[C++20][Modules][HU 1/5] Introduce header units as a module type.

[iains]

Summary

This is the first in a series of patches that introduce C++20 importable
header units.

These differ from clang header modules in that:
(a) they are identifiable by an internal name
(b) they represent the top level source for a single header - although that one might include or import other headers.

We name importable header units with the path by which they are specified
(although that need not be the absolute path for the file).

So "foo/bar.h" would have a name "foo/bar.h". Header units are made a
separate module type so that we can deal with diagnosing places where they
are permitted but a named module is not.

[iains]

This is a small patch set that implements the initial front end changes to support C++20 importable modules.
A second path series will follow that makes the driver changes to drive the fronted support.

This implementation makes the user-facing names, command line options and behaviour match the existing GCC implementation. There is no apparent engineering reason for divergence and this strategy minimises user learning curves and tooling differences.

1/5 - introduces the module type and build actions to compile them
2/5 - allows the user to specify that a header unit should be found from either the user or system search paths in force. In particular this avoids the user having to know the install path for system headers.
3/5 - handles emitting macros that are live at the end of the header unit TU, as required.
4/5 - handles pre-processed header unit sources
5/5 - introduces "-fdirectives-only" that can be combined with -E to produce pre-processor output that can be consumed as a second job to build an importable header (it can also be used to provide a scannable preprocesed file that has actioned any directives).

[iains]

please see first comment for a description of the patch series.

[ChuanqiXu]

clang/include/clang/Sema/Sema.h

2984–2986

From the implementation, I think it should be called only in ActOnStartOfTranslationUnit. So it would be better to move this function to private section to avoid accident calls. We should add such constraint as assumption or at least comment to tell it should be called by ActOnStartOfTranslationUnit.

---

The name ActOnStartOf* implies there should be a corresponding ActOnEndOf* methods. But there isn't one. So I would suggest to give another name to avoid ambiguity.

clang/lib/AST/Decl.cpp

1573–1575

How about

return InternalLinkage ? M->getTopLevelModule() : nullptr;

clang/lib/Parse/Parser.cpp

2475–2476

The comment is not accurate. header unit import is pre-processor too. http://eel.is/c++draft/cpp.import

clang/lib/Sema/SemaModule.cpp

519

I would feel better if we add an assertion below to assert ModuleScopes.back().Module->isGlobalModule() is true only if Mod is Header Unit.

[iains]

rebased, address review comments.

clang/include/clang/Sema/Sema.h

2984–2986

From the implementation, I think it should be called only in ActOnStartOfTranslationUnit. So it would be better to move this function to private section to avoid accident calls. We should add such constraint as assumption or at least comment to tell it should be called by ActOnStartOfTranslationUnit.

OK, I've made this private and updated the comment to note that this is a helper for ActOnStartOfTranslationUnit

---

The name ActOnStartOf* implies there should be a corresponding ActOnEndOf* methods. But there isn't one. So I would suggest to give another name to avoid ambiguity.

Is this a rule?

I think that the name ActOnStartOfHeaderUnit() says exactly what we are doing, of course, at some time we might need an ActOnEndOfHeaderUnit() - but we should not add an empty function for that reason.

(this is not a sticking point; if consensus is that the name is confusing, I will change it).

clang/lib/AST/Decl.cpp

1573–1575

that would alter the behaviour of the existing code.

getTopLevelModule() will return the ultimate parent:

const Module *Result = this;
while (Result->Parent)
  Result = Result->Parent;

where the existing code only returns the immediate parent (perhaps that is unintended, but it should be fixed separately if so).

clang/lib/Parse/Parser.cpp

2475–2476

The comment is not accurate. header unit import is pre-processor too.

the pre-processor 'import' is a pre-processor directive.

https://eel.is/c++draft/cpp.pre#1

I amended to clarify that we cannot import a named module in this position - only a header unit.

clang/lib/Sema/SemaModule.cpp

519

OK, done.
I also added a check for CPlusPlus modules, since modules-ts has an implicit GMF and slightly different rules.

[ChuanqiXu]

LGTM with suggested changes.

clang/include/clang/Sema/Sema.h

2984–2986

I don't find the rule that there must be a pair for ActOnStartOf* and ActOnEndOf*. I just find almost all other methods does so. For the consistency problem, I would still suggest to change the name to something else.

clang/lib/AST/Decl.cpp

1573–1575

I am sure that in C++20 modules, there would be 2 levels of hierarchy at most. I am OK to fix it later (Hope we don't forget it.)

