package IPC::Run::Win32Pump;
=pod
=head1 NAME
IPC::Run::Win32Pump - helper processes to shovel data to/from parent, child
=head1 SYNOPSIS
Internal use only; see IPC::Run::Win32IO and best of luck to you.
=head1 DESCRIPTION
See L<IPC::Run::Win32Helper|IPC::Run::Win32Helper> for details. This
module is used in subprocesses that are spawned to shovel data to/from
parent processes from/to their child processes. Where possible, pumps
are optimized away.
NOTE: This is not a real module: it's a script in module form, designed
to be run like
$^X -MIPC::Run::Win32Pumper -e 1 ...
It parses a bunch of command line parameters from IPC::Run::Win32IO.
=cut
use strict;
use vars qw{$VERSION};
BEGIN {
$VERSION = '20180523.0';
}
use Win32API::File qw(
OsFHandleOpen
);
my ( $stdin_fh, $stdout_fh, $debug_fh, $binmode, $parent_pid, $parent_start_time, $debug, $child_label );
BEGIN {
( $stdin_fh, $stdout_fh, $debug_fh, $binmode, $parent_pid, $parent_start_time, $debug, $child_label ) = @ARGV;
## Rather than letting IPC::Run::Debug export all-0 constants
## when not debugging, we do it manually in order to not even
## load IPC::Run::Debug.
if ($debug) {
eval "use IPC::Run::Debug qw( :default _debug_init ); 1;"
or die $@;
}
else {
eval <<STUBS_END or die $@;
sub _debug {}
sub _debug_init {}
sub _debugging() { 0 }
sub _debugging_data() { 0 }
sub _debugging_details() { 0 }
sub _debugging_gory_details() { 0 }
1;
STUBS_END
}
}
## For some reason these get created with binmode on. AAargh, gotta #### REMOVE
## do it by hand below. #### REMOVE
if ($debug) { #### REMOVE
close STDERR; #### REMOVE
OsFHandleOpen( \*STDERR, $debug_fh, "w" ) #### REMOVE
or print "$! opening STDERR as Win32 handle $debug_fh in pumper $$"; #### REMOVE
} #### REMOVE
close STDIN; #### REMOVE
OsFHandleOpen( \*STDIN, $stdin_fh, "r" ) #### REMOVE
or die "$! opening STDIN as Win32 handle $stdin_fh in pumper $$"; #### REMOVE
close STDOUT; #### REMOVE
OsFHandleOpen( \*STDOUT, $stdout_fh, "w" ) #### REMOVE
or die "$! opening STDOUT as Win32 handle $stdout_fh in pumper $$"; #### REMOVE
binmode STDIN;
binmode STDOUT;
$| = 1;
select STDERR;
$| = 1;
select STDOUT;
$child_label ||= "pump";
_debug_init(
$parent_pid,
$parent_start_time,
$debug,
fileno STDERR,
$child_label,
);
_debug "Entered" if _debugging_details;
# No need to close all fds; win32 doesn't seem to pass any on to us.
$| = 1;
my $buf;
my $total_count = 0;
while (1) {
my $count = sysread STDIN, $buf, 10_000;
last unless $count;
if (_debugging_gory_details) {
my $msg = "'$buf'";
substr( $msg, 100, -1 ) = '...' if length $msg > 100;
$msg =~ s/\n/\\n/g;
$msg =~ s/\r/\\r/g;
$msg =~ s/\t/\\t/g;
$msg =~ s/([\000-\037\177-\277])/sprintf "\0x%02x", ord $1/eg;
_debug sprintf( "%5d chars revc: ", $count ), $msg;
}
$total_count += $count;
$buf =~ s/\r//g unless $binmode;
if (_debugging_gory_details) {
my $msg = "'$buf'";
substr( $msg, 100, -1 ) = '...' if length $msg > 100;
$msg =~ s/\n/\\n/g;
$msg =~ s/\r/\\r/g;
$msg =~ s/\t/\\t/g;
$msg =~ s/([\000-\037\177-\277])/sprintf "\0x%02x", ord $1/eg;
_debug sprintf( "%5d chars sent: ", $count ), $msg;
}
print $buf;
}
_debug "Exiting, transferred $total_count chars" if _debugging_details;
## Perform a graceful socket shutdown. Windows defaults to SO_DONTLINGER,
## which should cause a "graceful shutdown in the background" on sockets.
## but that's only true if the process closes the socket manually, it
## seems; if the process exits and lets the OS clean up, the OS is not
## so kind. STDOUT is not always a socket, of course, but it won't hurt
## to close a pipe and may even help. With a closed source OS, who
## can tell?
##
## In any case, this close() is one of the main reasons we have helper
## processes; if the OS closed socket fds gracefully when an app exits,
## we'd just redirect the client directly to what is now the pump end
## of the socket. As it is, however, we need to let the client play with
## pipes, which don't have the abort-on-app-exit behavior, and then
## adapt to the sockets in the helper processes to allow the parent to
## select.
##
## Possible alternatives / improvements:
##
## 1) use helper threads instead of processes. I don't trust perl's threads
## as of 5.005 or 5.6 enough (which may be myopic of me).
##
## 2) figure out if/how to get at WaitForMultipleObjects() with pipe
## handles. May be able to take the Win32 handle and pass it to
## Win32::Event::wait_any, dunno.
##
## 3) Use Inline::C or a hand-tooled XS module to do helper threads.
## This would be faster than #1, but would require a ppm distro.
##
close STDOUT;
close STDERR;
1;
=pod
=head1 AUTHOR
Barries Slaymaker <barries@slaysys.com>. Funded by Perforce Software, Inc.
=head1 COPYRIGHT
Copyright 2001, Barrie Slaymaker, All Rights Reserved.
You may use this under the terms of either the GPL 2.0 ir the Artistic License.
=cut