Hmbown/CodeWhale · warning

tempdir

Error message

tempdir

What it means

A `tempfile::tempdir().expect("tempdir")` panic in a `#[tokio::test]` in `crates/tui/src/tools/subagent/tests.rs`. The test creates a temporary directory to back a `SubAgentManager` store and asserts creation succeeds; failure aborts the test before any assertions run. The same expect pattern also appears in `crates/app-server/src/chat_completions.rs` test helpers, so the failure family is 'the test harness could not obtain a temp dir'.

Solutions

1. Check `echo $TMPDIR` and confirm the directory exists and is writable; fix or unset it to fall back to /tmp.
2. Free disk space or raise quota if the filesystem holding temp is full.
3. Run the test outside the sandbox/with write permission for the temp location (e.g. grant the CI job tmpfs write access).
4. For a more actionable failure, replace expect with a panic message including the underlying io::Error.

Example fix

// before
let tmp = tempdir().expect("tempdir");
// after
let tmp = tempdir().unwrap_or_else(|e| panic!("tempdir creation failed: {e}; TMPDIR={:?}", std::env::var("TMPDIR")));

// Pre-flight in CI setup:
let probe = tempfile::tempdir().expect("temp filesystem must be writable for tests");

let tmp = tempdir().unwrap_or_else(|e| panic!("tempdir failed: {e}; check TMPDIR and disk space"));

When it happens

Trigger: `tempfile::tempdir()` returns Err when it cannot create a directory under the system temp root (TMPDIR): read-only /tmp, full disk, missing TMPDIR target, sandboxed CI denying tmpfs writes, or a very long path exceeding filename limits.

Common situations: Running tests in a restricted CI container or hermetic sandbox with no writable temp; TMPDIR pointing at a deleted or read-only path; disk quota exhausted during large test runs.

Understand the failure class

Background: "open() failed", "failed to open file", "cannot create file" — what a file open error means and how to fix it — this error's family across 42 libraries.

Related errors

• credential fixture
• event recovery chunk length fits u64
• event recovery chunk length fits usize
• first coordination interrupt
• first Stop

AI-assisted analysis of Hmbown/CodeWhale@73e0f67d83 (2026-09-22). Data as JSON: /api/errors/cc09cbf56ba152e0. Report an issue: GitHub.

Appendix: source

Thrown at crates/tui/src/tools/subagent/tests.rs:6545

        .await
        .get_worker_record("agent_cancel_probe")
        .expect("worker record remains inspectable");
    assert_eq!(
        record
            .events
            .iter()
            .filter(|event| event.status == AgentWorkerStatus::Cancelled)
            .count(),
        1,
        "repeated stop must not append a second terminal outcome"
    );
}

#[tokio::test]
async fn model_wait_cancel_fans_in_once_and_preserves_checkpoint() {
    use tokio_util::sync::CancellationToken;

    let tmp = tempdir().expect("tempdir");
    let mut manager = SubAgentManager::new(tmp.path().to_path_buf(), 2);
    let agent_id = "agent_model_wait_cancel".to_string();
    let (input_tx, _input_rx) = mpsc::unbounded_channel();
    let mut agent = SubAgent::new(
        agent_id.clone(),
        FleetRole::Worker,
        "cancel while waiting on provider".to_string(),
        make_assignment(),
        "deepseek-v4-flash".to_string(),
        None,
        None,
        input_tx,
        tmp.path().to_path_buf(),
        manager.current_session_boot_id.clone(),
    );
    agent.checkpoint = Some(make_checkpoint(
        &agent_id,
        1,

View on GitHub (pinned to 73e0f67d83)