test_ftplib now passes, but it only tests socket timeouts...
My corrections were based on runs like:
python -m ftplib ftp.gnu.org -l emacs

Note: In this kind of code, byte objects make the code more difficult:
- it is easy to forget the b prefix 
eg:    resp[:3] in (b'250', '200')
- getitem does not return a single byte:
eg:    resp[0] == b'2'

Or should ftplib decode all the lines it receives?

You're right, it's ugly. I wonder if it makes more sense to change the makefile() call to use 'r' for the mode, and adjust the various put* methods to convert their argument to bytes?  We could use "ASCII", "Latin-1" or "UTF-8" for encoding, I don't really know if there are ftp servers that do anything with non-ASCII characters or encodings.  Maybe adding the encoding as an option would be useful.

I've checked in minimal changes that make ftplib work along those lines.