DEPENDS semantics (was: removing useless dependencies)

[Edgar Fuß <ef%math.uni-bonn.de@localhost>]

DH> ...which is a bug.
Well, one could argue it's not a bug because it doesn't violate the spec.
Because there's no explicit spec, I guess.
Of course, it's surprising behaviour and should be prominently documented.

JS> Please stop asserting that random things are bugs without understanding
JS> both the big picture and the details, thanks.

DH> How does it make sense to compare version numbers of unrelated
DH> packages?

I'm afraid that, in the binary case, where you have no more information 
than "foo>=47.11 of bar>=0.8.15", it's the most un-unsensial tie-breaker 
to choose. It's the only thing, that, at least in some cases like 
"libfoo1>=1.11 or libfoo2>=2.22" may make sense.
Any suggestion for a better way of tie-breaking? It would make sense to 
think of a way of expressing the tie-breaking rule inside the DEPENDS line, 
but you would need a syntax expressing all common rules first.