This strategy avoids the non-intuitive (and other problematic aspects) of introducing the dev_t parameter in the dmabuf public functions, and also the strategy of including information in dmabuf module that really belongs to the gbm module. I'm not sure how to evaluate the tradeoffs involved (wayland_dmabuf using the exported wayland_gbm_get_render_dev() from wayland_gbm, thus introducing a dependency). I do like the much cleaner result of it, though.