Why is `if x is None: pass` faster than `x is None` alone?

March 26, 2024

Timing results in Python 3.12 (and similar with 3.11 and 3.13 on different machines):

When x = None:
13.8 ns  x is None
10.1 ns  if x is None: pass

When x = True:
13.9 ns  x is None
11.1 ns  if x is None: pass

Python: 3.12.0 (main, Oct  7 2023, 10:42:35) [GCC 13.2.1 20230801]

How can doing more take less time?

Why is if x is None: pass faster, when it has to do the same x is None check and then additionally checks the truth value of the result (and does or skips the pass)?

Times on other versions/machines:

Python 3.11: (12.4, 9.3) and (12.0, 8.8)
Python 3.13: (29.1, 24.6) and (29.1, 22.6)

Benchmark script (Attempt This Online!):

from timeit import repeat
import sys

for x in None, True:
    print(f'When {x = }:')
    for code in ['x is None', 'if x is None: pass'] * 2:
        t = min(repeat(code, f'{x=}', repeat=100))
        print(f'{t*1e3:4.1f} ns ', code)
    print()

print('Python:', sys.version)

>Solution :

Look at the disassembled code:

>>> import dis
>>> dis.dis('if x is None: pass')
  0           0 RESUME                   0

  1           2 LOAD_NAME                0 (x)
              4 POP_JUMP_IF_NOT_NONE     1 (to 8)
              6 RETURN_CONST             0 (None)
        >>    8 RETURN_CONST             0 (None)
>>> dis.dis('x is None')
  0           0 RESUME                   0

  1           2 LOAD_NAME                0 (x)
              4 LOAD_CONST               0 (None)
              6 IS_OP                    0
              8 RETURN_VALUE

The if case has a special POP_JUMP_IF_NOT_NONE operation, which is faster than a LOAD_CONST plus IS_OP. You can read the detailed discussion about it here: https://github.com/faster-cpython/ideas/discussions/154.