CHANGELOG.md

v0.2.2.rc3 (git develop)

• Experimental support for 1.9.3 (all tests do pass).

v0.2.2.rc2 (git master)

• Fixed some issues related to callbacks on embedded records.

v0.2.2.rc1

• ProxyCollection destroy_all and delete_all returns removed records.
• Changes are better detected on new records. user = User.new :email => "me@gmail.com"; user.email_changed? returns true.
• Mass assignment of attributes now support multi parameter so we should be able to support form helpers for date and time.
• Added support for embeds_many and embedded_in.
• IdFactory is now configurable per class. Some classes may use IdFactory::AtomicIncrementation, others might use IdFactory::Timestamp. By default we are using AtomicIncrementation. It is configurable via Person.id_factory = MassiveRecord::ORM::IdFactory::AtomicIncrementation
• Added an IdFactory which uses Time as its generator, instead of atomic incrementation of a value.
• ORM records now responds to raw_data(). it will give you a nested hash corresponding to column families and cells with RawData objects as values. These RawData objects contains the raw value and meta data like timestamp on the cell from Hbase.
• Give a string to a time attribute will now do a Time.zone.parse on the string to set it with correct time zone.
• MassiveRecord::ORM::Column is now named Embedded. You need to update your code! I guess in most cases not many have ever used Column, as its usage has been limited up until we implement the embedded relations.
• Reworked how the persistence module actually does the database specific calls like save, update and destroy. Before, all of the hbase-table-specific code lived inside of the Persistence module. It has now been extracted out into small Persistence::Operations classes. This should enable us to customize the save operation based on which context we are in (table or an embedded object).
• If you, for some reason, need to change id on an existing record, you may do so with change_id!("new_id").
• Optimization on references many proxy. It is now a bit smarter when you do any of: length, include?, present? and any?. Previously it loaded all the targets to figure out the length and if it included a record. Now it does these more efficient based on the proxy state.
• You can now give options when auto loading fields. For instance if all your fields are expected to be integers you can do column_family(:something) { autoload_fields :type => :integer }.
• Context of validations are now set to :create or :update automatically, so that validates :something, :validator => true, :on => :update works.
• Fetching multiple ids via Thrift is 4x faster.
• Assigning integers or float values as strings is now ran through to_i or to_f in the writer method.
• A references_many proxy now supports all(options). I would like to give better support for scopes on relations as well, as you right now cannot do a_person.cars.limit(2).offset("id-to-start-at"). The limit(2) will actually return the two first records.
• Scope methods like Person.limit(2).offset("some-id") now returns a cloned version of the previous one. This keeps state of one scope apart from the other and fixes obvious problems which can arise if building and using scopes.
• Added starts_with and offset to the scope, so now you can do: Person.starts_with("id-have-to-start-with-this").offset("id-have-to-start-with-this-and-starts-read-from-this").limit(1)
• Finder option :start has been deprecated and renamed to :starts_with.
• Added a basic IdentityMap. Calling User.find("an-id") twice will only load it once from the database. Calling User.find(["an-id", "another"]) will only load "another" from the database, and the "an-id" from the identity map. The identity map needs to be enabled. In Rails you can add a configuration option in your config/application.rb like: config.massive_record.identity_map = true. This will insert a Rack middleware which enables identity map per request. Note that the identity map only caches objects per request.
• When you do a find without any :select option given we will add known column families to be selected as default.
• do_find() (internal method) was re-factored to be more readable and easier to extend/hook in to.
• References many can now handle find_in_batches and find_each.
• Fixed a problem with utf-8 encoded strings in ids. The Thrift adapter will no longer blow up.
• Added support for Observers. See MassiveRecord::ORM::Observer.
• Fixed a nasty bug with default scope. Guess we have not used default scope that much not noticing this one until now.
• Fixed a couple of issues related to STI. We are filtering on type now, so doing find() on a subclass will no longer find superclasses. (Or, it will find it, but it will be filtered away before result are returned. Even beter if this could have been done in the database..)
• Atomic decrements where added to the Thrift adapter and the ORM.

