Equality for floating point numbers of points

[flippmoke]

https://github.com/mapbox/geometry.hpp/blob/master/include/mapbox/geometry/point.hpp#L25

template <typename T>
constexpr bool operator==(point<T> const& lhs, point<T> const& rhs)
{
    return lhs.x == rhs.x && lhs.y == rhs.y;
}

Would it be better in the case of double or float that we used an approximate equality here?

[jfirebaugh]

The appropriate epsilon for approximate equality varies by context, so I think if you need approximate equality you're best off creating your own function that compares the way you want. That said, I can see an argument for straight up disabling operator== for floating point coordinate types.

[flippmoke]

The appropriate epsilon for approximate equality varies by context

While I agree, I think a reasonable assumption for a general case could be made?

template <typename T>
bool values_are_equal(T a, T b) 
{
    return a == b;
}

template <>
bool values_are_equal<double>(double a, double b) 
{
    return std::fabs(a - b) < (3.0 * std::numeric_limits<double>::epsilon());
}   

However, if we could not find a reasonable balance I would suggest disabling as well.

[joto]

The Catch unit test framework has an Approx class which is used to compare doubles. Maybe the implementation and docs can provide some insight.

[artemp]

@joto - thanks.
There is also a comprehensive peer-reviewed docs and implementation :
http://www.boost.org/doc/libs/1_65_1/libs/math/doc/html/math_toolkit/float_comparison.html