THIS IS NO LONGER MAINTAINED!  Use the subversion repository instead.

This isn't really in proper GNU ChangeLog format, it just happens to
look that way.

2004-05-01 John J Lee <jjl@pobox.com>
	* Released 0.4.19 and 0.9.4a.

2004-04-24 John J Lee <jjl@pobox.com>
	* 0.9.x:
	* Fixed various MSIECookieJar bugs.
	* Fixed case of local Netscape cookie where domain is not an IP
	  address (was not getting returned).
	* Simplified CookieJar code :-)

2004-04-06 John J Lee <jjl@pobox.com>
	* 0.9.x:
	* Reverted to old httplib.HTTP instead of httplib.HTTPConnection
	  class, since use of the latter was broken
	* Added OpenerFactory.

2004-01-18 John J Lee <jjl@pobox.com>
	* 0.9.x:
	* Added urlretrieve() function and
	  ClientCookie.OpenerDirector.retrieve() method, which work like
	  their equivalents in urllib.
	* AbstractHTTPHandler.do_open() now uses httplib.HTTPConnection,
	  following Jeremy Hylton's patch in Python CVS.
	* Initial port of ConnCache.pm from libwww-perl.  Not yet tested
	  or integrated.

2004-01-16 John J Lee <jjl@pobox.com>
	* Added HTTPRobotRulesProcessor.
	* Rewrote one of the examples in the documentation to reflect
	  various changes that have gradually crept up on me.
	* Small doc fixes.

2004-01-05 John J Lee <jjl@pobox.com>
	* test_urllib.HandlerTests.test_file now passes on Windows.
	* Released 0.4.18 and 0.9.3a.

2004-01-04 John J Lee <jjl@pobox.com>
	* Changed return True to return 0 in BaseProcessor.__cmp__.
	* Fixed http_error_302.  Got rid of .original_url, treat
	  .origin_req_host correctly.
	* 0.9:
	* Renamed Request.iter_headers() to .header_items().
	* unverifiable and origin_req_host are now attributes initialised
	  by Request.__init__().  All arguments named unverifiable are now
	  gone.

2004-01-03 John J Lee <jjl@pobox.com>
	* For Location:, URI: and Refresh:, use _Util.getheaders to get
	  first header instead of just using dictionary-style lookup.
	  Some servers return multiple Location headers, and dict-lookup
	  gives a comma-separated string in that case, which is what's
	  needed in these cases.
	* Fixed spurious test_file failure in test_urllib2.py.

2004-01-02 John J Lee <jjl@pobox.com>
	* Fixed response_seek_wrapper (and started using it in 0.9.x, as
	  well as 0.4.x).  Thanks to Erno Kuusela for bug report.
	* Carry over .origin_req_host and .unverifiable attributes when
	  upgrading urllib2.Request to ClientCookie.Request.  Previously,
	  they were effectively clobbered.  Ugly, but having to set those
	  attributes in the first place is ugly...
	* Released 0.4.17 and 0.9.2a.

2004-01-01 John J Lee <jjl@pobox.com>
	* Fixed redirection yet again.  Thanks to Armand Lynch for bug
	  report.  This time, the problem was that the state introduced by
	  cookies makes it possible for redirection to occur to the same
	  URL multiple times without an infinite loop occurring
	  (regardless of redirection code).  Should have thought of that
	  before...
	* Fixed proxy handling.  _urllib2_support.methnames was still
	  broken.  Thanks to Robert Kiendl for bug report.
	* Released 0.4.16 and 0.9.1a.

2003-12-30 John J Lee <jjl@pobox.com>
	* Fixed request_path bug: missing brackets around format string
	  tuple (only triggered when ;foo=bar parameters are present in
	  URL.  Thanks to Ollie Rutherfurd for bug report.
	* Released 0.4.15.

2003-12-29 John J Lee <jjl@pobox.com>
	* Fixed MSIECookieJar initialiser argument passing to base class.
	* Fixed MSIECookieJar.load_cookie_data(), which was calling
	  self.set_cookie() instead of using the base class' method,
	  causing infinite recursion when called from .set_cookie().
	* 0.9:
	* Did some refactoring of MSIECookieJar stuff, and filled in more
	  of MSIEDBCookieJar to convince myself that the base classes are
	  good enough.
	* Finished BSDDBCookieJar (NOT FULLY TESTED!).
	* Released 0.4.14 and 0.9.0a.

2003-12-28 John J Lee <jjl@pobox.com>
	* Fixed a bug where https requests were missing important headers
	  (like Host).  The problem was that I'd assigned the do_request
	  method to the local scope instead of the HTTPSHandler class
	  scope!  Thanks to Robbie Vanbrabant for the bug report.
	* Another hack for AbstractHTTPHandler.do_open() (see 2003-12-21),
	  and renamed .do_request() to .do_request_(), for the same
	  reason.
	* Released 0.4.13.

2003-12-24 John J Lee <jjl@pobox.com>
	* Modified setup.py so can easily register with PyPI.
	* 0.9:
	* IOError was inappropriately raised (and documented to be
	  raised!) by .load() and .revert().  Now they raise
	  ClientCookie.LoadError when the file format is not understood
	  (and IOError in the usual circumstances).
	* Moved FileCookieJar.as_lwp_str() into derived class
	  LWPCookieJar, and updated tests as necessary.

