[RFC,v1] package/rpi-userland: add GLint64, GLuint64 and GLsync typedefs to GLES2/gl2ext.h

Fixes:

  - https://bugs.busybox.net/show_bug.cgi?id=13796

.../host/arm-linucleus-linux-gnueabihf/sysroot/usr/include/gstreamer-1.0/gst/gl/glprototypes/gstgl_compat.h:40:18: error: conflicting declaration ‘typedef void* GLsync’
   40 | typedef gpointer GLsync;
      |                  ^~~~~~

.../host/arm-linucleus-linux-gnueabihf/sysroot/usr/include/qt5/QtGui/qopengles2ext.h:24:26: note: previous declaration as ‘typedef struct __GLsync* GLsync’
   24 | typedef struct __GLsync *GLsync;
      |                          ^~~~~~

File gstgl_compat.h:

 39 #if !GST_GL_HAVE_GLSYNC
 40 typedef gpointer GLsync;
 41 #endif

File qopengles2ext.h:

   1 #ifndef __gles2_gl2ext_h_
   2 #define __gles2_gl2ext_h_ 1
   3
   4 #if 0
   5 #pragma qt_no_master_include
   6 #pragma qt_sync_skip_header_check
   7 #pragma qt_sync_stop_processing
   8 #endif
   9
  10 #ifdef __cplusplus
  11 extern "C" {
  12 #endif
  13
  14 #ifndef __gl3_h_
  15 /* These types are defined with reference to <inttypes.h>
  16  * in the Apple extension spec, but here we use the Khronos
  17  * portable types in khrplatform.h, and assume those types
  18  * are always defined.
  19  * If any other extensions using these types are defined,
  20  * the typedefs must move out of this block and be shared.
  21  */
  22 typedef khronos_int64_t GLint64;
  23 typedef khronos_uint64_t GLuint64;
  24 typedef struct __GLsync *GLsync;
  25 #endif

The problem is that gstreamer sees the original gl2ext.h file from
rpi-userland (rpi-userland-093b30bbc2fd083d68cc3ee07e6e555c6e592d11/interface/khronos/include/GLES2/gl2ext.h)
without the additional GLint64, GLuint64 and GLsync typedef definitions but
checks for existance and if not found enables its own versions in gstgl_compat.h,
incompatible with the later ones used from Qt (qopengles2ext.h).

Fix this by adding the same typedef definitions already to the
rpi-userland header version.

Signed-off-by: Peter Seiderer <ps.report@gmx.net>
---
Notes:
  - only compile tested, no run time test done yet
---
 ...dd-GLint64-GLuint64-and-GLsync-typed.patch | 93 +++++++++++++++++++
 1 file changed, 93 insertions(+)
 create mode 100644 package/rpi-userland/0006-GLES2-gl2ext.h-add-GLint64-GLuint64-and-GLsync-typed.patch

Hello,

On Mon, 10 May 2021 21:53:52 +0200
Peter Seiderer <ps.report@gmx.net> wrote:

> + interface/khronos/include/GLES2/gl2ext.h | 13 +++++++++++++
> + 1 file changed, 13 insertions(+)

Are you sure gl2ext.h is the right place to add those typedefs ? In the
mesa3d-headers package (which kind of is our reference for OpenGL
Khronos headers), these typedefs are in GLES2/gl2.h.

Another question is: why is GStreamer defining:

 39 #if !GST_GL_HAVE_GLSYNC
 40 typedef gpointer GLsync;
 41 #endif

Isn't that a mistake from the GStreamer code base ?

Did you report this issue to rpi-userland upstream to see what they think ?

Best regards,

Thomas

Hello,

Le 06/08/2021 à 00:07, Thomas Petazzoni a écrit :
> Hello,
> 
> On Mon, 10 May 2021 21:53:52 +0200
> Peter Seiderer <ps.report@gmx.net> wrote:
> 
>> + interface/khronos/include/GLES2/gl2ext.h | 13 +++++++++++++
>> + 1 file changed, 13 insertions(+)
> 

This issue can still be reproduced from Buildroot master branch.

> Are you sure gl2ext.h is the right place to add those typedefs ? In the
> mesa3d-headers package (which kind of is our reference for OpenGL
> Khronos headers), these typedefs are in GLES2/gl2.h.
> 
> Another question is: why is GStreamer defining:
> 
>  39 #if !GST_GL_HAVE_GLSYNC
>  40 typedef gpointer GLsync;
>  41 #endif
> 
> Isn't that a mistake from the GStreamer code base ?
> 
> Did you report this issue to rpi-userland upstream to see what they think ?

