root of the issue: unlike other languages, C compares the expected and actual values with approximate equality post-rounding, not strict equality post-rounding. and the epsilon used to compare and print the numbers is 1e-5, while the description asks for rounding to 1e-4. also, there is a bug in the random tests that make them generate the same coordinates for every test.

here is a fork that fixes all that, also improves the description and makes it language-agnostic + Markdown instead of HTML

I'm assuming he's talking about C, seeing as that's the language he solved it in.

In what language does this happen?

From the description:

Note: the result should be rounded down to the nearest integer.