CI: Stop using Miniconda on Travis

[effigies]

Suggested in #2598. Will allow us to test more vanilla installations of Python.

Should speed up Travis builds, too.

[effigies]

The failing test is related to the new numpy formatting.

nipype/nipype/interfaces/image.py

Lines 63 to 78 in a882e74

	>>> print(np.loadtxt(res.outputs.transform))
	[[1. 0. 0. 0.]
	[0. 1. 0. 0.]
	[0. 0. 1. 0.]
	[0. 0. 0. 1.]]
	>>> reorient.inputs.orientation = 'RAS'
	>>> res = reorient.run()
	>>> res.outputs.out_file # doctest: +ELLIPSIS
	'.../segmentation0_ras.nii.gz'
	>>> print(np.loadtxt(res.outputs.transform))
	[[-1. 0. 0. 60.]
	[ 0. -1. 0. 72.]
	[ 0. 0. 1. 0.]
	[ 0. 0. 0. 1.]]

I'm not really sure why this wouldn't have appeared under conda, but there we are.

I guess the question is whether to change the test or update print options like so.

[satra]

perhaps we can do something through the pytest config, just like we are doing doctests? we could also update the test to use our own format without relying on numpy.

[codecov-io]

Codecov Report

Merging #2600 into master will increase coverage by <.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #2600      +/-   ##
==========================================
+ Coverage   67.62%   67.62%   +<.01%     
==========================================
  Files         339      339              
  Lines       42787    42787              
  Branches     5288     5288              
==========================================
+ Hits        28934    28936       +2     
+ Misses      13171    13168       -3     
- Partials      682      683       +1

Flag	Coverage Δ
#smoketests	50.76% <ø> (ø)	⬆️
#unittests	65.09% <ø> (+0.01%)	⬆️

Impacted Files	Coverage Δ
nipype/interfaces/dcm2nii.py	48.13% <ø> (ø)	⬆️
nipype/interfaces/niftyreg/regutils.py	80.86% <ø> (ø)	⬆️
nipype/interfaces/io.py	54.47% <ø> (ø)	⬆️
nipype/interfaces/base/core.py	89.7% <ø> (ø)	⬆️
nipype/interfaces/image.py	90.19% <ø> (ø)	⬆️
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 205ea17...2025a06. Read the comment docs.

[effigies]

we could also update the test to use our own format without relying on numpy.

Yeah. Apparently the details of the effective formatting change vary by what version of Python the wheel was compiled for (while conda packages seemed reasonably consistent). Added a print function that replaces the square brackets with spaces and letting the whitespace normalization do the rest.

[effigies]

Passing Travis.

[effigies]

Hmm. Sphinx's .. testsetup:: blocks are showing up in the docs:

• https://5032-791477-gh.circle-artifacts.com/0/home/circleci/work/docs/interfaces/generated/nipype.interfaces.image.html#examples
• https://nipype.readthedocs.io/en/latest/interfaces/generated/nipype.interfaces.dcm2nii.html#examples

[effigies]

It's an indentation effect. .. testsetup:: and .. testcleanup:: are not treated correctly unless they start on the first character of the line. However, if we de-indent these, then the whole text of the docstring is treated as indented, changing appearance.

Is it acceptable to have a de-indented docstring?

[satra]

@effigies - sorry didn't see this earlier. if it's necessary, it's necessary. looks darn ugly though :(

[effigies]

Yeah. Could be possible to fix up the Sphinx extension, but that's beyond my current ambitions.

[effigies]

Actually, I think all but one testsetup block can be removed.