FIX: Set MRS type to Nifti1Extension for backwards compatibility

[effigies]

cc @wtclarke

[wtclarke]

Does Nifti1Extension still have the .json() methods? I've just updated the nifti-mrs package to use the new interface. So if the new methods aren't available for this, then I'm happy to just make the changes on my side and not have this fix.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 86.39%. Comparing base (96c8320) to head (56446e5).
Report is 5 commits behind head on maint/5.3.x.

❗ There is a different number of reports uploaded between BASE (96c8320) and HEAD (56446e5). Click for more details.

HEAD has 57 uploads less than BASE

Flag	BASE (96c8320)	HEAD (56446e5)
	58	1

Additional details and impacted files

@@               Coverage Diff               @@
##           maint/5.3.x    #1380      +/-   ##
===============================================
- Coverage        95.38%   86.39%   -8.99%     
===============================================
  Files              207      207              
  Lines            29672    29665       -7     
  Branches          3018     3018              
===============================================
- Hits             28302    25629    -2673     
- Misses             931     3594    +2663     
- Partials           439      442       +3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[effigies]

Yes, Nifti1Extension subclasses NiftiExtension[bytes] so _mangle() and _unmangle() are just identity functions, as was the previous default. The purpose of splitting the classes was so that new subclasses could derive from the base class and declare their types correctly.

[wtclarke]

Great, I see now, I was getting confused between Nifti1Extension and Nifti1Extensions. Thanks for the fix.

[effigies]

I guess with the problem solved on your end, we can take a minute to think about how you would like to evolve.

If you would like to offload more of your logic into nibabel and use a NiftiExtension directly, then it's possible that creating a JSONExtension subclass would make more sense. That would allow ext.get_object() to return a cached value that you can mutate. When it's time to save to a new image, that object will override the original bytes content.

If read-only access is all you want, then this PR will at least get you back what you had, without users having to upgrade your package if they upgrade nibabel. In practice, though, I expect most people just install the latest versions, and so making this release or not will impact very few people.

[effigies]

Well, regardless of the long-term path forward, this has the potential to reduce pain and doesn't make changing in the future harder.

[wtclarke]

Hi @effigies, Thanks for this fix. Sorry for the slow reply, four new starters in the group has me rushing about this week. For now I'm happy to keep the logic on my side.