Reland [C++20] [Modules] [Itanium ABI] Generate the vtable in the mod… (#102287)

Reland #75912

The differences of this PR between
#75912 are:

- Fixed a regression in `Decl::isInAnotherModuleUnit()` in DeclBase.cpp
pointed by @mizvekov and add the corresponding test.
- Fixed the regression in windows
#97447. The changes are in
`CodeGenModule::getVTableLinkage` from
`clang/lib/CodeGen/CGVTables.cpp`. According to the feedbacks from MSVC
devs, the linkage of vtables won't affected by modules. So I simply
skipped the case for MSVC.

Given this is more or less fundamental to the use of modules. I hope we
can backport this to 19.x.

Author: ChuanqiXu9

clang/include/clang/AST/DeclBase.h

@@ -670,6 +670,13 @@ class alignas(8) Decl {
   /// Whether this declaration comes from another module unit.
   bool isInAnotherModuleUnit() const;
 
+  /// Whether this declaration comes from the same module unit being compiled.
+  bool isInCurrentModuleUnit() const;
+
+  /// Whether the definition of the declaration should be emitted in external
+  /// sources.
+  bool shouldEmitInExternalSource() const;
+
   /// Whether this declaration comes from explicit global module.
   bool isFromExplicitGlobalModule() const;
 

clang/include/clang/Serialization/ASTBitCodes.h

@@ -721,6 +721,9 @@ enum ASTRecordTypes {
 
   /// Record code for \#pragma clang unsafe_buffer_usage begin/end
   PP_UNSAFE_BUFFER_USAGE = 69,
+
+  /// Record code for vtables to emit.
+  VTABLES_TO_EMIT = 70,
 };
 
 /// Record types used within a source manager block.

clang/include/clang/Serialization/ASTReader.h

@@ -790,6 +790,11 @@ class ASTReader
   /// the consumer eagerly.
   SmallVector<GlobalDeclID, 16> EagerlyDeserializedDecls;
 
+  /// The IDs of all vtables to emit. The referenced declarations are passed
+  /// to the consumers' HandleVTable eagerly after passing
+  /// EagerlyDeserializedDecls.
+  SmallVector<GlobalDeclID, 16> VTablesToEmit;
+
   /// The IDs of all tentative definitions stored in the chain.
   ///
   /// Sema keeps track of all tentative definitions in a TU because it has to
@@ -1500,6 +1505,7 @@ class ASTReader
   bool isConsumerInterestedIn(Decl *D);
   void PassInterestingDeclsToConsumer();
   void PassInterestingDeclToConsumer(Decl *D);
+  void PassVTableToConsumer(CXXRecordDecl *RD);
 
   void finishPendingActions();
   void diagnoseOdrViolations();

clang/include/clang/Serialization/ASTWriter.h

@@ -500,6 +500,10 @@ class ASTWriter : public ASTDeserializationListener,
   std::vector<SourceRange> NonAffectingRanges;
   std::vector<SourceLocation::UIntTy> NonAffectingOffsetAdjustments;
 
+  /// A list of classes which need to emit the VTable in the corresponding
+  /// object file.
+  llvm::SmallVector<CXXRecordDecl *> PendingEmittingVTables;
+
   /// Computes input files that didn't affect compilation of the current module,
   /// and initializes data structures necessary for leaving those files out
   /// during \c SourceManager serialization.
@@ -857,6 +861,8 @@ class ASTWriter : public ASTDeserializationListener,
     return PredefinedDecls.count(D);
   }
 
+  void handleVTable(CXXRecordDecl *RD);
+
 private:
   // ASTDeserializationListener implementation
   void ReaderInitialized(ASTReader *Reader) override;
@@ -951,6 +957,7 @@ class PCHGenerator : public SemaConsumer {
 
   void InitializeSema(Sema &S) override { SemaPtr = &S; }
   void HandleTranslationUnit(ASTContext &Ctx) override;
+  void HandleVTable(CXXRecordDecl *RD) override { Writer.handleVTable(RD); }
   ASTMutationListener *GetASTMutationListener() override;
   ASTDeserializationListener *GetASTDeserializationListener() override;
   bool hasEmittedPCH() const { return Buffer->IsComplete; }

clang/lib/AST/ASTContext.cpp

@@ -12431,8 +12431,7 @@ bool ASTContext::DeclMustBeEmitted(const Decl *D) {
       !isMSStaticDataMemberInlineDefinition(VD))
     return false;
 
