How Approx handle INF

[DarwinSenior]

So, I have a question regarding to Approx(INF), it seems that REQUIRE(Approx(INF) == INF) is false if I use float INF = std::numeric_limits<float>::infinity().

here is what I got

/Users/DarwinSenior/Desktop/cs419/tracer2/test/misc.cc:13: FAILED:
  REQUIRE( INF == Approx(INF) )
with expansion:
  inff == Approx( inf )

And my test case file

#include "catch.hpp"
#include <iostream>
#include <limits>

using namespace std;

namespace{
    float INF = std::numeric_limits<float>::infinity();

}

TEST_CASE("MISC", "[MISC]"){
    REQUIRE(INF == Approx(INF));
}

So, is there any way I could use Approx along with INF?

[lightmare]

Mathematically, Approx(inf) doesn't make sense. No finite number is "close enough" to infinity.

If you want to check that x is so large that adding something to it would overflow, resulting in +infinity, there's std::numeric_limits<float>::max().

CHECK(x >= std::numeric_limits<float>::max())

It's probably safe to use Approx with float max (not with double max, as that might overflow in operator==(double, Approx)). Although I still don't see the point, Approx is intended to check that calculation results are within a margin of error; not that they're (close to) overflowing.

[horenmar]

Why exactly do you want to have Approx of infinity?

[DarwinSenior]

That was long time ago for the course project when I decided to write test case for ray tracing. Using infinity for the the parallel cases. I will close this issue.

[horenmar]

Ok. I am not categorically opposed to making Approx infinity aware (although it would most likely start as "hey, don't do that" error on construction), but I wanted to know the motivation to do so.