[PATCH] D149650: Give NullabilityKind a printing operator<<

[Sam McCall via Phabricator via cfe-commits]

sammccall added a reviewer: aaron.ballman.
sammccall added a comment.

In D149650#4312389 <https://reviews.llvm.org/D149650#4312389>, @aaron.ballman wrote:

> I guess I'm not seeing much motivation for this change. We already have `clang::getNullabilitySpelling()` and `const StreamingDiagnostic &clang::operator<<(const StreamingDiagnostic &DB, DiagNullabilityKind nullability)` and now we're adding a third way to get this information. If this is just for debug/testing purposes, can we use existing debug formatters to convert the enumeration value into the enumerator name instead?

We're using NullabilityKind in our dataflow-based null-safety clang-tidy check <https://github.com/google/crubit/tree/main/nullability> that we hope to upstream when it works.
(The idea is to use `_Nullable` and `_Nonnull` annotations on API boundaries to gradually type pointers, and to provide a verification clang-tidy check and libraries to infer annotations for existing code. The actual clang-tidy check wrapper isn't in that repo yet, but it's trivial).

It would be convenient if e.g. EXPECT_THAT(someVectorOfNK, ElementsAre(Nullable, Unspecified)) <https://github.com/google/crubit/blob/main/nullability/type_nullability_test.cc> printed something useful when it fails, if we could write OS << NK and get Unspecified rather than OS << getSpelling(NK) and get _Unspecified_Nullability, etc.
This doesn't concern clang core (ha, there are no unit tests...) but there's no reasonable way to do this downstream without using a different type.

> `StreamingDiagnostic &clang::operator<<`

This actually wants the user-visible spelling, so I think it can just use getSpelling... if I make that change we're back to just two implementations :-)


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D149650/new/

https://reviews.llvm.org/D149650