Jan Vroonhof <vroonhof(a)math.ethz.ch&gt; writes: excellent hint -- thanks, Jan! I can now reproduce the problem from a vanilla xemacs, if no ttsession process is running on the machine. truss/dbx show that there are a couple of sigaction() calls from deep inside the libtt (apparently it forks and tries to execute something or other -- I didn't check in detail. These calls do NOT happen if a ttsession is running). All this is triggered by loading the tooltalk and the sunpro stuff and executing sunpro-startup. The original sigaction done by xemacs is: TRUSS> sigaction(SIGCLD, 0xEFFFF430, 0x0064DE3C) = 0 TRUSS> new: hand = 0xEF1B7D9C mask = 0 0 0 0 flags = 0x0000 TRUSS> old: hand = 0x00000000 mask = 0 0 0 0 flags = 0x20000 the calls from inside libtt are: TRUSS> sigaction(SIGCLD, 0xEFFFD790, 0xEFFFD890) = 0 TRUSS> new: hand = 0x00000000 mask = 0 0 0 0 flags = 0x20012 TRUSS> old: hand = 0xEF1B7D9C mask = 0 0 0 0 flags = 0x0000 TRUSS> Received signal #18, SIGCLD, in waitid() [default] TRUSS> siginfo: SIGCLD CLD_EXITED pid=11826 status=0x007F TRUSS> sigaction(SIGCLD, 0xEFFFD790, 0xEFFFD890) = 0 TRUSS> new: hand = 0xEF1B7D9C mask = 0 0 0 0 flags = 0x20012 TRUSS> old: hand = 0x00000000 mask = 0 0 0 0 flags = 0x20000 I think it's pretty obvious that libtt doesn't restore the old action correctly (i.e. it resets the handler but uses its own flags 0x20012 instead of the ones xemacs had originally installed 0x0000). from /usr/include/sys/signal.h these flags are: #define SA_NOCLDSTOP 0x00020000 /* don't send job control SIGCLD's */ #define SA_NODEFER 0x00000010 #define SA_RESETHAND 0x00000002 whatever that exactly means... I think this machine is not really up to date on patches, so maybe this was fixed by Sun at some point (any Sun employees reading this, who can peek at the libtt source to confirm/deny this?) So I suggest we reinitialise our handler again, if the tt_open() call in tooltalk.c has failed. Any comments? Gunnar P.S.: I haven't checked my HPUX builds yet, but maybe we have the same problem there. Has anybody ever actually USED tooltalk on HPUX? -- Gunnar Evermann Speech, Vision & Robotics Group Engineering Department Cambridge University