2012-05-19: version 1.30

	* Fix buck2sock corruption that causes errors like
	  "No map found matching for GLOB" or "No sock found for" (hachi@cpan.org)

	* Fix t/05_reconnect_timeout.t to not fail on networks that fast-reject
	  TEST-NET-1 https://rt.cpan.org/Ticket/Display.html?id=74500

	* Fix warning about undefined value in GetParser.pm (kost@ropnet.ru)

	* Add documentation of connect_timeout and select_timeout

2010-06-17: version 1.29

	* fix warnings spew if calling stats on a down server

	* fix buck2sock confusion (Eddie Canales)

	* quell ipv6-related warning
2009-10-21: version 1.28

	* IPv6 support (https://rt.cpan.org/Ticket/Display.html?id=50577)

	* Add Encode.pm requirement to Makefile.PL for perl 5.6.x (Ask).

2009-09-22: version 1.27

	* Fix get() with utf-8 keys  (athomason)

	* "stats malloc" command is gone in 1.4; remove those tests (athomason)

	* Add append/prepend support (dormando)
	* Fix occasional failure in the 100_flush_bug.t test (Ask Bjørn Hansen)

2009-05-04: version 1.26

	* don't include "stats sizes" by default in the stats method,
	  as that can hang big servers for a few seconds (Brad Fitzpatrick)
2009-05-02: version 1.25

	* Clear @buck2sock when calling disconnect_all.  (Dennis Stosberg,
	  [rt.cpan.org #45560]
	* Reconnects to a dead connection shouldn't happen every time when the
	  connection has never succeded. Apply the dead timeout to sockets that
	  never even came up. Add a test.

	* Warn when trying to put undef values into memcache.
	  (Henry Lyne <hlyne@livejournalinc.com>)
	* flush_all now only returns success if there is a proper reply from all
	  servers - Yann Kerherve <yann@sixapart.com>
	* 'noreply' support from Tomash Brechko <tomash.brechko@gmail.com>

	* various test updates from Ronald J Kimball <rkimball@pangeamedia.com>

2007-07-17: version 1.24

	* update the stats method, including tests for it
	  (Ronald J Kimball <rkimball+memcached@pangeamedia.com>)

	* arguments to 'new' constructor can be %hash or $hashref now
	  (previously was only $hashref)

	* work around a Perl segfault (Matthieu PATOU <mp@oxado.com>)
	  see http://lists.danga.com/pipermail/memcached/2007-June/004511.html

2007-06-19: version 1.23

	* add 'remove' as an alias for 'delete' (Dave Cardwell <dave@davecardwell.co.uk>)

2007-06-18: version 1.22

	* lost connection handling broken due to wrong %sock_map indexing
	  fix from RHESA

	* let parser_class be configured as a constructor option,
	  defaulting to XS if available, else regular.  (unless
	  $ENV{NO_XS} is set, in which case the default is regular)

2007-05-02: version 1.21

	* new faster optional interface for GetParser subclasses.  doing
	  this release so upcoming Cache::Memcached::GetParserXS can
	  depend on this.  otherwise this release isn't interesting.

2007-04-16: version 1.20

        * fix "Warning produced when flush_all called" from CDENT

	* support access via unix domain sockets. (Nathan Neulinger <nneul@umr.edu>)

	* abstract out response parsing into own class, and add XS-module
	  detection, so if you have the XS (C) version, things'll be faster.
	  that part's not done yet.

	* don't use dual scalar/glob sockets.  makes it all profilable
	  again under SmallProf, DProf, and Devel::Profiler, all three
	  of which used to barf on those weird sockets previously
	* only init_buckets once, when servers are changed
	* don't call sock_to_host and get_sock as much:  cache closer
	  in get_multi
	* more internal caching (buck2sock, etc)
	* fast paths for namespaces/single sock/etc in a few more places
	* general micro-speedups all over

	* patch from Maxim Dounin <mdounin@rambler-co.ru> to fix a typo
	  which caused no_rehash flag to not work.
	* release 1.18

	* flush_all command from Patrick Michael Kane <modus-memcache@pr.es.to>
	* document namespaces
	* release 1.17

	* fix stats method (people have only been asking since 2004  :-/)
	* add tests
	* move Memcached to lib/Cache directory to be more CPAN-friendly
	* release 1.16

	* configurable connect delays and callback on connect failure (brad)
	* release 1.15

        * _connect_sock never works in blocking mode because of a bug in setting
	the default timeout. (Gaal)

	* release 1.14

	* kill buggy, slow ord() _hashfunc, replace with crc32.
	  this adds String::CRC32 as a dependency.  thanks to
	  everybody's feedback on the mailing list.

