[Gnu-arch-users] Re: archive storage format comments on the size

[Pau Aliagas]

On Tue, 30 Sep 2003, Karel Gardas wrote:

> On Tue, 30 Sep 2003, Andrea Arcangeli wrote:
> 
> > The superpatchset makes sense only for the "main" repository for the
> > very old patchsets that you're not going to merge against anylonger, so
> > it's acceptable to ask the user who's merging that he needs to split the
> > superpatchset first if he really wants to merge patchsets in the middle
> > of the superpatchset.
> 
> I'm just curious, but isn't cached revision designed exactly to solve this
> issue? i.e. loong get with a tons of applied patches...

I was about to send the exact same response. Cache revisions every 100-200 
patches and you are done. See "tla cacherev -H". You can even delete the 
old ones you won't need anytime soon to save space, as it's a tar.gz of 
the complete revision.

This can easyly be automated with a ~/.arch-params/hook script that adds
and deletes revisions transparently (contact me privately if you need it).
Just decide your needs, modify the script and gets will be very fast.

Commits are already fast since the inode optimizations.

It looks like this woud solve your initial speed requirements flawlessly. 
Forget by now of the space trade-offs of the archive format, there are 
open discussion to improve it.
I'll file a bug in savannah to ask for the tla get --hardlinks.

> Also, setup revision library and your get will be as fast as possible...

The problem is that it can become very big with kernel archives. Maybe 
it's better to cache revisions.

Pau