Malcolm Purvis writes: Now I observe this, don't know why I didn't see it before (and don't know why I tried it again, for that matter!) Lstreams don't know anything about what they're driving, but AFAIK they are able to detect if what they're driving is dead. So it looks to me like where (in deactivate_process in process.c) we go /* ---------------- cut ---------------- */ if (!NILP (DATA_OUTSTREAM (p))) Lstream_close (XLSTREAM (DATA_OUTSTREAM (p))); if (!NILP (DATA_INSTREAM (p))) Lstream_close (XLSTREAM (DATA_INSTREAM (p))); if (!NILP (DATA_ERRSTREAM (p))) Lstream_close (XLSTREAM (DATA_ERRSTREAM (p))); /* Provide minimal implementation for deactivate_process if there's no process-specific one */ if (HAS_PROCMETH_P (deactivate_process)) PROCMETH (deactivate_process, (p, &in_usid, &err_usid)); else event_stream_delete_io_streams (p->pipe_instream, p->pipe_outstream, p->pipe_errstream, &in_usid, &err_usid); /* ---------------- cut ---------------- */ we should close the process's io_streams first (which is protected from SIGPIPE for exactly this reason in process-unix.c, see unix_deactivate_process), then close the DATA_*STREAMS (which are coding streams that protect the process streams from having to know about character encoding). Ben, can this cause any trouble? Steve P.S. Ben I got your other mail and will be working on it in a bit.