close_at_tolerance always returns false for comparisons of infinity about test

[Timmmm]

I have double checked and that test definitely passes on Boost 1.60.0 but not 1.67.0. Unfortunately brew doesn't have any versions between those so I can't easily narrow it down. :-(

[raffienficiaud]

Thanks for filing this bug and checking about the regression. Not so much happened on this part since some time, and I believe the fix is quite easy. The only thing I do not know about is how other floating point types (like multiprecision) handle infinity.

[Timmmm]

Do you happen to know what the fix is or which commit changed the behaviour?

[raffienficiaud]

I believe this is this one: https://svn.boost.org/trac10/ticket/13011
Rev bf703c2

I will check tonight.

[Timmmm]

Ah cool, I patched it like this and it works for us now:

inline assertion_result
compare_fpv( Lhs const& lhs, Rhs const& rhs, op::EQ<Lhs,Rhs>* )
{
    if (lhs == rhs) {
        return true;
    }
    ...

[raffienficiaud]

It is fixed on the branch topic/GH-156-comparison-to-infinity. However close_at_tolerance is using a tolerance value, and does not handle 0 nor infinity. I updated the doc as well.

Let me know if the branch works for you

[raffienficiaud]

In develop, thanks again for the report. Closing.

[Timmmm]

Seems like our tests have changed and I can't get them to fail now, but the fix looks good to me - thanks!

[raffienficiaud]

You're welcome!