I don't know where this code comes from, so I may only guess. I'd say that the author needed a binary buffer - a buffer of bytes which acts like a binary file. This is the io.BytesIO instance. But csv.writer() works with text and expects a text file. The io.StringIO instance acts like a text file. So the contents is written into the text file-like object (buff) first and then encoded into bytes and wrapped in a binary file-like object (buffer_2).