detail/voronoi_robust_fpt.hpp about polygon

[asydorchuk]

Yes, I will look into it.

Да, я посмотрю.

[LLlKuIIeP]

Good afternoon.

Are there any successes?

[eadf]

Is there really an issue with divide by zero here?
To me it looks like the err value intentionally is set to infinity whenever additions or subtractions results in zero.
What took me a while to figure out was if the divide by zero could result in NaN (0/0).

That would be a problem, since NaN compared to any ulp value would be false, but it doesn't look like it can happen

[LLlKuIIeP]

Good day.

Let's connect colleagues to clarify division by zero.

[meretin-as]

Is there really an issue with divide by zero here?
To me it looks like the err value intentionally is set to infinity whenever additions or subtractions results in zero.
What took me a while to figure out was if the divide by zero could result in NaN (0/0).

That would be a problem, since NaN compared to any ulp value would be false, but it doesn't look like it can happen

Problem occur when you test your code with some kind of sanitizers, which check division by zero. So in current code realization sanitizer fails even in correct data because of division. If I understood you right, your suggestion is to use isnan or something like that, but it will not help to solve our problem.

[eadf]

Your sanitizer problem sounds like a documentation/comment problem, not a problem with the program code itself.
Maybe there's a zero cost abstraction operation that can be used?
i_know_what_im_doing_let_me_divide_by_zero(n,d) :)

My worries was if that division by zero could result in NaN, but it does not seem (from what i can tell) like that can ever happen. That's why I closed the PR

My local copy has an assert test for that nan condition, and it has not triggered yet.