2003-12-21 John J Lee <jjl@pobox.com>
	* Corrected Cookie.__repr__ to use repr(arg), not str(arg) and to
	  do rest arg properly.
	* Added unpleasant hack to deal with the fact that
	  HTTPRedirectHandler appears to be a processor thanks to the
	  redirect_request method (which is documented in std. library
	  docs!).  Thankfully, Python CVS has a single base class for
	  processors and handlers, so this problem isn't fatal, but still
	  present (a URL like redirect://a.com/ would get redirect_request
	  called to process the request, resulting in great confusion
	  :-(( ).
	* Released 0.4.12.

2003-12-20 John J Lee <jjl@pobox.com>
	* Replaced all usage of seek_wrapper with new class
	  response_seek_wrapper, which knows about the response class
	  returned by urllib2.urlopen(), and uses that knowledge to avoid
	  clobbering all the response methods on .close().  This is a bit
	  of a hack to make it possible to cache response objects to make
	  a Browser class.
	* 0.9:
	* RFC 2965 is now off by default.

2003-12-18 John J Lee <jjl@pobox.com>
	* Changed order of to_read comparison in _Util.seek_wrapper.seek.
	  None was being compared with an integer (which seems, by chance,
	  not to cause a bug in 2.3, but is certainly wrong).  Thanks to
	  Jim Jewett.

2003-12-16 John J Lee <jjl@pobox.com>
	* Fixed bug in CookieJar.clear(): didn't work with three
	  arguments.
	* Fixed masses of test failures that had built up because I forgot
	  to run tests, and because of fix to redirection bug.
	* Released 0.4.11.

2003-12-14 John J Lee <jjl@pobox.com>
	* In stable and development versions (0.4.x and 0.9), fixed adding
	  of Cookie header so it's not left in incorrectly during
	  redirects.  Also, don't clobber Cookie or Cookie2 headers if
	  they're already present in the request.  Also also, added
	  .get_header() and to insulate users of Request against key
	  duplication between regular and unredirected headers.
	* Request is now documented to require .add_unredirected_header(),
	  .has_header(), .get_header() and .iter_headers() methods (but
	  these are added if missing by RequestUpgradeProcessor (for
	  compatibility with existing urllib2), so it doesn't actually
	  matter).  This will break code for people not using urllib2, but
	  I had to change it because it's needed for the security fix
	  above.
	* In 0.9:
	* Made CookieJar.cookies attribute private.
	* Moved file-based stuff into FileCookieJar abstract subclass of
	  CookieJar, and libwww-perl format stuff into LWPCookieJar.
	  Removed filename and delayload attributes and constructor args
	  from CookieJar, along with .save(), .load() and .revert().
	* Changed skip_discard arg to LWPCookieJar.as_lwp_str() to
	  ignore_discard (inverted meaning), and skip_expired to
	  ignore_expires.

2003-12-12 John J Lee <jjl@pobox.com>
	* Cleaned up tests a little bit.
	* Started 0.9:
	* Documented HTTPCookieProcessor.cookiejar and
	  HTTPRefreshProcessor.max_time / .honor_time as public
	  attributes.

2003-12-06 John J Lee <jjl@pobox.com>
	* Fixed unicode URL bug, I hope (replaced normalize_path with a
	  new escape_path function).  Thanks to Ronald Tschalar for
	  advice.
	* Started using warnings and logging.  As a result removed
	  CLIENTCOOKIE_DEBUG switch, added constants getLogger, NOTSET,
	  INFO, DEBUG, used exactly as per logging module in Python 2.3
	  (or you can just use logging module direct with 2.3).  Also
	  added USE_BARE_EXCEPT for old Pythons that don't have warnings
	  module, so can't turn warnings into errors.
	* More documentation updates and fixes.
	* Released 0.4.10.

2003-12-04 John J Lee <jjl@pobox.com>
	* Fixed a few doc bugs.

2003-12-01 John J Lee <jjl@pobox.com>
	* Fixed bug with ClientCookie.OpenerDirector.add_handler():
	  handler methods in base classes are now seen even with old
	  Python versions.
	* HTTPEquivProcessor now uses HTMLParser module when available
	  (should work with XHTML now).

2003-11-29 John J Lee <jjl@pobox.com>
	* Removed HTTPStandardHeadersProcessor: it's now part of
	  AbstractHTTPHandler.
	* Added test for FileHandler.

2003-11-28 John J Lee <jjl@pobox.com>
	* A bit more work on _urllib_support tests.
	* In _urllib2_support, cleaned up OpenerDirector.add_handler(),
	  removed Request.get() (dunno why it was there in the first
	  place...), renamed Request.has_key() to .has_header().
	* Got rid of ClientCookie.HTTP_DEBUG and .REDIRECT_DEBUG global
	  debug switches.  They're replaced by a set_http_debuglevel()
	  method for HTTP{S,}Handler, and an HTTPRedirectDebugProcessor.
	* Fixed 1.5.2 compatibility bug which broke ordering of
	  processors.

2003-11-14 John J Lee <jjl@pobox.com>
	* Fixed bug where no log message was generated for pre-expired
	  cookies where no cookie was previously present in the CookieJar.

2003-11-07 John J Lee <jjl@pobox.com>
	* seek_wrapper.close cleans up reference to read data.

