To be honest, your suggestion to put the check inside framebuffer_check() sounds like a better idea: we wouldn't even hit any driver-specific code-path when the check fails. Daniel, do you think there'd be an issue with this approach?