*  Added testing for Python 3.5 and 3.6 (PR from Tim Graham) #110, #131

   *  Fixed typos in docstrings (PR from Romuald Brunet, reviewed by Tim
      Graham) #105

   *  Removing Python 2.6, 3.2, and 3.3 testing (PR from Tim Graham) #115, #116

   *  Removing unnecessary parens in return statements (PR from Tim Graham)
      #113

   *  Remove unused _has_unicode/_str_cls vars (PR from Tim Graham) #111

   *  Add flake8 testing and cleanups (PR from Tim Graham, cleanups from Sean
      Reifschneider) #112

   * Fixed storing non-ASCII values on Python 2 and binary values on Python 3
     (PR from Nicolas Noé) #135

   * Fixed touch(..., time=0) command (PR from Nicolas Noé) #137

Fri, 27 May 2016 13:44:55 -0600  Sean Reifschneider  <jafo@tummy.com>

   *  1.58 release.

   *  Fixing a performance regression in v1.54 in Python 2, using cPickle again.
      Patch by edmorley #86

   *  Support for "stats slabs".
      Patch by grg350 #93

   *  get_stats encoding fix.
      Patch by bartTC #91

   *  Pin Six version to >= 1.4
      Patch by pipermirriam #81

   *  setup.py build process pulls version from memcached.py
      Patch by dieselmachine #72

   *  delete() and delete_multi() now default the "time" argument to None,
      since the protocol doesn't allow a time in some implementations.
      Patch by oremj #27

Fri, 31 Jul 2015 11:38:25 -0600  Sean Reifschneider  <jafo@tummy.com>

   *  1.57 release.

   * Fix for #75 and #76, mc.set(key, False) results in "ValueError:
     invalid literal for int()".  Reported by Carlos Sanchez on github.

Sun, 26 Jul 2015 14:44:20 -0600  Sean Reifschneider  <jafo@tummy.com>

   *  1.56 release.

   *  More patches for python3 from Victor Stinner #67.  Thanks!

   *  Tests moved out into their own unittest testcases.
      Patch by Jeremy Thurgood.

   *  Adding support for pluggable compressors.  Patch by cactus on github.

   *  Adding support for "noreply".  Patch by cactus on github.

   *  Allowing empty key in _multi() calls.  Suggested by sergio97 on github.

Sun, 21 Sep 2014 13:41:30 -0600  Sean Reifschneider  <jafo@tummy.com>

   *  1.54 release.

   *  Numerous patches to attempt to get Python 3 working, but
      tests are still failing under Python 3, so this code is not ready
      yet.

   *  MemcachedKeyCharacterError better describes key problem.
      Suggested by Roy Smith

   *  Added touch(), patch by erankor on github.

   *  Allow empty server list, patch by zewt on github.

   *  If unknown flags in get request, raises a value error rather than the
      code bombing out.

   *  Setting the SERVER_MAX_* values after import did not work, despite
      being documented to.  Reported by alexf101 on github.

Sun, 07 Jun 2013 11:12:18 -0600  Sean Reifschneider  <jafo@tummy.com>

   *  1.53 release.

   *  Fixing set_multi() so that if the server closes the connection
      it will no longer raise AttributeError.  Issue found and resolution
      reviewed by Ben Hoyt.

   *  readline() now will mark the connection dead if the read fails.
      It was just closing it before.  This is related to the set_multi()
      change but worth noting separately.  Thanks to Ben Hoyt.

   *  Discussion of the above:
      https://github.com/linsomniac/python-memcached/commit/b7054a964aed0e6d86e853e60aab09cd0183b9f6#commitcomment-3337557

Sun, 02 Jun 2013 01:08:26 -0600  Sean Reifschneider  <jafo@tummy.com>

   *  1.52 release.

   *  Changing check_keys to use re.match() instead of str.translate(),
      because re.match() works with Python < 2.6.
      Found by Giovanni Di Milia.

Mon, 06 May 2013 15:29:07 -0600  Sean Reifschneider  <jafo@tummy.com>

   *  1.51 release.

   *  Add a MANIFEST.in file, patch by Daniel Widerin.

Mon, 06 May 2013 07:20:21 -0600  Sean Reifschneider  <jafo@tummy.com>

   *  1.50 release.

   *  Client() now takes a "check_keys" option, which defaults to True.
      If False, it disables the checking of keys to ensure they have
      acceptable size and are composed of non-control characters.
      Suggested by Ben Hoyt.

   *  Converting control character checking of keys based on performance
      testing of alternatives by Ben Hoyt.

   *  Converted unicode tests from using u'', patch from Eren Güve.

   *  Included license file (pull request by "Philippe" pombredanne).

