Does this work for you?

diff --git a/eglonxbackend.cpp b/eglonxbackend.cpp
index 480e533..06e5a16 100644
--- a/eglonxbackend.cpp
+++ b/eglonxbackend.cpp
@@ -266,7 +266,7 @@ EGLSurface EglOnXBackend::createSurface(xcb_window_t window)
 
 bool EglOnXBackend::initBufferConfigs()
 {
-    const EGLint config_attribs[] = {
+    EGLint config_attribs[] = {
         EGL_SURFACE_TYPE,         EGL_WINDOW_BIT|EGL_SWAP_BEHAVIOR_PRESERVED_BIT,
         EGL_RED_SIZE,             1,
         EGL_GREEN_SIZE,           1,
@@ -280,8 +280,11 @@ bool EglOnXBackend::initBufferConfigs()
     EGLint count;
     EGLConfig configs[1024];
     if (eglChooseConfig(eglDisplay(), config_attribs, configs, 1024, &count) == EGL_FALSE) {
-        qCCritical(KWIN_CORE) << "choose config failed";
-        return false;
+        config_attribs[1] = EGL_WINDOW_BIT;
+        if (eglChooseConfig(eglDisplay(), config_attribs, configs, 1024, &count) == EGL_FALSE) {
+            qCCritical(KWIN_CORE) << "choose config failed";
+            return false;
+        }
     }
 
     ScopedCPointer<xcb_get_window_attributes_reply_t> attribs(xcb_get_window_attributes_reply(m_connection,

> Does this work for you?

Nope becouse EGL_BAD_CONFIG return only in eglCreatePlatformWindowSurfaceEXT and eglChooseConfig well done.

Created attachment 96244 [details]
Patch for fix swap behavior problem for intel

(In reply to Uriy Zhuravlev from comment #3)
> Created attachment 96244 [details]
> Patch for fix swap behavior problem for intel

All effects work but I don't like EGL_BAD_MATCH after EGL init...

Meh. Martin simply wanted to avoid it if buffer_age is supported (what makes sense).
Do you have EGL_EXT_buffer_age in the client extensions?

qDebug() << "buffer age?" << hasClientExtension(QByteArrayLiteral("EGL_EXT_buffer_age"));

in ::initRenderingContext()

(In reply to Thomas Lübking from comment #5)
> Meh. Martin simply wanted to avoid it if buffer_age is supported (what makes
> sense).
> Do you have EGL_EXT_buffer_age in the client extensions?
> 
> qDebug() << "buffer age?" <<
> hasClientExtension(QByteArrayLiteral("EGL_EXT_buffer_age"));
> 
> in ::initRenderingContext()

buffer age? false

Sorry for the long answer.

Errr... *after*  the "initClientExtensions()" call???
Can you in case post the output of "es2_info"?

Please:

