Bound Cursor project path resolution (#37747)

## Why

Resolving the working directory encoded in a Cursor project name could recursively scan large directory trees.

## What changed

- Probe a bounded set of path candidates using common filename separators instead of walking the directory tree.
- Stop after 128 probes and reject ambiguous matches or unsafe encoded components.
- Parse Windows drive prefixes separately from the encoded path.

## Testing

Added coverage for common separators, ambiguous leaf and ancestor paths, and Windows drive prefixes.

GitOrigin-RevId: 34b2a2bbe2d302fac3d2614a5e36dd3fa20cf2e5

codex-rs/app-server/tests/suite/v2/external_agent_config.rs

@@ -372,23 +372,13 @@ async fn external_agent_config_secondary_source_imports_session_and_plugin_end_t
 {
     let codex_home = TempDir::new()?;
     let source_home = secondary_external_agent_home(codex_home.path());
-    let project_root = codex_home.path().join("workspace with.dots_and-dashes");
+    let project_root = codex_home.path().join("my-project");
     std::fs::create_dir_all(&project_root)?;
 
     let encoded_project = project_root
         .to_string_lossy()
         .trim_start_matches(['/', '\\'])
-        .chars()
-        .map(|character| {
-            if character.is_ascii_alphanumeric() {
-                character
-            } else {
-                '-'
-            }
-        })
-        .collect::<String>();
-    #[cfg(windows)]
-    let encoded_project = encoded_project.replacen("--", "-", /*count*/ 1);
+        .replace([':', '/', '\\'], "-");
     let session_path = source_home
         .join("projects")
         .join(encoded_project)

codex-rs/external-agent-migration/src/detect/sessions/cur.rs

@@ -8,6 +8,10 @@ use std::io;
 use std::path::Path;
 use std::path::PathBuf;
 
+const MAX_CUR_PROJECT_PATH_PROBES: usize = 128;
+const CUR_PROJECT_SEPARATORS: [&str; 11] =
+    ["-", "_", ".", " ", "--", "..", "__", "  ", "+", "@", "&"];
+
 pub fn detect_recent_cur_sessions(
     external_agent_home: &Path,
     codex_home: &Path,
@@ -84,84 +88,126 @@ fn cur_project_cwd(project_storage: &Path) -> Option<PathBuf> {
     decode_cur_project_path(encoded)
 }
 
-#[cfg(not(windows))]
 fn decode_cur_project_path(encoded: &str) -> Option<PathBuf> {
-    let root = Path::new("/");
-    let mut matches = Vec::new();
-    collect_cur_project_paths(encoded, root, root, /*depth*/ 0, &mut matches);
-    if let Some(encoded) = encoded.strip_prefix('-') {
-        collect_cur_project_paths(encoded, root, root, /*depth*/ 0, &mut matches);
+    #[cfg(not(windows))]
+    let mut path = PathBuf::from("/");
+
+    #[cfg(windows)]
+    let (encoded, mut path) = {
+        let (drive, encoded) = decode_cur_windows_project_drive(encoded)?;
+        (encoded, PathBuf::from(format!("{drive}:\\")))
+    };
+
+    let encoded = encoded.strip_prefix('-').unwrap_or(encoded);
+    for component in encoded.split('-') {
+        if component.is_empty()
+            || matches!(component, "." | "..")
+            || component.contains(['/', '\\', ':'])
+        {
+            return None;
+        }
+        path.push(component);
     }
-    unique_path(matches)
+
+    let mut matched_path = None;
+    let mut probes = 0;
+    let mut inspect = |candidate: PathBuf| {
+        if probes >= MAX_CUR_PROJECT_PATH_PROBES {
+            return None;
+        }
+        probes += 1;
+        if candidate.is_dir() {
+            if matched_path
+                .as_ref()
+                .is_some_and(|matched_path| matched_path != &candidate)
+            {
+                return None;
+            }
+            matched_path = Some(candidate);
+        }
+        Some(())
+    };
+    inspect(path.clone())?;
+
+    for suffix_length in 2..=4 {
+        let mut parent = path.as_path();
+        let mut suffix = Vec::with_capacity(suffix_length);
+        for _ in 0..suffix_length {
+            let Some(component) = parent.file_name().and_then(|name| name.to_str()) else {
+                break;
+            };
+            suffix.push(component);
+            let Some(ancestor) = parent.parent() else {
+                break;
+            };
+            parent = ancestor;
+        }
+        if suffix.len() != suffix_length {
+            break;
+        }
+        suffix.reverse();
+
+        for separator in CUR_PROJECT_SEPARATORS {
+            inspect(parent.join(suffix.join(separator)))?;
+        }
+    }
+
+    let mut ancestor = path.parent();
+    while let Some(right) = ancestor {
+        let Some(right_name) = right.file_name().and_then(|name| name.to_str()) else {
+            break;
+        };
+        let Some(left) = right.parent() else {
+            break;
+        };
+        let Some(left_name) = left.file_name().and_then(|name| name.to_str()) else {
+            break;
+        };
+        let Some(prefix) = left.parent() else {
+            break;
+        };
+        let Ok(trailing) = path.strip_prefix(right) else {
+            return None;
+        };
+
+        for separator in CUR_PROJECT_SEPARATORS {
+            let merged_prefix = prefix.join(format!("{left_name}{separator}{right_name}"));
+            if probes >= MAX_CUR_PROJECT_PATH_PROBES {
+                return None;
+            }
+            probes += 1;
+            if !merged_prefix.is_dir() {
+                continue;
+            }
+
+            if probes >= MAX_CUR_PROJECT_PATH_PROBES {
+                return None;
+            }
+            probes += 1;
+            let candidate = merged_prefix.join(trailing);
+            if !candidate.is_dir()
+                || matched_path
+                    .as_ref()
+                    .is_some_and(|matched_path| matched_path != &candidate)
+            {
+                return None;
+            }
+            matched_path = Some(candidate);
+        }
+        ancestor = Some(left);
+    }
+
+    matched_path
 }
 
