Yeah, the change in Perl such that even the implicit closing of a popen-like file handle waits for the child to exit was a pretty sucky change in my book. For a very long time in Perl's history, only an explicit call to close would cause the wait. Super easy to get deadlock-like hangs with a modern Perl.

Yeah, the change in Perl such that even the implicit closing of a popen-like file handle waits for the child to exit was a pretty sucky change in my book.

Agreed. Sounds like a case of Perl trying to take care of too much for the programmer. Also sounds like there should be a method like:

$pipe->nowait(1);
[download]

to get around this.

POSIX::close()?

elsif ($pid > 0) {
    # write $pid to .pid file
    cleanup();
    exit(0);                 ## problem happens HERE
}
[download]

If $pid is defined here, you are in the parent process. The child doesn't get a PID from fork().

## now in sub-process
[download]

Right; but the parent is gone. What are you trying to do? This is a daemon exiting after fork. Lousy parent, I'd say.

Right; but the parent is gone. What are you trying to do? This is a daemon exiting after fork. Lousy parent, I'd say.

That's exactly what a daemon parent is supposed to do: fork() a child into a detached state (see Paul Barry's "Programming the Network with Perl" for examples, i.e. p. 125-126). Otherwise, the process call it would also block forever. But more to the point, I'm referring to the other child, then child resulting from the open($pipe, '-|', 'ip -o xfrm monitor')—that child.

That's exactly what a daemon parent is supposed to do: fork() a child into a detached state

Yes of course, a daemon closes all unnecessary file descriptors and forks itself and terminates, to detach from any terminal or I/O lines whatsoever. But in my book, the daemon does this right up front, before doing any other tasks.

But more to the point, I'm referring to the other child, then child resulting from the open($pipe, '-|', 'ip -o xfrm monitor')—that child.

Well, that child is kid of the daemons parent, so it might get reaped if grandfather exits. Perhaps reordering would help:

my $pipe;

my $pid = fork();

if (! defined $pid) {
    die "couldn't fork";
elsif ($pid > 0) {
    # write $pid to .pid file
    cleanup();
    exit(0);                 ## problem happens HERE
}

## now in sub-process
close(STDIN);
close(STDOUT);
close(STDERR);
...

# open the pipe in the child only after daemonizing
open($pipe, '-|', 'ip -o xfrm monitor') || die "couldn't start sub-com
+mand";

while (my $line = <$pipe>) {
   ...
}

close($pipe);
cleanup2();
exit(0);
[download]

perl -le'print map{pack c,($-++?1:13)+ord}split//,ESEL'