[iains]

rebased

[iains]

rebased, renamed helper method.

clang/include/clang/Basic/LangOptions.h

enum CompilingModuleKind {
     CMK_None,
 
     /// Compiling a module from a module map.
     CMK_ModuleMap,
 
     /// Compiling a module from a list of header files.
     CMK_HeaderModule,
 
+    /// Compiling a module header unit.
+    CMK_HeaderUnit,
+
     /// Compiling a C++ modules TS module interface unit.
     CMK_ModuleInterface,
   };
 
   enum PragmaMSPointersToMembersKind {
     PPTMK_BestCase,
     PPTMK_FullGeneralitySingleInheritance,
     PPTMK_FullGeneralityMultipleInheritance,

clang/include/clang/Basic/LangOptions.def

 LANGOPT(POSIXThreads      , 1, 0, "POSIX thread support")
 LANGOPT(Blocks            , 1, 0, "blocks extension to C")
 BENIGN_LANGOPT(EmitAllDecls      , 1, 0, "emitting all declarations")
 LANGOPT(MathErrno         , 1, 1, "errno in math functions")
 BENIGN_LANGOPT(HeinousExtensions , 1, 0, "extensions that we really don't like and may be ripped out at any time")
 LANGOPT(Modules           , 1, 0, "modules semantics")
 COMPATIBLE_LANGOPT(ModulesTS  , 1, 0, "C++ Modules TS syntax")
 COMPATIBLE_LANGOPT(CPlusPlusModules, 1, 0, "C++ modules syntax")
-BENIGN_ENUM_LANGOPT(CompilingModule, CompilingModuleKind, 2, CMK_None,
+BENIGN_ENUM_LANGOPT(CompilingModule, CompilingModuleKind, 3, CMK_None,
                     "compiling a module interface")
 BENIGN_LANGOPT(CompilingPCH, 1, 0, "building a pch")
 BENIGN_LANGOPT(BuildingPCHWithObjectFile, 1, 0, "building a pch which has a corresponding object file")
 BENIGN_LANGOPT(CacheGeneratedPCH, 1, 0, "cache generated PCH files in memory")
 BENIGN_LANGOPT(PCHInstantiateTemplates, 1, 0, "instantiate templates while building a PCH")
 COMPATIBLE_LANGOPT(ModulesDeclUse    , 1, 0, "require declaration of module uses")
 BENIGN_LANGOPT(ModulesSearchAll  , 1, 1, "searching even non-imported modules to find unresolved references")
 COMPATIBLE_LANGOPT(ModulesStrictDeclUse, 1, 0, "requiring declaration of module uses and all headers to be in modules")

clang/include/clang/Basic/Module.h

public:
   enum ModuleKind {
     /// This is a module that was defined by a module map and built out
     /// of header files.
     ModuleMapModule,
 
     /// This is a C++20 module interface unit.
     ModuleInterfaceUnit,
 
+    /// This is a C++ 20 header unit.
+    ModuleHeaderUnit,
+
     /// This is a C++ 20 module partition interface.
     ModulePartitionInterface,
 
     /// This is a C++ 20 module partition implementation.
     ModulePartitionImplementation,
 
     /// This is a fragment of the global module within some C++ module.
     GlobalModuleFragment,

clang/include/clang/Driver/Options.td

 def ast_view : Flag<["-"], "ast-view">,
   HelpText<"Build ASTs and view them with GraphViz">;
 def emit_module : Flag<["-"], "emit-module">,
   HelpText<"Generate pre-compiled module file from a module map">;
 def emit_module_interface : Flag<["-"], "emit-module-interface">,
   HelpText<"Generate pre-compiled module file from a C++ module interface">;
 def emit_header_module : Flag<["-"], "emit-header-module">,
   HelpText<"Generate pre-compiled module file from a set of header files">;
+def emit_header_unit : Flag<["-"], "emit-header-unit">,
+  HelpText<"Generate C++20 header units from header files">;
 def emit_pch : Flag<["-"], "emit-pch">,
   HelpText<"Generate pre-compiled header file">;
 def emit_llvm_bc : Flag<["-"], "emit-llvm-bc">,
   HelpText<"Build ASTs then convert to LLVM, emit .bc file">;
 def emit_llvm_only : Flag<["-"], "emit-llvm-only">,
   HelpText<"Build ASTs and convert to LLVM, discarding output">;
 def emit_codegen_only : Flag<["-"], "emit-codegen-only">,
   HelpText<"Generate machine code, but discard output">;

clang/include/clang/Frontend/FrontendActions.h

 private:
   bool PrepareToExecuteAction(CompilerInstance &CI) override;
   bool BeginSourceFileAction(CompilerInstance &CI) override;
 
   std::unique_ptr<raw_pwrite_stream>
   CreateOutputFile(CompilerInstance &CI, StringRef InFile) override;
 };
 
