Summary:
This is a radical change to the approach KWin used to have. So far KWin
fall back to XRender compositing if it detected llvmpipe. With this
change KWin will use OpenGL compositing also on llvmpipe.
This has drawbacks: it's CPU emulated which we actually do not want.
But the question is whether it still makes sense to keep the no-GL
enforcement.
There are several things to consider here:
* QtQuick will use llvmpipe anyway (including things like Plasmashell)
* glamor might be used in the xorg layer which also uses llvmpipe on
XRender
* more consistent look
* we get OpenGL working in tests (important for e.g. openQA)
* hardware which does not support proper GL gets less, there are
GL drivers for KVM, there are GL drivers for rPI, etc. We don't
need to block it anymore because of possibly bad results on those
hardware. Also they mostly use glamor.
* there is still the way to manually select XRender/QPainter compositing
Reviewers: #kwin, #plasma
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D2184
Summary:
This allows effects to animate when a window is shown again and when
a window gets hidden but not yet closed/destroyed. This situation
happens on X11 for e.g. auto hiding panels and on Wayland for pretty
much any window which properly unmaps (windowHidden) prior to destroy.
Reviewers: #kwin, #plasma_on_wayland
Subscribers: plasma-devel, kwin
Tags: #plasma_on_wayland, #kwin
Differential Revision: https://phabricator.kde.org/D2084
Summary:
The method xRenderBlendPicture created a static XRenderPicture on
first usage. To cleanup a XRenderPicture an xcb_connection_t* is needed.
As it's static the cleanup happens on exit handler and at that time Qt
already destroyed the xcb_connection_t*. With a certain chance this will
crash.
To expose the problem a Q_ASSERT(qApp) is added in the destructor of
XRenderPicture. Using xrenderBlendPicture() will hit this assert on
application exit. This is demonstrated by the added auto test.
The actual fix to the problem is moving the static variable out of
the method and introduce a global cleanup method just like the init
method. This is now called from Workspace dtor, so before application
goes down.
CCBUG: 363251
Reviewers: #plasma
Subscribers: plasma-devel
Tags: #plasma
Differential Revision: https://phabricator.kde.org/D1731
Summary:
Client uses a static Xcb::Window helper. This so far didn't get
explicitly destroyed, so the application finalize cleaned it up.
To destroy the window the xcb_connection_t* is used which the
QGuiApplication already destroyed.
This change ensures that the window gets destroyed before the xcb
connection gets destroyed.
In addition an assert is added to KWin::connection() to ensure that
we still have the QGuiApplication::instance() when it's invoked.
This way we'll notice if we have more cases where we call into xcb
after the application went down.
Reviewers: #plasma
Subscribers: plasma-devel
Projects: #plasma
Differential Revision: https://phabricator.kde.org/D1573
Allows to toggle Activity Manager and bring up KRunner by hitting the configured screen edge.
BUG: 358627
BUG: 245979
Differential Revision: https://phabricator.kde.org/D1105
using stack adresses makes it possible (even tough not frequent)
for two animations (onle just killed, one just created) to have
the same id, causing scripts to be potentially really confused.
this replaces the id with a global counter, and the
"flying tooltips" bug is gone.
REVIEW:127276
BUG:360068
CCBUG:352254
using stack adresses makes it possible (even tough not frequent)
for two animations (onle just killed, one just created) to have
the same id, causing scripts to be potentially really confused.
this replaces the id with a global counter, and the
"flying tooltips" bug is gone.
REVIEW:127276
BUG:360068
CCBUG:352254
The idea was to block shaders that operate on
Texture2D while Texture2DRectangle was use, but
that's indeed controlled by TextureNPOT while
LimitedNPOT indicates Texture2D but without support
for mipmapping and exotic clamping
FIXED-IN: 5.6
REVIEW: 126959
One resource is used for shader version 1.10 and one for version 1.40.
The ideas behind this change is to remove the locating of the shader
sources and also to fix that user provided shaders could be loaded
instead of the original ones (possible attack vector on Wayland).
To simplify the ShaderManager provides a new method call to load the
shader from the resource. This means the effects don't need to
duplicate the check for the shader version any more and also don't
need to duplicate the file reading functionality.
