yum cpio: md5sum errors

[seth vidal]

On Mon, 2009-03-02 at 14:48 -0800, Adam Williamson wrote:
> On Sat, 2009-02-28 at 05:12 -0500, Scott Robbins wrote:
> 
> > Ah, well,that's an important fix.  :)  
> > 
> > I'm going to (most respectfully) disagree that there's no helping it
> > after the fact.  There is.
> > 
> > On the page giving mirrors put in (again in large font and probably even
> > in red) a message that one should not run yum -y update but should,
> > instead, first do yum update rpm.  (Or, before it's fixed, I think the
> > simplest way would be yum install yum-downloadonly, then run yum
> > --download only rpm, and give the rest of the instructions.)
> 
> An idea occurs here. Again from Mandriva, sorry. :)
> 
> urpmi (Mandriva's yum-equivalent) handles the packages for itself and
> all the stuff it depends on - RPM, the perl interface to RPM, and a few
> critical things like glibc - specially. If any transaction involves
> upgrades to these packages, *just those packages* are automatically
> upgraded first, and then the urpmi transaction is re-started to handle
> all the other packages involved. This is designed to handle problems
> like this.
> 
> It also handles another case. Before this was introduced, we had a
> situation where there was a bug in urpmi itself, which caused an
> 'update' transaction to install stuff it shouldn't. So we sent out an
> update for the 'urpmi' package to fix it. Problem was, when you ran the
> updater application, you were still using the un-fixed version of urpmi,
> so the bug would happen during the same transaction that installed the
> updated urpmi package that fixed the bug...:)
> 
> Maybe yum should do something like this?

That really doesn't work. here's why:

new yum/rpm/rpm-python in the repo

yum internally does a:

yum update rpm\* yum\*

it pulls in a new python
which pulls in a new glibc

see where you are?

-sv