Allow custom precision in error reports for floating-point numbers

[mlimber]

Description
As in this SO Q&A, it would be nice to allow customization of the precision of floating-point values that fail. At present, the precision for float and double are hard-coded within Catch2 to 5 and 10, respectively:

std::string StringMaker<float>::convert(float value) {
    return fpToString(value, 5) + 'f';
}
std::string StringMaker<double>::convert(double value) {
    return fpToString(value, 10);
}

Sometimes things will fail (even with Approx()), but the precision of the failure message doesn't show enough precision to be meaningful.

prog.cc:22: FAILED:
  REQUIRE( TestType(0) == std::numeric_limits<TestType>::epsilon() )
with expansion:
  0.0f == 0.0f

The proposal here is to add a setPrecision() function to allow the user to control this. I have prototyped this, which you can see running live on Wandbox. Here's the test:

TEMPLATE_TEST_CASE( "Double, double, toil and trouble", "[double]", float, double ) 
{
    const auto precision = GENERATE( -1, 0, 3, std::numeric_limits<TestType>::max_digits10 );

    if( precision >= 0 )
    {
        Catch::StringMaker<TestType>::setPrecision( precision );
    }
    
    // Expected to fail to demonstrate the problem
    REQUIRE( TestType(0) == std::numeric_limits<TestType>::epsilon() ); 
}

The output shows as expected for floats and doubles. For instance:

prog.cc:15: FAILED:
  REQUIRE( TestType(0) == std::numeric_limits<TestType>::epsilon() )
with expansion:
  0.0f == 0.000000119f

Making a unit test for this is a little iffy since the actual string values are dependent on the usual floating-point fuzziness. I guess it could compare string length without looking too closely at the value.

[ecorm]

Is Catch::StringMaker<T>::setPrecision sticky, or does it only affect the current test case?

[horenmar]

It is sticky. For ease of implementation, it is a type-level static.