2003-11-03 John J Lee <jjl@pobox.com>
	* HTTPEquivProcessor's max_time constructor arg can now be None,
	  meaning no limit.
	* Added a functional test!  Tests cookies, refresh and http-equiv.
	  It does depend on an internet web page, though.  I should put up
	  my own page with some tests, on the SF site...

2003-11-03 John J Lee <jjl@pobox.com>
	* Bugfix: tests from test_urllib2.py were not run by test.py
	  script.
	* More _urllib2_support tests.
	* Bugfix: Host header is no longer clobbered by
	  HTTPStandardHeadersProcessor.
	* Fixed broken processor/handler merge which broke the whole
	  package.  Need some new functional tests!
	* Released 0.4.9.

2003-11-02 John J Lee <jjl@pobox.com>
	* Wrote a few unit tests for _urllib2_support.
	* Changed a couple of uses of urllib2.Request to
	  _urllib2_support.Request.
	* Fixed thoroughly broken HTTPRedirectHandler loop detection,
	  thanks to new tests!
	* Added (untested) thread synchronization.
	* Released 0.4.8.

2003-10-31 John J Lee <jjl@pobox.com>
	* Fixed missing import in _urllib2_support.py (for
	  HTTPRefreshProcessor) and 1.5.2-compatibility.  Apparently
	  nobody's using either...

2003-10-30 John J Lee <jjl@pobox.com>
	* Fixed implementation of redirection verifiability:
	  Request.original_url: was actually the URL of the previous
	  request iin the chain of redirections, not of the origin
	  request.

2003-10-29 John J Lee <jjl@pobox.com>
	* The processor and handler interfaces may now be implemented by a
	  single object.  This makes for more natural handling of things
	  like response caching.

2003-10-28 John J Lee <jjl@pobox.com>
	* Rehashed string representations so they behave in a more
	  standard way.  Cookie._debug_string is gone, to be replaced by a
	  new __repr__.  The old Cookie.__str__ is gone, to be
	  reincarnated as the lwp_cookie_str function.  Cookie.__repr__ is
	  renamed to __repr__.  CookieJar.as_string is renamed as_lwp_str.
	  CookieJar.__str__ and .__repr__ now simply return a string like
	  <CookieJar[...]>, where the ellipsis is replaced by a comma-
	  separated list of str(cookie) or repr(cookie) respectively.
	* domain_initial_dot is now False where domain_specified is False,
	  so that __repr__ is unchanged after save/load round-trip.

2003-10-23 John J Lee <jjl@pobox.com>
	* In DefaultCookiePolicy.set_ok_domain, added "int" to list of
	  top-level domains for strict_domain, and made string comparison
	  case-insensitive.
	* Cookies with missing or invalid expires cookie-attribute are now
	  treated as session cookies again, on reading Ronald's advice
	  about this.  Added test so I don't break it again.
	* Cleaned up CookieTests.test_secure.
	* Added xreadlines and iterator support to seek_wrapper.
	* Cleaned up _HeadersUtil.
	* Resolved long-standing question about /, ?, # and ; and cookie
	  paths (I was worried about this after reading Mozilla source) by
	  reading Mozilla bug 156725.  Nothing needed changing.

2003-10-23 John J Lee <jjl@pobox.com>
	* Fixed some confusing MSIECookieJar doc bugs.
	* Fixed delayloading bug that might cause set cookies to get
	  clobbered by those on disk.
	* Released 0.4.7a.

2003-10-22 John J Lee <jjl@pobox.com>
	* MSIECookieJar bug fixes: do case-insensitive search to find
	  cookie data (username was appearing in lower case), be more
	  liberal about index.dat 'magic'.  Also fixed an exception
	  message that gave a UnicodeDecodeError.
	* Released 0.4.6a.

2003-10-21 John J Lee <jjl@pobox.com>
	* Fixed Netscape parsing bug where cookie-attributes that should
	  have a value but don't would end up with a None key.  Also,
	  parse_ns_headers now works properly for empty headers, and no
	  longer drops expires cookie-attribute on the floor when the
	  value is invalid: instead, None is returned as the expires
	  value.
	* _normalized_cookie_tuples is now more robust to bad Set-Cookie
	  and Set-Cookie2 headers.
	* Cleaned up delayloading.  All methods should now work, but
	  anything that causes iteration over all cookies will not cause
	  loading of delayloaded cookies from disk.  If you really want to
	  iterate over *all* cookies (rather than just those currently in
	  memory), call read_all_cookies.

2003-10-20 John J Lee <jjl@pobox.com>
	* Split up set_ok and return_ok methods from DefaultCookiePolicy
	  to make it easier to inherit its implementation.

2003-10-10 John J Lee <jjl@pobox.com>
	* Corrected domain-blocking documentation ("acme.com" doesn't
	  match ".acme.com").  Also, reversed the ordering (x matches y ->
	  y matches x), to agree with that actually used in the code
	  (since the domain-matching algorithms used are not commutative,
	  this was confusing).
	* Implemented strict_domain argument to DefaultCookiePolicy
	  constructor.

2003-10-03 John J Lee <jjl@pobox.com>
	* Fixed loading of cookies with no name (previously, the name and
	  value were reversed).

2003-09-21 John J Lee <jjl@pobox.com>
	* Missing MSIE cookie files are now skipped, rather than raising
	  an exception (unless load_cookie_data is used directly, which
	  still raises IOError if the file is not found).  Thanks to Todd
	  Whiteman.
	* Released 0.4.5a.

