This patch makes that work using 1 of 2 options:
1. Add all missing enums to the switch statement
or
2. Cast the switch argument to a uint to avoid having to do that (mostly
for GdkEventType).
I even found a bug while doing that: clearing a GtkImage with a surface
did not notify thae surface property.
The reason for enabling this flag even though it is tedious at times is
that it is very useful when adding values to an enum, because it makes
GTK immediately warn about all the switch statements where this enum is
relevant.
And I expect changes to enums to be frequent during the GTK4 development
cycle.
-Wint-conversion is important because it checks casts from ints to
pointers.
-Wdiscarded-qualifiers is important to catch cases where we don't
strings when we should.
Since setting a clip is mandatory for almost all widgets, we can as well
change the size-allocate signature to include a out_clip parameter, just
like GtkCssGadget did. And since we now always propagate baselines, we
might as well pass that one on to size-allocate.
This way we can also make sure to transform the clip returned from
size-allocate to parent-coordinates, i.e. the same coordinate space
priv->allocation is in.
It's not necessary anymore for clipping nor receiving events. So just
remove it. The event handling code was expecting events in bin_window
coordinates, and have been updated to relying on widget-relative coords.
We now rely on toplevels receiving and forwarding all the events
the windowing should be able to handle. Event masks are no longer a
way to determine whether an event is deliverable ot a widget.
Events will always be delivered in the three captured/target/bubbled
phases, widgets can now just attach GtkEventControllers and let those
handle the events.
gtk_snapshot_pop() => removed
gtk_snapshot_pop_and_append() => gtk_snapshot_pop()
So now there is no way to get a rendernode out of the snapshotting API
until you gtk_snapshot_finish().
We already take ints when setting the translation, so it can't
currently take any other values. Additionally, I was seeing large
costs in int -> double -> int for the rects in
gtk_snapshot_clips_rect(), as all callers really are ints (widget
allocations) and the clip region is int-based.
This change completely cleared a 2% rectangle_init_from_graphene from
the profile and is likely to have nice performance effects elsewhere
too.
Add a new ::measure vfunc similar to GtkCssGadget's that widget
implementations have to override instead of the old get_preferred_width,
get_preferred_height, get_preferred_width_for_height,
get_preferred_height_for_width and
get_preferred_height_and_baseline_for_width.
And with it, gtk_widget_get_visual() and gtk_widget_set_visual() are
gone.
We now always use the RGBA visual (if available) and otherwise fall back
to the system visual.
Use G_PARAM_DEPRECATED with deprecated style properties.
This will make it easier to identify and remove such stale
properties from css, since it will now trigger warnings.
These days exposure happens only on the native windows (generally the
toplevel window) and is propagated down recursively. The expose event
is only useful for backwards compat, and in fact, for double buffered
widgets we totally ignore the event (and non-double buffering breaks
on wayland).
So, by not setting the mask we avoid emitting these events and then
later ignoring them.
We still keep it on eventbox, fixed and layout as these are used
in weird ways that want backwards compat.
Just use the last coordinates given on XdndPosition/drag_motion() in
order to trigger scrolling.
When running on Xwayland, the pointer position is unknown at this
stage on the X11 side, so the coordinates given here are bogus.
This change avoids both roundtrips and this situation.
https://bugzilla.gnome.org/show_bug.cgi?id=749679
The rubberband rendering code was assuming that we just have
a 1-pixel border and the rest of the rubberband is uniform.
That is not a safe assumption to make with css-styled
rubberbands, so remove it.
When trying to drag, we currently the position of the first motion
event to determine where the drag came from. This might be alright
in the case of the old animation, but the data will be inaccurate
if the user has moved the pointer quite a bit since pressing the
cursor to start dragging. While we could monkey patch the GdkEvent
at the widget layer, this is unintuitive and strange.
Add a new API that takes a set of pointer coordinates describing
the origin of the drag. Additionally, adapt most widgets to use
it and use it with correct coordinates.
https://bugzilla.gnome.org/show_bug.cgi?id=705605
The cell out argument to gtk_icon_view_get_cursor is a pass-through
for gtk_cell_area_get_focus_cell which is transfer none. Without
this explicit annotation, transfer full is defaulted and introspection
bindings will assume ownership of the GtkIconViews reference to the
cell, causing crashes. Additionally add explicit transfer full to
the path parameter because it is expected the caller will free the
returned memory.
https://bugzilla.gnome.org/show_bug.cgi?id=704700
This replaces the previously hardcoded calls to gdk_window_set_user_data,
and also lets us track which windows are a part of a widget. Old code
should continue working as is, but new features that require the
windows may not work perfectly.
