The dreaded M-R

[Benjamin Franksen]

On Saturday 28 January 2006 01:13, Twan van Laarhoven wrote:
> Benjamin Franksen wrote:
>  > My personal opinion is that it should be exactly the other way
>  > around:
>  >
>  > All normal bindings (i.e. using '=') should be as polymorphic and
>  > general as possible.
>
> Do you mean *all* bindings, 

Yes.

> or only top-level ones? If you really 
> mean
>
> all, wouldn't e be polymorphic (with type Num a=>a) in, say:
>  > f x = e + e
>  >    where e = very_expensive_polymorphic_function x
>
> That would be a Very Bad Thing.

Why? The compiler /might/ be able to decide that 'f' is in fact used 
only monomorphically and thus 'e' can be shared. Even if 'f' is 
exported (in which case I would most probably write a signature at one 
time), a compiler that does whole-program analysis can find out. It 
could even specialize 'f' for each type at which it is used, each 
specialized version being monomorphic, internally, so that 'e' can be 
'maximally shared'.

If all else fails, you can always change it to := (or whatever other 
symbol will be agreed upon) if you want to indicate to the 
compiler/interpreter: "I'd rather get an error message if I use this 
polymorphically/overloaded/whatever. This 'e' must be shared at all 
costs!".

Cheers,
Ben