Merge pull request #66 from SteelDynamite/claude/gracious-cray-yN12q

docs(api): clarify thread-safety bounds and multi-process limits

Audit.md

@@ -1,5 +1,29 @@
 # Audit Log
 
+## 2026-04-27
+
+Found and fixed 3 issues:
+
+1. **Perf: needless clone of upload payload** (sync.rs:733) — the `SyncAction::Upload` arm read the file into `data`, computed `compute_checksum(&data)`, then called `client.put_file(path, data.clone())`. The clone existed only because the next statement needed `data.len()` for the sync-state record. Captured `data.len() as u64` into `len` first, moved `data` into `put_file`, and used `len` afterwards — one full byte copy avoided per uploaded file.
+2. **Bug: Google Tasks sync silently drops metadata-write failures** (google_tasks.rs:361, 377) — both `.listdata.json` and `.onyx-workspace.json` were written via `if let Ok(meta_content) = serde_json::to_string_pretty(...) { let _ = atomic_write(...); }`, so a serialization or atomic-write error returned `Ok(GoogleSyncResult { downloaded: N, errors: [] })` even though list/workspace ordering was never persisted. Both writes now push their errors into the `errors` vec already returned in `GoogleSyncResult`.
+3. **Code quality: unreachable dead-error path in storage dedup** (storage.rs:447) — the dedup loop computed `Option<Task>` from each `by_id` group and then `ok_or_else(|| Error::InvalidData("Empty dedup entries for task"))?`. `by_id` is only populated by `entry(uuid).or_default().push(entry)`, so every group has ≥1 element and the `None` branch is unreachable. Replaced the `Option`+`?` with direct `expect` calls (one per branch) that document the non-empty invariant; the loop now yields `Task` directly.
+
+## 2026-04-25
+
+Found and fixed 3 issues:
+
+1. **Perf: O(n²) deletion-detection in `get_sync_status`** (sync.rs:918) — for every path tracked in `sync_state.files`, the loop scanned `local_files` linearly via `.any(|f| f.path == *path)` to decide whether to count it as a deleted-locally pending change. The earlier "modified or new" loop already used the inverse direction with `sync_state.files.get(...)` (O(1)), so the second loop was the inconsistent one. Built a `HashSet<&str>` of local paths once and used `contains` for the membership check.
+2. **Perf: cascade delete walks all_tasks per frontier pop** (tauri/lib.rs:460) — `delete_task`'s descendant BFS scanned the full task list on every parent popped from the frontier, making the work O(n × depth). Built a `parent_id -> [child_id]` `HashMap` once, then the BFS visits each descendant in O(1) amortised, dropping total cost to O(n).
+3. **Code quality: duplicate atomic-write in `AppConfig::save_to_file`** (config.rs:114) — the function had its own copy of the temp-file + rename + cleanup-on-failure dance even though `storage::atomic_write` is `pub(crate)` and was already shared by `google_tasks.rs`. Replaced the inline implementation with a call to `crate::storage::atomic_write` so the crate has one canonical atomic write path.
+
+## 2026-04-24
+
+Found and fixed 3 issues:
+
+1. **Bug: orphan base entries never cleaned from sync state** (sync.rs) — when a file was deleted both locally and remotely, `compute_sync_actions` emitted no action (the `(None, None, Some(_))` arm), so the base entry in `.syncstate.json` persisted forever. On each subsequent sync the same no-op case fired and the state file grew. Added `prune_orphan_bases` pass in `sync_workspace_inner` that drops base entries not present in either scan.
+2. **Code quality: redundant is_some_and on already-matched Option** (sync.rs:208) — the `(None, Some(_), Some(b))` arm re-checked `remote` via `remote.is_some_and(|r| ...)` even though the pattern had just proven `remote` is `Some(_)`. Bound the inner value with `Some(r)` in the pattern and used `r` directly.
+3. **Code quality: single-caller sanitize_filename wrapper** (storage.rs) — `FileSystemStorage::sanitize_filename` was a one-line forwarder to `crate::sanitize_filename` with one call site. Inlined the crate call and removed the method.
+
 ## 2026-04-20
 
 Found and fixed 4 issues:

CLAUDE.md

@@ -64,7 +64,7 @@ The GUI uses Svelte 5 runes mode (`$state`, `$derived`, `$effect`, `$props()`).
 
 Pre-alpha. No users, no released builds, no data to migrate. Breaking changes to on-disk formats, config structure, or sync conventions are free — do not add migration logic.
 
