Re: [stack] typing higher-order concatenative languages with standard hindley-milner

On May 31, 2008, at 1:36 AM, Christopher Diggins wrote:

> It can't infer much,

The system, as I am using it below, seems to do very well. Keep in  
mind that he's using it for an imperative forth-like language. We're  
using it for a functional Cat-like language. If you have an example of  
something Cat can infer using a combination of the functions below  
that the Haskell version cannot (ignoring self types), please put it  
forward. Otherwise, I'll have to waste my time finding the problems  
myself. :)

> and as Okasaki points out the whole system
> doesn't work in practice for even tiny non-trivial programs.

The efficiency issues, as far as I can tell, are only due to  
'begindef' and friends which are necessary to allow the postfix  
syntax. If you had a built-in postfix syntax, that becomes non-issue.  
The huge size of the types is, again, a purely syntactic issue.  
There's no reason they can't be rendered exactly the same as Cat's  
types, albeit without the need for rows. You just need to use the  
space as a left-associative pairing operator.

- John