Fredrik, you are obviously correct that most operations between normal and unicode strings don't work if the normal string contains non-ascii data.

I still do think that a UnicodeWarning like you get from "chr(128) == u'foo'" would be nicer than an exception and prevent problems like the one in telnetlib [1]. If an exception is raised I don't care too much about its type but a better message would make debugging possible problems easier.

[1] https://sourceforge.net/tracker/index.php?func=detail&aid=1772794&group_id=5470&atid=105470