My understanding is:

`loop` argument passed to sleep should be always the same as returned from `get_running_loop()`.

Passing it explicitly can be considered as microoptimization but `get_running_loop()` is pretty fast now, no need for such micro-opts.

On another hand passing *non-current* loop is a serious error: nothing prevents to do it but the code just hangs.