[clang][cas] Avoid ingeting framework binaries into cas-fs for modules

When importing a framework module we check if the framework binary
exists, and if so add an autolink for it. Ensure we canonicalize that
file in the cas-fs rather than depending on the exact contents to avoid
spurious cache misses every time a framework is rebuilt.

(cherry picked from commit e76e221)

Author: benlangmuir

clang/include/clang/Tooling/DependencyScanning/DependencyScanningCASFilesystem.h

@@ -63,6 +63,7 @@ class DependencyScanningCASFilesystem : public llvm::cas::ThreadSafeFileSystem {
   createThreadSafeProxyFS() override;
 
   llvm::ErrorOr<llvm::vfs::Status> status(const Twine &Path) override;
+  bool exists(const Twine &Path) override;
   llvm::ErrorOr<std::unique_ptr<llvm::vfs::File>>
   openFileForRead(const Twine &Path) override;
 

clang/lib/Lex/ModuleMap.cpp

@@ -962,7 +962,8 @@ static void inferFrameworkLink(Module *Mod, const DirectoryEntry *FrameworkDir,
   // for both before we give up.
   for (const char *extension : {"", ".tbd"}) {
     llvm::sys::path::replace_extension(LibName, extension);
-    if (FileMgr.getFile(LibName)) {
+    // Use VFS exists to avoid depending on the file's contents in cached builds
+    if (FileMgr.getVirtualFileSystem().exists(LibName)) {
       Mod->LinkLibraries.push_back(Module::LinkLibrary(Mod->Name,
                                                        /*IsFramework=*/true));
       return;

clang/lib/Tooling/DependencyScanning/DependencyScanningCASFilesystem.cpp

@@ -299,6 +299,13 @@ DependencyScanningCASFilesystem::status(const Twine &Path) {
   return Result.Entry->Status;
 }
 
+bool DependencyScanningCASFilesystem::exists(const Twine &Path) {
+  // Existence check does not require caching the result at the dependency
+  // scanning level. The CachingOnDiskFileSystem tracks the exists call, which
+  // ensures it is included in any resulting CASFileSystem.
+  return getCachingFS().exists(Path);
+}
+
 IntrusiveRefCntPtr<llvm::cas::ThreadSafeFileSystem>
 DependencyScanningCASFilesystem::createThreadSafeProxyFS() {
   llvm::report_fatal_error("not implemented");

clang/test/CAS/fcas-fs-framework-autolink.c

@@ -0,0 +1,51 @@
+// Check that when scanning framework modules, changing the framework binary
+// does not change the cache key.
+
+// 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: clang-scan-deps -compilation-database %t/cdb.json \
+// RUN:   -cas-path %t/cas -action-cache-path %t/cache -module-files-dir %t/outputs \
+// RUN:   -format experimental-full -mode preprocess-dependency-directives \
+// RUN:   > %t/deps_no_fw.json
+
+// RUN: echo 'build 1' > %t/Foo.framework/Foo
+
+// RUN: clang-scan-deps -compilation-database %t/cdb.json \
+// RUN:   -cas-path %t/cas -action-cache-path %t/cache -module-files-dir %t/outputs \
+// RUN:   -format experimental-full -mode preprocess-dependency-directives \
+// RUN:   > %t/deps_fw1.json
+
+// The existince of the framework is significant, since it affects autolinking.
+// RUN: not diff -u %t/deps_fw1.json %t/deps_fw2.json
+
+// RUN: echo 'build 2' > %t/Foo.framework/Foo
+
+// RUN: clang-scan-deps -compilation-database %t/cdb.json \
+// RUN:   -cas-path %t/cas -action-cache-path %t/cache -module-files-dir %t/outputs \
+// RUN:   -format experimental-full -mode preprocess-dependency-directives \
+// RUN:   > %t/deps_fw2.json
+
+// But the contents of the binary are not.
+// RUN: diff -u %t/deps_fw1.json %t/deps_fw2.json
+
+//--- cdb.json.template
+[{
+  "directory" : "DIR",
+  "command" : "clang_tool -fsyntax-only DIR/tu.c -fmodules -fimplicit-modules -fimplicit-module-maps -fmodules-cache-path=DIR/module-cache -Rcompile-job-cache -F DIR",
+  "file" : "DIR/tu.c"
+}]
+
+//--- Foo.framework/Modules/module.modulemap
+framework module Foo {
+  umbrella header "Foo.h"
+  export *
+}
+
+//--- Foo.framework/Headers/Foo.h
+
+//--- tu.c
+#include "Foo/Foo.h"