Let Approx::epsilon also take float

[tomvierjahn]

Description

• Approx::Approx and its comparison operators can now take float
• Adding this feature to Approx::epsilon would improve consistency and code

Somehow related to #873

Extra information

• Catch version: v1.9.1 (Generated: 2017-04-09 21:21:06.285364)

[horenmar]

I spent some time thinking about this, then suddenly got really busy and couldn't respond.

The reason why this wasn't a no-brainer change was that it isn't quite so simple as to add a float overload for the function, because that would trigger compile-time error for code like Approx(2).margin(2). The solution that we are using is to add a templated overload, that uses SFINAE to only work with types that can be converted to double.

Now, for Approx::margin this isn't semantically problematic, as Approx(temperature(72)).margin(temperature(2)), makes perfect sense. However for Approx::epsilon the situation is not quite so clear-cut, as Approx(temperature(72)).epsilon(temperature(0.001)) does not make semantic sense.

This could be solved in two ways.

1. Specifically make Approx::epsilon only accept floating point types (using std::is_floating_point).
2. Damn the semantics, make Approx::epsilon accept anything.

Personally I like the second option better.

There is also Approx::scale, but I don't think anybody is actually using that one.

[horenmar]

After discussion with Phil, I decided to go with option 2.