This is the most obfuscated “Hello, world!” program I have ever seen!

The code

Here is a monstrous piece of Python code:

___ = -8
@lambda _: _(__ for _, __ in _.__dict__.items() if -len(_) > ___)
class _(list):
    """Monty Python is obviously arousing parrots!"""
    def __format__(_, __):
        _.__class__._ = property(lambda _: eval(_[0][___::___])(__))
        return "Spam, spam, spam, ..."

def __() -> f"{_:Hello, world!}": ...

_._

This is a valid Python program, in Python 3.9+. If you run this program, the output you get is "Hello, world!", but how?

This program is the most obfuscated “Hello, world!” program I have ever seen, taking the spot from this very obscure “Hello, world!” Python program that I wrote about yesterday.

Background

I will explain this program assuming you've read the article about this simpler version:

@lambda _: _()
class _:
    def __format__(_, __):
        _.__class__._ = property(lambda _: print(__))
        return ""

def __() -> f"{_:Hello, world!}": ...

_._

As a quick reminder, the simpler version works because we define a class that is immediately replaced by an instance of itself. Then, that instance's dunder method __format__ is called when the return annotation is evaluated in the function definition, which ends up binding the call to print to a property called _. That call to print has the format specification __ as argument, which happens to be "Hello, world!". Finally, the property of the instance is accessed, triggering the print.

If you would like a step-by-step explanation of how the simpler version works, you can find it here.

Understanding the source

The version we want to study now is mostly the same as the simpler one, apart from a few changes:

___ = -8  # <- new; v different v
@lambda _: _(__ for _, __ in _.__dict__.items() if -len(_) > ___)
class _(list):  # <- now inheriting from `list`
    """Monty Python is obviously arousing parrots!"""
    def __format__(_, __):
        # The property here is slightly different:
        _.__class__._ = property(lambda _: eval(_[0][___::___])(__))
        # The return was empty:
        return "Spam, spam, spam, ..."

# ...

I started by trying to remove docstring from the class and by making the return the empty string again, but removing the docstring broke the program. So, the docstring is needed but the return isn't:

___ = -8
@lambda _: _(__ for _, __ in _.__dict__.items() if -len(_) > ___)
class _(list):
    """Monty Python is obviously arousing parrots!"""
    def __format__(_, __):
        _.__class__._ = property(lambda _: eval(_[0][___::___])(__))
        return ""

def __() -> f"{_:Hello, world!}": ...

_._

Then, I looked at the property definition. It is quite similar to the previous one, apart from what follows the lambda _: ...:

# old v
property(lambda _:        print        (__))
property(lambda _: eval(_[0][___::___])(__))
# new ^

So, it looks like eval(_[0][___::___]) is supposed to be replacing print... But how?

If we look at the decorator, we can see it also looks similar:

# old v
@lambda _: _()
@lambda _: _(__ for _, __ in _.__dict__.items() if -len(_) > ___)
# new ^

The difference is that we build an instance of the class with an argument, and the class itself inherits from list, so we are building a list with some stuff in it.

The good news is that we can just print the instance itself to see what it contains! If we add print(_) to the end of the program and if we run it, this is what we get:

['Monty Python is obviously arousing parrots!']

Interesting! The generator expression inside the lambda decorator is the iterable that will be used to build the list, and therefore that generator expression must be filtering everything out of _.__dict__.items() except for the docstring.

The lambda decorator receives the class and then looks at the items in its dunder name __dict__. We can inspect that __dict__ by running the program with python -i, which runs the whole program and then drops us into a REPL where everything from the program has been defined. If we do that, we can play around with a couple of things:

>>> list(_.__class__.__dict__.keys())
['__module__', '__doc__', '__format__', '__dict__', '__weakref__', '_']
>>> list(name for name in _.__class__.__dict__.keys() if -len(name) > ___)
['__doc__', '_']

We can see that the constant ___ = -8 defined at the top of the program is being used to filter the keys of this dictionary. Only two keys are short enough: "__doc__" and _. However, _ is the property that only gets added after the lambda decorator runs, so when the program runs for the first time, the generator expression

@lambda _: _(__ for _, __ in _.__dict__.items() if -len(_) > ___)

only finds the key "__doc__" and then uses its corresponding value, which is the string "Monty Python is obviously arousing parrots!".

We know how this funky string ends up inside the list, now we just need to figure out how it is used to build the function print.

The property that we define contains an interesting expression inside the lambda:

eval(_[0][___::___])

_ refers to the list itself and _[0] is the string "Monty Python is obviously arousing parrots!". Then, [___::___] is slicing syntax, and it is equivalent to [-8::-8] because that is the value of ___. Incidentally, the slice builds the string "print":

>>> s = "Monty Python is obviously arousing parrots!"
>>> s[-8::-8]
'print'

The final step is to use eval to transform the string into the built-in print:

>>> eval("print")
<built-in function print>

That is how this works! Insane, right?

If you come up with an even crazier “Hello, world!” program, feel free to comment it below or send it to me on Twitter, for example.

The short explanation

The short explanation is that the lambda decorator is creating a list whose only element is the docstring of the class. Then, with some funky slicing, we use that string to build the string "print" and then eval it into the print function, which then prints the formatting specification as the simpler version of this program.

Source

The program we studied was not authored by me. In fact, it was sent to me in the comment section of a Reddit post where I was sharing another obscure “Hello, world!” Python program.

Become a better Python 🐍 developer 🚀

+35 chapters. +400 pages. Hundreds of examples. Over 30,000 readers!

My book “Pydon'ts” teaches you how to write elegant, expressive, and Pythonic code, to help you become a better developer. >>> Download it here 🐍🚀.