Today when I am trying to solve acmsguru-113 with PyPy3 (because when I was submitting Python3, CodeForces noticed me that "Almost always, if you send a solution on PyPy, it works much faster"), I got lots of Wrong Answer on test 1. I thought it impossible because it's actually a very simple problem and I have confidence in my code. Then I rewrite it in C++, and with identical implementation it's accepted.

Then I realize there may be problems in input format, so I write a tokenizer to solve this problem. However, it's just another WA on 1.

After one hour of struggling in finding bugs and differences, I found nothing but submitted it with Python3 — and it's a accepted!

Here are the python code with tokenizer, and submissions with this code : 63721685 63721667

now_token = input().split()
now_token.reverse()
def nxt_token():
    global now_token
    if len(now_token) > 0:
        ret = now_token[-1]
        now_token.pop()
        return ret
    else:
        now_token = input().split()
        now_token.reverse()
        return nxt_token();

n = int(nxt_token())
while n > 0:
    n -= 1
    x = int(nxt_token())
    ans = False
    i = 2
    while i * i <= x:
        if x % i == 0:
            y = x // i
            ans = (y > 1)
            j = 2
            while j * j <= y:
                if y % j == 0:
                    ans = 0
                    break
                j += 1
            break
        i += 1
    if ans:
        print("Yes")
    else:
        print("No")

It's true that the data is not corresponding to the input format, but there must be something else with the PyPy3 interpreter.