read-event and unexpected buffer changes

read-event and unexpected buffer changes

[Julien Danjou]

[-- Attachment #1: Type: text/plain, Size: 1485 bytes --]

Hi there,

I'm having a nasty bug and just spent an hour tracking it down.

I'm using `erc-desktop-notifications' to get IRC messages sent with my
nickname over D-Bus using `notifications-notify'. This function calls
`dbus-call-method' which contains the following snippet at the end:

    (with-timeout ((if timeout (/ timeout 1000.0) 25))
      (while (eq (gethash key dbus-return-values-table :ignore) :ignore)
	(let ((event (let (unread-command-events) (read-event nil nil 0.1))))
	  (when (and event (not (ignore-errors (dbus-check-event event))))
	    (setq unread-command-events
		  (append unread-command-events (list event)))))))

This `read-event' call causes me trouble. Basically what happens, is
that it changes the ERC buffer in a very subtle way so that the last
line is then on top. Basically, the buffer is supposed to be shown as:

/-----
| <jd> ohla
| <foo> hey
| <foo> how are you jd?
| [#bar]
\-----

But because of this `read-event' call, the buffer is "recentered to the
top" and I finally see this instead:

/-----
| <foo> how are you jd?
| [#bar]
|
| 
\-----

Which is utterly inconvenient. I've tried adding `save-excursion' around
(read-event), but that doesn't change anything. I wonder what this
function can do that changes the buffer, and in any way, that doesn't
look like an expected and right side effect.

Hint?
-- 
Julien Danjou
# Free Software hacker # freelance consultant
# http://julien.danjou.info

[-- Attachment #2: Type: application/pgp-signature, Size: 835 bytes --]

Re: read-event and unexpected buffer changes

[Stefan Monnier]

> 	(let ((event (let (unread-command-events) (read-event nil nil 0.1))))
> 	  (when (and event (not (ignore-errors (dbus-check-event event))))
> 	    (setq unread-command-events
> 		  (append unread-command-events (list event)))))))

Why not use sit-for?

> But because of this `read-event' call, the buffer is "recentered to the
> top" and I finally see this instead:

The only expected side-effect is a redisplay (which could also trigger
things like jit-lock).

> Hint?

Sorry,


        Stefan

Re: read-event and unexpected buffer changes

[Michael Albinus]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

>> 	(let ((event (let (unread-command-events) (read-event nil nil 0.1))))
>> 	  (when (and event (not (ignore-errors (dbus-check-event event))))
>> 	    (setq unread-command-events
>> 		  (append unread-command-events (list event)))))))
>
> Why not use sit-for?

<http://debbugs.gnu.org/cgi/bugreport.cgi?bug=11447#26>

>         Stefan

Best regards, Michael.

DBus events processing (was: read-event and unexpected buffer changes)

[Stefan Monnier]

>>> (let ((event (let (unread-command-events) (read-event nil nil 0.1))))
>>> (when (and event (not (ignore-errors (dbus-check-event event))))
>>> (setq unread-command-events
>>> (append unread-command-events (list event)))))))
>> Why not use sit-for?
> <http://debbugs.gnu.org/cgi/bugreport.cgi?bug=11447#26>

Ah, right!
Thinking about this problem some more, I think the problem is that dbus
events should be processed during sit-for, just like process filters are
run there.

DBus events come concurrently with keyboard&mouse events and then we
forcefully linearize them into the event queue, which is not right in
your case where you want to process the dbus events before the user's
key presses.

So we should change the C code for our dbus interface.  It could either
be shoehorned into the way process filters/sentinels are handled, or we
could add a secondary event queue for "asynchronous events".


        Stefan

Re: DBus events processing

[Michael Albinus]

Stefan Monnier <monnier@IRO.UMontreal.CA> writes:

> So we should change the C code for our dbus interface.  It could either
> be shoehorned into the way process filters/sentinels are handled, or we
> could add a secondary event queue for "asynchronous events".

Gimme some days, I'll try to understand how it works with process events.

However, I don't know whether such a change would solve Julien's problem.

>         Stefan

Best regards, Michael.

Re: DBus events processing

[Stefan Monnier]

> However, I don't know whether such a change would solve Julien's problem.

It probably won't.  I just hijacked that part of your thread by focusing
on another part (hence the subject modification).


        Stefan

Re: read-event and unexpected buffer changes

[Michael Albinus]

Julien Danjou <julien@danjou.info> writes:

> Hi there,

Hi,

> I'm using `erc-desktop-notifications' to get IRC messages sent with my
> nickname over D-Bus using `notifications-notify'. This function calls
> `dbus-call-method' which contains the following snippet at the end:
>
>     (with-timeout ((if timeout (/ timeout 1000.0) 25))
>       (while (eq (gethash key dbus-return-values-table :ignore) :ignore)
> 	(let ((event (let (unread-command-events) (read-event nil nil 0.1))))
> 	  (when (and event (not (ignore-errors (dbus-check-event event))))
> 	    (setq unread-command-events
> 		  (append unread-command-events (list event)))))))
>
> This `read-event' call causes me trouble. Basically what happens, is
> that it changes the ERC buffer in a very subtle way so that the last
> line is then on top.
>
> Hint?

Does the following patch suffices?

--8<---------------cut here---------------start------------->8---
*** /home/albinmic/src/emacs/lisp/net/dbus.el.~112739~    2013-05-27 13:51:51.844960427 +0200
--- /home/albinmic/src/emacs/lisp/net/dbus.el          2013-05-27
13:51:33.992671583 +0200
***************
*** 270,276 ****
      ;; default 25".  Events which are not from D-Bus must be restored.
      (with-timeout ((if timeout (/ timeout 1000.0) 25))
        (while (eq (gethash key dbus-return-values-table :ignore) :ignore)
!       (let ((event (let (unread-command-events) (read-event nil nil 0.1))))
          (when (and event (not (ignore-errors (dbus-check-event event))))
            (setq unread-command-events
                  (append unread-command-events (list event)))))))
--- 270,277 ----
      ;; default 25".  Events which are not from D-Bus must be restored.
      (with-timeout ((if timeout (/ timeout 1000.0) 25))
        (while (eq (gethash key dbus-return-values-table :ignore) :ignore)
!       (let ((event (let ((inhibit-redisplay t) unread-command-events)
!                      (read-event nil nil 0.1))))
          (when (and event (not (ignore-errors (dbus-check-event event))))
            (setq unread-command-events
                  (append unread-command-events (list event)))))))
--8<---------------cut here---------------end--------------->8---

Best regards, Michael.

Re: read-event and unexpected buffer changes

[Julien Danjou]

[-- Attachment #1: Type: text/plain, Size: 269 bytes --]

On Mon, May 27 2013, Michael Albinus wrote:

Hi Michael,

> Does the following patch suffices?

Yes indeed, it fixes the problem as far as I can tell!

Thanks!

-- 
Julien Danjou
// Free Software hacker / freelance consultant
// http://julien.danjou.info

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 835 bytes --]

Re: read-event and unexpected buffer changes

[Michael Albinus]

Julien Danjou <julien@danjou.info> writes:

> Hi Michael,

Hi Julien,

>> Does the following patch suffices?
>
> Yes indeed, it fixes the problem as far as I can tell!

Thanks for confirmation. I don't use erc myself, so I couldn't check.

The patch is committed to the trunk. I'll continue to investigate
Stefan's proposal about changed D-Bus event handling.

> Thanks!

Best regards, Michael.