dis.dis(ob) currently accepts "a module, a class, a method, a function,
or a code object." But for most uses I have seen on python-list, people
start with a code snippet. They must then wrap that in a function or
remember (or lookup) a call such as compile(code, '', 'exec') to make a
code object. I propose that dis do the latter automatically. Dis already
has to branch on the input class, so I assume adding another branch
should not be difficult.

On the Python ideas list, Steven D'Aprano raised the issue of 'exec'
versus 'single' versus 'eval'. As far as dis is concerned, there seems
to be no difference between 'exec' and 'single'.

>>> dis(compile('x = x+1', '', 'single'))
  1           0 LOAD_NAME                0 (x) 
              3 LOAD_CONST               0 (1) 
              6 BINARY_ADD           
              7 STORE_NAME               0 (x) 
             10 LOAD_CONST               1 (None) 
             13 RETURN_VALUE         

>>> dis(compile('x = x+1', '', 'exec'))
  1           0 LOAD_NAME                0 (x) 
              3 LOAD_CONST               0 (1) 
              6 BINARY_ADD           
              7 STORE_NAME               0 (x) 
             10 LOAD_CONST               1 (None) 
             13 RETURN_VALUE         

Using 'exec' instead of 'eval' adds two spurious, but easily ignored, lines.

>>> dis(compile('x+1','', 'eval'))
  1           0 LOAD_NAME                0 (x) 
              3 LOAD_CONST               0 (1) 
              6 BINARY_ADD           
              7 RETURN_VALUE    
     
>>> dis(compile('x+1', '', 'exec'))
  1           0 LOAD_NAME                0 (x) 
              3 LOAD_CONST               0 (1) 
              6 BINARY_ADD           
              7 POP_TOP              
              8 LOAD_CONST               1 (None) 
             11 RETURN_VALUE         

Between the current doc sentences "For a single code sequence, it prints
one line per bytecode instruction." and "If no object is provided, it
disassembles the last traceback." I propose adding the following two
sentences.

"Strings are first compiled as statements to code objects with
compile(string,'','exec'). For expressions, this adds a spurious POP_TOP
and LOAD_CONST at the end."

'compile' should be cross-referenced to its listing under built-in
functions.

Copying my suggestion (minus examples) over from the python-ideas thread:

We could define it as trying the three modes in order (first 'eval',
then 'single', then 'exec') moving on to the next option if it raises
syntax error:

from dis import dis
def dis_str(source):
  modes = ('eval', 'single', 'exec')
  for mode in modes:
    try:
      c = compile(source, '', mode)
      break
    except SyntaxError:
      if mode is modes[-1]:
        raise
  return dis(c)

As I explained on python-ideas, 'single' should not be tried.

As per Georg's suggestion, a better approach would look like:

from dis import dis
def dis_str(source):
  try:
    c = compile(source, '', 'eval')
  except SyntaxError:
    c = compile(source, '', 'exec')
  return dis(c)

Trying both 'eval' and 'exec' looks fine to me.
Marking 'easy' because I expect it should be ;-)

I've made a patch, which adds a disassemble_str function to the dis module. The dis.dis function calls this function if x is a string. Added the following sentence to the documentation: "Strings are first compiled to code objects with the :func:`compile` built-in function." Added two simle unittests.

+1

Any chance, that my patch will be accepted? Is there a problem with it?

Thanks.

disassemble_str should be private with an underscore.

Done. Attached new patch as issue6507_2_priv.diff.

Missed the window for 2.7, but should be right for 3.2.

There's a minor error in the documentation (strings need to be mentioned in the list of acceptable types), but I can fix that on commit.

disassemble_str should be private with an underscore.

disassemble_string should've been private as well, while we are editing this module.

disassemble_string should've been private as well, while we are editing this module.

Attached the updated patch.

Just today, someone posted the result of dis.dis('somebytes') and did not notice the error because dis blithely disassembles bytes as bytecodes, even in 3.x. (The person actually dissed a 2.x string).

>>> from dis import dis
>>> dis(b'cat')
          0 DUP_TOPX        29793

It is a natural thing to do, so I hope this is put in 3.2.

Since the undocumented 'disassemble_string' now disassembles bytes, I think it should be renamed '_disassemble_bytes' instead of '_disassemble_string'. This would accord with the general effort to remove 2.x fossils from 3.x.

Aside from that, it looks ready, from a reading review, to apply and test: doc addition, added tests, new function and else case, and rename.

Committed (with some minor modifications) in r82471.

Inspired by Yanov Aknin's ssc() tool, I've opened a new RFE (bpo-9147) for a similarly enhanced show_code() implementation.