REVIEW: 126905
This method allows to generate a shader with ShaderTraits but one
part (e.g. vertex and/or fragment) to be replaced by custom shader
code.
Thus it can be used as a replacement for the load*(ShaderType) variants
and allows more effects to transition to shaders using traits.
The screen projection matrix is needed if an effect wants to customize
the modelview projection matrix while rendering a window and keeping
the currently applied screen projection and transformation.
With this change a new ctor overload is added to ScreenPaintData
which allows passing a projection matrix through the effects. This
allows effects to put up custom shaders with a shared projection matrix
and without having to calculate it themselves.
The projection matrix is a read-only information for the effects. There
is no way to change or overwrite it.
The way nearestPowerOfTwo is currently defined allows for the possibility of
left-shifting an int by 32 bits or more, which is undefined behavior on
platforms where int is 32 bits, and is something that can happen here if `s`
is equal to 31 by the end of the loop. Noted by Coverity as CID 1291191.
This patch takes an algorithm to perform the same operation from Hank Warren
Jr.'s book "Hacker's Delight", which should avoid UB.
REVIEW:126540
The extensions specify that you are only allwoed to use the robust
functions if the context is robust. Given that we need to query
whether the robust is context and fall back to our workaround if
the context is not robust.
REVIEW: 126051
For the functions from GL_FOO_robustness we want to resolve it by
ourselves in order to add a custom implementation if it's not available.
Unfortunately once epoxy.h is included this breaks as epoxy defines the
names and so through the preprocessor epoxy always wins.
So we need different names: all functions from robustness get a "kwin"
prefix and the usage is changed everywhere in kwin source code.
REVIEW: 125883
-use qstringliteral only when necessary (i.e. not in concat or comparison)
-use qbytearray instead of qstring when dealing with latin1 input and output (glplatform)
-use qstringref to extract numbers from strings (glplatform)
-define qt_use_qstringbuilder to optimize all string concatenations
-anidata: use ctor init lists, add windowType member initialization
REVIEW: 125933
So far it was bound to whether we build for GLES. But this is
semantically wrong. It might be possible that even on desktop gl
epoxy is built without GLX support, thus we need to reflect this.
This change ensures that epoxy/glx.h is only included if available,
that relevant code is bound to it and that checks are in place to
enforce EGL if not build with glx support.
In addtion the glxbackend.cpp is now only included in the build set
if available.
Only those which truly are different are kept as compile time checks.
In addition the index buffer is made available to GLES as in principle
all required functionality is available on gles.
REVIEW: 125865
It was broken on so many ways, it's unbelievable:
* action was read but did nothing
* config was saved into a different file than read from
REVIEW: 125701
CCBUG: 331841
Removes the ElectricAction for ShowDashboard and the relevant code
in screenedges and the KCM.
Also a leftover in the glide effect.
REVIEW: 125700
BUG: 353928
A quad might have a null size in case it has a null width or null height.
Not just one dimension.
BUG: 236353
BUG: 210467
FIXED-IN: 5.4.2
REVIEW: 125131
It doesn't make much sense any more as we do no longer link EGL since the
switch to epoxy. And epoxy pulls it in at runtime if needed.
Even more on Wayland it's just plain stupid to have EGL disabled. So
removing the option just simplifies our code base without any
disadvantages.
REVIEW: 124695
Internally QX11Info::display() calls into QPA native interface and
tries to resolve the nativeResourceForIntegration for "display".
Unfortunately this key also exists in other QPA plugins, e.g. in
QtWayland.
Calling without a check on platform results on wayland qpa in
wl_display* being casted into Display*.
REVIEW: 124203
The glVersion was not set at all for gles causing any code doing a
hasGLVersion check to fail which means that the compositor doesn't
start at all, because it has a hasGLVersion(2, 0) check.
The complete ifdef is no longer needed. The used additional glGetStringi
is also available in gles 3.0, thus epoxy can handle it quite well
without a need for ifdef.
Unfortunately the version string can be "strange" on GLES. On desktop
it looks like: "3.0 some driver foo", on GLES it could also be:
"OpenGL ES 3.0 some driver bar". Thus to make the logic work we are
first removing any leading "OpenGL ES " and hope that then the version
is encoded just like on GL.
On wayland OpenGL might get initialized before we have a connection
to the XServer and we might only have an xcb connection. So let's
guard the access.
