CI: Test against minimum versions

[effigies]

We're currently testing against the most recent versions of our dependencies, which makes it harder to identify when we accidentally depend on newer features. See, e.g., #2181.

[codecov-io]

Codecov Report

Merging #2576 into master will increase coverage by 0.02%.
The diff coverage is 100%.

@@            Coverage Diff            @@
##           master   #2576      +/-   ##
=========================================
+ Coverage   67.57%   67.6%   +0.02%     
=========================================
  Files         333     333              
  Lines       42547   42550       +3     
  Branches     5266    5267       +1     
=========================================
+ Hits        28750   28764      +14     
+ Misses      13124   13113      -11     
  Partials      673     673

Flag	Coverage Δ
#smoketests	50.7% <0%> (-0.01%)	⬇️
#unittests	65.05% <100%> (+0.04%)	⬆️

Impacted Files	Coverage Δ
nipype/pipeline/plugins/base.py	61.04% <100%> (+1.16%)	⬆️
nipype/pipeline/engine/utils.py	82.58% <0%> (+0.11%)	⬆️
nipype/pipeline/engine/workflows.py	79.76% <0%> (+0.68%)	⬆️
nipype/interfaces/matlab.py	86.86% <0%> (+1.01%)	⬆️
nipype/interfaces/dipy/setup.py	25% <0%> (+25%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5a96ea5...62e1734. Read the comment docs.

[effigies]

Looks like we're running up against issues when numpy==1.9 and scipy is 0.14.0, 0.16.0 or 0.18.1.

Python < 3.6, scipy == {0.14.0, 0.16.0}

E           RuntimeError: Traceback (most recent call last):
E           
E             File "/home/travis/build/nipy/nipype/nipype/pipeline/plugins/base.py", line 370, in _local_hash_check
E               self._task_finished_cb(jobid, cached=True)
E           
E             File "/home/travis/build/nipy/nipype/nipype/pipeline/plugins/base.py", line 394, in _task_finished_cb
E               rowview[rowview.nonzero()] = 0
E           
E             File "/home/travis/conda/lib/python2.7/site-packages/scipy/sparse/base.py", line 536, in nonzero
E               A = self.tocoo()
E           
E             File "/home/travis/conda/lib/python2.7/site-packages/scipy/sparse/base.py", line 643, in tocoo
E               return self.tocsr().tocoo()
E           
E             File "/home/travis/conda/lib/python2.7/site-packages/scipy/sparse/lil.py", line 389, in tocsr
E               indices = np.asarray(indices, dtype=idx_dtype)
E           
E             File "/home/travis/conda/lib/python2.7/site-packages/numpy/core/numeric.py", line 462, in asarray
E               return array(a, dtype, copy=False, order=order)
E           
E           ValueError: setting an array element with a sequence.

Python == 3.6, scipy == 0.18.1

../../../conda/lib/python3.6/site-packages/scipy/linalg/__init__.py:174: in <module>
    from .misc import *
../../../conda/lib/python3.6/site-packages/scipy/linalg/misc.py:5: in <module>
    from .blas import get_blas_funcs
../../../conda/lib/python3.6/site-packages/scipy/linalg/blas.py:155: in <module>
    from scipy.linalg import _fblas
E   ImportError: numpy.core.multiarray failed to import

nipype/algorithms/modelgen.py:23: in <module>
    from scipy.special import gammaln
../../../conda/lib/python3.6/site-packages/scipy/special/__init__.py:638: in <module>
    from .basic import *
../../../conda/lib/python3.6/site-packages/scipy/special/basic.py:15: in <module>
    from ._ufuncs import (ellipkm1, mathieu_a, mathieu_b, iv, jv, gamma,
E   ImportError: cannot import name 'ellipkm1'

[satra]

@effigies - do we know what the python/numpy/scipy version compatibility matrix looks like? my simple searches did not turn up anything concrete.

[effigies]

I do not. I picked the scipy minimum versions based on which ones had wheels on PyPI.

Judging by the current scipy/setup.py, there should be compatibility back through numpy 1.8 and for all versions that we test. But that doesn't mean the wheels were built properly.

[effigies]

While I think it's a good goal to test against minimum versions, I don't think I'm going to have the time to track this one down.

[satra]

@effigies - this is exactly the kind of thing that regtests is being built for. @djarecka - perhaps you can add this as an example.

[djarecka]

@satra I added this issue to regtest to not forget