[Pkg-gpm-devel] Bug#326644: gpm: modifies conffile?

[Peter Samuelson]

[Justin Pryzby]
> Hmm.  dpkg automatically handles it for conffiles.  I suppose what
> you mean is that, for configuration files, admin changes must not be
> lost on upgrade.  Instead, the file should be parsed, possibly
> storing the values temporarily via debconf database, and the file
> rewritten, if necessary.  Right?

Correct.  And that is exactly what we do.  And the simplest way to take
into account changes the user makes with debconf is to rewrite the file
in postinst unconditionally.  This, too, we do.  This also keeps the
file comments in sync with our latest version, which is in general a
Good Thing.  (You may have noticed the comment update telling you to
use dpkg-reconfigure rather than gpmconfig.  That comment is for your
benefit, not ours.)

> > The problem here is that we didn't forcibly add the md5sum of the
> > previous non-ucf file to the ucf database. So it was marked as
> > modified, then you didn't install any new config file version handled
> > by ucf, so it kept thinking it was manually generated.
> Can you fix it in the preinst, by adding it now?

Won't help.  We cannot tell the difference, as Guillem said, between a
file converted to ucf in sarge and later modified by hand, and a file
converted to ucf in sarge and *not* later modified by hand.

The only thing we could fix would be upgrades from woody to present
day.  And that isn't particularly supported anyway (although in the
case of gpm, it will work), so I don't see much point.

> I don't understand..  It affected _me_, upgrading from post-sarge
> testing machine to currents testing.  AFAICT that implies that it
> will affect other users upgrading from sarge to etch, when it is
> released as such, unless something is changed to close this bug.

Right, you don't understand.  See above for why this would have to be
fixed in sarge (which means it will not happen).

> As above, isn't the md5sum of the previous nonucf file known?

What md5sum?  The one corresponding to the woody /etc/gpm.conf on your
computer?  How would we know whether that was locally modified or not?
We don't.

We didn't even know *at the time* whether the config file was "modified
locally".  The gpmconfig script could be run independently of the
postinst, and in any case it did not record what md5sum the file had
when last written out.  debconf + ucf is a great improvement in that
regard.

Peter
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://lists.alioth.debian.org/pipermail/pkg-gpm-devel/attachments/20050906/0e7f5b7b/attachment.pgp