Full precision formatting is desirable for floating point numbers

[9il]

It is essential to see the whole number when debugging numeric algorithms.

• If the printed numbers are the same, I can't guess how different their actual values are.
• It is very useful to see the printed number and copy past it to the test: that is a typical case when the corresponding numeric algorithm has been updated.

[atilaneves]

Unless you give me an example of what happens now and what should happen instead, I don't know how I can fix this.

[9il]

It would be much more helpfull to print floating points in assert messages of should and shouldApproxEqual in sciencific notation with 16 decimal points after the dot using C library. Phobos gives wrong results (if it doesn't use C library for sciencific notation, I am not sure about if it is). Ofcourse, you are wellcome to use Mir's to!string, it will give the full precision in minimal form. C's sciencific notation is good too.

[9il]

Unitthreaded example:

unittest
{
    import unit_threaded;
    auto a = 1e-7;
    a.shouldApproxEqual(0, 1e-8, 1e-8);
}
unit_threaded.exception.UnitTestException@source/mir/complex/math.d(376): Expected approx: 0
     Got       : 0.000000

The same example with mir.test

unittest 
{
    import mir.test;
    auto a = 1e-7;
    a.shouldApprox(1e-8, 1e-8) == 0;
}
mir.exception.MirError@source/mir/date.d(4473): expected approximately 0.0, got 1e-7, maxRelDiff = 1e-8, maxAbsDiff = 1e-8