+class GenerateHeaderUnitAction : public GenerateModuleAction {
+
+private:
+  bool BeginSourceFileAction(CompilerInstance &CI) override;
+
+  std::unique_ptr<raw_pwrite_stream>
+  CreateOutputFile(CompilerInstance &CI, StringRef InFile) override;
+};
+
 class SyntaxOnlyAction : public ASTFrontendAction {
 protected:
   std::unique_ptr<ASTConsumer> CreateASTConsumer(CompilerInstance &CI,
                                                  StringRef InFile) override;
 
 public:
   ~SyntaxOnlyAction() override;
   bool hasCodeCompletionSupport() const override { return true; }

clang/include/clang/Frontend/FrontendOptions.h

enum ActionKind {
   GenerateModule,
 
   /// Generate pre-compiled module from a C++ module interface file.
   GenerateModuleInterface,
 
   /// Generate pre-compiled module from a set of header files.
   GenerateHeaderModule,
 
+  /// Generate a C++20 header unit module from a header file.
+  GenerateHeaderUnit,
+
   /// Generate pre-compiled header.
   GeneratePCH,
 
   /// Generate Interface Stub Files.
   GenerateInterfaceStubs,
 
   /// Only execute frontend initialization.
   InitOnly,

clang/include/clang/Lex/ModuleMap.h

public:
   ///
   /// \returns The newly-created module.
   Module *createModuleForInterfaceUnit(SourceLocation Loc, StringRef Name,
                                        Module *GlobalModule);
 
   /// Create a header module from the specified list of headers.
   Module *createHeaderModule(StringRef Name, ArrayRef<Module::Header> Headers);
 
+  /// Create a C++20 header unit.
+  Module *createHeaderUnit(SourceLocation Loc, StringRef Name,
+                           Module::Header H);
+
   /// Infer the contents of a framework module map from the given
   /// framework directory.
   Module *inferFrameworkModule(const DirectoryEntry *FrameworkDir,
                                bool IsSystem, Module *Parent);
 
   /// Create a new top-level module that is shadowed by
   /// \p ShadowingModule.
   Module *createShadowedModule(StringRef Name, bool IsFramework,

clang/include/clang/Sema/Sema.h

enum class ModuleImportState {
     FirstDecl,       ///< Parsing the first decl in a TU.
     GlobalFragment,  ///< after 'module;' but before 'module X;'
     ImportAllowed,   ///< after 'module X;' but before any non-import decl.
     ImportFinished,  ///< after any non-import decl.
     PrivateFragment, ///< after 'module :private;'.
     NotACXX20Module  ///< Not a C++20 TU, or an invalid state was found.
   };
 
+private:
+  /// The parser has begun a translation unit to be compiled as a C++20
+  /// Header Unit, helper for ActOnStartOfTranslationUnit() only.
+  void ActOnStartOfHeaderUnit();
+
+public:
   /// The parser has processed a module-declaration that begins the definition
   /// of a module interface or implementation.
   DeclGroupPtrTy ActOnModuleDecl(SourceLocation StartLoc,
                                  SourceLocation ModuleLoc, ModuleDeclKind MDK,
                                  ModuleIdPath Path, ModuleIdPath Partition,
                                  ModuleImportState &ImportState);
 
   /// The parser has processed a global-module-fragment declaration that begins
   /// the definition of the global module fragment of the current module unit.
   /// \param ModuleLoc The location of the 'module' keyword.
   DeclGroupPtrTy ActOnGlobalModuleFragmentDecl(SourceLocation ModuleLoc);
 
