Example:

>>> from poplib import POP3
>>> p = POP3("localhost")
>>> p.user("user")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/py3k/Lib/poplib.py", line 179, in user
    return self._shortcmd('USER %s' % user)
  File "/py3k/Lib/poplib.py", line 151, in _shortcmd
    self._putcmd(line)
  File "/py3k/Lib/poplib.py", line 98, in _putcmd
    self._putline(line)
  File "/py3k/Lib/poplib.py", line 91, in _putline
    self.sock.sendall('%s%s' % (line, CRLF))
TypeError: sendall() argument 1 must be string or buffer, not str
>>> p.user(b"user")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/py3k/Lib/poplib.py", line 179, in user
    return self._shortcmd('USER %s' % user)
  File "/py3k/Lib/poplib.py", line 151, in _shortcmd
    self._putcmd(line)
  File "/py3k/Lib/poplib.py", line 98, in _putcmd
    self._putline(line)
  File "/py3k/Lib/poplib.py", line 91, in _putline
    self.sock.sendall('%s%s' % (line, CRLF))
TypeError: sendall() argument 1 must be string or buffer, not str

Here is a patch proposition:
 - a socket uses bytes
 - makefile() creates an unicode file using 'r' mode
 - default encoding ISO-8859-1 because I guess that most servers use 
this encoding, but you can change the encoding using "encoding" 
constructor optioan argument
 - read unicode and write unicode: convert convert from/to bytes at 
the last moment (just after/before reading/writing the socket)
 - cosmetic: use .startswith() instead of for example b[:2] == '..'

Test updates:
 - replace "localhost" by HOST
 - write a test for a logging (user + password)

Missing: no SSL unit test. I tested SSL on my personal POP3 account, 
but only the login.

Ooops, my previous patch was wrong (startswith => not startswith).

I tested python trunk test for poplib: with minor changes, all tests 
are ok except tests using SSL. I get a "select.error: (9, 'Bad file 
descriptor')" from asyncore. So I tried to synchronize python3 ssl 
with python2 trunk, but it depends on python2 trunk version of the 
socket module and this module is very complex and hard to port to 
python3.

About EBADF error from select(), it may comes from missing makefile() 
method of the ssl socket wrapper.

New version:
 - fix SSL: self.file contains the SSL socket and not the raw socket!
 - upgrade test_poplib.py from Python trunk

poplib should be refactored to reuse the new IO library. But I don't 
know how to build a TextIO wrapper for a socket. Using TextIO, it 
would be possible to remove newline (CR/LF) and unicode 
encoding/decoding code.

I haven't tried the patch but I think that "encoding" should be a class
attribute as it is in ftplib and similar py3k network related modules.

New version:
 - remove duplicate methods of POP3_SSL()
 - use makefile('r', encoding=self.encoding) to get a nice text 
wrapper with universal newline
 - remove newline conversion (done by TextIOWrapper)

Finally my patch removes more code in poplib.py than it adds :-D I 
like such patch.

Python3 new I/O library rocks!

@giampaolo.rodola: Right, I also prefer encoding as a class attribute. 
So I wrote a new patch:
 - encoding is now a class attribute
 - continue SSL code factorization: SSL code is now around 10 lines 
instead of 70 lines!

I like this patch.

Oooops, I removed the message74562 from giampaolo.rodola, sorry:
"As for issue #3911 this is another module for which an actual test 
suite would be very necessary."

After testing real world message, my patch using pure unicode doesn't 
work. The problem comes with message encoding with "8-bit" encoding. If 
the email charset is different than POP3.encoding, the message in not 
decoded correctly.

New patch: resp() returns bytes
 - self.file is now a binary file
 - encode commands using POP3.encoding charset, default is UTF-8
 - use md5.hexdigest()
 - factorize POP3_SSL code: code specific for SSL is just the creation 
of the socket

The default charset is UTF-8, but most servers only accept pure ASCII 
login/password (eg. gmail.com) or a smaller subset of ASCII (eg. only 
A-Z, a-z, 0-9 and some ponctuation signs :-/). If you user non-ASCII 
login/password and your server doesn't use UTF-8, change POP3.encoding 
or <your pop object>.encoding (encoding is not used in the 
constructor).

welcome attribute (and getwelcome() results) is a bytes string.

You have to parse the message headers to get the right charset to 
decode bytes to unicode characters. A multipart message may contains 
two or more charsets and different encoding. But poplib is not 
responsible to decode messages, just to download a message as bytes.

Arguments (username (login), password, a message identifier) are 
unicode strings. For a message identifier, you can also use an integer 
(nothing new, it was already possible).

I hope that apop() works. No Python error is raised but no server does 
support his authentication method. I tested 3 servers 
(pop3.haypocalc.com, pop.laposte.net and pop.gmail.com) and none 
supports APOP. I tested POP3 and POP3_SSL (gmail requires SSL).

About apop(): the second argument is the user password, not a "shared 
password" which is the local variable "timestamp" read from welcome 
attribute.

I forgot the new unit tests. New patch:
 - port python trunk unit tests

Can anyone review my last patch (poplib-bytes-2.patch)?

I'm happy with Victor's patch.

Benjamin's reviewed this and the only thing that jumps out at me is some
funky indentation at about line 331.  If you fix that, you can land this
patch.

Le Tuesday 04 November 2008 01:03:42 Barry A. Warsaw, vous avez écrit :
> Benjamin's reviewed this and the only thing that jumps out at me is some
> funky indentation at about line 331

It's not related to my patch (I did'nt change POP3_SSL comment). But well, as 
you want: a new patch re-indent the "See the methods of the parent (...)" 
line (331).

Patch applied in r67109