If your transcript is accurate, there is certainly a bug. The fact that e."0
_ works probably indicates it lies in the special code for e. . However,
since you're running J5, it may already have been fixed. Can you try
downloading the latest J6 beta and trying your experiment there?
I would do so for you, but I cannot reproduce your results even in J5. It
maybe that I'm on a Windows machine, i.e.
9!:12''
6
and you're on Linux. But I bet that's not the difference. I believe your data
is very specific. For one thing, it's complex. Do you expect that? Complex
numbers are implemented with floating points, which has been a historical
sticking point for comparatives.
Try the following and see if you get the results you expect:
p1 e.&:(]&.":) p2
The verb ]&": ensures the numbers "are" what they "look like". If that
solves the problem, then you can either use it in your scripts, or we can
explore your data further. A few phrases which will help:
p1 e.&:(]&.":) p2 NB. Ignore internal reps: still all 0s?
p2 e. p1 NB. Switch arguments: still all 0s? (order
sensitive?)
p1 -:!.0 p2 NB. Zero tolerance: still 1?
p1 =!.0 p2 NB. Zero tolerance: still all 1s?
p1 e.!.(2^_34) p2 NB. Max tolerance: still all 0s?
9!:11[20 NB. Max print precision
p1 NB. Display p1 to full precision
p2 NB. Display p2 to full precision
p1 -:&:(3!:3) p2 NB. Compare internal reps byte-for-byte
p1 ;&:(3!:3) p2 NB. Display the internal reps for comparison.
Run these sentences and send the results back to the Forum.
-Dan
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
