Logged In: YES 
user_id=6380

Methinks that the fix isn't quite right: it would
incorrectly recognize as a boundary a very long line
starting with "--" followed by the appropriate random string
at offset 2**16. This could probably be taken care of by
adding a flag that is true initially and after that keeps
track of whether the previous line ended in \n.

Also, there's a call to fp.readline() in parse_multipart()
that you didn't patch -- it wouldn't help because that code
is saving the lines in a list anyway, but isn't that code
vulnerable as well? Or is it not used?