2003-08-16 John J Lee <jjl@pobox.com>
	* Fixed MSIECookieJar._cookies for_domain (dict lookup instead of
	  dict.get).  Also added an optional username argument to load and
	  load_from_registry.  Thanks to Mark Carter for bug reports.
	* Released 0.4.4a.

2003-08-13 John J Lee <jjl@pobox.com>
	* Released 0.4.3a.

2003-08-10 John J Lee <jjl@pobox.com>
	* Drastically reduced the requirements on response objects.

2003-07-28 John J Lee <jjl@pobox.com>
	* Added a couple of missing return False's to domain_return_ok and
	  return_ok.  Also, verifiability logic was too strict, due to
	  return False in wrong place.  Thanks to Armand Lynch for the bug
	  report.
	* Documented cookies attribute of CookieJar as being not publicly
	  writable.  This was just an oversight.  Your code won't
	  currently break if you do this, but it's certainly a bad idea.

2003-06-23 John J Lee <jjl@pobox.com>
	* Seekable response wrapper stuff changed so it doesn't trip over
	  the typecheck in pre-2.3 urllib2s (actually I did this ages ago,
	  and forgot to record it).
	* Fixed test for old-style issubclass.

2003-06-23 John J Lee <jjl@pobox.com>
	* Much of AbstractHTTPHandler has been rewritten in terms of
	  processors (a bit like handlers).
	* MozillaCookies.load now gives a more helpful exception message
	  (thanks to 'Steve').
	* Another fix to MozillaCookies: first field can be empty, so
	  don't lstrip (thanks to 'Steve').
	* Response objects are no longer seekable by default.  Pass
	  SeekableProcessor to build_opener if you want that.
	* HTTPRefreshProcessor now has some constructor arguments to
	  allow non-zero time Refreshes, and to control whether or not
	  the requested pause is actually honoured, or just skipped.

2003-06-17 John J Lee <jjl@pobox.com>
	* Applied some bugfixes for AbstractHTTPHandler from Python CVS.
	  This fixes proxy support.
	* Renamed Cookies class to CookieJar (and analogously for
	  MozillaCookies and MSIECookies).  This was inevitable given the
	  existence of the Cookie class -- too confusing otherwise.

2003-06-14 John J Lee <jjl@pobox.com>
	* Fixed bug in redirect code that would have allowed infinite
	  loop.
	* Added support for automatically adding Referer HTTP header.  I
	  was really getting tired of adding that all the time...
	* Cleaned up redirect code.
	* Added some more debugging switches.
	* Changed license to BSD, so I can use a Python Cookbook recipe.
	  The only difference is the addition of a non-endorsement clause.
	* Changed all raise statements to raise instances, not classes.

2003-06-13 John J Lee <jjl@pobox.com>
	* 0.4.2a changes (these were the only changes from 0.4.1a to
	  0.4.2a; VERSION attribute was incorrectly set to "0.4.1a" in the
	  0.4.2a release).
	* Fixed CookiePolicy.path_return_ok: raise True --> return True!
	* Fixed HTTP-EQUIV support: moved CHUNK to _urllib2_support.py.
	* Added missing imports to _urllib2_support.py.
	* Pychecker found a bug in _normalized_cookie_info (comparison
	  instead of assignment).
	* Pychecker found a bug in _cookie_from_cookie_tuple (non-existent
	  local in debug statement).
	* Pychecker found a bug in _MSIECookies and _MozillaCookies:
	  missing import of MISSING_FILENAME_TEXT.
	* Pychecker found a bug in _MSIECookies._really_load: non-existent
	  local in assert statement.
	* Pychecker found a bug in _urllib2_support.HTTPRedirectHandler:
	  non-existent local in raise statement.  Aaaaghhh...
	* Redirect now allows return to a url previously seen, because
	  that doesn't necessarily imply a loop if HTTP-EQUIV and 30x
	  occur in a single redirect sequence.
	* Released 0.4.2a.

2003-06-08 John J Lee <jjl@pobox.com>
	* 0.4.1a changes:
	* Simplified pointlessly complicated imports (having read Guido's
	  import docs, finally).
	* Simplified add_cookie_header, while preserving efficiency for
	  MSIECookies delayload mode (to stop loading lots of cookie files
	  from disk unless necessary).
	* Uploaded porting docs.
	* Uploaded example code (also uses ClientForm).
	* Fixed a few remaining local references that lynx was writing in
	  README.txt.
	* Cleared up weird file permissions.
	* A few more doc fixes to reflect changes.
	* Released 0.4.1a.

2003-06-08 John J Lee <jjl@pobox.com>
	* 0.4.1a changes:
	* Netscape parser now works for Mixed-case Netscape cookie-
	  attributes.
	* Netscape parser now accepts "port" (no equals sign), and
	  cookies with no name (again, no equals sign).
	* Suddenly realised that the port attribute of urllib2.Request
	  objects is always None -- it's unused.  Added request_port
	  function.
	* Fixed path ordering of returned cookies (they were only ordered
	  within each domain, not overall).  Happily, this change made the
	  code much clearer, too.
	* Improved parse_ns_headers whitespace handling.
	* parse_ns_headers: expires cookie-attribute can now be quoted.

