autodoc tries to import classes as modules?

[amueller]

After updating sphinx and numpydoc, I get

WARNING: /home/andy/checkout/scikit-learn/doc/modules/generated/sklearn.preprocessing.RobustScaler.rst:92: (WARNING/2) autodoc: failed to import method 'RobustScaler.__init__' from module 'sklearn.preprocessing.data.RobustScaler'; the following exception was raised:
Traceback (most recent call last):
  File "/home/andy/anaconda3/lib/python3.5/site-packages/sphinx/ext/autodoc.py", line 658, in import_object
    __import__(self.modname)
ImportError: No module named 'sklearn.preprocessing.data.RobustScaler'; 'sklearn.preprocessing.data' is not a package

for every class in the package.

It looks like the modname is wrong as it is the class and not the module. I'm not sure what should be imported at this point, though. The class? Importing a method doesn't seem to make a lot of sense...

[tk0miya]

Could you share your project to reproduce the error?
It would be nice if you provide small reproducible examples :-)

Thanks,

[amueller]

The project is scikit-learn. I'll try to extract a minimal example, but I'll be traveling. Sent from phone. Please excuse spelling and brevity.

…

On May 19, 2017 11:01 AM, "Takeshi KOMIYA" ***@***.***> wrote: Could you share your project to reproduce the error? It would be nice if you provide small reproducible examples :-) Thanks, — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#3768 (comment)>, or mute the thread <https:/notifications/unsubscribe-auth/AAbcFp4AHDTYMsTf-7NGmbiut4mJXfGcks5r7a7hgaJpZM4NfxIc> .

[tk0miya]

Hmm.. scikit-learn's doc is too large and complex to investigate...
I failed to do that today.

[goosemania]

@amueller , I suggest to test your case with this recently merged fix #3780 . It is very small in terms of the changes to the code but it has a huge impact on how import works in certain cases. Hopefully it covers your case as well.

[amueller]

Thanks for looking into it. I'll have a look next week when I'm back from vacation ;)

[amueller]

The issue is still present on master.

[amueller]

I stripped down the sklearn build:
https:/amueller/sphinx-debug

you need to add the doc folder to the pythonpath so the blablub.py is found as a module.

I get

WARNING: /home/andy/checkout/sphinx-debug/doc/modules/generated/blablub.BaseEstimator.rst:28: (WARNING/2) autodoc: failed to import method 'BaseEstimator.__init__' from module 'blablub.BaseEstimator'; the following exception was raised:
Traceback (most recent call last):
  File "/home/andy/checkout/sphinx/sphinx/ext/autodoc.py", line 657, in import_object
    __import__(self.modname)

[tk0miya]

Yes, autodoc imports the target module internally on generating document.

It seems your example surely raises ImportError, but the error reason is different with first case. Is this intended?

WARNING: /docs/sphinx-debug/doc/modules/classes.rst:1: (WARNING/2) autodoc: failed to import module 'blablub'; the following exception was raised:
Traceback (most recent call last):
  File "/usr/local/lib/python3.5/dist-packages/sphinx/ext/autodoc.py", line 658, in import_object
    __import__(self.modname)
ImportError: No module named 'blablub'

The message on first case is here:

ImportError: No module named 'sklearn.preprocessing.data.RobustScaler'; 'sklearn.preprocessing.data' is not a package

[amueller]

@tk0miya as I said, you need to add the doc path to the python path for the module to be found. Not sure how to autodoc a module that's not installed.

[amueller]

anything I can do to make reproducing this easier? solving this will make our release process much easier, currently it prevents us from using stock packages for building our docs.

[tk0miya]

Not sure how to autodoc a module that's not installed.

Please add your library's path to sys.path in conf.py like following:

import os
import sys

sys.path.insert(0, os.path.abspath("."))

[amueller]

done, not that that's any different from export PYTHONPATH=`pwd`

[jschueller]

maybe it was fixed by numpy/numpydoc#93 ?

[amueller]

@jschueller indeed it seems to be fixed on numpydoc master. Thanks!