[RTM] Support GIFTI outputs in SampleToSurface

[effigies]

This PR creates "implicit" output types for Freesurfer utilities, where no explicit --out_type is supported, but a filename with the associated extension will be correctly produced as that type.

At present, I've only tested this with GIFTI on mri_vol2surf. I expect mri_surf2surf will have the same level of support. I also suspect that any of the mris_convert outputs would work fine.

Note that .gii.gz actually produces a .mgz file with .gii.gz extension, and so any additional output types need to be checked for correctness. Similarly, not all files seem to produce a correct GIFTI; for example, --srchit test.gii produces a valid GIFTI file with completely different data.

[codecov-io]

Codecov Report

Merging #1886 into master will decrease coverage by 0.02%.
The diff coverage is 25.92%.

@@            Coverage Diff             @@
##           master    #1886      +/-   ##
==========================================
- Coverage   70.94%   70.92%   -0.03%     
==========================================
  Files        1057     1057              
  Lines       53142    53164      +22     
  Branches     7699     7709      +10     
==========================================
+ Hits        37703    37705       +2     
- Misses      14065    14096      +31     
+ Partials     1374     1363      -11

Flag	Coverage Δ
#smoketests	43.41% <ø> (ø)	⬆️
#unittests	70.36% <25.92%> (-0.03%)	⬇️

Impacted Files	Coverage Δ
nipype/interfaces/freesurfer/utils.py	62.72% <25.92%> (-0.81%)	⬇️
nipype/interfaces/base.py	83.28% <0%> (-0.19%)	⬇️
nipype/interfaces/fsl/utils.py	65.6% <0%> (ø)	⬆️
nipype/interfaces/spm/model.py	41.76% <0%> (ø)	⬆️
nipype/workflows/fmri/fsl/preprocess.py	72.77% <0%> (ø)	⬆️
nipype/interfaces/io.py	48.16% <0%> (ø)	⬆️
nipype/interfaces/spm/preprocess.py	55.93% <0%> (ø)	⬆️
nipype/workflows/dmri/fsl/artifacts.py	81.28% <0%> (ø)	⬆️
nipype/interfaces/fsl/base.py	77.17% <0%> (ø)	⬆️
nipype/utils/docparse.py	53.04% <0%> (ø)	⬆️

---

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 22ca05c...f786fce. Read the comment docs.

[effigies]

I've done some playing, and mri_vol2surf with .gii file as output produces identical data as mri_vol2surf with .mgz output, followed by mris_convert.

Also, mri_surf2surf similarly accepts .gii in output filenames, so I've updated the SurfaceTransform interface.

[effigies]

I guess one issue here is that, currently, setting output type allows you to get away with non-matching names. Here, setting the target type to gii would not set an output type, but would change the filename, provided that an output filename is not given. If output filename is set, setting the target type would have no effect.

So if this change is to be accepted, I should notify users. Would something in the interface docstring, trait description, or a runtime check for output filename and target type == 'gii' and emitting a warning? All of the above?

[chrisgorgo]

We can throw an error in format_args if the type and extensions are inconsistent. Nothing good can come out of such combinations.

BTW is .gii.gz a thing?

[effigies]

.gii.gz. I suspect this is just for compression on NITRC, not an officially supported extension. Since .gii can contain gzip-base64 encoded blocks, it's probably seen as a bit pointless.

For the error, I'd say a warning if they mismatch, an error if they mismatch and the extension belongs to a supported type. There are some extension-less FreeSurfer files, and catching all of those could be tricky.

[chrisgorgo]

sounds good to me

[satra]

if there are no other changes requested, this looks good to me. feel free to merge.