On Thu, Jun 20, 2002 at 03:22:18PM +0200, Buchan Milne wrote:
> 
> So, all you want to do is update the rpm db?

Right!  Bonus points would be awarded to an rpm (i.e. imagine this as
a feature of rpm) that did a "rpm -V" on the files in /usr "just to
be doubly sure" they are the same as the files in the rpm being
installed on the client.  Either warnings can be issued for files that
are not the same, or rpm could even fail to install and require a
--force.

This "verification" is of course an rpm enhancement and therefore
bigger than Mandrake and just a gravy item.  I would be happy if /usr
were simply for strictly non-volatile files.  i.e. if it don't come
straight out of an RPM, it don't belong in /usr.

> Do you need to 
> install/upgrade any of the files in the RPMs?

Yes.  Files that are not in /usr.  For instance, /lib, /etc, /bin,
/sbin, etc.
> 
> How about:
> 
> # rpm -Fvh *.rpm --justdb

Not quite.  As per the above.  But for files in /usr, yes, indeed.

> Otherwise, share /var/lib/rpm ro by NFS.

No, that would be very bad.  The RPM db lives in /var, where it should
be, because it is volatile.

> Not running a cluster of machines myself, I don't know what the exact 
> problem is (if you are upgrading packages, why does rpm not complain 
> about not being able to write to /usr/bin?).

Because I use "--excludepath /usr".  That tells RPM not to put files
in /usr, but still to update the database.  It does exactly what I
want it to in fact.  It's rpm scriplets and other utilities (such as
the menuing tools) that muck about in /usr that is the problem.

> And if it really only is the scripts that give you problems, how about 
> --noscripts

Well, scripts are a culprit, but it was the update-menus that fired
off this whole thread.  Menus local to a given machine should not be
in /usr.  But --noscripts is not the right answer either.  The
majority of scripts don't actually want to touch /usr, so eliminating
them for the small percentage of cases that do is just wrong.

> Maybe there should be some urpmi options for this?

I have thought that with the lack of a static method (rpmrc for
instance) for specifying --exludepath to rpm, urpmi could take this
approach.  But the root of this problem lies in rpm, not urpmi.

b.

-- 
Brian J. Murrell

Attachment: msg66338/pgp00000.pgp
Description: PGP signature

Reply via email to