   /// The parser has processed a private-module-fragment declaration that begins
   /// the definition of the private module fragment of the current module unit.
   /// \param ModuleLoc The location of the 'module' keyword.
   /// \param PrivateLoc The location of the 'private' keyword.
   DeclGroupPtrTy ActOnPrivateModuleFragmentDecl(SourceLocation ModuleLoc,
                                                 SourceLocation PrivateLoc);

clang/lib/AST/Decl.cpp

case Module::ModuleMapModule:
     // Module map modules have no special linkage semantics.
     return nullptr;
 
   case Module::ModuleInterfaceUnit:
   case Module::ModulePartitionInterface:
   case Module::ModulePartitionImplementation:
     return M;
 
+  case Module::ModuleHeaderUnit:
   case Module::GlobalModuleFragment: {
     // External linkage declarations in the global module have no owning module
     // for linkage purposes. But internal linkage declarations in the global
     // module fragment of a particular module are owned by that module for
     // linkage purposes.
     // FIXME: p1815 removes the need for this distinction -- there are no
     // internal linkage declarations that need to be referred to from outside
     // this TU.
     if (IgnoreLinkage)
       return nullptr;
     bool InternalLinkage;
     if (auto *ND = dyn_cast<NamedDecl>(this))
       InternalLinkage = !ND->hasExternalFormalLinkage();
     else
       InternalLinkage = isInAnonymousNamespace();
-    return InternalLinkage ? M->Parent : nullptr;
+    return InternalLinkage ? M->Kind == Module::ModuleHeaderUnit ? M : M->Parent
+                           : nullptr;
   }
 
   case Module::PrivateModuleFragment:
     // The private module fragment is part of its containing module for linkage
     // purposes.
     return M->Parent;
   }
 
   llvm_unreachable("unknown module kind");

clang/lib/Frontend/CompilerInvocation.cpp

static const std::pair<frontend::ActionKind, unsigned> Table[] = {
       {frontend::ExtractAPI, OPT_extract_api},
 
       {frontend::FixIt, OPT_fixit_EQ},
       {frontend::FixIt, OPT_fixit},
 
       {frontend::GenerateModule, OPT_emit_module},
       {frontend::GenerateModuleInterface, OPT_emit_module_interface},
       {frontend::GenerateHeaderModule, OPT_emit_header_module},
+      {frontend::GenerateHeaderUnit, OPT_emit_header_unit},
       {frontend::GeneratePCH, OPT_emit_pch},
       {frontend::GenerateInterfaceStubs, OPT_emit_interface_stubs},
       {frontend::InitOnly, OPT_init_only},
       {frontend::ParseSyntaxOnly, OPT_fsyntax_only},
       {frontend::ModuleFileInfo, OPT_module_file_info},
       {frontend::VerifyPCH, OPT_verify_pch},
       {frontend::PrintPreamble, OPT_print_preamble},
       {frontend::PrintPreprocessedInput, OPT_E},
       {frontend::TemplightDump, OPT_templight_dump},
       {frontend::RewriteMacros, OPT_rewrite_macros},
       {frontend::RewriteObjC, OPT_rewrite_objc},
       {frontend::RewriteTest, OPT_rewrite_test},
       {frontend::RunAnalysis, OPT_analyze},
       {frontend::MigrateSource, OPT_migrate},
       {frontend::RunPreprocessorOnly, OPT_Eonly},
       {frontend::PrintDependencyDirectivesSourceMinimizerOutput,
-          OPT_print_dependency_directives_minimized_source},
   };
 
   return Table;
 }
 
 /// Maps command line option to frontend action.
 static Optional<frontend::ActionKind> getFrontendAction(OptSpecifier &Opt) {
   for (const auto &ActionOpt : getFrontendActionTable())
static bool isStrictlyPreprocessorAction(frontend::ActionKind Action) {
   case frontend::EmitLLVMOnly:
   case frontend::EmitCodeGenOnly:
   case frontend::EmitObj:
   case frontend::ExtractAPI:
   case frontend::FixIt:
   case frontend::GenerateModule:
   case frontend::GenerateModuleInterface:
   case frontend::GenerateHeaderModule:
+  case frontend::GenerateHeaderUnit:
   case frontend::GeneratePCH:
   case frontend::GenerateInterfaceStubs:
   case frontend::ParseSyntaxOnly:
   case frontend::ModuleFileInfo:
   case frontend::VerifyPCH:
   case frontend::PluginAction:
   case frontend::RewriteObjC:
   case frontend::RewriteTest:

clang/lib/Frontend/FrontendActions.cpp

 }
 
 std::unique_ptr<raw_pwrite_stream>
 GenerateHeaderModuleAction::CreateOutputFile(CompilerInstance &CI,
                                              StringRef InFile) {
   return CI.createDefaultOutputFile(/*Binary=*/true, InFile, "pcm");
 }
 
