[MAYBE] kill rpipe early

[Eric Wong <e@80x24.org>]

Eric Wong <e@80x24.org> wrote:
> Still haven't found anything which is causing problems on korg,
> but found a bunch of worthwhile cleanups, clarifications
> and perhaps minor efficiency improvements that I'm comfortable
> putting on 'master'.

This might help narrow down where the FD leak is if it stops
pipes (but not regular files) from leaking.  However, if it
fixes things, I also don't understand why it'd work, but I'm
mostly running on fumes at this point...

diff --git a/lib/PublicInbox/Qspawn.pm b/lib/PublicInbox/Qspawn.pm
index 5a30064..3bf5c03 100644
--- a/lib/PublicInbox/Qspawn.pm
+++ b/lib/PublicInbox/Qspawn.pm
@@ -224,7 +224,7 @@ sub filter_fh ($$) {
 #              immediately (or streamed via ->getline (pull-based)).
 sub psgi_return {
 	my ($self, $env, $limiter, $parse_hdr) = @_;
-	my ($fh, $rpipe);
+	my $fh;
 	my $end = sub {
 		my $err = $_[0]; # $!
 		log_err($env, "psgi_return: $err") if defined($err);
@@ -233,6 +233,7 @@ sub psgi_return {
 	};
 
 	my $buf = '';
+	my $rpipe;
 	my $rd_hdr = sub {
 		# typically used for reading CGI headers
 		# we must loop until EAGAIN for EPOLLET in HTTPD/Async.pm
@@ -275,6 +276,7 @@ sub psgi_return {
 			$fh = $wcb->($r); # scalar @$r == 2
 			$fh = filter_fh($fh, $filter) if $filter;
 			$async->async_pass($env->{'psgix.io'}, $fh, \$buf);
+			$rpipe = undef;
 		} else { # for synchronous PSGI servers
 			require PublicInbox::GetlineBody;
 			$r->[2] = PublicInbox::GetlineBody->new($rpipe, $end,