Function spawn_publisher_loop 

fn spawn_publisher_loop(
    args: &Args,
    mps: Option<WpsClient>,
    buffered: Arc<Mutex<Vec<u8>>>,
    rx: Receiver<LineEvent>,
) -> JoinHandle<()> 

Shared publisher loop — drains the LineEvent channel, aggregates into the model-visible buffer, and publishes each line via MPS.

Leading-\r spinner handling: throttled spinner frames (npm, cargo, ora, tqdm at >50 ms/frame) are collapsed upstream — in the pending_cr_override slot inside stream_reader (pipe path) and pty_reader_loop (PTY path) — before reaching this loop. The pending_cr_line slot here remains as a secondary layer: when a frame arrives with a leading \r, the prior stored frame is published first (so no frame is silently dropped), then the new frame is stored and flushed at the next non-\r line or EOF. Consecutive spinner frames therefore appear as sequential published chunks in the live log.

The model-visible buffer (buffered) receives the post-collapse LineEvent bytes. Intermediate spinner frames collapsed by pending_cr_override never become LineEvents and are not present in buffered; the model sees only the surviving final frame for each overwrite sequence.