[C23] Implement N3018: The constexpr specifier for object definitions (#73099)

The implementation mostly reuses C++ code paths where possible,
including narrowing check in order to provide diagnostic messages in
case initializer for constexpr variable is not exactly representable in
target type.

The following won't work due to lack of support for other features:
- Diagnosing of underspecified declarations involving constexpr
- Constexpr attached to compound literals

Also due to lack of support for char8_t some of examples with utf-8
strings don't work properly.

Fixes #64742

Author: Fznamznon

clang/docs/ReleaseNotes.rst

@@ -143,6 +143,9 @@ C23 Feature Support
   macros typically exposed from ``<inttypes.h>``, such as ``PRIb8``.
   (`#81896: <https://github.com/llvm/llvm-project/issues/81896>`_).
 
+- Clang now supports `N3018 The constexpr specifier for object definitions`
+  <https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3018.htm>`_.
+
 Non-comprehensive list of changes in this release
 -------------------------------------------------
 

clang/include/clang/AST/Expr.h

@@ -1875,6 +1875,17 @@ class StringLiteral final
     llvm_unreachable("Unsupported character width!");
   }
 
+  // Get code unit but preserve sign info.
+  int64_t getCodeUnitS(size_t I, uint64_t BitWidth) const {
+    int64_t V = getCodeUnit(I);
+    if (isOrdinary() || isWide()) {
+      unsigned Width = getCharByteWidth() * BitWidth;
+      llvm::APInt AInt(Width, (uint64_t)V);
+      V = AInt.getSExtValue();
+    }
+    return V;
+  }
+
   unsigned getByteLength() const { return getCharByteWidth() * getLength(); }
   unsigned getLength() const { return *getTrailingObjects<unsigned>(); }
   unsigned getCharByteWidth() const { return StringLiteralBits.CharByteWidth; }

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -2946,6 +2946,18 @@ def warn_private_extern : Warning<
 def note_private_extern : Note<
   "use __attribute__((visibility(\"hidden\"))) attribute instead">;
 
+// C23 constexpr
+def err_c23_constexpr_not_variable : Error<
+  "'constexpr' can only be used in variable declarations">;
+def err_c23_constexpr_invalid_type : Error<
+  "constexpr variable cannot have type %0">;
+def err_c23_constexpr_init_not_representable : Error<
+  "constexpr initializer evaluates to %0 which is not exactly representable in type %1">;
+def err_c23_constexpr_init_type_mismatch : Error<
+  "constexpr initializer for type %0 is of type %1">;
+def err_c23_constexpr_pointer_not_null : Error<
+  "constexpr pointer initializer is not null">;
+
 // C++ Concepts
 def err_concept_decls_may_only_appear_in_global_namespace_scope : Error<
   "concept declarations may only appear in global or namespace scope">;

clang/include/clang/Basic/TokenKinds.def

@@ -393,7 +393,7 @@ CXX11_KEYWORD(alignas               , KEYC23)
 CXX11_UNARY_EXPR_OR_TYPE_TRAIT(alignof, AlignOf, KEYC23)
 CXX11_KEYWORD(char16_t              , KEYNOMS18)
 CXX11_KEYWORD(char32_t              , KEYNOMS18)
-CXX11_KEYWORD(constexpr             , 0)
+CXX11_KEYWORD(constexpr             , KEYC23)
 CXX11_KEYWORD(decltype              , 0)
 CXX11_KEYWORD(noexcept              , 0)
 CXX11_KEYWORD(nullptr               , KEYC23)

clang/lib/AST/Decl.cpp

@@ -2465,7 +2465,7 @@ bool VarDecl::mightBeUsableInConstantExpressions(const ASTContext &C) const {
 
   // OpenCL permits const integral variables to be used in constant
   // expressions, like in C++98.
-  if (!Lang.CPlusPlus && !Lang.OpenCL)
+  if (!Lang.CPlusPlus && !Lang.OpenCL && !Lang.C23)
     return false;
 
   // Function parameters are never usable in constant expressions.
@@ -2487,14 +2487,19 @@ bool VarDecl::mightBeUsableInConstantExpressions(const ASTContext &C) const {
   if (!getType().isConstant(C) || getType().isVolatileQualified())
     return false;
 
