Dataclass descriptors + conformance test problems

[stroxler]

Describe the Bug

This issue tracks an interaction between Pyrefly descriptor behavior and the conformance suite.

See this sandbox example, which is adapted from dataclasses_descriptors.py.

The actual behavior doesn't match what the conformance tests expect almost at all: I think that for a dataclass that has a descriptor with no __set__, instances will frequently wind up with raw descriptors as instance attributes although in the presence of optionality I suspect the behavior is unpredictable and you get a union at runtime.

If the descriptor does have a __set__, I'm highly uncertain the runtime behavior; likely the constructor winds up invoking it but I have not yet checked.

In addition, if no value is defined at all and the attribute is annotation only, then it's a pure instance-only attribute and no descriptor behavior exists (this would be true even for a normal class, it's not dataclass-specific) so the conformance tests are 100% wrong here.

Sandbox Link

No response

(Only applicable for extension issues) IDE Information

No response

[stroxler]

Self-assigned: I'm working on fixes to Pyrefly that seem to best approximate runtime behavior, I'll reach out to the typing repo once my fixes are in trunk because that will make sharing context easier

While working on the fix I realized there are more edge cases (data descriptors, self-assignable descriptors) where types potentially work out okay, but the conformance tests are still asserting what seems like wrong behavior to me.

[github-actions]

This issue has someone assigned, but has not had recent activity for more than 2 weeks.

If you are still working on this issue, please add a comment so everyone knows. Otherwise, please unassign yourself and allow someone else to take over.

Thank you for your contributions!

[migeed-z]

@stroxler I'm going through some of the conformance related issues and came across this one! is this still something you are working on and would you think it's a good idea to open it up to the community?

[stroxler]

It's not suitable for a contributor who wants to hack on Pyrefly, it's a typing community problem because I think the existing conformance tests are wrong.

I'd be fine with someone else taking it on, but most of the people looking at our issues want to hack on Pyrefly, not propose changes that need to go through the typing council