Bug#714634: lsb-core: Remove lsb-invalid-mta as a dependency of lsb-core; require an actual MTA instead

[Aaron Sowry <aeneby@gmail.com>]

Didier, Jeff,

> I think the summary is not the above statement, but that your _opinion_ 
> is that lsb-invalid-mta does not fulfil the requirements of the LSB 
> specification. I don't agree, fwiw. Can you point to a specific LSB 
> requirement not fulfilled by lsb-invalid-mta, please?

Of course it's my opinion, and I don't think it's a completely unfounded
one. From the LSB sendmail description: 

"Note: The name sendmail was chosen for historical reasons, but the
sendmail command specified here is intended to *reflect functionality
provided by smail, exim and other implementations*, not just the
sendmail implementation." (emphasis mine)

Anyway, we could sit here and discuss the semantics of the LSB
specification all week, however I'm of the firm belief that the authors
of the specification did not include a sendmail command just because it
might be fun to have a command called "sendmail" which does nothing;
rather, it is intended to be a functional interface through which
applications can send email. I've CC'd Jeff Licquia on this mail,
hopefully he can chime in with his thoughts.

> I don't see the existance of lsb-invalid-mta as a problem, why should it 
> be removed? I think it _is_ useful for some users of the lsb-* packages 
> and therefore don't understand why we should take it off them.

I would love to hear an example of where a binary which simply exits
non-zero could be reasonably considered "useful" in this case. We
already have /bin/false.

> That's probably where I'd be open to changes: making lsb-core depend on 
> "default-mta | mail-transport-agent" on Debian (and still lsb-invalid-
> mta | mail-transport-agent" on Ubuntu) might be a worthwhile change.

Agreed. But let's focus on Debian and let downstream deal with their own
problems.

> That said, we released Wheezy with both lsb-invalid-mta and the 
> dependency on it from lsb-core so we would need to respect the choice of 
> admins that actually _want_ a non-working sendmail in their lsb 
> dependencies across upgrades.

This isn't really an administrator's problem, it's a problem for
developers of applications designed to be run on LSB-compliant systems.
I can't think of any reason an administrator would ever "want" a
non-functional sendmail command on their system.

Thanks,
Aaron