[clang] [DependencyScanningFilesystem] Make sure the local/shared cache filename lookups use only absolute paths (PR #66122)

Ben Langmuir via cfe-commits cfe-commits at lists.llvm.org
Fri Sep 15 11:34:28 PDT 2023


================
@@ -330,3 +353,20 @@ DependencyScanningWorkerFilesystem::openFileForRead(const Twine &Path) {
     return Result.getError();
   return DepScanFile::create(Result.get());
 }
+
+std::error_code DependencyScanningWorkerFilesystem::setCurrentWorkingDirectory(
+    const Twine &Path) {
+  updateWorkingDirForCacheLookup(Path.str());
+  return ProxyFileSystem::setCurrentWorkingDirectory(Path);
+}
+
+void DependencyScanningWorkerFilesystem::updateWorkingDirForCacheLookup(
+    llvm::ErrorOr<std::string> CWD) {
+  if (CWD && !CWD->empty()) {
+    WorkingDirForCacheLookup = *CWD;
+  } else {
+    // The cache lookup functions will not accept relative paths for safety, so
+    // at least make it absolute from a "root".
+    WorkingDirForCacheLookup = "/";
----------------
benlangmuir wrote:

> What is the specific concern? If multiple TUs got errors for working directory and subsequently used the same standard prefix then they may resolve to same path for a relative filename but these TUs are already in an erroneous state already, it's not invalidating their correctness.

The dummy path you're generating can overlap with a real file path.  E.g. failing on "include/foo.h" poisons "/include/foo.h".

> My general point is that I don't think we should make the assumption that if a VFS implementation emits an error for setCurrentWorkingDirectory() then that VFS instance should not ever get used again subsequently.

I think this is fine as long as our cache cannot poison the value for a real path.

https://github.com/llvm/llvm-project/pull/66122


More information about the cfe-commits mailing list