RT66882 is an interesting bug. A single special function test fails due to loss of precision, only when DUSELONGDOUBLE is defined in Perl. It seems to have the same behavior across a wide number of Perl versions, from 5.6 to 5.13.x. Why are long doubles causing this loss of precision? Haven't the slightest, currently. It could be a bug in Perl 5, GSL or Math::GSL. It's a bug hunt.
This should hopefully fix this failing CPANtesters report: http://www.cpantesters.org/cpan/report/5967e710-ea21-11df-9996-c802470775b1
Attempt to fix the test suite to deal with most of these issues: http://www.cpantesters.org/cpan/report/7328313
… tests work, illvilja++