Re: [Bug: 21.4.13] Core dump using EFS

APA> Hi David, APA> please use M-x efs-report-bug to make sure to get the EFS developers APA> attention. Michael> Except it's not an EFS bug if XEmacs crashes (in the process Michael> code). I'm sort of clueless: I run under FreeBSD myself, and Michael> I've never seen this bug. I've tracked down at least one facet of the problem. It is not an EFS problem, it's a pty problem that EFS triggers due to its use of ptys. The problem is that, occasionally, an opened pty does NOT retain its pty-ness after a fork. Here is an excerpt from process-unix.c showing where the problem occurs: process-unix.c:894 set_descriptor_non_blocking (inchannel); set_descriptor_non_blocking (outchannel); /* Record this as an active process, with its channels. As a result, child_setup will close Emacs's side of the pipes. */ init_process_io_handles (p, (void*)inchannel, (void*)outchannel, pty_flag ? STREAM_PTY_FLUSHING : 0); /* Record the tty descriptor used in the subprocess. */ UNIX_DATA(p)->subtty = forkin; fprintf(stderr, "2: isatty(forkin): %d\n", isatty(forkin)); { #if !defined(CYGWIN) #warning "!CYGWIN" /* child_setup must clobber environ on systems with true vfork. Protect it from permanent change. */ char **save_environ = environ; #endif pid = fork (); if (pid == 0) { /**** Now we're in the child process ****/ int xforkin = forkin; int xforkout = forkout; fprintf(stderr, "3: isatty(forkin): %d\n", isatty(forkin)); I've actually seen this happening. Here're the prints: 2: isatty(forkin): 1 3: isatty(forkin): 0 If I close and reopen the pty, it seems to fix things, but it smells like a Paclid hack: ``It needs this to make it go.'' And, of course, now that I've added even more debugging code, the problem is no longer occurring. But I did see the pty-ness change and running the hack prevented the core dump. I don't know if xemacs is using FreeBSD's ptys incorrectly, or if there is a bug in FreeBSD's pty or fork code. I'm running: FreeBSD baloo.meduseld.net 4.8-RELEASE FreeBSD 4.8-RELEASE #1 and XEmacs 21.4 (patch 13) "Rational FORTRAN" [Lucid] (i386-unknown-freebsd4.8, Mule) of Thu Jul 17 2003 on baloo.meduseld.net Anyone have any ideas??? thanks, davep -- Decision maker, n.: The person in your office who was unable to form a task force before the music stopped.