Raul wrote:
> but I wonder if you have thought through
> the implications of your requirements for extended precision
> results?
The hurdle occurred to me. And I came up with some ideas, but I did
not want to be too prescriptive; I am hoping to leverage the community's
creativity. I envisioned an iterative process to resolve such issues.
(What fun is a competition with no action?)
But, if it helps, the ideas I had were:
(1) For values that are extended but could be represented by
non-negative decimal integers (n <: <: 2^<:word_size), just
prefix an x: , as in 'x:1234' -: pc_xxx '1234x' .
(2) For values that don't fit into machine words, then
something like:
(a) ' 10 #. x: '0123456789' i. ''9999999999999999'' '
-: pc_xxx '9999999999999999x' .
(i) One benefit of this format is it's
generalizable to all integral values, not
just extended ones (i.e. for normal
integers, elide the x: ).
(b) or maybe ' 10 #. x: 9 9 9 9 9 9 9 9 9 ... '.
(c) or maybe put the x: on the left instead, as in
(x:10) #. 9 9 9 9 9 ... .
(d) These considerations also apply to large floating
point values, too, as in 99999999999999999999 (i.e.
no trailing x, resulting in a FP instead of an
int).
(3) For rationals, I can't decide if (x:p) % (x:q:) is more
readable than p %&x: q but my gut says choose the former
(unless there's some clearer expression I'm not thinking
of).
Other ideas welcome (encouraged!).
-Dan
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
