[CAS] Improve the cc1 flag round trip speed when CAS is used. NFCI

Delay initialize the inputs from CAS to after ParseArgs to speed up the
performance of round tripping arguments. This will save the CAS
initialization/destruction costs when round tripping cc1 arguments,
which is often used to modify/update cc1 arguments.

rdar://134363755

Author: cachemeifyoucan

clang/include/clang/Basic/DiagnosticDriverKinds.td

@@ -716,6 +716,8 @@ def err_drv_inputs_and_include_tree : Error<
   "passing input files is incompatible with '-fcas-include-tree'">;
 def err_drv_incompatible_option_include_tree : Error<
   "passing incompatible option '%0' with '-fcas-include-tree'">;
+def err_drv_include_tree_miss_input_kind : Error<
+  "missing '-x' input kind when using '-fcas-include-tree'">;
 
 def err_drv_invalid_directx_shader_module : Error<
   "invalid profile : %0">;

clang/include/clang/Frontend/CompilerInstance.h

@@ -824,6 +824,9 @@ class CompilerInstance : public ModuleLoader {
                        bool RemoveFileOnSignal, bool UseTemporary,
                        bool CreateMissingDirectories);
 
+  /// Initialize inputs from CAS.
+  void initializeDelayedInputFileFromCAS();
+
 public:
   std::unique_ptr<raw_pwrite_stream> createNullOutputFile();
 

clang/lib/Frontend/CompilerInstance.cpp

@@ -89,6 +89,10 @@ CompilerInstance::~CompilerInstance() {
 void CompilerInstance::setInvocation(
     std::shared_ptr<CompilerInvocation> Value) {
   Invocation = std::move(Value);
+
+  /// Initialize the input from CAS when setting the invocation to preserve
+  /// the same behavior when perform all kinds of FrontendActions.
+  initializeDelayedInputFileFromCAS();
 }
 
 bool CompilerInstance::shouldBuildGlobalModuleIndex() const {
@@ -1000,6 +1004,57 @@ CompilerInstance::createOutputFile(StringRef OutputPath, bool Binary,
   return nullptr;
 }
 
+void CompilerInstance::initializeDelayedInputFileFromCAS() {
+  auto &Opts = Invocation->getFrontendOpts();
+  // Return if no need to initialize or already initialized.
+  if (Opts.CASIncludeTreeID.empty() || !Opts.Inputs.empty())
+    return;
+
+  // If there is include tree, initialize the inputs from CAS.
+  auto reportError = [&](llvm::Error &&E) {
+    Diagnostics->Report(diag::err_fe_unable_to_load_include_tree)
+        << Opts.CASIncludeTreeID << llvm::toString(std::move(E));
+  };
+  auto CAS = Invocation->getCASOpts().getOrCreateDatabases(*Diagnostics).first;
+  if (!CAS)
+    return;
+  auto ID = CAS->parseID(Opts.CASIncludeTreeID);
+  if (!ID)
+    return reportError(ID.takeError());
+  auto Object = CAS->getReference(*ID);
+  if (!Object)
+    return reportError(llvm::cas::ObjectStore::createUnknownObjectError(*ID));
+  auto Root = cas::IncludeTreeRoot::get(*CAS, *Object);
+  if (!Root)
+    return reportError(Root.takeError());
+  auto MainTree = Root->getMainFileTree();
+  if (!MainTree)
+    return reportError(MainTree.takeError());
+  auto BaseFile = MainTree->getBaseFile();
+  if (!BaseFile)
+    return reportError(BaseFile.takeError());
+  auto FilenameBlob = BaseFile->getFilename();
+  if (!FilenameBlob)
+    return reportError(FilenameBlob.takeError());
+
+  auto InputFilename = FilenameBlob->getData();
+
+  if (InputFilename != Module::getModuleInputBufferName()) {
+    Opts.Inputs.emplace_back(Root->getRef(), InputFilename, Opts.DashX,
+                             /*isSystem=*/false);
+  } else {
+    assert(Opts.ProgramAction == frontend::GenerateModule);
+
+    auto Kind = Opts.DashX.withFormat(InputKind::Source);
+    auto Contents = BaseFile->getContents();
+    if (!Contents)
+      return reportError(Contents.takeError());
+    auto Buffer = llvm::MemoryBufferRef(Contents->getData(), InputFilename);
+    Opts.Inputs.emplace_back(Root->getRef(), Buffer, Kind,
+                             (bool)Opts.IsSystemModule);
+  }
+}
+
 Expected<std::unique_ptr<llvm::raw_pwrite_stream>>
 CompilerInstance::createOutputFileImpl(StringRef OutputPath, bool Binary,
                                        bool RemoveFileOnSignal,
@@ -1127,6 +1182,8 @@ bool CompilerInstance::ExecuteAction(FrontendAction &Act) {
   if (getFrontendOpts().ShowStats || !getFrontendOpts().StatsFile.empty())
     llvm::EnableStatistics(false);
 
+  initializeDelayedInputFileFromCAS();
+
   for (const FrontendInputFile &FIF : getFrontendOpts().Inputs) {
     // Reset the ID tables if we are reusing the SourceManager and parsing
     // regular files.

