Tolerance is an ugly thing, I guess.

http://code.jsoftware.com/wiki/Essays/Tolerant_Comparison

Thanks,

-- 
Raul


On Wed, Sep 13, 2017 at 11:23 AM, Erling Hellenäs
<[email protected]> wrote:
> Floor and ceiling and eq in j all uses comparison tolerance, So when the
> result is set to 0 if floor and ceiling are equal in the J sense, we get the
> zero result. The question s then if this is the correct behaviour.
>
> As we can see floor uses comparison tolerance. It is not obvious to me why.
> It is not a comparison. If it didn't, we would get the correct result.
>
> Another possibility is to use comparison tolerance = 0 for floor and ceiling
> only here, if that would be what we consider correct behaviour.
>
>      q=: 5729082486784839 % 196
>
>      30j5":q
> 29230012687677.75000
>      30j5":<.q
> 29230012687678.00000
>      30j5":>.q
> 29230012687678.00000
>      30j5":(>.q) = <.q
> 1.00000
>
>
> 9!:19]0
>
>
> 30j5":q
>
> 29230012687677.75000
>
> 30j5":<.q
>
> 29230012687677.00000
>
> 30j5":>.q
>
> 29230012687678.00000
>
> 30j5":(>.q) = <.q
>
> 0.00000
>
>
> Cheers,
>
>
> Erling
>
>
>
> Den 2017-09-13 kl. 16:34, skrev Raul Miller:
>>
>> Hmm... actually... no...
>>
>> The relevant implementation here is in ve.c:
>>
>> static D jtremdd(J jt,D a,D b){D q,x,y;
>>   if(!a)R b;
>>   ASSERT(!INF(b),EVNAN);
>>   if(a==inf )R 0<=b?b:a;
>>   if(a==infm)R 0>=b?b:a;
>>   q=b/a; x=tfloor(q); y=tceil(q); R teq(x,y)?0:b-a*x;
>> }
>>
>> No hint of fmod here (though it does get used elsewhere).
>>
>> FYI,
>>
>
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to