-  // Variables in other module units shouldn't be forced to be emitted.
-  if (VD->isInAnotherModuleUnit())
+  if (VD->shouldEmitInExternalSource())
     return false;
 
   // Variables that can be needed in other TUs are required.

clang/lib/AST/DeclBase.cpp

@@ -1124,20 +1124,36 @@ bool Decl::isInAnotherModuleUnit() const {
   if (!M)
     return false;
 
+  // FIXME or NOTE: maybe we need to be clear about the semantics
+  // of clang header modules. e.g., if this lives in a clang header
+  // module included by the current unit, should we return false
+  // here?
+  //
+  // This is clear for header units as the specification says the
+  // header units live in a synthesised translation unit. So we
+  // can return false here.
   M = M->getTopLevelModule();
-  // FIXME: It is problematic if the header module lives in another module
-  // unit. Consider to fix this by techniques like
-  // ExternalASTSource::hasExternalDefinitions.
-  if (M->isHeaderLikeModule())
+  if (!M->isNamedModule())
     return false;
 
-  // A global module without parent implies that we're parsing the global
-  // module. So it can't be in another module unit.
-  if (M->isGlobalModule())
+  return M != getASTContext().getCurrentNamedModule();
+}
+
+bool Decl::isInCurrentModuleUnit() const {
+  auto *M = getOwningModule();
+
+  if (!M || !M->isNamedModule())
     return false;
 
-  assert(M->isNamedModule() && "New module kind?");
-  return M != getASTContext().getCurrentNamedModule();
+  return M == getASTContext().getCurrentNamedModule();
+}
+
+bool Decl::shouldEmitInExternalSource() const {
+  ExternalASTSource *Source = getASTContext().getExternalSource();
+  if (!Source)
+    return false;
+
+  return Source->hasExternalDefinitions(this) == ExternalASTSource::EK_Always;
 }
 
 bool Decl::isFromExplicitGlobalModule() const {

clang/lib/CodeGen/CGVTables.cpp

@@ -1081,29 +1081,41 @@ llvm::GlobalVariable::LinkageTypes
 CodeGenModule::getVTableLinkage(const CXXRecordDecl *RD) {
   if (!RD->isExternallyVisible())
     return llvm::GlobalVariable::InternalLinkage;
-
-  // We're at the end of the translation unit, so the current key
-  // function is fully correct.
-  const CXXMethodDecl *keyFunction = Context.getCurrentKeyFunction(RD);
-  if (keyFunction && !RD->hasAttr<DLLImportAttr>()) {
+  
+  // In windows, the linkage of vtable is not related to modules.
+  bool IsInNamedModule = !getTarget().getCXXABI().isMicrosoft() &&
+        RD->isInNamedModule();
+  // If the CXXRecordDecl is not in a module unit, we need to get
+  // its key function. We're at the end of the translation unit, so the current
+  // key function is fully correct.
+  const CXXMethodDecl *keyFunction =
+      IsInNamedModule ? nullptr : Context.getCurrentKeyFunction(RD);
+  if (IsInNamedModule || (keyFunction && !RD->hasAttr<DLLImportAttr>())) {
     // If this class has a key function, use that to determine the
     // linkage of the vtable.
     const FunctionDecl *def = nullptr;
-    if (keyFunction->hasBody(def))
+    if (keyFunction && keyFunction->hasBody(def))
       keyFunction = cast<CXXMethodDecl>(def);
 
