On Thu, Nov 29, 2007 at 10:03:04AM -0500, seth vidal wrote:
On Thu, 2007-11-29 at 16:03 +0100, Florian La Roche wrote:
> On Wed, Nov 28, 2007 at 08:57:09AM -0500, seth vidal wrote:
> >
> > On Wed, 2007-11-28 at 14:54 +0100, Florian La Roche wrote:
> > > On Wed, Nov 28, 2007 at 08:39:21AM -0500, seth vidal wrote:
> > > >
> > > > On Wed, 2007-11-28 at 14:25 +0100, Florian La Roche wrote:
> > > >
> > > > >
> > > > > Hello Seth,
> > > > >
> > > > > below is a patch to sort the output, otherwise createrepo
> > > > > depends on the order of the local filesystem and how rpms
> > > > > are stored there.
> > > >
> > > > Why does the order matter? Doesn't this just take more time to do
the
> > > > sort?
> > >
> > > It makes the metadata predictable and not dependent on the fs of the
> > > server.
> >
> > Again, how does the predictability of order the filelists and deps/reqs
> > are put into the metadata impact the functionality?
>
>
> Hello Seth,
>
> This is about stability of the data. Currently you cannot easily
> run createrepo and quickly check the output against the existing data.
yes, you can - use the metadata diff tool in createrepo's source -
dmd.py
Hello Seth,
dmd.py is nice, this wasn't part of earlier releases. I still think
putting this into the format you write out to disk is a plus.
>
> >
> > > The additional time should not matter at all here.
> >
> > There are as many as 11000 pkgs in some of the repositories we create.
> > How does it NOT matter if it takes extra time to sort each of them?
>
>
> I am not aware of such limits for python .sort(). 11000 is not a huge
> amount of data compared to other computations done in createrepo.
>
except that your code has us sorting all the files in each of the 11000
pkgs.
Same data stability problem as above with filename sorting, to not
depend on python ordering.
also - why did you cc fedora-devel on this thread?
To open up the discussion, not sure how many people are on
rpm-metadata.
regards,
Florian La Roche