On Thu, Jan 31, 2008 at 11:07:51PM +0000, Joey Hess wrote:
> Cyril Brulebois wrote:
> > since the second argument of commit would be a treeish, is that correct?
> 
> Yes.
> 
> (Although it looks like it won't be able to use the tarball created by
> git-archive directly, since git-archive does not provide a way to order
> the files correctly. Also since I don't want to rely on the git guys to
> not add or change special pax headers that would break reproducing the
> exact same tarball later. Oh well, it'll just be a bit slower.)

  Well, just better don't rely on git-archive indeed. FWIW you don't
really regenerate your orig.tar.gz often, so slow isn't that bad.

> > The interface looks quite nice, but how would data be stored? At the
> > moment, Pierre and I are using an “independent” pristine-tar branch
> > where successive deltas get added incrementally. Do you plan to use
> > something similar? And what if the deltas stored there get refreshed?
> > You just create another commit at the tip of the branch, and always use
> > the tip when you checkout?
> > 
> > But indeed, it would be *very* interesting to have pristine-tar handle
> > the bootstrap (generating an extra initial commit for a
> > “pristine-tar” branch).

> Yes, madcoder has helped me figure out how to initially populate and
> update the branch.

  Yay, I'm glad to have been of help.

> > That would be perfect. And for people having already stored their
> > tarballs with the current version (and which might not be able to update
> > the deltas with the above technique), I guess that it might make sense
> > to add a backward-compatibility flag which would create the empty
> > directories, so that tarballs gets regenerated if needed? Once they got
> > their original tarballs back, it's sufficient to commit them to get
> > appropriate delta, making the use of the compatibility flag obsolete
> > from this moment.
> 
> I'll consider adding such a backwards compatability flag, not sure yet.

  Bah, I don't think there are that many users of the technique yet, and
they probably are all in the recipients of that mail, or at one hop ;)

  In any case, this is very exciting stuff, be quick, I'm doing a talk
at FOSDEM (Feb 23-24) and I'd _LOVE_ being able to demonstrate such
nifty tools !

-- 
·O·  Pierre Habouzit
··O                                                [EMAIL PROTECTED]
OOO                                                http://www.madism.org

Attachment: pgpeCIFM3FOJa.pgp
Description: PGP signature

Reply via email to