Testing Fedora - small (?) suggestion.

Nicolas Mailhot nicolas.mailhot at laposte.net
Sat Nov 11 22:20:15 UTC 2006


Le samedi 11 novembre 2006 à 12:05 -0900, Jeff Spaleta a écrit :
> On 11 Nov 2006 14:10:08 +0100, Benny Amorsen <benny+usenet at amorsen.dk> wrote:
> > Would it be possible to automatically detect when the rawhide tree is
> > in shape for installation, and then spin an ISO?
> 
> Sure you could detect that... that's not the issue. The issue is the
> probability of having a fully consistent tree is very low from day to
> day and you could never predict the frequency nor the actual days that
> rawhide would spontenously be self-consistent.

rawhide will never be spontaneously self-consistent with the tools we
use today. That's why people ask autoblocking of the packageset pushes
that break its consistency (not blocking *all* the updates just those
not complete yet)

> Self-consistency of
> any merit will require release engineering and the establishment of
> freeze deadlines and post freeze reviews.

Self consistency only requires holding packageset pushes that break the
repo till they're complete, and packagers talking to each other instead
of communicating through angry testers. Release engineering is only
required if a package is not updated timely enough to unblock the
packageset but :
1. that's not the general case
2. the worst offenders will be AWOL at the next test mass rebuild, so
you're spreading the load, not adding to it
3. if the blocking package is not critical it can always be dropped from
the devel repo. That's better than keeping it without the needed deps
anyway (both for users and would-be new maintainers). If the blocking
package is critical breaking blocking till it's fixed is the only sane
thing to do.

-- 
Nicolas Mailhot




More information about the fedora-devel-list mailing list