Update from the latest gnome-control-center version.

This reverts commit 69b121b0.

https://docs.mesa3d.org/envvars.html#envvar-DRI_PRIME clearly says:
"It applies to OpenGL and Vulkan"

For machines with dual-GPUs by 2 different vendors, setting the
VK_LOADER_DRIVER_SELECT envvar reduces the GPUs it can detect to those
of that vendor, effectively forcing a particular GPU to be chosen.

For machines with dual-non-Mesa-GPUs, the variable will have no effect,
and things will work as if no selection had been made, just like for
NVidia GPUs.

Note that this will only work when the VulkanLoader in the application's
libraries are new enough to include this change:
https://github.com/KhronosGroup/Vulkan-Loader/pull/984

Closes: #23

Use MESA_VK_DEVICE_SELECT to select a particular Mesa supported GPU when
launching Vulkan applications.

This configuration option is available since Mesa 20.0 from commit
9bc5b2d ("vulkan: add initial device selection layer. (v6.1)")

In file included from /usr/include/glib-2.0/glib.h:114,
                 from ../src/info-cleanup.c:21:
In function ‘g_autoptr_cleanup_generic_gfree’,
    inlined from ‘info_cleanup’ at ../src/info-cleanup.c:123:20:
/usr/include/glib-2.0/glib/glib-autocleanups.h:28:3: warning: ‘pretty’ may be used uninitialized [-Wmaybe-uninitialized]
   28 |   g_free (*pp);
      |   ^~~~~~~~~~~~
../src/info-cleanup.c: In function ‘info_cleanup’:
../src/info-cleanup.c:123:20: note: ‘pretty’ was declared here
  123 |   g_autofree char *pretty;
      |                    ^~~~~~

Closes: #27

We never used autotools or libtool.

In https://github.com/python/cpython/commit/664448d81f41c5fa971d8523a71b0f19e76cc136
the "errors" attribute was removed from the _Outcome object. Adapt the
code to print the daemon log after a failed run().

This reverts commit de360742.

The Vulkan support breaks when the exact paths don't exist in the app's
view of the filesystem, such as when containers are used and have a
different hierarchy than the host.

Closes: #30

This reverts commit d3d83998.

This works around non-Intel platforms not setting the "boot_vga"
sysfs attribute.

On a Raspberry Pi 3.

Before:
$ switcherooctl
$
After:
$ switcherooctl
Device: 0
  Name:        Unknown Graphics Controller
  Default:     no
  Environment: DRI_PRIME=platform-soc_gpu

/usr/include/glib-2.0/glib/gmacros.h: In function ‘get_drm_cards’:
/usr/include/glib-2.0/glib/gmacros.h:1212:17: warning: ‘parent’ may be used uninitialized [-Wmaybe-uninitialized]
 1212 |     { if (_ptr) (cleanup) ((ParentName *) _ptr); }                                                              \
      |                 ^
../src/switcheroo-control.c:367:32: note: ‘parent’ was declared here
  367 |         g_autoptr(GUdevDevice) parent;
      |                                ^~~~~~
/usr/include/glib-2.0/glib/gmacros.h:1212:17: warning: ‘parent’ may be used uninitialized [-Wmaybe-uninitialized]
 1212 |     { if (_ptr) (cleanup) ((ParentName *) _ptr); }                                                              \
      |                 ^
../src/switcheroo-control.c:339:32: note: ‘parent’ was declared here
  339 |         g_autoptr(GUdevDevice) parent;
      |                                ^~~~~~
/usr/include/glib-2.0/glib/gmacros.h:1212:17: warning: ‘parent’ may be used uninitialized [-Wmaybe-uninitialized]
 1212 |     { if (_ptr) (cleanup) ((ParentName *) _ptr); }                                                              \
      |                 ^
../src/switcheroo-control.c:289:32: note: ‘parent’ was declared here
  289 |         g_autoptr(GUdevDevice) parent;
      |                                ^~~~~~

display-manager.target doesn't exist, but display-manager.service does.
That's the one we want to start before.

Closes: #21
Fixes: 3592c1c4

For machines with dual-GPUs by 2 different vendors, setting the
VK_ICD_FILENAMES envvar reduces the GPUs it can detect to those of that
vendor, effectively forcing a particular GPU to be chosen.

For machines with dual-GPUs, the variable will have no effect, and
things will work as if no selection had been made, just like for NVidia
GPUs.

Closes: #23

WARNING: You should add the boolean check kwarg to the run_command call.
         It currently defaults to false,
         but it will default to true in future releases of meson.
         See also: https://github.com/mesonbuild/meson/issues/9300

/etc should be left for sysadmin overrides.

Closes: #26

unittest_inspector.py lists the tests in the integration-test.py script,
which are then added as individual tests.

Add the python suffix.

Filippe LeMarchand authored Jun 02, 2021 and Bastien Nocera committed Jun 09, 2021

The __NV_PRIME_RENDER_OFFLOAD env var causes the special Vulkan layer
VK_LAYER_NV_optimus to be loaded which, in turn, reorders GPUs reported
to the app so that Nvidia GPUs come first. The app is supposed to
select the first GPU, but it appears to be not always the case,
especially for Wine.

To make sure that Vulkan apps always use the Nvidia dGPU, set
__VK_LAYER_NV_optimus=NVIDIA_only, which tells the special layer
to report only the Nvidia GPUs, thus, leaving the app no other choice.

Closes: #22

Two tests with the same name meant that one of those wasn't getting run.

--gpu=X was documented, but --gpu X was expected. Fix both to work.

The COPYING file mentions GPLv3+ but this file said GPLv3 only. I'm the
sole author of this file, so there aren't any problems fixing this
problem summarily.

Closes: #18

It shouldn't need to always be relative to /usr (eg. the prefix).