On Fri, 2005-01-28 at 07:43 -0800, Shawn Garbett wrote:

> Does avoiding polymorphism and static type checking
> exclude things like type classing in Haskell?

I don't know. Perhaps Swaroop does.

Perhaps I should explain to everyone here that our group isn't really
very knowledgeable about type systems, and in this regard our objective
is to learn as little as we can and still competently get the job done.
In this area, we want to reuse established work, not invent it. Our goal
isn't to get sucked into the type system world. Our goal is to get a
pragmatic solution built *quickly* for safe systems programs.

As principle investigator in all this, my greatest fear concerning BitC
is that it is extremely easy to get sucked into examination of type
systems for *decades*. There are people who spend their entire careers
on this, and they are much better trained at it than any of us are.

So: I'm happy for us to consider anything simple and compelling, but it
has to be both. I'm also willing to look at languages that take
different approaches than ML, but I cannot say that I'm very eager to do
so. It would slow us down alot.

As it is, I'm not sure that polymorphism is going to be compatible with
the object system I have in mind.

But here is my litmus test: is it better to have a working, safe
language a year from now with a mature type system that we can reason
about two years from now, or is it better to spent the next decade
screwing around to find the best type system on Earth, and the following
decade formalizing it and figuring out how to prove stuff about it?

Now if there is a type system out there that has the pragmatic utility
of ML-style polymorphism but is known to be easier to reason about
formally, *that* would be of great interest to me.


shap

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

Reply via email to