GtkWindow now requests the minimum width for the natural height
of its child, or the opposite; depending on ->get_height_for_width.
Currently its not done for GTK_WINDOW_POPUP windows as the menushells
aren't working right yet.
This patch adds another commented algorythm to find the collective
minimum and natural height for a said width of a horizontally oriented
box (or the opposite). The algorithm works quite well and can be optimized
a bit more - currently its commented because GtkLabel cannot effectively
do width-for-height calculations (doh).
Further, this patch returns an is_height_for_width() preference depending
on the boxes orientation (vertical boxes are width-for-height).
Added an indicator telling whether a widget prefers to be allocated
as height-for-width or width-for-height. Usually this depends on the
orientation of a container or the nature of a content widget like GtkLabel.
This indicator is only used in the seldom case where a parent is allocating
free space to the child and the child can flow in either direction, GtkWindow
and GtkScrolledWindow are users of this api.
In order for natural size information to cascade correctly up through
the ancestry GtkBox needs to report height-for-width and width-for-height,
this patch includes an implementation for both in both orientations, one
of them is commented for now as its much too cpu intensive to actually use.
Merged in fixes from the old branch in a patch prepared by Matthias Clasen,
added some fixes of my own to make sure that label wrapping follows allocation
and not requisition at show time (allocate time).
Instead of implementing ->get_desired_size() on GtkBin, which
cant really be done because border widths are in the domain of
the concrete subclasses; here we implement only the get_height_for_width
and get_width_for_height apis - GtkBin subclasses whom might have a
variable border width depending on allocations need to write their
own height-for-width implementations.
Fixed gtksizegroup.c:do_size_request() to never force an initial requisition
as some widgets expect it to remain unchanged across resizes (GtkImage
with pixbufs/filenames assigned is one of these cases).
gtk_box_size_allocate() was forgetting to fetch the minimum
size for children when allocating in homogeneous mode and then
accessing the uninitialized allocated values in that case, fixed.
With this commit it is possible to use ellipsizing text
in the combobox's cell renderers and have them desire to
expand to natural size when placed in a GtkBox.
Ever since size groups gtk_widget_size_request() can be used in place
of gtk_widget_get_child_requisition(), deprecating both now in favor
of gtk_extended_layout_get_desired_size().
collapse get_fast_child_requisition && get_fast_natural_size into
a single get_fast_size() function which uses the possibly hard coded
minimum request as a minimum value for the returned desired size
(this fixes cases where widgets unwarily return height-for-width
with minimum size > natural size).
Added gtk_cell_view_get_desired_size_of_row() to report full
extended layout information for a said row, this obsoletes
gtk_cell_view_get_size_of_row(). Also make GtkCellView queue
a resize on itself when cell renderers are added/removed.
GtkAlignment is already using the height for width api to align
the child in the cases where space is available, but also needed
to implement the extended layout in order to reflect natural sizes
to parent widgets.
Added a new interface to extend layout information of GtkCellRenderers,
GtkCellRenderer is a prerequisite of GtkExtendedCellIface and provides
a base implementation of ->get_desired_size() with ->get_size().
Note that the api for GtkExtendedCell differs from GtkExtendedLayout
since the cells need to be passed an additional widget to calculate
sizes contextual to the widget/screen (fonts etc).
As it is not allowed to called gtk_extended_layout_get_desired_size()
on a GtkWidget directly; gtk_widget_get_desired_size() was really an
ambiguous api. This patch removes the added GtkWidget api, calls
the appropriate sizegroup code which in turn envokes the interface
vtable, this patch also accordingly makes GtkWidget a prerequisite
of GtkExtendedLayout (the api doesnt work for cell renderers anyway,
patch comming...).
This commit fixes the loop to take into consideration the
child->padding that will be allocated to the child while calculating
full available size. Additionally it fixes the initial positioning
of child widgets when packed at the end (a special case because the
x position used starts from the end of the box).
When doing size-requests and get_natural_size() calls, use 0 as
the base value instead of -1, because size_request code assumes
that widget->requesition is initialized at 0.
Because "size-request" signal can be handled by signal connections as well
as by third party derived classes it is necessary to always fire the
"size-request" signal for every recalculation of the widget requisition,
now gtksizegroup.c:do_size_request() fires the signal first and allows it
to overflow the minimum/natural sizes returned by extended layout
implementations.
GtkWidget->get_natural_size() is now left assigning default -1 values to
ensure they both get overridden by the size-request handling.
Getting the desired size of a GtkWidget must always be done with
gtk_widget_get_desired_size() and never with
gtk_extended_layout_get_desired_size() directly as the former passes
through size group logic and updates the widget->requisition cache.
Removed the size_request() handling on the class and calculate
minimum size from get_desired_size(); base the natural width on
either the minimum of column headers, or a greater natural request
of the columns renderers.
Applied patch to add height-for-width/width-for-height suppot
to GtkScrolledWindow plus modifications to use gtk_widget_get_desired_size()
to query children sized instead of directly calling the layout api, plus
removed the ->size_request() implementation interfering with the mess.
We should not unref the model here, it might not even exist anymore.
Instead check if it exists and only use it if it does.
The unref was leftover from a previous fix in
ba9f53397f.
Spotted by Matthias Clasen in
https://bugzilla.gnome.org/show_bug.cgi?id=614099
