为什么python子模块没有显示在dir（module）的列表中？

Question

This is probably a more general question, because I have seen similar behavior with other modules. But I will show the issue as I am seeing it with module matplotlib . I thought that dir(module) lists all attributes of the specified module. So I can do the following:

import matplotlib
dir(matplotlib)

Among the results listed is 'dateutil', but 'dates' is NOT listed by dir(matplotlib) . But then I do this:

import matplotlib.dates   ## why does this work? `dir(matplotlib)` did not list it
dir(matplotlib)           ## this now lists both 'dates' and 'dateutil'

Why does dir(matplotlib) now list 'dates' as one of its attributes?

What's going on here?

I just noticed now, it appears to me that dateutil is not actually part of matplotlib, but is imported by matplotlib:

In [11]: matplotlib.dateutil.__file__
Out[11]: '/anaconda3/lib/python3.6/site-packages/dateutil/__init__.py'

In [12]: matplotlib.__file__
Out[12]: '/anaconda3/lib/python3.6/site-packages/matplotlib/__init__.py'

In [13]: matplotlib.dates.__file__
Out[13]: '/anaconda3/lib/python3.6/site-packages/matplotlib/dates.py'

Notice in the output above, matplotlib.dateutil.__file__ shows a file that is not even in the matplotlib directory. Furthermore I can see in file '/anaconda3/lib/python3.6/site-packages/matplotlib/__init__.py' that dateutil is imported by matplotlib/__init__.py .

So when I run dir(module) is there an easy way to tell which attributes are part of the module itself, and which were imported by the module? And how could I have know (other than reading documentation, or "ls"ing the matplotlib directory) that 'dates' is also a valid "attribute" and "module" within matplotlib?

Answer 1

Simply matplotlib does not itself import .dates . What dir() shows you is a list of attributes of an object. If matplotlib does not import .dates , then there's no local name dates on the matplotlib module object.

Now, if you do import matplotlib.dates , Python's import system resolves that to the file/folder matplotlib/dates.py or matplotlib/dates/__init__.py , whichever exists. It's important to emphasise that matplotlib itself has nothing to do with that file resolution and hence does not need to "know" in any way about the existence of that submodule.

If you load a module, it does not automatically load all possible submodules as well. That only happens if something explicitly import s that name.