GitLab

Fix runtime DRI pagesize assumptions in the r128 and radeon drivers.  This is
particularly important on ia64 and other architectures that have a configurable
pagesize.  Otherwise, the X server is hard coded at build time to a particular
pagesize which may not match the installed kernel.

Reported by: Dieter Nutzel

Hein for pointing out the problem.

Bug report:     http://freedesktop.org/bugzilla/show_bug.cgi?id=788

extensions are *not* currently enabled.  This is for several reaons.  First,
the server-side does not support texture compression, so there's no
advantage to supporting it (yet) on the client-side.  Second, I tried
testing it with an IHV's closed-source driver (that advertises GL 1.3), but
their server-side driver doesn't support the protocol.  That means this code
is largely untested.

following discussion on mailing list.

MESABUILDDIR

__glXGLClientExtensions.  This makes it so that the list of extensions
supported by the client only has to be maintained in one place.

xc tree. The driver is built as via_dri.so as opposed to unichrome_dri.so when
it is built in the Mesa tree. This is because all known 2d drivers expects
this name.

__glXGetStringFromServer.  Other functions will likely be added in the
not too distant future.  Modify surrounding code to use
__glXGetStringFromServer.

As an aside, this fixes a long standing, dormant bug in
__glXInternalQueryServerString.  If memory could not be allocated to
hold the string retured by the server, _XEatData would be called with
'length'.  This is wrong.  'length' measures the size of the packet in
CARD32 units, but _XEatData expects the size to be in CARD8 units.

__glXGLClientVersion is not needed.

string but are implied by the core GL version number.  This code is
not actually enabled, but it's all there.  Comments in glGetString (in
single2.c) explain it all.

enabled.

can also be used for GL extensions.  Modify glGetString to use the new
mechanism.  This allows us to, among other things, enable extensions
that are implemented entirely client-side (i.e.,
GL_ARB_transpose_matrix).

reports GLX 1.3 support.

rendering ONLY.  It was tested with Mesa's pbdemo and an Nvidia driver.
Both the SGIX and GLX 1.3 protocol seemed to work.

* Fill in the size/mask values for the framebuffer by hand, because the
  automatic filling didn't seem to be complete.  Matches other drivers.
* Don't use an alpha channel in accum if framebuffer lacks alpha.
* Mark visuals with accum buffer as GLX_SLOW_CONFIG.

in the Mesa tree so that the Rage128 DRI driver will build.

bump the API version.  This is done so that DRI_NEW_INTERFACE_ONLY
drivers, which *need* that function can bail early with a clear message.

that are called directly by DRI drivers.

XF86DRIDestroyDrawable are directly called by drivers using the new
DRI interface, convert them to types available in the driver.  The
prototypes will also be duplicated in dri_interface.h so that drivers
do not need xf86dri.h anymore.

drm_handle_t, drm_context_t, drm_drawable_t, drm_magic_t.

drm_clip_rect_t.  All of the Imakefile changes are so that drm.h can
be found.  DRM SOURCES (i.e., shared/drm.h) MUST ALSO BE UPDATED FOR
THIS TO COMPILE!

Soon the files in lib/GL/dri/drm will be removed (and libdrm.a will be
built from the "pristine" DRM source in the drm module of DRI CVS) and
the while lib/GL/dri tree will be gone.