Floating-point comparison of multiprecision values fails if expression template is on

[Paul A. Bristow]

Floating-point comparison of multiprecision values fails if expression template option is on (which is the default for example with

boost::multiprecision::cpp_bin_float_50; 

typedef boost::multiprecision::number<boost::multiprecision::cpp_bin_float<50>,
  boost::multiprecision::et_on>
  cpp_bin_float_50_et_on;  // et_on is default so is same as cpp_bin_float_50.)

The error appears inside Boost.test floating_point_comparision.hpp at an apparently impossible place fpc_detail::fpt_abs.

The actual problem is in class close_at_tolerance operator() line 204

   FPT diff = fpc_detail::fpt_abs(left - right);

in order to allow use of expression templates, it is necessary to cast the variable diff to the FPT explicitly, thus:

   FPT diff = fpc_detail::fpt_abs( static_cast<FPT>(left - right) );

I can provide an explicit test for this, but it seems such a trivial change.

This change is vital to allow Boost.Test to be used fully with Boost.Multiprecision.

[Raffi Enficiaud]

From the email exchanges, seems to work on the branch local/doc_floating_point_fixes. What remains for the issue is to have a proper unit test on this for multiprecision.

[Raffi Enficiaud]

merged to develop 47829d34593d9ca52de0d0a8b798b128f9ffa71f