[clang] [Attributes] Support Attributes being declared as only supporting late parsing when passing an experimental feature flag (PR #88596)

[Yeoul Na via cfe-commits]

================
@@ -592,6 +592,46 @@ class AttrSubjectMatcherAggregateRule<AttrSubject subject> {
 
 def SubjectMatcherForNamed : AttrSubjectMatcherAggregateRule<Named>;
 
+// Late Attribute parsing mode enum
+class LateAttrParseKind <int val> {
+  int Kind = val;
+}
+
+// Never late parsed
+def LateAttrParseNever : LateAttrParseKind<0>;
+
+// Standard late attribute parsing
+//
+// This is language dependent. For example:
+//
+// * For C++ enables late parsing of a declaration attributes
+// * For C does not enable late parsing of attributes
+//
+def LateAttrParseStandard: LateAttrParseKind<1>;
+
+// Experimental extension to standard late attribute parsing
+//
+// This extension behaves like `LateAttrParseStandard` but allows
+// late parsing attributes in more contexts.
+//
+// This extension changes behavior depending on whether the
+// `-fexperimental-late-parse-attributes`
+// (`LangOpts.ExperimentalLateParseAttributes`) flag is enabled.
+//
+// * If the flag is disabled then the attribute is parsed just like
+//   `LateAttrParseStandard`.
+//
+// * If the flag is enabled and if the attribute is being used in a context
+//   supported by this extension it will be late parsed. Otherwise the attribute
+//   will be parsed just like `LateAttrParseStandard`.
+//
+// Currently this extension extends `LateAttrParseStandard` by allowing:
+//
+// * For C, late parsing of the `counted_by` as a type or decl attribute.
+//   (TODO: Not implemented yet).
----------------
rapidsna wrote:

I don't have a strong opinion here. Having TODO in the comment seems okay to me.

https://github.com/llvm/llvm-project/pull/88596