   *  Doing a "set" after server goes away, raised AttributeError:
      'NoneType' object has no attribute 'sendall'.  Patch by Ken Lalonde

   *  incr/decr return None instead of 0 on server connection failure.
      Suggested by Ivan Virabyan

   *  Supports IPv6 connections using: "inet6:[fd00::32:19f7]:11000".
      Patch by Romain Courteaud

   *  Switching over to github for this project:

         https://github.com/linsomniac

   * Bug #974632: _ConnectionDeadError sometimes was escaping the get/set
   code.  This should return to readline() not raising an exception, except
   in the case that it's called from the get/set().  Report from Gary
   Poster, proposed patch by Brad Crittenden.

   Misc fixes by Brad Crittenden: fixing a docstring, if "port" is set to
   any false-like value it will default to 11211.

Mon, 29 Nov 2011 12:37:32 -0700  Sean Reifschneider  <jafo@tummy.com>

   * Bug #887765: Interrupted connection to memcache server can cause
   inconsistencies.
   Added "flush_on_reconnect" (defaults to off) to Client() which will
   cause a client that has lost connection to a server and then reconnects
   to flush the cache on the reconnect so that it doesn't get old values
   from that server.  Patch by Daniel Benamy.

Sun, 27 Nov 2011 18:15:32 -0700  Sean Reifschneider  <jafo@tummy.com>

   * Bug #745633: Values of maximum size are not stored
   API inconsistency, max value length was tested for <= while max KEY
   length was <.  So I picked that keys and values *LONGER* than the
   specified max value are what is used, and added documentation and tests
   to that effect.  The test for max value tested that length plus 4, so
   I've changed that to be that value plus 1.  Issue found by matt-quru.

   * Bug #713488: Issues Invalid "delete" command.
   Protocol has changed so that the "delete" operation no longer takes a
   "time" argument.  It seems that some servers will refuse a "delete key
   0" while others will accept it, but the official server will NOT accept
   "delete key 1".  So I've changed it so that if no "time" argument is
   specified, no time argument is sent to the server.

   * Bug #713451: server.expect("END") needs to be in a finally block
   Expect an "END" when the _recv_value() raises an exception.
   Patch by Jay Farrimond.

   * Bug: #741090: cas cache can grow unbounded.  Default now is that the
   cache is not used, unless the "Client()" object is created with
   "cache_cas=True".  In that case, you need to have your own cas clearing
   code, a simple one would be to use Client().reset_cas() to completely
   clear the cas_ids cache.  Problem pointed out by Shaun Cutts.

   * Bug #728359: Make python-memcache work on memcache restarts.
   Patch by Tarek Ziade', reviewed and further patches submitted by Hugo
   Beauze'e-Luysse and Neganov Alexandr.

   * Bug #798342: If memcached server sends unknown flag in response for
   "get", results in:
      "UnboundLocalError: local variable 'val' referenced before assignment"
   Now returns "None" instead.  Patch by Sharoon Thomas

Mon, 20 Dec 2010 19:14:17 -0700  Sean Reifschneider  <jafo@tummy.com>

   * Bug #680359: useOldServerHashFunction() is broken.  It now correctly
     switches back to the old memcache hash function.

Thu, 16 Dec 2010 02:07:40 -0700  Sean Reifschneider  <jafo@tummy.com>

   * Bug #471727: Changed the delete() code to explicitly check for both
     NOT_FOUND and DELETED as the responses and return successful for both.
     It also logs an error if one of these two responses is not found.
     Also added a test to ensure that delete() works.

   * When using set_multi and one value is too big, traceback
     TypeError: 'int' object is unsubscriptable
     Patch by Orjan Persson

   * Fixing Bug #529855: Server host can now be bare host without ":<port>".
     Fix proposed by Roger Binns.

   * Fixing Bug #491164: Typo fix, "compession" -> "compRession".

   * Fixing Bug #509712: "TypeError: 'NoneType' object is unsubscriptable"
     Also fixed some other similar code to not have issues with that.

   * Also related to 509712 and 628339: readline() now returns '' instead
     of None when a server dies.  This should be safer.  Patch suggested by
     Denis Otkidach.

   * Fixing Bug #628339: Read from server sometimes fails.  Patch by Jeremy
     Cowles.

   * Fixing Bug #633553: Add stat arguments support to get_stats().  Patch
     by Ryan Lane.

   * Changing the license to the PSF License.

   * Removing Evan's e-mail address at his request, changing authorship to
     Sean.

Sat, 28 Nov 2009 01:07:42 -0700  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.45

   * Per-connection max server key length.  Patch by Nicolas Delaby

   * Patches to make memcached more garbage-collectable.  Removes
     "debugfunc" argument from _Host objects and changed to "debug"
     boolean.  Patches by John McFarlane and Aryeh Katz.