+bool GenerateHeaderUnitAction::BeginSourceFileAction(CompilerInstance &CI) {
+  if (!CI.getLangOpts().CPlusPlusModules) {
+    CI.getDiagnostics().Report(diag::err_module_interface_requires_cpp_modules);
+    return false;
+  }
+  CI.getLangOpts().setCompilingModule(LangOptions::CMK_HeaderUnit);
+  return GenerateModuleAction::BeginSourceFileAction(CI);
+}
+
+std::unique_ptr<raw_pwrite_stream>
+GenerateHeaderUnitAction::CreateOutputFile(CompilerInstance &CI,
+                                           StringRef InFile) {
+  return CI.createDefaultOutputFile(/*Binary=*/true, InFile, "pcm");
+}
+
 SyntaxOnlyAction::~SyntaxOnlyAction() {
 }
 
 std::unique_ptr<ASTConsumer>
 SyntaxOnlyAction::CreateASTConsumer(CompilerInstance &CI, StringRef InFile) {
   return std::make_unique<ASTConsumer>();
 }
 
static StringRef ModuleKindName(Module::ModuleKind MK) {
   case Module::ModuleMapModule:
     return "Module Map Module";
   case Module::ModuleInterfaceUnit:
     return "Interface Unit";
   case Module::ModulePartitionInterface:
     return "Partition Interface";
   case Module::ModulePartitionImplementation:
     return "Partition Implementation";
+  case Module::ModuleHeaderUnit:
+    return "Header Unit";
   case Module::GlobalModuleFragment:
     return "Global Module Fragment";
   case Module::PrivateModuleFragment:
     return "Private Module Fragment";
   }
   llvm_unreachable("unknown module kind!");
 }
 

clang/lib/FrontendTool/ExecuteCompilerInvocation.cpp

case ExtractAPI:
     return std::make_unique<ExtractAPIAction>();
   case FixIt:                  return std::make_unique<FixItAction>();
   case GenerateModule:
     return std::make_unique<GenerateModuleFromModuleMapAction>();
   case GenerateModuleInterface:
     return std::make_unique<GenerateModuleInterfaceAction>();
   case GenerateHeaderModule:
     return std::make_unique<GenerateHeaderModuleAction>();
+  case GenerateHeaderUnit:
+    return std::make_unique<GenerateHeaderUnitAction>();
   case GeneratePCH:            return std::make_unique<GeneratePCHAction>();
   case GenerateInterfaceStubs:
     return std::make_unique<GenerateInterfaceStubsAction>();
   case InitOnly:               return std::make_unique<InitOnlyAction>();
   case ParseSyntaxOnly:        return std::make_unique<SyntaxOnlyAction>();
   case ModuleFileInfo:         return std::make_unique<DumpModuleInfoAction>();
   case VerifyPCH:              return std::make_unique<VerifyPCHAction>();
   case TemplightDump:          return std::make_unique<TemplightDumpAction>();

clang/lib/Lex/ModuleMap.cpp

for (const Module::Header &H : Headers) {
     // Header modules are implicitly 'export *'.
     M->Exports.push_back(Module::ExportDecl(nullptr, true));
     addHeader(M, H, NormalHeader);
   }
 
   return Result;
 }
 
+Module *ModuleMap::createHeaderUnit(SourceLocation Loc, StringRef Name,
+                                    Module::Header H) {
+  assert(LangOpts.CurrentModule == Name && "module name mismatch");
+  assert(!Modules[Name] && "redefining existing module");
+
+  auto *Result = new Module(Name, Loc, nullptr, /*IsFramework*/ false,
+                            /*IsExplicit*/ false, NumCreatedModules++);
+  Result->Kind = Module::ModuleHeaderUnit;
+  Modules[Name] = SourceModule = Result;
+  addHeader(Result, H, NormalHeader);
+  return Result;
+}
+
 /// For a framework module, infer the framework against which we
 /// should link.
 static void inferFrameworkLink(Module *Mod, const DirectoryEntry *FrameworkDir,
                                FileManager &FileMgr) {
   assert(Mod->IsFramework && "Can only infer linking for framework modules");
   assert(!Mod->isSubFramework() &&
          "Can only infer linking for top-level frameworks");
 

