Public SourceMage Discussion List

[Eric Sandall <eric AT sandall.us>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Sun, 11 Sep 2005, Sergey A. Lipnevich wrote:
> Quoting Seth Alan Woolley <seth AT positivism.org>:
>
> > I replied to your blog, but talking it over with Andrew made me realize
> > the problem was probably caused initially by underspecified dependencies
> > which caused our full-tree dependency analysis to be incomplete, thus
> > wrong, thus leading to upgrade breakage.
>
> The solution I suggest here is to give people a workable system during
> the time
> they recompile stuff (and maybe longer if something else goes wrong and recast
> stops in the middle of the queue), which would be hours for many of the users.
> I'm not trying to circumvent or improve a dependency system.

If we do the full-dependency information then we can 'fix' /all/ of
the library issues and not just openssl's.

> > What actually cast too soon (or too late) on your system for the double
> > upgrades to not work?
> >
> > Can you send a portion of the activity log to a bug report and cc me on
> > it?
>
> Maybe, if I recover the drive. One can play with factors like wget
> dependency on
> openssl (pointed by Andrew), gnome-vfs2 incompatibility with samba (recently
> fixed), spell compilations failing because /their/ dependents are missing
> openssl (remember, cleanse --fix is only run after install queue is
> empty), and
> maybe some other things, which combined may have a very bad effect.

I believe this is what the new UP_TRIGGERS code is supposed to fix as
well as Seth's fixes to openssl.

> > P.S. Allowing a trigger to rebuild a spell using the tablet version of
> > the spell might be another workaround I don't think I mentioned in the
> > blog, preventing the double-upgrade issue, but again, we need to fix the
> > problem before adding a patch that doesn't really fix the core problem.
>
> There are two problems here: recasting all dependents when shared object links
> are broken (and doing it in the right order), and keeping the system
> operational in the process (a few hours to a day). Dependency system
> deals with
> the former (now with help of UP_TRIGGER file I guess), and I'm trying to think
> of how to solve the latter.

That is a problem that I'm not sure how to fix, other than your
'generic' .so idea.

> If we do make openssl-0.9.7-so, it has to be cast /after/ openssl, so
> it must be
> in TRIGGERS file of this hypothetical openssl-0.9.7-so spell I think. Contrary
> to what I said earlier, openssl spell should not depend on openssl-0.9.7-so.

Probably have openssl-0.9.7.so depend on openssl, then have packages
that depend on openssl now depend on OPENSSL-WRAPPER and have
openssl-0.9.7.so provide OPENSSL-WRAPPER (for when we add
openssl-0.9.8.so, etc.).

I feel that this idea is just working around the problem rather than
fixing it, though it would be a quick way to fix openssl. I'd hope
that with a proper dependency tree and UP_TRIGGERS we'd be able to fix
all of the libraries instead of just openssl.

- -sandalle

- --
Eric Sandall                     |  Source Mage GNU/Linux Developer
eric AT sandall.us                  |  http://www.sourcemage.org/
http://eric.sandall.us/          |  SysAdmin @ Inst. Shock Physics @ WSU
http://counter.li.org/  #196285  |  http://www.shock.wsu.edu/
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)

iD8DBQFDJcBRHXt9dKjv3WERAv8nAJ0cm85ygZGg6CnZu6mRlEGor9TxTgCfcYbW
BeRaFhVxFWT8BLclm4xi/pY=
=E2Op
-----END PGP SIGNATURE-----