CodeGen: Properly null-check typeid expressions

[majnemer]

Summary

Thoroughly check for a pointer dereference which yields a glvalue. Look
through casts, comma operators, conditional operators, paren
expressions, etc.

This was originally D4416.

[rsmith]

lib/CodeGen/CGExprCXX.cpp

1632 ↗

(On Diff #11669)

Do you need to walk through OpaqueValueExprs to support this for binary conditional operators? Does this work:

struct T { operator bool(); } *p, t, &q = t;
typeid(*p ?: q);

[majnemer]

• Handle OpaqueValueExprs in typeid correctly

[rsmith]

LGTM

test/CodeGenCXX/typeid-should-throw.cpp

62–64 ↗

(On Diff #11671)

You could make t and r parameters to match the style of other tests.

[majnemer]

Closed by commit rL213434 (authored by @majnemer).

cfe/trunk/lib/CodeGen/CGExprCXX.cpp

void CodeGenFunction::EmitCXXDeleteExpr(const CXXDeleteExpr *E) {
   } else {
     EmitObjectDelete(*this, E->getOperatorDelete(), Ptr, DeleteTy,
                      E->isGlobalDelete());
   }
 
   EmitBlock(DeleteEnd);
 }
 
+static bool isGLValueFromPointerDeref(const Expr *E) {
+  E = E->IgnoreParens();
+
+  if (const auto *CE = dyn_cast<CastExpr>(E)) {
+    if (!CE->getSubExpr()->isGLValue())
+      return false;
+    return isGLValueFromPointerDeref(CE->getSubExpr());
+  }
+
+  if (const auto *OVE = dyn_cast<OpaqueValueExpr>(E))
+    return isGLValueFromPointerDeref(OVE->getSourceExpr());
+
+  if (const auto *BO = dyn_cast<BinaryOperator>(E))
+    if (BO->getOpcode() == BO_Comma)
+      return isGLValueFromPointerDeref(BO->getRHS());
+
+  if (const auto *ACO = dyn_cast<AbstractConditionalOperator>(E))
+    return isGLValueFromPointerDeref(ACO->getTrueExpr()) ||
+           isGLValueFromPointerDeref(ACO->getFalseExpr());
+
+  // C++11 [expr.sub]p1:
+  //   The expression E1[E2] is identical (by definition) to *((E1)+(E2))
+  if (isa<ArraySubscriptExpr>(E))
+    return true;
+
+  if (const auto *UO = dyn_cast<UnaryOperator>(E))
+    if (UO->getOpcode() == UO_Deref)
+      return true;
+
+  return false;
+}
+
 static llvm::Value *EmitTypeidFromVTable(CodeGenFunction &CGF, const Expr *E,
                                          llvm::Type *StdTypeInfoPtrTy) {
   // Get the vtable pointer.
   llvm::Value *ThisPtr = CGF.EmitLValue(E).getAddress();
 
   // C++ [expr.typeid]p2:
   //   If the glvalue expression is obtained by applying the unary * operator to
   //   a pointer and the pointer is a null pointer value, the typeid expression
   //   throws the std::bad_typeid exception.
-  bool IsDeref = false;
-  if (const UnaryOperator *UO = dyn_cast<UnaryOperator>(E->IgnoreParens()))
-    if (UO->getOpcode() == UO_Deref)
-      IsDeref = true;
-
+  //
+  // However, this paragraph's intent is not clear.  We choose a very generous
+  // interpretation which implores us to consider comma operators, conditional
+  // operators, parentheses and other such constructs.
   QualType SrcRecordTy = E->getType();
-  if (CGF.CGM.getCXXABI().shouldTypeidBeNullChecked(IsDeref, SrcRecordTy)) {
+  if (CGF.CGM.getCXXABI().shouldTypeidBeNullChecked(
+          isGLValueFromPointerDeref(E), SrcRecordTy)) {
     llvm::BasicBlock *BadTypeidBlock =
         CGF.createBasicBlock("typeid.bad_typeid");
     llvm::BasicBlock *EndBlock = CGF.createBasicBlock("typeid.end");
 
     llvm::Value *IsNull = CGF.Builder.CreateIsNull(ThisPtr);
     CGF.Builder.CreateCondBr(IsNull, BadTypeidBlock, EndBlock);
 
