Paul A. Bristow schrieb:
-----Original Message----- From: boost-users-bounces@lists.boost.org [mailto:boost-users-bounces@lists.boost.org] On Behalf Of Steven Watanabe Sent: Wednesday, March 31, 2010 3:35 PM To: boost-users@lists.boost.org Subject: Re: [Boost-users] [Boost.UTF] Strange (failed) result on comparing floats
AMDG
Olaf Peter wrote:
I'm checking some data of consistence using boost's unit test framework (1.42):
BOOST_CHECK_CLOSE( *v[1].y.max, -43.0e-3, std::numeric_limits<double>::epsilon() );
which fails with:
difference{1.6137e-14%} between *v[1].y.max{-0.043000000000000003} and -43.0e-3{-0.042999999999999997} exceeds 2.2204460492503131e-16%
where v is a std::vector,y a struct and max a boost::optional<double>. The data where parsed by boost.spirit. This is the only failed test case from 16 parsed doubles. What's gone wrong here? Do I miss something?
epsilon is too small. BOOST_CHECK_CLOSE treats the argument as a percent, so there's an extra factor of 0.01, which means that this comparison is effectively for equality.
Well spotted Steven.
BOOST_CHECK_CLOSE_FRACTION( *v[1].y.max, -43.0e-3, std::numeric_limits<double>::epsilon() );
is what you meant?
Yes, Thanks!
Our esteemed and illustrious author of the invaluable Test library was begged not to use percents but ...
why this? Olaf