EGL_VERSION: 1.4 (DRI2)
EGL_VENDOR: Mesa Project
EGL_EXTENSIONS:
    EGL_CHROMIUM_sync_control, EGL_EXT_buffer_age, 
    EGL_EXT_create_context_robustness, EGL_EXT_image_dma_buf_import, 
    EGL_KHR_create_context, EGL_KHR_fence_sync, 
    EGL_KHR_get_all_proc_addresses, EGL_KHR_gl_renderbuffer_image, 
    EGL_KHR_gl_texture_2D_image, EGL_KHR_gl_texture_cubemap_image, 
    EGL_KHR_image, EGL_KHR_image_base, EGL_KHR_image_pixmap, 
    EGL_KHR_surfaceless_context, EGL_KHR_wait_sync, 
    EGL_MESA_configless_context, EGL_MESA_drm_image, 
    EGL_MESA_image_dma_buf_export, EGL_NOK_texture_from_pixmap, 
    EGL_WL_bind_wayland_display
EGL_CLIENT_APIS: OpenGL OpenGL_ES OpenGL_ES2 OpenGL_ES3 
GL_VERSION: OpenGL ES 3.0 Mesa 11.2.0-devel (git-b201a6e)
GL_RENDERER: Mesa DRI Intel(R) Haswell Mobile 
GL_EXTENSIONS:
    GL_EXT_blend_minmax, GL_EXT_multi_draw_arrays, 
    GL_EXT_texture_filter_anisotropic, GL_EXT_texture_compression_dxt1, 
    GL_EXT_texture_format_BGRA8888, GL_OES_compressed_ETC1_RGB8_texture, 
    GL_OES_depth24, GL_OES_element_index_uint, GL_OES_fbo_render_mipmap, 
    GL_OES_mapbuffer, GL_OES_rgb8_rgba8, GL_OES_standard_derivatives, 
    GL_OES_stencil8, GL_OES_texture_3D, GL_OES_texture_float, 
    GL_OES_texture_float_linear, GL_OES_texture_half_float, 
    GL_OES_texture_half_float_linear, GL_OES_texture_npot, GL_OES_EGL_image, 
    GL_OES_depth_texture, GL_OES_packed_depth_stencil, 
    GL_EXT_texture_type_2_10_10_10_REV, GL_OES_get_program_binary, 
    GL_APPLE_texture_max_level, GL_EXT_discard_framebuffer, 
    GL_EXT_read_format_bgra, GL_NV_fbo_color_attachments, 
    GL_OES_EGL_image_external, GL_OES_EGL_sync, GL_OES_vertex_array_object, 
    GL_ANGLE_texture_compression_dxt3, GL_ANGLE_texture_compression_dxt5, 
    GL_EXT_texture_rg, GL_EXT_unpack_subimage, GL_NV_draw_buffers, 
    GL_NV_read_buffer, GL_NV_read_depth, GL_NV_read_depth_stencil, 
    GL_NV_read_stencil, GL_EXT_draw_buffers, GL_EXT_map_buffer_range, 
    GL_KHR_debug, GL_OES_depth_texture_cube_map, GL_OES_surfaceless_context, 
    GL_EXT_color_buffer_float, GL_EXT_separate_shader_objects, 
    GL_EXT_shader_integer_mix, GL_INTEL_performance_query, 
    GL_EXT_draw_elements_base_vertex, GL_KHR_context_flush_control, 
    GL_OES_draw_elements_base_vertex, GL_EXT_blend_func_extended

(In reply to Uriy Zhuravlev from comment #8)
> Please:

"Here you are" ;-)

The extension seems there, to be clear, this:

diff --git a/eglonxbackend.cpp b/eglonxbackend.cpp
index 480e533..d7e2fa0 100644
--- a/eglonxbackend.cpp
+++ b/eglonxbackend.cpp
@@ -166,6 +166,7 @@ void EglOnXBackend::init()
 bool EglOnXBackend::initRenderingContext()
 {
     initClientExtensions();
+    qDebug() << "buffer age?" << hasClientExtension(QByteArrayLiteral("EGL_EXT_buffer_age"));
     EGLDisplay dpy;
 
     // Use eglGetPlatformDisplayEXT() to get the display pointer


prints "false" on your side?

(In reply to Thomas Lübking from comment #9)
> prints "false" on your side?

Yes. My past result printed after initClientExtensions. :)

which mesa version has that new behavior? When we discussed it on the mailing list my mesa didn't have it yet, so I didn't look into the implementation.

(In reply to Martin Gräßlin from comment #11)
> which mesa version has that new behavior? When we discussed it on the
> mailing list my mesa didn't have it yet, so I didn't look into the
> implementation.

Actual git master.

What puzzles me is that EGL_EXT_buffer_age seems to be available, so why would it not be in the list of client extensions?

(In reply to Thomas Lübking from comment #13)
> What puzzles me is that EGL_EXT_buffer_age seems to be available, so why
> would it not be in the list of client extensions?

