diff options
| -rw-r--r-- | src/sysdep.c | 51 |
1 files changed, 42 insertions, 9 deletions
diff --git a/src/sysdep.c b/src/sysdep.c index 79bac258c3b..933f599bbc9 100644 --- a/src/sysdep.c +++ b/src/sysdep.c @@ -314,16 +314,49 @@ wait_for_termination (pid) status = SYS$FORCEX (&pid, 0, 0); break; #else /* not VMS */ - - /* Exit if the process has terminated. */ - if (!synch_process_alive) +#if defined (BSD) || (defined (HPUX) && !defined (HPUX_5)) + /* Note that kill returns -1 even if the process is just a zombie now. + But inevitably a SIGCHLD interrupt should be generated + and child_sig will do wait3 and make the process go away. */ + /* There is some indication that there is a bug involved with + termination of subprocesses, perhaps involving a kernel bug too, + but no idea what it is. Just as a hunch we signal SIGCHLD to see + if that causes the problem to go away or get worse. */ + sigsetmask (sigmask (SIGCHLD)); + if (0 > kill (pid, 0)) + { + sigsetmask (SIGEMPTYMASK); + kill (getpid (), SIGCHLD); + break; + } + if (wait_debugging) + sleep (1); + else + sigpause (SIGEMPTYMASK); +#else /* not BSD, and not HPUX version >= 6 */ +#ifdef UNIPLUS + if (0 > kill (pid, 0)) break; - /* Otherwise wait 1 second or until a signal comes in. */ - signal (SIGALRM, wait_for_termination_signal); - alarm (1); - pause (); - alarm (0); - signal (SIGALRM, SIG_IGN); + wait (0); +#else /* neither BSD nor UNIPLUS: random sysV */ +#ifdef HAVE_SYSV_SIGPAUSE + sighold (SIGCHLD); + if (0 > kill (pid, 0)) + { + sigrelse (SIGCHLD); + break; + } + sigpause (SIGCHLD); +#else /* not HAVE_SYSV_SIGPAUSE */ + if (0 > kill (pid, 0)) + break; + /* Using sleep instead of pause avoids timing error. + If the inferior dies just before the sleep, + we lose just one second. */ + sleep (1); +#endif /* not HAVE_SYSV_SIGPAUSE */ +#endif /* not UNIPLUS */ +#endif /* not BSD, and not HPUX version >= 6 */ #endif /* not VMS */ #else /* not subprocesses */ #ifndef BSD4_1 |