Add Clang attribute to ensure that fields are initialized explicitly (#102040)

This is a new Clang-specific attribute to ensure that field
initializations are performed explicitly.

For example, if we have
```
struct B {
  [[clang::explicit]] int f1;
};
```
then the diagnostic would trigger if we do `B b{};`:
```
field 'f1' is left uninitialized, but was marked as requiring initialization
```

This prevents callers from accidentally forgetting to initialize fields,
particularly when new fields are added to the class.

Author: higher-performance

clang/include/clang/AST/Decl.h

@@ -4285,6 +4285,14 @@ class RecordDecl : public TagDecl {
     RecordDeclBits.HasNonTrivialToPrimitiveCopyCUnion = V;
   }
 
+  bool hasUninitializedExplicitInitFields() const {
+    return RecordDeclBits.HasUninitializedExplicitInitFields;
+  }
+
+  void setHasUninitializedExplicitInitFields(bool V) {
+    RecordDeclBits.HasUninitializedExplicitInitFields = V;
+  }
+
   /// Determine whether this class can be passed in registers. In C++ mode,
   /// it must have at least one trivial, non-deleted copy or move constructor.
   /// FIXME: This should be set as part of completeDefinition.

clang/include/clang/AST/DeclBase.h

@@ -1446,6 +1446,9 @@ class DeclContext {
   /// hasLazyLocalLexicalLookups, hasLazyExternalLexicalLookups
   friend class ASTWriter;
 
+protected:
+  enum { NumOdrHashBits = 25 };
+
   // We use uint64_t in the bit-fields below since some bit-fields
   // cross the unsigned boundary and this breaks the packing.
 
@@ -1667,6 +1670,14 @@ class DeclContext {
     LLVM_PREFERRED_TYPE(bool)
     uint64_t HasNonTrivialToPrimitiveCopyCUnion : 1;
 
+    /// True if any field is marked as requiring explicit initialization with
+    /// [[clang::requires_explicit_initialization]].
+    /// In C++, this is also set for types without a user-provided default
+    /// constructor, and is propagated from any base classes and/or member
+    /// variables whose types are aggregates.
+    LLVM_PREFERRED_TYPE(bool)
+    uint64_t HasUninitializedExplicitInitFields : 1;
+
     /// Indicates whether this struct is destroyed in the callee.
     LLVM_PREFERRED_TYPE(bool)
     uint64_t ParamDestroyedInCallee : 1;
@@ -1681,7 +1692,7 @@ class DeclContext {
 
     /// True if a valid hash is stored in ODRHash. This should shave off some
     /// extra storage and prevent CXXRecordDecl to store unused bits.
-    uint64_t ODRHash : 26;
+    uint64_t ODRHash : NumOdrHashBits;
   };
 
   /// Number of inherited and non-inherited bits in RecordDeclBitfields.

clang/include/clang/Basic/Attr.td

@@ -1902,6 +1902,13 @@ def Leaf : InheritableAttr {
   let SimpleHandler = 1;
 }
 
+def ExplicitInit : InheritableAttr {
+  let Spellings = [Clang<"requires_explicit_initialization">];
+  let Subjects = SubjectList<[Field], ErrorDiag>;
+  let Documentation = [ExplicitInitDocs];
+  let SimpleHandler = 1;
+}
+
 def LifetimeBound : DeclOrTypeAttr {
   let Spellings = [Clang<"lifetimebound", 0>];
   let Subjects = SubjectList<[ParmVar, ImplicitObjectParameter], ErrorDiag>;

