[Clang] Implement P2718R0 "Lifetime extension in range-based for loop…

…s" (#76361)

Implement P2718R0 "Lifetime extension in range-based for loops"
(https://wg21.link/P2718R0)

Differential Revision: https://reviews.llvm.org/D153701

---------

Signed-off-by: yronglin <yronglin777@gmail.com>

clang/docs/ReleaseNotes.rst

@@ -70,6 +70,9 @@ C++20 Feature Support
 C++23 Feature Support
 ^^^^^^^^^^^^^^^^^^^^^
 
+- Implemented `P2718R0: Lifetime extension in range-based for loops <https://wg21.link/P2718R0>`_. Also
+  materialize temporary object which is a prvalue in discarded-value expression.
+
 C++2c Feature Support
 ^^^^^^^^^^^^^^^^^^^^^
 

clang/include/clang/Parse/Parser.h

@@ -2402,7 +2402,7 @@ class Parser : public CodeCompletionHandler {
   struct ForRangeInit {
     SourceLocation ColonLoc;
     ExprResult RangeExpr;
-
+    SmallVector<MaterializeTemporaryExpr *, 8> LifetimeExtendTemps;
     bool ParsedForRangeDecl() { return !ColonLoc.isInvalid(); }
   };
   struct ForRangeInfo : ForRangeInit {

clang/include/clang/Sema/Sema.h

@@ -1342,6 +1342,12 @@ class Sema final {
     /// context not already known to be immediately invoked.
     llvm::SmallPtrSet<DeclRefExpr *, 4> ReferenceToConsteval;
 
+    /// P2718R0 - Lifetime extension in range-based for loops.
+    /// MaterializeTemporaryExprs in for-range-init expressions which need to
+    /// extend lifetime. Add MaterializeTemporaryExpr* if the value of
+    /// InLifetimeExtendingContext is true.
+    SmallVector<MaterializeTemporaryExpr *, 8> ForRangeLifetimeExtendTemps;
+
     /// \brief Describes whether we are in an expression constext which we have
     /// to handle differently.
     enum ExpressionKind {
@@ -1361,6 +1367,39 @@ class Sema final {
     // VLAs).
     bool InConditionallyConstantEvaluateContext = false;
 
+    /// Whether we are currently in a context in which all temporaries must be
+    /// lifetime-extended, even if they're not bound to a reference (for
+    /// example, in a for-range initializer).
+    bool InLifetimeExtendingContext = false;
+
+    /// Whether we are currently in a context in which all temporaries must be
+    /// materialized.
+    ///
+    /// [class.temporary]/p2:
+    /// The materialization of a temporary object is generally delayed as long
+    /// as possible in order to avoid creating unnecessary temporary objects.
+    ///
+    /// Temporary objects are materialized:
+    ///   (2.1) when binding a reference to a prvalue ([dcl.init.ref],
+    ///   [expr.type.conv], [expr.dynamic.cast], [expr.static.cast],
+    ///   [expr.const.cast], [expr.cast]),
+    ///
+    ///   (2.2) when performing member access on a class prvalue ([expr.ref],
+    ///   [expr.mptr.oper]),
+    ///
+    ///   (2.3) when performing an array-to-pointer conversion or subscripting
+    ///   on an array prvalue ([conv.array], [expr.sub]),
+    ///
+    ///   (2.4) when initializing an object of type
+    ///   std​::​initializer_list<T> from a braced-init-list
+    ///   ([dcl.init.list]),
+    ///
+    ///   (2.5) for certain unevaluated operands ([expr.typeid], [expr.sizeof])
+    ///
+    ///   (2.6) when a prvalue that has type other than cv void appears as a
+    ///   discarded-value expression ([expr.context]).
+    bool InMaterializeTemporaryObjectContext = false;
+
     // When evaluating immediate functions in the initializer of a default
     // argument or default member initializer, this is the declaration whose
     // default initializer is being evaluated and the location of the call
@@ -5253,22 +5292,17 @@ class Sema final {
     BFRK_Check
   };
 
