RFC: Junk in Fedora Extras

[Hans de Goede]

Michael Schwendt wrote:
> So far, we have been running "repomanage" on the Fedora Extras repository
> trees to get rid of old package releases automatically. In development, we
> keep only the latest release, for older trees we keep at most two
> releases.
> 
> However, occasionally I've run into obsolete/orphaned sub-packages where
> the packager changed the BuildArch or deleted a sub-package without asking
> the repository staff to delete old package releases manually. It can also
> happen that the package removal requests, which are processed by humans,
> lead to incomplete removal of packages, when e.g. "debuginfo" packages are
> not removed or version-information was misleading.
> 
> The resulting obsolete package releases live in the repository forever, as
> repomanage will never decide to delete them.
> 
> As a solution, I've looked into porting my old Perl rpm-expire script to
> Python -- I've lost the script in the single place where it was running
> and have only recovered an early prototype I'm not satisfactory with.
> 
> The technique is based on this:
> 
>  * We build from src.rpm packages.
> 
>  * For all binary rpms there must be a corresponding src.rpm.
> 
>  * The src.rpm name is in the %{sourcerpm} header of a binary rpm.
> 
>  * If the src.rpm for a binary rpm in the repository does not exist,
>    where does the binary rpm come from? --> We delete it!
> 
>  * Instead of expiring arbitrary packages based on "number of pkgs found",
>    we expire only src.rpms and clean up the binary rpms based on whether
>    their src.rpms still exist.
> 
> Running a quick hack of this on the current FE repositories, it creates a
> long list of "lost souls" as appended below. I've checked a few already
> and so far could not find any false positives. Of course, letting such a
> script remove these packages automatically actually, will require a few
> good sanity checks, so run-time errors don't empty the entire
> repository. ;)
> 

Sounds like a good plan, but what about huge disttag-less noarch packages which get copied over from dist to dist?

Regards,

Hans