Issue in sin/cos_pi:

[jzmaddock]

Lurking in my inbox:

I sent a long long time ago (for boost 1.72) a proposal of correction of a very small bug of the cos_pi and sin_pi implementations

and I have just see that in the last 1.78 beta 1 nothing has been corrected yet.

The bug is from a conversion, line 36 for cos_pi.hpp and line 41 for sin_pi.hpp:

if(iconvert(rem, pol) & 1)
invert = !invert;

if rem is not representable in the integral type issued by iconvert, the result can be incorrect.

I proposed a a one liner correction, replacing the wrong test by

if(floor(rem/2)*2 != rem)
invert = !invert;

which does not make use of integer conversion.

Of course any other way to correctly compute the parity of rem will do the job.

Looking forward for your answer and correction perhaps for boost 1.79 ?

Thanks for the great boost::math, and apologies for this direct mailing.

Jean-Thierry Lapresté 

[jzmaddock]

Fixed in referenced PR