Implicit parameter constraints not inferred

[Wolfgang Thaller]

Benjamin Franksen wrote:

>> main = let ?b = True in use_b
>
>> --use_b :: (?g::Bool) => IO ()
>> use_b = print ?b
>
> It isn't: ghc -fimplicit-params says
>
>     Unbound implicit parameter (?b :: a)
>       arising from use of implicit parameter `?b' at TestBug.hs:4
>     In the first argument of `print', namely `?b'
>     In the definition of `use_b': use_b = print ?b
>
> It works if I uncomment the signature.
>
> Using ghc-6.2.2, btw.
>
> My question: Is this as it should be or is it a bug?

The Monomorphism Restriction.
It's intended to prevent people from shooting themselves in the foot by 
defining things that look like constants (like use_b) but really are 
polymorphic values that still depend on some context and therefore are 
not constant.
So in this case (use_b defined with no arguments), you need to specify 
a type signature.

The monomorphism restriction would be useful in situations like:

foo = expensiveOp bar

How often will foo be evaluated?
With the monomorphism restriction, at most once. Without it, well, that 
depends on the type of bar. With bar :: (?g :: Bool) => Int, foo also 
gets the (?g :: Bool) => context, and it's reevaluated whenever it's 
used; but you won't be able to tell that by looking at foo alone.

Cheers,

Wolfgang