-  StmtResult ActOnCXXForRangeStmt(Scope *S, SourceLocation ForLoc,
-                                  SourceLocation CoawaitLoc,
-                                  Stmt *InitStmt,
-                                  Stmt *LoopVar,
-                                  SourceLocation ColonLoc, Expr *Collection,
-                                  SourceLocation RParenLoc,
-                                  BuildForRangeKind Kind);
-  StmtResult BuildCXXForRangeStmt(SourceLocation ForLoc,
-                                  SourceLocation CoawaitLoc,
-                                  Stmt *InitStmt,
-                                  SourceLocation ColonLoc,
-                                  Stmt *RangeDecl, Stmt *Begin, Stmt *End,
-                                  Expr *Cond, Expr *Inc,
-                                  Stmt *LoopVarDecl,
-                                  SourceLocation RParenLoc,
-                                  BuildForRangeKind Kind);
+  StmtResult ActOnCXXForRangeStmt(
+      Scope *S, SourceLocation ForLoc, SourceLocation CoawaitLoc,
+      Stmt *InitStmt, Stmt *LoopVar, SourceLocation ColonLoc, Expr *Collection,
+      SourceLocation RParenLoc, BuildForRangeKind Kind,
+      ArrayRef<MaterializeTemporaryExpr *> LifetimeExtendTemps = {});
+  StmtResult BuildCXXForRangeStmt(
+      SourceLocation ForLoc, SourceLocation CoawaitLoc, Stmt *InitStmt,
+      SourceLocation ColonLoc, Stmt *RangeDecl, Stmt *Begin, Stmt *End,
+      Expr *Cond, Expr *Inc, Stmt *LoopVarDecl, SourceLocation RParenLoc,
+      BuildForRangeKind Kind,
+      ArrayRef<MaterializeTemporaryExpr *> LifetimeExtendTemps = {});
   StmtResult FinishCXXForRangeStmt(Stmt *ForRange, Stmt *Body);
 
   StmtResult ActOnGotoStmt(SourceLocation GotoLoc,
@@ -10010,6 +10044,18 @@ class Sema final {
     return currentEvaluationContext().isImmediateFunctionContext();
   }
 
+  bool isInLifetimeExtendingContext() const {
+    assert(!ExprEvalContexts.empty() &&
+           "Must be in an expression evaluation context");
+    return ExprEvalContexts.back().InLifetimeExtendingContext;
+  }
+
+  bool isInMaterializeTemporaryObjectContext() const {
+    assert(!ExprEvalContexts.empty() &&
+           "Must be in an expression evaluation context");
+    return ExprEvalContexts.back().InMaterializeTemporaryObjectContext;
+  }
+
   bool isCheckingDefaultArgumentOrInitializer() const {
     const ExpressionEvaluationContextRecord &Ctx = currentEvaluationContext();
     return (Ctx.Context ==
@@ -10049,6 +10095,32 @@ class Sema final {
     return Res;
   }
 
+  /// keepInLifetimeExtendingContext - Pull down InLifetimeExtendingContext
+  /// flag from previous context.
+  void keepInLifetimeExtendingContext() {
+    if (ExprEvalContexts.size() > 2 &&
+        ExprEvalContexts[ExprEvalContexts.size() - 2]
+            .InLifetimeExtendingContext) {
+      auto &LastRecord = ExprEvalContexts.back();
+      auto &PrevRecord = ExprEvalContexts[ExprEvalContexts.size() - 2];
+      LastRecord.InLifetimeExtendingContext =
+          PrevRecord.InLifetimeExtendingContext;
+    }
+  }
+
+  /// keepInMaterializeTemporaryObjectContext - Pull down
+  /// InMaterializeTemporaryObjectContext flag from previous context.
+  void keepInMaterializeTemporaryObjectContext() {
+    if (ExprEvalContexts.size() > 2 &&
+        ExprEvalContexts[ExprEvalContexts.size() - 2]
+            .InMaterializeTemporaryObjectContext) {
+      auto &LastRecord = ExprEvalContexts.back();
+      auto &PrevRecord = ExprEvalContexts[ExprEvalContexts.size() - 2];
+      LastRecord.InMaterializeTemporaryObjectContext =
+          PrevRecord.InMaterializeTemporaryObjectContext;
+    }
+  }
+
   /// RAII class used to determine whether SFINAE has
   /// trapped any errors that occur during template argument
   /// deduction.

clang/lib/Frontend/InitPreprocessor.cpp

@@ -643,7 +643,9 @@ static void InitializeCPlusPlusFeatureTestMacros(const LangOptions &LangOpts,
                                                                   : "200704");
     Builder.defineMacro("__cpp_constexpr_in_decltype", "201711L");
     Builder.defineMacro("__cpp_range_based_for",
-                        LangOpts.CPlusPlus17 ? "201603L" : "200907");
+                        LangOpts.CPlusPlus23   ? "202211L"
+                        : LangOpts.CPlusPlus17 ? "201603L"
+                                               : "200907");
     Builder.defineMacro("__cpp_static_assert", LangOpts.CPlusPlus26 ? "202306L"
                                                : LangOpts.CPlusPlus17
                                                    ? "201411L"

