The sync version of the TestKit backend was over 4 times as slow as its async counterpart. The reason is that the TCPServer shipped with Python hands out an unbuffered writer by default.
I don't quite understand why this wasn't a problem earlier. The sync backend code did not change, neither did the TCPServer implementation in the standard library. It shall remain a mystery insert mystical sound effect.
The sync version of the TestKit backend was over 4 times as slow as its async counterpart. The reason is that the TCPServer shipped with Python hands out an unbuffered writer by default.
I don't quite understand why this wasn't a problem earlier. The sync backend code did not change, neither did the TCPServer implementation in the standard library. It shall remain a mystery insert mystical sound effect.