Rawhide stability

[Roland McGrath]

> # yum install yum-skip-broken and use the --skip-broken parameter.

My hero!  A "smallest possible transaction" mode would still make me
happier for rawhide upgrades.  I have too many times had the giant yum
upgrade not have any dep problems, but have some %post error or kernel
crash in the middle, that left me with an unholy mess of half-upgraded crap
to sort out (usually both versions of a billion rpms in the db, with the
files maybe from the new one or maybe still some files from the old one,
etc).  If each transaction were only as big as it needed to be to be
coherent, with the rpm database written in a clear state, then the
incremental damage done at any instance of crashing in the middle would be
much more manageable.


Thanks,
Roland