[clang] [APINotes] Match function-like Where.Parameters in Sema (PR #205307)

[John Hui via cfe-commits]

================
@@ -993,6 +993,87 @@ UnwindTagContext(TagDecl *DC, api_notes::APINotesManager &APINotes) {
   return std::nullopt;
 }
 
+static void stripAPINotesParameterNullability(QualType &ParamType) {
+  while (true) {
+    if (!AttributedType::stripOuterNullability(ParamType))
+      return;
+  }
+}
+
+// Print the APINotes selector spelling for one parameter. The source-spelled
+// selector is tried first. The desugared spelling is only a permissive
+// fallback.
+static std::string getAPINotesParameterSelectorSpelling(
+    QualType ParamType, const ASTContext &Context, const PrintingPolicy &Policy,
+    bool Desugar) {
+  ParamType.removeLocalConst();
+  stripAPINotesParameterNullability(ParamType);
+
+  if (Desugar) {
+    ParamType = ParamType.getDesugaredType(Context);
+    ParamType.removeLocalConst();
+    stripAPINotesParameterNullability(ParamType);
+  }
+
+  return ParamType.getAsString(Policy);
+}
+
+static std::optional<SmallVector<SmallVector<std::string, 4>, 2>>
+getAPINotesParameterSelectorCandidates(const Sema &S, const FunctionDecl *FD) {
+  const auto *FPT = FD->getType()->getAs<FunctionProtoType>();
+  if (!FPT)
+    return std::nullopt;
+
+  SmallVector<std::string, 4> SourceParameters;
+  SmallVector<std::string, 4> DesugaredParameters;
+  SourceParameters.reserve(FPT->getNumParams());
+  DesugaredParameters.reserve(FPT->getNumParams());
+
+  const PrintingPolicy &Policy = S.Context.getPrintingPolicy();
----------------
j-hui wrote:

I'm fine with this being done on the basis of string comparisons for now!

We can talk more about whether we should be passing around strings vs `QualType`s when we meet next. Doing so may reduce the amount you have to copy around string vectors and perform (potentially fragile) string normalization, but that might not be a good trade-off if it comes at the cost of needing to repeatedly print a `QualType` into a string to do the comparison. The trade-off might not become clear until later, so no need to do any premature optimizations in this PR (:

https://github.com/llvm/llvm-project/pull/205307