Because in:
void AbstractEglBackend::initClientExtensions()

eglQueryString(EGL_NO_DISPLAY, EGL_EXTENSIONS) returned only:
EGL_EXT_client_extensions EGL_EXT_platform_base EGL_EXT_platform_wayland EGL_EXT_platform_x11 EGL_KHR_client_get_all_proc_addresses EGL_MESA_platform_gbm

and yes es2_info relay difference between eglinfo:
EGL API version: 1.4
EGL vendor string: Mesa Project
EGL version string: 1.4 (DRI2)
EGL client APIs: OpenGL OpenGL_ES OpenGL_ES2 OpenGL_ES3 
EGL extensions string:
    EGL_CHROMIUM_sync_control EGL_EXT_buffer_age
    EGL_EXT_create_context_robustness EGL_EXT_image_dma_buf_import
    EGL_KHR_create_context EGL_KHR_fence_sync
    EGL_KHR_get_all_proc_addresses EGL_KHR_gl_renderbuffer_image
    EGL_KHR_gl_texture_2D_image EGL_KHR_gl_texture_cubemap_image
    EGL_KHR_image EGL_KHR_image_base EGL_KHR_image_pixmap
    EGL_KHR_surfaceless_context EGL_KHR_wait_sync
    EGL_MESA_configless_context EGL_MESA_drm_image
    EGL_MESA_image_dma_buf_export EGL_NOK_texture_from_pixmap
    EGL_WL_bind_wayland_display
EGL client extensions string:
    EGL_EXT_client_extensions EGL_EXT_platform_base
    EGL_EXT_platform_wayland EGL_EXT_platform_x11
    EGL_KHR_client_get_all_proc_addresses EGL_MESA_platform_gbm

"EGL client extensions string"  -- It solves your puzzle.

Hey. Do you have any thoughts? I think es2 != egl. 

PS Now I have to use a LXQT instead KDE. Plasma5 + kwin instability at Intel. :(

We need to pre-fetch egl extensions init to know wheter BA is supported, untested patch:

diff --git a/eglonxbackend.cpp b/eglonxbackend.cpp
index 480e533..abacdaa 100644
--- a/eglonxbackend.cpp
+++ b/eglonxbackend.cpp
@@ -89,6 +89,8 @@ EglOnXBackend::~EglOnXBackend()
 
 void EglOnXBackend::init()
 {
+    initEgl();       // required to toggle
+    initBufferAge(); // EGL_SWAP_BEHAVIOR_PRESERVED_BIT
     if (!initRenderingContext()) {
         setFailed(QStringLiteral("Could not initialize rendering context"));
         return;
@@ -267,7 +269,7 @@ EGLSurface EglOnXBackend::createSurface(xcb_window_t window)
 bool EglOnXBackend::initBufferConfigs()
 {
     const EGLint config_attribs[] = {
-        EGL_SURFACE_TYPE,         EGL_WINDOW_BIT|EGL_SWAP_BEHAVIOR_PRESERVED_BIT,
+        EGL_SURFACE_TYPE,         EGL_WINDOW_BIT|(supportsBufferAge() ? 0 : EGL_SWAP_BEHAVIOR_PRESERVED_BIT),
         EGL_RED_SIZE,             1,
         EGL_GREEN_SIZE,           1,
         EGL_BLUE_SIZE,            1,

1. initEGL(); insted initEgl(); 
2. Nope... it's not work.

EGL_WINDOW_BIT|(supportsBufferAge() ? 0 : EGL_SWAP_BEHAVIOR_PRESERVED_BIT)
maybe you mean:
EGL_WINDOW_BIT|(supportsBufferAge() ? EGL_SWAP_BEHAVIOR_PRESERVED_BIT : 0)

?

No, we only want to skip that bit if we've buffer_age support anyway.
I've not looked into it, but guess the extension check requires a context, what means we cannot depend this bit on the absence of buffer_age - or we cannot avoid surface recreation - either because it failed due to that bit or because we skipped the bit initially and afterwards figure we need it because we lack buffer_age support. This sucks.