-    switch (keyFunction->getTemplateSpecializationKind()) {
-      case TSK_Undeclared:
-      case TSK_ExplicitSpecialization:
+    bool IsExternalDefinition =
+        IsInNamedModule ? RD->shouldEmitInExternalSource() : !def;
+
+    TemplateSpecializationKind Kind =
+        IsInNamedModule ? RD->getTemplateSpecializationKind()
+                        : keyFunction->getTemplateSpecializationKind();
+
+    switch (Kind) {
+    case TSK_Undeclared:
+    case TSK_ExplicitSpecialization:
       assert(
-          (def || CodeGenOpts.OptimizationLevel > 0 ||
+          (IsInNamedModule || def || CodeGenOpts.OptimizationLevel > 0 ||
            CodeGenOpts.getDebugInfo() != llvm::codegenoptions::NoDebugInfo) &&
-          "Shouldn't query vtable linkage without key function, "
-          "optimizations, or debug info");
-      if (!def && CodeGenOpts.OptimizationLevel > 0)
+          "Shouldn't query vtable linkage without the class in module units, "
+          "key function, optimizations, or debug info");
+      if (IsExternalDefinition && CodeGenOpts.OptimizationLevel > 0)
         return llvm::GlobalVariable::AvailableExternallyLinkage;
 
-      if (keyFunction->isInlined())
+      if (keyFunction && keyFunction->isInlined())
         return !Context.getLangOpts().AppleKext
                    ? llvm::GlobalVariable::LinkOnceODRLinkage
                    : llvm::Function::InternalLinkage;
@@ -1122,7 +1134,7 @@ CodeGenModule::getVTableLinkage(const CXXRecordDecl *RD) {
 
       case TSK_ExplicitInstantiationDeclaration:
         llvm_unreachable("Should not have been asked to emit this");
-    }
+      }
   }
 
   // -fapple-kext mode does not support weak linkage, so we must use
@@ -1216,22 +1228,20 @@ bool CodeGenVTables::isVTableExternal(const CXXRecordDecl *RD) {
       TSK == TSK_ExplicitInstantiationDefinition)
     return false;
 
+  // Otherwise, if the class is attached to a module, the tables are uniquely
+  // emitted in the object for the module unit in which it is defined.
+  if (RD->isInNamedModule())
+    return RD->shouldEmitInExternalSource();
+
   // Otherwise, if the class doesn't have a key function (possibly
   // anymore), the vtable must be defined here.
   const CXXMethodDecl *keyFunction = CGM.getContext().getCurrentKeyFunction(RD);
   if (!keyFunction)
     return false;
 
-  const FunctionDecl *Def;
   // Otherwise, if we don't have a definition of the key function, the
   // vtable must be defined somewhere else.
-  if (!keyFunction->hasBody(Def))
-    return true;
-
-  assert(Def && "The body of the key function is not assigned to Def?");
-  // If the non-inline key function comes from another module unit, the vtable
-  // must be defined there.
-  return Def->isInAnotherModuleUnit() && !Def->isInlineSpecified();
+  return !keyFunction->hasBody();
 }
 
 /// Given that we're currently at the end of the translation unit, and

clang/lib/CodeGen/ItaniumCXXABI.cpp

@@ -2308,6 +2308,9 @@ bool ItaniumCXXABI::canSpeculativelyEmitVTable(const CXXRecordDecl *RD) const {
   if (!canSpeculativelyEmitVTableAsBaseClass(RD))
     return false;
 
+  if (RD->shouldEmitInExternalSource())
+    return false;
+
   // For a complete-object vtable (or more specifically, for the VTT), we need
   // to be able to speculatively emit the vtables of all dynamic virtual bases.
   for (const auto &B : RD->vbases()) {

clang/lib/Sema/SemaDecl.cpp

@@ -18096,6 +18096,15 @@ void Sema::ActOnTagFinishDefinition(Scope *S, Decl *TagD,
       if (NumInitMethods > 1 || !Def->hasInitMethod())
         Diag(RD->getLocation(), diag::err_sycl_special_type_num_init_method);
     }
+
+    // If we're defining a dynamic class in a module interface unit, we always
+    // need to produce the vtable for it, even if the vtable is not used in the
+    // current TU.
+    //
+    // The case where the current class is not dynamic is handled in
+    // MarkVTableUsed.
+    if (getCurrentModule() && getCurrentModule()->isInterfaceOrPartition())
+      MarkVTableUsed(RD->getLocation(), RD, /*DefinitionRequired=*/true);
   }
 
