FIX: Correctly reorient dim_info when reorienting NIfTI images

[effigies]

Starting with a test to demonstrate the failure. Will push a fix shortly.

cc @constracti

Fixes #716.

[effigies]

Anybody up for a review on this?

[effigies]

@markhymers This was originally your contribution. Would you care to check my logic, and make sure that I'm not breaking things?

[codecov-io]

Codecov Report

Merging #741 into master will decrease coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #741      +/-   ##
==========================================
- Coverage   89.55%   89.54%   -0.01%     
==========================================
  Files          93       93              
  Lines       11474    11470       -4     
  Branches     1992     1991       -1     
==========================================
- Hits        10275    10271       -4     
+ Misses        859      858       -1     
- Partials      340      341       +1

Impacted Files	Coverage Δ
nibabel/nifti1.py	91.47% <100%> (-0.06%)	⬇️
nibabel/cmdline/parrec2nii.py	32.73% <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 9430224...7c37276. Read the comment docs.

[constracti]

This commit does the same thing in a more pythonic way, as you note, with the list comprehension. It does not fix the bug.

[effigies]

Thanks for the review.

I thought the confusion was that the freq/phase/slice ordering will only sometimes rotate in the same way as the shapes, but it turns out that orientations also didn't work how I thought... The issue was that I keep thinking of the orientation produced by axcodes2ornt as how to get from RAS to a given axis code, but it's exactly the reverse. So my consistency check was off, which was why your fix didn't seem to work. By instead cycling through orientations and getting the axis codes from the affine, I can now verify your proposed fix was correct.

[effigies]

@constracti I found 46476cc in your history, so I went ahead and cherry-picked it and got rid of my flailing about for a wrong solution. I did still tack on the refactor to a list comprehension and casting to int(), as orientation matrices may not be integer-typed. If you want to let me know your preferred name for publication, your ORCID and your affiliation, I'll update the mailmap and Zenodo file as needed.

[effigies]

Any more comments?

[effigies]

Thanks for the contribution and feedback.