If you pass csv.reader() a filename as its first argument:
csv.reader('filename')
instead of a file object like you're supposed to, you don't get an error.
You instead get a reader object which returns the characters which make up
the filename.
Technically, this is not a bug, since the documentation says, "csvfile can
be any object which supports the iterator protocol and returns a string
each time its next method is called", and a string meets that definition.
Still, this is unexpected behavior, and is almost certainly not what the
user intended. It would be useful if a way could be devised to catch this
kind of mistake.