On Wed, Jul 9, 2014 at 9:34 PM, Matt Oliveri <[email protected]> wrote:

> On Wed, Jul 9, 2014 at 10:06 PM, Jonathan S. Shapiro <[email protected]>
> wrote:
> > Since the Tofte-Talpin region system also pushes you to semi-unification,
> > we'll at least end up with a set of deficiencies that manifest in a
> > consistent way in different parts of the type system. I think it would
> feel
> > awkward if region type inference tended to fail for qualitatively
> different
> > reasons than (e.g.) polymorphic recursion. It makes it much harder to get
> > any intuitive handle on what's going on.
>
> Type inference doesn't have to be intuitive.


Correct. But type inference *failures* have to be intuitive. When the
inference engine is unable to solve yshaour problem, you need to be able to
develop a sense of what to do.

If we choose an incomplete inference strategy, what I suspect will happen
is that subprograms which require non-inferrable types will be relegated to
libraries written by clever experts. As long as those idioms are things
that only need to work in corners, that will be fine. If they turn out to
be central to getting effective use out of the language, that's another
matter.

I disagree with you very strongly about the need for principal types. Every
principal typing you can't obtain is a place where genericity fails.


shap
_______________________________________________
bitc-dev mailing list
[email protected]
http://www.coyotos.org/mailman/listinfo/bitc-dev

Reply via email to