Implement VFS Path normalization for improved cross-platform tree synchronization (#1201)

crates/memofs/CHANGELOG.md

@@ -1,6 +1,7 @@
 # memofs Changelog
 
 ## Unreleased Changes
+* Added `Vfs::canonicalize`. [#1201]
 
 ## 0.3.1 (2025-11-27)
 * Added `Vfs::exists`. [#1169]

crates/memofs/Cargo.toml

@@ -19,3 +19,6 @@ crossbeam-channel = "0.5.12"
 fs-err = "2.11.0"
 notify = "4.0.17"
 serde = { version = "1.0.197", features = ["derive"] }
+
+[dev-dependencies]
+tempfile = "3.10.1"

crates/memofs/src/in_memory_fs.rs

@@ -232,6 +232,33 @@ impl VfsBackend for InMemoryFs {
         }
     }
 
+    // TODO: We rely on Rojo to prepend cwd to any relative path before storing paths
+    // in MemoFS. The current implementation will error if no prepended absolute path
+    // is found. It really only normalizes paths within the provided path's context.
+    // Example: "/Users/username/project/../other/file.txt" ->
+    // "/Users/username/other/file.txt"
+    // Erroneous example: "/Users/../../other/file.txt" -> "/other/file.txt"
+    // This is not very robust. We should implement proper path normalization here or otherwise
+    // warn if we are missing context and can not fully canonicalize the path correctly.
+    fn canonicalize(&mut self, path: &Path) -> io::Result<PathBuf> {
+        let mut normalized = PathBuf::new();
+        for component in path.components() {
+            match component {
+                std::path::Component::ParentDir => {
+                    normalized.pop();
+                }
+                std::path::Component::CurDir => {}
+                _ => normalized.push(component),
+            }
+        }
+
+        let inner = self.inner.lock().unwrap();
+        match inner.entries.get(&normalized) {
+            Some(_) => Ok(normalized),
+            None => not_found(&normalized),
+        }
+    }
+
     fn event_receiver(&self) -> crossbeam_channel::Receiver<VfsEvent> {
         let inner = self.inner.lock().unwrap();
 

crates/memofs/src/lib.rs

@@ -77,6 +77,7 @@ pub trait VfsBackend: sealed::Sealed + Send + 'static {
     fn metadata(&mut self, path: &Path) -> io::Result<Metadata>;
     fn remove_file(&mut self, path: &Path) -> io::Result<()>;
     fn remove_dir_all(&mut self, path: &Path) -> io::Result<()>;
+    fn canonicalize(&mut self, path: &Path) -> io::Result<PathBuf>;
 
     fn event_receiver(&self) -> crossbeam_channel::Receiver<VfsEvent>;
     fn watch(&mut self, path: &Path) -> io::Result<()>;
@@ -225,6 +226,11 @@ impl VfsInner {
         self.backend.metadata(path)
     }
 
+    fn canonicalize<P: AsRef<Path>>(&mut self, path: P) -> io::Result<PathBuf> {
+        let path = path.as_ref();
+        self.backend.canonicalize(path)
+    }
+
     fn event_receiver(&self) -> crossbeam_channel::Receiver<VfsEvent> {
         self.backend.event_receiver()
     }
@@ -413,6 +419,19 @@ impl Vfs {
         self.inner.lock().unwrap().metadata(path)
     }
 
+    /// Normalize a path via the underlying backend.
+    ///
+    /// Roughly equivalent to [`std::fs::canonicalize`][std::fs::canonicalize]. Relative paths are
+    /// resolved against the backend's current working directory (if applicable) and errors are
+    /// surfaced directly from the backend.
+    ///
+    /// [std::fs::canonicalize]: https://doc.rust-lang.org/stable/std/fs/fn.canonicalize.html
+    #[inline]
+    pub fn canonicalize<P: AsRef<Path>>(&self, path: P) -> io::Result<PathBuf> {
+        let path = path.as_ref();
+        self.inner.lock().unwrap().canonicalize(path)
+    }
+
     /// Retrieve a handle to the event receiver for this `Vfs`.
     #[inline]
     pub fn event_receiver(&self) -> crossbeam_channel::Receiver<VfsEvent> {
@@ -540,6 +559,13 @@ impl VfsLock<'_> {
         self.inner.metadata(path)
     }
 
