GNU bug report logs - #31901
Incorrect make-network-process + nowait state handling for non-existing unix sockets in emacs-26.1

Previous Next

Full log

Message #49 received at 31901 <at> debbugs.gnu.org (full text, mbox):

From: Mike Kazantsev <mk.fraggod <at> gmail.com>
To: Eli Zaretskii <eliz <at> gnu.org>
Cc: Lars Ingebrigtsen <larsi <at> gnus.org>, npostavs <at> gmail.com,
 31901 <at> debbugs.gnu.org
Subject: Re: bug#31901: Incorrect make-network-process + nowait state
 handling for non-existing unix sockets in emacs-26.1
Date: Sun, 22 Jul 2018 21:42:07 +0500

On Sun, 22 Jul 2018 18:56:31 +0300
Eli Zaretskii <eliz <at> gnu.org> wrote:

> > From: Lars Ingebrigtsen <larsi <at> gnus.org>
> > Cc: Eli Zaretskii <eliz <at> gnu.org>,  npostavs <at> gmail.com,  31901 <at> debbugs.gnu.org
> > Date: Sun, 22 Jul 2018 17:23:45 +0200
> > 
> > Mike Kazantsev <mk.fraggod <at> gmail.com> writes:
> >   
> > > Given either of these guidelines, it'd be more surprising to not get
> > > error state in the sentinel call with ":nowait t" in emacs-26.  
> > 
> > I agree, it would be better to get a sentinel callback.  But I don't
> > think that's how it's ever worked, so that's not a regression.  But it
> > could be a new feature for Emacs 27...  
> 
> Yes, I think we should just fix the bug on emacs-26, and introduce the
> sentinel callback on master.

I just built emacs from git to test how it works, and have to report
that unfortunately current behavior is not consistent with what
emacs had either before this bug was introduced.

More details follow.


== How it worked in emacs-25:

  (make-network-process
    ...
    :family 'local
    :service "/tmp/does-not-exist.sock"
    :nowait t
    :sentinel #'unix-socket-test-func)

  => nil
  [sentinel never called]

  (process-live-p socket) => nil (because "socket" will be nil)
  (process-status socket) => error

I.e. nil was returned to indicate failure to create network process.


== How it works in emacs-26.1 (with the bug in question):

  (make-network-process
    ...
    :family 'local
    :service "/tmp/does-not-exist.sock"
    :nowait t
    :sentinel #'unix-socket-test-func)

  => #<process unix-socket-test>
  [sentinel never called]

  (process-live-p socket) => [non-nil value]
  (process-status socket) => 'open

This is the case described in the original bug report.


== How it works in emacs-26.1 with ":nowail nil":

  (make-network-process
    ...
    :family 'local
    :service "/tmp/does-not-exist.sock"
    :nowait nil
    :sentinel #'unix-socket-test-func)

  => [file-missing signaled]
  [sentinel never called]

  (process-live-p socket) => [no socket value]
  (process-status socket) => [no socket value]

Provided here to illustrate the difference with current ":nowait t"
case.


== How it works in current emacs-git [8f3bca3ad51]:

  (make-network-process
    ...
    :family 'local
    :service "/tmp/does-not-exist.sock"
    :nowait t
    :sentinel #'unix-socket-test-func)

  => #<process unix-socket-test>
  [sentinel never called]

  (process-live-p socket) => nil
  (process-status socket) => 'file-missing

Note that this is:

- Not consistent with what emacs did before this bug.

- Not consistent with what emacs-26 does with ":nowait nil".

- Does not seem to be consistent with what documentation describes
  (as indicated in this thread earlier, might be subjective).


So wanted to note that as it is, entirely new unexpected behavior was
introduced, which does not line up neither with how this worked
previously (and will certainly break the code there), nor how you'd
expect it to work after emacs-26 changes (either signal error same as
":nowait nil" does, or via sentinel).

As far as I can tell, this applies to both missing unix socket and when
it exists, but connection to it is refused (nothing listening there).


-- 
Mike Kazantsev // fraggod.net

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.