clang/lib/Parse/Parser.cpp

Decl *Parser::ParseModuleImport(SourceLocation AtLoc,
   case Sema::ModuleImportState::NotACXX20Module:
     // We can only import a partition within a module purview.
     if (IsPartition)
       Diag(ImportLoc, diag::err_partition_import_outside_module);
     else
       SeenError = false;
     break;
   case Sema::ModuleImportState::GlobalFragment:
     // We can only have pre-processor directives in the global module
-    // fragment.  We can, however have a header unit import here.
-    if (!HeaderUnit)
+    // fragment.  We cannot import a named modules here, however we have a
+    // header unit import.
+    if (!HeaderUnit || HeaderUnit->Kind != Module::ModuleKind::ModuleHeaderUnit)
       Diag(ImportLoc, diag::err_import_in_wrong_fragment) << IsPartition << 0;
     else
       SeenError = false;
     break;
   case Sema::ModuleImportState::ImportFinished:
     if (getLangOpts().CPlusPlusModules)
       Diag(ImportLoc, diag::err_import_not_allowed_here);
     else

clang/lib/Sema/Sema.cpp

void Sema::emitAndClearUnusedLocalTypedefWarnings() {
   }
   UnusedLocalTypedefNameCandidates.clear();
 }
 
 /// This is called before the very first declaration in the translation unit
 /// is parsed. Note that the ASTContext may have already injected some
 /// declarations.
 void Sema::ActOnStartOfTranslationUnit() {
-  if (getLangOpts().ModulesTS &&
-      (getLangOpts().getCompilingModule() == LangOptions::CMK_ModuleInterface ||
+  if (getLangOpts().CPlusPlusModules &&
+      getLangOpts().getCompilingModule() == LangOptions::CMK_HeaderUnit)
+    ActOnStartOfHeaderUnit();
+  else if (getLangOpts().ModulesTS &&
+           (getLangOpts().getCompilingModule() ==
+                LangOptions::CMK_ModuleInterface ||
             getLangOpts().getCompilingModule() == LangOptions::CMK_None)) {
     // We start in an implied global module fragment.
     SourceLocation StartOfTU =
         SourceMgr.getLocForStartOfFile(SourceMgr.getMainFileID());
     ActOnGlobalModuleFragmentDecl(StartOfTU);
     ModuleScopes.back().ImplicitGlobalModuleFragment = true;
   }
 }
 

clang/lib/Sema/SemaModule.cpp

Sema::ActOnGlobalModuleFragmentDecl(SourceLocation ModuleLoc) {
   auto *TU = Context.getTranslationUnitDecl();
   TU->setModuleOwnershipKind(Decl::ModuleOwnershipKind::Visible);
   TU->setLocalOwningModule(GlobalModule);
 
   // FIXME: Consider creating an explicit representation of this declaration.
   return nullptr;
 }
 
+void Sema::ActOnStartOfHeaderUnit() {
+  assert(getLangOpts().CPlusPlusModules &&
+         "Header units are only valid for C++20 modules");
+  SourceLocation StartOfTU =
+      SourceMgr.getLocForStartOfFile(SourceMgr.getMainFileID());
+
+  StringRef HUName = getLangOpts().CurrentModule;
+  if (HUName.empty()) {
+    HUName = SourceMgr.getFileEntryForID(SourceMgr.getMainFileID())->getName();
+    const_cast<LangOptions &>(getLangOpts()).CurrentModule = HUName.str();
+  }
+
+  auto &Map = PP.getHeaderSearchInfo().getModuleMap();
+  // TODO: Make the C++20 header lookup independent.
+  Module::Header H{getLangOpts().CurrentModule, getLangOpts().CurrentModule,
+                   SourceMgr.getFileEntryForID(SourceMgr.getMainFileID())};
+  Module *Mod = Map.createHeaderUnit(StartOfTU, HUName, H);
+  assert(Mod && "module creation should not fail");
+  ModuleScopes.push_back({}); // No GMF
+  ModuleScopes.back().BeginLoc = StartOfTU;
+  ModuleScopes.back().Module = Mod;
+  ModuleScopes.back().ModuleInterface = true;
+  ModuleScopes.back().IsPartition = false;
+  VisibleModules.setVisible(Mod, StartOfTU);
+
+  // From now on, we have an owning module for all declarations we see.
+  // All of these are implicitly exported.
+  auto *TU = Context.getTranslationUnitDecl();
+  TU->setModuleOwnershipKind(Decl::ModuleOwnershipKind::Visible);
+  TU->setLocalOwningModule(Mod);
+}
+
 Sema::DeclGroupPtrTy
 Sema::ActOnModuleDecl(SourceLocation StartLoc, SourceLocation ModuleLoc,
                       ModuleDeclKind MDK, ModuleIdPath Path,
                       ModuleIdPath Partition, ModuleImportState &ImportState) {
   assert((getLangOpts().ModulesTS || getLangOpts().CPlusPlusModules) &&
          "should only have module decl in Modules TS or C++20");
 