clang/lib/Parse/ParseDecl.cpp

@@ -2312,12 +2312,38 @@ Parser::DeclGroupPtrTy Parser::ParseDeclGroup(ParsingDeclSpec &DS,
     bool IsForRangeLoop = false;
     if (TryConsumeToken(tok::colon, FRI->ColonLoc)) {
       IsForRangeLoop = true;
+      EnterExpressionEvaluationContext ForRangeInitContext(
+          Actions, Sema::ExpressionEvaluationContext::PotentiallyEvaluated,
+          /*LambdaContextDecl=*/nullptr,
+          Sema::ExpressionEvaluationContextRecord::EK_Other,
+          getLangOpts().CPlusPlus23);
+
+      // P2718R0 - Lifetime extension in range-based for loops.
+      if (getLangOpts().CPlusPlus23) {
+        auto &LastRecord = Actions.ExprEvalContexts.back();
+        LastRecord.InLifetimeExtendingContext = true;
+
+        // Materialize non-`cv void` prvalue temporaries in discarded
+        // expressions. These materialized temporaries may be lifetime-extented.
+        LastRecord.InMaterializeTemporaryObjectContext = true;
+      }
+
       if (getLangOpts().OpenMP)
         Actions.startOpenMPCXXRangeFor();
       if (Tok.is(tok::l_brace))
         FRI->RangeExpr = ParseBraceInitializer();
       else
         FRI->RangeExpr = ParseExpression();
+
+      // Before c++23, ForRangeLifetimeExtendTemps should be empty.
+      assert(
+          getLangOpts().CPlusPlus23 ||
+          Actions.ExprEvalContexts.back().ForRangeLifetimeExtendTemps.empty());
+
+      // Move the collected materialized temporaries into ForRangeInit before
+      // ForRangeInitContext exit.
+      FRI->LifetimeExtendTemps = std::move(
+          Actions.ExprEvalContexts.back().ForRangeLifetimeExtendTemps);
     }
 
     Decl *ThisDecl = Actions.ActOnDeclarator(getCurScope(), D);

clang/lib/Parse/ParseStmt.cpp

@@ -2288,11 +2288,11 @@ StmtResult Parser::ParseForStatement(SourceLocation *TrailingElseLoc) {
     ForRangeStmt = Actions.ActOnCXXForRangeStmt(
         getCurScope(), ForLoc, CoawaitLoc, FirstPart.get(),
         ForRangeInfo.LoopVar.get(), ForRangeInfo.ColonLoc, CorrectedRange.get(),
-        T.getCloseLocation(), Sema::BFRK_Build);
-
-  // Similarly, we need to do the semantic analysis for a for-range
-  // statement immediately in order to close over temporaries correctly.
+        T.getCloseLocation(), Sema::BFRK_Build,
+        ForRangeInfo.LifetimeExtendTemps);
   } else if (ForEach) {
+    // Similarly, we need to do the semantic analysis for a for-range
+    // statement immediately in order to close over temporaries correctly.
     ForEachStmt = Actions.ActOnObjCForCollectionStmt(ForLoc,
                                                      FirstPart.get(),
                                                      Collection.get(),

clang/lib/Sema/SemaExpr.cpp

@@ -6259,7 +6259,7 @@ ExprResult Sema::BuildCXXDefaultArgExpr(SourceLocation CallLoc,
   assert(Param->hasDefaultArg() && "can't build nonexistent default arg");
 
   bool NestedDefaultChecking = isCheckingDefaultArgumentOrInitializer();
-
+  bool InLifetimeExtendingContext = isInLifetimeExtendingContext();
   std::optional<ExpressionEvaluationContextRecord::InitializationContext>
       InitializationContext =
           OutermostDeclarationWithDelayedImmediateInvocations();
@@ -6292,9 +6292,17 @@ ExprResult Sema::BuildCXXDefaultArgExpr(SourceLocation CallLoc,
     ImmediateCallVisitor V(getASTContext());
     if (!NestedDefaultChecking)
       V.TraverseDecl(Param);
-    if (V.HasImmediateCalls) {
-      ExprEvalContexts.back().DelayedDefaultInitializationContext = {
-          CallLoc, Param, CurContext};
+
+    // Rewrite the call argument that was created from the corresponding
+    // parameter's default argument.
+    if (V.HasImmediateCalls || InLifetimeExtendingContext) {
+      if (V.HasImmediateCalls)
+        ExprEvalContexts.back().DelayedDefaultInitializationContext = {
+            CallLoc, Param, CurContext};
+      // Pass down lifetime extending flag, and collect temporaries in
+      // CreateMaterializeTemporaryExpr when we rewrite the call argument.
+      keepInLifetimeExtendingContext();
+      keepInMaterializeTemporaryObjectContext();
       EnsureImmediateInvocationInDefaultArgs Immediate(*this);
       ExprResult Res;
       runWithSufficientStackSpace(CallLoc, [&] {
@@ -18655,6 +18663,16 @@ void Sema::PopExpressionEvaluationContext() {
     }
   }
 
