From mgoland@optonline.net

Hello guys,

I think that perl 5.8 on winXP has a problem with file descripter tables. It fails to properly handle parents STDOUT when I dup2 in the child. Here is my script that runs fine on my FreeBSD box. It properly redirects childs STDOUT to a file , while leaving parents STDOUT to the screen. On an XP box, it redirects both parents and childs STDOUT to a file. Any ideas on how to work around it ??

_____cut script______

#!/usr/bin/perl -w

# $? child error
# $$ pid

use POSIX qw ( :sys_wait_h :unistd_h);
use IO​::Handle;

my $CHILD_SPAWNED=0;
my $CHILD_TO_CREATE=5;
my $MAX_CHILD=3;
my $CHILD_CREATED=0;
$|=1;

#open a file for childs redirection
open(FD,"> ./output") || myerrno();

while( $CHILD_SPAWNED < $CHILD_TO_CREATE ){

print "-------\t\t RUNING $CHILD_CREATED------------\n";

if( $CHILD_CREATED < $MAX_CHILD ){

die "couldnt not spawn​: $!" unless defined ( my $PID=fork() );

$CHILD_CREATED++;
$CHILD_SPAWNED++;

unless( $PID ){
  $|=1;
  dup2(fileno(FD) , fileno(STDOUT) ) || myerrno();
  close(FD);
  $^F=3; #make the redirection get past exec
  print "I am child $$...about to exec\n";

{ exec ("perl count.pl") }; print STDERR "could not exec​: $!";

}

}

$status=waitpid(-1,WNOHANG);

if( $status != 0 ){ #some child has exited
$CHILD_CREATED--;

print "child $status has exited \n";
}

}

print "TOTAL CHILD SPAWNED IS $CHILD_SPAWNED\n";

sub myerrno{
foreach $Val (keys( %!) ){
  print "$Val $!{$Val}\n" if $!{$Val};

}
exit(1);
}

____paste script_____

From @jkeenan

On Mon Mar 31 12​:58​:25 2003, mgoland wrote​:

Hello guys,

I think that perl 5.8 on winXP has a problem with file descripter
tables. It fails to properly handle parents STDOUT when I dup2 in
the child. Here is my script that runs fine on my FreeBSD box. It
properly redirects childs STDOUT to a file , while leaving parents
STDOUT to the screen. On an XP box, it redirects both parents and
childs STDOUT to a file. Any ideas on how to work around it ??

In the hope that someone with Win32 access and experience can respond to
this ticket, I have cleaned up the original poster's program somewhat
and am attaching it as 21776.pl. I'm also attaching a simple count.pl
program to serve as the argument for the 'exec' call. On Darwin, the
program appears to execute as the OP intended.

(And, should someone be curious​: No, it didn't get a response when it
was cross-posted on beginners@​perl.org 8-1/2 years ago.)

Thank you very much.
Jim Keenan

From @jkeenan

count.pl

From @jkeenan

21776.pl

The RT System itself - Status changed from 'new' to 'open'

From @tonycoz

On Thu, Dec 01, 2011 at 05​:00​:10PM -0800, James E Keenan via RT wrote​:

On Mon Mar 31 12​:58​:25 2003, mgoland wrote​:

Hello guys,

I think that perl 5.8 on winXP has a problem with file descripter
tables. It fails to properly handle parents STDOUT when I dup2 in
the child. Here is my script that runs fine on my FreeBSD box. It
properly redirects childs STDOUT to a file , while leaving parents
STDOUT to the screen. On an XP box, it redirects both parents and
childs STDOUT to a file. Any ideas on how to work around it ??

In the hope that someone with Win32 access and experience can respond to
this ticket, I have cleaned up the original poster's program somewhat
and am attaching it as 21776.pl. I'm also attaching a simple count.pl
program to serve as the argument for the 'exec' call. On Darwin, the
program appears to execute as the OP intended.

(And, should someone be curious​: No, it didn't get a response when it
was cross-posted on beginners@​perl.org 8-1/2 years ago.)

I suspect what's happening is the *process* stdout is being modified -
fork is emulated on Win32 using threads, so when the child code
modifies stdout it's modified for both the main thread and the child
thread.

C​:\Users\tony>perl -MPOSIX=dup2 -le "if (fork) { print STDERR 'before ', -t STDO
UT; sleep 1; print STDERR 'after ', -t STDOUT; } else { open my $f, '>', 'test.t
xt'; dup2(fileno($f), fileno(*STDOUT)) || die; }"
before 1
after

The workaround is probably IPC​::Open[23].

Tony

@tonycoz isn't the advice to not fork on windows then?

No. it is use IPC​::Open[23].

Ideally we'd provide something vaguely like posix_spawn(), though it would really need to be dual-life (there's POSIX::RT::Spawn but that's very limited, and broken)