git-send-email and "mailhost" misbehavior

[Mike Frysinger <vapier@gentoo.org>]

i have sendemail in my ~/.gitconfig setup like so:
[sendemail]
    smtpserver = localhost:1111

and i have a ssh tunnel running there to forward my e-mail to another
machine's localhost:25

however, when i attempt to send e-mail from some systems, the `git
send-email` process pauses for like 2 minutes without doing anything.
after hassling my e-mail admin about greylisting (which is disabled
for localhost), i looked at git-send-email a bit closer.

running it through a heavy strace shows the source of the pause:
connect(6, {sa_family=AF_INET, sin_port=htons(25),
sin_addr=inet_addr("208.68.139.38")}, 16)
then there's a timeout after 120 seconds trying to connect to this guy

so wtf is "208.68.139.38" !?  well, my shitty ISP (comcast) is doing
DNS hijacking for unresolved DNS names.  so git-send-email tried to
look up some host and it got back "208.68.139.38".  while comcast is
wrong here, why is git-send-email looking up anything at all
considering my sendemail.smtpserver is configured to localhost ?

it seems that buried in the bowels of git-send-email and totally
undocumented is this nugget:
/usr/libexec/git-core/git-send-email:
...
sub maildomain_mta {
    my $maildomain;

    if (eval { require Net::SMTP; 1 }) {
        for my $host (qw(mailhost localhost)) {
            my $smtp = Net::SMTP->new($host);
...

so git-send-email is so kind as to attempt port 25 connections to
"mailhost:25" and "localhost:25" while attempting to locate its FQDN
!?  this doesnt sound right at all.  DNS lookups do not require actual
TCP/IP connection attempts.

stubbing out "mailhost" to 127.0.0.1 in /etc/hosts fixes my troubles.
-mike