The series ending with mesa d61d259 has the following effect on vulkan tests as run in Intel's CI: Crucible: fails to link, with link errors such as crucible/src/framework/test/t_phases.c:87: undefined reference to `vkQueueWaitIdle' Vulkancts: all tests crash
This is because, with this series, you really have to start using the loader which I don't think we always do with CI. I think we actually want to make whether or not we expose the API a compile-time flag. When we start to develop extensions or 1.1 features (whenever that happens), being able to use the driver without a loader is a useful feature. We should definitely hide the API by default and exposing it should yield loud warnings but I think it's useful none the less.
Wouldn't it be better to use the loader ? It would seem counter intuitive to test in a way unlike any Linux distro. Sure that means that the CI might be ever so slower due to the loader overhead, but the tests would be performed in a closer to a real world scenario. Furthermore I'd imagine that CTS has/will get a test for the negotiation API (between loader and ICD) which one wants to cover as well ?
So presumably we're now using the loader in the CI system? Is there anything else to do here?
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.