>>> def func(test):
...     if test == 1:
...         return 1
...     elif test == 2:
...         return 2
...     return 3
... 
>>> import dis
>>> dis.dis(func)

Python 3.5:

2 0 LOAD_FAST 0 (test)
3 LOAD_CONST 1 (1)
6 COMPARE_OP 2 (==)
9 POP_JUMP_IF_FALSE 16

3 12 LOAD_CONST 1 (1)
15 RETURN_VALUE

4 >> 16 LOAD_FAST 0 (test)
19 LOAD_CONST 2 (2)
22 COMPARE_OP 2 (==)
25 POP_JUMP_IF_FALSE 32

5 28 LOAD_CONST 2 (2)
31 RETURN_VALUE

6 >> 32 LOAD_CONST 3 (3)
35 RETURN_VALUE

Python 3.6:

2 0 LOAD_FAST 0 (test)
2 LOAD_CONST 1 (1)
4 COMPARE_OP 2 (==)
6 POP_JUMP_IF_FALSE 14

3 8 LOAD_CONST 1 (1)
10 RETURN_VALUE
12 JUMP_FORWARD 12 (to 26)

4 >> 14 LOAD_FAST 0 (test)
16 LOAD_CONST 2 (2)
18 COMPARE_OP 2 (==)
20 POP_JUMP_IF_FALSE 26

5 22 LOAD_CONST 2 (2)
24 RETURN_VALUE

6 >> 26 LOAD_CONST 3 (3)
28 RETURN_VALUE

Note JUMP_FORWARD after RETURN_VALUE in 3.6 listing.

Proposed patch fixes removing unreachable code after RETURN_VALUE in peephole optimizer.

Hi Serhiy,

Could you help me, because I don't understand your patch ? because a RETURN_VALUE will go to the end of the block, and in this case, I don't understand why there is a JUMP_FORWARD at 12 in Python 3.6.

+0 The situation this addresses isn't common and the patch will rarely produce a perceptable benefit (just making the disassembly look a little nicer). That said, change looks simple enough and doesn't add any overhead.

New changeset 5784cc37b5f4 by Serhiy Storchaka in branch '3.6':
Issue bpo-28517: Fixed of-by-one error in the peephole optimizer that caused
https://hg.python.org/cpython/rev/5784cc37b5f4

New changeset 8d571fab4d66 by Serhiy Storchaka in branch 'default':
Issue bpo-28517: Fixed of-by-one error in the peephole optimizer that caused
https://hg.python.org/cpython/rev/8d571fab4d66

The main problem was that this bug caused unnecessary breakage of tests in Victor's bytecode [1].

Stéphane, the number of opcodes to the end of the block was counted incorrectly. Just off-by-one error. One unreachable opcode always was left.

[1] https://github.com/haypo/bytecode

ah ok, thanks for the explanation.

the off by one error fix here introduced a new off by one error. PR coming, follow https://bugs.python.org/issue35193 for that.