Ben:

This is a very useful pointer, and I'll read what LippMeier has to say, but
in general terms I don't think that adding implicit thunkification changes
anything from an effects standpoint.

The reason is simple:

At any point where the language introduces implicit thunkification, we could
have written the same program explicitly using lambda forms, and we know how
to do type-and-effect for that.

At the moment, it appears to me that all of the thunk injection can be done
cleanly, using purely syntactic transforms, before the type checker even
runs. So I think it's just sugar.

You also wrote:

On Fri, Sep 3, 2010 at 10:29 PM, Ben Karel <[email protected]> wrote:

> On the other hand, for a language that's supposed to have a transparent
> mapping to the machine level, it's not immediately clear that giving
> indistinguishable syntax to such semantically different operations is a good
> idea.


Mark Miller was making the same point, and went further to observe that
lightweight lambdas (which go back to Smalltalk) make explicit syntax much
more palatable. He also noted that if the only use cases are
AND/OR/IF-THEN-ELSE, then this isn't sufficient to justify new mechanism in
the language. And I agreed.

But then I read Pitman's paper, and it reminded me of the many other uses
for thunkification that people have found useful in LISP. And I started
digging in to mixfix, and that brought my attention to the fact that FEXPRs
like these are right smack in the middle of the expression hierarchy, and
play a more central rule than people want to admit.

And there is certainly something attractive about reducing the core
language...


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

Reply via email to