Introduce new queries on ObjCRuntime for how to interpret subscripts

on object pointers and whether pointer arithmetic on object pointers
is supported.  Make ObjFW interpret subscripts as pseudo-objects.
Based on a patch by Jonathan Schleifer.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@161028 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 121 insertions, 64 deletions

diff --git a/include/clang/Basic/ObjCRuntime.h b/include/clang/Basic/ObjCRuntime.h
index da3ecca0f8..b24fe7cd91 100644
--- a/include/clang/Basic/ObjCRuntime.h
+++ b/include/clang/Basic/ObjCRuntime.h
@@ -170,6 +170,34 @@ public:
     llvm_unreachable("bad kind");
   }
 
+  /// \brief Does this runtime allow sizeof or alignof on object types?
+  bool allowsSizeofAlignof() const {
+    return isFragile();
+  }
+
+  /// \brief Does this runtime allow pointer arithmetic on objects?
+  ///
+  /// This covers +, -, ++, --, and (if isSubscriptPointerArithmetic()
+  /// yields true) [].
+  bool allowsPointerArithmetic() const {
+    switch (getKind()) {
+    case FragileMacOSX:
+    case GCC:
+      return true;
+    case MacOSX:
+    case iOS:
+    case GNUstep:
+    case ObjFW:
+      return false;
+    }
+    llvm_unreachable("bad kind");
+  }
+
+  /// \brief Is subscripting pointer arithmetic?
+  bool isSubscriptPointerArithmetic() const {
+    return allowsPointerArithmetic();
+  }
+
   /// \brief Does this runtime provide an objc_terminate function?
   ///
   /// This is used in handlers for exceptions during the unwind process;
diff --git a/lib/CodeGen/CGObjCGNU.cpp b/lib/CodeGen/CGObjCGNU.cpp
index 9993f448f7..166bb9a2c0 100644
--- a/lib/CodeGen/CGObjCGNU.cpp
+++ b/lib/CodeGen/CGObjCGNU.cpp
@@ -653,8 +653,11 @@ class CGObjCGNUstep : public CGObjCGNU {
     }
 };
 
-/// Class used when targeting the ObjFW runtime.
-class CGObjCObjFW: public CGObjCGCC {
+/// The ObjFW runtime, which closely follows the GCC runtime's
+/// compiler ABI.  Support here is due to Jonathan Schleifer, the
+/// ObjFW maintainer.
+class CGObjCObjFW : public CGObjCGCC {
+  /// Emit class references unconditionally as direct symbol references.
   virtual llvm::Value *GetClassNamed(CGBuilderTy &Builder,
                                      const std::string &Name, bool isWeak) {
     if (isWeak)
diff --git a/lib/Sema/SemaExpr.cpp b/lib/Sema/SemaExpr.cpp
index 19259b30ec..f13408e632 100644
--- a/lib/Sema/SemaExpr.cpp
+++ b/lib/Sema/SemaExpr.cpp
@@ -2902,8 +2902,9 @@ static bool CheckObjCTraitOperandConstraints(Sema &S, QualType T,
                                              SourceLocation Loc,
                                              SourceRange ArgRange,
                                              UnaryExprOrTypeTrait TraitKind) {
-  // Reject sizeof(interface) and sizeof(interface<proto>) in 64-bit mode.
-  if (S.LangOpts.ObjCRuntime.isNonFragile() && T->isObjCObjectType()) {
+  // Reject sizeof(interface) and sizeof(interface<proto>) if the
+  // runtime doesn't allow it.
+  if (!S.LangOpts.ObjCRuntime.allowsSizeofAlignof() && T->isObjCObjectType()) {
     S.Diag(Loc, diag::err_sizeof_nonfragile_interface)
       << T << (TraitKind == UETT_SizeOf)
       << ArgRange;
@@ -3193,6 +3194,22 @@ Sema::ActOnPostfixUnaryOp(Scope *S, SourceLocation OpLoc,
   return BuildUnaryOp(S, OpLoc, Opc, Input);
 }
 
