ git pull git://$u/$tree

but unfortunately servers do not run git-daemon yet.

What this means is that using objects/info/alternates mechanism
in your repository is a bit premature as things currently stand,
if you intend your repository to be used by the general public.

Daniel started working on teaching http transport to be able to
read from objects/info/alternates last night, and I am expecting
that would be proven stable and usable sometime next week at the
latest. HPA is helping us in the discussion to whip git-daemon
into a shape usable on to enable it there. I'll be
fixing rsync transport sometime over this weekend.

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-09-16 01:55    [from the cache]
©2003-2011 Jasper Spaans