469756 - GDK_EXTENSION_EVENTS_CURSOR option is not enabled for gdk window

Status: RESOLVED WONTFIX

(Core :: Widget: Gtk, defect)

Description

[Oleg Romashin (:romaxa)]

Attachment: Enable gdk extensions for inner window

After some changes in bug 428988, small part related to enabling GDK EXTENSIONS was missing.

We need to enable GDK_EXTENSION_EVENTS_ALL to have pressure events generation.

Comment 1

[Oleg Romashin (:romaxa)]

Pushed in:
http://hg.mozilla.org/mozilla-central/rev/0c0bf7bd8e7b

Comment 2

[Joe Drew (not getting mail)]

This was backed out to see if it caused bug 469809.

Comment 3

[Joe Drew (not getting mail)]

Preliminary results are that this backout did fix bug 469809.

Comment 4

[Oleg Romashin (:romaxa)]

Attachment: change EVENTS_ALL -> GDK_EXTENSION_EVENTS_CURSOR

Seems GDK_EXTENSION_EVENTS_ALL enable also other extensions which is affect on performance....
For pressure events generation enough to have GDK_EXTENSION_EVENTS_CURSOR mode.
Can we try again with CURSOR mode?

Comment 5

[Oleg Romashin (:romaxa)]

Not sure but may be it also reasonable to add some preference which will enable and disable that modes...

Comment 6

[Oleg Romashin (:romaxa)]

Pushed in:
http://hg.mozilla.org/mozilla-central/rev/bdef726fb656

Comment 7

[Peter Van der Beken [:peterv]]

Backed out again (Twinopen jumped). Next time please test first on try server and be around after checkin.

Comment 8

[Oleg Romashin (:romaxa)]

Sadly, but it seems jumping always when any gdk_window extensions enabled, for any events...
roc do you have any ideas is it gdk problem or moz gtk widgets implementation?

Comment 9

[Oleg Romashin (:romaxa)]

On maemo it is also the same problem topen jump ~10% :(

Comment 10

[Karl Tomlinson (:karlt)]

The description of GDK Input Devices says

"Because handling extended input devices may involve considerable overhead,
 they need to be turned on for each GdkWindow individually using
 gdk_input_set_extension_events()."

I had a quick look at the GDK Input Device code and saw that enabling
extension events would cause these functions that require X server round trips
to be used perhaps frequently from the callers indicated.

XTranslateCoordinates
 _gdk_input_get_root_relative_geometry
   gdk_input_set_extension_events
   _gdk_input_crossing_event if enter && extension_events
     send_crossing_event
       _gdk_synthesize_crossing_events in enter/leave situations (geometry change)
       proxy_pointer_event in some leave and enter events
         _gdk_windowing_got_event if is_motion_type(event->type)
   _gdk_input_configure_event if extension_events
     gdk_event_translate with ConfigureNotify events

XQueryDeviceState
  gdk_input_check_proximity for each enabled non-core device until one is found with InProximity
    _gdk_input_crossing_event if enter && extension_events
    _gdk_input_other_event with any input extension event
       gdk_event_translate
  gdk_device_get_state if !core
     gdk_event_request_motions if GDK_MOTION_NOTIFY && is_hint

These may or may not be the cause of the performance hit, but it is not
surprising that performance is affected.

In order to enable extension events, I suspect we would need to be very
selective about when they are enabled and/or have them controlled by a pref,
off by default - or do some performance work on the GDK code.

BTW, it seems that GTK needs to be configured --with-xinput for extension events.  That doesn't seem to be the default though distros may turn that on.

Comment 11

[BugBot [:suhaib / :marco/ :calixte]]

The bug assignee didn't login in Bugzilla in the last 7 months.
:stransky, could you have a look please?
For more information, please visit auto_nag documentation.

Comment 12

[Martin Stránský [:stransky] (ni? me)]

AFAIK outdated. Please reopen if you want to work on that.