Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors
Rust
fix(store): say "does not exist" when the project id is dangling
Samir Hanna Verza53e6c0b0c955
diff --git a/crates/ai-memory-store/src/ops.rs b/crates/ai-memory-store/src/ops.rsindex 65a6c07c..cb228bfb 100644--- a/crates/ai-memory-store/src/ops.rs+++ b/crates/ai-memory-store/src/ops.rs@@ -532,6 +532,15 @@ pub fn ensure_project_with_id(/// Wiki writes call this before touching the filesystem so a stale hook/cache/// carrying the old workspace for a moved project fails before it can create an/// orphan file. The pairing INSERT triggers are still the final SQL backstop.+///+/// The two ways this fails need different words, because they send a reader+/// looking in different places. A project that moved workspaces is a stale+/// pairing: the row exists, and the caller's copy of its scope is out of date.+/// A project id with no row at all is a dangling reference — nothing moved,+/// and there is no scoping question to answer. Reporting the second as "does+/// not belong to workspace X" costs whoever reads that line a hunt for a+/// workspace/project mismatch that does not exist. The disambiguating query+/// runs only on the failure path, so the common case still pays one lookup.pub fn ensure_project_workspace(conn: &Connection,workspace_id: &WorkspaceId,@@ -545,12 +554,23 @@ pub fn ensure_project_workspace().optional()?;if found.is_some() {- Ok(())- } else {- Err(StoreError::NotFound(format!(- "project {project_id} does not belong to workspace {workspace_id}"- )))+ return Ok(());}+ let owner: Option<Vec<u8>> = conn+ .query_row(+ "SELECT workspace_id FROM projects WHERE id = ?1",+ params![project_id.as_bytes()],+ |row| row.get(0),+ )+ .optional()?;+ Err(StoreError::NotFound(match owner {+ Some(owner) => format!(+ "project {project_id} does not belong to workspace {workspace_id}; \+ it belongs to {}",+ WorkspaceId::from_slice(&owner)?+ ),+ None => format!("project {project_id} does not exist"),+ }))}/// Upsert a batch of pages inside one transaction. Either *all* pages@@ -8063,6 +8083,38 @@ pub(crate) mod tests {);}+ #[test]+ fn ensure_project_workspace_separates_a_moved_project_from_a_missing_one() {+ let (_tmp, conn, ws, proj) = fresh_db();+ let other_ws = WorkspaceId::new();+ let ghost = ProjectId::new();++ // The project exists, but the caller carries a stale workspace: the+ // message has to name where it actually lives, or the reader cannot+ // tell a stale cache from a corrupt one.+ let moved = ensure_project_workspace(&conn, &other_ws, &proj).unwrap_err();+ let moved = moved.to_string();+ assert!(+ moved.contains(&ws.to_string()) && moved.contains(&other_ws.to_string()),+ "a moved project must name both the real and the supplied workspace, got: {moved}"+ );++ // No row at all. Saying this "does not belong to workspace X" sends+ // the reader hunting a scoping bug that is not there — the id is+ // simply dangling.+ let missing = ensure_project_workspace(&conn, &ws, &ghost)+ .unwrap_err()+ .to_string();+ assert!(+ missing.contains("does not exist"),+ "a dangling project id must be reported as missing, got: {missing}"+ );+ assert!(+ !missing.contains("does not belong to workspace"),+ "a dangling project id must not be described as a workspace mismatch, got: {missing}"+ );+ }+#[test]fn ensure_workspace_with_id_rejects_id_name_mismatch() {let (_tmp, mut conn, _ws, _proj) = fresh_db();