I have observed the following behavior in python 3.4.2, and I am unableto explain it. Hopefully someone could shed some light on the matter:
In IPython:
In [129]: import urllib
In [130]: print(urllib.parse)
<module 'urllib.parse' from '/Users/ashwin/.pyenv/versions/3.4.2/lib/python3.4/urllib/parse.py'>
I've imported a module, and printed one of its attributes. Everything works as expected. So far, life is good.
Now, I do the same thing from the command line:
$ python -c 'import urllib; print(urllib.parse)'
Traceback (most recent call last):
File "<string>", line 1, in <module>
AttributeError: 'module' object has no attribute 'parse'
Say what?! that's not how that's supposed to work.
Ok, maybe this is a python-wide behavior; maybe modules are not immediately imported when using the -c
flag. Let's try another module:
$ python -c 'import datetime; print(datetime.datetime)'
<class 'datetime.datetime'>
What?! How does it work for datetime
and not for urllib
? I'm using the same version of python in both places (3.4.2)
Does anyone have any thoughts on this?
EDIT:
Per one of the comments:
$ which -a ipython
/Users/ashwin/.pyenv/shims/ipython
/Library/Frameworks/Python.framework/Versions/2.7/bin/ipython
/usr/local/bin/ipython
/usr/local/bin/ipython
And
$ which -a python
/Users/ashwin/.pyenv/shims/python
/Library/Frameworks/Python.framework/Versions/2.7/bin/python
/usr/bin/python
/usr/bin/python
When you run import urllib
, it creates the module object of the urllib
module (which is actually a package) without importing its submodules (parse, request etc.).
You need the parent module object (urllib
) to be in your namespace if you want to access its submodule using attribute access. In addition to that, that submodule must already be loaded (imported). From the documentation:
if package
spam
has a submodulefoo
, after importingspam.foo
,spam
will have an attributefoo
which is bound to the submodule. [...] The invariant holding is that if you havesys.modules['spam']
andsys.modules['spam.foo']
(as you would after the above import), the latter must appear as thefoo
attribute of the former.
There is only one instance of each module, thus any changes made to the urllib
module object (stored in sys.modules['urllib']
) get reflected everywhere.
You don't import urllib.parse
, but IPython does. To prove this I'm going to create a startup file:
import urllib
print('Running the startup file: ', end='')
try:
# After importing 'urllib.parse' ANYWHERE,
# 'urllib' will have the 'parse' attribute.
# You could also do "import sys; sys.modules['urllib'].parse"
urllib.parse
except AttributeError:
print("urllib.parse hasn't been imported yet")
else:
print('urllib.parse has already been imported')
print('Exiting the startup file.')
and launch ipython
vaultah@base:~$ ipython
Running urllib/parse.py
Running the startup file: urllib.parse has already been imported
Exiting the startup file.
Python 3.6.0a0 (default:089146b8ccc6, Sep 25 2015, 14:16:56)
Type "copyright", "credits" or "license" for more information.
IPython 4.0.0 -- An enhanced Interactive Python.
It is the side effect of importing pydoc
during the startup of IPython (which ipython
is /usr/local/bin/ipython):
/usr/local/bin/ipython, line 7:
from IPython import start_ipython
/usr/local/lib/python3.6/site-packages/IPython/__init__.py, line 47:
from .core.application import Application
/usr/local/lib/python3.6/site-packages/IPython/core/application.py, line 24:
from IPython.core import release, crashhandler
/usr/local/lib/python3.6/site-packages/IPython/core/crashhandler.py, line 28:
from IPython.core import ultratb
/usr/local/lib/python3.6/site-packages/IPython/core/ultratb.py, line 90:
import pydoc
/usr/local/lib/python3.6/pydoc.py, line 68:
import urllib.parse
This explains why the below code fails - you only import urllib
and nothing seems to import urllib.parse
:
$ python -c 'import urllib; print(urllib.parse)'
On the other hand, the following command works because datetime.datetime
is not a module. It's a class that gets imported during import datetime
.
$ python -c 'import datetime; print(datetime.datetime)'