Seriously, what's the difference? When you first fire up the Python interpreter, __builtins__ is in your namespace for free:

>>> globals().keys()
['__builtins__', '__name__', '__doc__', '__package__']
>>> __builtins__
<module '__builtin__' (built-in)>

But it appears to be the __builtin__ module (singular)! If you:

>>> import __builtin__
>>> __builtin__ is __builtins__
True

Hrm. So they are both names that point to the same object, the module __builtin__. Weird. Why does Python do this? Do they always behave the same?

I read on StackOverflow that

By default, when in the __main__ module, __builtins__ is the built-in module __builtin__ (note: no 's'); when in any other module, __builtins__ is an alias for the dictionary of the __builtin__ module itself.

What. What does that mean.

This talk of the "__main__ module" and "any other module" reminds me of a sequence of words that I've known for quite a while, but haven't completely grokked:

We can access the name of the current module with the builtin variable __name__.

You're probably familiar with the related canonical statement:

if __name__ == '__main__':
    main()

But what does "current module" mean? What does the __name__ variable look like when it does not equal __main__?

I happen to know, because I've obsessively read about the import statement, another sequence of words:

Any code executed as a result of an import isn't executed in the __main__ module.

Let's use these bits of knowledge to observe the behavior of __builtins__ both inside and outside of the __main__ module. We can also check out the __name__ variable while we're at it.

First, let's make a script, a.py, which will allow us to observe the behavior of __builtin__, __builtins__, and __name__.

import __builtin__

print "In a"
print "__name__ is:", __name__
print "__builtin__ is __builtins__:", __builtin__ is __builtins__
print "type(__builtin__):", type(__builtin__)
print "type(__builtins__):", type(__builtins__)

Let's see what happens when we execute a.py:

$ python a.py
In a
__name__ is: __main__
__builtin__ is __builtins__: True
type(__builtin__): <type 'module'>
type(__builtins__): <type 'module'>

Okay. So we're in the __main__ module, and in here __builtin__ is pointing to the same module object as __builtins__.

What happens if we import a in another script? The code in a will execute, but it won't be executed within the __main__ module. Instead, it'll be executed within the a module. Let's write another script, b.py, to find out what happens to __builtins__ outside of __main__:

import __builtin__

print "In b, before importing a"

# the output from this should be the same as when we ran
# $ python a.py
print "__name__ is:", __name__
print "__builtin__ is __builtins__:", __builtin__ is __builtins__
print "type(__builtin__):", type(__builtin__)
print "type(__builtins__):", type(__builtins__)
print "\n"

import a
# code from a will execute here

Let's see what happens when we run b.py:

$ python b.py
In b, before importing a
__name__ is: __main__
__builtin__ is __builtins__: True
type(__builtin__): <type 'module'>
type(__builtins__): <type 'module'>


In a
__name__ is: a
__builtin__ is __builtins__: False
type(__builtin__): <type 'module'>
type(__builtins__): <type 'dict'>

Aha. So when we're outside the context of the __main__ module, __name__ is just equal to the name of the module where code is currently being executed. That seems logical. And outside of __main__, __builtins__ is a dict, rather than a module.

We were told earlier that, outside the context of __main__, "__builtins__ is an alias for the dictionary of the __builtin__ module". I think that means that __builtins__ is __builtin__.__dict__. Let's see if my hypothesis is true, by adding another line to the bottom of our a.py file:

print "__builtins__ is __builtin__.__dict__", __builtins__ is __builtin__.__dict__

Running b.py again, we get:

$ python b.py
In b, before importing a
__name__ is: __main__
__builtin__ is __builtins__: True
type(__builtin__): <type 'module'>
type(__builtins__): <type 'module'>


In a
__name__ is: a
__builtin__ is __builtins__: False
type(__builtin__): <type 'module'>
type(__builtins__): <type 'dict'>
__builtins__ is __builtin__.__dict__ True

Yes! My hypothesis was correct. Okay. So now I get why using __builtin__ is better than __builtins__:

The type, and thus behavior, of __builtins__ changes based on the context of where it's being executed, while the type and behavior of __builtin__ is constant. Rad.

Thanks, stranger who suggested I look into this, for the learning opportunity. And thanks, always, to Allison Kaptur, for exploring this topic with me.

The code for this blog post is on GitHub, of course.

tags: python builtins hacker school python internals