   * Switching to a cmemcache compatible hash function.  Implemented by
     André Cru and Ludvig Ericson.  To switch back to the old style, use:

        memcached.useOldServerHashFunction()

   * Rejecting keys that have spaces in them.  Patch by Etienne Posthumus.

   * Fixing exception raising syntax.  Patch by Samuel Stauffer.

   * Optimizations in read code.  Patch by Samuel Stauffer.

   * Changing classes to be newstyle.  Patch by Samuel Stauffer.

   * Changed "has_key" to "in".  Patch by Samuel Stauffer.

   * incr/decr were raising ValueError if the key did not exist, the
     docstring said it returned none.  Patch by Chihiro Sakatoku.

   * Adding cas method, submitted by Ben Gutierrez.

   * Fix in the docstring for how to use the "set" method.  Found and fixed
     by William McVey

Thu, 02 Apr 2009 13:37:49 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.44

   * Allowing spaces in the key.  (Patch provided by xmm on Launchpad)

   * Detecting when the pickler needs a positional argument.  (Patch
     provided by Brad Clements on Launchpad)

   * Moving length check after the compression.  (Patch provided by user
     Tom on Launchpad)

   * Fixing arguments passed to the _Error if invalid read length.

   * Fixing the representation of domain sockets.  (Patch provided by user
     MTB on Launchpad)

   * Changing a typo of dead_until.  (Patch provided by Shane R. Spencer)

   * Providing better error messages (patch provided by Johan Euphrosine).

   * Adding get_slabs() function to get stats.  (Patch provided
     by Nick Verbeck)

Sun, 01 Jun 2008 15:05:11 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.43

   *  eliott reported a bug in the 1.42 related to the socket timeout code
      causing a traceback due to the timeout value not being set.

Sat, 31 May 2008 02:09:17 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.42

   *  Paul Hummer set up a Launchpad project which I'm going to start using
      to track patches and allow users to set up their own bzr branches and
      manage marging in the upstream patches with their own.

         https://launchpad.net/python-memcached

   *  Patch from Jehiah Czebotar which does: Changing the calls to
      mark_dead() to make them dereference tuples, reducing timeout on
      sockets to 3 seconds, settable via setting Host._SOCKET_TIMEOUT.

   *  Patches from Steve Schwarz for set_multi() to return the full set of
      keys if all servers are down.  Previously would not report any keys.

   *  Fix from Steve Schwarz delete_multi() argument "seconds" not being
      correctly handled.  Changed it to "time" to match all other calls.

   *  Patch from Peter Wilkinson to support using unix domain sockets.
      He reports that tests succeed with with memcached daemons running,
      the normal and a domain socket started via
      "memcached -s memcached.socket".  I massaged it quite a bit.

      To use domain sockets, use a connect string of "unix:/path/to/socket"
      Note however that if you are using a host name of "unix", it will now
      detect "unix:11211" as being a domain socket with the name "11211".
      In this case, please use "inet:unix:11211".

      Because of this, it is now preferred to use a connect string prefix
      of "inet:" or "unix:".

Tue, 29 Apr 2008 21:03:53 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.41

   * Patch from Jehiah Czebotar to catch an additional server disconnect
     situation.

   * Patch from Andrey Petrov to add the "append" and "replace" commands.

Tue, 18 Sep 2007 20:52:09 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.40

   * Updated setup.py file that uses distutils provided by Kai Lautaportti.

   * Prevent keys from containing ASCII character 127 as well, patch provided
     by Philip Neustrom.

   * Added ability to overload the persistent_load/id, patch provided by
     Steve Schwarz.

   * Fixed ability to pass (server_hash,key) in place of key in Client.set()
     Reported by Alexander Klyuev.

Tue, 14 Aug 2007 14:43:27 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.39

   * Michael Krause reports the previous version doesn't work for
     _val_to_store_info() calls because it's defined as a staticmethod.
     Removing staticmethod decorator.  Also confirmed by Kai Lautaportti,
     with suggested fix of removing staticmethod.

Fri, 10 Aug 2007 17:50:13 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.38

   * Matt McClanahan submitted a patch that allow add() to have a
     min_compress_len argument.

   * Steve Schwarz submitted a patch allowing user-defined picklers.

   * Michael Krause suggested checking the return value to prevent an
     exception from being raised in _set() when a value is too large to be
     stored.

Fri, 27 Jul 2007 01:55:48 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.37

   * Fixing call from add() to _set() with parameter for min_compress_len.
     Reported by Jeff Fisher.

Thu, 07 Jun 2007 04:10:31 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.36

   * Patch by Dave St.Germain to make the Client() class sub-class
     threadlocal to help with multi-threading issues.  Only available in
     Python 2.4 and above.

