For the second part of your question, of avoiding the bugs — it was possible to solve this problem using long long arithmetic alone, there was no need for doubles / floating point arithmetic.

I often use this technique: instead of storing a fraction a/b, I store 2 integers a and b. Then to check if a/b < c/d, I check if a*d < b*c, as its "if and only if".

Special care needs to be taken for no overflow, but thats easier than debugging floating point errors.

Yep, exact same thing happened to me with my python code which behaved same as your WA code.

I noticed while testing that this

#include <stdio.h>

int main(void)
{
    printf("%Lf\n", 1.0l);
}

Outputs 0.000000 while changing stdio.h into cstdio and adding std:: results in 1.000000. It doesn't seem to be related to your problem but I find this very weird. It also outputs 1.000000 when I compile as C.

The first two numbers outputted by this:

#include <iostream>
#include <iomanip>
using namespace std;

int main()
{
	cout << setprecision(20);
	cout << 999977.0 * 249944 / 249943 - 499998.0 << '\n';
	for (int i = 249944; i <= 249946; i++)
	{
		cout << 999977.0 * i / 249943 - 499998.0 << '\n';
	}
}

Are different on custom invocation with GNU G++17 7.3.0. I'm really puzzled on what's going on since I don't see any undefined behavior and IEEE 754 conformant floats should round consistently. IIRC GCC optimizations shouldn't affect the output unless a flag is used to specifically allow optimizations affecting floating point rounding.

Edit: It's because Codeforces G++17 7.3.0 apparently isn't IEEE 754 conformant for some reason. See https://codeforces.me/blog/entry/21844.