Re: [Toss-devel] Hierarchical Terms and Parametric Polymorphism

[Lukasz S. <luk...@gm...>]

On Wed, Jul 4, 2012 at 3:29 PM, Lukasz Kaiser <luk...@gm...> wrote:
>
> Well - it is hard to say. I was never very concerned with lack
> of expressiveness of the old basic speagram type system - it
> was the speed of parsing we were aiming for. And my hope was
> that 2 conceptual things will make it faster: keeping types with
> terms (to avoid recomputing them) and subtyping. I am now
> quite convinced that the first thing failed - we will have to get
> back to it and do something to half-undo it to bring speed back.
> But maybe it is not so relevant if the second thing works - how
> about that? If you have GLB done, we should be able to test
> it quite soon, right? And we will see then - I really hope that
> some gains will start to show up!

I'm not aiming at the speed of parsing, I'm aiming at a language for
expressing hypergraphs directly. But I understand that you aim at the
speed of parsing and I hope that the speed will return somewhat even
before introducing subtyping to the library (the substitution algo
might be to blame). I'll repeat the options:

(1) drop the proposal, go back and start again with the simplest type
system with subtyping

(2) drop parametric polymorphism and the idea of a functional-like
typed programming language, instead go for TFS-based language for
expressing hypergraphs (still a programming language based on
rewriting)

(3) have distinct TFS-variables (with GLB) and type-variables (with LUB)

I understand that you oppose (2), just wanted to make sure. I'm not
interested in (1). I'll work towards (3) then.