[clang][deps] Stop sharing FileManager across module builds in scanner

[benlangmuir]

Summary

Sharing the FileManager across implicit module builds currently leaks
paths looked up in an importer into the built module itself. This can
cause non-deterministic results across scans. It is especially bad for
modules since the path can be saved into the pcm file itself, leading to
stateful behaviour if the cache is shared.

This should not impact the number of real filesystem accesses in the
scanner, since it is already caching in the
DependencyScanningWorkerFilesystem.

Note: this change does not affect whether or not the FileManager is
shared across TUs in the scanner, which is a separate issue.

[jansvoboda11]

LGTM, thanks. (For context, this patch build on top of D131076).

clang/lib/Tooling/DependencyScanning/DependencyScanningWorker.cpp

bool runInvocation(std::shared_ptr<CompilerInvocation> Invocation,
     if (!ScanInstance.hasDiagnostics())
       return false;
 
     ScanInstance.getPreprocessorOpts().AllowPCHWithDifferentModulesCachePath =
         true;
 
     ScanInstance.getFrontendOpts().GenerateGlobalModuleIndex = false;
     ScanInstance.getFrontendOpts().UseGlobalModuleIndex = false;
+    ScanInstance.getFrontendOpts().ModulesShareFileManager = false;
 
     FileMgr->getFileSystemOpts().WorkingDir = std::string(WorkingDirectory);
     ScanInstance.setFileManager(FileMgr);
     ScanInstance.createSourceManager(*FileMgr);
 
     llvm::StringSet<> PrebuiltModulesInputFiles;
     // Store the list of prebuilt module files into header search options. This
     // will prevent the implicit build to create duplicate modules and will

clang/test/ClangScanDeps/modules-file-path-isolation.c

+// Ensure that the spelling of a path seen outside a module (e.g. header via
+// symlink) does not leak into the compilation of that module unnecessarily.
+// Note: the spelling of the modulemap path still depends on the includer, since
+// that is the only source of information about it.
+
+// REQUIRES: shell
+
+// RUN: rm -rf %t
+// RUN: split-file %s %t
+// RUN: sed -e "s|DIR|%/t|g" %t/cdb.json.in > %t/cdb.json
+// RUN: ln -s A.h %t/Z.h
+
+// RUN: clang-scan-deps -compilation-database %t/cdb.json -j 1 -format experimental-full \
+// RUN:   -mode preprocess-dependency-directives -generate-modules-path-args > %t/output
+// RUN: FileCheck %s < %t/output
+
+// CHECK:      "modules": [
+// CHECK-NEXT:   {
+// CHECK:          "file-deps": [
+// CHECK-NEXT:       "{{.*}}A.h",
+// CHECK-NEXT:       "{{.*}}module.modulemap"
+// CHECK-NEXT:     ],
+// CHECK-NEXT:     "name": "A"
+// CHECK-NEXT:   }
+
+//--- cdb.json.in
+[{
+  "directory": "DIR",
+  "command": "clang -fsyntax-only DIR/tu.c -fmodules -fmodules-cache-path=DIR/module-cache -fimplicit-modules -fimplicit-module-maps",
+  "file": "DIR/tu.c"
+}]
+
+//--- module.modulemap
+module A { header "A.h" }
+module B { header "B.h" }
+module C { header "C.h" }
+
+//--- A.h
+
+//--- B.h
+#include "Z.h"
+
+//--- tu.c
+#include "B.h"