Hi Sergei,
It turns out that --disable-threads didn't resolve the issue after all. It
triggered again this
afternoon. Analysis of the strace of output of expect process and some other
commands
was done by Google AI:
Dave Anglin followed up on Debian Bug #1154002 to report that even with
--disable-threads
applied to tcl8.6 (8.6.18+dfsg-2), expect still experiences intermittent
300-second DejaGnu
timeouts on native PA-RISC SMP hardware (HP rp3440) [Debian Bug #1154002].
Attaching strace
to the hanging process reveals it stuck in pselect6() polling /dev/ptmx master
descriptors
(32 and 33) [Debian Bug #1154002]:
pselect6(34, [32 33], [], [32 33], {tv_sec=42, tv_nsec=682563857}, NULL) = 0
(Timeout)
Inspecting /proc/27620/fd and system-wide /proc/[0-9]*/fd confirms that the
child process (xgcc)
has exited and been reaped via wait4(), and zero processes hold the
corresponding /dev/pts/*
slave descriptors open [Debian Bug #1154002]. Because the PTY slave is closed
yet pselect6() on
the master never receives a POLLHUP or POLLIN notification, the issue appears
to be a Linux
kernel PTY lost-wakeup race on SMP PA-RISC (between pty_close() and
n_tty_poll()) rather than
a Tcl/Expect bug [Debian Bug #1154002].
Regards,
Dave Anglin
--
John David Anglin [email protected]