🚀 go-pugleaf

In article <DELACOUR.92Jul16110219@waxwing.parc.xerox.com> delacour@waxwing.parc.xerox.com (Vincent Delacour) writes:

>	   2- Another issues that has been raised is that of efficiency.
>   Despite the belief expressed by some that "Laziness is theoretically
>   more efficient even." (J. Farrel), I don't see how you can avoid
>   running "at the speed of the continental drift" as will from now on be
>   remembered as "Turner's confession", unless he denies vigorously
>   having said so.

It should be noted that the Miranda implementation is notoriously slow
and based on interpretive techniques. Indeed, even back in 1987, the
Chalmers LML implementation ran 100 times faster than Miranda on a
number of benchmarks (quicksort, 8 queens, that kind of thing)
[Augustsson's thesis]. So perhaps Miranda shouldn't be taken as a
representative example of such implementations. I would guess
that lazy languages are perhaps 10 times slower than imperative
ones on the average (feel free to correct me), with a smaller
factor on some benchmarks. (E.g., procedures that use
arithmetic are usually compiled well today, as I understand it.)
I would expect this factor to decrease with time, as the issues of
compilation are better understood (the basis of much of the work in
the area was laid in the mid-80:s, which isn't all that far-off).

As has been mentioned previously by Peyton Jones, there are good books
out on the subject of outrunning the continental drift.

>	   V. Delacour

			Thomas

--
--------------------------------------------------------------------------
Thomas Lindgren				thomasl@meryl.csd.uu.se
Computing Science Dept.,		Thomas.Lindgren@csd.uu.se
Uppsala University			lindgren@sics.se

	"I can call procedures in the vasty kernel."
	"Why so can I, or so can any program. But will you
	 return when you do call them?"
-----------------------------\\\\-----------------------------------------