clang/include/clang/Basic/AttrDocs.td

@@ -1684,6 +1684,55 @@ is not specified.
   }];
 }
 
+def ExplicitInitDocs : Documentation {
+  let Category = DocCatField;
+  let Content = [{
+The ``clang::requires_explicit_initialization`` attribute indicates that a
+field of an aggregate must be initialized explicitly by the user when an object
+of the aggregate type is constructed. The attribute supports both C and C++,
+but its usage is invalid on non-aggregates.
+
+Note that this attribute is *not* a memory safety feature, and is *not* intended
+to guard against use of uninitialized memory.
+
+Rather, it is intended for use in "parameter-objects", used to simulate,
+for example, the passing of named parameters.
+The attribute generates a warning when explicit initializers for such
+variables are not provided (this occurs regardless of whether any in-class field
+initializers exist):
+
+.. code-block:: c++
+
+  struct Buffer {
+    void *address [[clang::requires_explicit_initialization]];
+    size_t length [[clang::requires_explicit_initialization]] = 0;
+  };
+
+  struct ArrayIOParams {
+    size_t count [[clang::requires_explicit_initialization]];
+    size_t element_size [[clang::requires_explicit_initialization]];
+    int flags = 0;
+  };
+
+  size_t ReadArray(FILE *file, struct Buffer buffer,
+                   struct ArrayIOParams params);
+
+  int main() {
+    unsigned int buf[512];
+    ReadArray(stdin, {
+      buf
+      // warning: field 'length' is not explicitly initialized
+    }, {
+      .count = sizeof(buf) / sizeof(*buf),
+      // warning: field 'element_size' is not explicitly initialized
+      // (Note that a missing initializer for 'flags' is not diagnosed, because
+      // the field is not marked as requiring explicit initialization.)
+    });
+  }
+
+  }];
+}
+
 def NoUniqueAddressDocs : Documentation {
   let Category = DocCatField;
   let Content = [{

clang/include/clang/Basic/DiagnosticASTKinds.td

@@ -396,6 +396,16 @@ def note_constexpr_assumption_failed : Note<
 def err_experimental_clang_interp_failed : Error<
   "the experimental clang interpreter failed to evaluate an expression">;
 
+def warn_attribute_needs_aggregate : Warning<
+  "%0 attribute is ignored in non-aggregate type %1">,
+  InGroup<IgnoredAttributes>;
+
+def warn_cxx20_compat_requires_explicit_init_non_aggregate : Warning<
+  "explicit initialization of field %1 will not be enforced in C++20 and later "
+  "because %2 has a user-declared constructor, making the type no longer an "
+  "aggregate">,
+  DefaultIgnore, InGroup<CXX20Compat>;
+
 def warn_integer_constant_overflow : Warning<
   "overflow in expression; result is %0 with type %1">,
   InGroup<DiagGroup<"integer-overflow">>;

clang/include/clang/Basic/DiagnosticGroups.td

@@ -796,6 +796,7 @@ def Trigraphs      : DiagGroup<"trigraphs">;
 def UndefinedReinterpretCast : DiagGroup<"undefined-reinterpret-cast">;
 def ReinterpretBaseClass : DiagGroup<"reinterpret-base-class">;
 def Unicode  : DiagGroup<"unicode">;
+def UninitializedExplicitInit : DiagGroup<"uninitialized-explicit-init">;
 def UninitializedMaybe : DiagGroup<"conditional-uninitialized">;
 def UninitializedSometimes : DiagGroup<"sometimes-uninitialized">;
 def UninitializedStaticSelfInit : DiagGroup<"static-self-init">;

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -2347,6 +2347,9 @@ def err_init_reference_member_uninitialized : Error<
   "reference member of type %0 uninitialized">;
 def note_uninit_reference_member : Note<
   "uninitialized reference member is here">;
+def warn_field_requires_explicit_init : Warning<
+  "field %select{%1|in %1}0 requires explicit initialization but is not "
+   "explicitly initialized">, InGroup<UninitializedExplicitInit>;
 def warn_field_is_uninit : Warning<"field %0 is uninitialized when used here">,
   InGroup<Uninitialized>;
 def warn_base_class_is_uninit : Warning<

clang/lib/AST/Decl.cpp

@@ -5031,6 +5031,7 @@ RecordDecl::RecordDecl(Kind DK, TagKind TK, const ASTContext &C,
   setHasNonTrivialToPrimitiveDefaultInitializeCUnion(false);
   setHasNonTrivialToPrimitiveDestructCUnion(false);
   setHasNonTrivialToPrimitiveCopyCUnion(false);
+  setHasUninitializedExplicitInitFields(false);
   setParamDestroyedInCallee(false);
   setArgPassingRestrictions(RecordArgPassingKind::CanPassInRegs);
   setIsRandomized(false);
@@ -5231,9 +5232,10 @@ unsigned RecordDecl::getODRHash() {
   // Only calculate hash on first call of getODRHash per record.
   ODRHash Hash;
   Hash.AddRecordDecl(this);
-  // For RecordDecl the ODRHash is stored in the remaining 26
-  // bit of RecordDeclBits, adjust the hash to accomodate.
-  setODRHash(Hash.CalculateHash() >> 6);
+  // For RecordDecl the ODRHash is stored in the remaining
+  // bits of RecordDeclBits, adjust the hash to accommodate.
+  static_assert(sizeof(Hash.CalculateHash()) * CHAR_BIT == 32);
+  setODRHash(Hash.CalculateHash() >> (32 - NumOdrHashBits));
   return RecordDeclBits.ODRHash;
 }
 

