Reveal previously hidden AttributeErrors and TypeErrors

[akx]

This ensures AttributeErrors and TypeErrors do not get hidden when DRF is just looking for a get_queryset().

[lovelydinosaur]

Prob worth issuing as two separate PRs.
The get_queryset ones are simple enough and would prob accept those as-is.
The .count() I'd want to give more review to.

[akx]

Sure, I'll snip that one out.

[lovelydinosaur]

Addressing any windows test failures should probably be treated as a separate PR.
I assume that's what this change is about, right?

[akx]

Yeah -- though it's a bit odd that repr would act differently across platforms...

[akx]

I'll split the test stuff into yet another PR. :)

[lovelydinosaur]

Just a regular AttributeError would be okay here.

[lovelydinosaur]

Or perhaps return self.this_attribute_does_not_exist?

[akx]

I subclassed AttributeError for the test for the purpose of being able to check for that exact exception being propagated up, which was the point of the patch in the first place :)

[lovelydinosaur]

Okay, on review at this point I'd actually rather have this, but without the test cases.
It's a case of evidently improved behavior, and I think they'd actually add more noise than value in this case. Otherwise fix looks good, so happy to move once tests have been dropped. 👍

[akx]

Well, if you're sure! I'll take the test cases out.

[akx]

Tests are now gone, @tomchristie.

[lovelydinosaur]

May seem counter-intuitive, but they come with an associated cost, and can be okay not to strictly test against obviously better/cleaner behavior.