Bug#898136: lintian: Reduce depends-on-mail-transport-agent-without-alternatives to pedantic

[Scott Kitterman <debian@kitterman.com>]

On Monday, May 07, 2018 01:35:30 PM Russ Allbery wrote:
> Scott Kitterman <debian@kitterman.com> writes:
> > Package: lintian
> > Version: 2.5.85
> > Severity: normal
> > 
> > Also, please reduce the certainty from certain.  It's not.
> > 
> > I'd just noticed depends-on-mail-transport-agent-without-alternatives.
> > I mainain approximately 10% of the packages affected by the check (3 of
> > 33) and in all those cases the check is wrong.  A cursory review of some
> > of the others clearly show it's incorrect for them as well (at least
> > 10).  I don't think a check with a false positive rate of a minimum of
> > nearly 50% is that useful.
> > 
> > In the case of my three, they depend on postfix without alternative
> > becuase they are only for postfix.
> 
> It sounds like there's both a bug and a certainty error here, but I don't
> think this check is a good example of something that should be pedantic.
> The dependency structure for depending on a generic MTA should be
> documented in Policy and only isn't because no one has found the time to
> write the patch.
> 
> A simple check for whether the depended-on MTA is also present in the name
> of the package would make a lot of these false positives go away.  If the
> package name contains "postfix" or "exim4" and depends on those MTAs, it's
> probably not a mistake.  :)
> 
> More generally, I suspect this tag should only affect packages that depend
> on the default (exim4).  If the package is already depending on a
> non-default MTA, I think it's highly likely that was intentional and
> Lintian is being more annoying than helpful here.
> 
> Pedantic isn't a dumping ground for buggy or uncertain checks.  If a check
> is known to be buggy or produce a lot of false positives but we don't want
> to delete it entirely because we think we can make it better in the
> future, that's what experimental is for.  Pedantic is for best-practice
> advice that's controversial, that is correct but may not be fixable (no
> upstream changelog, for instance), or that is minor
> quality-of-implementation details that a lot of maintainers aren't
> interested in messing with (upstream/metadata, for instance).

Thanks.  Experimental seems better.  Mostly I was thinking "not on the list of 
stuff most people see".  

I don't think that package name is a great trigger since, while many MTA 
specfic packages have the MTA name in the package name, not all do and there's 
no requirement for it.

Limiting it to packages that depend on exim4 would address all the false 
positives I saw (since I mostly know about postfix stuff).

Scott K