Thanks. That makes sense. I confess to not having needed extended precision from J yet. I'm relieved to know it's implemented that way. Automatically switching, like you say, would be expensive. On 12 Jan 2016 4:46 am, "Raul Miller" <[email protected]> wrote:
> On Mon, Jan 11, 2016 at 6:11 AM, Matthew Baulch <[email protected]> > wrote: > > 2^.!21 is greater than 64 and 2^.!20 is less, so I'm guessing roll is not > > implemented with arbitrary precision arithmetic. > > As Roger Hui pointed out, roll is implemented for arbitrary precision > numbers (or: you can try that and see that it works). > > What it's not implemented for is floating point numbers which require > arbitrary precision results. > > Hopefully that makes sense? (If you think about an array > implementation, the switch from densely packed numbers to arbitrary > precision numbers is comparable to the switch from numbers to boxed - > it's not something people should want to have happen by mistake.) > > Thanks, > > -- > Raul > ---------------------------------------------------------------------- > For information about J forums see http://www.jsoftware.com/forums.htm ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
