attach

You can install this Python package using pip install attach.

How does this work?

Here is an example session:

>>> from attach import Namespace, attach
>>>
>>> foo = Namespace()
>>> with attach(foo):
...    bar = 'bar'
...    baz = 'baz'
...
>>>
>>> bar
Traceback (most recent call last):
    ...
NameError: name 'bar' is not defined

Notice that since we defined bar inside the attach(foo) context, bar does not exist as a global. Instead, it's value is saved to the namespace foo:

>>> foo.bar
'bar'

We can reattach the namespace later:

>>> with attach(foo):
...     print(bar)
...
bar

What’s the point?

The module is especially useful inside Jupyter notebooks. Quite often, we have constructions like this:

X = np.array([1, 2, 3])
def increment(X):
    return X + 1

In other words, we have globals and function parameters or local variables by the same name. However, this can lead to unintended references in the case of a typo, or difficulty keeping everything modular.

By keeping variables out of globals and inside namespaces, you can force functions to only use the variables that have been explicitly passed in, thereby preventing many tricky bugs!

Isn’t hacking globals evil?

Yes, for good reason, but this is not meant to be used in code files or in production, bur rather in exploratory Jupyter notebooks.

What’s this Namespace class?

You can pass any dictionary-like object into attach(); it does not have an instance of the Namespace class defined in this package.

The Namespace class just has a few niceties, like a nice string representation, and allowing you to attributes both as foo.bar and foo['bar'].

Stuff you should know

• By default, variables beginning with an underscore are not saved to the namespace; they are lost. To change this behaviour, set skip_underscored=False in attach().

• attach() only concerns itself with globals. If you call it inside a function, beware unexpected behaviour.

To Do

• Add a read_only paramater to attach(), which enables using namespaces in a nested way when neeed.
• Check if being called inside a function, and error out unless read_only=True since we can’t modify function locals, only globals.
• Move tests into their own file.