   * Patch by James Robinson with:
      1) new set_multi method.
      2) factored out determining the flags, length, and value to store
         from set() into method _val_to_store_info() for use by both set()
         and set_multi().
      3) send_cmds() method on host which doesn't apply the trailing '\r\n'
         for use by set_multi.
      4) check_key() extended a bit to allow for testing the prefix passed
         to set_multi just once, not once per each key.
      5) Patch also enables support for auto compression in set, set_multi,
         and replace.

   * Suggestion by Helge Tesdal, fixes in check_key for non-string keys.

   * NOTE: On a farm of clients with multiple servers, all clients will
     need to be upgraded to this version.  The next patch changes the
     server hash.

   * Philip Neustrom supplied a patch to change the server hash function to
     binascii.crc32.  The original "hash()" call is not cross-platform, so
     big and little endian systems accessing the same memcache may end up
     hitting different servers.  Restore the old functionality by calling:
     "memcached.serverHashFunction = hash" after importing memcache.

   * Philip Neustrom points out that passing Unicode keys or values causes
     problems because len(key) or len(value) is not equal to the number of
     bytes that are required to store the key/value.  Philip provides a
     patch which raises an exception in this case.  Raises
     memcache.Client.MemcachedStringEncodingError exception in this case.

   * NOTE: If you recompiled memcached to increase the default 1MB max
     value size, you will need to call "memcached.MAX_SERVER_VALUE_LENGTH = N"
     or memcached will not store values larger than the default 1MB.

   * Philip Neustrom includes another patch which checks that the key
     doesn't exceed the memcache server's max size.  If it does, the item
     is silently not stored.

   * Philip Neustrom added a bunch of sanity checks.

   * Jehiah Czebotar provided a patch to make the add() and replace()
     functions return 0 when the add or replace fails, similar to how set()
     works.

Sat, 16 Sep 2006 18:31:46 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.34

   * In get_multi, if the recv loop reads 0 bytes, raising an EOFError.
     Identified by Jim Baker.

Tue, 05 Sep 2006 14:06:50 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.33

   * Including patch from Yoshinori K. Okuji to read in larger chunks for
   readline() calls.  This should dramatically improve performance under
   some circumstances.

Sun, 03 Sep 2006 14:02:03 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.32

   * Including patch from Philip Neustrom which checks keys sent to the
   server for length and bad characters.

Sat, 20 May 2006 14:51:28 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Version 1.31

   *  Rolled version 1.30 since the Danga folks are now listing this
   version as the official version.  Removing the "tummy" from the version
   number, and incrementing so that it's clear it's more recent than "1.2".

   * Patch applied from Simon Forman for handling of weighted hosts.

   * Added a little more meat to the README.

Sat, 28 Jan 2006 15:59:50 -0700  Sean Reifschneider  <jafo@tummy.com>

   * cludwin at socallocal suggested that the write-combining with
   sendall() may not be beneficial.  After testing on both SMP and non-SMP
   machines, I can't see a significant benefit to not doing the
   write-combining, even on large strings.  The benefits of write-combining
   on smaller strings seems to be significant on UP machines in tight loops.
   Even on strings that are larger than 2MB, there seems to be no benefit to
   splitting out the writes.

Sun, 18 Sep 2005 18:56:31 -0600  Sean Reifschneider  <jafo@tummy.com>

   * Changing a printf to debuglog and catching a pickle exception, patch
   submitted by Justin Azoff.

Thu, 14 Jul 2005 11:17:30 -0700  Sean Reifschneider  <jafo@tummy.com>

   * Alex Stapleton found that the sendall call was slow for writing data
   larger than several kilobytes.  I had him test a change to his patch,
   which worked as well, but was simpler.  The code now does two sendall
   calls, one for the data and one for the line termination, if the data is
   larger than 100 bytes.

Thu, 7 Apr 2005 14:45:44 -0700  Sean Reifschneider  <jafo@tummy.com>

   * Incorporating some fixes to get_multi() from Bo Yang

Mon, 13 Dec 2004 02:35:17 -0700  Sean Reifschneider  <jafo@tummy.com>

   * Simplifying the readline() function and speeding it up ~25%.
   * Fixing a bug in readline() if the server drops, mark_dead() was not
      being properly called.

Sun, 12 Dec 2004 18:56:33 -0700  Sean Reifschneider  <jafo@tummy.com>

   * Adding "stats()" and "flush_all()" methods.

Thu, 10 Aug 2003 12:17:50 -0700  Evan Martin  <martine@danga.com>

   * Slightly more verbose self-test output.
   * Fix mark_dead() to use proper classname.
   * Make pooltest.py run from the test directory.

Thu, 07 Aug 2003 16:32:32 -0700  Evan Martin  <martine@danga.com>

   * Add incr, decr, and delete.
   * Better Python (based on comments from Uriah Welcome).
   * Docs, using epydoc.

Thu, 07 Aug 2003 14:20:27 -0700  Evan Martin  <martine@danga.com>

   * Initial prerelease.