BUGS
OpenerDirector.retrieve() should be a function, not a method.
HTTPRobotRulesProcessor uses RobotFileParser.read(), which uses urllib.
 Switch to urllib2.
Max-Age handling should take account of RFC 2616 age rules, according to
 RFC 2965.  Would be nice to get rid of ._now instance attribute if &
 when fix this (be careful with maxage=0 case -- means delete cookie).
De-facto rules may have gotten slightly saner with recent MSIEs, so might
 be possible to simplify and/or tighten up some things.  I haven't tested
 much.  eg.:
 MS docs claim that (eg.) .co.jp (two dots but country-code TLD) is
  blocked.
 MS docs: "Session cookies that specify a domain are rejected".  I don't
  remember that, and I don't understand what the purpose would be...
 Support any more bits of RFC 2109 that IE / Mozilla have added to their
  Netscape protocol implementations (there may not be any, dunno).
Sketch overlap with P3P protocol in docs, once I figure out exactly what
 the relationship is.  I think IE 6 and Mozilla implement it.

TESTS
Better round-trip tests for LWPCookieJar and MozillaCookieJar.
Test MSIECookies on Windows (only tested on wine recently).
Test thread safety.  How??
Add doctest running to test.py -- see Zope's test.py.
Add pychecker to release script somehow?  Need to figure out how to
 suppress warnings -- see Zope's test.py.
Write more tests.

PLANNED FEATURES
(some of this stuff is really for the standard library rather than
 this package -- eg. file upload function)
Add a little processor in the docs that shows how to deal with pages that
 should have had a non-200 code, but don't.
HTTP connection caching.
Timeouts (PEP 42, bug 723287).  Document lack of DNS timeout (say should
 use separate process?? -- that's what squid does).
File upload function(s).  Remember need content-type HTTP header, not
 just request body data.
 Conditional fetches?  Negotiation / charset?
Embedded object fetcher (eg. images) for functional testing.
305 redirections.
HTTPS / CONNECT (see 515003)?
gzip / deflate / compress encoding processors (send Accept-Encoding (but
 doesn't httplib send this too?), unzip returned data if it *is*
 gzipped/zipped (Transfer-Encoding & optionally Content-Encoding)).
Weird schemes:
 data: urllib implements it, but does anybody use it?  Not in MSIE.
 nttp?
 imap??
TidyProcessor -- mxTidy / uTidylib (wait until tidylib & Python wrappers
 stabilise a bit).
Konqueror cookie file support (~/.kde/share/apps/kcookiejar/cookies).
Opera cookie file support (http://www.opera.com/docs/fileformats/).
Support for cookperm.txt (Mozilla file format; cookie permissions,
 presumably block / accept list).
Add a save method to MSIECookieJar? -- just iterate through
 self._cookies calling InternetSetCookie() for each.
Use wininet InternetSetCookie / InternetGetCookie (ctypes).
Link DOMForm's JavaScript handling to CookieJar.
Maybe MSIE stuff is appropriate for win32all once rest is in std. lib?

TODO FOR PYTHON STD. LIB.
Before 2.4b1:
 Upload final patch to docs, tests (?) and code.

Put up a page documenting 2.4 / 0.9.x differences on web, link from
 README.html.

Release 1.0

Python 2.4 version known warts:
 Thread synchronization needs checking by someone with more of a clue
  about this.
  ._now attributes!

Future:
 Rest of stuff: UserAgent, HTTPEquivProcessor, HTTPRefreshProcessor,
  etc.
 What should happen with seek_wrapper?  It's needed for
  HTTPEquivProcessor.  Could use Andrew Dalke's ReseekFile instead,
  which is simpler.  Should it be a separate module?