On Tue, May 2, 2017 at 12:00 AM, John Cowan <xxxxxx@ccil.org> wrote:

On Mon, May 1, 2017 at 10:12 AM, Alex Shinn <xxxxxx@gmail.com> wrote:

Is it worth supporting hypothetical non-twos-complement implementations?

Frankly, my dear, I don't *care* about people who want to run Scheme on their Unisys ClearPath Dorado mainframes. They typically pay per CPU cycle, anyway. The IBM 360 (1964) and the DEC PDP-1 (1963) were already 2's complement when the remote ancestor of the Dorado, the Univac 1100, came out.

I was thinking more of emulated environments. Specifically

in javascript all numbers are double precision floats. These

can be used to implement fixnums easily enough - the fixnum

procedures are just normal javascript arithmetic, and the

"error" cases overflow into inexacts (with loss of precision).

In this case, the fixnum range is [-2^53, 2^53], inclusive.

On the other hand, we could also just as easily define 2^53

as outside the fixnum range to agree with the standard.

It would depend on the potential use cases of fixnum-width

which is more useful.

Alex