On 11/8/11 1:42 PM, Richard Fateman wrote:
> On 11/8/2011 12:55 PM, Edwin Woollett wrote:
>> Is it possible to return an error in bessel.lisp
>> when, for example, the numerical value of
>> a bessel function is so large that the code/system
>> can't cope, and the answer returned for a calculation
>> is completely wrong?
>>
> The "right" solution is to roll over to another computer program that
> uses bigfloats and an appropriate algorithm for the argument values,
> and gives the right answer. (This is one of the things that
> Mathematica claims to do.)
>
On the other hand, I think there's an expectation that floats are
reasonably fast. If they automatically overflowed to bfloats, then
suddenly everything gets slower. That would go against my expectations.
But if we signaled an error, as Ted suggests, then the user could catch
it and do something about it, including conversion to bfloats.
Ray