2009/04/26 John A. De Goes
2009/04/26 Jason Dusek:
This was what I was originally writing in about, yeah. However, thinking it over I realize there are some serious problems. We really do want to work with a restricted subset of Haskell that is more amenable to translation -- one with finite arrays and most every type translatable to JSON.
For purposes of cross-platform development, it's important the subset be minimally restricted (at least in my case).
Cross-platform in what sense? Like browser/GTK/QT?
I fear that requires us to know, statically, whether potentially infinite structures like lists are finite or infinite. I'm not sure about this but I suspect that's a major stumbling block.
I think it's easy to determine if something is definitely finite (certainly any list passed into an exported Haskell function must be finite, since JS doesn't have infinite lists).
That's fair; but internal lists may not be.
Determining if something is definitely infinite is not as easy, but it might be sufficient to use lazy evaluation whenever there is a possibility that a structure might be infinite. Any function exported to JavaScript must return a finite list.
That is, I think, a most reasonable restriction.
Annotations would be another way of handling this.
Why would annotations be preferable to a restricted set of types that implement, say,`Foldable` or `Traversible`? I think finite collections would be of use for other environments where we can't expect a working RTS. In fact I would not be surprised if that this sort of thing is not what's been done for GPUs by other Haskellers. -- Jason Dusek