[Clang] Diagnose invalid function types in dependent contexts

[cor3ntin]

When forming an invalid function type, we were not diagnosing it if the call was dependent.

However, we later rely on the function type to be sensible during argument deduction.

We now diagnose anything that is not a potential function type, to avoid constructing bogus call expressions.

Fixes #138657
Fixes #115725
Fixes #68852

[llvmbot]

@llvm/pr-subscribers-clang

Author: cor3ntin (cor3ntin)

Changes

When forming an invalid function type, we were not diagnosing it if the call was dependent.

However, we later rely on the function type to be sensible during argument deduction.

We now diagnose anything that is not a potential function type, to avoid constructing bogus call expressions.

Fixes #138657
Fixes #115725
Fixes #68852

---

Full diff: https://github.com/llvm/llvm-project/pull/138731.diff

3 Files Affected:

• (modified) clang/docs/ReleaseNotes.rst (+1)
• (modified) clang/lib/Sema/SemaExpr.cpp (+19)
• (modified) clang/test/SemaTemplate/fun-template-def.cpp (+50-1)

diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
index 203958dab7430..bdd54f6a52b05 100644
--- a/clang/docs/ReleaseNotes.rst
+++ b/clang/docs/ReleaseNotes.rst
@@ -655,6 +655,7 @@ Bug Fixes to C++ Support
 - Fixed an assertion when trying to constant-fold various builtins when the argument
   referred to a reference to an incomplete type. (#GH129397)
 - Fixed a crash when a cast involved a parenthesized aggregate initialization in dependent context. (#GH72880)
+- Fixed a crash when forming an invalid function type in a dependent context. (#GH138657) (#GH115725) (#GH68852)
 
 Bug Fixes to AST Handling
 ^^^^^^^^^^^^^^^^^^^^^^^^^
diff --git a/clang/lib/Sema/SemaExpr.cpp b/clang/lib/Sema/SemaExpr.cpp
index be3f145f3c5f1..f463bffa83545 100644
--- a/clang/lib/Sema/SemaExpr.cpp
+++ b/clang/lib/Sema/SemaExpr.cpp
@@ -6541,6 +6541,15 @@ ExprResult Sema::ActOnCallExpr(Scope *Scope, Expr *Fn, SourceLocation LParenLoc,
   return Call;
 }
 
+// Any type that could be used to form a callable expression
+static bool MayBeFunctionType(const ASTContext &Context, QualType T) {
+  return T == Context.BoundMemberTy || T == Context.UnknownAnyTy ||
+         T == Context.BuiltinFnTy || T == Context.OverloadTy ||
+         T->isFunctionType() || T->isFunctionReferenceType() ||
+         T->isMemberFunctionPointerType() || T->isFunctionPointerType() ||
+         T->isBlockPointerType() || T->isRecordType();
+}
+
 ExprResult Sema::BuildCallExpr(Scope *Scope, Expr *Fn, SourceLocation LParenLoc,
                                MultiExprArg ArgExprs, SourceLocation RParenLoc,
                                Expr *ExecConfig, bool IsExecConfig,
@@ -6594,6 +6603,16 @@ ExprResult Sema::BuildCallExpr(Scope *Scope, Expr *Fn, SourceLocation LParenLoc,
             *this, dyn_cast<UnresolvedMemberExpr>(Fn->IgnoreParens()),
             Fn->getBeginLoc());
 
