[clang][cas] Don't include PGO-related files in the include-tree when they are not used

PGO options are ignored for PCH invocations and may not be treated as dependencies from the build
system, so don't add them in the include-tree. Otherwise we may get in a situation where such a file
gets updated and an incremental build fails because of mismatched files in the include-tree of the
PCH and the translation unit that uses it.

rdar://125629656
(cherry picked from commit 4ccf152)

Author: akyrtzi

clang/include/clang/Basic/DiagnosticCASKinds.td

@@ -14,8 +14,7 @@ def err_cas_cannot_be_initialized : Error<
 def err_cas_cannot_parse_root_id : Error<
   "CAS cannot parse root-id '%0' specified by -fcas-fs">, DefaultFatal;
 def err_cas_filesystem_cannot_be_initialized : Error<
-  "CAS filesystem cannot be initialized from root-id '%0' specified by"
-  " -fcas-fs">, DefaultFatal;
+  "CAS filesystem cannot be initialized from root-id '%0': %1">, DefaultFatal;
 def err_cas_filesystem_cannot_set_working_directory : Error<
   "CAS filesystem cannot set working directory to '%0' specified by"
   " -fcas-fs-working-directory">, DefaultFatal;

clang/lib/Frontend/CompilerInvocation.cpp

@@ -1552,9 +1552,8 @@ createBaseFS(const FileSystemOptions &FSOpts, const FrontendOptions &FEOpts,
   auto ExpectedFS = IsIncludeTreeFS ? makeIncludeTreeFS(std::move(CAS), *RootID)
                                     : makeCASFS(std::move(CAS), *RootID);
   if (!ExpectedFS) {
-    llvm::consumeError(ExpectedFS.takeError());
     Diags.Report(diag::err_cas_filesystem_cannot_be_initialized)
-        << RootIDString;
+        << RootIDString << llvm::toString(ExpectedFS.takeError());
     return makeEmptyCASFS();
   }
   IntrusiveRefCntPtr<llvm::vfs::FileSystem> FS = std::move(*ExpectedFS);

clang/lib/Tooling/DependencyScanning/DependencyScanningWorker.cpp

@@ -926,7 +926,6 @@ void DependencyScanningWorker::computeDependenciesFromCompilerInvocation(
 
   // Adjust the invocation.
   auto &Frontend = Invocation->getFrontendOpts();
-  Frontend.ProgramAction = frontend::RunPreprocessorOnly;
   Frontend.OutputFile = "/dev/null";
   Frontend.DisableFree = false;
 

clang/lib/Tooling/DependencyScanning/IncludeTreeActionController.cpp

@@ -381,6 +381,10 @@ Error IncludeTreeActionController::finalizeModuleBuild(
     CompilerInstance &ModuleScanInstance) {
   // FIXME: the scan invocation is incorrect here; we need the `NewInvocation`
   // from `finalizeModuleInvocation` to finish the tree.
+  resetBenignCodeGenOptions(
+      frontend::GenerateModule,
+      ModuleScanInstance.getInvocation().getLangOpts(),
+      ModuleScanInstance.getInvocation().getCodeGenOpts());
   auto Builder = BuilderStack.pop_back_val();
   auto Tree = Builder->finishIncludeTree(ModuleScanInstance,
                                          ModuleScanInstance.getInvocation());
@@ -589,6 +593,8 @@ IncludeTreeBuilder::finishIncludeTree(CompilerInstance &ScanInstance,
 
   auto addFile = [&](StringRef FilePath,
                      bool IgnoreFileError = false) -> Error {
+    if (FilePath.empty())
+      return Error::success();
     llvm::ErrorOr<const FileEntry *> FE = FM.getFile(FilePath);
     if (!FE) {
       if (IgnoreFileError)

clang/lib/Tooling/DependencyScanning/ScanAndUpdateArgs.cpp

@@ -29,12 +29,8 @@ void tooling::dependencies::configureInvocationForCaching(
   // Clear this otherwise it defeats the purpose of making the compilation key
   // independent of certain arguments.
   CI.getCodeGenOpts().DwarfDebugFlags.clear();
-  if (FrontendOpts.ProgramAction == frontend::GeneratePCH) {
-    // Clear paths that are emitted into binaries; they do not affect PCH.
-    // For modules this is handled in ModuleDepCollector.
-    CI.getCodeGenOpts().CoverageDataFile.clear();
-    CI.getCodeGenOpts().CoverageNotesFile.clear();
-  }
+  resetBenignCodeGenOptions(FrontendOpts.ProgramAction, CI.getLangOpts(),
+                            CI.getCodeGenOpts());
 
