[ObjC] Emit a boxed expression as a compile-time constant if the

expression inside the parentheses is a valid UTF-8 string literal.

Previously clang emitted an expression like @("abc") as a message send
to stringWithUTF8String. This commit makes clang emit the boxed
expression as a compile-time constant instead.

This commit also has the effect of silencing the nullable-to-nonnull
conversion warning clang started emitting after r317727, which
originally motivated this commit (see https://oleb.net/2018/@keypath).

rdar://problem/42684601

Differential Revision: https://reviews.llvm.org/D58729

llvm-svn: 355662

clang/include/clang/AST/ExprObjC.h

@@ -138,6 +138,12 @@ public:
     return BoxingMethod;
   }
 
+  // Indicates whether this boxed expression can be emitted as a compile-time
+  // constant.
+  bool isExpressibleAsConstantInitializer() const {
+    return !BoxingMethod && SubExpr;
+  }
+
   SourceLocation getAtLoc() const { return Range.getBegin(); }
 
   SourceLocation getBeginLoc() const LLVM_READONLY { return Range.getBegin(); }

clang/include/clang/Basic/DiagnosticGroups.td

@@ -404,6 +404,7 @@ def ObjCPointerIntrospectPerformSelector : DiagGroup<"deprecated-objc-pointer-in
 def ObjCPointerIntrospect : DiagGroup<"deprecated-objc-pointer-introspection", [ObjCPointerIntrospectPerformSelector]>;
 def ObjCMultipleMethodNames : DiagGroup<"objc-multiple-method-names">;
 def ObjCFlexibleArray : DiagGroup<"objc-flexible-array">;
+def ObjCBoxing : DiagGroup<"objc-boxing">;
 def OpenCLUnsupportedRGBA: DiagGroup<"opencl-unsupported-rgba">;
 def DeprecatedObjCIsaUsage : DiagGroup<"deprecated-objc-isa-usage">;
 def ExplicitInitializeCall : DiagGroup<"explicit-initialize-call">;

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -918,6 +918,9 @@ def err_inconsistent_ivar_count : Error<
   "inconsistent number of instance variables specified">;
 def warn_undef_method_impl : Warning<"method definition for %0 not found">,
   InGroup<DiagGroup<"incomplete-implementation">>;
+def warn_objc_boxing_invalid_utf8_string : Warning<
+  "string is ill-formed as UTF-8 and will become a null %0 when boxed">,
+  InGroup<ObjCBoxing>;
 
 def warn_conflicting_overriding_ret_types : Warning<
   "conflicting return type in "

clang/lib/AST/ExprConstant.cpp

@@ -1743,6 +1743,8 @@ static bool IsGlobalLValue(APValue::LValueBase B) {
   case Expr::CXXTypeidExprClass:
   case Expr::CXXUuidofExprClass:
     return true;
+  case Expr::ObjCBoxedExprClass:
+    return cast<ObjCBoxedExpr>(E)->isExpressibleAsConstantInitializer();
   case Expr::CallExprClass:
     return IsStringLiteralCall(cast<CallExpr>(E));
   // For GCC compatibility, &&label has static storage duration.
@@ -5794,6 +5796,8 @@ public:
   bool VisitObjCStringLiteral(const ObjCStringLiteral *E)
       { return Success(E); }
   bool VisitObjCBoxedExpr(const ObjCBoxedExpr *E) {
+    if (E->isExpressibleAsConstantInitializer())
+      return Success(E);
     if (Info.noteFailure())
       EvaluateIgnoredValue(Info, E->getSubExpr());
     return Error(E);

clang/lib/CodeGen/CGExprConstant.cpp

@@ -1613,6 +1613,7 @@ private:
   ConstantLValue VisitConstantExpr(const ConstantExpr *E);
   ConstantLValue VisitCompoundLiteralExpr(const CompoundLiteralExpr *E);
   ConstantLValue VisitStringLiteral(const StringLiteral *E);
+  ConstantLValue VisitObjCBoxedExpr(const ObjCBoxedExpr *E);
   ConstantLValue VisitObjCEncodeExpr(const ObjCEncodeExpr *E);
   ConstantLValue VisitObjCStringLiteral(const ObjCStringLiteral *E);
   ConstantLValue VisitPredefinedExpr(const PredefinedExpr *E);
@@ -1775,10 +1776,24 @@ ConstantLValueEmitter::VisitObjCEncodeExpr(const ObjCEncodeExpr *E) {
   return CGM.GetAddrOfConstantStringFromObjCEncode(E);
 }
 
