On 5/16/12 2:50 PM, Edwin Woollett wrote:
> On May 16, 2012, Raymond Toy wrote:
> -------------------------
>>> testing Raymond Toy's bug #3526359
>>> fix contained in my gamma-new.lisp.
>>>
>>> float and bfloat of gamma_incomplete(1/5,-32)
>>> good to 15 digits (gcl).
>>>
>>> compile-file complains about two items.
>>>
>>> the bfloat evaluation now contains the
>>> comment "new stuff", and also should
>>>
>>
>> Oops. I'll fix that.
>>
>>> have an extra expansion of the result.
>>>
>>
>> Oh, I guess previously you didn't have to call expand to get the expected
>> result. I can fix that.
> -------------------------------------------------------
> another problem, found while playing with some integrals,
> is that bfloat is not processing (-1)^(4/5) with domain
> equal to complex.
> -------------------------------------------------------
That's not new, I think. There's been a longstanding issue about
bfloat((-1)^(1/4)). You have to rectform it to get a bfloat value.
IIRC, there was no resolution on what to do.
>
> (%i1) display2d:false$
>
> (%i2) load("c:/work2/gamma-new.o");
> (%o2) "c:/work2/gamma-new.o"
Just a note that if you update from git again, you'll have to get
gamma.lisp and numeric.lisp for this to work. I moved the code for the
Lentz algorithm from gamma.lisp to numeric.lisp.
Ray