bob5972 authored 4 years ago and Olivier Fourdan committed 4 years ago

On my system, 2-4% of the CPU time in xfwm4 is being spend in
find_cwindow_in_screeni() pulling in entries from the linked list on
a lightly updating desktop.

After this change, 25-50% of that CPU is gone, presumably because the
slice allocators have better locality of the list-entries.

Closes: #416

bob5972 authored 4 years ago and Olivier Fourdan committed 4 years ago

The function free_win_data() may generate X11 errors as the client
window may already be gone.

To avoid those, the entire function code is supposed to be enclosed
within an error trap/push.

A typo in the code would however push twice instead of a balanced
push/pop which causes the list of event sequences in GDK to grow
indefinitely, leading to slowdowns as the list grows over time.

Fix the leak by balancing the error push with an error pop at the end
of the function.

#351

The function compositorCheckCMSelection() would use the atom index in
our table of atoms, instead of the actual atom, so this is unlikely to
work as expected.

Fix the code to use the appropriate atom as it should.

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

Prefer using memory allocation functions which initialize the memory
with 0's to be on the safe side.

No functional change.

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

A pid is not necessarily a long on all systems, so casting the value of
NET_WM_PID directly into a GPid may cause a stack overflow.

Fixes: 49c1a33a
Closes: #411


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

Dridi Boukelmoune authored 5 years ago and Olivier Fourdan committed 4 years ago

If the user selected a primary monitor, the Alt+Tab window will only
appear on that monitor, otherwise on all of them.

To quote the GDK documentation for gdk_display_get_primary_monitor:

> [...], specialized desktop applications such as panels should place
> themselves on the primary monitor.

It is however not possible to use gdk_display_get_primary_monitor and
also respect the user's setting. Despite what the online documentation
says, recent versions of GTK+3 may no longer return a null value and
will instead pick one monitor as the primary if there isn't already one
defined. The nullable marker [1] for the return value was eventually
dropped.

In other words, whether the user selected a primary monitor or not
gdk_display_get_primary_monitor will always return one.

Signed-off-by: Dridi Boukelmoune <dridi@fedoraproject.org>

[1] https://gitlab.gnome.org/GNOME/gtk/commit/76d95c312dde

Dridi Boukelmoune authored 4 years ago and Olivier Fourdan committed 4 years ago

Signed-off-by: Dridi Boukelmoune <dridi@fedoraproject.org>

Add more XError traps related to XConfigureWindow requests.

#405


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

To be on the safe side.

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

The gvalues strings were considered statics, hence not duplicated.

As the caller frees the strings after setting the gvalue, the actual
content would end up pointing at free data.

Consider string gvalues as regular strings, meaning that the given
string value is duplicated as a gvalue.

Signed-off-by: Olivier Fourdan <fourdan@xfce.org>
Closes: #406

Commit 821ee7cc would not actually send the synthetic configure notify
events anymore. That obviously breaks plenty of applications…

While the logic of the commit itself is sound, the problem is that it
bases the descision to send the configure notify event by comparing
against the current window size and location.

The problem is that the current window size and location is modified in
various places, defeating the logic.

Add an “applied_geometry” rectangle that we update only in a single
place in the code, so that we know exactly when to send a synthetic
configure notify event.

Closes: #403


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

The size and position of the client are stored at multiple occasions.

But the naming of the variables used is a bit confusing, rename the
variables and use GdkRectangles instead.

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

xfwm4 would send a synthetic configure event while resizing the client
window, which is not what the ICCCM states.

Fix this by sending the synthetic configure event only when necessary,
as per the ICCCM specification.

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

Avoid setting the same flag twice in a row, we already set the
CFG_REQUEST just a few line above.

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

Some client do not process events on their temporary windows, causing
the client unresponsive dialog to show up.

Avoid pinging the client on focus in, even though this is clearly a
client bug.

This reverts commit 637d70b2.

To make sure that they are still alive and detect hung clients.

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

When xfwm4 needs to kill an unresponsive client, it will use the its PID
is known.

However, the _NET_WM_PID as defined in the extended window manager hint
mechanism is unreliable when the client runs within a sandbox, as it is
the client itself who sets that property using its own view of the PID.

The Xserver can provide a more accurate value of the client PID using
the XRes X11 extension, so use that protocol if available.