-### Current state (2026-04-15)
+### Current state (2026-04-27)
 
 - **Phase 1** (Core + CLI): Complete
 - **Phase 2** (WebDAV sync): Complete — remote folder browsing, checksum-based conflict resolution, auto-sync lifecycle, per-workspace sync interval
@@ -106,7 +106,7 @@ Pre-alpha. No users, no released builds, no data to migrate. Breaking changes to
 - Task deduplication on load (handles sync conflict duplicates)
 - Subtask hierarchy: subtask count shown on parent tasks in list, subtask detail via three-panel slide navigation, inline add at top of subtask list (new subtasks prepend), collapsible completed subtasks section, cascade delete (parent deletion removes all subtasks with confirmation warning)
 - Custom confirmation dialogs (ConfirmDialog component replaces native confirm())
-- Workspace path validation (rejects system directories)
+- Workspace path validation (rejects filesystem root `/` and system directories: `/etc`, `/usr`, `/bin`, `/sbin`, `/var`, `/proc`, `/sys`, `/dev`)
 - Task detail auto-cleanup (taskStack clears when viewed task is deleted or list switches)
 - Swipe gestures on mobile: swipe left/right on a task to toggle completion (swipe direction depends on current status)
 - Accessibility: ARIA labels/roles on interactive components, keyboard handlers, `prefers-reduced-motion` CSS support

PLAN.md

@@ -765,7 +765,7 @@ WorkspaceConfig {
 - [x] List rename (inline input via list kebab menu in drawer)
 - [x] Keyboard shortcuts (Escape closes settings → detail → drawer → menus in priority order)
 - [x] Sync status indicators (last-sync time + upload/download counts chip in TasksScreen)
-- [x] Push/pull sync mode selection (session-only sync direction selector in SettingsScreen)
+- [ ] Push/pull sync mode selection (session-only sync direction selector in SettingsScreen)
 - [x] Group-by-date toggle per list (checkmark toggle in list kebab menu)
 - [x] Subtask hierarchy (expand/collapse, inline add, cascade toggle/delete)
 - [ ] Search/filter tasks
@@ -1058,6 +1058,6 @@ This project is free and open-source software licensed under GPL v3.
 
 ---
 
-**Last Updated**: 2026-04-23
-**Document Version**: 4.4
+**Last Updated**: 2026-04-27
+**Document Version**: 4.5
 **Status**: Ready to Implement - Milestone-Driven Plan

apps/tauri/src-tauri/src/lib.rs

@@ -455,12 +455,23 @@ fn delete_task(
     // so deleting a parent can't leave grandchildren orphaned with a
     // parent_id pointing at a deleted task.
     let all_tasks = repo.list_tasks(lid).map_err(|e| e.to_string())?;
+    // Build a parent -> children index in one pass so the BFS below is O(n)
+    // instead of O(n * depth) scanning all tasks for each frontier pop.
+    let mut children_by_parent: std::collections::HashMap<Uuid, Vec<Uuid>> =
+        std::collections::HashMap::new();
+    for t in &all_tasks {
+        if let Some(pid) = t.parent_id {
+            children_by_parent.entry(pid).or_default().push(t.id);
+        }
+    }
     let mut to_delete: std::collections::HashSet<Uuid> = std::collections::HashSet::new();
     let mut frontier: Vec<Uuid> = vec![tid];
     while let Some(parent) = frontier.pop() {
-        for t in &all_tasks {
-            if t.parent_id == Some(parent) && to_delete.insert(t.id) {
-                frontier.push(t.id);
+        if let Some(children) = children_by_parent.get(&parent) {
+            for &child_id in children {
+                if to_delete.insert(child_id) {
+                    frontier.push(child_id);
+                }
             }
         }
     }

crates/onyx-core/src/config.rs

@@ -116,13 +116,7 @@ impl AppConfig {
             std::fs::create_dir_all(parent)?;
         }
         let content = serde_json::to_string_pretty(&self)?;
-        // Atomic write: write to temp file then rename to prevent corruption on crash
-        let temp = path.with_extension("tmp");
-        std::fs::write(&temp, &content)?;
-        if let Err(e) = std::fs::rename(&temp, path) {
-            let _ = std::fs::remove_file(&temp);
-            return Err(e.into());
-        }
+        crate::storage::atomic_write(path, content.as_bytes())?;
         Ok(())
     }
 

crates/onyx-core/src/google_tasks.rs

@@ -358,8 +358,15 @@ pub async fn sync_google_tasks(
         list_meta.task_order = task_order;
         list_meta.updated_at = Utc::now();
 
-        if let Ok(meta_content) = serde_json::to_string_pretty(&list_meta) {
-            let _ = atomic_write(&listdata_path, meta_content.as_bytes());
+        match serde_json::to_string_pretty(&list_meta) {
+            Ok(meta_content) => {
+                if let Err(e) = atomic_write(&listdata_path, meta_content.as_bytes()) {
+                    errors.push(format!("Failed to write metadata for list '{}': {}", gt_list.title, e));
+                }
+            }
+            Err(e) => {
+                errors.push(format!("Failed to serialize metadata for list '{}': {}", gt_list.title, e));
+            }
         }
     }
 