-#[cfg(windows)]
-fn decode_cur_project_path(encoded: &str) -> Option<PathBuf> {
+#[cfg(any(windows, test))]
+fn decode_cur_windows_project_drive(encoded: &str) -> Option<(char, &str)> {
     let drive = encoded.as_bytes().first().copied()?;
     if !drive.is_ascii_alphabetic() || encoded.as_bytes().get(1) != Some(&b'-') {
         return None;
     }
-    let encoded = encoded.get(2..)?;
-    let base = PathBuf::from(format!("{}:\\", char::from(drive)));
-    let mut matches = Vec::new();
-    collect_cur_project_paths(encoded, &base, &base, /*depth*/ 0, &mut matches);
-    unique_path(matches)
-}
 
-fn collect_cur_project_paths(
-    encoded: &str,
-    base: &Path,
-    root: &Path,
-    depth: usize,
-    matches: &mut Vec<PathBuf>,
-) {
-    if encoded.is_empty() || depth > 32 || matches.len() > 1 {
-        return;
-    }
-    let Ok(entries) = fs::read_dir(base) else {
-        return;
-    };
-    for entry in entries.flatten() {
-        if matches.len() > 1 {
-            break;
-        }
-        let candidate = entry.path();
-        if !candidate.is_dir() {
-            continue;
-        }
-        let Ok(candidate_from_root) = candidate.strip_prefix(root) else {
-            continue;
-        };
-        let candidate_slug = cur_project_path_slug(candidate_from_root);
-        if candidate_slug == encoded {
-            if !matches.contains(&candidate) {
-                matches.push(candidate);
-            }
-        } else if encoded
-            .strip_prefix(&candidate_slug)
-            .is_some_and(|remaining| remaining.starts_with('-'))
-        {
-            collect_cur_project_paths(encoded, &candidate, root, depth + 1, matches);
-        }
-    }
-}
-
-fn cur_project_path_slug(path: &Path) -> String {
-    path.to_string_lossy()
-        .trim_start_matches(['/', '\\'])
-        .chars()
-        .map(|character| {
-            if character.is_ascii_alphanumeric() {
-                character
-            } else {
-                '-'
-            }
-        })
-        .collect()
-}
-
-fn unique_path(mut matches: Vec<PathBuf>) -> Option<PathBuf> {
-    (matches.len() == 1).then(|| matches.swap_remove(0))
+    Some((char::from(drive), encoded.get(2..)?))
 }
 
