Revert https://github.com/apple/swift/pull/40269

[hborla]

#40269 is a source breaking change, as noted here:

protocol P { associatedtype A = Int }
protocol Q { associatedtype B }
protocol R: P, Q where A == B {}

struct S: R {}

This code compiles today, but breaks with the associated type inference refactoring. Unfortunately, there's real-world code relying on the old behavior.

Resolves: rdar://90402522

[hborla]

@AnthonyLatsis do you think it'd be relatively straightforward to put the new behavior behind a flag? I'm not too familiar with your change, but with your help, I'll happily attempt to put it behind a flag instead of the revert.

[AnthonyLatsis]

Bummer! Let us see... We would need to revert this to resurrect the old code and conditionalise the inference method in AssociatedTypeInference::inferAbstractTypeWitnesses with respect to these changes (disregard all the const additions). Sounds relatively straightforward from what I see. The backup plan would be to trade duplication for easiness and define a second inferAbstractTypeWitnesses. Everything else should be good to keep.

---

Let me know if you need any help!

[AnthonyLatsis]

@hborla In case you haven't started working on the flag yet — just want to be honest and let you know that I don't feel right about having you fix a problem that I'm responsible for unless you are actually interested, and would be glad to resolve this if given the opportunity. 🙂

[hborla]

@AnthonyLatsis I have not started working on the flag yet! I was working on a couple other PRs that I wanted to get up before doing that, but if you'd like to add the flag instead, please go for it 🙂

[AnthonyLatsis]

#42004

[hborla]

@AnthonyLatsis Thank you! Closing this in favor of your PR 🙏🏻