Fix filterset multiple inheritance bug

[rpkilby]

Declared filter resolution was mostly copied from DRF. This fixes a multiple inheritance bug described in encode/django-rest-framework#6980

[codecov-io]

Codecov Report

Merging #1131 into master will increase coverage by 1.33%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #1131      +/-   ##
==========================================
+ Coverage    98.1%   99.44%   +1.33%     
==========================================
  Files          15       15              
  Lines        1216     1255      +39     
==========================================
+ Hits         1193     1248      +55     
+ Misses         23        7      -16

Impacted Files	Coverage Δ
django_filters/filterset.py	100% <100%> (ø)	⬆️
django_filters/utils.py	100% <0%> (ø)	⬆️
django_filters/rest_framework/filterset.py	100% <0%> (ø)	⬆️
django_filters/filters.py	100% <0%> (+0.29%)	⬆️
django_filters/conf.py	100% <0%> (+2.94%)	⬆️
django_filters/widgets.py	100% <0%> (+4.72%)	⬆️
django_filters/rest_framework/backends.py	100% <0%> (+8.53%)	⬆️

---

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 da4b64e...f7e7af3. Read the comment docs.

[carltongibson]

We sure we’re not changing a behaviour from the dawn of time here? 🙂

[rpkilby]

It looks like the original behavior was mirrored in this PR.

django-filter/django_filters/filterset.py

Lines 33 to 40 in 62516ad

	if with_base_filters:
	for base in bases[::-1]:
	if hasattr(base, 'base_filters'):
	filters = list(base.base_filters.items()) + filters
	else:
	for base in bases[::-1]:
	if hasattr(base, 'declared_filters'):
	filters = list(base.declared_filters.items()) + filters

However, I refactored get_declared_filters at some point and iirc I copied the code largely from DRF. Currently looking into this.

[rpkilby]

oh wait, I read that completely wrong. 🤦‍♂

Either way, like DRF, I think the test case demonstrates that the current behavior is incorrect. I just don't think this multiple inheritance use case is incredibly common, which is why it hasn't been raised here yet (and why it hasn't seen much attention in DRF either).

[carltongibson]

OK... Let me look into it. I want to be 100% sure. (And we'll need a release note stating the BC change.)

[rpkilby]

Just as a note, this PR needs to be updated, pending review of the DRF PR.

[rpkilby]

@carltongibson updated this from the DRF pr. This should be good to go assuming tests pass.

[carltongibson]

Super, yes. Good work!