On Wed, Sep 29, 2010 at 04:42:56AM -0500, Marcus Hüwe wrote:
On 2010-09-28 11:31:38 -0500, Michael E Brown wrote:
On Tue, Sep 28, 2010 at 11:14:12AM -0500, Robert Xu wrote:
On Tue, Sep 28, 2010 at 06:24, Marcus Hüwe <suse-tux@gmx.de> wrote:
So would this let us get RPMs that are continually updated, like from a repo such as Fedora Updates?
Absolutely not. In fact, this seriously breaks and wont work at all due to some limitations of DoD. The repo you pull from has to be completely static. The build service never re-downloads the repo metadata, and caches a lot of info. Can you summarize which features do you miss? If I understand you correctly the obs should fetch the required metadata on its own?
Correct. Today you have to manually download primary.xml.gz and uncompress even though the information exists where OBS could/should get this itself. It should download repomd.xml and parse that to get the url for primary.xml.gz, then download/uncompress.
In order to support non static metadata (without "hacks" like removing :full* and triggering a scanrepo event) we would just need to compare the old metadata with the new metadata and remove some entries from :full.solv (and their correspondent binaries from the :full/ dir). I think this shouldn't be too hard to implement:)
Well, if you put it that way, maybe. But you still need to teach OBS to download the metadata in the first place. Pat Dignan wrote up a wiki page with the concept on it, which I cant lookup right now as I'm on my way to a meeting... -- Michael