We have been struggling for a week on an issue involving open3 in several modules that are run out of mod_perl. A rough version of the code is shown below, but it is important to note that it WORKS, at least for some people.

Each developer runs their own sandbox under RedHat Linux, comprising the code and apache. For some folks, these modules work fine. For others, the sysread always returns 0 bytes, with no warnings or errors. The difference is in who starts apache/mod_perl.

We became desperate enough wrap the called utility in strace, and the one clue we've found is a difference in fstat64 calls. The users whose modules work see:

fstat64(1, . . . st_mode=S_IFIFO . . .) = 0;
[download]

For users whose calls fail:

fstat64(1, . . . st_mode=S_IFSOCK . . .) = 0;
[download]

We've compared profiles, environment variables, and version numbers and see no differences.

In addition, users whose modules don't work from inside mod_perl can run standalone scripts like this with no problems.

Has anyone encountered anything like this? Thanks in advance.

use IPC::Open3;
use Symbol qw /gensym /;
use IO::File;
use IO::Select;

my $pid;
my ($in,$out,$err) = map gensym, 1..3;
eval {
  $pid = open3($in,$out,$err,"/usr/bin/<utility>",("/home/me/infile"))
+;
};
die "Bad error" if ($@);
my $sel = IO::Select->new();
$sel->add($out);
$sel->add($err);
while (my @ready = $sel->can_read(5)) {
  for my $handle (@ready) {
    my $buf;
    if ($handle == $out) {
      my $bytes = sysread($handle,$buf,1024);
      if ($bytes == 0) {
        $sel->remove($handle);
        next;
      }
      $results .= $buf;
    }
  }
}
if (length($results) > 0) {
  waitpid($pid);
}
else {
  kill(9,$pid);
}
[download]