+static ConstantLValue emitConstantObjCStringLiteral(const StringLiteral *S,
+                                                    QualType T,
+                                                    CodeGenModule &CGM) {
+  auto C = CGM.getObjCRuntime().GenerateConstantString(S);
+  return C.getElementBitCast(CGM.getTypes().ConvertTypeForMem(T));
+}
+
 ConstantLValue
 ConstantLValueEmitter::VisitObjCStringLiteral(const ObjCStringLiteral *E) {
-  auto C = CGM.getObjCRuntime().GenerateConstantString(E->getString());
-  return C.getElementBitCast(CGM.getTypes().ConvertTypeForMem(E->getType()));
+  return emitConstantObjCStringLiteral(E->getString(), E->getType(), CGM);
+}
+
+ConstantLValue
+ConstantLValueEmitter::VisitObjCBoxedExpr(const ObjCBoxedExpr *E) {
+  assert(E->isExpressibleAsConstantInitializer() &&
+         "this boxed expression can't be emitted as a compile-time constant");
+  auto *SL = cast<StringLiteral>(E->getSubExpr()->IgnoreParenCasts());
+  return emitConstantObjCStringLiteral(SL, E->getType(), CGM);
 }
 
 ConstantLValue

clang/lib/CodeGen/CGObjC.cpp

@@ -14,6 +14,7 @@
 #include "CGObjCRuntime.h"
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
+#include "ConstantEmitter.h"
 #include "TargetInfo.h"
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/DeclObjC.h"
@@ -60,7 +61,12 @@ CodeGenFunction::EmitObjCBoxedExpr(const ObjCBoxedExpr *E) {
   // Get the method.
   const ObjCMethodDecl *BoxingMethod = E->getBoxingMethod();
   const Expr *SubExpr = E->getSubExpr();
-  assert(BoxingMethod && "BoxingMethod is null");
+
+  if (E->isExpressibleAsConstantInitializer()) {
+    ConstantEmitter ConstEmitter(CGM);
+    return ConstEmitter.tryEmitAbstract(E, E->getType());
+  }
+
   assert(BoxingMethod->isClassMethod() && "BoxingMethod must be a class method");
   Selector Sel = BoxingMethod->getSelector();
 

clang/lib/Sema/SemaExprObjC.cpp

@@ -25,6 +25,7 @@
 #include "clang/Sema/Scope.h"
 #include "clang/Sema/ScopeInfo.h"
 #include "llvm/ADT/SmallString.h"
+#include "llvm/Support/ConvertUTF.h"
 
 using namespace clang;
 using namespace sema;
@@ -524,6 +525,30 @@ ExprResult Sema::BuildObjCBoxedExpr(SourceRange SR, Expr *ValueExpr) {
         NSStringPointer = Context.getObjCObjectPointerType(NSStringObject);
       }
 
+      // The boxed expression can be emitted as a compile time constant if it is
+      // a string literal whose character encoding is compatible with UTF-8.
+      if (auto *CE = dyn_cast<ImplicitCastExpr>(ValueExpr))
+        if (CE->getCastKind() == CK_ArrayToPointerDecay)
+          if (auto *SL =
+                  dyn_cast<StringLiteral>(CE->getSubExpr()->IgnoreParens())) {
+            assert((SL->isAscii() || SL->isUTF8()) &&
+                   "unexpected character encoding");
+            StringRef Str = SL->getString();
+            const llvm::UTF8 *StrBegin = Str.bytes_begin();
+            const llvm::UTF8 *StrEnd = Str.bytes_end();
+            // Check that this is a valid UTF-8 string.
+            if (llvm::isLegalUTF8String(&StrBegin, StrEnd)) {
+              BoxedType = Context.getAttributedType(
+                  AttributedType::getNullabilityAttrKind(
+                      NullabilityKind::NonNull),
+                  NSStringPointer, NSStringPointer);
+              return new (Context) ObjCBoxedExpr(CE, BoxedType, nullptr, SR);
+            }
+
+            Diag(SL->getBeginLoc(), diag::warn_objc_boxing_invalid_utf8_string)
+                << NSStringPointer << SL->getSourceRange();
+          }
+
       if (!StringWithUTF8StringMethod) {
         IdentifierInfo *II = &Context.Idents.get("stringWithUTF8String");
         Selector stringWithUTF8String = Context.Selectors.getUnarySelector(II);

clang/test/CodeGenObjC/boxing.m

@@ -53,6 +53,9 @@ typedef signed char BOOL;
 + (id)stringWithUTF8String:(const char *)nullTerminatedCString;
 @end
 