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

The X-Resource extension client library (XRes) allows a client to query
the Xserver for various resources associated with an X11 client,
including its PID.

Add (optional) support for XRes in xfwm4 and use it for getting the
client's PID instead of relying on _NET_WM_PID.

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

It's easier on the eyes...

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

Bug: 16713

If one of the required X11 extension is not available, the composiutor
cannot be used.

Yet xfwm4 would still try to enable the compositor for each screen,
which would not work and prevent xfwm4 from starting.

If we can't use the compositor, well, we'd better not use it...

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

Clicking outside the tab-win would cancel the selection, make sure to
revert to the original first window of the list in that case, it feels
more natural and predictible.

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

Bug: 16713

Without XRender, the mask would be misrendered and completely
transparent.

As a result, the entire pixmap would become totally transparent and all
window decorations become invisible.

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

Bug: 16708

With Xi2, the clients needs to reset the valuators (for mouse wheel)
themselves, usually on focus events.

Qt instead does that on EnterNotify, but ignores the EnterNotify events
resulting from grabs first. This is a bug in Qt as that prevents the
valuators to be reset if the pointer enters the window while grabbed.

Precisely, xfwm4 would issue a pointer grab while cycling to prevent
from random button events, and would only ungrab after the client window
is raised and focused.

While this is a bug in Qt, we can still work around the issue in xfwm4
by not issuing a pointer grab and simply abort the window cycling when a
click occurs outside of the alt-tab window.

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

Update the Pango attributes whent he widget font factor is updated, then
repaint all frames.

With this, changing the scaling factor in the appearance settings
updates the title font dynamically as well.

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

Bug: 16067

The font height is not used anymore, remove it.

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

Bug: 16067

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

Bug: 16711

We need to keep the pango font attributes for the scale with HiDPI.

This reverts commit 34a31e27.

Bug: 16383

If the window has vanished, the XShape request will fail.

Make sure all XShape requests are protected by error traps.

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

As we already do an error trap/push with X11 core devices, do the same
with Xi2 grabs and remove the check from the callers as it is not needed
anymore.

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

Bug: 15891

When raising a transient window we would automatically raise its parent
windows.

However, if the transient and parent are not placed on the same layer,
we might raise the parent from a lower layer.

To avoid that issue, limit the scope of search to the same or higher
layers.

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

When using automatic vblank mode, use a message instead of a warning.

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

With vblank mode set to auto (default), use the GL renderer to select
XPresent on Intel and AMD if available.

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

glEnable() is not once the texture type is chosen, no need to do that in
advance.

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

An `else` statement was left after the recent code rework.

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

Bug: 15852

Previous attempt to fix a crash with a screensize of 0, which can occur
without any output connected was incomplete (commit 3925109d) as a
client might still trigger a repaint by queuing up damages.

Make sure we take this case into account as well.

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

Simon Steinbeiss authored 5 years ago and Olivier Fourdan committed 4 years ago

Bug: #16382

After hovering the currently selected item the tabwin would be left
without a selected item. This patch makes sure that on the leave event
the selected item regains its selected style.

Also switch from GTK_STATE_FLAG_ACTIVE to GTK_STATE_FLAG_CHECKED.

Andre Miranda authored 4 years ago and Olivier Fourdan committed 4 years ago

Bug: 16644

GTimeVal is deprecated, use the GDK API du jour instead.

Bug: 16649

With Xi2 we need to use Xi2 routines.

But some client may issue standard X11 core protocol grabs the on the
client window.

Grabs from the standard core protocol and Xi2 grabs do not mix well.

So instead of ditching the Xi2 grabs, we issue the core protocol first,
check that it succeeds and only if that works we can issue the Xi2
grabs.

That keeps the Xi2 grabs logic while fixing the issue with X-AIR-Edit
which uses a passive grab on the pointer using core X11 protocol.

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

Bug: 16649

This reverts commit aa3e3cac.

Bug: 16347

XInput2 passive button grab would not fail if the button is already
grabbed by the client.

X-AIR-Edit is such a client which issues a button grab on its own
window, meaning that with XI2 button it would not react to button events
anymore.

Revert to the standard grab button for the passive grabs so that such
apps can still work with xfwm4.

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