   // Exit this scope of this tag's definition.

clang/lib/Sema/SemaDeclCXX.cpp

@@ -18484,11 +18484,15 @@ bool Sema::DefineUsedVTables() {
 
     bool DefineVTable = true;
 
-    // If this class has a key function, but that key function is
-    // defined in another translation unit, we don't need to emit the
-    // vtable even though we're using it.
     const CXXMethodDecl *KeyFunction = Context.getCurrentKeyFunction(Class);
-    if (KeyFunction && !KeyFunction->hasBody()) {
+    // V-tables for non-template classes with an owning module are always
+    // uniquely emitted in that module.
+    if (Class->isInCurrentModuleUnit()) {
+      DefineVTable = true;
+    } else if (KeyFunction && !KeyFunction->hasBody()) {
+      // If this class has a key function, but that key function is
+      // defined in another translation unit, we don't need to emit the
+      // vtable even though we're using it.
       // The key function is in another translation unit.
       DefineVTable = false;
       TemplateSpecializationKind TSK =
@@ -18533,7 +18537,7 @@ bool Sema::DefineUsedVTables() {
     DefinedAnything = true;
     MarkVirtualMembersReferenced(Loc, Class);
     CXXRecordDecl *Canonical = Class->getCanonicalDecl();
-    if (VTablesUsed[Canonical])
+    if (VTablesUsed[Canonical] && !Class->shouldEmitInExternalSource())
       Consumer.HandleVTable(Class);
 
     // Warn if we're emitting a weak vtable. The vtable will be weak if there is

clang/lib/Serialization/ASTReader.cpp

@@ -3927,6 +3927,13 @@ llvm::Error ASTReader::ReadASTBlock(ModuleFile &F,
       }
       break;
 
+    case VTABLES_TO_EMIT:
+      if (F.Kind == MK_MainFile ||
+          getContext().getLangOpts().BuildingPCHWithObjectFile)
+        for (unsigned I = 0, N = Record.size(); I != N;)
+          VTablesToEmit.push_back(ReadDeclID(F, Record, I));
+      break;
+
     case IMPORTED_MODULES:
       if (!F.isModule()) {
         // If we aren't loading a module (which has its own exports), make
@@ -8122,6 +8129,10 @@ void ASTReader::PassInterestingDeclToConsumer(Decl *D) {
     Consumer->HandleInterestingDecl(DeclGroupRef(D));
 }
 
+void ASTReader::PassVTableToConsumer(CXXRecordDecl *RD) {
+  Consumer->HandleVTable(RD);
+}
+
 void ASTReader::StartTranslationUnit(ASTConsumer *Consumer) {
   this->Consumer = Consumer;
 

clang/lib/Serialization/ASTReaderDecl.cpp

@@ -4209,6 +4209,13 @@ void ASTReader::PassInterestingDeclsToConsumer() {
 
   // If we add any new potential interesting decl in the last call, consume it.
   ConsumingPotentialInterestingDecls();
+
+  for (GlobalDeclID ID : VTablesToEmit) {
+    auto *RD = cast<CXXRecordDecl>(GetDecl(ID));
+    assert(!RD->shouldEmitInExternalSource());
+    PassVTableToConsumer(RD);
+  }
+  VTablesToEmit.clear();
 }
 
 void ASTReader::loadDeclUpdateRecords(PendingUpdateRecord &Record) {

clang/lib/Serialization/ASTWriter.cpp

@@ -927,6 +927,7 @@ void ASTWriter::WriteBlockInfoBlock() {
   RECORD(DECLS_TO_CHECK_FOR_DEFERRED_DIAGS);
   RECORD(PP_ASSUME_NONNULL_LOC);
   RECORD(PP_UNSAFE_BUFFER_USAGE);
+  RECORD(VTABLES_TO_EMIT);
 