-  // In C++, const, non-volatile variables of integral or enumeration types
-  // can be used in constant expressions.
-  if (getType()->isIntegralOrEnumerationType())
+  // In C++, but not in C, const, non-volatile variables of integral or
+  // enumeration types can be used in constant expressions.
+  if (getType()->isIntegralOrEnumerationType() && !Lang.C23)
     return true;
 
+  // C23 6.6p7: An identifier that is:
+  // ...
+  // - declared with storage-class specifier constexpr and has an object type,
+  // is a named constant, ... such a named constant is a constant expression
+  // with the type and value of the declared object.
   // Additionally, in C++11, non-volatile constexpr variables can be used in
   // constant expressions.
-  return Lang.CPlusPlus11 && isConstexpr();
+  return (Lang.CPlusPlus11 || Lang.C23) && isConstexpr();
 }
 
 bool VarDecl::isUsableInConstantExpressions(const ASTContext &Context) const {
@@ -2572,11 +2577,11 @@ APValue *VarDecl::evaluateValueImpl(SmallVectorImpl<PartialDiagnosticAt> &Notes,
   bool Result = Init->EvaluateAsInitializer(Eval->Evaluated, Ctx, this, Notes,
                                             IsConstantInitialization);
 
-  // In C++, this isn't a constant initializer if we produced notes. In that
+  // In C++/C23, this isn't a constant initializer if we produced notes. In that
   // case, we can't keep the result, because it may only be correct under the
   // assumption that the initializer is a constant context.
-  if (IsConstantInitialization && Ctx.getLangOpts().CPlusPlus &&
-      !Notes.empty())
+  if (IsConstantInitialization &&
+      (Ctx.getLangOpts().CPlusPlus || Ctx.getLangOpts().C23) && !Notes.empty())
     Result = false;
 
   // Ensure the computed APValue is cleaned up later if evaluation succeeded,
@@ -2634,7 +2639,9 @@ bool VarDecl::checkForConstantInitialization(
   // std::is_constant_evaluated()).
   assert(!Eval->WasEvaluated &&
          "already evaluated var value before checking for constant init");
-  assert(getASTContext().getLangOpts().CPlusPlus && "only meaningful in C++");
+  assert((getASTContext().getLangOpts().CPlusPlus ||
+          getASTContext().getLangOpts().C23) &&
+         "only meaningful in C++/C23");
 
   assert(!getInit()->isValueDependent());
 

clang/lib/AST/ExprConstant.cpp

@@ -4133,6 +4133,10 @@ static CompleteObject findCompleteObject(EvalInfo &Info, const Expr *E,
     }
 
     bool IsConstant = BaseType.isConstant(Info.Ctx);
+    bool ConstexprVar = false;
+    if (const auto *VD = dyn_cast_if_present<VarDecl>(
+            Info.EvaluatingDecl.dyn_cast<const ValueDecl *>()))
+      ConstexprVar = VD->isConstexpr();
 
     // Unless we're looking at a local variable or argument in a constexpr call,
     // the variable we're reading must be const.
@@ -4152,6 +4156,9 @@ static CompleteObject findCompleteObject(EvalInfo &Info, const Expr *E,
         return CompleteObject();
       } else if (VD->isConstexpr()) {
         // OK, we can read this variable.
+      } else if (Info.getLangOpts().C23 && ConstexprVar) {
+        Info.FFDiag(E);
+        return CompleteObject();
       } else if (BaseType->isIntegralOrEnumerationType()) {
         if (!IsConstant) {
           if (!IsAccess)
@@ -15826,7 +15833,8 @@ bool Expr::EvaluateAsInitializer(APValue &Value, const ASTContext &Ctx,
   EStatus.Diag = &Notes;
 
   EvalInfo Info(Ctx, EStatus,
-                (IsConstantInitialization && Ctx.getLangOpts().CPlusPlus)
+                (IsConstantInitialization &&
+                 (Ctx.getLangOpts().CPlusPlus || Ctx.getLangOpts().C23))
                     ? EvalInfo::EM_ConstantExpression
                     : EvalInfo::EM_ConstantFold);
   Info.setEvaluatingDecl(VD, Value);

clang/lib/Parse/ParseDecl.cpp

@@ -4265,6 +4265,8 @@ void Parser::ParseDeclarationSpecifiers(
 
     // constexpr, consteval, constinit specifiers
     case tok::kw_constexpr:
+      if (getLangOpts().C23)
+        Diag(Tok, diag::warn_c23_compat_keyword) << Tok.getName();
       isInvalid = DS.SetConstexprSpec(ConstexprSpecKind::Constexpr, Loc,
                                       PrevSpec, DiagID);
       break;

