On 09/10/2012 03:58 AM, Erich Rast wrote:
I agree with that one. Regarding this I also have a question. Sometimes
I just need one of multiple values, but this always bothers me somewhat.
Take for example
(let-values (([a b c] (values 1 2 3)))
(computation-with a))
Are the spurious bindings to b and c optimized away if they aren't used?
Is this even possible in general?
Matthew pushed a change to the optimizer a little while ago that
transforms your example into
(let-values ([(a) 1] [(b) 2] [(c) 3]))
(computation-with a))
If `b' and `c' are unused and the optimizer can prove their values are
side-effect-free, they'll be optimized away. They are in your example,
according to the timings output by this program:
#lang racket
(require racket/unsafe/ops)
(define (make-x) 4)
(define x (make-x)) ; keeps the optimizer from inlining x
(define bx (box #f)) ; mutated to make sure optimizer leaves `+' alone
(for ([_ (in-range 5)])
(time (for ([_ (in-range 10000000)])
(let-values ([(a b c) (values 1 2 3)])
(unsafe-set-box! bx (+ a x))))))
(newline)
(for ([_ (in-range 5)])
(time (for ([_ (in-range 10000000)])
(let ([a 1])
(unsafe-set-box! bx (+ a x))))))
I don't know how smart the optimizer is about proving computations are
side-effect-free. It's probably limited to arithmetic and a few other
basic primitives.
Neil ⊥
____________________
Racket Users list:
http://lists.racket-lang.org/users