clang/lib/AST/DeclCXX.cpp

@@ -29,6 +29,7 @@
 #include "clang/AST/TypeLoc.h"
 #include "clang/AST/UnresolvedSet.h"
 #include "clang/Basic/Diagnostic.h"
+#include "clang/Basic/DiagnosticAST.h"
 #include "clang/Basic/IdentifierTable.h"
 #include "clang/Basic/LLVM.h"
 #include "clang/Basic/LangOptions.h"
@@ -457,6 +458,10 @@ CXXRecordDecl::setBases(CXXBaseSpecifier const * const *Bases,
     if (BaseClassDecl->hasMutableFields())
       data().HasMutableFields = true;
 
+    if (BaseClassDecl->hasUninitializedExplicitInitFields() &&
+        BaseClassDecl->isAggregate())
+      setHasUninitializedExplicitInitFields(true);
+
     if (BaseClassDecl->hasUninitializedReferenceMember())
       data().HasUninitializedReferenceMember = true;
 
@@ -1113,6 +1118,9 @@ void CXXRecordDecl::addedMember(Decl *D) {
     } else if (!T.isCXX98PODType(Context))
       data().PlainOldData = false;
 
+    if (Field->hasAttr<ExplicitInitAttr>())
+      setHasUninitializedExplicitInitFields(true);
+
     if (T->isReferenceType()) {
       if (!Field->hasInClassInitializer())
         data().HasUninitializedReferenceMember = true;
@@ -1372,6 +1380,10 @@ void CXXRecordDecl::addedMember(Decl *D) {
         if (!FieldRec->hasCopyAssignmentWithConstParam())
           data().ImplicitCopyAssignmentHasConstParam = false;
 
+        if (FieldRec->hasUninitializedExplicitInitFields() &&
+            FieldRec->isAggregate())
+          setHasUninitializedExplicitInitFields(true);
+
         if (FieldRec->hasUninitializedReferenceMember() &&
             !Field->hasInClassInitializer())
           data().HasUninitializedReferenceMember = true;
@@ -2188,6 +2200,33 @@ void CXXRecordDecl::completeDefinition(CXXFinalOverriderMap *FinalOverriders) {
   for (conversion_iterator I = conversion_begin(), E = conversion_end();
        I != E; ++I)
     I.setAccess((*I)->getAccess());
+
+  ASTContext &Context = getASTContext();
+
+  if (isAggregate() && hasUserDeclaredConstructor() &&
+      !Context.getLangOpts().CPlusPlus20) {
+    // Diagnose any aggregate behavior changes in C++20
+    for (const FieldDecl *FD : fields()) {
+      if (const auto *AT = FD->getAttr<ExplicitInitAttr>())
+        Context.getDiagnostics().Report(
+            AT->getLocation(),
+            diag::warn_cxx20_compat_requires_explicit_init_non_aggregate)
+            << AT << FD << Context.getRecordType(this);
+    }
+  }
+
+  if (!isAggregate() && hasUninitializedExplicitInitFields()) {
+    // Diagnose any fields that required explicit initialization in a
+    // non-aggregate type. (Note that the fields may not be directly in this
+    // type, but in a subobject. In such cases we don't emit diagnoses here.)
+    for (const FieldDecl *FD : fields()) {
+      if (const auto *AT = FD->getAttr<ExplicitInitAttr>())
+        Context.getDiagnostics().Report(AT->getLocation(),
+                                        diag::warn_attribute_needs_aggregate)
+            << AT << Context.getRecordType(this);
+    }
+    setHasUninitializedExplicitInitFields(false);
+  }
 }
 
 bool CXXRecordDecl::mayBeAbstract() const {

clang/lib/Sema/SemaDecl.cpp

@@ -19232,6 +19232,8 @@ void Sema::ActOnFields(Scope *S, SourceLocation RecLoc, Decl *EnclosingDecl,
         if (FT.hasNonTrivialToPrimitiveCopyCUnion() || Record->isUnion())
           Record->setHasNonTrivialToPrimitiveCopyCUnion(true);
       }
+      if (FD->hasAttr<ExplicitInitAttr>())
+        Record->setHasUninitializedExplicitInitFields(true);
       if (FT.isDestructedType()) {
         Record->setNonTrivialToPrimitiveDestroy(true);
         Record->setParamDestroyedInCallee(true);