Re: [Mesa3d-dev] Initializing extensions

[Brian P. <br...@tu...>]

Ian Romanick wrote:
> Brian Paul wrote:
> 
>> This appears to be a bug in the XFree86/DRI libGL.  I just 
>> double-checked and libGL doesn't have the GLX encoder support for 
>> glPixelTexGenSGIX() so it shouldn't be reporting GL_SGIX_pixel_texture 
>> in the extensions string (which is the intersection of the client-side 
>> and server-side extensions lists).
>>
>> Ian Romanick might have fixed this recently.
>>
>> By the way, Mesa/tests/getprocaddress.c is a test program which should 
>> detect problems like this.  It has a list of GL extensions and the 
>> corresponding extension functions.  It uses glXGetProcAddress() to be 
>> sure all the functions that are supposed to be present really are.
>>
>> It works fine with stand-alone Mesa.
>>
>> With NVIDIA's 41.91 driver it segfaults.  Running with GDB it seems to 
>> get stuck somewhere in the driver.
>>
>> I don't have a DRI configuration up and running at the moment to 
>> test.  Maybe someone else can.  Ian?
> 
> 
> All of the stuff that I was doing was with the GLX extension string.  I 
> took some time to look at this while I was merging the DRI trunk to the 
> texmem-0-0-1 branch.  It seems that the problem is in the implementation 
> of glGetString (line 927, lib/GL/glx/single2.c).  Basically, it just 
> returns to the client EXACTLY what the server gives it.  This is wrong 
> for both GL_EXTENSIONS and GL_VERSION.
> 
> Fixing GL_EXTENSIONS should be as easy as making combine_strings (in 
> glxcmds.c) global and calling it on the returned string and the internal 
> extension list.  If that is done, then __glXGLClientExtensions should be 
> expanded to include all of the extensions that are supported by libGL.
> 
> Fixing GL_VERSION is a bit more tricky.  Drivers classically add on 
> other stuff to the version string (driver dates, settings, etc).  Would 
> something like this be okay?
> 
>    if ( atof( __glXGLClientVersion ) < atof( server_version ) ) {
>       gc->version = __glXGLClientVersion;
>    }
>    else {
>       gc->version = server_version;
>    }
> 
> Or should something be done to add in the extra struff after the proper 
> version?

I could have sworn this was working correctly at one point.

Sounds like you've got the right idea for fixing it.  Go ahead.

-Brian