[clang-repl] Add a interpreter-specific overload of operator new for C++ (#76218)

This patch brings back the basic support for C by inserting the required
for value printing runtime only when we are in C++ mode. Additionally,
it defines a new overload of operator placement new because we can't
really forward declare it in a library-agnostic way.

Fixes the issue described in #69072.

Author: vgvassilev

clang/include/clang/Interpreter/Interpreter.h

@@ -129,7 +129,7 @@ class Interpreter {
   llvm::Expected<llvm::orc::ExecutorAddr>
   getSymbolAddressFromLinkerName(llvm::StringRef LinkerName) const;
 
-  enum InterfaceKind { NoAlloc, WithAlloc, CopyArray };
+  enum InterfaceKind { NoAlloc, WithAlloc, CopyArray, NewTag };
 
   const llvm::SmallVectorImpl<Expr *> &getValuePrintingInfo() const {
     return ValuePrintingInfo;
@@ -144,7 +144,7 @@ class Interpreter {
 
   llvm::DenseMap<CXXRecordDecl *, llvm::orc::ExecutorAddr> Dtors;
 
-  llvm::SmallVector<Expr *, 3> ValuePrintingInfo;
+  llvm::SmallVector<Expr *, 4> ValuePrintingInfo;
 };
 } // namespace clang
 

clang/lib/Interpreter/Interpreter.cpp

@@ -249,23 +249,26 @@ Interpreter::~Interpreter() {
 // can't find the precise resource directory in unittests so we have to hard
 // code them.
 const char *const Runtimes = R"(
-    void* operator new(__SIZE_TYPE__, void* __p) noexcept;
+#ifdef __cplusplus
     void *__clang_Interpreter_SetValueWithAlloc(void*, void*, void*);
     void __clang_Interpreter_SetValueNoAlloc(void*, void*, void*);
     void __clang_Interpreter_SetValueNoAlloc(void*, void*, void*, void*);
     void __clang_Interpreter_SetValueNoAlloc(void*, void*, void*, float);
     void __clang_Interpreter_SetValueNoAlloc(void*, void*, void*, double);
     void __clang_Interpreter_SetValueNoAlloc(void*, void*, void*, long double);
     void __clang_Interpreter_SetValueNoAlloc(void*,void*,void*,unsigned long long);
+    struct __clang_Interpreter_NewTag{} __ci_newtag;
+    void* operator new(__SIZE_TYPE__, void* __p, __clang_Interpreter_NewTag) noexcept;
     template <class T, class = T (*)() /*disable for arrays*/>
     void __clang_Interpreter_SetValueCopyArr(T* Src, void* Placement, unsigned long Size) {
       for (auto Idx = 0; Idx < Size; ++Idx)
-        new ((void*)(((T*)Placement) + Idx)) T(Src[Idx]);
+        new ((void*)(((T*)Placement) + Idx), __ci_newtag) T(Src[Idx]);
     }
     template <class T, unsigned long N>
     void __clang_Interpreter_SetValueCopyArr(const T (*Src)[N], void* Placement, unsigned long Size) {
       __clang_Interpreter_SetValueCopyArr(Src[0], Placement, Size);
     }
+#endif // __cplusplus
 )";
 
 llvm::Expected<std::unique_ptr<Interpreter>>
@@ -280,7 +283,7 @@ Interpreter::create(std::unique_ptr<CompilerInstance> CI) {
   if (!PTU)
     return PTU.takeError();
 
-  Interp->ValuePrintingInfo.resize(3);
+  Interp->ValuePrintingInfo.resize(4);
   // FIXME: This is a ugly hack. Undo command checks its availability by looking
   // at the size of the PTU list. However we have parsed something in the
   // beginning of the REPL so we have to mark them as 'Irrevocable'.
@@ -501,7 +504,7 @@ Interpreter::CompileDtorCall(CXXRecordDecl *CXXRD) {
 static constexpr llvm::StringRef MagicRuntimeInterface[] = {
     "__clang_Interpreter_SetValueNoAlloc",
     "__clang_Interpreter_SetValueWithAlloc",
-    "__clang_Interpreter_SetValueCopyArr"};
+    "__clang_Interpreter_SetValueCopyArr", "__ci_newtag"};
 
 bool Interpreter::FindRuntimeInterface() {
   if (llvm::all_of(ValuePrintingInfo, [](Expr *E) { return E != nullptr; }))
@@ -531,6 +534,9 @@ bool Interpreter::FindRuntimeInterface() {
   if (!LookupInterface(ValuePrintingInfo[CopyArray],
                        MagicRuntimeInterface[CopyArray]))
     return false;
+  if (!LookupInterface(ValuePrintingInfo[NewTag],
+                       MagicRuntimeInterface[NewTag]))
+    return false;
   return true;
 }
 
