Echoing the comment of Thomas Viehmann, the current behaviour makes it impossible to use this library to generate a correct multipart/mixed message with 7_or_8_bit encoding on Unix.

The MIME standard specifies that headers are to be CRLF terminated - this is independent of any lower-level transport encoding.  

Binary bodies are simply octet sequences - so may contain \n or \r characters.

The correct MIME encoding would have the headers terminated with CRLF, and the bodies as raw byte sequences.  However on Unix, since the headers are generated with \n not CRLF, you can't even post-process this (e.g. message.as_string().replace('\n', '\r\n') as that will screw up the binary body data.

The current behaviour is basically wrong and should be fixed.