LambdaRewriter.SubstituteTypeArguments should make progress

[jcouv]

Fix #37456

@gafter Looking back at #35284, I still think the use of reference equality is correct, but I want to check with you in case you think we should return to the original .Equals(..., ConsiderEverything) comparison.

[RikkiGibson]

is this the right file for this test?

[jcouv]

I think it's ok because the regression came from nullability work.

[jcouv]

Moved the test ;-)

[cston]

v [](start = 20, length = 1)

Uppercase first char.

[cston]

            while ((object)oldTypeArg.Type != newTypeArg.Type);

Consider changing this back to !TypeSymbol.Equals(oldTypeArg.Type, newTypeArg.Type, TypeCompareKind.ConsiderEverything) (as it was before #35284) in case SubstituteType() does return an identical copy rather the original instance. We could assert the instances are identical rather than relying on that.

---

Refers to: src/Compilers/CSharp/Portable/Lowering/LambdaRewriter/LambdaRewriter.cs:946 in ded9135. [](commit_id = ded9135, deletion_comment = False)

[jcouv]

Fix merged for 16.3p2. We're discussing porting it back to a servicing fix for 16.2.

[gafter]

I believe the reference equality is correct; type substitution should not return a new instance when it makes no substitutions.