That's what the GP is saying. You can't really tell a priori what encoding to use for a stream of "text" (bytes). Without some sort of metadata about the stream

That's what the GP is saying. You can't really tell a priori what encoding to use for a stream of "text" (bytes). Without some sort of metadata about the stream you just have to guess. Convention will help you make an informed guess but it's not guaranteed to be correct. Then stuff breaks in unexpected and stupid ways.