+    /// Normalize a path via the underlying backend.
+    #[inline]
+    pub fn normalize<P: AsRef<Path>>(&mut self, path: P) -> io::Result<PathBuf> {
+        let path = path.as_ref();
+        self.inner.canonicalize(path)
+    }
+
     /// Retrieve a handle to the event receiver for this `Vfs`.
     #[inline]
     pub fn event_receiver(&self) -> crossbeam_channel::Receiver<VfsEvent> {
@@ -555,7 +581,9 @@ impl VfsLock<'_> {
 
 #[cfg(test)]
 mod test {
-    use crate::{InMemoryFs, Vfs, VfsSnapshot};
+    use crate::{InMemoryFs, StdBackend, Vfs, VfsSnapshot};
+    use std::io;
+    use std::path::PathBuf;
 
     /// https://github.com/rojo-rbx/rojo/issues/899
     #[test]
@@ -571,4 +599,62 @@ mod test {
             "bar\nfoo\n\n"
         );
     }
+
+    /// https://github.com/rojo-rbx/rojo/issues/1200
+    #[test]
+    fn canonicalize_in_memory_success() {
+        let mut imfs = InMemoryFs::new();
+        let contents = "Lorem ipsum dolor sit amet.".to_string();
+
+        imfs.load_snapshot("/test/file.txt", VfsSnapshot::file(contents.to_string()))
+            .unwrap();
+
+        let vfs = Vfs::new(imfs);
+
+        assert_eq!(
+            vfs.canonicalize("/test/nested/../file.txt").unwrap(),
+            PathBuf::from("/test/file.txt")
+        );
+        assert_eq!(
+            vfs.read_to_string(vfs.canonicalize("/test/nested/../file.txt").unwrap())
+                .unwrap()
+                .to_string(),
+            contents.to_string()
+        );
+    }
+
+    #[test]
+    fn canonicalize_in_memory_missing_errors() {
+        let imfs = InMemoryFs::new();
+        let vfs = Vfs::new(imfs);
+
+        let err = vfs.canonicalize("test").unwrap_err();
+        assert_eq!(err.kind(), io::ErrorKind::NotFound);
+    }
+
+    #[test]
+    fn canonicalize_std_backend_success() {
+        let contents = "Lorem ipsum dolor sit amet.".to_string();
+        let dir = tempfile::tempdir().unwrap();
+        let file_path = dir.path().join("file.txt");
+        fs_err::write(&file_path, contents.to_string()).unwrap();
+
+        let vfs = Vfs::new(StdBackend::new());
+        let canonicalized = vfs.canonicalize(&file_path).unwrap();
+        assert_eq!(canonicalized, file_path.canonicalize().unwrap());
+        assert_eq!(
+            vfs.read_to_string(&canonicalized).unwrap().to_string(),
+            contents.to_string()
+        );
+    }
+
+    #[test]
+    fn canonicalize_std_backend_missing_errors() {
+        let dir = tempfile::tempdir().unwrap();
+        let file_path = dir.path().join("test");
+
+        let vfs = Vfs::new(StdBackend::new());
+        let err = vfs.canonicalize(&file_path).unwrap_err();
+        assert_eq!(err.kind(), io::ErrorKind::NotFound);
+    }
 }

crates/memofs/src/noop_backend.rs

@@ -1,5 +1,5 @@
 use std::io;
-use std::path::Path;
+use std::path::{Path, PathBuf};
 
 use crate::{Metadata, ReadDir, VfsBackend, VfsEvent};
 
@@ -50,6 +50,10 @@ impl VfsBackend for NoopBackend {
         Err(io::Error::other("NoopBackend doesn't do anything"))
     }
 
+    fn canonicalize(&mut self, _path: &Path) -> io::Result<PathBuf> {
+        Err(io::Error::other("NoopBackend doesn't do anything"))
+    }
+
     fn event_receiver(&self) -> crossbeam_channel::Receiver<VfsEvent> {
         crossbeam_channel::never()
     }

crates/memofs/src/std_backend.rs

@@ -106,6 +106,10 @@ impl VfsBackend for StdBackend {
         })
     }
 
+    fn canonicalize(&mut self, path: &Path) -> io::Result<PathBuf> {
+        fs_err::canonicalize(path)
+    }
+
     fn event_receiver(&self) -> crossbeam_channel::Receiver<VfsEvent> {
         self.watcher_receiver.clone()
     }