     CGF.EmitBlock(BadTypeidBlock);

cfe/trunk/test/CodeGenCXX/typeid-should-throw.cpp

+// RUN: %clang_cc1 %s -triple %itanium_abi_triple -Wno-unused-value -emit-llvm -o - -std=c++11 | FileCheck %s
+namespace std {
+struct type_info;
+}
+
+struct A {
+  virtual ~A();
+  operator bool();
+};
+struct B : A {};
+
+void f1(A *x) { typeid(false, *x); }
+// CHECK-LABEL: define void @_Z2f1P1A
+// CHECK:       icmp eq {{.*}}, null
+// CHECK-NEXT:  br i1
+
+void f2(bool b, A *x, A *y) { typeid(b ? *x : *y); }
+// CHECK-LABEL: define void @_Z2f2bP1AS0_
+// CHECK:       icmp eq {{.*}}, null
+// CHECK-NEXT:  br i1
+
+void f3(bool b, A *x, A &y) { typeid(b ? *x : y); }
+// CHECK-LABEL: define void @_Z2f3bP1ARS_
+// CHECK:       icmp eq {{.*}}, null
+// CHECK-NEXT:  br i1
+
+void f4(bool b, A &x, A *y) { typeid(b ? x : *y); }
+// CHECK-LABEL: define void @_Z2f4bR1APS_
+// CHECK:       icmp eq {{.*}}, null
+// CHECK-NEXT:  br i1
+
+void f5(volatile A *x) { typeid(*x); }
+// CHECK-LABEL: define void @_Z2f5PV1A
+// CHECK:       icmp eq {{.*}}, null
+// CHECK-NEXT:  br i1
+
+void f6(A *x) { typeid((B &)*(B *)x); }
+// CHECK-LABEL: define void @_Z2f6P1A
+// CHECK:       icmp eq {{.*}}, null
+// CHECK-NEXT:  br i1
+
+void f7(A *x) { typeid((*x)); }
+// CHECK-LABEL: define void @_Z2f7P1A
+// CHECK:       icmp eq {{.*}}, null
+// CHECK-NEXT:  br i1
+
+void f8(A *x) { typeid(x[0]); }
+// CHECK-LABEL: define void @_Z2f8P1A
+// CHECK:       icmp eq {{.*}}, null
+// CHECK-NEXT:  br i1
+
+void f9(A *x) { typeid(0[x]); }
+// CHECK-LABEL: define void @_Z2f9P1A
+// CHECK:       icmp eq {{.*}}, null
+// CHECK-NEXT:  br i1
+
+void f10(A *x, A *y) { typeid(*y ?: *x); }
+// CHECK-LABEL: define void @_Z3f10P1AS0_
+// CHECK:       icmp eq {{.*}}, null
+// CHECK-NEXT:  br i1
+
+void f11(A *x, A &y) { typeid(*x ?: y); }
+// CHECK-LABEL: define void @_Z3f11P1ARS_
+// CHECK:       icmp eq {{.*}}, null
+// CHECK-NEXT:  br i1
+
+void f12(A &x, A *y) { typeid(x ?: *y); }
+// CHECK-LABEL: define void @_Z3f12R1APS_
+// CHECK:       icmp eq {{.*}}, null
+// CHECK-NEXT:  br i1
+
+void f13(A &x, A &y) { typeid(x ?: y); }
+// CHECK-LABEL: define void @_Z3f13R1AS0_
+// CHECK-NOT:   icmp eq {{.*}}, null
+
+void f14(A *x) { typeid((const A &)(A)*x); }
+// CHECK-LABEL: define void @_Z3f14P1A
+// CHECK-NOT:   icmp eq {{.*}}, null
+
+void f15(A *x) { typeid((A &&)*(A *)nullptr); }
+// CHECK-LABEL: define void @_Z3f15P1A
+// CHECK-NOT:   icmp eq {{.*}}, null