Needs to delay injections for not yet imported submodules until their import time about duecredit HOT 9 OPEN

2016-05-04 16:14:32,911 [WARNING] Could not find fsl in module <module 'nipype.interfaces' from '/home/alexandre/Software/nipype/nipype/interfaces/init.py'>: module 'nipype.interfaces' has no attribute 'fsl' (injector.py:193)
that is the location -- "injector.py:193" ;)

from duecredit.

could you give a minimal snapshot (or just point to your branch of nipype or whatnot) on how to provoke such behavior? the problem is that we are checking too early since fsl not being imported yet is not available within interfaces:

In [1]: import nipype.interfaces

In [2]: dir(nipype.interfaces)
Out[2]: 
['DataGrabber',
 'DataSink',
 'Function',
 'IdentityInterface',
 'Merge',
 'Rename',
 'Select',
 'SelectFiles',
 '__builtins__',
 '__doc__',
 '__docformat__',
 '__file__',
 '__name__',
 '__package__',
 '__path__',
 'base',
 'io',
 'traits_extension',
 'utility']

*In [3]: import nipype.interfaces.fsl

In [4]: dir(nipype.interfaces)
Out[4]: 
['DataGrabber',
 'DataSink',
 'Function',
 'IdentityInterface',
 'Merge',
 'Rename',
 'Select',
 'SelectFiles',
 '__builtins__',
 '__doc__',
 '__docformat__',
 '__file__',
 '__name__',
 '__package__',
 '__path__',
 'base',
 'fsl',
 'io',
 'traits_extension',
 'utility']

from duecredit.

Ooops I see, injector.py line 193. 👍

I am using the nipype master branch and Python 3.5. I will have a look too.

from duecredit.

Another approach could be try and __import__, instead of dir ?

from duecredit.

well -- the point is that duecredit should nohow interfer or do anything custom (such as imports)... matteo has mentioned that you are using python 3.5 which we apparently don't travis against yet... we will rerun our tests under 3.5 (#79) and see it could be something specific to it

interestingly that nipype/init.py also doesn't import those submodules... so it would be interesting to replicate to see what causes the warnings and how dangerous it is all is or could be ignored ;-)

from duecredit.

at least our tests didn't detect anything 3.5 special https://travis-ci.org/duecredit/duecredit/builds/127833554

from duecredit.

ok -- replicated even with python2.7. we will look into it

from duecredit.

I wonder if the nipype injector makes sense now that we will use duecredit directly from nipype.

from duecredit.

you are correct -- nipy injection would not be needed as soon as nipype supports it directly.
So far I thought to support decision on either to inject or not based on the module version, but may be I could/should add additional ability to check, e.g. if nipype.due is available etc, so even in development snapshots could avoid unnecessary injection.
But I believe that this issue is independent to the warnings -- they poped up even without trying your PR with proper duecredit entry

from duecredit.