+        if (!Fn->getType()->isDependentType()) {
+          // If the type of the function itself is not dependent
+          // check that it is a reasonable as a function, as type deduction
+          // later assume the CallExpr has a sensible TYPE.
+          if (!MayBeFunctionType(Context, Fn->getType()))
+            return ExprError(
+                Diag(LParenLoc, diag::err_typecheck_call_not_function)
+                << Fn->getType() << Fn->getSourceRange());
+        }
+
         return CallExpr::Create(Context, Fn, ArgExprs, Context.DependentTy,
                                 VK_PRValue, RParenLoc, CurFPFeatureOverrides());
       }
diff --git a/clang/test/SemaTemplate/fun-template-def.cpp b/clang/test/SemaTemplate/fun-template-def.cpp
index de77901b5b601..716296e72bc44 100644
--- a/clang/test/SemaTemplate/fun-template-def.cpp
+++ b/clang/test/SemaTemplate/fun-template-def.cpp
@@ -1,6 +1,7 @@
 // RUN: %clang_cc1 -fsyntax-only -verify %s
 // RUN: %clang_cc1 -fsyntax-only -verify -std=c++98 %s
 // RUN: %clang_cc1 -fsyntax-only -verify -std=c++11 %s
+// RUN: %clang_cc1 -fsyntax-only -verify -std=c++20 %s
 
 // Tests that dependent expressions are always allowed, whereas non-dependent
 // are checked as usual.
@@ -32,7 +33,7 @@ T f1(T t1, U u1, int i1, T** tpp)
   i1 = t1[u1];
   i1 *= t1;
 
-  i1(u1, t1); // error
+  i1(u1, t1); // expected-error {{called object type 'int' is not a function or function pointer}}
   u1(i1, t1);
 
   U u2 = (T)i1;
@@ -60,3 +61,51 @@ void f3() {
   f2<int*>(0);
   f2<int>(0); // expected-error {{no matching function for call to 'f2'}}
 }
+
+#if __cplusplus >= 202002L
+namespace GH138657 {
+template <auto V> // #gh138657-template-head
+class meta {};
+template<int N>
+class meta<N()> {}; // expected-error {{called object type 'int' is not a function or function point}}
+
+template<int N[1]>
+class meta<N()> {}; // expected-error {{called object type 'int *' is not a function or function point}}
+
+template<char* N>
+class meta<N()> {}; // expected-error {{called object type 'char *' is not a function or function point}}
+
+struct S {};
+template<S>
+class meta<S()> {}; // expected-error {{template argument for non-type template parameter is treated as function type 'S ()'}}
+                    // expected-note@#gh138657-template-head {{template parameter is declared here}}
+
+}
+
+namespace GH115725 {
+template<auto ...> struct X {};
+template<typename T, typename ...Ts> struct A {
+  template<Ts ...Ns, T *...Ps>
+  A(X<0(Ps)...>, Ts (*...qs)[Ns]);
+  // expected-error@-1{{called object type 'int' is not a function or function pointer}}
+
+};
+}
+
+namespace GH68852 {
+template <auto v>
+struct constexpr_value {
+  template <class... Ts>
+  constexpr constexpr_value<v(Ts::value...)> call(Ts...) {
+    //expected-error@-1 {{called object type 'int' is not a function or function pointer}}
+    return {};
+  }
+};
+
+template <auto v> constexpr static inline auto c_ = constexpr_value<v>{};
+// expected-note@-1 {{in instantiation of template}}
+auto k = c_<1>; // expected-note {{in instantiation of variable}}
+
+}
+
+#endif

[erichkeane]

I might suggest putting these in teh same 'if' to make it more obvious what is happening (or perhaps have MayBeFunctionType return true for a dependent type?

[yuxuanchen1997]

Hi @cor3ntin,

This patch seems to have made expression SFINAEs in variable template partial specialization an hard error: https://godbolt.org/z/EosEEnMsf

Upstream clang rejects the program while clang-20, gcc, msvc accepted it. This change breaks folly invocability suite, which is used by large amount of currently working code.

Can we revert until this issue gets sorted out?

[ayermolo]

This is causing major breakages in our internal builds in multiple services, please revert.

[mizvekov]

This is reverted now.

[cor3ntin]

Sorry about that, folks. And thanks for the minimal repro, it's greatly appreciated.
I relanded the PR here #139246
I hope that resolves the issue you found