* 1.3.6
    * Implementation of all Hash commands
    * Pipelines now wrap their execution with MULTI and EXEC commands to
      process all commands atomically.
    * Connections can now set timeout. If command execution exceeds the
      timeout, an exception is raised.
    * Numerous bug fixes and more tests.
* 1.3.4
    * Skipped version numbers ahead so that the client version matches the
      Redis version it is feature-compatible with. Going forward, the client
      will stay in sync with Redis version numbers when client updates are
      made.
    * Completely refactored the client library. It's now trivial to maintain
      and add new commands. The library is also much more consistent.
    * With the exception of "Response value type inference" (see below), the
      client should be backwards compatible with 0.6.1. Some older, less
      consistent methods will emit DeprecationWarnings, indicating that you
      should use another command or option, but these should continue to
      work as expected for the next few releases.
    * WARNING: BACKWARDS INCOMPATIBLE CHANGE: "Response value type inference"
      Previously, all values returned from Redis went through a decoding
      process. In this process, if the response was numeric, it would be
      automatically converted to an int or float type prior to being returned.
      Otherwise the response would be decoded as a unicode string. This meant
      that storing the string "123" would actually return an integer 123, and
      that the string "foo" would be returned as the unicode object u"foo".
      This fundamentally breaks the retrieval of binary data (byte strings) and
      values that might accidentally look like a number (a hash value). After
      discussing this in detail with a number of users and on the Redis mailing
      list (http://groups.google.com/group/redis-db/browse_thread/thread/9888eb9ff383c90c/ec44fe80b6400f7b#ec44fe80b6400f7b)
      *ALL* values returned from methods such as get() now return raw
      Python strings. It is now your responsibility to convert that data to
      whatever datatype you need. Other methods that *always* return integer
      or float values, such as INCR, DECR, LLEN, ZSCORE, etc., will continue
      returning values of the appropriate type. This resolves issue #2, #8
      and #11:
      http://github.com/andymccurdy/redis-py/issues#issue/2
      http://github.com/andymccurdy/redis-py/issues#issue/8
      http://github.com/andymccurdy/redis-py/issues#issue/11
    * The "select" method now takes a "host" and "port" argument in addition
      to the database. Behind the scenes, select() swaps out the underlying
      socket connection. This resolves issue #4:
      http://github.com/andymccurdy/redis-py/issues#issue/4
    * The client now supports pipelining of Redis commands. Use the pipeline()
      method to create a new Pipeline object. Each command called on the
      pipeline object will be buffered until the pipeline if executed.
      A list of each command's results will be returned by execution.  Use
      this for batch processing in order to eliminate multiple request/response
      cycles.
      
* 0.6.1
    * Added support for ZINCRBY via the `zincr` command
    * Swapped score and member parameters to zadd to make it more similar to other commands.
    * Added support for Python 2.4 (thanks David Moss)
* 0.6.0 Changed to Andy McCurdy's codebase on github
* 0.5.5 Patch from David Moss, SHUTDOWN and doctest bugfix
* 0.5.1-4 Bugfixes, no code changes, just packaging, 10/2/09
* 0.5 Initial release, redis.py version 1.0.1, 10/2/09