On Sun, Sep 13, 2015 at 7:21 PM, Arthur A. Gleckler <xxxxxx@speechcode.com> wrote:
On Sun, Sep 13, 2015 at 8:27 PM, Arthur A. Gleckler <xxxxxx@speechcode.com> wrote:
On Sun, Sep 13, 2015 at 4:28 PM, Shiro Kawai <xxxxxx@gmail.com> wrote:
Go's map actually randomizes iterating order every time.  It has a merit
that it tends to catch the code depending on the interation order early,
and it won't be surprising if some Scheme implementations adopt it.

If there's a cheap way to do that, that sounds like a good idea, assuming that jumping around doesn't lead to poor performance, once again due to cache effects.  I'd certainly be nervous about requiring such behavior until careful measurements had been taken on several implementations that did that.

Sorry, I meant for this message to go to the list, too.

And this was my response:
 
Yeah, I agree that *requiring* randomized behavior is overkill.  It's just that we don't
want to exclude it (i.e.. implying unspecified, but consistent order under the same 
boundary condition).
btw, Go uses some internal "lightweight PRNG" for the purpose.