Bug: 13607

If XPresentPixmap() fails, we would disable Present rather that waiting
for a notification that will never arrive.

But that relies on the actual window to be reported in the XError, which
is not necessarily the case.

Disable Xpresent for all screens as we are not sure we can tell which
one from the resource id.

Błażej Szczygieł authored 7 years ago and Olivier Fourdan committed 7 years ago

Bug: 13613

Since commit 7115d1b4, we can't disable second (broken) vblank waiting
code.

glXSwapBuffers() already waits for v-blank (if enabled in driver), so
any other "synthetic" waiting for v-blank can reduce performance
(de-sync) without actual benefit.

This patch removes old code to prevent waiting for vblank twice.

Bug: 13608

Commit 58955779 (Use ximerama indices for _NET_WM_FULLSCREEN_MONITORS)
introduced a regression because the monitor index was wrongly compared
against the total number of monitors, which would falsely assume that
the Xrandr query failed whereas it actually worked.

Fix the false comparison and also use the screen size as fallback values
in case the Xrandr really failed so that we don't end up using random
uninitialized values.

Prefer Present over GLX if both are available, and remove the
environment variable "XFWM4_USE_PRESENT".

Eventually, we'd rather use a command line option instead (TODO).

This is a leftover of the old (broken) vblank code.

Otherwise the current WM window might have vanished and we wouldn't
know.

Bug: 9351

_NET_WM_FULLSCREEN_MONITORS uses Xinerama indices for multi-monitor
fullscreen, but xfwm4 uses gdk indices which may differ.

As a result, depending on the actual layout and which monitor is marked
as primary, a fullscreen window could be placed on the wrong outputs.

Fetch the Xinerama list and use the monitor size from there instead of
relying on gdk monitor indices for _NET_WM_FULLSCREEN_MONITORS

Bug: 12534

This reverts commit 0dfb5540.

Bug: 13454

We would use the first monitor by default, but we'd rather use the
primary monitor.

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

NET_WM_STATE_FOCUS is set by the WM itself, it's not an unknown
property.

Bug: 13344

If a window is unmapped while the user is moving or resizing it, we
would cancel the move but swallow the event, which would leave the
window frame on screen as the main event handle would not get the
unmap notify.

Mark the client as gone when detected in the move/resize event loop,
let the event get through to the main event handler that will free the
client and shortcut the move/resize operation if the client is marked
as gone.

The panel is a dock window, and may appear differently as it is not
focused by default by the window manager.

Set the NET_WM_STATE_FOCUSED property on windows of type dock or
spashscreen so that they don't get the state GTK_STATE_FLAG_BACKDROP set
by default.

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

Bug: 13257

If for some reason XPresentPixmap() fails, we might end up waiting
forever for a present complete notification that will never arrive.

Check for possible errors after XPresentPixmap() and disable XPresent in
case of error.

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

Add an API to retrieve the screen from the output window.

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

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

Store the default display and add an API to retrieve it.

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

haarp authored 8 years ago and Olivier Fourdan committed 8 years ago

Bug: 6427

Some mice have additional forward and backward buttons, use those with
the easy-move key to cycle through the workspaces.

Bug: 12634

xfwm4 would raise a client upon receiving the _NET_ACTIVE_WINDOW message
but if the client provided an old timestamp, the window won't be focused
(as the timestamp will be older than the last current X server time).

To avoid this, only raise the window on focus event notification and
when using click-to-focus mode.

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

Bug: 12587

Relocating any window on the current monitor in multi-monitor setup
when centered can cause issues with Citrix.

Relocate only dialogs, and when not the exact same size as the screen.

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

Keith Packard authored 8 years ago and Olivier Fourdan committed 8 years ago

LibreOffice incorrectly sets its top level windows win_gravity field
to StaticGravity. Resizing these windows to the left or upwards
results in a corrupted application with the application window moved
within the parent by the parent window resize.

Metacity forces window gravity for all application windows to
NorthWest, masking this application bug. xfwm4 should do the same.

A bug has been filed against LibreOffice as well.

Signed-off-by: Keith Packard <keithp@keithp.com>

Restore modifiers masks removed by mistake in previous commit.

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

Bug: 10760

If a given keyboard modifier is not supported, ignore the shortcut.

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

That's sufficient for what we do.

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

Andreas Müller authored 11 years ago and Olivier Fourdan committed 9 years ago

Bug: 10184

Signed-off-by: Andreas Müller <schnitzeltony@googlemail.com>

Stig Inge Lea Bjørnsen authored 9 years ago and Olivier Fourdan committed 9 years ago

Bug: 12344

Use gdk_screen_get_system_colormap() instead of
gdk_screen_get_rgb_colormap() which returns NULL on a 30-bit screen and
therefore breaks further rendering.

Originally reported in Debian bug #800883 where this patch was
attached.

Bug: 10413

Some applications may temporarily set a transient relationship betwen
their toplevel windows to maintain a stacking order, in which case we
would change the decorations (as transients may not have all buttons
available).

However, we ought to restore the controls when this transient
relationship is removed by the application, otherwise we leave the
toplevels with reduced controls.

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

Bug: 11861

In some cases, the user may prefer Xpresent over GLX for vsync, if both
are present, add an environment variable XFWM4_USE_PRESENT for this
purpose.

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

Bug: 12331

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

Bug: 11962

Previous commit has the comment backwards.

Bug: 11962

If all we have is GLX_TEXTURE_2D, we ought to have NPOT.

Thanks again to Ilia Mirkin <imirkin@alum.mit.edu> for pointing this
out!

Bug 11962

The compositor using GL for tear-free rendering would generate corrupted
output on some older nv34 card because it wrongly uses GL_TEXTURE_2D
instead of GL_TEXTURE_RECTANGLE_ARB which is supported on this hardware.

Remove detection of "GL_ARB_texture_rectangle" and
"GL_ARB_texture_non_power_of_two" which are GL extensions, not GLX.

Instead, rely on GLX_BIND_TO_TEXTURE_TARGETS_EXT returned values to
determine the texture type and target to use.

See also: https://bugs.freedesktop.org/show_bug.cgi?id=90871


Thanks to Hans de Goede <jwrdegoede@fedoraproject.org> and Ilia Mirkin
<imirkin@alum.mit.edu> for their help on this issue.

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

To reduce the number of motion notify events, allowing for better event
compression.

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

Bug: 12176

If, upon mapping, WM_SIZE_HINTS specifies a non-resizable window, xfwm4
ignores all subsequent updates to WM_SIZE_HINTS set by the application.

This is a regression introduced by commit 78b5c3c4 (bug 10910).

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

Guus Sliepen authored 9 years ago and Olivier Fourdan committed 9 years ago

Bug: 11995

In wireframeDrawXlib(), there was a check to not redraw the wireframe
if it had the same size as the previously drawn wireframe. This check is
also in wireframeDrawCairo(), but in the Xlib case it prevents correct
updates of the wireframe when cycling through identically sized windows.

Signed-off-by: Guus Sliepen <guus@debian.org>