Hello,

I have a function "angle" that returns the angle (in degrees) of a right triangle given the length of the its lengths as type "real". It works, but when trying to test the equality of the angle with the seemingly identical value, I'm informed that the two are not equal. Thus, when angleDeg(sqrt(3.0)/2.0, 1.5) returns 60, the comparison "angleDeg(sqrt(3.0)/2.0, 1.5) == 60" resolves to false.

I'm most likely stupid and sleep deprived, but if I may have some elucidation, I'd be very grateful.

Some code:

real angleDeg(real x, real y)
{
	import std.math : PI, abs, atan2, sgn;

	auto angle = atan2(y, x);
	
	// Simplify angle if vector resides in quadrants II or III
	if (x < 0)
	{
		angle = (PI - abs(angle)) * sgn(angle);
	}

	return angle * 180 / PI;
}

import std.math : sqrt;

assert(angleDeg(sqrt(3.0)/2.0, 1.5) == 60.0); // Fails

Typo in the first paragraph: angle -> angleDeg

16.11.2019 13:01, Avery Stupidman пишет:
> Hello,
> 
> I have a function "angle" that returns the angle (in degrees) of a right triangle given the length of the its lengths as type "real". It works, but when trying to test the equality of the angle with the seemingly identical value, I'm informed that the two are not equal. Thus, when angleDeg(sqrt(3.0)/2.0, 1.5) returns 60, the comparison "angleDeg(sqrt(3.0)/2.0, 1.5) == 60" resolves to false.
> 
> I'm most likely stupid and sleep deprived, but if I may have some elucidation, I'd be very grateful.
> 
> Some code:
> 
> real angleDeg(real x, real y)
> {
>      import std.math : PI, abs, atan2, sgn;
> 
>      auto angle = atan2(y, x);
> 
>      // Simplify angle if vector resides in quadrants II or III
>      if (x < 0)
>      {
>          angle = (PI - abs(angle)) * sgn(angle);
>      }
> 
>      return angle * 180 / PI;
> }
> 
> import std.math : sqrt;
> 
> assert(angleDeg(sqrt(3.0)/2.0, 1.5) == 60.0); // Fails
In general you shouldn't compare floating point values for equality because of their nature, you can use something like approxEqual https://dlang.org/library/std/math/approx_equal.html

P.S. see also https://floating-point-gui.de/errors/comparison/

On Saturday, 16 November 2019 at 10:17:12 UTC, drug wrote:
> 16.11.2019 13:01, Avery Stupidman пишет:
>> [...]
> In general you shouldn't compare floating point values for equality because of their nature, you can use something like approxEqual https://dlang.org/library/std/math/approx_equal.html
>
> P.S. see also https://floating-point-gui.de/errors/comparison/

Yup, I'm dumb. Thank you very much for the help and the link. I should know better. Also apologies for the un-concise example.

On Saturday, 16 November 2019 at 10:01:25 UTC, Avery Stupidman wrote:
> assert(angleDeg(sqrt(3.0)/2.0, 1.5) == 60.0); // Fails

On my computer the results are very close, that is:

import std.math: nextUp;

assert(nextUp(angleDeg(sqrt(3.0)/2.0, 1.5)) == 60.0); // OK


On Saturday, 16 November 2019 at 10:17:12 UTC, drug wrote:
> you can use something like approxEqual https://dlang.org/library/std/math/approx_equal.html

With approxEqual, one needs to be very carefull. The default parameters allow for very odd results, accepting values to be equal, which are actually far apart considering the precision available; this is especially true, when using double or real.

For example, here are about 12 significant fractional digits involved, which are completely ignored:

assert(approxEqual(100.0,101.0)); // OK

I wrote a replacement with better handling of the default parameters [1], but it looks like no one wants to take the burden to review it, probably because that would involve to make a decision. (Or maybe it has some other reasons I do not see.)

[1] https://github.com/dlang/phobos/pull/7241