I agree about inexactness being contagious, so that if you divide (/ #i2 #i7) then you should certainly get an inexact bit set on the result, but that alone doesn't say whether that result is stored as a flonum or as a pair of ints.

Wraith Scheme (my R5 implementation) approaches this problem as described here:

http://jayreynoldsfreeman.com/Aux/WraithScheme.64/Wraith%20Scheme%20Help.2.25.html#R5%20Section%206

Jay Reynolds Freeman
-------------------------------
Jay_Reynolds_Freeman@mac.com
http://JayReynoldsFreeman.com (personal web site)

On Apr 29, 2014, at 12:55 AM, Taylan Ulrich Bayirli/Kammer <taylanbayirli@gmail.com> wrote:

Jay Freeman <jay_reynolds_freeman@mac.com> writes:

> Thus I suggest that issue #1 not be addressed without also
> simultaneously addressing means whereby Scheme users who wish to trade
> accuracy for speed in numerical work, may do so.

I thought that was done simply by introducing an inexact number anywhere
in the calculation, since inexactness is "contagious".  For example by
starting out with inexact constants ("1.0" instead of "1"), using the
`inexact' procedure on initial inputs, etc.

Taylan

_______________________________________________
Scheme-reports mailing list
Scheme-reports@scheme-reports.org
http://lists.scheme-reports.org/cgi-bin/mailman/listinfo/scheme-reports