> What about the possibility to include possible exceptions into a 
> function signature (a la java)? Does this have problems with type 
> inference? 

I would love an (optional?) way to get the type signature of my functions to
reflect their non-totalness (exceptions + anything else), as well as
reflecting their 'imperative' content [ie which state variables are used].

In fact, any such 'monadic' information that can be automatically inferred
would be really useful to have (optionally).  I guess these are known as
'types and effects' systems.

Jacques