ERR is used to indicate internal inconsistencies in vkd3d. Here that's
not the case, we simply have to forward the error condition to the
caller.
This fixes failures on the CI with llvmpipe, because the build we use is
compiled without support for VK_KHR_surface and related extensions.
This unfortunately introduces a lot of failures, because it turns
out there is still work to do there. But at least we can
estimate how bad we're doing.
The previous Intel-based macOS runner was replaced with another one
running as a virtual machine on top of an Apple Silicon host.
Since the current macOS runner is not yet able to deal with
different exit codes, we temporarily allow failure unconditionally.
This will be reverted as soon as the runner issue is fixed.
Nowadays vkd3d is probably most commonly used compiled as PE rather then
ELF, so it makes sense to at least ensure that compilation succeeds.
In the future it would be nice to somehow test these binaries as well.
Mostly to avoid polluting other logs and artifacts, and also to avoid
recompiling crosstests over and over. Eventually the artifacts produced
at this stage should be run on native Windows.
In order for this to work it is expected that the Docker host exposes
the devices in /dev/dri to the guest system, and that the render nodes
have GID 800 (usually that would be the "render" group, but the GID
for that group is dynamically assigned, so we explicitly agree on a
fixed number).