   bool IsFirstDecl = ImportState == ModuleImportState::FirstDecl;
case LangOptions::CMK_ModuleInterface:
     MDK = ModuleDeclKind::Interface;
     break;
 
   case LangOptions::CMK_ModuleMap:
     Diag(ModuleLoc, diag::err_module_decl_in_module_map_module);
     return nullptr;
 
   case LangOptions::CMK_HeaderModule:
+  case LangOptions::CMK_HeaderUnit:
     Diag(ModuleLoc, diag::err_module_decl_in_header_module);
     return nullptr;
   }
 
   assert(ModuleScopes.size() <= 1 && "expected to be at global module scope");
 
   // FIXME: Most of this work should be done by the preprocessor rather than
   // here, in order to support macro import.
Sema::ActOnPrivateModuleFragmentDecl(SourceLocation ModuleLoc,
   //   A private-module-fragment shall appear only in a primary module
   //   interface unit.
   switch (ModuleScopes.empty() ? Module::GlobalModuleFragment
                                : ModuleScopes.back().Module->Kind) {
   case Module::ModuleMapModule:
   case Module::GlobalModuleFragment:
   case Module::ModulePartitionImplementation:
   case Module::ModulePartitionInterface:
+  case Module::ModuleHeaderUnit:
     Diag(PrivateLoc, diag::err_private_module_fragment_not_module);
     return nullptr;
 
   case Module::PrivateModuleFragment:
     Diag(PrivateLoc, diag::err_private_module_fragment_redefined);
     Diag(ModuleScopes.back().BeginLoc, diag::note_previous_definition);
     return nullptr;
 
DeclResult Sema::ActOnModuleImport(SourceLocation StartLoc,
   if (!ModuleScopes.empty())
     Context.addModuleInitializer(ModuleScopes.back().Module, Import);
 
   // A module (partition) implementation unit shall not be exported.
   if (getLangOpts().CPlusPlusModules && ExportLoc.isValid() &&
       Mod->Kind == Module::ModuleKind::ModulePartitionImplementation) {
     Diag(ExportLoc, diag::err_export_partition_impl)
         << SourceRange(ExportLoc, Path.back().second);
-  } else if (!ModuleScopes.empty() && ModuleScopes.back().ModuleInterface) {
+  } else if (!ModuleScopes.empty() &&
+             (ModuleScopes.back().ModuleInterface ||
+              (getLangOpts().CPlusPlusModules &&
+               ModuleScopes.back().Module->isGlobalModule()))) {
+    assert((!ModuleScopes.back().Module->isGlobalModule() ||
+            Mod->Kind == Module::ModuleKind::ModuleHeaderUnit) &&
+           "should only be importing a header unit into the GMF");
     // Re-export the module if the imported module is exported.
     // Note that we don't need to add re-exported module to Imports field
     // since `Exports` implies the module is imported already.
     if (ExportLoc.isValid() || getEnclosingExportDecl(Import))
       getCurrentModule()->Exports.emplace_back(Mod, false);
     else
       getCurrentModule()->Imports.insert(Mod);
   } else if (ExportLoc.isValid()) {

