Date: Mon, 27 Jun 2011 23:49:37 -0400
From: Austin Clements <amdragon@MIT.EDU>
To: Dmitry Kurochkin <dmitry.kurochkin@gmail.com>
Subject: Re: [PATCH 2/2] test: use emacsclient(1) for Emacs tests
Message-ID: <20110628034937.GB4120@mit.edu>
References: <1309132379-24089-1-git-send-email-dmitry.kurochkin@gmail.com>
	<1309146846-30991-1-git-send-email-dmitry.kurochkin@gmail.com>
	<1309146846-30991-2-git-send-email-dmitry.kurochkin@gmail.com>
	<BANLkTi=nwHRndz367K5oh4=tPaU-KNeZ8Q@mail.gmail.com>
	<87hb7byqge.fsf@gmail.com> <87d3hyzs1c.fsf@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <87d3hyzs1c.fsf@gmail.com>
User-Agent: Mutt/1.5.20 (2009-06-14)
Cc: notmuch@notmuchmail.org
Precedence: list

Quoth Dmitry Kurochkin on Jun 28 at  5:03 am:
> > >  The only way I know to
> > > reliably kill a child process is to open a pipe to it and have it exit
> > > on its own when it reads EOF.  Unfortunately, I couldn't find a way to
> > > do this with an emacs daemon (it appears daemon mode aggressively
> > > cleans up things like pipes), but here's a different approach:
> > > 
> > >     coproc emacs --batch --eval "(while t (eval (read)))"
> > >     EMACSFD=${COPROC[1]}
> > >     trap "echo '(kill-emacs)' >&$EMACSFD" EXIT
> > > 
> > >     echo '(message "Hi")' >&$EMACSFD
> > >     # ...
> > > 
> > > This is, basically, a poor man's emacs server, but the coprocess pipe
> > > binds it tightly to the shell.  If the shell exits for *any* reason,
> > > the pipe will be closed by the kernel, emacs will read an EOF, and
> > > exit.
> > 
> > I like this idea.
> > 
> > >  The trap is there just to cleanly shut down in case of a normal
> > > exit [1].
> > 
> > For normal exit we should just put this into test_done.  Otherwise it is
> > not a normal exit and we do not care about Emacs error message.  No?
> > 
> > >  This also has the advantage that read-from-minibuffer still
> > > works:
> > > 
> > >     echo '(message (read-from-minibuffer ""))' >&$EMACSFD
> > >     echo 'Test' >&$EMACSFD
> > > 
> > > Thoughts?
> > > 
> > 
> > I like it and I will implement it.  Thanks for the idea.
> > 
> 
> While implementing the idea, I stumbled upon a problem: we need to know
> when Emacs finished what we echoed or failed with an error.  At the
> moment tests fail because they check for OUTPUT before Emacs creates it.
> 
> We can tell Emacs to print some special marker and wait for it.  But
> there may be exceptions and errors which may make it difficult.  I did
> not found a good solution yet.  Would love to hear your thoughts :)

Oof, yes, of course.  How about making the one-line poor man's emacs
server slightly less poor?  Use a FIFO to communicate completion.
Something like,

EMACSDONE=$TEST_DIRECTORY/emacsdone
mkfifo $EMACSDONE
coproc emacs --batch --eval '(while t (eval (read)) (write-region "\n" nil "'$EMACSDONE'" t 0))'
EMACSFD=${COPROC[1]}

test_emacs() {
    echo "$1" >&$EMACSFD
    read <$EMACSDONE
}

test_emacs '(sleep-for 2)'
test_emacs '(message "Hi")'

echo '(kill-emacs)' >&$EMACSFD