Date: Sat, 12 Nov 2011 10:41:28 -0500
From: Austin Clements <amdragon@MIT.EDU>
To: Tomi Ollila <tomi.ollila@iki.fi>
Subject: Re: [PATCH] test: attempt to send QUIT to smtp-dummy in case mail
	send failed
Message-ID: <20111112154128.GB2658@mit.edu>
References: <yf6ipmr7dki.fsf@taco2.nixu.fi> <20111112024941.GY2658@mit.edu>
	<yf662iplkv3.fsf@taco2.nixu.fi>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <yf662iplkv3.fsf@taco2.nixu.fi>
User-Agent: Mutt/1.5.21 (2010-09-15)
Cc: notmuch@notmuchmail.org
Precedence: list

Quoth Tomi Ollila on Nov 12 at 12:02 pm:
> On Fri, 11 Nov 2011 21:49:41 -0500, Austin Clements <amdragon@MIT.EDU> wrote:
> > Good idea, but this introduces a dependency on netcat.  Would it work
> > to simply kill the SMTP dummy?
> 
> I cannot guarantee 6 nines of probability (but can guarantee as many eights
> as one migh wish ;)
> 
> Socket buffering and timing might cause smtp dummy to be killed before
> it has read all content...
> 
> But... i recall bash has a way to do tcp stuff .../dev/tcp/host/port
> is mentioned in manual page.
> I'll investigate this later (or, anyone of you, provide a patch :)

Actually, since sending mail is synchronous, there shouldn't be any
issues with buffering or timing.  If Emacs successfully sends the
message, it will wait for the OK response (`smtpmail-via-smtp'), which
should indicate that the message is in stable storage on the SMTP
server (this is rather fundamental to SMTP's reliability as a
protocol, after all).  Thus, the smtp-dummy *should* be done by the
time Emacs exits.  If Emacs fails to send the message, then it doesn't
really matter and you just want the smtp-dummy to go away.

I say "should" because smtp-dummy needs a one line patch to add an
fflush at the end of receive_data_to_file.  Any real SMTP server would
do this (not to mention a full fsync) before acknowledging the
message.