+/// \brief Diagnose if arithmetic on the given ObjC pointer is illegal.
+///
+/// \return true on error
+static bool checkArithmeticOnObjCPointer(Sema &S,
+                                         SourceLocation opLoc,
+                                         Expr *op) {
+  assert(op->getType()->isObjCObjectPointerType());
+  if (S.LangOpts.ObjCRuntime.allowsPointerArithmetic())
+    return false;
+
+  S.Diag(opLoc, diag::err_arithmetic_nonfragile_interface)
+    << op->getType()->castAs<ObjCObjectPointerType>()->getPointeeType()
+    << op->getSourceRange();
+  return true;
+}
+
 ExprResult
 Sema::ActOnArraySubscriptExpr(Scope *S, Expr *Base, SourceLocation LLoc,
                               Expr *Idx, SourceLocation RLoc) {
@@ -3223,7 +3240,6 @@ Sema::ActOnArraySubscriptExpr(Scope *S, Expr *Base, SourceLocation LLoc,
   return CreateBuiltinArraySubscriptExpr(Base, LLoc, Idx, RLoc);
 }
 
-
 ExprResult
 Sema::CreateBuiltinArraySubscriptExpr(Expr *Base, SourceLocation LLoc,
                                       Expr *Idx, SourceLocation RLoc) {
@@ -3261,13 +3277,21 @@ Sema::CreateBuiltinArraySubscriptExpr(Expr *Base, SourceLocation LLoc,
     IndexExpr = RHSExp;
     ResultType = PTy->getPointeeType();
   } else if (const ObjCObjectPointerType *PTy =
-             LHSTy->getAs<ObjCObjectPointerType>()) {
+               LHSTy->getAs<ObjCObjectPointerType>()) {
     BaseExpr = LHSExp;
     IndexExpr = RHSExp;
-    Result = BuildObjCSubscriptExpression(RLoc, BaseExpr, IndexExpr, 0, 0);
-    if (!Result.isInvalid())
-      return Owned(Result.take());
+
+    // Use custom logic if this should be the pseudo-object subscript
+    // expression.
+    if (!LangOpts.ObjCRuntime.isSubscriptPointerArithmetic())
+      return BuildObjCSubscriptExpression(RLoc, BaseExpr, IndexExpr, 0, 0);
+
     ResultType = PTy->getPointeeType();
+    if (!LangOpts.ObjCRuntime.allowsPointerArithmetic()) {
+      Diag(LLoc, diag::err_subscript_nonfragile_interface)
+        << ResultType << BaseExpr->getSourceRange();
+      return ExprError();
+    }
   } else if (const PointerType *PTy = RHSTy->getAs<PointerType>()) {
      // Handle the uncommon case of "123[Ptr]".
     BaseExpr = RHSExp;
@@ -3279,6 +3303,11 @@ Sema::CreateBuiltinArraySubscriptExpr(Expr *Base, SourceLocation LLoc,
     BaseExpr = RHSExp;
     IndexExpr = LHSExp;
     ResultType = PTy->getPointeeType();
+    if (!LangOpts.ObjCRuntime.allowsPointerArithmetic()) {
+      Diag(LLoc, diag::err_subscript_nonfragile_interface)
+        << ResultType << BaseExpr->getSourceRange();
+      return ExprError();
+    }
   } else if (const VectorType *VTy = LHSTy->getAs<VectorType>()) {
     BaseExpr = LHSExp;    // vectors: V[123]
     IndexExpr = RHSExp;
@@ -3351,13 +3380,6 @@ Sema::CreateBuiltinArraySubscriptExpr(Expr *Base, SourceLocation LLoc,
                           diag::err_subscript_incomplete_type, BaseExpr))
     return ExprError();
 
-  // Diagnose bad cases where we step over interface counts.
-  if (ResultType->isObjCObjectType() && LangOpts.ObjCRuntime.isNonFragile()) {
-    Diag(LLoc, diag::err_subscript_nonfragile_interface)
-      << ResultType << BaseExpr->getSourceRange();
-    return ExprError();
-  }
-
   assert(VK == VK_RValue || LangOpts.CPlusPlus ||
          !ResultType.isCForbiddenLValueType());
 
