> From a quick look at the patch, if you call login() twice, the socket
> will be wrapped twice as well? Perhaps auth_tls() should have a
> protection against this.

You're right. Done.

> In prot_p() and prot_c(), it seems that self._prot_p is updated
> unconditionally, regardless of the FTP response.

Both PROT P and PROT C commands expect a 2xx response.
That's why I used voidcmd(). If a response different than 2xx is 
received voidcmd() automatically raises an error_reply exception already 
and self._prot_p doesn't get set.

> One more question, why is "ssl_version=ssl.PROTOCOL_TLSv1" hardwired?

You're right, it shouldn't be.
This is now provided as a class attribute.
The reason I used PROTOCOL_TLSv1 instead of PROTOCOL_SSLv23 (the ssl.py 
default) is because RFC-4217 recommends it.

> In retrbinary(), retrlines(), storbinary() and storlines(), you
> certainly want some try/finally blocks so that the data connection
> always gets closed at the end.

I agree, it's a good practice.
I avoided to do that because the original FTP class is coded like that.
Anyway, this is done too now.

Modified patch is in attachment.

I hope I'll be able to work on tests and documentation during this week-
end.