@@ -608,7 +614,9 @@ class RuntimeInterfaceBuilder
                 .getValuePrintingInfo()[Interpreter::InterfaceKind::CopyArray],
             SourceLocation(), Args, SourceLocation());
       }
-      Expr *Args[] = {AllocCall.get()};
+      Expr *Args[] = {
+          AllocCall.get(),
+          Interp.getValuePrintingInfo()[Interpreter::InterfaceKind::NewTag]};
       ExprResult CXXNewCall = S.BuildCXXNew(
           E->getSourceRange(),
           /*UseGlobal=*/true, /*PlacementLParen=*/SourceLocation(), Args,
@@ -629,8 +637,9 @@ class RuntimeInterfaceBuilder
           Interp.getValuePrintingInfo()[Interpreter::InterfaceKind::NoAlloc],
           E->getBeginLoc(), Args, E->getEndLoc());
     }
+    default:
+      llvm_unreachable("Unhandled Interpreter::InterfaceKind");
     }
-    llvm_unreachable("Unhandled Interpreter::InterfaceKind");
   }
 
   Interpreter::InterfaceKind VisitRecordType(const RecordType *Ty) {
@@ -815,3 +824,15 @@ __clang_Interpreter_SetValueNoAlloc(void *This, void *OutVal, void *OpaqueType,
   VRef = Value(static_cast<Interpreter *>(This), OpaqueType);
   VRef.setLongDouble(Val);
 }
+
+// A trampoline to work around the fact that operator placement new cannot
+// really be forward declared due to libc++ and libstdc++ declaration mismatch.
+// FIXME: __clang_Interpreter_NewTag is ODR violation because we get the same
+// definition in the interpreter runtime. We should move it in a runtime header
+// which gets included by the interpreter and here.
+struct __clang_Interpreter_NewTag {};
+REPL_EXTERNAL_VISIBILITY void *
+operator new(size_t __sz, void *__p, __clang_Interpreter_NewTag) noexcept {
+  // Just forward to the standard operator placement new.
+  return operator new(__sz, __p);
+}

clang/test/Interpreter/incremental-mode.cpp

@@ -1,3 +1,4 @@
 // RUN: clang-repl -Xcc -E
-// RUN: clang-repl -Xcc -emit-llvm 
+// RUN: clang-repl -Xcc -emit-llvm
+// RUN: clang-repl -Xcc -xc
 // expected-no-diagnostics

clang/unittests/Interpreter/InterpreterTest.cpp

@@ -248,28 +248,10 @@ TEST(IncrementalProcessing, FindMangledNameSymbol) {
 #endif // _WIN32
 }
 
-static void *AllocateObject(TypeDecl *TD, Interpreter &Interp) {
+static Value AllocateObject(TypeDecl *TD, Interpreter &Interp) {
   std::string Name = TD->getQualifiedNameAsString();
-  const clang::Type *RDTy = TD->getTypeForDecl();
-  clang::ASTContext &C = Interp.getCompilerInstance()->getASTContext();
-  size_t Size = C.getTypeSize(RDTy);
-  void *Addr = malloc(Size);
-
-  // Tell the interpreter to call the default ctor with this memory. Synthesize:
-  // new (loc) ClassName;
-  static unsigned Counter = 0;
-  std::stringstream SS;
-  SS << "auto _v" << Counter++ << " = "
-     << "new ((void*)"
-     // Windows needs us to prefix the hexadecimal value of a pointer with '0x'.
-     << std::hex << std::showbase << (size_t)Addr << ")" << Name << "();";
-
-  auto R = Interp.ParseAndExecute(SS.str());
-  if (!R) {
-    free(Addr);
-    return nullptr;
-  }
-
+  Value Addr;
+  cantFail(Interp.ParseAndExecute("new " + Name + "()", &Addr));
   return Addr;
 }
 
@@ -317,7 +299,7 @@ TEST(IncrementalProcessing, InstantiateTemplate) {
   }
 
   TypeDecl *TD = cast<TypeDecl>(LookupSingleName(*Interp, "A"));
-  void *NewA = AllocateObject(TD, *Interp);
+  Value NewA = AllocateObject(TD, *Interp);
 
   // Find back the template specialization
   VarDecl *VD = static_cast<VarDecl *>(*PTUDeclRange.begin());
@@ -328,8 +310,7 @@ TEST(IncrementalProcessing, InstantiateTemplate) {
   typedef int (*TemplateSpecFn)(void *);
   auto fn =
       cantFail(Interp->getSymbolAddress(MangledName)).toPtr<TemplateSpecFn>();
-  EXPECT_EQ(42, fn(NewA));
-  free(NewA);
+  EXPECT_EQ(42, fn(NewA.getPtr()));
 }
 
 #ifdef CLANG_INTERPRETER_NO_SUPPORT_EXEC