Khronos headers are the same since 10 years:

https://github.com/raspberrypi/userland/tree/master/interface/khronos/include/GLES2

In 2014, mesa3d updated Khronos headers to revision 24614 where GLsync moved
from gl2ext.h to gl2.h.

https://gitlab.freedesktop.org/mesa/mesa/-/commit/117d8ce27b7929c42f46f818ac038580877b3086

https://gitlab.freedesktop.org/mesa/mesa/-/commit/d519ebb34cd21cfc13987f2038d7a2ec353a5cfd

I guess it's not only a GLsync issue but a too old khronos headers.

Best regards,
Romain


> 
> Best regards,
> 
> Thomas

Hello Peter,

On Mon, 10 May 2021 21:53:52 +0200
Peter Seiderer <ps.report@gmx.net> wrote:

> The problem is that gstreamer sees the original gl2ext.h file from
> rpi-userland (rpi-userland-093b30bbc2fd083d68cc3ee07e6e555c6e592d11/interface/khronos/include/GLES2/gl2ext.h)
> without the additional GLint64, GLuint64 and GLsync typedef definitions but
> checks for existance and if not found enables its own versions in gstgl_compat.h,
> incompatible with the later ones used from Qt (qopengles2ext.h).
> 
> Fix this by adding the same typedef definitions already to the
> rpi-userland header version.

Romain and I finally had a look at this. I was a bit confused by your
explanation, so after doing our investigation, I rewrote it this way:

    The problem is that rpi-userland doesn't define GLsync, and both
    GStreamer and Qt have their own definition of GLsync in this case, but
    they are not the same.
    
    We reported this issue to:
    
     * rpi-userland, to get the headers updated:
       https://github.com/raspberrypi/userland/issues/469#issuecomment-1193864294
    
     * gstreamer, to get their bogus definition of GLsync fixed:
       https://gitlab.freedesktop.org/gstreamer/gst-plugins-base/-/issues/973
    
    In the mean time, fix this by adding the missing definitions to
    rpi-userland, so that GStreamer and Qt don't try to provide their own.
    
As you can see, I also reported the issue back to rpi-userland (more
precisely pinged in an existing issue) and to GStreamer.

But for now, what you proposed is the most immediate fix, so I've
applied your patch to master with an updated commit log and patch
description.

Thanks a lot!

Thomas

>>>>> "Thomas" == Thomas Petazzoni via buildroot <buildroot@buildroot.org> writes:

 > Hello Peter,
 > On Mon, 10 May 2021 21:53:52 +0200
 > Peter Seiderer <ps.report@gmx.net> wrote:

 >> The problem is that gstreamer sees the original gl2ext.h file from
 >> rpi-userland (rpi-userland-093b30bbc2fd083d68cc3ee07e6e555c6e592d11/interface/khronos/include/GLES2/gl2ext.h)
 >> without the additional GLint64, GLuint64 and GLsync typedef definitions but
 >> checks for existance and if not found enables its own versions in gstgl_compat.h,
 >> incompatible with the later ones used from Qt (qopengles2ext.h).
 >> 
 >> Fix this by adding the same typedef definitions already to the
 >> rpi-userland header version.

 > Romain and I finally had a look at this. I was a bit confused by your
 > explanation, so after doing our investigation, I rewrote it this way:

 >     The problem is that rpi-userland doesn't define GLsync, and both
 >     GStreamer and Qt have their own definition of GLsync in this case, but
 >     they are not the same.
    
 >     We reported this issue to:
    
 >      * rpi-userland, to get the headers updated:
 >        https://github.com/raspberrypi/userland/issues/469#issuecomment-1193864294
    
 >      * gstreamer, to get their bogus definition of GLsync fixed:
 >        https://gitlab.freedesktop.org/gstreamer/gst-plugins-base/-/issues/973
    
 >     In the mean time, fix this by adding the missing definitions to
 >     rpi-userland, so that GStreamer and Qt don't try to provide their own.
    
 > As you can see, I also reported the issue back to rpi-userland (more
 > precisely pinged in an existing issue) and to GStreamer.

 > But for now, what you proposed is the most immediate fix, so I've
 > applied your patch to master with an updated commit log and patch
 > description.

Committed to 2022.05.x and 2022.02.x, thanks.