Hi Kacper,
thanks, fixed in SVN 1302.
Best regards,
Jürgen
On 6/23/20 5:34 PM, Kacper Gutowski
wrote:
When
dividing a number relatively close to zero by an integer, the
result unexpectedly gets clamped to zero:
(2⋆¯34)÷1
0
Using ⌹ or ×÷ gives expected results instead.
Looks like it happens at the line IntCell.cc:543 which should be
removed. But I don't think there should be a check for near-zero
even in the zero-divisor branch. With float and complex divisors,
an exact zero is tested instead.
(2⋆¯34)÷0 ⍝ shouldn't that still be a domain error?
1
(2⋆¯34)÷0J0 ⍝ like here
DOMAIN ERROR
(2⋆¯34)÷0
^ ^
The complex one is missing a check for finiteness:
(2⋆999)÷2⋆¯999 ⍝ correct: 2⋆1998 is not representable
DOMAIN ERROR
(2⋆999)÷2⋆¯999
^ ^
(2⋆999)÷2⋆¯999J0 ⍝ ??
∞
When trying to see if I can trigger the above, I also noticed that
something is off with power too:
2⋆¯999J0 ⍝ correct
1.866527237E¯301
2J0⋆¯999 ⍝ ??
0
2⋆9999 ⍝ this is probably wrong too
∞
There's also some error in parser:
1.000000001E¯301
1000000001
1.000000001E¯302
1.000000001E10
1.000000001E¯308 ⍝ this is still representable as double
0
-k
|