2003-06-06 John J Lee <jjl@pobox.com>
	* 0.4.1a changes:
	* Split str2time into httpstr2time and isostr2time.  str2time
	  is now an alias for httpstr2time.
	* Fixed MSIECookies 1.5.2-compatibility bug.  Hmm, must install
	  1.5.2 on Windows...
	* Cleared up confusion between third-party and unverifiable
	  transactions in docs.

2003-06-03 John J Lee <jjl@pobox.com>
	* Fixed trivially broken MSIECookies cookie loading (was
	  attempting to call long() on value and domain/path).
	* 0.4.1a changes:
	* Fixed handling of Max-Age=0 (actually, I think it was correct
	  before, but fragile).  Max-Age=0 means discard (old and new)
	  cookie.
	* MozillaCookies.save now saves expires as "" if cookie.expires is
	  None.
	* Complete rehash of domain-matching code, to follow Mozilla and
	  IE for V0 cookies.
	* MozillaCookies.save and .load now agree with each other, with
	  the Cookies class and with Mozilla about domain handling.
	* Changed eff_request_host so .local is only added for HDNs (ie.
	  not for IP addresses) -- since this is presumably what will be
	  in RFC 2965 errata.
	* Removed requirement that Netscape Domain cookie-attribute must
	  not be IP address.
	* Renamed test_cookies.LWPTests.test_9 (there were two of them) so
	  the second doesn't clobber the first!
	* Returned Domain attribute must now exactly match that originally
	  set, including presence or otherwise of initial dot, according
	  to draft RFC 2965 errata.
	* Added some Netscape strictness switches.
	* RFC 2109 cookies are now treated as RFC 2965 cookies (but still
	  parsed by Netscape parser).

2003-06-02 John J Lee <jjl@pobox.com>
	* 0.4.1a changes:
	* Renamed Cookies.clear_temporary_cookies to
	  clear_session_cookies.
	* Fixed clear_session_cookies -- was incorrectly clearing expired
	  cookies, too.  Added is_expired method to Cookie class.
	* Fixed MSIECookies delayloading (still needs testing).
	* Cookies.load, .restore and MSIECookies.load_from_registry
	  methods now have ignore_discard and ignore_expires arguments.
	* Added Cookies.clear_expired_cookies method.

2003-05-27 John J Lee <jjl@pobox.com>
	* Uploaded 0.4.0a, in case the person who grumbled about lack
	  of thread-safety wants to write a patch.
	* MozillaCookies is currently broken (no doubt other stuff too)
	  in 0.4.0a -- and slightly broken in 0.3.4b, for that matter.
	* Uploaded 0.3.5b, with fixed NetscapeCookies (the class that's
	  renamed MozillaCookies in 0.4.0a), thanks to a demi-nonymous
	  'Steve'.

2003-05-22 John J Lee <jjl@pobox.com>
	* 0.4.0a changes:
	* Removed autosave constructor arg from Cookies class, and
	  removed __del__ methods.  Also removed ignore_discard and
	  ignore_expires args from constructor and moved to save
	  method.
	* MozillaCookies and Cookies are now consistent on expires /
	  discard behaviour on saving -- they both follow the
	  ignore_expires and ignore_discard args.
	* Cookies objects no longer automatically try to load cookies
	  when you supply a filename to the cookies constructor, and
	  the save and load methods no longer alter the value of the
	  filename attribute.
	* delayload is now a public attribute.
	* All files are now explicitly close()d in finally: blocks.
	* Another minor redirect change to synchronise with current SF
	  patch (still not applied!).  The actual behaviour is
	  unchanged.

2003-05-18 John J Lee <jjl@pobox.com>
	* 0.4.0a changes:
	* Fixed Cookie2 header -- before, it wouldn't have been sent
	  unless the first matching cookie was V0, and V2 (heaven forfend)
	  would not have caused it to be sent.  Really, though, the whole
	  issue of multiple versions is still to be clarified in RFC 2965
	  errata.
	* Add set_cookie_if_ok method, which works like set_cookie, but
	  only if policy.set_ok method says it's OK.
	* Moved check for cookies starting with $ to set_ok, limited it to
	  V0 cookies, and removed check for empty cookies.
	* Added accept list -- this is a 'whitelist'.  Blocked domains are
	  considered first, then accepted domains.
	* Changed domain match algorithm for blocking (and accepting)
	  cookies.  I've copied Ronald Tschalar's algorithm, which is
	  better than mine, because it allows more control.  Now a domain
	  in a block / accept list like "acme.com" with no initial dot
	  only matches "acme.com" (not as before), but ".acme.com" (with
	  initial dot) matches "good.acme.com" and "bad.acme.com" (as
	  before).  Empty domains no longer have any special meaning.
	* Another change to file format -- added 'domain_dot' which
	  records whether or

2003-05-16 John J Lee <jjl@pobox.com>
	* 0.4.0a changes:
	* CommentURL and Comment are now exposed as the comment_url and
	  comment attributes of the Cookie class.  They are no longer part
	  of the Cookie.rest dictionary.  This is also reflected in the
	  string representation of the Cookies class and of the Cookie
	  class.
	* Removed scan method, in favour of iterator (& 1.5.2 __getitem__
	  hackery).

