Recover a job when submission cannot proceed

submit waits for queue space. submit_nowait instead returns the unstarted job in SubmitError.Full(job) when full. Every submission error owns the job, so a generic helper can run it locally when submission is unavailable:

def run_or_submit[F: OnceCallable[[], i64]](
    pool: &ThreadPoolExecutor, job: F
) -> Result[i64, FutureError]:
    match pool.submit_nowait(job):
        case Ok(future):
            future.result()
        case Err(
            SubmitError.Full(job)
            | SubmitError.Shutdown(job)
            | SubmitError.OutOfMemory(job)
            | SubmitError.CapacityOverflow(job)
        ):
            Ok(job())

def main() -> Result[(), AllocError | IoError | FutureError | PoolError]:
    with ThreadPoolExecutor(1, 1)? as pool:
        pool.shutdown()
        number = 42
        job = def once [number]() -> i64:
            number
        print(run_or_submit(pool, job)?)?
    Ok(())

42

When you do not need the job back, call drop_job() on the submission, as in the executor lesson: the error keeps its variant and names no job. Erasing it into Failure also drops the job but loses the error type; that is a last resort.