   // SourceManager Block.
   BLOCK(SOURCE_MANAGER_BLOCK);
@@ -3961,6 +3962,10 @@ void ASTWriter::WriteIdentifierTable(Preprocessor &PP,
     Stream.EmitRecord(INTERESTING_IDENTIFIERS, InterestingIdents);
 }
 
+void ASTWriter::handleVTable(CXXRecordDecl *RD) {
+  PendingEmittingVTables.push_back(RD);
+}
+
 //===----------------------------------------------------------------------===//
 // DeclContext's Name Lookup Table Serialization
 //===----------------------------------------------------------------------===//
@@ -5163,6 +5168,13 @@ void ASTWriter::PrepareWritingSpecialDecls(Sema &SemaRef) {
   // Write all of the DeclsToCheckForDeferredDiags.
   for (auto *D : SemaRef.DeclsToCheckForDeferredDiags)
     GetDeclRef(D);
+
+  // Write all classes that need to emit the vtable definitions if required.
+  if (isWritingStdCXXNamedModules())
+    for (CXXRecordDecl *RD : PendingEmittingVTables)
+      GetDeclRef(RD);
+  else
+    PendingEmittingVTables.clear();
 }
 
 void ASTWriter::WriteSpecialDeclRecords(Sema &SemaRef) {
@@ -5317,6 +5329,17 @@ void ASTWriter::WriteSpecialDeclRecords(Sema &SemaRef) {
   }
   if (!DeleteExprsToAnalyze.empty())
     Stream.EmitRecord(DELETE_EXPRS_TO_ANALYZE, DeleteExprsToAnalyze);
+
+  RecordData VTablesToEmit;
+  for (CXXRecordDecl *RD : PendingEmittingVTables) {
+    if (!wasDeclEmitted(RD))
+      continue;
+
+    AddDeclRef(RD, VTablesToEmit);
+  }
+
+  if (!VTablesToEmit.empty())
+    Stream.EmitRecord(VTABLES_TO_EMIT, VTablesToEmit);
 }
 
 ASTFileSignature ASTWriter::WriteASTCore(Sema &SemaRef, StringRef isysroot,
@@ -6559,10 +6582,12 @@ void ASTRecordWriter::AddCXXDefinitionData(const CXXRecordDecl *D) {
   // computed.
   Record->push_back(D->getODRHash());
 
-  bool ModulesDebugInfo =
-      Writer->Context->getLangOpts().ModulesDebugInfo && !D->isDependentType();
-  Record->push_back(ModulesDebugInfo);
-  if (ModulesDebugInfo)
+  bool ModulesCodegen =
+      !D->isDependentType() &&
+     (Writer->Context->getLangOpts().ModulesDebugInfo ||
+      D->isInNamedModule());
+  Record->push_back(ModulesCodegen);
+  if (ModulesCodegen)
     Writer->AddDeclRef(D, Writer->ModularCodegenDecls);
 
   // IsLambda bit is already saved.

clang/lib/Serialization/ASTWriterDecl.cpp

@@ -1529,8 +1529,14 @@ void ASTDeclWriter::VisitCXXRecordDecl(CXXRecordDecl *D) {
   if (D->isThisDeclarationADefinition())
     Record.AddCXXDefinitionData(D);
 
+  if (D->isCompleteDefinition() && D->isInNamedModule())
+    Writer.AddDeclRef(D, Writer.ModularCodegenDecls);
+
   // Store (what we currently believe to be) the key function to avoid
   // deserializing every method so we can compute it.
+  //
+  // FIXME: Avoid adding the key function if the class is defined in
+  // module purview since in that case the key function is meaningless.
   if (D->isCompleteDefinition())
     Record.AddDeclRef(Context.getCurrentKeyFunction(D));