Erm . . . since you're already using the shell to do job control rather than doing fork/exec yourself and since you're not on Wintendo you have a perfectly good shell with job control built in already why not just have the shell do the wait for you?

system( qq{ command1 & command2 & command3 & wait } );
[download]

Well, on a Real™ operating system, you could replace that with:

my %child;
for my $cmd (  "command1", "command2", "command3"  ) {
    $child{system(1,$cmd)}= $cmd;
}
while(  %child  ) {
    my $cmd= delete $child{wait()};
    warn "$cmd failed: $?\n"
        if  0 != $?;
}
[download]

But if you are stuck on Unix, then it becomes much harder to "spawn" properly, unfortunately. I'll see if I can write up the helper function for that, but I suspect somebody else will beat me to it.

Okay, a close approximation to system(1,...) for less fortunate builds of Perl is:

sub spawn
{
    my( $cmd )= @_;
    die "Usage: spawn($cmdLine) or spawn([$cmd,@args])"
        if  1 != @_;
    die "You don't need this; use system(1,...) instead."
        if  $^O =~ /win32/i;    # Not an accurate test
    my $cpid= fork();
    die "Can't fork: $!\n";
        if  ! defined $cpid;
    return $cpid
        if  $cpid;
    if(  ! ref $cmd  ) {
        exec( $cmd );
    } elsif(  1 == @$cmd  ) {
        $cmd= $cmd->[0];
        exec( $cmd $cmd );
    } else {
        exec( @$cmd );
    }
    die "exec failed: $!\n";
}
[download]

But note that system(1,...) has some non-trivial advantages still; I didn't replicate the "pipe to child process for making 'exec fails' cases" saner.

So you can use spawn($cmd) in place of system(1,$cmd) in the code I provided in the previous node. I didn't write spawn() to handle Win32 because it was a quick hack.

- tye        

That seems like a perl limitation, not a unix one, since there are plenty of ways to do this in perl and without.

For whatever reason, you have to use fork or modules (like IPC::Open3 or whatever) in unix. Actually, why does system(1,$cmd) work at all? I don't see that documented. It's possible I'm blind. I'm cutting down on coffee.

UPDATE: I still don't see how that's a unix problem, rather than a perl one; but I guess I understand your animosity/reciprocity.

UPDATE2 (nov 1st): Your example (in a sibling post) also seems intentionally complex...

Read more... (1135 Bytes)

-Paul

Sorry, it is long-standing tradition that, whenever anyone asks a question where the answer in Win32 is harder than the answer in Unix, for several Unix lovers to make jokes about "get a Real™ operating system". I assumed, based on the Golden Rule, that they must want this same treatment when the situation is reversed. However, it appears to often result in several Unix lovers getting their panties in a wad when it is pointed out that their favorite operating system is not perfect, in fact, in some ways inferior to some other operating system (especially if that operating system is from a particular vendor, it seems).

Oh well, not everybody has a sense of humor about all subjects.

In my experience, the use of fork or some module will be quite a bit more complex than what I wrote and is likely to be buggy (and is very unlikely to do the nice stuff system(1,...) does with the pipe to the child so the parent can distinguish 'exec' failing from the launched program failing).

Update: Note that I don't replicate the childish part of the tradition of mispelling the name of the operating system to make fun of it. (:

- tye        

The node you linked to doesn't properly handle fork failing. That is typical.

See my reply in this thread for a solution that does handle fork failing and for more information about how it still isn't as nice as system(1,...).

- tye