2003-05-13 John J Lee <jjl@pobox.com>
	* 0.4.0a changes:
	* Renamed Cookies.clear method key argument to name.
	* Renamed 'redirect' arguments of Cookies.add_cookie_headers and
	  .extract_cookies methods to 'unverifiable'.
	* Even if there are some Netscape cookies in a response, still
	  check path according to RFC 2965 rules for version 1 cookies.
	* Created new Cookie class.
	* set_cookie method now takes a Cookie as its argument.  The old
	  method signature was actually broken (there was no way to
	  specify whether the domain and port were specified in the HTTP
	  headers), so this is unavoidable.
	* Cookies.cookies three-level dict now contains Cookie instances
	  as its leaf values, rather than sequences.
	* Major code reorganisation: moved all accept / reject logic into
	  nice shiny new CookiePolicy classes.  The Cookies class is now
	  simpler.  :-)
	* Removed blocked_domains arg from Cookies class.  This is now
	  handled by a CookiePolicy object.
	* Removed netscape_only and hide_cookie2 args from Cookies
	  constructor.  This is now handled by a CookiePolicy object.
	* RFC 2965 port bug fixed: before, a port cookie-attribute with no
	  value was incorrectly returned with a value.
	* Changed Cookies class' save / load format to remember whether
	  or not port was specified.  Loading old-format files containing
	  RFC 2965 cookies (if any such even exist, which is highly
	  unlikely) may now cause ClientCookie to get returned port values
	  wrong in a slighly different way to how it got them wrong
	  before. ;-)  In any case, you almost certainly only have
	  Netscape cookies, in which case the two file formats are
	  degenerate.

2003-05-09 John J Lee <jjl@pobox.com>
	* Simplified and refactored _HTTPDate and tests.

2003-05-06 John J Lee <jjl@pobox.com>
	* Fixed parsing (by NetscapeCookies) of cookies.txt with empty
	  cookie value.

2003-04-30 John J Lee <jjl@pobox.com>
	* Fixed MSIECookies -- apparently NT and XP, and IE 5 and 6 *do*
	  store cookies in the same format.  When hit 0x0df0ad0b padding,
	  just ignore it and read again instead of deciding that the
	  sector size is wrong.
	* Fixed MSIECookies.__setattr__, which was too strict about
	  autosave.
	* Added base URL to README.html, so README.txt comes out with http
	  URL references rather than pointers to files local to my
	  machine.
	* Released 0.3.4b.

2003-04-29 John J Lee <jjl@pobox.com>
	* Small change to HTTPRedirectHandler: 301 from a POST is now
	  allowed, and redirected as a GET.
	* More minor documentation fixes.
	* Hmm, cleaning up the tests seems to have revealed a whole bunch
	  of 1.5.2 compatibility bugs.  Not sure what was wrong with old
	  test running system, but don't care.  Fixed bugs.
	* Added INSTALL file to MANIFEST.in -- hadn't noticed it was
	  missing from the distribution.
	* Released 0.3.3b.

2003-04-28 John J Lee <jjl@pobox.com>
	* Cleaned up MSIE code, and got it working on NT.  Hopefully it
	  still works on XP...
	* Cleaned up cookies tests, and ported all tests to use unittest.
	* Minor documentation fixes.

2003-03-27 John J Lee <jjl@pobox.com>
	* Applied Eric Woudenberg's fixes to previously untested
	  MSIECookies code (Eric is using Windows XP).

2003-03-23 John J Lee <jjl@pobox.com>
	* Added documentation explaining the reasons for the existence of
	  the various cookie protocols.
	* Released 0.3.2b.

2003-03-10 John J Lee <jjl@pobox.com>
	* Fixed documentation bug (changing User-Agent header involves
	  setting addheaders, not add_headers, attribute on
	  urllib2.OpenerDirector).  Thanks to Christoph Schwarz.

2003-03-05 John J Lee <jjl@pobox.com>
	* Fixed 1.5.2 compatibility bug.

2003-02-17 John J Lee <jjl@pobox.com>
	* Released 0.3.1b.

2003-02-16 John J Lee <jjl@pobox.com>
	* Removed Netscape three-dots-for-non-special-toplevel-domains
	  check, again on the advice of Ronald Tschalar, who informs me
	  this is another place where browsers don't follow Netscape spec.
	* Fixed tests so they work with new request_host function and
	  insistence on version attribute in Set-Cookie2 header.
	* Changes to domain-matching code.  Host prefixes are allowed to
	  contain a dot for Netscape cookies.  Code dealing with
	  dot-normalisation and check for embedded dots now explicitly
	  follows RFC 2965 algorithm -- I can't be bothered to figure out
	  whether the end result is equivalent in general to the way way
	  it was before.  I feel no guilt about this :)
	* add_cookie_header now checks for request absoluteURI with
	  missing abs_path (as defined by RFC 2396) and corrects it if
	  necessary.
	* IP addresses in the domain attribute of Set-Cookie are now only
	  illegal for Netscape cookies.
	* Cookies.blocked_domains now returns an empty tuple rather than
	  None, as the docstring says it should.

2003-02-12 John J Lee <jjl@pobox.com>
	* Altered request_host, to match my interpretation of RFCs 2965
	  and 2616 (Host: header no longer checked).

2003-02-11 John J Lee <jjl@pobox.com>
	* Removed the Netscape bad-path hack.  Not convinced it's
	  generally useful, or even non-harmful.
	* Cookies.extract_cookies now disallows Set-Cookie2 headers with
	  no Version attribute.

2003-02-10 John J Lee <jjl@pobox.com>
	* Made MSIECookies importable from ClientCookie package.
	* Fixed Cookies.__len__!