Can you debug the result of initEGL?

diff --git a/eglonxbackend.cpp b/eglonxbackend.cpp
index 480e533..abacdaa 100644
--- a/eglonxbackend.cpp
+++ b/eglonxbackend.cpp
@@ -89,6 +89,8 @@ EglOnXBackend::~EglOnXBackend()
 
 void EglOnXBackend::init()
 {
+    initEgl();       // required to toggle
+    initBufferAge(); // EGL_SWAP_BEHAVIOR_PRESERVED_BIT
     if (!initRenderingContext()) {
         setFailed(QStringLiteral("Could not initialize rendering context"));
         return;
@@ -267,7 +269,7 @@ EGLSurface EglOnXBackend::createSurface(xcb_window_t window)
 bool EglOnXBackend::initBufferConfigs()
 {
     const EGLint config_attribs[] = {
-        EGL_SURFACE_TYPE,         EGL_WINDOW_BIT|EGL_SWAP_BEHAVIOR_PRESERVED_BIT,
+        EGL_SURFACE_TYPE,         EGL_WINDOW_BIT|(supportsBufferAge() ? 0 : EGL_SWAP_BEHAVIOR_PRESERVED_BIT),
         EGL_RED_SIZE,             1,
         EGL_GREEN_SIZE,           1,
         EGL_BLUE_SIZE,            1,
diff --git a/libkwineffects/kwinglutils.cpp b/libkwineffects/kwinglutils.cpp
index 438b785..b44b821 100644
--- a/libkwineffects/kwinglutils.cpp
+++ b/libkwineffects/kwinglutils.cpp
@@ -101,6 +101,7 @@ void initEGL()
     eglVersion = MAKE_GL_VERSION(major, minor, 0);
     const QByteArray string = eglQueryString(dpy, EGL_EXTENSIONS);
     eglExtensions = string.split(' ');
+    qDebug() << eglExtensions;
     eglResolveFunctions();
 }

I have two outputs, first:
("EGL_EXT_client_extensions", "EGL_EXT_platform_base", "EGL_EXT_platform_wayland", "EGL_EXT_platform_x11", "EGL_KHR_client_get_all_proc_addresses", "EGL_MESA_platform_gbm")
extension "EGL_EXT_buffer_age"    () () ("EGL_EXT_client_extensions", "EGL_EXT_platform_base", "EGL_EXT_platform_wayland", "EGL_EXT_platform_x11", "EGL_KHR_client_get_all_proc_addresses", "EGL_MESA_platform_gbm") 

and second:
("EGL_CHROMIUM_sync_control", "EGL_EXT_buffer_age", "EGL_EXT_create_context_robustness", "EGL_EXT_image_dma_buf_import", "EGL_KHR_create_context", "EGL_KHR_fence_sync", "EGL_KHR_get_all_proc_addresses", "EGL_KHR_gl_renderbuffer_image", "EGL_KHR_gl_texture_2D_image", "EGL_KHR_gl_texture_cubemap_image", "EGL_KHR_image", "EGL_KHR_image_base", "EGL_KHR_image_pixmap", "EGL_KHR_surfaceless_context", "EGL_KHR_wait_sync", "EGL_MESA_configless_context", "EGL_MESA_drm_image", "EGL_MESA_image_dma_buf_export", "EGL_NOK_texture_from_pixmap", "EGL_WL_bind_wayland_display", "")

The first one looks like other debug out (on your local system, not caused by the patch) of the client extensions and a match of EGL_EXT_buffer_age against all three extension lists.

The second one looks promising and things should work. As the first one suggest a local patch - what's your exact diff agains origin/master and/or would it work as expected if you strip everything except for the patch in comment #16 ?