clang/lib/Sema/DeclSpec.cpp

@@ -1377,6 +1377,20 @@ void DeclSpec::Finish(Sema &S, const PrintingPolicy &Policy) {
       ThreadStorageClassSpec = TSCS_unspecified;
       ThreadStorageClassSpecLoc = SourceLocation();
     }
+    if (S.getLangOpts().C23 &&
+        getConstexprSpecifier() == ConstexprSpecKind::Constexpr) {
+      S.Diag(ConstexprLoc, diag::err_invalid_decl_spec_combination)
+          << DeclSpec::getSpecifierName(getThreadStorageClassSpec())
+          << SourceRange(getThreadStorageClassSpecLoc());
+    }
+  }
+
+  if (S.getLangOpts().C23 &&
+      getConstexprSpecifier() == ConstexprSpecKind::Constexpr &&
+      StorageClassSpec == SCS_extern) {
+    S.Diag(ConstexprLoc, diag::err_invalid_decl_spec_combination)
+        << DeclSpec::getSpecifierName(getStorageClassSpec())
+        << SourceRange(getStorageClassSpecLoc());
   }
 
   // If no type specifier was provided and we're parsing a language where

clang/lib/Sema/SemaDecl.cpp

@@ -5147,6 +5147,8 @@ Decl *Sema::ParsedFreeStandingDeclSpec(Scope *S, AccessSpecifier AS,
       Diag(DS.getConstexprSpecLoc(), diag::err_constexpr_tag)
           << GetDiagnosticTypeSpecifierID(DS)
           << static_cast<int>(DS.getConstexprSpecifier());
+    else if (getLangOpts().C23)
+      Diag(DS.getConstexprSpecLoc(), diag::err_c23_constexpr_not_variable);
     else
       Diag(DS.getConstexprSpecLoc(), diag::err_constexpr_wrong_decl_kind)
           << static_cast<int>(DS.getConstexprSpecifier());
@@ -8649,6 +8651,38 @@ static bool checkForConflictWithNonVisibleExternC(Sema &S, const T *ND,
   return false;
 }
 
+static bool CheckC23ConstexprVarType(Sema &SemaRef, SourceLocation VarLoc,
+                                     QualType T) {
+  QualType CanonT = SemaRef.Context.getCanonicalType(T);
+  // C23 6.7.1p5: An object declared with storage-class specifier constexpr or
+  // any of its members, even recursively, shall not have an atomic type, or a
+  // variably modified type, or a type that is volatile or restrict qualified.
+  if (CanonT->isVariablyModifiedType()) {
+    SemaRef.Diag(VarLoc, diag::err_c23_constexpr_invalid_type) << T;
+    return true;
+  }
+
+  // Arrays are qualified by their element type, so get the base type (this
+  // works on non-arrays as well).
+  CanonT = SemaRef.Context.getBaseElementType(CanonT);
+
+  if (CanonT->isAtomicType() || CanonT.isVolatileQualified() ||
+      CanonT.isRestrictQualified()) {
+    SemaRef.Diag(VarLoc, diag::err_c23_constexpr_invalid_type) << T;
+    return true;
+  }
+
+  if (CanonT->isRecordType()) {
+    const RecordDecl *RD = CanonT->getAsRecordDecl();
+    if (llvm::any_of(RD->fields(), [&SemaRef, VarLoc](const FieldDecl *F) {
+          return CheckC23ConstexprVarType(SemaRef, VarLoc, F->getType());
+        }))
+      return true;
+  }
+
+  return false;
+}
+
 void Sema::CheckVariableDeclarationType(VarDecl *NewVD) {
   // If the decl is already known invalid, don't check it.
   if (NewVD->isInvalidDecl())
@@ -8899,6 +8933,12 @@ void Sema::CheckVariableDeclarationType(VarDecl *NewVD) {
     return;
   }
 
+  if (getLangOpts().C23 && NewVD->isConstexpr() &&
+      CheckC23ConstexprVarType(*this, NewVD->getLocation(), T)) {
+    NewVD->setInvalidDecl();
+    return;
+  }
+
   if (NewVD->isConstexpr() && !T->isDependentType() &&
       RequireLiteralType(NewVD->getLocation(), T,
                          diag::err_constexpr_var_non_literal)) {
@@ -9281,6 +9321,22 @@ static FunctionDecl *CreateNewFunctionDecl(Sema &SemaRef, Declarator &D,
   FunctionDecl *NewFD = nullptr;
   bool isInline = D.getDeclSpec().isInlineSpecified();
 
