SF [2159499] Full bigfloat precision for Gamma after the second call



On Tue, Oct 14, 2008 at 2:03 PM, Raymond Toy <raymond.toy at ericsson.com>wrote:

> Ok, ok. :-)
>

OK, a Maxima-level demo like this is convincing.  Thanks.

I think fppi is correct.  It's the computation in fppi1 that is
> incorrect because it uses $fpprec instead of fpprec.
>

Try your test case again, with an addition:

zot(z,fpprec):= bfloat(%pi*z/sin(%pi*z))$
fpprec:64$
zot(bfloat(1/2),fpprec)$
fpprec:128$
bfloat(%pi);        <<<<<<< fppi returns without calling fppi1 for a new
value of %pi
bfloat(sin(2))      <<<<<<< add this to test; fppi calls fppi1 and gets
correct value
bfloat(%pi);        <<<<<<< now gives correct result

So it looks as though the problem is not the pi calculation itself (i.e.
fppi1), but something in the logic of fppi or fpprec1.

          -s