Revert "Merge pull request #61606 from NuriAmari/fwd-declarations"

[Azoy]

This introduced 12 failing tests:

Failed Tests (12):
  Swift(macosx-x86_64) :: ClangImporter/incomplete_objc_types_availability.swift
  Swift(macosx-x86_64) :: ClangImporter/incomplete_objc_types_base_interface_swift_6.swift
  Swift(macosx-x86_64) :: ClangImporter/incomplete_objc_types_compatibility_complete_incomplete.swift
  Swift(macosx-x86_64) :: ClangImporter/incomplete_objc_types_compatibility_incomplete_incomplete.swift
  Swift(macosx-x86_64) :: ClangImporter/incomplete_objc_types_full_availability_swift_6.swift
  Swift(macosx-x86_64) :: ClangImporter/incomplete_objc_types_importing_limitations_diagnostics.swift
  Swift(macosx-x86_64) :: ClangImporter/incomplete_objc_types_no_reference.swift
  Swift(macosx-x86_64) :: ClangImporter/incomplete_objc_types_no_root_protocol.swift
  Swift(macosx-x86_64) :: ClangImporter/incomplete_objc_types_nsproxy.swift
  Swift(macosx-x86_64) :: ClangImporter/incomplete_objc_types_swift_definition_imported.swift
  Swift(macosx-x86_64) :: ClangImporter/incomplete_objc_types_swift_definition_imported_implementationOnly.swift
  Swift(macosx-x86_64) :: ClangImporter/incomplete_objc_types_swift_definition_imported_name_conflict.swift

https://ci.swift.org/job/oss-swift_tools-R_stdlib-RD_test-simulator/2677/

[Azoy]

@swift-ci please test

[drodriguez]

It seems that all that's missing is adding a REQUIRES: asserts to those test to avoid --swift-version 6 from being rejected.

Can we fix forward instead of reverting? We could have a fix ready in some minutes.

[Azoy]

That would explain the --swift-version 6 errors, but I don't think it solves the other errors in the failures no?

[NuriAmari]

That would explain the --swift-version 6 errors, but I don't think it solves the other errors in the failures no?

I will double check, but I don't see other failures? Do you have an example

[NuriAmari]

Every failing test uses -swift-version 6, there is other noise from diagnostics, but those are intended as far as I see. If there were other issues, I think we'd have seen it on the assert toolchains.

#64357

[drodriguez]

We have the asserts preset building and passing in #64357. To avoid a landing race, is everyone ok closing this PR and going with the other PR?

Edit: correct wrong PR number.