Fix for the previous commit: we have to count the bytes of the header that we skip to properly continue the next reading. Not only files can provide header