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

Stavros Macrakis wrote:
> On Mon, Oct 13, 2008 at 12:09 PM, Raymond Toy <raymond.toy at ericsson.com
> <mailto:raymond.toy at ericsson.com>> wrote:
> 
>     ... It seems to me that the problem is actually in the value of pi.
>      I put a break point at *fpsin, and this is what I see, when we
>     change the precision from 64 to 128.  
> 
> 
> Internally, if code wants bfloat pi to the current precision, it should
> be calling (fppi), not looking at bigfloat%pi.
> 
>     I think the problem is that when fpprec is set, fpprec1 is called.  But
>     this only updates the bfloat versions of 1, 0, +/- 1/2.  Perhaps we need
>     to update the bigfloat values of %pi at least, and probably also %e and
>     %gamma.
> 
> 
> No!  fppi takes care of calculating the value of pi *if needed* and
> caching it.  If it is not needed, it should not be calculated, as that
> can be very expensive.

Fair enough.  That makes sense.

Now I'm not exactly sure what is happening.  I can see fppi is called
before *fpsin is called.  And fppi is called again from *fpsin (because
*fpsin adds 8 to fpprec).

Not quite sure what happens then.  But it's clear the argument to fpsin
is off a bit.

Ray