[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Should the car be a stream?
The following seems wrongly typed to me:
((stream-cons obj strm)
(stream-delay obj) ;???
Specifically, I don't think the car of a stream should be a stream.
I agree, from the types given, this doesn't seem right.
Why weren't there types given for the streams primitives library? This
seems like a valuable part of the spec to me.
But even I am wondering about the rationale for making the stream
elements themselves promises. It may be useful to leave the types of
the elements (including whether they are promises or ordinary values)
orthogonal to the backbone of the stream itself.
I suspect that if the elements were not themselves delayed by default,
you would see a huge difference in performance.
If the elements were not themselves delayed by default, computing the
nth element of a stream is proportional to computing each element
previous to it. In the lazy version, it's simply proportional to n.
It's easy to imagine cases that perform infinitely worse using the first
model compared to the second.
To me at least, it's important that no element of a lazy list is
evaluated until it is needed. This is what SRFI 40 specifies in prose
"no element of the stream is evaluated until it is accessed," but the
code does otherwise. I'm glad to see SRFI 41 fix the problem.