+  ConstexprSpecKind ConstexprKind = D.getDeclSpec().getConstexprSpecifier();
+  if (ConstexprKind == ConstexprSpecKind::Constinit ||
+      (SemaRef.getLangOpts().C23 &&
+       ConstexprKind == ConstexprSpecKind::Constexpr)) {
+
+    if (SemaRef.getLangOpts().C23)
+      SemaRef.Diag(D.getDeclSpec().getConstexprSpecLoc(),
+                   diag::err_c23_constexpr_not_variable);
+    else
+      SemaRef.Diag(D.getDeclSpec().getConstexprSpecLoc(),
+                   diag::err_constexpr_wrong_decl_kind)
+          << static_cast<int>(ConstexprKind);
+    ConstexprKind = ConstexprSpecKind::Unspecified;
+    D.getMutableDeclSpec().ClearConstexprSpec();
+  }
+
   if (!SemaRef.getLangOpts().CPlusPlus) {
     // Determine whether the function was written with a prototype. This is
     // true when:
@@ -9314,15 +9370,6 @@ static FunctionDecl *CreateNewFunctionDecl(Sema &SemaRef, Declarator &D,
   }
 
   ExplicitSpecifier ExplicitSpecifier = D.getDeclSpec().getExplicitSpecifier();
-
-  ConstexprSpecKind ConstexprKind = D.getDeclSpec().getConstexprSpecifier();
-  if (ConstexprKind == ConstexprSpecKind::Constinit) {
-    SemaRef.Diag(D.getDeclSpec().getConstexprSpecLoc(),
-                 diag::err_constexpr_wrong_decl_kind)
-        << static_cast<int>(ConstexprKind);
-    ConstexprKind = ConstexprSpecKind::Unspecified;
-    D.getMutableDeclSpec().ClearConstexprSpec();
-  }
   Expr *TrailingRequiresClause = D.getTrailingRequiresClause();
 
   SemaRef.CheckExplicitObjectMemberFunction(DC, D, Name, R);
@@ -13909,7 +13956,9 @@ void Sema::AddInitializerToDecl(Decl *RealDecl, Expr *Init, bool DirectInit) {
       VDecl->setStorageClass(SC_Extern);
 
     // C99 6.7.8p4. All file scoped initializers need to be constant.
-    if (!getLangOpts().CPlusPlus && !VDecl->isInvalidDecl())
+    // Avoid duplicate diagnostics for constexpr variables.
+    if (!getLangOpts().CPlusPlus && !VDecl->isInvalidDecl() &&
+        !VDecl->isConstexpr())
       CheckForConstantInitializer(Init, DclT);
   }
 
@@ -14520,17 +14569,21 @@ void Sema::CheckCompleteVariableDeclaration(VarDecl *var) {
   QualType baseType = Context.getBaseElementType(type);
   bool HasConstInit = true;
 
+  if (getLangOpts().C23 && var->isConstexpr() && !Init)
+    Diag(var->getLocation(), diag::err_constexpr_var_requires_const_init)
+        << var;
+
   // Check whether the initializer is sufficiently constant.
-  if (getLangOpts().CPlusPlus && !type->isDependentType() && Init &&
-      !Init->isValueDependent() &&
+  if ((getLangOpts().CPlusPlus || (getLangOpts().C23 && var->isConstexpr())) &&
+      !type->isDependentType() && Init && !Init->isValueDependent() &&
       (GlobalStorage || var->isConstexpr() ||
        var->mightBeUsableInConstantExpressions(Context))) {
     // If this variable might have a constant initializer or might be usable in
     // constant expressions, check whether or not it actually is now.  We can't
     // do this lazily, because the result might depend on things that change
     // later, such as which constexpr functions happen to be defined.
     SmallVector<PartialDiagnosticAt, 8> Notes;
-    if (!getLangOpts().CPlusPlus11) {
+    if (!getLangOpts().CPlusPlus11 && !getLangOpts().C23) {
       // Prior to C++11, in contexts where a constant initializer is required,
       // the set of valid constant initializers is described by syntactic rules
       // in [expr.const]p2-6.