[Boost-bugs] [Boost C++ Libraries] #10241: Please relax template constraints of boost::math::binomial_coefficient

Subject: [Boost-bugs] [Boost C++ Libraries] #10241: Please relax template constraints of boost::math::binomial_coefficient
From: Boost C++ Libraries (noreply_at_[hidden])
Date: 2014-07-24 15:46:22

#10241: Please relax template constraints of boost::math::binomial_coefficient
------------------------------+--------------------------
 Reporter:  der-storch-85@…   |      Owner:  johnmaddock
     Type:  Feature Requests  |     Status:  new
Milestone:  To Be Determined  |  Component:  math
  Version:  Boost 1.55.0      |   Severity:  Optimization
 Keywords:                    |
------------------------------+--------------------------
 boost::math::binomial_coefficient may up to now only be used with floating
 point types. See also ticket #3100.

 This constraint has two problems:

 It yields in inaccurate calculations, i. e.
 {{{
 static_cast<long long
 unsigned>(boost::math::binomial_coefficient<double>(60, 30));
 }}}
 is
 118264581564861408
 but should be
 118264581564861424.
 (long long unsigned and double are both 8 byte on my machine.)

 And it does not make sense for very large or arbitrary-precision integer
 types: boost::math::binomial_coefficient<boost::multiprecision::cpp_int>
 would work great if you let it do so.

 At the moment it is not possible just to release integer types as template
 argument, you have to add an extra implementation, i. e. this one with
 unsigned k and n as input:
 {{{
 Inttype result = 1;
 k = std::min(k, n-k);
 for (unsigned i = 1; i < k+1; ++i)
 {
         result *= Inttype(n-k+i);
         result /= Inttype(i);
 }
 }}}

-- 
Ticket URL: <https://svn.boost.org/trac/boost/ticket/10241>
Boost C++ Libraries <http://www.boost.org/>
Boost provides free peer-reviewed portable C++ source libraries.