[Bioclusters] how are the Redhat product changes affecting existing and future plans?

[Tim Cutts]

On 06-Nov-03, Matt Temple wrote:
> I think time will tell how well and if YUM or APT-RPM will
> make it there.    up2date, while a big improvement on ... uh ...
> nothing, would often choke if I had done any "non-standard"
> installations and claim that package this or that needed to be
> removed when I couldn't even find it.   These things are often
> a matter of habit.   I got used to rpm.   Strangely, there's
> a little lurking thingie that goes off when looking at
> Debian -- sort of like "it's too easy.   It must not be taking
> something into account."

Debian's package management system does work extremely well, and I agree
it's hard to shake off the fear.  :-)  A colleague of mine found a
really nice example of how sensible apt/dpkg is the other week, when he
discovered that when you upgrade a package, the first thing that happens
is that Debian tries to run the pre-rm script for the installed package.
If that fails, rather than abort the upgrade, it runs the pre-rm script
from the *new* package, which it hasn't installed yet, and continues the
installation if that succeeds.  That gives the package maintainer a
really simple way of fixing brokenness in existing installed versions of
his/her package; just issue a new package with a pre-rm script which
knows about the problems in the old script, and apt will sort it out.

> I don't want to encourage divisions here.   Since I wouldn't like to see
> there to be the One True Linux, I think we'll probably be looking
> at Red Hat, Fedora, Suse, and Debian and (throw in some other
> favorites.)    I find that people wind up comfortable and  competent
> in the distro they've used most.

Actually, I've probably used Red Hat more than anything else (certainly
on more machines, if not for a longer length of time), but I'm most
comfortable with Debian.  I agree it's good to avoid monoculture.

My main concern is that most distributions lack ISV support.  I would
like to encourage distributions to do what Platform Computing do, which
is to support a particular kernel/C library version combination,
which is a rather more sensible approach.

Tim

-- 
Dr Tim Cutts
Informatics Systems Group
Wellcome Trust Sanger Institute, Hinxton, Cambridge, CB10 1SA, UK