For the implementation strategy a somewhat relevant question on a longer time scale is how we want to handle multi-platform in the future. Currently there is only one Platform instance active per session and in the important use case this is mostly the Drm backend. But is there still a case to be made, where more than one Platform at the same time could make sense?

See also T5709.

API-wise, the best option is to let platform do PRIME stuff, i.e. have only one instance of platform. Apparently this would require some refactoring in the drm backend, e.g. there have to be a thing that represents each drm device, some manager that orchestrates devices (if this part gets done correctly, it should be quite easy to add support for GPU hot-plugging), etc.