Is there any objection to making the DEV_TTY_P macro (and any similar things that get added) take a struct stat * argument instead of dev_t? We might want to add the optimized isatty check for the Hurd, but the test would be on different st_* fields, not just on st_rdev.