The previous fix checked downcast_ref::<io::Error> directly on the
OtherError payload, but wayland-client wraps the io::Error inside
WaylandError::Io — one level deeper. Downcast failed, is_closed
stayed false, and the panic arm fired anyway.
Replace the single downcast with a source()-chain walk that finds
the io::Error at any depth, matching the same BrokenPipe /
ConnectionReset check already used for the direct IoError arm.
calloop surfaces the closed-socket condition either as
Error::IoError(BrokenPipe) — already handled — or as
Error::OtherError wrapping an std::io::Error with kind BrokenPipe,
which previously fell through to panic!(). Add a matching arm that
downcasts the inner error and treats both forms the same: log and
set exit_requested so the run loop terminates gracefully instead
of unwinding.
yamabush1