We want internal state ensured to be forgotten, and the controller
to be ready for use for any future time it might be added to a
widget again.

Keep a hard copy of the widget event controllers to handle state
changes across, and check for controllers that might have been detached.
This makes us:

- Tiptoe over controllers that might have been removed directly or
  indirectly as a result of a signal emitted here, which is great and
  fixes possible crashes.
- Ignore new controllers that might have been added in the handling
  of these signals, which is fair enough since these controllers did
  not handle any input related to the state change.

Closes: https://gitlab.gnome.org/GNOME/gtk/-/issues/6924

This lets us compare the cost of realizing the new renderers vs
the old GL renderer.

gsk: Add another profiler mark

See merge request GNOME/gtk!7712

We were just exiting the loop, but not remembering the index.

Speeds up various memory operations, sometimes by quite a lot.

If no ops are recorded, then we don't need to wait for any ops to
finish.

Also fix the initial fence creation on Vulkan - we no longer need to
create it fixed because of the random cleanup() call at startup does no
longer happen.

This is useful because cleaning up will do the final copies of texture
data.

It also means we use less memory, as we're going to release images that
were used in ops.

No functional changes.

Instead of recreating frames from scratch every time, use an existing one.

This ensures that renderers don't need to recreate GPU resources every
time (like buffers and everything else that frames manage). It also
speeds up occasional render_texture() calls in default renderers.

This speeds up in particular the Vulkan renderer.

It turns out we lost the check to see if color states are equal when we
multithreaded things in !7657

Whoops.

Cairo is using BGRA8 but we want to use RGBA8 because it's the canonical
GL format.

So getting this fast sounds useful.

We don't want to get slower in the fast path.

When downloading textures, download the in the actual format of the
texture. We don't want to benchmark conversion, we want to benchmark
renderers.

In the colorize and texture ops, print the tex rect. This is useful
because when adding new features with textures (like atlas usage), these
are the ops that I use for testing.

We might want to use it outside of the nodeprocessor.

The function is now called gsk_rect_snap_to_grid().

vulkan: Actually return the preferred memory format

See merge request GNOME/gtk!7713

Some nodes like `GtkBox` need to process removed-in-GTK4 attributes to
correctly convert their contents. If the node children are processed
first, then those attributes are removed prematurely.

Just use two individual fields, so we can track if we've already
created each one. This also matches the individual fields we have
for the dmabuf formats.

And change preference order of downloaders

Previously, our order was mmap > vulkan > egl.

But depending on the hw (discrete vs integrated gpu), mmap
can be catastrophically slower (on the order of 20ms vs 1.5s).

So, change the order to egl > vulkan > mmap.

Note that this currently has less effect than we'd like to,
since we don't let the downloaders claim linear formats.

This takes the `position` property and ensures children are sorted by
it, splits children by `pack-type` (also reversing the order of `end`
children), and handles children with `type="center"`.

If either a center child or end children exist, then the `GtkBox` is
converted to a `GtkCenterBox`, with `start-widget`/`end-widget` being a
nested `GtkBox` with the relevant children.

The splitting does cause some non-`object` children to sort differently
(hence the change to `office-runner.expected`.)

builder-tool: Handle pack-type, position, and center child in GtkBox

See merge request GNOME/gtk!7715

Change the xxx_get_downloader() functions to init_dmabuf_xxx(),
and make them initialize both xxx_dmabuf_formats and xxx_downloader.

Only initialize the Vulkan or EGL parts where possible.

When dmabufs or dmabuf formats are actually used, we still
initialize fully by creating both a Vulkan and EGL downloader.

This shortens the time to first commit from 149ms to 108ms.

This is similar to the checks in tools/gtk-rendernode-tool-render.c.

resolves #5072

Signed-off-by: Thomas Devoogdt <thomas@devoogdt.com>

This is similar to the checks in tools/gtk-rendernode-tool-render.c.

resolves #5072

Signed-off-by: Thomas Devoogdt <thomas@devoogdt.com>

resolves #5072

Signed-off-by: Thomas Devoogdt <thomas@devoogdt.com>

Refactor dmabuf initialization

See merge request GNOME/gtk!7714

gtk/print/gtkprintoperation{,-unix,-portal}.c: add CAIRO_HAS checks

Closes #5072

See merge request GNOME/gtk!7717

For VK_DEBUG_REPORT_WARNING_BIT_EXT we should always have used
g_warning().

For VK_DEBUG_REPORT_ERROR_BIT_EXT g_critical() is technically the right
choice, but Mesa has been using this flag for normal warnings, so until
that gets fixed, we don't want to throw criticals.

Related: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31292
Fixes: !7020

vulkan: Demote g_critical() to g_warning()

See merge request GNOME/gtk!7723

When the text is rtl, pango will put the text at the right end of
its given width, causing the logical.x to be big, and in turn, our
computed position to be negative. If we don't allow that, centered
text will end up at the right side if it is rtl.

Fixes: #6836

We need to fix all the textures we own. This broke in 65c8320a.

Fixes: #7013

It is not necessary.
Worse, it is reentrant and causes all sorts of avoc when processing
events halfway through initializing the context.

It only exists because in commit 3887548d the Wayland Vulkan code
was copy/pasted from X11.

Fixes #7022
Fixes #7025

Fix centered text in labels

Closes #6836

See merge request GNOME/gtk!7724

wayland: Don't call gdk_display_sync()

Closes #7022 and #7025

See merge request GNOME/gtk!7726

Sometimes, jokers send us 'mimetypes' like DELETE or
org.webkitgtk.WebKit.custom-pasteboard-data, and gdk_intern_mime_type
will return NULL for such things. Handle that by just closing the fd.

Better than running into an assertion further down.

We keep a pointer to the GdkDrop object without a reference, and
then it dies on us. Be more careful, and clean up the dnd section
when the we drop object goes away.

Fixes: #7026

wayland: Be more careful with mimetypes

See merge request GNOME/gtk!7729

inspector: Avoid criticals in the clipboard code

Closes #7026

See merge request GNOME/gtk!7730

gtkwidget: Check for destroyed controllers while propagating state

Closes #6924

See merge request GNOME/gtk!7668

gsk: Don't leak gl textures

Closes #7013

See merge request GNOME/gtk!7725

stack-sidebar: Hold reference to page

Closes #5917

See merge request GNOME/gtk!6144