Re: [Savannah-hackers-public] Re: lilypond development moved

[Johannes Schindelin]

Hi,

On Mon, 13 Nov 2006, Han-Wen Nienhuys wrote:

> Sylvain Beucler escreveu:
> > On Sun, Nov 12, 2006 at 01:17:43PM +0100, Han-Wen Nienhuys wrote:
> > > Werner LEMBERG escreveu:
> > > > > However, in collaboration with the savannah hackers, I've now
> > > > > successfully set up a git repository at savannah.  Check it out at
> > > > > 
> > > > >   http://git.sv.gnu.org/gitweb/
> > > > It seems that this host is quite loaded.  The remote part of doing a
> > > > `git-clone' took ages...
> > > I think this machine is just savannah, which is loaded indeed.
> > 
> > It's probably due to the fact we have a 1.3GB/180000obj repository
> > that was never git-repack'ed. I just did so, how about now? :)
> > 
> > If you know about a commit hook that repacks safely (apparently, every
> > 4-5MB is recommended), I'll be happy to install it.
> 
> I thought GIT was inherently safe. Johannes?

As mentioned elsewhere: repacking is safe; there were some concerns with 
concurrent repacks, but I think they are already squashed in the current 
version.

But please keep in mind: if anybody is downloading via HTTP, and you 
repack with "-d", i.e. delete the files when they are obsoleted by a pack, 
it can be necessary to load the whole pack, in spite of needing only a 
fraction of it. Hopefully this will be fixed in a future version, but 
don't hold your breath for it.

> (Why not use a cron job?)

The most important thing is the initial packing. The whole repository _is_ 
big, but the incremental updates are relatively small.

Note that with recent versions of git (IIRC >= 1.4) repacking is not as 
expensive as it sounds: if there is already a pack, then the implicit 
information (which pairs of objects are most similar) is reused.

That is also the reason why a git-clone on a completely unpacked repo 
takes ages, but that of a mostly packed does not: the server side of 
git-clone creates a pack.

Ciao,
Dscho