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

[Dan Liew via cfe-commits]

================
@@ -1822,28 +1822,106 @@ void WriteSemanticSpellingSwitch(const std::string &VarName,
   OS << "  }\n";
 }
 
+enum class LateAttrParseKind {
+  LateAttrParsingNever = 0,
+  LateAttrParsingAlways = 1,
+  LateAttrParsingExperimentalOnly = 2
+};
+
+static LateAttrParseKind getLateAttrParseKind(const Record *Attr) {
+  // This function basically does
+  // `Attr->getValueAsDef("LateParsed")->getValueAsInt("Mode")` but does
+  // a bunch of sanity checking to ensure that
+  // `LateAttrParseMode` in `Attr.td` is in sync with the `LateAttrParseKind`
+  // enum in this source file.
+
+  const char *LateParsedStr = "LateParsed";
+  auto *LAPK = Attr->getValueAsDef(LateParsedStr);
+
+  // Typecheck the `LateParsed` field.
+  SmallVector<Record *, 1> SuperClasses;
+  LAPK->getDirectSuperClasses(SuperClasses);
+  if (SuperClasses.size() != 1) {
+    PrintFatalError(Attr, "Field `" + llvm::Twine(LateParsedStr) +
+                              "`should only have one super class");
+  }
+  const char *LateAttrParseKindStr = "LateAttrParseKind";
----------------
delcypher wrote:

I'll give this a try. Guess I'll find out if `llvm::Twine` is happy working with `constexpr`.

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