clang/test/Modules/cxx20-hu-01.cpp

+// Test generation and import of simple C++20 Header Units.
+
+// RUN: rm -rf %t
+// RUN: mkdir -p %t
+// RUN: split-file %s %t
+
+// RUN: %clang_cc1 -std=c++20 -emit-header-unit -xc++-header %t/hu-01.h \
+// RUN:  -o %t/hu-01.pcm
+
+// RUN: %clang_cc1 -std=c++20 -module-file-info %t/hu-01.pcm | \
+// RUN: FileCheck --check-prefix=CHECK-HU %s -DTDIR=%t
+
+// RUN: %clang_cc1 -std=c++20 -emit-module-interface %t/imp-hu-01.cpp \
+// RUN:  -fmodule-file=%t/hu-01.pcm -o %t/B.pcm -Rmodule-import 2>&1  | \
+// RUN: FileCheck --check-prefix=CHECK-IMP %s -DTDIR=%t
+
+// RUN: %clang_cc1 -std=c++20 -emit-module-interface %t/imp-hu-02.cpp \
+// RUN:  -fmodule-file=%t/hu-01.pcm -o %t/C.pcm -Rmodule-import 2>&1  | \
+// RUN: FileCheck --check-prefix=CHECK-GMF-IMP %s -DTDIR=%t
+
+// RUN: %clang_cc1 -std=c++20 -emit-header-unit -xc++-header %t/hu-02.h \
+// RUN:  -o %t/hu-02.pcm
+
+// RUN: %clang_cc1 -std=c++20 -emit-module-interface %t/imp-hu-03.cpp \
+// RUN:  -fmodule-file=%t/hu-01.pcm -fmodule-file=%t/hu-02.pcm -o %t/D.pcm \
+// RUN: -Rmodule-import 2>&1 | \
+// RUN: FileCheck --check-prefix=CHECK-BOTH %s -DTDIR=%t
+
+// RUN: %clang_cc1 -std=c++20 -emit-header-unit -xc++-header %t/hu-03.h \
+// RUN: -fmodule-file=%t/hu-01.pcm  -o %t/hu-03.pcm
+
+// RUN: %clang_cc1 -std=c++20 -module-file-info %t/hu-03.pcm | \
+// RUN: FileCheck --check-prefix=CHECK-HU-HU %s -DTDIR=%t
+
+// RUN: %clang_cc1 -std=c++20 -emit-module-interface %t/imp-hu-04.cpp \
+// RUN:  -fmodule-file=%t/hu-03.pcm -o %t/E.pcm -Rmodule-import 2>&1 | \
+// RUN: FileCheck --check-prefix=CHECK-NESTED %s -DTDIR=%t
+
+//--- hu-01.h
+int foo(int);
+
+// CHECK-HU:  ====== C++20 Module structure ======
+// CHECK-HU-NEXT:  Header Unit '[[TDIR]]/hu-01.h' is the Primary Module at index #1
+
+//--- imp-hu-01.cpp
+export module B;
+import "hu-01.h";
+
+int bar(int x) {
+  return foo(x);
+}
+// CHECK-IMP: remark: importing module '[[TDIR]]/hu-01.h' from '[[TDIR]]/hu-01.pcm'
+// expected-no-diagnostics
+
+//--- imp-hu-02.cpp
+module;
+import "hu-01.h";
+
+export module C;
+
+int bar(int x) {
+  return foo(x);
+}
+// CHECK-GMF-IMP: remark: importing module '[[TDIR]]/hu-01.h' from '[[TDIR]]/hu-01.pcm'
+// expected-no-diagnostics
+
+//--- hu-02.h
+int baz(int);
+
+//--- imp-hu-03.cpp
+module;
+export import "hu-01.h";
+
+export module D;
+import "hu-02.h";
+
+int bar(int x) {
+  return foo(x) + baz(x);
+}
+// CHECK-BOTH: remark: importing module '[[TDIR]]/hu-01.h' from '[[TDIR]]/hu-01.pcm'
+// CHECK-BOTH: remark: importing module '[[TDIR]]/hu-02.h' from '[[TDIR]]/hu-02.pcm'
+// expected-no-diagnostics
+
+//--- hu-03.h
+export import "hu-01.h";
+int baz(int);
+// CHECK-HU-HU:  ====== C++20 Module structure ======
+// CHECK-HU-HU-NEXT:  Header Unit '[[TDIR]]/hu-03.h' is the Primary Module at index #2
+// CHECK-HU-HU-NEXT:   Exports:
+// CHECK-HU-HU-NEXT:    Header Unit '[[TDIR]]/hu-01.h' is at index #1
+
+// expected-no-diagnostics
+
+//--- imp-hu-04.cpp
+module;
+import "hu-03.h";
+
+export module E;
+
+int bar(int x) {
+  return foo(x) + baz(x);
+}
+// CHECK-NESTED: remark: importing module '[[TDIR]]/hu-03.h' from '[[TDIR]]/hu-03.pcm'
+// expected-no-diagnostics