`present_flip()` now takes the buffer number instead of the rendering
buffer.

Fix the caller to pass the appropriate expected value.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Bug: 14741

Clicking a window title button would keep the button pressed even after
the pointer leaves the button.

This is just a simple case of receiving multiple enter/leave events
without actually checking for the window, so any additional enter events
received after the leave event would let the button in a pressed state.

Fix the issue by checking for the actual event window against the button
window.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Bug: 15400

Dialog windows which have no parent should not be skipped from taskbar.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Bug: 15399

If a client specifies an invalid NET_WM_WINDOW_TYPE, fallback to normal
window type.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Bug: 14741

For button enter/leave events, checking the crossing mode prevents the
events from being handled.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Bug: 15384

Commit f5da3d66 cause a serious regression by preventing the filter to
be run on any window but the one being grabbed.

This reverts commit f5da3d66.

The number of buffers we need to allocate/manage depends on the feature
enabled at build time.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Basically, we want the fence to prevent GL from accessing the pixmap
buffer until its update is complete.

However, depending on the code path, we may update the buffer more than
once, so to prevent any flickering, we need a finer grain control over
the fence operations.

Split the fence support into different actions as we need them, i.e.
create, destroy, reset, trigger and wait.

Make sure we reset the fence before each time we are to update the
pixmap buffer and wait for the fence just before doing the GL swap.

Also create a different fence for each buffer, so if ever decide to use
multiple buffers with GL, the existing code will be able to support that
and not use the same fence for multiple pixmaps.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Rename `scale_glx_texture()` as `set_glx_scale()`.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Fences support was added in XSync 3.1, make sure the XServer has support
for this.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

This is not needed and breaks the fence mechanism, hence causing
flickering.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Alistair Buxton authored 5 years ago and Olivier Fourdan committed 5 years ago

Bug: 15391

The compositor zoom refresh rate is currently hardcoded at half the
refresh rate, or 30 FPS.

Increase the zoom refresh to the same as the refresh rate, or 60 FPS.

Bug: 15391

Previously, the zoom update was based on XRandr refresh rate ÷ 2, or
30 by default if not available.

But that XRandr code was removed along with the removal of the fake
vblank sync code in favor of XPresent or GLX.

Base the zoom update on half the refresh rate of the primary monitor (if
available from GDK), so we would get the same results as with 4.12.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Bug: 15392

The texture filter changes based on the zoom value to give best results
when zooming at higher levels.

As we do not unbind/rebind at each repaint anymore, the texture filter
would not be updated, hence the chosen filter is ignored.

Make sure we apply the texture filter when zooming.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

We always query the root window for pointer location, there is no point
in passing it around through the API really.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

And a bit of cleanup.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

This change is incompatible with NVidia closed source driver and
causes a blank screen.

Thanks-to: Philip Müller <philm@manjaro.org>

This reverts commit 97c051b8.

Oddly, `XGetWMProtocols()` return a non-zero value on success, checking
against `Success` would lead to the wrong conclusion.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>
Fixes: 0bf68c53 - "Protect against XErrors"

Places where the window manager uses the client window are all possible
sources of XError because the window may have vanished and the window
manager won't know until it gets the destroy notification from the
Xserver.

As GDK error handler is smart and uses the serial of the requests to
check for errors, if the XError is not trapped in time it will end up in
GDK and cause an abort.

Protect against such a possibility by making sure cases where the XID
of the client or one of its dependent resources is used remains withing
an error trap/push.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

It's done on unbind.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

If we don't find a FBconfig without caveats, don't use GLX.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

That should not be needed.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Reading from the front buffer is not advisable and gives random results
depending on the driver and/or hardware.

So instead of relying on the front buffer, re-copy the entire content to
the back buffer again just after we swapped buffers.

Also, by doing so, we don't need to accumulate damages with GLX backend.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>
Fixes: a818cc06 "compositor: Optimize GLX code path"

The serial on an XError is an unsigned long.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Using just “x” for the name of the generic `XEvent` is really confusing.

Rename it as “xevent” for clarity.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Some more XErrors trap.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

When using GLX as backend, we are not using Render for zooming and the
associated `zoomBuffer` is not created, so trying to use that for any
Render operation will generate XErrors.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

When not zooming, use the damage region to update the back buffer
instead of copying the entire texture.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Draw the texture to the back buffer instead of the front buffer and
`glXSwapBuffers()` will take care of swapping the back to front.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Those are fairly common errors, no need to actually send warnings.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

The window and its resources may be gone already on the Xserver side, in
which case we might generate XErrors. Ignore those to keep on the safe
side.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

That should not be needed as GDK already handles serial ranges for us.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Helpful for debugging purpose.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Remove unused variable added by mistake.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

GTK would update the size increment to match the scale, so when using a
scale 2x the size increment of all windows is set to 2 as well.

As we use the size increment to decide whether to show the position and
resize indicator window, we would end up showing it for all windows
when scale is 2.

Take the scale into account for size increment as well.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

The poswin window may become unrealized, use our own instead.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

The position/size indicator would be misplaced when a scale is applied.

Take the scale into account to place the indicator on screen.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>

Small tweak, keep timeout active but to its minimum.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>