   HeaderSearchOptions &HSOpts = CI.getHeaderSearchOpts();
   // Avoid writing potentially volatile diagnostic options into pcms.

clang/test/CAS/modules-include-tree-pgo-option.c

@@ -0,0 +1,42 @@
+// REQUIRES: ondisk_cas
+
+// RUN: rm -rf %t
+// RUN: split-file %s %t
+// RUN: sed "s|DIR|%/t|g" %t/cdb.json.template > %t/cdb.json
+
+// RUN: llvm-profdata merge -o %t/instrumentation.profdata %S/Inputs/pgo.profraw
+// RUN: clang-scan-deps -compilation-database %t/cdb.json \
+// RUN:   -cas-path %t/cas -module-files-dir %t/outputs \
+// RUN:   -format experimental-include-tree-full -mode preprocess-dependency-directives \
+// RUN:   > %t/deps.json
+
+// RUN: %deps-to-rsp %t/deps.json --module-name Top > %t/Top.rsp
+// RUN: FileCheck %s --input-file=%t/Top.rsp
+
+// RUN: cat %t/Top.rsp | sed -E 's|.*"-fcas-include-tree" "(llvmcas://[[:xdigit:]]+)".*|\1|' > %t/Top.casid
+// RUN: clang-cas-test -cas %t/cas -print-include-tree @%t/Top.casid | FileCheck %s
+// CHECK-NOT: instrumentation.profdata
+
+//--- cdb.json.template
+[{
+  "file": "DIR/tu.m",
+  "directory": "DIR",
+  "command": "clang -fsyntax-only DIR/tu.m -I DIR -fmodules -fimplicit-modules -fimplicit-module-maps -fmodules-cache-path=DIR/module-cache -Rcompile-job-cache -fprofile-instr-use=DIR/instrumentation.profdata"
+}]
+
+//--- module.modulemap
+module Top { header "Top.h" export *}
+
+//--- Top.h
+#pragma once
+struct Top {
+  int x;
+};
+void top(void);
+
+//--- tu.m
+#import "Top.h"
+
+void tu(void) {
+  top();
+}

clang/test/CAS/pgo-profile-with-pch.c

@@ -0,0 +1,28 @@
+// RUN: rm -rf %t.dir && mkdir -p %t.dir
+
+// Check that use of profile data for PCH is ignored
+
+// RUN: llvm-profdata merge -o %t.profdata %S/Inputs/pgo.profraw
+// RUN: %clang -cc1depscan -fdepscan=inline -fdepscan-include-tree -o %t-pch.rsp -cc1-args -cc1 -triple x86_64-apple-macosx12.0.0 -emit-pch -O3 -Rcompile-job-cache \
+// RUN:   -x c-header %s -o %t.h.pch -fcas-path %t.dir/cas -fprofile-instrument-use-path=%t.profdata
+// RUN: %clang @%t-pch.rsp 2>&1 | FileCheck %s --check-prefix=CACHE-MISS
+// RUN: FileCheck %s -check-prefix=PCHPROF -input-file %t-pch.rsp
+// PCHPROF-NOT: -fprofile-instrument-use-path
+
+// Update profdata file contents
+// RUN: llvm-profdata merge -o %t.profdata %S/Inputs/pgo2.profraw
+
+// Use the modified profdata file for the main file along with the PCH.
+// RUN: %clang -cc1depscan -fdepscan=inline -fdepscan-include-tree -o %t.rsp -cc1-args -cc1 -triple x86_64-apple-macosx12.0.0 -emit-obj -O3 -Rcompile-job-cache \
+// RUN:   -x c %s -o %t.o -fcas-path %t.dir/cas -fprofile-instrument-use-path=%t.profdata -include-pch %t.h.pch
+// RUN: %clang @%t.rsp 2>&1 | FileCheck %s --check-prefix=CACHE-MISS
+// RUN: FileCheck %s -check-prefix=TUPROF -input-file %t.rsp
+// TUPROF: -fprofile-instrument-use-path
+
+// Check that the modified profdata is ignored when re-scanning for the PCH.
+// RUN: %clang -cc1depscan -fdepscan=inline -fdepscan-include-tree -o %t-pch2.rsp -cc1-args -cc1 -triple x86_64-apple-macosx12.0.0 -emit-pch -O3 -Rcompile-job-cache \
+// RUN:   -x c-header %s -o %t.h.pch -fcas-path %t.dir/cas -fprofile-instrument-use-path=%t.profdata
+// RUN: diff -u %t-pch.rsp %t-pch2.rsp
+
+// CACHE-MISS: remark: compile job cache miss
+// CACHE-HIT: remark: compile job cache hit