@julliard: what's the status of this MR?
thinking about it, perhaps you'd prefer that we introduce a seond method for guessing the format (based on fd, not requiring to read the whole file in memory), so that we don't change the detection code for all the supported formats but only for PDB