(unreleased)

- Fetching a batch of objects is now done in batches of 1000 objects at
  a time. If you are iterating over large collections, this change should
  provide a significant performance boost both in used memory and total
  execution time.
- MutableSet#<< is now an alias for #add.

1.3.1

- Improve memory consumption when indexing persisted attributes.

  No migration is needed and old indices will be cleaned up as you save
  instances.

1.3.0

- Add Model.attributes.

1.2.0

- Enumerable fix.
- Merge Ohm::PipelinedFetch into Ohm::Collection.
- Fix Set, MultiSet, and List enumerable behavior.
- Change dependencies to use latest cutest.

1.1.0

- Compatible with redis-rb 3.

1.0.0

- Fetching a batch of objects is now done through one pipeline, effectively
  reducing the IO to just 2 operations (one for SMEMBERS / LRANGE, one for
  the actual HGET of all the individual HASHes.)
- write_remote / read_remote have been replaced with set / get respectively.
- Ohm::Model.unique has been added.
- Ohm::Model::Set has been renamed to Ohm::Set
- Ohm::Model::List has been renamed to Ohm::List
- Ohm::Model::Collection is gone.
- Ohm::Validations is gone. Ohm now uses Scrivener::Validations.
- Ohm::Key is gone. Ohm now uses Nest directly.
- No more concept of volatile keys.
- Ohm::Model::Wrapper is gone.
- Use Symbols for constants instead of relying on Ohm::Model.const_missing.
- #sort / #sort_by now uses `limit` as it's used in redis-rb, e.g. you
  have to pass in an array like so: sort(limit: [0, 1]).
- Set / List have been trimmed to contain only the minimum number
  of necessary methods.
- You can no longer mutate a collection / set as before, e.g. doing
  User.find(...).add(User[1]) will throw an error.
- The #union operation has been added. You can now chain it with your filters.
- Temporary keys when doing finds are now automatically cleaned up.
- Counters are now stored in their own key instead, i.e. in
  User:<id>:counters.
- JSON support has to be explicitly required by doing `require
  "ohm/json"`.
- All save / delete / update operations are now done using
  transactions (see http://redis.io/topics/transactions).
- All indices are now stored without converting the values to base64.