[clang] [llvm] [HLSL][RootSignature] Define and integrate rootsig clang attr and decl (PR #137690)

[Aaron Ballman via cfe-commits]

================
@@ -5311,6 +5313,90 @@ void Parser::ParseMicrosoftUuidAttributeArgs(ParsedAttributes &Attrs) {
   }
 }
 
+void Parser::ParseMicrosoftRootSignatureAttributeArgs(ParsedAttributes &Attrs) {
+  assert(Tok.is(tok::identifier) &&
+         "Expected an identifier to denote which MS attribute to consider");
+  IdentifierInfo *RootSignatureIdent = Tok.getIdentifierInfo();
+  assert(RootSignatureIdent->getName() == "RootSignature" &&
+         "Expected RootSignature identifier for root signature attribute");
+
+  SourceLocation RootSignatureLoc = Tok.getLocation();
+  ConsumeToken();
+
+  // Ignore the left paren location for now.
+  BalancedDelimiterTracker T(*this, tok::l_paren);
+  if (T.consumeOpen()) {
+    Diag(Tok, diag::err_expected) << tok::l_paren;
+    return;
+  }
+
+  auto ProcessStringLiteral = [this]() -> std::optional<StringLiteral *> {
+    if (!isTokenStringLiteral())
+      return std::nullopt;
+
+    ExprResult StringResult = ParseUnevaluatedStringLiteralExpression();
+    if (StringResult.isInvalid())
+      return std::nullopt;
+
+    if (auto Lit = dyn_cast<StringLiteral>(StringResult.get()))
+      return Lit;
+
+    return std::nullopt;
+  };
+
+  auto StrLiteral = ProcessStringLiteral();
+  if (!StrLiteral.has_value()) {
+    Diag(Tok, diag::err_expected_string_literal)
+        << /*in attributes...*/ 4 << RootSignatureIdent->getName();
+    SkipUntil(tok::r_paren, StopAtSemi | StopBeforeMatch);
+    T.consumeClose();
+    return;
+  }
+
+  // Construct our identifier
+  StringRef Signature = StrLiteral.value()->getString();
----------------
AaronBallman wrote:

Sorry for the late post-commit review, but I just noticed this patch because it's doing a lot of sema work from within the parser. Is there a reason this wasn't done using an `ActOnWhatever` function in Sema? It seems to be the only use of `PushOnScopeChains` in the parser, which is how I noticed this. CC @llvm-beanz 

https://github.com/llvm/llvm-project/pull/137690