[Haskell-cafe] package show needs upper version bound for smallcheck?

[Johannes Waldmann]

> Could the maintainers of "show"
> http://hackage.haskell.org/package/show-0.4.1.2
> please add some version bound (< 1  or similar)
> for the smallcheck dependency?

Thanks for the quick reaction.
show-0.5 now depends on smallcheck>=1.0 
This works until the next API-breaking change in smallcheck ...

I admit that I find too-detailed upper bounds on dependencies
impractical as well, because they make upgrades really painful; 
so mostly I'm too lazy to write any bounds. 
Which is a bad idea, as the above example shows.

Perhaps a dependency on a major version (only) 
could be a reasonable policy (smallcheck == 1.*)


PS: Anyway the underlying problems that these policies
try to solve (did the API "really" change? 
did the observable behaviour of the implementation change?)
can only be solved by formal specification and verification.

(e.g., you write down the full specification of a "sort" function,
then a machine can check whether some library contains some function 
that claims to fulfil this spec, and it can also check
whether the claim holds true - if the implementation comes with a proof)

(Corollary: Agda would not need a package versioning policy
because it can solve these problems at the language level :-)