 #[cfg(test)]

codex-rs/external-agent-migration/src/detect/sessions/cur_tests.rs

@@ -138,6 +138,94 @@ fn rejects_ambiguous_encoded_project_cwd() {
     );
 }
 
+#[test]
+fn resolves_cur_project_names_with_common_separators() {
+    for (project_name, encoded_name) in [
+        ("project", "project"),
+        ("my-project", "my-project"),
+        ("my--project", "my-project"),
+        ("my project", "my-project"),
+        ("my.project", "my-project"),
+        ("my..project", "my-project"),
+        ("my_project", "my-project"),
+        ("my+project", "my-project"),
+        ("my@project", "my-project"),
+        ("my&project", "my-project"),
+        ("my-awesome-project", "my-awesome-project"),
+        ("my-awesome-cool-project", "my-awesome-cool-project"),
+    ] {
+        let root = TempDir::new().expect("tempdir");
+        let project = root.path().join(project_name);
+        fs::create_dir_all(&project).expect("project root");
+        let encoded = format!("{}-{encoded_name}", encode_project_path(root.path()));
+
+        assert_eq!(decode_cur_project_path(&encoded), Some(project));
+    }
+}
+
+#[test]
+fn rejects_ambiguous_cur_project_without_a_direct_match() {
+    let root = TempDir::new().expect("tempdir");
+    for project_name in ["my-project", "my project", "my+project"] {
+        fs::create_dir_all(root.path().join(project_name)).expect("project root");
+    }
+    let encoded = format!("{}-my-project", encode_project_path(root.path()));
+
+    assert_eq!(decode_cur_project_path(&encoded), None);
+}
+
+#[test]
+fn rejects_ambiguous_cur_project_with_punctuated_ancestor() {
+    let root = TempDir::new().expect("tempdir");
+    let punctuated_ancestor = root.path().join("a-b").join("c");
+    let punctuated_leaf = root.path().join("a").join("b-c");
+    fs::create_dir_all(&punctuated_ancestor).expect("punctuated ancestor project");
+    fs::create_dir_all(&punctuated_leaf).expect("punctuated leaf project");
+    let encoded = encode_project_path(&punctuated_ancestor);
+
+    assert_eq!(encoded, encode_project_path(&punctuated_leaf));
+    assert_eq!(decode_cur_project_path(&encoded), None);
+}
+
+#[test]
+fn rejects_ambiguous_cur_project_with_multiple_punctuated_ancestors() {
+    for (first, second) in [
+        (&["a-b", "c-d", "e"][..], &["a", "b", "c", "d-e"][..]),
+        (&["a-b", "c-d"][..], &["a", "b", "c", "d"][..]),
+    ] {
+        let root = TempDir::new().expect("tempdir");
+        let first = first
+            .iter()
+            .fold(root.path().to_path_buf(), |path, component| {
+                path.join(component)
+            });
+        let second = second
+            .iter()
+            .fold(root.path().to_path_buf(), |path, component| {
+                path.join(component)
+            });
+        fs::create_dir_all(&first).expect("first project");
+        fs::create_dir_all(&second).expect("second project");
+        let encoded = encode_project_path(&first);
+
+        assert_eq!(encoded, encode_project_path(&second));
+        assert_eq!(decode_cur_project_path(&encoded), None);
+    }
+}
+
+#[test]
+fn parses_windows_cursor_fixture_project_directory() {
+    assert_eq!(
+        decode_cur_windows_project_drive("C--Users-fixture-Cursor"),
+        Some(('C', "-Users-fixture-Cursor"))
+    );
+    assert_eq!(
+        decode_cur_windows_project_drive("C-Users-fixture-Cursor"),
+        Some(('C', "Users-fixture-Cursor"))
+    );
+    assert_eq!(decode_cur_windows_project_drive("1-Users-fixture"), None);
+}
+
 #[test]
 fn ignores_cur_sessions_older_than_import_window() {
     let root = TempDir::new().expect("tempdir");
@@ -296,10 +384,12 @@ fn set_modified_at(path: &Path, modified_at: SystemTime) {
 
 #[cfg(windows)]
 fn encode_project_path(path: &Path) -> String {
-    cur_project_path_slug(path).replacen("--", "-", 1)
+    path.to_string_lossy().replace([':', '\\', '/'], "-")
 }
 
 #[cfg(not(windows))]
 fn encode_project_path(path: &Path) -> String {
-    cur_project_path_slug(path)
+    path.to_string_lossy()
+        .trim_start_matches('/')
+        .replace('/', "-")
 }