[racket-dev] member like assoc

On 2013-05-31 19:40:52 -0400, Asumu Takikawa wrote:
> Is it feasible to get `member` to have the same optional argument
> behavior as `assoc`? That is, to have an equality predicate as the third
> argument.

I went ahead and implemented this behavior and submitted a pull request:
  https://github.com/plt/racket/pull/366

To see if the change is "pay as you go", I ran some microbenchmarks to
see if old-style `member` calls would slow down (see below). There
doesn't seem to be any significant slow-down AFAICT.

I was surprised that the non-JIT version was faster in the first case
but much slower in the second microbenchmark though. (note: I had to
change the C implementation of `member` for the non-JIT path)

I'd appreciate any suggestions on the code.

Cheers,
Asumu

  ;;; NEW BRANCH
  $ racket member-benchmark.rkt
  cpu time: 1748 real time: 1752 gc time: 0
  $ racket --no-jit member-benchmark.rkt
  cpu time: 1524 real time: 1526 gc time: 0

  ;;; MASTER
  $ racket member-benchmark.rkt
  cpu time: 1712 real time: 1716 gc time: 0
  $ racket --no-jit member-benchmark.rkt
  cpu time: 1524 real time: 1528 gc time: 0

The microbenchmark is just this:
  #lang racket/base
  (require (only-in racket/list range))
  (define lst (range 1 5000))
  (time
    (for ([i 30000])
        (member 2500 lst)))

On a slightly different microbenchmark:

  ;;; NEW BRANCH
  $ racket member-benchmark-2.rkt
  cpu time: 2396 real time: 2402 gc time: 0
  $ racket --no-jit member-benchmark-2.rkt
  cpu time: 7156 real time: 7174 gc time: 0

  ;;; MASTER
  $ racket member-benchmark-2.rkt
  cpu time: 2412 real time: 2416 gc time: 0
  $ racket --no-jit member-benchmark-2.rkt
  cpu time: 6892 real time: 6911 gc time: 0

  #lang racket/base
  (define lst '(a b c))
  (time
    (for ([i 30000000])
        (member 'b lst)))