+  // Append the collected materialized temporaries into previous context before
+  // exit if the previous also is a lifetime extending context.
+  auto &PrevRecord = ExprEvalContexts[ExprEvalContexts.size() - 2];
+  if (getLangOpts().CPlusPlus23 && isInLifetimeExtendingContext() &&
+      PrevRecord.InLifetimeExtendingContext && !ExprEvalContexts.empty()) {
+    auto &PrevRecord = ExprEvalContexts[ExprEvalContexts.size() - 2];
+    PrevRecord.ForRangeLifetimeExtendTemps.append(
+        Rec.ForRangeLifetimeExtendTemps);
+  }
+
   WarnOnPendingNoDerefs(Rec);
   HandleImmediateInvocations(*this, Rec);
 

clang/lib/Sema/SemaExprCXX.cpp

@@ -8222,21 +8222,6 @@ ExprResult Sema::IgnoredValueConversions(Expr *E) {
     E = result.get();
   }
 
-  // C99 6.3.2.1:
-  //   [Except in specific positions,] an lvalue that does not have
-  //   array type is converted to the value stored in the
-  //   designated object (and is no longer an lvalue).
-  if (E->isPRValue()) {
-    // In C, function designators (i.e. expressions of function type)
-    // are r-values, but we still want to do function-to-pointer decay
-    // on them.  This is both technically correct and convenient for
-    // some clients.
-    if (!getLangOpts().CPlusPlus && E->getType()->isFunctionType())
-      return DefaultFunctionArrayConversion(E);
-
-    return E;
-  }
-
   if (getLangOpts().CPlusPlus) {
     // The C++11 standard defines the notion of a discarded-value expression;
     // normally, we don't need to do anything to handle it, but if it is a
@@ -8257,11 +8242,32 @@ ExprResult Sema::IgnoredValueConversions(Expr *E) {
     //   If the expression is a prvalue after this optional conversion, the
     //   temporary materialization conversion is applied.
     //
-    // We skip this step: IR generation is able to synthesize the storage for
-    // itself in the aggregate case, and adding the extra node to the AST is
-    // just clutter.
-    // FIXME: We don't emit lifetime markers for the temporaries due to this.
-    // FIXME: Do any other AST consumers care about this?
+    // We do not materialize temporaries by default in order to avoid creating
+    // unnecessary temporary objects. If we skip this step, IR generation is
+    // able to synthesize the storage for itself in the aggregate case, and
+    // adding the extra node to the AST is just clutter.
+    if (isInMaterializeTemporaryObjectContext() && getLangOpts().CPlusPlus17 &&
+        E->isPRValue() && !E->getType()->isVoidType()) {
+      ExprResult Res = TemporaryMaterializationConversion(E);
+      if (Res.isInvalid())
+        return E;
+      E = Res.get();
+    }
+    return E;
+  }
+
+  // C99 6.3.2.1:
+  //   [Except in specific positions,] an lvalue that does not have
+  //   array type is converted to the value stored in the
+  //   designated object (and is no longer an lvalue).
+  if (E->isPRValue()) {
+    // In C, function designators (i.e. expressions of function type)
+    // are r-values, but we still want to do function-to-pointer decay
+    // on them.  This is both technically correct and convenient for
+    // some clients.
+    if (!getLangOpts().CPlusPlus && E->getType()->isFunctionType())
+      return DefaultFunctionArrayConversion(E);
+
     return E;
   }
 

clang/lib/Sema/SemaInit.cpp

@@ -8491,6 +8491,10 @@ Sema::CreateMaterializeTemporaryExpr(QualType T, Expr *Temporary,
   // are done in both CreateMaterializeTemporaryExpr and MaybeBindToTemporary,
   // but there may be a chance to merge them.
   Cleanup.setExprNeedsCleanups(false);
+  if (isInLifetimeExtendingContext()) {
+    auto &Record = ExprEvalContexts.back();
+    Record.ForRangeLifetimeExtendTemps.push_back(MTE);
+  }
   return MTE;
 }