@@ -374,8 +381,15 @@ pub async fn sync_google_tasks(
         RootMetadata::default()
     };
     root_meta.list_order = new_list_order;
-    if let Ok(meta_content) = serde_json::to_string_pretty(&root_meta) {
-        let _ = atomic_write(&root_meta_path, meta_content.as_bytes());
+    match serde_json::to_string_pretty(&root_meta) {
+        Ok(meta_content) => {
+            if let Err(e) = atomic_write(&root_meta_path, meta_content.as_bytes()) {
+                errors.push(format!("Failed to write workspace metadata: {}", e));
+            }
+        }
+        Err(e) => {
+            errors.push(format!("Failed to serialize workspace metadata: {}", e));
+        }
     }
 
     Ok(GoogleSyncResult { downloaded, errors })

crates/onyx-core/src/storage.rs

@@ -236,12 +236,8 @@ impl FileSystemStorage {
         Ok(path)
     }
 
-    fn sanitize_filename(name: &str) -> String {
-        crate::sanitize_filename(name)
-    }
-
     fn task_file_path(&self, list_dir: &Path, task: &Task) -> PathBuf {
-        let safe_title = Self::sanitize_filename(&task.title);
+        let safe_title = crate::sanitize_filename(&task.title);
         let filename = if safe_title.is_empty() {
             task.id.to_string()
         } else {
@@ -458,7 +454,9 @@ impl Storage for FileSystemStorage {
 
         let mut tasks = Vec::new();
         for (_id, entries) in by_id {
-            let winner = if entries.len() > 1 {
+            // `by_id` only inserts non-empty groups, so each `entries` has at
+            // least one element.
+            let task = if entries.len() > 1 {
                 // Read mtime once per file so sort_by doesn't hit the filesystem
                 // O(n log n) times and can't produce inconsistent orderings if a
                 // file is touched mid-sort.
@@ -483,12 +481,14 @@ impl Storage for FileSystemStorage {
                         eprintln!("Warning: failed to remove stale duplicate task file {:?}: {}", stale_path, e);
                     }
                 }
-                with_mtime.into_iter().next().map(|(_, t, _)| t)
+                let (_, t, _) = with_mtime.into_iter().next()
+                    .expect("dedup group is non-empty after drain(1..)");
+                t
             } else {
-                entries.into_iter().next().map(|(_, t)| t)
+                let (_, t) = entries.into_iter().next()
+                    .expect("dedup group is non-empty");
+                t
             };
-            let task = winner
-                .ok_or_else(|| Error::InvalidData("Empty dedup entries for task".to_string()))?;
             tasks.push(task);
         }
 

crates/onyx-core/src/sync.rs

@@ -204,8 +204,9 @@ pub fn compute_sync_actions(
             }
 
             // Remote present, local gone, base known: local was deleted
-            (None, Some(_), Some(b)) => {
-                let remote_changed = remote.is_some_and(|r| r.size != b.size || !timestamps_equal(r.last_modified.as_deref(), b.modified_at.as_deref()));
+            (None, Some(r), Some(b)) => {
+                let remote_changed = r.size != b.size
+                    || !timestamps_equal(r.last_modified.as_deref(), b.modified_at.as_deref());
                 if remote_changed {
                     // deleted locally + modified remotely -> download (remote wins)
                     actions.push(SyncAction::Download { path: path.to_string() });
@@ -229,6 +230,22 @@ pub fn compute_sync_actions(
     actions
 }
 
+/// Remove base entries for files that are gone from both local and remote.
+/// `compute_sync_actions` emits no action for the both-deleted case, so without
+/// this pass those entries would persist in `.syncstate.json` indefinitely.
+fn prune_orphan_bases(
+    sync_state: &mut SyncState,
+    local_files: &[LocalFileInfo],
+    remote_files: &[RemoteFileSnapshot],
+) {
+    let live_paths: std::collections::HashSet<&str> = local_files
+        .iter()
+        .map(|f| f.path.as_str())
+        .chain(remote_files.iter().map(|f| f.path.as_str()))
+        .collect();
+    sync_state.files.retain(|p, _| live_paths.contains(p.as_str()));
+}
+
 /// Compare two timestamps for equality by parsing both, tolerating format differences.
 fn timestamps_equal(a: Option<&str>, b: Option<&str>) -> bool {
     match (a, b) {
@@ -604,6 +621,12 @@ async fn sync_workspace_inner(
         }
     };
 
