Bug#370720: virtual-package-depends-without-real-package-depends is bodus

[Kevin Glynn <kevin.glynn@gmail.com>]

Russ Allbery writes:
 > martin f krafft <madduck@debian.org> writes:
 > 
 > > Package: lintian
 > > Version: 1.23.21
 > > Severity: normal
 > 
 > > W: mdadm: virtual-package-depends-without-real-package-depends recommends: mail-transport-agent
 > > N:
 > > N:   The package declares a depends on a virtual package without listing a
 > > N:   real package as an alternative first.
 > > N:   
 > > N:   A real package should be listed in the first part of the | dependency
 > > N:   in order for the package to be installable by package management
 > > N:   programs that can't or won't guess which alternative to select by
 > > N:   default. In particular, it helps build daemons rebuild the package
 > > N:   without manual overrides.
 > > N:   
 > > N:   Refer to Policy Manual, section 7.4 for details.
 > 
 > > This makes no sense. Build daemons are only concerned about
 > > build-depends, and there isn't one package management programme in
 > > Debian that can't pick an alternative.
 > 
 > > Please at least don't show this warning for recommends/suggests.
 > 
 > If package A Build-Depends on package B which in turn then Depends on
 > virtual package C, I can see two possibilities:  either the package system
 > will want to prompt someone for what package to install, which won't work
 > because it's non-interactive, or the installed package to satisfy the
 > dependency on virtual package C will be chosen essentially at random.  The
 > latter leaves open the possibility that a different choice will be made
 > next time, meaning that we no longer have a consistent and reproducible
 > build environment.  That strikes me as bad, although I'm not sure it's bad
 > enough to warrant the lintian warning.  I'd at least want to see some more
 > discussion of that first.
 > 

Since you ask for discussion :-).  

I just got this lintian warning (I replaced emacs21 | xemacs21 by
emacsen in my depends field) and I have a little complaint as
well. Policy Manual, section 7.4 does not say that a real package must
be included in the dependency.  It just says "The effect is as if the
package(s) which provide a particular virtual package name had been
listed by name everywhere the virtual package name appears".

I understand the usefulness of this requirement but then shouldn't the
Policy Manual be updated to say this?

thanks
k