Re: Piuparts testing status update

[Justin Pryzby <justinpryzby@users.sourceforge.net>]

On Tue, Nov 14, 2006 at 08:07:27PM +0100, Frank Lichtenheld wrote:
> On Tue, Nov 14, 2006 at 06:44:38PM +0100, Lucas Nussbaum wrote:
> > Here is a list of packages that fail when all packages except essential
> > ones, apt and debfoster wxhere removed, but didn't fail when all
> > important&required packages were kept. This indicates a missing
> > dependency on an important or required package, and is considered an RC bug.
> > Packages from the list above were removed from that list, so all of them
> > should be RC bugs (pending confirmation from the release team, of course).
> 
> Here some preliminary investigation of these to give a better feel
> for the reasons behind the failures.

The package needs to check for the existence of the dependency before making
use of it, right?

|Note, however, that the postrm cannot rely on any non-essential packages to be
|present during the purge phase.

This is at least true for the "debconf" errors; since "debconf is a cache", it
isn't useful to fail when the cache has already been removed.  I guess this is
why some packages use:

   . /usr/share/debconf/confmodule || true

They should really use 

   f=/usr/share/debconf/confmodule
   [ ! -e $f ] || . $f;

For the "adduser" errors, it might be reasonable to intentionally fail, as a
mechanism to alert the admin that "the user hasn't and can't be removed".  Same
for update-inetd.  Is that the intent?

| > gpar2
| update-mime-database on purge, no missing dep
|
| > lukemftpd
| update-inetd on purge, no missing dep
|
| > gtk-im-libthai
| update-gtk-immodules on purge no missing dep
|
| > ndtpd
| syslog-facility on purge, no missing dep

These can all be fixed by using just if [ "$1" = remove ], since "purge" is
only ever called after "remove".

Justin