[Yum-devel] Fedora update metadata

[Jeremy Katz]

Bringing an old thread back from the dead given we're actually going to
need something here for pup (or at least, it'll make things work
nicer :)

On Mon, 2005-06-20 at 16:45 -0400, seth vidal wrote:
> Like I responded to Matt, doing this via a cgi (xml-rpc) that createrepo
> can access for the right information is probably the most simple way
> forward.

Sure, although how the data gets pulled in by createrepo is less of the
first concern (imho).  If we have the format defined, then it's easy
enough to get things generated even if there's some Magic (tm)

> The questions are all in the formatting:
> - do we put one package per file (that gets bulky on disk)
> - one file for all the update info for all packages  (that gets bulky to
> read through and download)
> - a combination of both so in the first case downloading only one is
> easier and in the second case searching is easier.

One file per update is probably going to scale the best for the actual
textual information -- the single file solution is going to grow quite a
bit.  If I just look at the fedora-announce-list archives, it's about
200k of gzipped text per month.  Granted, that's two releases worth, but
it still makes for basically doubling the size of the metadata pretty
quickly.  Then just mapping per package -> the text of the update with
an identifier (/location) in the primary.xml.  We'll also probably want
an easy index of the update "titles". 

That then makes it so the full information can be grabbed on an
as-needed basis which will help cut down on what needs downloading for
the common case.  It also makes it so that translation can be done just
by having files in lang-specific paths without having to regen all of
the metadata to add a translation.

Jeremy