We need this for the transparent widget support to work, as we need
to specially mark the windows of child widgets.
https://bugzilla.gnome.org/show_bug.cgi?id=687842
Commit da09447914 removed the call to
gtk_style_context_set_background() in favour of always rendering it with
gtk_render_background() during the draw vfunc.
This has the side effect of making the backing window always
transparent, which blocks GTK from applying some optimizations during
the paint cycle. The result is that, especially in clutter-gtk
applications, scrolling performance gets really bad.
This commit partially reverts da09447914
and changes the code so that both gtk_style_context_set_background() and
gtk_render_background() are called.
This commit exposes the get_type() functions and standard
headers for accessible implementations. This makes it possible
to derive from the GTK accessible implementations without
GType magic tricks. This is necessary, because we require the
a11y type hierarchy to be parallel to the widget type hierarchy.
So, if you derive a widget and need to adjust its a11y implementation,
you have to be able to derive its accessible implementation.
This commit probably exposes more than is absolutely necessary,
it also exposes accessibles of widgets that are unlikely candidates
for deriving from.
This is needed for the SELECTION_NONE mode where nothing is ever
selected, but its also needed for CTRL-<key> keynav that moves the
focus without changing the selection.
https://bugzilla.gnome.org/show_bug.cgi?id=684984
This gets the current cell area of a particular item. Its similar
to gtk_tree_view_get_cell_area().
The code is extracted from gtk_icon_view_set_tooltip_cell which now
just calls the old code.
https://bugzilla.gnome.org/show_bug.cgi?id=678418
If the icon view is empty, we cannot get a reasonable size request from
the cell renderers. So all values we would compute are pretty much
useless.
So we special case it.
This also gets rid of a bunch of crashers from div-by-0 in corner cases.
https://bugzilla.gnome.org/show_bug.cgi?id=677809
This is useful for 2 reasons:
(1) Items actually exist and are clickable
(2) Size computations don't divide by 0
I've not seen problems with this in the wild (mostly because
item-padding defaults to non-0), but noticed this while fixing other
bugs.
Minimum size is necessary so you can see the item. If we can't get that
we need to scroll.
Natural as the maximum is used so that the spacing between items doesn't
increase when resizing the iconview, but empty space is added to the
right/bottom instead.
https://bugzilla.gnome.org/show_bug.cgi?id=677809
While working on the "iconview: Don't shrink items" patch I noticed that
gtk_icon_view_compute_n_items_for_size modifies the natural and minimum
item sizes it got from gtk_icon_view_get_preferred_item_size when
calculating the max number of items which will fit, but later on it
checks against these sizes when calculating the item_size, and these
checks expect these values to be unmodified.
This patch fixes this by modifying the natural and minimum values in
advance and doing all computations with modified values.
https://bugzilla.gnome.org/show_bug.cgi?id=677809
The previous code assumed that the width was always enough for more than
one column, which is obviously not correct when a number of columns is
hardcoded.
With this patch, it will now always check that the width is enough and
otherwise cause scrolling.
https://bugzilla.gnome.org/show_bug.cgi?id=677809
Code was using different places to compute the icon size when layouting
and when doing size requests. This resulted in non-matching behavior in
obscure cases. And that lead to iconviews that were too small.
This ensures that items stay left-aligned instead of slowly expanding into
empty space when widening the iconview. It's also what the iconview did
pre-refactoring.
Note that for cases where natural width != minimum width, the cells
might still expand and shrink back.
Always assume max-columns and min-rows. The old approach was kinda
insane.
As an example, try to write an algorithm that optimizes the minimum size
for infinite (take a reasonably large number like 2520) word-wrapped
Monospace text cells containing the text "XXXXX XXX XXX XXXXX" (keep in
mind that this is the easy problem, because it's assuming equal cell
renderers). There's 4 ways to reasonably lay out this text:
19 glyphs (19x1):
XXXXX XXX XXX XXXXX
18 glyphs (9x2):
XXXXX XXX
XXX XXXXX
21 glyphs (7x3):
XXXXX
XXX XXX
XXXXX
20 glyphs (5x4):
XXXXX
XXX
XXX
XXXXX
The best thing to do usually is using the 9x2 approach, but that's
neither the one using the natural nor the one using the minimum size.
