Why does Python's dis dislike lists?

python, python-2.x

Solution

Short Answer

In Python 2.x, the `str` type holds raw bytes, so `dis` assumes that if you pass it a string it is getting compiled bytecode. It tries to disassemble the string you pass it as bytecode and -- purely due to the implementation details of Python bytecode -- succeeds for `i in (2,3)`. Obviously, though, it returns gibberish.

In Python 3.x, the `str` type is for strings and the `bytes` types is for raw bytes, so `dis` can distinguish between compiled bytecode and strings -- and assumes it is getting source code if it gets a string.

Long Answer

Here's the thought process I followed to work this one out.

I tried it on my Python (3.2):

>>> import dis
>>> dis.dis("i in (2,3)")  
  1           0 LOAD_NAME                0 (i)
              3 LOAD_CONST               2 ((2, 3))
              6 COMPARE_OP               6 (in)
              9 RETURN_VALUE
>>> dis.dis("i in [2,3]")
  1           0 LOAD_NAME                0 (i)
              3 LOAD_CONST               2 ((2, 3))
              6 COMPARE_OP               6 (in)
              9 RETURN_VALUE

Obviously, this works.

I tried it on Python 2.7:

>>> import dis
>>> dis.dis("i in (2,3)")
          0 BUILD_MAP       26912
          3 JUMP_FORWARD    10272 (to 10278)
          6 DELETE_SLICE+0
          7 <44>
          8 DELETE_SLICE+1
          9 STORE_SLICE+1
>>> dis.dis("i in [2,3]")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Python27\lib\dis.py", line 45, in dis
    disassemble_string(x)
  File "C:\Python27\lib\dis.py", line 112, in disassemble_string
    labels = findlabels(code)
  File "C:\Python27\lib\dis.py", line 166, in findlabels
    oparg = ord(code[i]) + ord(code[i+1])*256
IndexError: string index out of range

Aha! Notice also that the generated bytecode in Python 3.2 is what you would expect ("load `i`, load `(2,3)`, test for membership, return the result") whereas what you have got in Python 2.7 is gibberish. Clearly, `dis` is decompiling the string as bytecode in 2.7 but compiling it as Python in 3.2.

I had a look in the source code for `dis.dis`. Here are the key points:

Python 2.7:

elif isinstance(x, str):
    disassemble_string(x)

Python 3.2:

   elif isinstance(x, (bytes, bytearray)): # Raw bytecode
       _disassemble_bytes(x)
   elif isinstance(x, str):    # Source code
       _disassemble_str(x)

Just for fun, let's check this by passing the same bytes to `dis` in Python 3:

>>> dis.dis("i in (2,3)".encode())
          0 BUILD_MAP       26912
          3 JUMP_FORWARD    10272 (to 10278)
          6 <50>
          7 <44>
          8 <51>
          9 <41>

Aha! Gibberish! (Though note that it's slightly different gibberish -- the bytecode has changed with the Python version!)

Problem

Original source