clang/lib/Frontend/CompilerInvocation.cpp

@@ -3046,50 +3046,6 @@ static void GenerateFrontendArgs(const FrontendOptions &Opts,
       Consumer(Input.getFile());
 }
 
-static void determineInputFromIncludeTree(
-    StringRef IncludeTreeID, CASOptions &CASOpts, DiagnosticsEngine &Diags,
-    std::optional<cas::IncludeTreeRoot> &IncludeTree,
-    std::optional<llvm::MemoryBufferRef> &Buffer, StringRef &InputFilename) {
-  assert(!IncludeTreeID.empty());
-  auto reportError = [&](llvm::Error &&E) {
-    Diags.Report(diag::err_fe_unable_to_load_include_tree)
-        << IncludeTreeID << llvm::toString(std::move(E));
-  };
-  auto CAS = CASOpts.getOrCreateDatabases(Diags).first;
-  if (!CAS)
-    return;
-  auto ID = CAS->parseID(IncludeTreeID);
-  if (!ID)
-    return reportError(ID.takeError());
-  auto Object = CAS->getReference(*ID);
-  if (!Object)
-    return reportError(llvm::cas::ObjectStore::createUnknownObjectError(*ID));
-  auto Root = cas::IncludeTreeRoot::get(*CAS, *Object);
-  if (!Root)
-    return reportError(Root.takeError());
-  auto MainTree = Root->getMainFileTree();
-  if (!MainTree)
-    return reportError(MainTree.takeError());
-  auto BaseFile = MainTree->getBaseFile();
-  if (!BaseFile)
-    return reportError(BaseFile.takeError());
-  auto FilenameBlob = BaseFile->getFilename();
-  if (!FilenameBlob)
-    return reportError(FilenameBlob.takeError());
-
-  InputFilename = FilenameBlob->getData();
-  IncludeTree = *Root;
-
-  if (InputFilename != Module::getModuleInputBufferName())
-    return;
-
-  // Handle <module-include> buffer
-  auto Contents = BaseFile->getContents();
-  if (!Contents)
-    return reportError(Contents.takeError());
-  Buffer = llvm::MemoryBufferRef(Contents->getData(), InputFilename);
-}
-
 static bool ParseFrontendArgs(FrontendOptions &Opts, ArgList &Args,
                               CASOptions &CASOpts, DiagnosticsEngine &Diags,
                               bool &IsHeaderFile) {
@@ -3308,17 +3264,18 @@ static bool ParseFrontendArgs(FrontendOptions &Opts, ArgList &Args,
   std::vector<std::string> Inputs = Args.getAllArgValues(OPT_INPUT);
   Opts.Inputs.clear();
 
-  std::optional<cas::IncludeTreeRoot> Tree;
-  std::optional<llvm::MemoryBufferRef> TreeInputBuffer;
   if (!Opts.CASIncludeTreeID.empty()) {
-    if (!Inputs.empty()) {
+    if (!Inputs.empty())
       Diags.Report(diag::err_drv_inputs_and_include_tree);
+
+    if (DashX.isUnknown()) {
+      Diags.Report(diag::err_drv_include_tree_miss_input_kind);
+      DashX = Language::C; // set to default C to avoid crashing.
     }
-    StringRef InputFilename;
-    determineInputFromIncludeTree(Opts.CASIncludeTreeID, CASOpts, Diags, Tree,
-                                  TreeInputBuffer, InputFilename);
-    if (!InputFilename.empty())
-      Inputs.push_back(InputFilename.str());
+
+    // Quit early as include tree will delay input initialization.
+    Opts.DashX = DashX;
+    return Diags.getNumErrors() == NumErrorsBefore;
   }
 
   if (Inputs.empty())
@@ -3353,20 +3310,6 @@ static bool ParseFrontendArgs(FrontendOptions &Opts, ArgList &Args,
     Opts.Inputs.emplace_back(std::move(Inputs[i]), IK, IsSystem);
   }
 
-  if (Tree) {
-    FrontendInputFile &InputFile = Opts.Inputs.back();
-    if (TreeInputBuffer) {
-      // This is automatically set to modulemap when building a module; revert
-      // to a source file for the module includes buffer.
-      auto Kind = InputFile.getKind().withFormat(InputKind::Source);
-      InputFile = FrontendInputFile(Tree->getRef(), *TreeInputBuffer, Kind,
-                                    InputFile.isSystem());
-    } else {
-      InputFile = FrontendInputFile(Tree->getRef(), InputFile.getFile(),
-                                    InputFile.getKind(), InputFile.isSystem());
-    }
-  }
-
   Opts.DashX = DashX;
 
   return Diags.getNumErrors() == NumErrorsBefore;

clang/test/ClangScanDeps/modules-include-tree-api-notes.c

@@ -51,6 +51,7 @@
 // RUN:   -fcas-path %t/cas -fsyntax-only 2>&1 | \
 // RUN:   FileCheck %s -check-prefix=INCOMPATIBLE
 
+// INCOMPATIBLE: error: missing '-x' input kind when using '-fcas-include-tree'
 // INCOMPATIBLE: error: passing incompatible option '-fapinotes' with '-fcas-include-tree'
 
 //--- cdb.json.template