Jan> Didier Verna <didier(a)xemacs.org&gt; writes: Jan> This is going to get ugly. XEmacs cannot handle library code Jan> playing with timers. Any such code must be wrapped in Jan> stop/start_async_timeouts (void) which unfortunately is also only available Jan> as part of stop/start_interrupts. Jan> A similar problem occurs with modern libesd.so's.. I have a patch for Jan> that. However it seems that your socks library calls the fairly Jan> innocent getlogin and that uses a timeout in glibc(). Apparently on Jan> Linux the timers are fair game for libraries. In addition most of them Jan> seem to be using alarm which has much lower resolution than the itimer Jan> stuff. Jan> We need a better solution for this. Andreas any ideas..? Not directly. For now I just checked what we're doing in glibc. The problem seems to be sysdeps/generic/utmp_file.c. We try to lock the utmp file and save/restore the alarm values (we don't mess around with the values - but if XEmacs overwrites alarm...). That's AFAIK (besides sleep) the only place where glibc uses alarm. Jan> That means that with socks every call to the networking stuff can call Jan> alarm(). Some more general solution must be found. Jan> Jan Jan> P.S. In the mean time play around with your nsswitch.conf to disable Jan> to the setting that is triggering the alarm call. I don't think that this is might help here. Andreas -- Andreas Jaeger SuSE Labs aj(a)suse.de private aj(a)arthur.rhein-neckar.de