RPM Support

Server is C72, upgraded 2.4 -> 2.6 -> 2.8rc9 -> 2.8, but not in production yet.

The public Fedora 23 mirrors use lzma compression for group_gz in repomod, i.e. *-comps.xml.xz.

When pulp syncs a repo, it reads comps from somewhere (it looks like the .xz should work), and it output the uncompressed *-comps.xml file as the type="group" data, and ignores group_gz altogether in the published repomd.xml. Unfortunately, in Fedora 23, Yum is deprecated, and DNF/libcomps does not handle uncompressed comps data. So Pulp won't publish compressed + DNF won't read uncompressed = no groups for F23 dnf users.

One thing I notice is that a mirror of Adobe's x86_64 repo keeps the type="group_gz" data, albeit as comps.xml.gz (which just has an empty comps tag). It goes a step further, because while trying to hack around the DNF limitation I changed the "comps.xml" in the distributor python to use "comps.xml.gz", which humorously caused the adobe repo to start using "comps.xml.gz.gz". I've also tried the other obvious naive approach: adding various endswith('.xz') alongside the endswith('.gz').

Mostly, this has served to confirm that I don't know how this stuff is supposed to be working.

Anyways, if someone could clue me in, that would be awesome---my further naive thought is that it should simply always output group_gz, regardless of whether upstream includes it or not.