I haven't seen the problem you describe, oddly enough. That might be OS / version specific. My routines in the past would exit because of time or state limits exceeded (I'd have them start over after four hours or four billion states examined) or signals (I'd have them print and die if I sent them an INT or HUP). They would exit quickly and normally on the latter (some flavor of linux and ActiveState, 5.8.something.)
I did take the excellent advice presented here. My code now "uses memory management via fork".
while(1) {
if (my $pid = fork) {
open F, ">>run.log" or die "Can't open log file: $!";
print F scalar(localtime(time)), " Launched process to play cards.
+\n";
my ($user,$system,$cuser,$csystem) = times;
print F "user $user system $system children $cuser, $csystem\n";
close F;
waitpid($pid, 0);
open F, ">>run.log" or die "Can't open log file: $!";
print F scalar(localtime(time)), " Child process finished.\n";
my ($user,$system,$cuser,$csystem) = times;
print F "user $user system $system children $cuser, $csystem\n";
close F;
} else {
die "Cannot fork: $!" unless defined $pid;
initialize();
while ($current < $MAXMETRIC) {
goodbye('INT') if -e $DIEFILE;
## etc
} ## end while ($current < $MAXMETRIC)
}
}
A run shows the following:
Mon Jan 14 12:36:28 2008 Launched process to play cards.
user 0.031 system 0.031 children 0, 0
Mon Jan 14 16:29:34 2008 Child process finished.
user 9774.546 system 556.765 children 0, 0
Mon Jan 14 16:29:35 2008 Launched process to play cards.
user 9774.546 system 556.781 children 0, 0
The "Child process finished" is after the waitpid, so that should mean that my kid is dead. There is a caveat on times() that says "times for children are included only after they terminate" which should have occurred. Why didn't I see any value on the line for 16:29:34 (and similarly for 16:29:35). Is this some sort of race condition? Or is my OS (oh, look at the Vista out this window!) so brain dead that the feature isn't implemented?