Created attachment 96500 [details]
Fix init compositing for intel egl

Excuse me, I'm a little tired of this debug. So I decided to investigate the problem itself.
Attached is my new patch.
The main difference from your proposal is what I use:
EGLDisplay dpy = eglGetDisplay(EGL_DEFAULT_DISPLAY);
insted:
EGLDisplay dpy = eglGetCurrentDisplay();

because from docs:
"eglGetCurrentDisplay returns the current EGL display connection for the current EGL rendering context, as specified by eglMakeCurrent. If no context is current, EGL_NO_DISPLAY is returned."

and eglMakeCurrent call only from initRenderingContext. 

For EGL_NO_DISPLAY glQueryString(dpy, EGL_EXTENSIONS) returns only EGL client extensions. 

This patch solve my problem... While it may be ugly. It may be necessary to add arg to initEGL?

if (dpy == EGL_NO_DISPLAY)
   dpy = eglGetDisplay(EGL_DEFAULT_DISPLAY);

should do.
Martin?

> if (dpy == EGL_NO_DISPLAY)
>   dpy = eglGetDisplay(EGL_DEFAULT_DISPLAY);

I will try this. Although it seems to me a very good solution.

Martin?

if it works I'm fine with it.

Let's get this off the list
https://git.reviewboard.kde.org/r/126783/

>Let's get this off the list

Good! Patch works well.

There seem to be a mistake in the patch:

    EGLDisplay dpy = eglGetDisplay(EGL_DEFAULT_DISPLAY);
    if (dpy == EGL_NO_DISPLAY)
        dpy = eglGetDisplay(EGL_DEFAULT_DISPLAY);

I guess the first line was not supposed to be changed, as this is now calling the same function again with the same parameters in case it fails.

Hehe - many thanks for covering my ass =)

No problem :)

I don't know anything about EGL, but shouldn't the call be eglGetCurrentDisplay() instead of eglGetDisplay() like it is in the updated patched version? Seems like eglGetDisplay always take a parameter.

(In reply to AnAkkk from comment #30)
> No problem :)
> 
> I don't know anything about EGL, but shouldn't the call be
> eglGetCurrentDisplay() instead of eglGetDisplay() like it is in the updated
> patched version? Seems like eglGetDisplay always take a parameter.

Yep, current version is wrong. :)

This kinda starts to become embarrassing... :-\

(In reply to Thomas Lübking from comment #32)
> This kinda starts to become embarrassing... :-\

It happens to everyone. The current version is going and work perfectly. I hope it will be quick in the upstream.

Git commit e641022bf9482a11209577b5654cd43231be0755 by Thomas Lübking.
Committed on 18/01/2016 at 10:02.
Pushed by luebking into branch 'Plasma/5.5'.

skip SWAP_BEHAVIOR_PRESERVED for supportsBufferAge

pointless and unsupported on latter MESA/DRI3
http://lists.freedesktop.org/archives/mesa-dev/2015-November/100869.html

Thanks to Uriy Zhuravlev for reporting and investigation
FIXED-IN: 5.5.4

M  +3    -1    eglonxbackend.cpp
M  +2    -0    libkwineffects/kwinglutils.cpp

http://commits.kde.org/kwin/e641022bf9482a11209577b5654cd43231be0755

Git commit c7aefc6b6b62c704929fc74338da2ca0d07b547a by Thomas Lübking.
Committed on 18/01/2016 at 21:32.
Pushed by luebking into branch 'master'.

skip SWAP_BEHAVIOR_PRESERVED for supportsBufferAge

pointless and unsupported on latter MESA/DRI3
http://lists.freedesktop.org/archives/mesa-dev/2015-November/100869.html

Thanks to Uriy Zhuravlev for reporting and investigation
Related: bug 356882
FIXED-IN: 5.5.4
REVIEW: 126783

M  +3    -1    eglonxbackend.cpp
M  +2    -0    libkwineffects/kwinglutils.cpp

http://commits.kde.org/kwin/c7aefc6b6b62c704929fc74338da2ca0d07b547a

*** Bug 356882 has been marked as a duplicate of this bug. ***