This takes a bit more work to compute, but since we cache all of this anyway
it isn't too big of deal. Using average delay instead of last delay will be
a bit more fair on mirrors that have odd syncing schedules, as well as
exposing those that only sync once a day. Also fix an issue that will arise
with cutoff_time being calculated once, and adjust mirror score to treat
hours delay as a float rather than an integer.
Signed-off-by: Dan McGee <dan@archlinux.org>
This should get this to the point where it is releasable to the general
public for their use and pleasure. Still not sure on how often the check
should be run, and we probably want to incorporate this mined data into some
other things like the mirror list generator.
Signed-off-by: Dan McGee <dan@archlinux.org>
