Hi,
It seems that functions requiring “near-real numbers” as arguments throw
DOMAIN ERROR when given a value represented internally as complex even
if imaginary part is exactly equal zero.
1 > 0J0
DOMAIN ERROR
1>0
^^
1 ⌊ 0J0
DOMAIN ERROR
1⌊0
^^
etc.
Note: from ISO's definition of near-real one can infer that 0J0 is always
near-real regardless of real-tolerance. On the other hand 0J0 is never
demoted by GNU APL from complex to real. This isn't relevant to examples
above but the problem initially hit me this way:
⌈/(3⍴2)⊤2
DOMAIN ERROR
⌈/(3⍴2)⊤2
^ ^
-k