+// CHECK: [[V0:%.*]] = type opaque
+// CHECK: [[STRUCT_NSCONSTANT_STRING_TAG:%.*]] = type { i32*, i32, i8*, i64 }
+
 // CHECK: [[WithIntMeth:@.*]] = private unnamed_addr constant [15 x i8] c"numberWithInt:\00"
 // CHECK: [[WithIntSEL:@.*]] = private externally_initialized global i8* getelementptr inbounds ([15 x i8], [15 x i8]* [[WithIntMeth]]
 // CHECK: [[WithCharMeth:@.*]] = private unnamed_addr constant [16 x i8] c"numberWithChar:\00"
@@ -65,8 +68,12 @@ typedef signed char BOOL;
 // CHECK: [[WithUnsignedIntegerSEL:@.*]] = private externally_initialized global i8* getelementptr inbounds ([27 x i8], [27 x i8]* [[WithUnsignedIntegerMeth]]
 // CHECK: [[stringWithUTF8StringMeth:@.*]] = private unnamed_addr constant [22 x i8] c"stringWithUTF8String:\00"
 // CHECK: [[stringWithUTF8StringSEL:@.*]] = private externally_initialized global i8* getelementptr inbounds ([22 x i8], [22 x i8]* [[stringWithUTF8StringMeth]]
+// CHECK: [[STR0:.*]] = private unnamed_addr constant [4 x i8] c"abc\00", section "__TEXT,__cstring,cstring_literals", align 1
+// CHECK: [[UNNAMED_CFSTRING:.*]] = private global [[STRUCT_NSCONSTANT_STRING_TAG]] { i32* getelementptr inbounds ([0 x i32], [0 x i32]* @__CFConstantStringClassReference, i32 0, i32 0), i32 1992, i8* getelementptr inbounds ([4 x i8], [4 x i8]* [[STR0]], i32 0, i32 0), i64 3 }, section "__DATA,__cfstring", align 8
 
 int main() {
+  // CHECK: [[T:%.*]] = alloca [[V0]]*, align 8
+
   // CHECK: load i8*, i8** [[WithIntSEL]]
   int i; @(i);
   // CHECK: load i8*, i8** [[WithCharSEL]]
@@ -92,4 +99,7 @@ int main() {
   Color col = Red;
   // CHECK: load i8*, i8** [[WithIntegerSEL]]
   @(col);
+
+  // CHECK: store [[V0]]* bitcast ([[STRUCT_NSCONSTANT_STRING_TAG]]* [[UNNAMED_CFSTRING]] to [[V0]]*), [[V0]]** [[T]], align 8
+  NSString *t = @("abc");
 }

clang/test/SemaObjC/boxing-illegal.m

@@ -1,4 +1,4 @@
-// RUN: %clang_cc1 -fsyntax-only -verify -Wattributes %s
+// RUN: %clang_cc1 -fsyntax-only -verify -Wattributes -fpascal-strings %s
 
 typedef long NSInteger;
 typedef unsigned long NSUInteger;
@@ -57,6 +57,19 @@ void testEnum(void *p) {
   box = @(*(enum ForwE*)p); // expected-error {{incomplete type 'enum ForwE' used in a boxed expression}}
 }
 
+@interface NSString
+@end
+
+void testStringLiteral() {
+  NSString *s;
+  s = @("abc");
+  s = @(u8"abc");
+  s = @(u"abc"); // expected-error {{illegal type 'unsigned short *' used in a boxed expression}}
+  s = @(U"abc"); // expected-error {{illegal type 'unsigned int *' used in a boxed expression}}
+  s = @(L"abc"); // expected-error {{illegal type 'int *' used in a boxed expression}}
+  s = @("\pabc"); // expected-error {{illegal type 'unsigned char *' used in a boxed expression}}
+}
+
 // rdar://13333205
 @class NSMutableDictionary;
 

clang/test/SemaObjC/objc-literal-sig.m

@@ -39,6 +39,8 @@ typedef _Bool BOOL;
 