+    // Purge orphan base entries: files we previously tracked that are now gone
+    // from both local and remote. Without this, `.syncstate.json` accumulates
+    // ghost entries forever because the both-deleted diff case emits no action
+    // and so nothing else would clean them.
+    prune_orphan_bases(&mut sync_state, &local_files, &remote_files);
+
     // Compute actions from three-way diff
     let fresh_actions = compute_sync_actions(&local_files, &remote_files, &sync_state);
 
@@ -701,19 +724,20 @@ async fn execute_action(
                 Err(e) => return Err(e.into()),
             };
             let checksum = compute_checksum(&data);
+            let len = data.len() as u64;
 
             if let Some(parent) = path_parent(path) {
                 client.ensure_dir(parent).await?;
             }
 
             report(&format!("  ^ Uploading {}", path));
-            client.put_file(path, data.clone()).await?;
+            client.put_file(path, data).await?;
 
             // Record in sync state using local file metadata
             let modified = std::fs::metadata(&local_path).ok()
                 .and_then(|m| m.modified().ok())
                 .map(|t| { let dt: DateTime<Utc> = t.into(); dt.to_rfc3339() });
-            sync_state.record_file(path, &checksum, modified.as_deref(), data.len() as u64);
+            sync_state.record_file(path, &checksum, modified.as_deref(), len);
         }
 
         SyncAction::Conflict { path } => {
@@ -891,9 +915,15 @@ pub fn get_sync_status(workspace_path: &Path) -> Result<SyncStatusInfo> {
         }
     }
 
-    // Count files in base that are now missing locally (deleted)
+    // Count files in base that are now missing locally (deleted).
+    // Build a set of local paths once so the membership check is O(1) per
+    // tracked file instead of scanning local_files linearly each time.
+    let local_paths: std::collections::HashSet<&str> = local_files
+        .iter()
+        .map(|f| f.path.as_str())
+        .collect();
     for path in sync_state.files.keys() {
-        if !local_files.iter().any(|f| f.path == *path) {
+        if !local_paths.contains(path.as_str()) {
             pending_changes += 1;
         }
     }
@@ -1106,6 +1136,22 @@ mod tests {
         assert!(actions.is_empty());
     }
 
+    #[test]
+    fn test_prune_orphan_bases() {
+        let mut state = SyncState::default();
+        state.files.insert("kept_local.md".to_string(), make_base("a"));
+        state.files.insert("kept_remote.md".to_string(), make_base("b"));
+        state.files.insert("orphan.md".to_string(), make_base("c"));
+
+        let local = vec![make_local("kept_local.md", "a")];
+        let remote = vec![make_remote("kept_remote.md")];
+        prune_orphan_bases(&mut state, &local, &remote);
+
+        assert!(state.files.contains_key("kept_local.md"));
+        assert!(state.files.contains_key("kept_remote.md"));
+        assert!(!state.files.contains_key("orphan.md"));
+    }
+
     #[test]
     fn test_multiple_files_mixed() {
         let local = vec![

docs/API.md

@@ -456,7 +456,7 @@ All metadata and state files use an atomic write pattern (write to `.tmp` then r
 
 - **List names**: Rejected if they contain `/`, `\`, or `..` components. Canonicalized and verified to stay within workspace root.
 - **Sync paths**: Validated to reject `..` components and backslashes anywhere in the path before any file system operation.
-- **Workspace paths** (Tauri): Rejected if they point to system directories (`/etc`, `/usr`, `/bin`, etc.).
+- **Workspace paths** (Tauri): Rejected if they point to the filesystem root (`/`) or system directories (`/etc`, `/usr`, `/bin`, `/sbin`, `/var`, `/proc`, `/sys`, `/dev`).
 - **Filenames**: Sanitized to replace `/ \ : * ? " < > |` and control characters with `_`.
 
 ## Example: Complete Workflow
@@ -523,9 +523,9 @@ Key test areas:
 
 ## Thread Safety
 
-The `Storage` trait requires `Send + Sync`, and `TaskRepository` wraps `Box<dyn Storage + Send + Sync>`, so repository instances can be shared across threads behind a `Mutex`. The Tauri GUI uses `Mutex<AppState>` for this purpose.
+`TaskRepository` holds its storage as `Box<dyn Storage + Send + Sync>`, so any concrete storage implementation passed in must be `Send + Sync`. Repository instances can be shared across threads behind a `Mutex` — the Tauri GUI uses `Mutex<AppState>` for this purpose.
 
 For concurrent access:
 
 1. Wrap `TaskRepository` in `Mutex` or `RwLock` (the Tauri app does this)
-2. Or create separate repository instances per thread (file system handles locking)
+2. Or create separate repository instances per thread. Note that `FileSystemStorage` does not coordinate writes between processes — concurrent multi-process writes to the same workspace are not supported outside the WebDAV sync flow, which uses a `.sync.lock` file.