	* don't use pos() because it doesn't seem to work in
	  taint mode.  use $+[0] instead.  (Dave Evans <..@djce.org.uk>)

	* support multiple paths to memcache nodes (Brad)
	  see 'set_pref_ip'

	* release version 1.13

2004-05-26 (Whitaker <whitaker@danga.com>)
	* quiet warning

2004-05-25 (Whitaker <whitaker@danga.com>)
	* get_multi shouldn't modify caller's @_

2004-05-18 (Michael <ged@danga.com>)
	* namespace support
	* use fields

2004-05-16 (Alexei Kozlov <kozlov@w-m.ru>)
	* remove warnings with vec

2004-04-09 (brad)
	* in perl 5.6, trap errors dethawing 5.8 storable objects
	  and instead treat it like a cache miss

	* use $! and not %! for perl 5.6 compat (Dave Evans <..@djce.org.uk>)
	* don't mark whole IP dead anymore when a node is down (Jason Titus <jtitus@postini.com>)
	* start version numbering (Jamie McCarthy <jamie@mccarthy.vg>)

2004-03-09 (Brad/Avva)
	* _oneline can return more than one line (but always on a line break),
	  so caller must decide when it's really time to quit.  had to modify
	  run_command to know that.  (which is used by stats)

2004-03-05 (Dave Evans <..@djce.org.uk>)
	* Here's a really trivial patch for the Perl binding,
	Cache::Memcached.  The bug is that the module assumes that the
	currently select()ed filehandle is STDOUT, but this might not be
	the case.  So this patch ensures that the select()ed filehandle is
	preserved, not forced to STDOUT.

2004-02-29 (Brad)
	* add readonly option

2004-02-27 (Avva)
	* Cleaner handling of the case when _oneline is called without a
	  line parameter (i.e. not to send anything, just read a line from
	  the socket).  Make it depend on $line being defined only,
	  regardless of its content (thanks Brad!).

2004-02-25 (Avva)
	* Asyncify all I/O, finally get rid of alarm() yuckiness, unify all
	  one-liner command/responses into a single internal API.
	* document in POD the delete method
	* fix bug with 2k read boundaries falling in the middle
	  of "VALUE ..." or "END" lines, thus halting future
	  parsing and responses.  (eek!)
	* version 1.0.12
	* merge stats/stats_reset patch from Jamie McCarthy
	* trailing whitespace cleanup

	* work on Solaris/BSD where there's no MSG_NOSIGNAL.
	  the expense is extra syscalls to change the local
	  SIGPIPE handler all the time.  in the future, it'd 
	  be nice to have an option so Solaris/BSD callers
	  can say, "Hey, I've turned off SIGPIPE globally,
	  don't worry about it."

	* add a test file, so automated CPAN test hosts are happy
	* check MSG_NOSIGNAL immediately on module load, not on use,
	  so Solaris dies early.  (still on TODO to fix, but better
	  to fail loudly)
	* version 1.0.11

	* version 1.0.10, rename to Cache::Memcached, upload to CPAN
	* implement read/write timeouts everywhere. Now the client shouldn't
          hang if the server machine goes down unexpectedly. (avva)

	* use Storable::nfreeze instead of freeze, so hosts from different
	  architectures can all use the same data.  (all must use Perl, though.
	  the different memcache APIs all store/pickle/serialize data differently)
	  Suggestion by Jason Titus <jtitus@postini.com>
	* fix _incrdecr to return complete number, not just first
	  digit (thanks to Ryan T. Dean)
	* release version 1.0.9

	* document expiration times in POD (thanks to Tim Bunce
	  for noting the omission)
	* release version 1.0.8
	* add connect timeout of 0.25s, for dead host detection.
	  We had 1 second a couple revs ago, but lost it when
	  ditching IO::Socket module.  (avva)
	* fix _incrdecr with explicit-hashvalue keys (whitaker)
	* add run_command API call.  TODO: document, and document
	  the $exptime on the setters
	* use send instead of print, so we can set MSG_NOSIGNAL
	  and not get SIGPIPES, which avoids 3 syscalls of localizing
	  $SIG{PIPE} and sends everything at once, instead of 4k
	  stdio chunks.  in review:  stdio buffered in, send unbuffered
	  out.  TODO:  setvbuf so reads are buffered at more than 4k.

	* yet faster parsing
	* switch to stdio/perlio instead of raw io: more correct,
	  simpler parsing code.

	* prevent some warnings
	* faster get() call that doesn't use get_multi()
	* optimizations for single-server case
	* use socket APIs directly, instead of uber-slow IO::* modules
	* new faster _load_items parsing
	* emit debug when set/add/replace fails, in addition to succeed
Version 1.0.7
	-- compression support (Brad Whitaker)
Version 1.0.6
	-- incr/decr client support
	-- make delete optionally take second argument (server now supports
	   a delay time on delete)
	-- doc updates from Jamie McCarthy
	-- better hashing after dead host detection:  new requests go to different
	   remaining hosts, instead of all to the same one.

Version 1.0.2
	-- initial release, about.