30

If I open up interactive mode and type:

__builtins__ = 0 # breaks everything

have I completely broken the session? If so, what is going on behind the scenes to assign __builtins__ to the builtin module that can't be handled by the interpreter? If not, how can I recover from this?

Just a few of my own attempts to fix it:

  • Any attempt to import anything results in an error "ImportError __import__ not found"
  • all functions I might use to do anything other than evaluate numerical expressions are broken
  • There is another variable __package__ still accessible, but I don't know if/how it can be used.
Sven Marnach
  • 574,206
  • 118
  • 941
  • 841
Hart Simha
  • 862
  • 8
  • 14

3 Answers3

32

You can usually get access to anything you need, even when __builtins__ has been removed. It's just a matter of digging far enough. For example:

Python 2.7.3 (default, Apr 10 2012, 23:31:26) [MSC v.1500 32 bit (Intel)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> __builtins__ = 0
>>> open
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
NameError: name 'open' is not defined
>>> dir
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
NameError: name 'dir' is not defined
>>> int
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
NameError: name 'int' is not defined
>>> float
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
NameError: name 'float' is not defined
>>>
>>> __builtins__ = [t for t in ().__class__.__bases__[0].__subclasses__() if 'warning' in t.__name__][0]()._module.__builtins__
>>>
>>> open
<built-in function open>
>>> int
<type 'int'>
>>> float
<type 'float'>
>>>

For an explanation of what the heck just happened here, read Eval really is dangerous, where similar techniques are used to demonstrate that you cannot safely execute untrusted Python code.

Ned Batchelder
  • 364,293
  • 75
  • 561
  • 662
  • Just tried running it in python 3.. got an error. Notably, this: [t for t in ().__class__.__bases__[0].__subclasses__() if 'warning' in t.__name__] expands to [] when I try running it. Maybe this is a problem with python 3 vs 2? – Hart Simha Nov 09 '12 at 11:47
  • Slick! Is there any way to actually completely break a session? – acjay Nov 09 '12 at 13:29
  • 4
    @HartSimha: Python 2 enables a restricted mode if builtins in the current frame differs from the interpreter builtins. Since restricted mode was removed in Python 3, it's easy to modify this answer to get `__builtins__` from a function: `__builtins__ = [t for t in ().__class__.__base__.__subclasses__() if t.__name__ == 'Sized'][0].__len__.__globals__['__builtins__']`. – Eryk Sun Nov 09 '12 at 15:52
  • 9
    @acjohnson55, how about `__builtins__.__dict__.clear()`. I tried it, and now my shell is so broken it doesn't even try to execute what I give it. – Winston Ewert Nov 09 '12 at 16:14
  • 1
    @WinstonEwert: that is awful, in the sense of inducing awe bordering on terror. Such knowledge is not for mortals. – DSM Nov 09 '12 at 23:28
4

Basically messing with protected and reserved names means breaking your session, sometimes without a way to recover from.

For example, you can type in shell:

True = False # The chaos begins!

These are not possible with other programming languages, but python lets you do what you want, even if it'll break everything.

iTayb
  • 12,373
  • 24
  • 81
  • 135
  • I get a SyntaxError when I type True = False, even in the broken shell. If the session is completely broken, how does \_\_builtins\_\_ get assigned to the module in python in the first place. I assume the interpreter initializes it using an obscure statement and maybe some kind of low-level file access. But if not, how **IS** builtins assigned. – Hart Simha Nov 09 '12 at 11:34
  • 9
    This only works in Python 2. Python 3 will not allow you to assign things to keywords. But even in Python 2 this is easily fixed: `False = (False is not False)` and `True = not False` and you’re back to the originals. – poke Nov 09 '12 at 11:49
3

You're right; you can practically break a Python session. I doubt there's a way to completely destroy it - seeing Ned's answer was quite the revelation to me.

Being a very dynamic language, Python gives you a lot of rope to hang yourself with. Don't look at this as a flaw, though; a common Python slogan states that "we're all consenting adults here." If you understand the language and really know what you're doing, you have an insane amount of control over basically every aspect of Python.

Thane Brimhall
  • 9,256
  • 7
  • 36
  • 50
  • 2
    This is really true. That’s also a reason for why there are no accessibility modificators like `public` and `private` for members. When using something, you are expected to stay out of the internal implementation unless you know exactly what you do. – poke Nov 09 '12 at 11:51