Bug#659822: RFS: mpd-sima/0.9.0-1 (New upstream version)

[Jakub Wilk <jwilk@debian.org>]

* Benoît Knecht <benoit.knecht@fsfe.org>, 2012-02-17, 12:01:
> > > > - In debian/mpd-sima.postrm, you use awk but you don't Depend on it.
> > > Right, I replaced it with a “grep | cut” alternative.
> >
> > While I personally don't like awk :P, but if you like it, you _can_ 
> > use it in maintainer scripts without depending on it. awk is (in a 
> > way) an essential package:
> > http://lists.debian.org/debian-mentors/2005/11/msg00193.html
> >
> > I believe that lintian would yell at you if you had unversioned 
> > dependency on awk. (And versioned dependency on awk would render your 
> > package uninstallable. :P)
>
> Interesting. In my package (minidlna), I'm depending on (gawk | mawk), 
> because I'm using awk in one of the maintainers scripts too; I take it 
> I should remove it, right?

If you use only basic functionality of awk, you can remove it, yes.

> > > > You're also checking if /usr/sbin/deluser is executable and silently 
> > > > not removing the user if it's not (same thing for delgroup). Since 
> > > > you Depend on adduser, you should assume these commands exist, and 
> > > > it should be an error visible to the user if they don't.
> > > Corrected as well.
> > Err. What Policy §6.5 says: “[…] all ‘postrm’ actions may only rely on 
> > essential packages and must gracefully skip any actions that require 
> > the package's dependencies if those dependencies are unavailable.”
> >
> > So checking for existence of /usr/sbin/deluser _is_ the right thing if 
> > you want to use it. See this however:
> > http://lists.debian.org/debian-devel/2012/02/msg00094.html
>
> Yes, I realize now my advice was misleading at best.

Or I might have misread it. :)

> What I meant is that you should try running the command anyway, so that 
> the error becomes visible to the user, but then catch the exit status; 
> something like:
>
>  deluser ${USER} || echo "Removing ${USER} failed."
>
> This way, if the user is watching the output, they will know they need 
> to remove the user by hand afterwards. Does it make sense? (Again, I'm 
> doing this in my package, so I'd like to know if that's wrong.)

You should redirect the message to stderr. Other than that, I doesn't 
look crazy, but I don't have a strong opinion about that.

The only example that is included in the Policy is:

if [ "$1" = purge ] && [ -e /usr/share/debconf/confmodule ]; then
         . /usr/share/debconf/confmodule
         db_purge
fi

(However debconf, in contrast to deluser, is supposed to work even when 
its dependencies are not installed.)

I took a look at packages installed on my system, and there appears to 
be great variety of style of this “gracefully skipping” in the wild. 
Perhaps it's something worth standardising.

--
Jakub Wilk