KWin::connection() uses the property to resolve the value instead of
using QX11Info. In practice this doesn't change anything at the moment,
but allows kwin_wayland to provide an xcb connection without depending
on QX11Info.
As we cannot make xcb_connection_t* available as a metatype, the
property's type is set to void*.
KWin::rootWindow() uses the property to resolve the value instead of
using QX11Info. In practice this doesn't change anything at the moment,
but allows kwin_wayland to provide a root window without depending on
QX11Info.
The porting to Qt5 broke the timestamp handling in many areas. A deeper
look into Qt's xcb plugin shows that the appTime handling is not
sufficient for KWin's need. E.g. the time is only updated in response to
a property notify event if it's for a Qt created window, which is hardly
ever the case in KWin. Another example is that key press/release events
never updated the appTime.
As the functionality in Qt is rather trivial we can do the timestamp
handling ourselves. We filter all events anyway and it is slightly faster
as we don't have to go through the QPA interface any more.
REVIEW: 122636
Resetting glPixelStore is bound to the variable useUnpack, but setting
was only bound to a subset of the checks going into useUnpack variable.
This could cause an assert if one updated with a QImage not in format
ARGB32_Premultiplied.
REVIEW: 122521
it makes no sense and there's a good chance
that in this case m_vbo is still nullptr and
it will not be created since d->m_cachedSize
will be an empty size as well
-> nullptr access -> crash
BUG: 337090
FIXED-IN: 5.2
Use glTexStorage2D() to allocate storage for the texture. The structure
of the resulting texture becomes immutable and the texture is always
mipmap complete. This allows the driver to skip the mipmap consistency
checks when validating the texture at draw time.
Prior to this commit we didn't know if mipmaps were going to be used
when we created the GL texture, which meant that we couldn't tell the
driver whether to allocate storage for mipmaps or not.
This resulted in one of two things happening depending on the driver;
either it would allocate storage for mipmaps that in most cases would
never be used, or it wouldn't and would later be forced to reallocate
the texture when mipmaps were added.
By adding this parameter we can now explicitly tell the driver how
many mipmap levels will be used.
The parameter is only added to the non-image constructor for now. The
image constructor is changed to only allocate a single level, which
matches how textures created from images are used in kwin. This may
need to be revisited in the future.
GLTexture would set both the minification and magnification filters
to GL_NEAREST if the texture filter was set to any mipmap filter other
than GL_LINEAR_MIPMAP_LINEAR.
This code is broken in a number of different ways; firstly by assuming
that the mipmaps need to be regenerated when the texture filter has
changed. Secondly by preventing mipmaps from being specified by other
means.
This commit removes the code from bind() and adds a generateMipmaps()
method instead.
Effects currently modify the matrices by reading back the uniform
values from the shader they assume will be used to paint the window,
set new values for the uniforms, call paintWindow(), and then restore
the uniforms to their previous values.
This is both slow and unreliable, and will not work with dynamically
generated shaders.
This patch solves the problem by putting the matrices in
WindowPaintData and making it the responsibility of the paintWindow()
implementation to set the uniforms.
This adds new API in ShaderManager that makes it possible to request a
shader based on a set of traits. ShaderManager generates these shaders
on demand and caches them in a hash table.
Instead of getting size from displayWidth() and displayHeight() use
the information we have from Screens. This means there is only one
place to have the information and by that we can ensure that all
components use the same data to rely on. displayWidth/displayHeight
seem to provide the wrong information when unplugging an output
without disabling the output. This results in rendering artefacts.
But KWin::Screens has the correct information available.
This is the only place where it is called.
There was also no need for this function to be virtual since it's not
reimplemented anywhere.
This also fixes the GLTexturePrivate being created only to be
immediately deleted and replaced by a new GLTexturePrivate when the
GLTexture(QImage) constructor was used.
With this change the structure of the GLTexture effectively becomes
immutable to the outside world; the contents of the texture images can
change, but the texture images themselves cannot be reallocated.
QImage::Format_RGBA8888_Premultiplied is the equivalent of
GL_RGBA/GL_UNSIGNED_BYTE, so use QImage::convertToFormat()
instead.
This commit also makes the texture upload code work correctly
on big-endian systems.
