On Friday 27 January 2006 10:51, Simon Marlow wrote:
How about an even simpler solution:
*All* pattern and variable bindings are monomorphic unless a type signature is given.
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. An alternative symbol (':=') should be available, but strictly as a means to optimize binding of monomorphic ground values (that are then /guaranteed/ to be computed only once). I think that the language should /never/ force the programmer to consider performance issues from the outset. I see ':=' as something similar to SPECIALIZE or INLINE pragmas -- such things should never be mandatory to get a program to work. Thus, I think, for all programs that are accepted by the compiler, replacing all ':=' by '=' should result in a program that gets accepted as well, it should be semantically equivalent (apart from possibly being slower or consuming more memory). Furthermore, in Haskell there are /thousands/ of other possibilities to ruin the efficiency of your program, most of them a lot less obvious and in fact very hard to optimize away. For instance, compared to all those subtle too-much-laziness performance problems, this one could be easily diagnosed by profiling and also easily fixed by introducing the occasional ':='. Also, the compiler could help by issuing warnings (but only if prompted to do so by some compiler flag). On the other hand, forcing the programmer to write signatures in order to get the most general type of a bound variable -- when this would be avoidable in principle -- is a very bad idea, IMO. Type inference is an extremely valuable feature when prototyping and experimenting. It is also of great help for people who want to explore some of more advanced aspects of Haskell's type system. Asking ghci or hugs to give me the inferred type of a complicated function definition has often given me the decisive insight about how to write a correct version. Just my (hopefully not too uninformed) 2 cents. Ben