2003-02-08 John J Lee <jjl@pobox.com>
	* Removed 'del True, False' from __init__.py -- 2.3 doesn't like
	  it, understandably!  Thanks to Martijn Faasen.
	* Cookie headers are now added for two-component domain names like
	  lwn.net (moved checking for number of dots to extract_cookies,
	  and made the check specific to Netscape cookies).  Thanks to
	  Martijn Faasen.
	* Fixed Cookies.revert method, which didn't always leave object
	  state unchanged on failure.
	* Fixed netscape_only mode of Cookies class in the case where
	  RFC 2965 Set-Cookie2 headers are present (Set-Cookie2 now never
	  clobbers Set-Cookie cookies in netscape_only mode).
	* Assorted minor refactoring.

2003-02-05 John J Lee <jjl@pobox.com>
	* Made blocked_domains attribute private again, added accessor
	  methods (to avoid mutability issues).
	* Released 0.3.0b.

2003-02-03 John J Lee <jjl@pobox.com>
	* Fixed handling of quoted Netscape cookies values.  Thanks to Rob
	  Dunstan for the bug report.

2003-02-02 John J Lee <jjl@pobox.com>
	* Added install_opener to public interface exported by __init__ --
	  it should have been there in 0.2.0, of course.
	* Fixed several bugs in lazily-seekable response code, and added
	  tests.

2003-01-03 John J Lee <jjl@pobox.com>
	* Fixed bug in interaction of Netscape and RFC 2965 cookies.
	  Previously, RFC 2965 cookies overwrote Netscape cookies if both
	  cookie name and value matched.  Now, domain, path and name must
	  match (this was broken in original Perl). [2003-03-27 But note
	  that this is one of the topics currently under discussion on
	  http-state mailing list for the RFC 2965 errata]
	* Fixed bug in overwriting of cookies with new cookies: domain
	  matching is now case-insensitive (this was broken in original
	  Perl).
	* Removed incorrect information about urllib2 (OpenerDirector does
	  not have an attribute named server_version).
	* Added optional HTTP-EQUIV handling (in which META HTTP-EQUIV
	  headers in HTML are transparently read and added to your regular
	  HTTP headers).
	* Added optional HTTP Refresh header handling.  If a Refresh
	  header is present and has zero refresh time, it is translated
	  into a 302 redirect.  Refresh headers with non-zero refresh
	  times are ignored.
	* Made seekable responses optional (but on by default).  If you're
	  using HTTP-EQUIV handling, seekable responses are required.  At
	  the moment, the parser for HTTP-EQUIV will slurp in the whole
	  file unless the </HEAD> closing tag is explicitly present.
	* HTTPHandler and HTTPSHandler constructors now have options to
	  control the above: seekable_responses, handle_http_equiv and
	  handle_refresh.

2003-01-02 John J Lee <jjl@pobox.com>
	* Refactored Cookies.add_cookie_header and .extract_cookies
	  methods.  Now I have some hope of checking the code is
	  correct...
	* Updated MSIE support with bug fixes from Johnny Lee, the author
	  of the MSIE Perl code.
	* Made blocked_domains a public attribute.
	* Fixed quoting of backslashes in cookie values (previously, they
	  were incorrectly left unquoted).
	* Improved documentation.

2002-12-19 John J Lee <jjl@pobox.com>
	* Added first attempt at Microsoft Internet Explorer (MSIE) cookie
	  support.  Not working yet.

2002-12-18 John J Lee <jjl@pobox.com>
	* Response objects now have lazy read, readline, readlines and
	  seek methods.
	* Removed Apple Mac-specific code, since it has never been tested.
	  If anybody wants it back, let me know.
	* Fixed 1.5.2-compatibility bug (calendar.timegm not available).

2002-12-13 John J Lee <jjl@pobox.com>
	* Improved README.html.
	* Removed Python 1.5.2-compatible urllib2/urllib.  Now separately
	  downloadable, because useful for all three web client modules
	  (ClientCookie, ClientForm, ClientTable).

2002-11-19 John J Lee <jjl@pobox.com>
	* Released 0.2.0b.

2002-11-17 John J Lee <jjl@pobox.com>
	* Removed README, created README.html and INSTALL.  README mostly
	  just restated what was in the web page, so README.html is now
	  just a copy of the web page.

2002-11-16 John J Lee <jjl@pobox.com>
	* Removed _TimeLocal.py.  :-)
	* Changed license to MIT (from Perl Artistic).  Thanks, Gisle.

2002-10-13 John J Lee <jjl@pobox.com>
	* Removed cookies argument from urlopen, and added install_opener
	  function.  This is the way urllib2 does things, and is also
	  simpler.  This won't change again, honest!
	* Added documentation to explain how to change User-Agent header,
	  and how to provide a cookies instance in the new way.
	* Improved documentation layout, and corrected a few minor
	  mistakes in docs.

2002-10-11 John J Lee <jjl@pobox.com>
	* Fixed silly Python 2.3 forwards-compatibility bug (True / False
	  constants were defined, overwriting the new builtin versions in
	  2.3).