As a side note, this does not include spacing and padding, which might
also influence the decision. Nor does it include height-for-width
considerations. Look at this table (numbers given in glyphs, not pixels,
as for pixel-sizes it gets even more interesting):
given best solution
width columns sizing glyphs per cell
6 1 6x4 20
7 1 7x3 21
8 1 7x3 24
9 1 9x2 18
10 1/2 9x2/5x4 20
11 1/2 9x2/5x4 22
12 1/2 9x2/5x4 24
13 1/2 9x2/5x4 26
14 2 7x3 21
15 3 5x4 20
16 3 5x4 21.3
17 3 5x4 22.7
18 2 9x2 18
19 1/2 19x1/8x2 19
20 1/2/4 19x1/8x2/5x4 20
21 1-4 any 21
22 1-4 any 22
23 1-4 any 23
24 1-4 any 24
25 5 5x4 20
26 5 5x4 20.8
27 3 9x2 18
28 3 9x2 18.7
29 3 9x2 19.3
30 3/6 9x2/5x4 20
Now of course, nobody wants the number of columns to randomly change in
inexplicable ways while they enlarge or shrink an iconview, so we not
only have to optimize for smallest or other size measurements, but we
also have to optimize for "most pleasing to the eye".
And last but not least, I'd like to once again remind you - if you kept
up until now - that this discussion was for identically-sized cells
only.
Instead of just returning the last allocated numbers, we now compute the
proper sizes from scratch. This is a bit less trivial, but it results in
proper height-for-width handling.
This is a huge quest to remove all caching from GtkIconview to simplify
the code. As it turns out, iconview performance is a joke, so the caches
are kinda unnecessary.
If we need caching, we can add it in a useful way later.
If we use
&item->cell_area
instead of
(GdkRectangle *) item
there is no need anymore to keep the cell_area as the first member of
the the item. And we cget compile-time checks for changes to the item
struct.
This can cause lagging when scrolling as it causes us to repaint
on every scroll event. This wasn't historically a great problem,
but with smooth scrolling we get a lot more events, so this
now creates visible lagging on slower machines.
The widget window is usually covered by the bin_window.
Its background color will become relevant when we introduce
kinetic scrolling with overshooting.
As the draw handler expects the items to be laid out already, we cannot
queue a layout here to avoid a race condition with the resize that is
queued immediately after, which in turn would lead to a segfault later
in the paint_item() implementation.
https://bugzilla.gnome.org/show_bug.cgi?id=663138
which are SHIFT and MOD2 on the Mac, and SHIFT and CONTROL otherwise.
Use the new define all over the place and rename variables and
members to not say "shift" or "control".
Render GtkIconView cell items with the prelight state flag when they're
being mouse hovered.
This works basically in the same way it's done for GtkTreeView cells,
and e.g. GtkCellRendererPixbuf will need to have its follow-state
property to opt in to prelight rendering.
https://bugzilla.gnome.org/show_bug.cgi?id=615501
The code sets old_adj_ptr to the location containing the old weak ref,
but then wants to remove a weak ref from &view->old_hadj, causing warnings
when disposing the widget.
This commit introduces a new setting, gtk-visible-focus, backed
by the Gtk/VisibleFocus X setting. Its three values control how
focus rectangles are displayed.
'always' is equivalent to the traditional GTK+ behaviour of always
rendering focus rectangles.
'never' does what it says, and is intended for keyboardless
situations, e.g. tablets.
'automatic' hides focus rectangles initially, until the user
interacts with the keyboard, at which point focus rectangles
become visible.
https://bugzilla.gnome.org/show_bug.cgi?id=649567
In this patch we adress rows_reordered() and row_deleted() callbacks
(since some icon view subclasses manipulate the connected treemodel
from _init()).
Adding these cases here to cater to icon view subclasses that want to
access icon view APIs from the instance structure initializer instead
of properly waiting for the super class to initialize and adding renderers
from the ->constructor() vfunc.
