FW: mr

[Simon Peyton-Jones]

I thought you might be interested in this brief exchange between John
and myself.  (I was starting from Simon Marlow's idea of pattern binding
=> monomorphic.)

Simon

-----Original Message-----
From: John Hughes [mailto:rjmh at cs.chalmers.se] 
Sent: 31 January 2006 14:40
To: Simon Peyton-Jones
Subject: Re: mr

Simon Peyton-Jones wrote:

>John
>
>How about this:
>
>A binding for a pattern or variable, that lacks a type signature, is
>treated as follows:
>
>* At top level: polymorphic, but illegal if overloaded
>* Nested: monomorphic
>
>The reason for the difference is that at top level there's no context,
>so forcing monomorphic is bad.  Whereas for nested bindings it seems
>like the right thing to force a signature, which will either resolve
the
>overloading or declare it.
>
>Simon
>  
>
I can see the attraction--it seems to "do the right thing" in each case.

But aesthetically,
I think it's ghastly. Now to understand a variable definition, one need 
not look (through
the entire module) to see if it has a type signature, but also look to 
see whether it is top
level or not! Refactoring that moves definitions in and out of where 
blocks (pretty
common!) would also risk introducing type errors. That would be a trap 
for the unwary
(and we were all unwary once upon a time).

John