On Oct 18, AndrevanTonder wrote:
> On Mon, 16 Oct 2006, Eli Barzilay wrote:
>
> > I was surprised to see that the bullets at the end of section 20.3
> > were kept.
>
> I don't know if the existing bullets should be kept.
Just to clarify things: I don't think that I oppose having these
bullets in, I just don't think that they're the kind of stuff that can
be thrown in as a side remark. I think that my Lazy Scheme
implementation is a natural result of taking these bullets to an
extreme -- which means that I *like* them, but I think that they lead
to something that is different enough from a `plain' Scheme that it
requires more discussion. In its current form, I think that they are
questionable.
> However, I would like to suggest a bullet option allowing
> implementations to throw an exception for reentrant promises,
> reading as follows:
>
> - An implementation may raise a runtime contract violation if a
> promise refers to its own value, i.e., if forcing a promise may
> cause the promise to be forced a second time before the value
> of the first force has been computed.
>
> In other words, the example in the middle of page 121 may be allowed
> to throw a runtime exception.
>
> I think being able to detect this kind of thing is a useful feature,
> and an implementation should not be disallowed from providing this
> feature. I have an implementation that does this, if someone is
> interested.
>
> I might make this a formal comment. However, I'll wait and see
> first - maybe someone knows of useful applications of reentrant
> promises of which I am not aware.
It sounds to me like a probably useful feature, but I wouldn't bet on
not having any use for the kind of self-reference that will be
forbidden by it. I will try to do this in my Lazy Scheme thing, and
see if any problems pop up.
--
((lambda (x) (x x)) (lambda (x) (x x))) Eli Barzilay:
http://www.barzilay.org/ Maze is Life!
Received on Thu Oct 19 2006 - 21:07:16 UTC