refactor the window attributes and signal code

[Antoine Martin]

As it's a bit of a mess...

In particular:

• keep in mind #41..
• try to reduce the duplication between the window models, WM and the server's view
• cleaner abstraction of the "window model" interface
• remove redundant attributes (is_OR vs "override-redirect" property, is_shadow, etc..)
• move some wm-state attributes so OR windows can use them (at least focus?)
• more consistent property change code
• fixup geometry issues (and try to get rid of the ownership election stuff)
• make it less dependent on gtk2 whilst we're at it
• maybe move all _handle_XXXX methods to a util class?
• generalize NET_WM_ALLOWED_ACTIONS?
• maybe consider the eventual move to GTK3? meh, if so we will need to replace add_event_receiver / remove_event_receiver with a WeakKeyDictionary
• maybe a new form of _internal_set_property can check if the value has actually changed? (no need to send a notify() if it hasn't)

List of API calls to deal with:

• get_default_window_icon
• _guess_window_type
• do_get_property_can_focus just update the "can-focus" property instead
• x11_get_server_time - rewrite in Cython
• read_initial_properties and X11 properties, it's a mess
• call_setup() is inconsistent, it's called for OR and tray, but we call it for "normal" windows
• WM_PROTOCOLS -> XGetWMProtocols
• WM_CLIENT_MACHINE -> XGetWMClientMachine
• attention-requested: shouldn't we clear this when "urgency" is not in the wm_hints?
• set_active - maybe this should be a method on WM instead?

[Antoine Martin]

better patch

[Antoine Martin]

much cleaner patch

[Antoine Martin]

Huge changeset in r9822 - see commit message.
If needed, it should be possible to revert those changes easily enough for testing.. so far.

Still TODO:

• WM_STATE and withdrawn windows - we only handle "iconic" (should we forward "withdrawn" to client? and emulate it as a destroy + new for older clients?)
• geometry (big!) - see also #846, #988 and many others
• "attention-request"
• maybe too many messages are now handled in the window base class (not relevant to OR windows?)
• should use "managed_connect" more
• seem to be getting double notifications for a single notify() call!?
• "can-focus" is not used?
• #846
• #881?

[Antoine Martin]

Moving the geometry bits to #990 (the rest is in pretty good shape already)

[Antoine Martin]

Note: on centos 7.x, we get tons of warnings like these:

/usr/lib64/python2.7/site-packages/gobject/__init__.py:115: Warning: Attempt to add property xpra+x11+gtk2+models+or_window+OverrideRedirectWindowModel::size-hints after class was initialised
  type_register(cls, namespace.get('__gtype_name__'))
/usr/lib64/python2.7/site-packages/gobject/__init__.py:115: Warning: Attempt to add property xpra+x11+gtk2+models+or_window+OverrideRedirectWindowModel::frame after class was initialised
  type_register(cls, namespace.get('__gtype_name__'))

Which is caused by this bug: ​https://bugzilla.gnome.org/show_bug.cgi?id=698614 (hopefully will be backported to centos 7.x soon).

AFAICT, it is not possible to workaround this warning with pygobject (see the two attempts at subclassing attached below).

[Antoine Martin]

example triggering the centos 7 warning

[Antoine Martin]

another example triggering the centos 7 warning

[Antoine Martin]

Lots more work on this done as part of #990 - this will do for this release.

Will follow up in #1034