Difference between BOOST_CHECK_CLOSE and BOOST_CHECK_CLOSE_FRACTION?

Question

Can anyone describe the difference in behavior between BOOST_CHECK_CLOSE and BOOST_CHECK_CLOSE_FRACTION? The documentation implies the that both m

Answer 1

According to this discussion, one (BOOST_CHECK_CLOSE) treats the third parameter as expressing a percent, while the other (BOOST_CHECK_CLOSE_FRACTION) treats it as expressing a fraction. So, .01 in the first should be equivalent to .0001 in the second.

Not certain if that explains your problem -- do you get the same odd result with BOOST_CHECK_CLOSE? I wouldn't be shocked if the 0 caused an issue -- but I don't have first hand experience with the macros.

Answer 2

Yes. Zero is not "close" to any value. You can use BOOST_CHECK_SMALL instead.

Answer 3

@Gennadiy: Zero can be close to any small value. :-) Relative differences grow arbitrarily large if the expected value is very close to zero.

Here is a workaround function I use to unit-test double values: if the expected value is very small or zero then I check the smallness of the observed value, otherwise I check closeness:

void dbl_check_close(
    double expected, double observed, 
    double small, double pct_tol
) {
    if (std::fabs(expected) < small) {
        BOOST_CHECK_SMALL(observed, small);
    } else {
        BOOST_CHECK_CLOSE(expected, observed, pct_tol);
    }
}

Of course it would be great to have a BOOST_CHECK_SMALL_OR_CLOSE macro that does this automatically. Gennadiy could perhaps talk to the author of Boost.Test ;-)