[Clang] Look through type sugar when accessing FunctionProtoType

[Sirraide]

This fixes a bug introduced by #84473: if a lambda’s type is type sugar (e.g. an AttributedType), we need to use getAs() instead of cast() to retrieve the FunctionProtoType.

[llvmbot]

@llvm/pr-subscribers-clang

Author: None (Sirraide)

Changes

This fixes a bug introduced by #84473: if a lambda’s type is type sugar (e.g. an AttributedType), we need to use getAs() instead of cast() to retrieve the FunctionProtoType.

---

Full diff: https://github.com/llvm/llvm-project/pull/88428.diff

4 Files Affected:

• (modified) clang/lib/Sema/SemaExpr.cpp (+1-1)
• (modified) clang/lib/Sema/SemaExprCXX.cpp (+1-1)
• (modified) clang/test/SemaCXX/cxx2b-deducing-this.cpp (+12)
• (modified) clang/test/SemaCXX/lambda-expressions.cpp (+9)

diff --git a/clang/lib/Sema/SemaExpr.cpp b/clang/lib/Sema/SemaExpr.cpp
index 45acbf197ea6b4..2d8b09df939db9 100644
--- a/clang/lib/Sema/SemaExpr.cpp
+++ b/clang/lib/Sema/SemaExpr.cpp
@@ -20743,7 +20743,7 @@ static void FixDependencyOfIdExpressionsInLambdaWithDependentObjectParameter(
       if (MD->getType().isNull())
         continue;
 
-      const auto *Ty = cast<FunctionProtoType>(MD->getType());
+      const auto *Ty = MD->getType()->getAs<FunctionProtoType>();
       if (!Ty || !MD->isExplicitObjectMemberFunction() ||
           !Ty->getParamType(0)->isDependentType())
         continue;
diff --git a/clang/lib/Sema/SemaExprCXX.cpp b/clang/lib/Sema/SemaExprCXX.cpp
index 9822477260e592..58b3607b2fdc43 100644
--- a/clang/lib/Sema/SemaExprCXX.cpp
+++ b/clang/lib/Sema/SemaExprCXX.cpp
@@ -1489,7 +1489,7 @@ void Sema::MarkThisReferenced(CXXThisExpr *This) {
         if (MD->getType().isNull())
           return false;
 
-        const auto *Ty = cast<FunctionProtoType>(MD->getType());
+        const auto *Ty = MD->getType()->getAs<FunctionProtoType>();
         return Ty && MD->isExplicitObjectMemberFunction() &&
                Ty->getParamType(0)->isDependentType();
       }
diff --git a/clang/test/SemaCXX/cxx2b-deducing-this.cpp b/clang/test/SemaCXX/cxx2b-deducing-this.cpp
index c14c971afd2359..5f29a955e053c3 100644
--- a/clang/test/SemaCXX/cxx2b-deducing-this.cpp
+++ b/clang/test/SemaCXX/cxx2b-deducing-this.cpp
@@ -312,6 +312,18 @@ void TestMutationInLambda() {
       l1();
       l2();
     }
+
+    // Check that we don't crash if the lambda has type sugar.
+    const auto l15 = [=](this auto&&) [[clang::annotate_type("foo")]] [[clang::annotate_type("bar")]] {
+        return x;
+    };
+
+    const auto l16 = [=]() [[clang::annotate_type("foo")]] [[clang::annotate_type("bar")]] {
+        return x;
+    };
+
+    l15();
+    l16();
 }
 
 struct Over_Call_Func_Example {
diff --git a/clang/test/SemaCXX/lambda-expressions.cpp b/clang/test/SemaCXX/lambda-expressions.cpp
index 151d74f21d64dc..6be338064452ed 100644
--- a/clang/test/SemaCXX/lambda-expressions.cpp
+++ b/clang/test/SemaCXX/lambda-expressions.cpp
@@ -762,3 +762,12 @@ template auto t::operator()<int>(int a) const; // expected-note {{in instantiati
 
 }
 #endif
+
+namespace GH84473_bug {
+void f1() {
+  int b;
+  (void) [=] [[gnu::regcall]] () { // expected-warning {{an attribute specifier sequence in this position is a C++23 extension}}
+    (void) b;
+  };
+}
+}

[erichkeane]

Ooof, yeah, good catch. I should have caught that in the review, I've fallen for that one before :)

[Sirraide]

Ooof, yeah, good catch. I should have caught that in the review, I've fallen for that one before :)

I was just now made aware of the fact that this asserts and the only reason I found the problem fairly quickly is because I happen to have a pr open at the moment that deals w/ type sugar on lambdas.