Right now we default to "all" which gives us a situation much like before,
but when the "typical" option is implemented, we can change the default and
reduce the number of visuals the GLX module bloats the X server with.
XIDs for GLX visuals and FBConfigs used to be interchangable and the list of
GLX visuals was identical to the list for FBConfigs. This patch splits handling
of these two data types and allows the X server to pick and choose the FBConfigs
that are exposed as visuals.
Instead of the fragile setup where we filter the modes common between the
DDX generated GLX visuals and the DRI driver generated fbconfigs, we now
just take the fbconfigs returned by the DRI driver to be our supported set.
* GL/glx/glxvisuals.c: added boolean to disable
calling init_visuals(). This gives a chance to Xephyr
to take over visuals manipulation and avoid a crash at
server shutdown in __glXMesaScreenDestroy(), due to the fact
that mesa might sees more visual than what it has actually created in
init_visuals(). It might see more visuals because Xephyr can augment
the number of visuals, dynamically.
* os/utils.c: the boolean is actually defined here.
This reverts commit 2243b30e54. The existing
DRI interface doesn't let us get from a __DRIdrawable to the corresponding
X drawable, and thus, we can't implement AIGLX damage tracking with the
current interface.
This prevents situations where the server doesn't use the target the
client thinks it does, usually resulting in the texture being sampled as all
white.
The previous scheme didn't work when the client didn't create the core drawable,
e.g. the root or composite overlay window. Use refcounting via special client
resources to fix that.
This is to avoid issues with redirected windows which are located partly or
fully outside of a screen edge, resulting in unusual cliprects which the 3D
drivers generally can't handle. The symptoms in such cases would be incorrect
rendering or even crashes or hangs.
When available, use the 2D driver texOffsetStart hook and the 3D driver
setTexOffset hook to save the overhead of passing the pixmap data to
glTex(Sub)Image.
The basic idea is to update the driver specific 'offset' for bound pixmaps
before dispatching a GLX render request and to flush immediately afterwards
if there are any pixmaps bound. This should ensure that the 3D driver can
use pixmaps for texturing directly regardless of the X server moving them
around.
Previously, the new mode was added at the head of the list. This caused the
positional correspondence between modes and the XMesaVisuals array to be off
by one. The net result was GLX clients failing when they tried to use the
last GLX mode/visual.
We still have the problem of DRI drivers not being able to use the extra
mode/visual introduced by __glXCreateARGBConfig(). glXCreateContext fails
with BadAlloc if it's attempted. This is also the source of the often-
seen warning "libGL warning: 3D driver claims to not support visual xxx"
Look into fixing that someday...
As the driver EnterVT function generally re-enables the hardware and
prepares it for rendering, it must be called before any gl functions are
called which could touch the hardware.
Use newly added XMesaCopyContext() and drop the GlxSetRenderTables() call
for Xgl, as this is now done inside XMesaForceCurrent(). This leaves xmesaP.h
but only for the declarations of the three XMesa/XFree86 functions. Also,
GlxSetRenderTables() stays but is only used in hw/xgl/glxext/ .
Also drop xf86glxint.h, no longer used.
Depends on mesa commit 7439a36785b6a2783e80a40a96c09db8f56dc2bc of 2007-03-30.
Move the byte-order related wrappers out of the individual source
files into a dedicated header file. Modify the single hand-coded
source file that uses the byte-order wrappers to use the new header
file.
- Use autoconf tests instead of platform-specific #ifdef's to decide
which macros to use.
- Provide fallbacks for platforms like Solaris that don't provide any
of the existing known forms.
defining __GLIBC__, as done already for Linux and Hurd (cf. bug #5613).
Also includes some more __GNU__ checks on top of those made in
commit ade4bf0907.
Thanks Petr Salinger (Debian bug #400869).
The logic for freeing GL contexts introduced by "Fix AIGLX VT switching." is
inverted. As it is now, GL context destruction is deferred for glxDRIEnterVT().