@@ -6173,17 +6195,12 @@ static void diagnoseArithmeticOnFunctionPointer(Sema &S, SourceLocation Loc,
 /// \returns True if pointer has incomplete type
 static bool checkArithmeticIncompletePointerType(Sema &S, SourceLocation Loc,
                                                  Expr *Operand) {
-  if ((Operand->getType()->isPointerType() &&
-       !Operand->getType()->isDependentType()) ||
-      Operand->getType()->isObjCObjectPointerType()) {
-    QualType PointeeTy = Operand->getType()->getPointeeType();
-    if (S.RequireCompleteType(
-          Loc, PointeeTy,
-          diag::err_typecheck_arithmetic_incomplete_type,
-          PointeeTy, Operand->getSourceRange()))
-      return true;
-  }
-  return false;
+  assert(Operand->getType()->isAnyPointerType() &&
+         !Operand->getType()->isDependentType());
+  QualType PointeeTy = Operand->getType()->getPointeeType();
+  return S.RequireCompleteType(Loc, PointeeTy,
+                               diag::err_typecheck_arithmetic_incomplete_type,
+                               PointeeTy, Operand->getSourceRange());
 }
 
 /// \brief Check the validity of an arithmetic pointer operand.
@@ -6254,26 +6271,14 @@ static bool checkArithmeticBinOpPointerOperands(Sema &S, SourceLocation Loc,
     return !S.getLangOpts().CPlusPlus;
   }
 
-  if (checkArithmeticIncompletePointerType(S, Loc, LHSExpr)) return false;
-  if (checkArithmeticIncompletePointerType(S, Loc, RHSExpr)) return false;
+  if (isLHSPointer && checkArithmeticIncompletePointerType(S, Loc, LHSExpr))
+    return false;
+  if (isRHSPointer && checkArithmeticIncompletePointerType(S, Loc, RHSExpr))
+    return false;
 
   return true;
 }
 
-/// \brief Check bad cases where we step over interface counts.
-static bool checkArithmethicPointerOnNonFragileABI(Sema &S,
-                                                   SourceLocation OpLoc,
-                                                   Expr *Op) {
-  assert(Op->getType()->isAnyPointerType());
-  QualType PointeeTy = Op->getType()->getPointeeType();
-  if (!PointeeTy->isObjCObjectType() || S.LangOpts.ObjCRuntime.isFragile())
-    return true;
-
-  S.Diag(OpLoc, diag::err_arithmetic_nonfragile_interface)
-    << PointeeTy << Op->getSourceRange();
-  return false;
-}
-
 /// diagnoseStringPlusInt - Emit a warning when adding an integer to a string
 /// literal.
 static void diagnoseStringPlusInt(Sema &Self, SourceLocation OpLoc,
@@ -6350,13 +6355,26 @@ QualType Sema::CheckAdditionOperands( // C99 6.5.6
     return compType;
   }
 
-  // Put any potential pointer into PExp
-  Expr* PExp = LHS.get(), *IExp = RHS.get();
-  if (IExp->getType()->isAnyPointerType())
-    std::swap(PExp, IExp);
+  // Type-checking.  Ultimately the pointer's going to be in PExp;
+  // note that we bias towards the LHS being the pointer.
+  Expr *PExp = LHS.get(), *IExp = RHS.get();
 
-  if (!PExp->getType()->isAnyPointerType())
-    return InvalidOperands(Loc, LHS, RHS);
+  bool isObjCPointer;
+  if (PExp->getType()->isPointerType()) {
+    isObjCPointer = false;
+  } else if (PExp->getType()->isObjCObjectPointerType()) {
+    isObjCPointer = true;
+  } else {
+    std::swap(PExp, IExp);
+    if (PExp->getType()->isPointerType()) {
+      isObjCPointer = false;
+    } else if (PExp->getType()->isObjCObjectPointerType()) {
+      isObjCPointer = true;
+    } else {
+      return InvalidOperands(Loc, LHS, RHS);
+    }
+  }
+  assert(PExp->getType()->isAnyPointerType());
 
   if (!IExp->getType()->isIntegerType())
     return InvalidOperands(Loc, LHS, RHS);
@@ -6364,8 +6382,7 @@ QualType Sema::CheckAdditionOperands( // C99 6.5.6
   if (!checkArithmeticOpPointerOperand(*this, Loc, PExp))
     return QualType();
 
-  // Diagnose bad cases where we step over interface counts.
-  if (!checkArithmethicPointerOnNonFragileABI(*this, Loc, PExp))
+  if (isObjCPointer && checkArithmeticOnObjCPointer(*this, Loc, PExp))
     return QualType();
 