2002-10-05 John J Lee <jjl@pobox.com>
	* Fixed handling of "secure" attribute for Netscape protocol,
	  which was broken in original Perl code (thanks to Rob Hall and
	  Levente Sandor for bug report and fix respectively).
	* Fixed debug switches, which were in the wrong place.
	* Fixed revert method of Cookies class, which didn't raise IOError
	  when loading failed.
	* Fixed trivially-broken 1.5.2-compatibility.
	* Fixed quick exit for no-cookies case.
	* Removed spurious print statement from _HTTPDate.py.
	* Removed non-functional mxDateTime `support', until I manage to
	  get mxDateTime installed, at least.
	* Improved documentation a little (including removing some
	  disinformation about the differences in cookie expiration
	  between Cookies and NetscapeCookies).
	* 0.1.1b release.

2002-07-12 John J Lee <jjl@pobox.com>
	* Removed stupid cookie_kwargs arguments to urlopen, HTTPHandler
	  and HTTPSHandler, and replaced with a cookies argument that
	  lets you simply provide your own Cookies instance.  This allows
	  you to get at the cookies, for example.
	* Fixed redirect fix (thanks to Jeremy Hylton for pointing this
	  out).  See Python bug #549151 on www.sourceforge.net.  30x
	  redirects should now be properly handled for all x (which may
	  mean leaving them for the user of ClientCookie to deal with, by
	  raising an exception).  This is a fix to the Python standard
	  library, and is a separate issue from that of cookies being
	  turned off during redirections when using the RFC 2965 protocol.
	  This fix was also applied to the 1.5.2 backport.
	* Fixed negative expiration time case (thanks to Johann Hibschmann
	  and David Lashar for bug reports).
	* 0.1.0b release.

2002-05-31 John J Lee <jjl@pobox.com>
	* Added import types in _ClientCookie.py.
	* Used time-machine to fix ClientCookie.VERSION (was erroneously
	  "0.0.6b" in this release; was also wrong in README).
	* 0.0.7b release.

2002-05-27 John J Lee <jjl@pobox.com>
	* RFC 2965 cookie handling may now be switched off entirely.  This
	  is useful because few browsers implement it, so it is likely
	  that many servers incorrectly implement it.  This is also
	  useful because ClientCookie does not yet fully implement
	  redirects with RFC 2965 cookies.
	* The redirect argument to Cookies.add_cookie_header and
	  Cookies.extract_cookies is going to stay: you are allowed to use
	  it now.
	* Arguments can now be passed to the Cookie constructor when using
	  urlopen, HTTPHandler and HTTPSHandler.
	* Added domain blocking.
	* Improved documentation.
	* Applied redirect+POST bugfix (bug #549151 on SF) to 1.5.2
	  back-port of urllib2.py (not necessary for correct functioning
	  of ClientCookie itself, since this is also fixed elsewhere, in
	  _ClientCookie; this will only affect anyone who is using this
	  urllib2 separately from ClientCookie).
	* Reversed ChangeLog time ordering!
	* Changed version numbering system to integer triple x.y.z from
	  fixed-point x.yz: this release is 0.0.6b, not 0.06b.
	* 0.0.6b release.

2002-05-06 John J Lee <jjl@pobox.com>
	* Back-ported to Python 1.5.2.
	* Fixed Windows issues with _HTTPDate and _TimeLocal; all tests now
	  pass on both Unix and Windows (thanks to Irmen de Jong for bug
	  reports).
	* Tested with Python versions 1.5.2, 2.0.0, 2.1.1 and 2.2.0.
	* Set line endings to Unix convention.
	* 0.05b release.

2002-05-01 John J Lee <jjl@pobox.com>
	* Fixed _HTTPDate to give up for timezones given as strings, unless
	  mxDateTime is installed (mxDateTime support is untested).
	* Removed use of strptime -- Windows doesn't have it.
	* Fixed use of localtime and gmtime in _HTTPDate and _TimeLocal for
	  2.0-compatibility.
	* Fixed _host again (properly this time, I hope).
	* Changed tests to reflect work-around for incorrect Netscape path
	  attributes.
	* 0.04b release.

2002-04-26 John J Lee <jjl@pobox.com>
	* Interface is stable.
	* Has now been used in anger, by me.
	* Fixed _ClientCookie._host function (now works with proxies).
	* Fixed HTTPHandler, etc (handling of redirections is not correct yet,
	  though).
	* Work-around for Netscape cookies with bad paths.
	* Automatic redirections (with urllib2) now work (for Netscape cookies
	  only).
	* Handling of RFC 2965 cookies is currently turned off during automatic
	  redirections, until I figure out exactly which cookies should be
	  blocked.
	* Included fixed version of HTTPRedirectHandler: does GET instead of
	  POST during redirects, as required by RFC 1945 (bug #549151 on SF).
	* str2time function is now importable directly from ClientCookie.
	* 0.03b release.

2002-04-23 John J. Lee <jjl@pobox.com>
	* Renamed top-level directory to ClientCookie, from LWPy, and added
	  names to __init__.py, so eg. from ClientCookie import Cookies works.
	  ClientCookie.py is now _ClientCookie.py.
	* Renamed all private modules to have initial underscore.
	* Fixed _HeadersUtil bugs.
	* Removed all `return self's from Cookie.
	* Return values and exceptions raised are more-or-less okay now, I
	  think.
	* urlopen, HTTPHandler, etc. now actually work.
	* Miscellaneous cleanup and bug fixes.
	* Tried actually using the thing on the internet a few more times.
	* 0.02a release.

2002-04-19 John J. Lee <jjl@pobox.com>
	* Managed to get a cookie out of http://uk.yahoo.com/, so something
	  must be working...
	* Initial 0.01a release.