 // All tests are doubled to make sure that a bad method is not saved
 // and then used un-checked.
+const char *getStr(void);
+
 void test_sig() {
   (void)@__objc_yes; // expected-error{{literal construction method 'numberWithBool:' has incompatible signature}}
   (void)@__objc_yes; // expected-error{{literal construction method 'numberWithBool:' has incompatible signature}}
@@ -46,6 +48,6 @@ void test_sig() {
   id array2 = @[ @17 ]; // expected-error{{literal construction method 'arrayWithObjects:count:' has incompatible signature}}
   id dict = @{ @"hello" : @17 }; // expected-error{{literal construction method 'dictionaryWithObjects:forKeys:count:' has incompatible signature}}
   id dict2 = @{ @"hello" : @17 }; // expected-error{{literal construction method 'dictionaryWithObjects:forKeys:count:' has incompatible signature}}
-  id str = @("hello"); // expected-error{{literal construction method 'stringWithUTF8String:' has incompatible signature}}
-  id str2 = @("hello"); // expected-error{{literal construction method 'stringWithUTF8String:' has incompatible signature}}
+  id str = @(getStr()); // expected-error{{literal construction method 'stringWithUTF8String:' has incompatible signature}}
+  id str2 = @(getStr()); // expected-error{{literal construction method 'stringWithUTF8String:' has incompatible signature}}
 }

clang/test/SemaObjC/transfer-boxed-string-nullability.m

@@ -16,13 +16,23 @@
 void takesNonNull(NSString * _Nonnull ptr);
 
 void testBoxedString() {
+  // No diagnostic emitted as this doesn't need a stringWithUTF8String message
+  // send.
+  takesNonNull(@("hey"));
+  takesNonNull(@(u8"hey"));
+
+  // If the string isn't a valid UTF-8 string, a diagnostic is emitted since the
+  // boxed expression turns into a message send.
+  takesNonNull(@(u8"\xFF")); // expected-warning {{string is ill-formed as UTF-8}}
+  takesNonNull(@(u8"\xC0\x80")); // expected-warning {{string is ill-formed as UTF-8}}
+
   const char *str = "hey";
   takesNonNull([NSString stringWithUTF8String:str]);
   takesNonNull(@(str));
 #ifndef NOWARN
-  // expected-warning@-3 {{implicit conversion from nullable pointer 'NSString * _Nullable' to non-nullable pointer type 'NSString * _Nonnull'}}
-  // expected-warning@-3 {{implicit conversion from nullable pointer 'NSString * _Nullable' to non-nullable pointer type 'NSString * _Nonnull'}}
-#else
-  // expected-no-diagnostics
+  // expected-warning@-7 {{implicit conversion from nullable pointer 'NSString * _Nullable' to non-nullable pointer type 'NSString * _Nonnull'}}
+  // expected-warning@-7 {{implicit conversion from nullable pointer 'NSString * _Nullable' to non-nullable pointer type 'NSString * _Nonnull'}}
+  // expected-warning@-5 {{implicit conversion from nullable pointer 'NSString * _Nullable' to non-nullable pointer type 'NSString * _Nonnull'}}
+  // expected-warning@-5 {{implicit conversion from nullable pointer 'NSString * _Nullable' to non-nullable pointer type 'NSString * _Nonnull'}}
 #endif
 }

clang/test/SemaObjCXX/literals.mm

@@ -50,6 +50,9 @@ void x() {
 + (id)dictionaryWithObjects:(const id [])objects forKeys:(const id<NSCopying> [])keys count:(unsigned long)cnt;
 @end
 
+@interface NSString
+@end
+
 template<typename T>
 struct ConvertibleTo {
   operator T();
@@ -185,3 +188,8 @@ id value;
 void test_dictionary_colon() {
   id dict = @{ key : value };
 }
+
+void testConstExpr() {
+  constexpr NSString *t0 = @"abc";
+  constexpr NSString *t1 = @("abc");
+}