   // Check array bounds for pointer arithemtic
@@ -6414,7 +6431,8 @@ QualType Sema::CheckSubtractionOperands(ExprResult &LHS, ExprResult &RHS,
     QualType lpointee = LHS.get()->getType()->getPointeeType();
 
     // Diagnose bad cases where we step over interface counts.
-    if (!checkArithmethicPointerOnNonFragileABI(*this, Loc, LHS.get()))
+    if (LHS.get()->getType()->isObjCObjectPointerType() &&
+        checkArithmeticOnObjCPointer(*this, Loc, LHS.get()))
       return QualType();
 
     // The result type of a pointer-int computation is the pointer type.
@@ -7746,14 +7764,16 @@ static QualType CheckIncrementDecrementOperand(Sema &S, Expr *Op,
     S.Diag(OpLoc, diag::warn_increment_bool) << Op->getSourceRange();
   } else if (ResType->isRealType()) {
     // OK!
-  } else if (ResType->isAnyPointerType()) {
+  } else if (ResType->isPointerType()) {
     // C99 6.5.2.4p2, 6.5.6p2
     if (!checkArithmeticOpPointerOperand(S, OpLoc, Op))
       return QualType();
-
-    // Diagnose bad cases where we step over interface counts.
-    else if (!checkArithmethicPointerOnNonFragileABI(S, OpLoc, Op))
-      return QualType();
+  } else if (ResType->isObjCObjectPointerType()) {
+    // On modern runtimes, ObjC pointer arithmetic is forbidden.
+    // Otherwise, we just need a complete type.
+    if (checkArithmeticIncompletePointerType(S, OpLoc, Op) ||
+        checkArithmeticOnObjCPointer(S, OpLoc, Op))
+      return QualType();    
   } else if (ResType->isAnyComplexType()) {
     // C99 does not support ++/-- on complex types, we allow as an extension.
     S.Diag(OpLoc, diag::ext_integer_increment_complex)
diff --git a/lib/Sema/SemaExprObjC.cpp b/lib/Sema/SemaExprObjC.cpp
index d57329b69e..638a30f13a 100644
--- a/lib/Sema/SemaExprObjC.cpp
+++ b/lib/Sema/SemaExprObjC.cpp
@@ -575,27 +575,33 @@ ExprResult Sema::BuildObjCBoxedExpr(SourceRange SR, Expr *ValueExpr) {
   return MaybeBindToTemporary(BoxedExpr);
 }
 
+/// Build an ObjC subscript pseudo-object expression, given that
+/// that's supported by the runtime.
 ExprResult Sema::BuildObjCSubscriptExpression(SourceLocation RB, Expr *BaseExpr,
                                         Expr *IndexExpr,
                                         ObjCMethodDecl *getterMethod,
                                         ObjCMethodDecl *setterMethod) {
-  // Subscripting is only supported in the non-fragile ABI.
-  if (LangOpts.ObjCRuntime.isFragile())
-    return ExprError();
+  assert(!LangOpts.ObjCRuntime.isSubscriptPointerArithmetic());
 
-  // If the expression is type-dependent, there's nothing for us to do.
-  assert ((!BaseExpr->isTypeDependent() && !IndexExpr->isTypeDependent()) &&
-          "base or index cannot have dependent type here");
+  // We can't get dependent types here; our callers should have
+  // filtered them out.
+  assert((!BaseExpr->isTypeDependent() && !IndexExpr->isTypeDependent()) &&
+         "base or index cannot have dependent type here");
+
+  // Filter out placeholders in the index.  In theory, overloads could
+  // be preserved here, although that might not actually work correctly.
   ExprResult Result = CheckPlaceholderExpr(IndexExpr);
   if (Result.isInvalid())
     return ExprError();
   IndexExpr = Result.get();
   
-  // Perform lvalue-to-rvalue conversion.
+  // Perform lvalue-to-rvalue conversion on the base.
   Result = DefaultLvalueConversion(BaseExpr);
   if (Result.isInvalid())
     return ExprError();
   BaseExpr = Result.get();
+
+  // Build the pseudo-object expression.
   return Owned(ObjCSubscriptRefExpr::Create(Context, 
                                             BaseExpr,
                                             IndexExpr,