On Tue, May 01, 2007 at 10:09:18AM +0200, u+codalist-p4pg_at_chalmers.se wrote:
> On Mon, Apr 30, 2007 at 10:53:55PM -0400, Jan Harkes wrote:
> > There are actually some changes in git that are not as friendly for a
> > shared repository, however most of those are optional or don't apply to
> > 'bare' repositories. Specifically the newly introduced packed refs
> > format (which is optional) and the fact that pruning a repository tries
> > to remove the individual object directories.
> 
> Which means that the idea of shared repositories has not crossed the minds
> of git developers :(

Well, I guess there are different types of sharing. And one typical
model which works well is to have a repository that can be modified by
anyone in some group. What I'm trying to achieve is a finer grained
access model which is possible because of Coda's directory ACls where
developers only add new commits and update their own branch references.
The ACLs protect the repository so that you normally cannot remove or
change any already committed data or modify branch refs of other
developers.

Once you pack all references into a single file, the directory acls (or
even file acls) won't be able to enforce the required isolation. The
same is true when the individual objects in the repository are packed
into one file. However, once an object is written it will never be
modified, only read. So all developers only need read permission on the
object-pack file and they never have to update only some parts of the
file. And maybe the same count for packed references, I just haven't
really looked at it yet.

Jan