We are adding the x value twice (from the allocation and from the origin)
in get_rect of the WidgetInvoker which we derive from in the FrameInvoker.
In the toolkit-gtk2 code [1] we did add the allocation.x value when
the widget does not provide its own gtk.gdk.Window (gtk.NO_WINDOW) [2]. This
is the same check we do above if the widget has a window and we set x and y
to 0 there which sounds sane enough to me.
[1] d1f68419e7/src/sugar/graphics/palettewindow.py (line716)
[2] http://www.pygtk.org/docs/pygtk/class-gtkobject.html#method-gtkobject--flags
Signed-off-by: Simon Schampijer <simon@laptop.org>
Acked-by: Manuel Quiñones <manuq@laptop.org>
- the GtkCellRenderer (which our CellRenderer derives from) is not a
GtkWidget and therefor the 'get_display' method does not exist, we
fallback to the default display in that case [1]
- the get_rect method should return a Gdk.Rectangle now, see other
invokers
- check if event.mode is Gdk.CrossingMode.NORMAL to trigger a mouse
leave see 289787e8c6 for a similar
case
- todo: the Palette does not go away when the mouse leaves the
widget
[1] http://developer.gnome.org/gtk3/3.4/GtkCellRenderer.html
Signed-off-by: Simon Schampijer <simon@laptop.org>
Acked-by: Manuel Quiñones <manuq@laptop.org>
Like in the WidgetInvoker we check now first if the leave
event has the mode Gdk.CrossingMode.NORMAL, only then we trigger
a mouse leave to popdown the Palette. This stops the Palette
to appear/disappear in a loop.
Signed-off-by: Simon Schampijer <simon@laptop.org>
The _ToolbarPalette does get passed the invoker on
initialisation. But we do create the PaletteWindowWidget
later. We do attach the invoker to the widget when calling
_setup_widget that is why it was still working without that
patch.
This patch prevents the traceback that we had because of not
having a widget at this point.
Signed-off-by: Simon Schampijer <simon@laptop.org>
Tested-by: Manuel Quiñones <manuq@laptop.org>
set_data/get_data not available anymore is not available anymore [1]. The
recommended approach is using a python attribute, which we do.
Changing to use the attribute in the activity class slipped by mistake
into 6330204e91.
[1] https://bugzilla.gnome.org/show_bug.cgi?id=641944
Signed-off-by: Simon Schampijer <simon@laptop.org>
Tested-by: Manuel Quiñones <manuq@laptop.org>
The accelerator in the primary information in the Palette
has not been drawn because there was not enough space
reserved for it. The preferred size we get back for the
Palette window does not include the accelerator of the
Gtk.AccelLabel. We need to include that in our calculation for
the Palette size.
In order to make that information available which is part
of the Palette class we need to pass the instance to the
PaletteWindowWidget instance.
Signed-off-by: Simon Schampijer <simon@laptop.org>
Acked-by: Daniel Drake <dsd@laptop.org>
First we needed to port the Palette code to use a minimum size. The default size
is two times the GRID_CELL_SIZE. Since the request-phase of the traditional GTK+
geometry management has been replaced by a height-for-width system [1] we have
to compensate for that. Furthermore we need to pass the invoker from the
PaletteWindow to the _PaletteWindowWidget for the gap calculation code for
drawing the border around the Palette.
We do the drawing of the border for the toolbutton in the base class, moved
this from the ToolbarButton and made sure we are drawing in the right order.
In the ToolButton we draw as well a black background for the ToolButton when
the Palette is up. While the mouse is over the button we can do that in the
theme, but not when the mouse moves over the Palette.
[1] http://developer.gnome.org/gtk3/3.0/ch25s02.html#id1525688
Signed-off-by: Simon Schampijer <simon@laptop.org>
Moving from GTK2 to GTK3 has presented various challenges regarding
palettes.
In GTK2, we were able to access some internal API of the GtkMenu class
and use it to embed a GtkMenu in a regular window. As of GTK3, that API
has become private and we can no longer access it.
We still want to use GtkMenu for the advanced functionality it provides
(multiple-level menus, keyboard navigation, etc), but we are now limited
to popping it up with its own (internal) window, rather than being able
to pack it into one of our own.
Our palettes can historically be used either as a menu, or as a general
area where widgets can be added, or both. The new restrictions upon
GtkMenu force some changes here, but we work hard to stick to the old
API as far as possible.
A Palette instance now acts as a controller of either a "window widget"
(where any type of widget can be displayed as usual) or a "menu widget"
which just pops up a GtkMenu. A Palette defaults to the window mode, but
dynamically switches to menu mode if/when the user attempts to access
the menu element.
As a result of this, palettes can now pack either a user-defined collection
of widgets, or a menu, but types can no longer be mixed. This should
only affect a handful of palettes which will need to pick a single
approach and convert to it.
Some further challenges are presented by the fact that GtkMenu performs a
grab on the whole screen, meaning that all input events are delivered to
the GtkMenu widget. Through some careful event filtering and examination
of the mouse cursor position we are still able to determine when the mouse
has entered or left the invoker or menu areas.
This work is authored by Benjamin Berg, Marco Pesenti Gritti, Simon
Schampijer and Daniel Drake.
CanvasIcon and CanvasInvoker were removed in a previous GTK3-porting commit
as they were based on hippocanvas.
However, this leaves the toolkit with some missing functionality:
there is no longer a trivial way to show an icon which can receive mouse
events and pop up a palette. Such functionality is used in various
places throughout the shell and activities.
Reimplement this functionality as EventIcon and CursorInvoker.
Instead of reimplementing much of the Icon class (like CanvasIcon did),
EventIcon opts for a more simplistic encapsulation of an Icon object.
This means trivial API changes for CanvasIcon users who must now
use the 'icon' property with the Icon API.
Signed-off-by: Daniel Drake <dsd@laptop.org>
Acked-by: Simon Schampijer <simon@laptop.org>
The following data fields that were provided by PyGTK are not accessible
directly in GTK3+pygi and need to be replaced by accessor calls:
Adjustment.lower
Adjustment.page_size
Adjustment.upper
Adjustment.value
Bin.child
Widget.parent
Widget.style
Widget.window
Based on patches by Daniel Drake <dsd@laptop.org>.
Signed-off-by: Sascha Silbe <silbe@activitycentral.com>
This is only on a best-effort basis; the code will be in a broken state after
this patch and need to be fixed manually.
The purpose of committing the intermediate, non-working output is to make it
reproducible. It's impractical to manually review the changes.
The exact version used was 4f637212f13b197a95c824967a58496b9e3b877c from the
main pygobject repository [1] plus a custom patch [2] that hasn't been sent
upstream yet.
[1] git://git.gnome.org/pygobject
[2] https://sascha.silbe.org/patches/pygobject-convert-sugar-20111122.patch
Signed-off-by: Sascha Silbe <silbe@activitycentral.com>
Some parts of GTK moved to GDK, so we need to import the latter for things to
work after the conversion script runs.
Signed-off-by: Sascha Silbe <silbe@activitycentral.com>
hippo-canvas isn't available in the GTK3 world, so we need to remove
anything that depends on it. Activities that still use it will need replace
hippo-canvas based widgets with native GTK ones before they can be ported to
GTK3.
[replaced description]
Signed-off-by: Sascha Silbe <silbe@activitycentral.com>
The old gtk-2 based module will be present in
the 0.94 branch in the sugar-toolkit.
Signed-off-by: Simon Schampijer <simon@laptop.org>
Acked-by: Sascha Silbe <silbe@activitycentral.com>
