From: Eric Wong <e@80x24.org>
Date: Tue, 12 May 2026 22:31:49 +0000 (+0000)
Subject: listener: bless all accepted sockets
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=8cfeeff83a1db46998daff1eccc6580dbc446bc1;p=thirdparty%2Fpublic-inbox.git

listener: bless all accepted sockets

Newer IO::Socket::SSL (tested 2.098 on FreeBSD) seems to trigger
undefined variable warnings on TLS negotiation failures and
other problems.  Since I've been meaning to unify the {rbuf}
used by PublicInbox::DS with the {pi_io_rbuf} field anyways, it
seems best to bless these as PublicInbox::IO objects as early as
possible and avoid adding branches into in later stages.

I originally considered specialized changes to accept_tls_step
to stash the FD and ensure proper cleanup after IO::Socket::SSL
closes the socket behind our back, but it was too complex.
---

diff --git a/lib/PublicInbox/Listener.pm b/lib/PublicInbox/Listener.pm
index 364301d0f..a87581db5 100644
--- a/lib/PublicInbox/Listener.pm
+++ b/lib/PublicInbox/Listener.pm
@@ -34,6 +34,10 @@ sub event_step {
 	do {
 		if (my $addr = accept(my $c, $sock)) {
 			IO::Handle::blocking($c, 0); # no accept4 :<
+			# newer IO::Socket::SSL (tested 2.098) doesn't like
+			# unblessed sockets on TLS handshake failures;
+			# and plan to unify DS{rbuf} with {pi_io_rbuf}
+			bless $c, 'PublicInbox::IO';
 			eval { $self->{post_accept}->($c, $addr, $sock) };
 			warn "E: $@\n" if $@;
 		} elsif ($! == EAGAIN || $! == ECONNABORTED) {