[rchen152]

Parts of the dataclasses_descriptors.py conformance test expect behavior that doesn't align with the runtime (see issue: #2259, PR: #2299). This PR deletes the questionable parts: Desc2, DC2, and the associated assert_types.

[carljm]

Sorry, I don't think this change makes sense :/ A good argument can be made that the current type expectation a few lines down is currently incorrect (at least for the code exactly as shown), but that doesn't clearly justify an error on this line. I don't think there is a good general rationale for making all non-data descriptors error if used as dataclass field defaults. So I think this just makes an already-confusing part of the conformance suite even more confusing.

If the goal is to weaken the conformance suite so Pyrefly isn't marked down for its behavior in an area where the suite isn't clearly correct anyway, I think we should just do that by deleting this DC2 section of the suite instead. I don't feel like that should be controversial (I would happily approve it). It doesn't preclude us reintroducing more tests in this area in future if someone is motivated to add them, but that clearly will require some significant work on proposing behavior and building consensus around it.

(In other words, if someone feels strongly that the conformance suite should specify the behavior of descriptors as dataclass defaults, I think they should feel free to work on that topic, but in the meantime they should not stand in the way of deleting the current clearly-inadequate conformance test.)

[rchen152]

@carljm Done! The PR just deletes the DC2 section now.

[JelleZijlstra]

I wanted to do this too, agree with just deleting the questionable tests.

[jorenham]

Thanks Rebecca

[rchen152]

Thanks all!