After all this absolutely enjoyable discussion of repodata format I
figured I'd mention something that would help users tremendously to
reducing the amount of data they might have to download.
right now when a repository is created for use it's just made on ALL
packages in the install tree.
so you end up with srpms and debuginfo rpms in there.
So you get 2623 pkgs to read through for fc3 base
but if you just read through binary rpms you only have 1653 pkgs to read
through. A pretty serious savings, almost 1000 packages.
Now, yum prunes out the packages it can't use right away, but it is
still having to read through all that data.
so why not make the repo like this:
createrepo -x *.src.rpm -x *.debuginfo* -g Fedora/base/comps.xml .
and then make another srpm repository in the SRPMS dir all on its own.
Ditto with debuginfo.
but then we should add disabled-by-default srpm and debuginfo repos to
each .repo file in yum.repos.d
fedora.repo:
[base]
....
enabled=1
[base-srpm]
...
enabled=0
[base-debug]
....
enabled=0
then the user, if they need the debuginfo rpms, for example, can run:
yum --enablerepo=base-debug install foo-debuginfo
We should do the same for extras and updates, updates-testing repos, and
rawhide too.
I think, on the whole, we'd see a huge increase in speed for going
through repos b/c we simply would have less data for 95% of the users to
go through.
-sv