v0.2.1

• Models without any default_id will now by default get an id via next_id(). You can turn it off via the setting set_id_from_factory_before_create on ORM::Base or on the Model class itself.
• record.reload now resets relations.
• If you have a persisted record and you set one attribute to nil that attribute will be deleted from HBase to represent the nil-value. The fact that the schema of that record class knows of that attribute will be the reason for it to still respond and return nil.
• Thrift adapter will no longer return UTF-8 encoded strings. The reason for this is that binary representation of integers cannot be set with a UTF-8 encoding; it must be BINARY. It is the client`s responsibility of the adapter to set correct encoding, and the ORM now does this too.
• If you do have a database where integers are stored as string, you should enable ORM::Base.backward_compatibility_integers_might_be_persisted_as_strings. It will, before for instance atomic_increment! is called, ensure that the integers has been persisted as hex.
• Based on the previous change we are now able to do real atomic incrementation of integer values. HBase will do the incrementation and guarantee the integrity of that incrementation.
• Fixnum and Bignum are stored as a 64 bit signed integer representation, not as strings. Fixnum and bignum values are no longer encoded by the ORM; that responsibility has been moved down to the adapter.
• We can now, by setting Base.check_record_uniqueness_on_create to true, do a quick and simple (and kinda insecure) check if the record id exists on a new_record when doing a create call. Just a simple sanity check.
• We now have mass assignment of attributes, just like ActiveRecord. It uses the same module, so attr_accessible and attr_protected methods are available. By default the id and inheritable attribute are protected. If you where doing Person.new(:id => 'ID') you might want to change this now as it will no longer work.
• Ids can now be assigned on new/create/create! with: Person.new("id", attributes). Person.new(id: "id") will is soon to be disallowed.
• Polymorphic type attribute is no longer called underscore on. Should be backwards compatible when finding records.
• Time can now be time zone aware. Times are being presented in the Time.zone, and persisted in UTC. You enable it with MassiveRecord::ORM::Base.time_zone_aware_attributes = true. In a Rails application this will be set to true by default.
• No longer checking if table exist on a find. Instead we rescue error the few times a table does not exist, and create it based on current schema. This gained a lot of speed on our production server.
• Changed the way Thrift::Table executes find(). It is now doing the work ~3 times faster.
• In Rails, we are now included in the 200-ok-completed-log like: Completed 200 OK in 798ms (Views: 277.5ms | MassiveRecord: 9.2ms)
• Subscribed to events from query instruments and printing out time spent in database per call.
• Added ActiveSupport Notifications instruments around database query calls in the ORM level.
• Removed inclusion of ActiveModel::Translation in ORM::Base class, as including ActiveModel::Validations extends ORM::Base with Translation as well (and it should never have been included; it should have been extended with..)

v0.2.0

• Intersection and union operations on arrays containing MassiveRecord objects is now working as expected.
• You can now disallow nil values, and in that case we will ensure that given field has its default value.
• Rails will now handle MassiveRecord::ORM::RecordNotFound correctly in a production environment, rendering 404.
• record.attributes no longer returns internal @attributes-hash directly. Instead it iterates over all attributes, fetches their values through read_attribute and then returns a new hash with these attribute-name-values pairs.
• Fixed problem with STI sub classes in Rails development environment: Attributes defined in a sub class was not loaded correctly first time you loaded a record through its parent class.

v0.2.0.beta2

• We are now raising error if MassiveRecordClass.new(attributes) receives unknown attributes.
• Added support for Record.new(:references_many => [record, record]) and a_record.references_many = [record, record]

v0.2.0.beta

• ORM will now take care of serialize and de-serialize of attributes like arrays, hashes etc. It is doing so based on the type of your fields. You can select either JSON or YAML serialization for your data. As a default it will use JSON. You can also, by chaining multiple coders together add support for multiple serialization types when reading data from the database.
• Thrift-adapter will no longer auto-serialize objects likes hashes and arrays. Its vlaues must now be strings, and it will only take care of encoding/decoding it to and from what Thrift expects (binary encoding).
• Compare Person === proxy_targeting_a_person will now be true. Makes case-when-constructions doable.
• Single table inheritance is supported. By default you can have an attribute called type to give you support for it in a table.
• A default_scope is possible to set on classes. For instance: Calling default_scope select(:only_this_column_family) inside of a class will execute finder operations with this as default scope. If you need to fetch records of class without your preset default scope you can use Model.unscoped.
• We now have some ActiveRecord like chaining of method calls when we do find-operations. Like Person.select(:column_family).limit(2) is the same as Person.all(:select => ['column_family', :limit => 2])
• references_many has first() and limit() which uses the target array if loaded, or load only what it needs from the database.
• Wrapper::Thrift has been moved into Adapter::Thrift. Adding more adapters should be not that hard now.
• References many is now possible. We have to strategies: Store an array of foreign keys in the proxy_owner, or supply a ids-starts-with and open up a scanner and read from that point.
• Setting a non-parsable value on date/time field will no longer raise an error.
• Scanner no longer fetches with a limit of 10 by default. It is set to 100000000.
• References one relations support polymorphic relations.
• Simple implementation of references_one relation. This is where you have a foreign key you will look up in a different table.

v0.1.2

• Fixed, or at least made better, the is_yaml? method in Wrapper::Cell.This functionality of serialize/de-serialize should be moved up into the ORM asap, but for now a hot fix has been applied.

v0.1.1

• A ORM record now now if it is read only or not.
• Added a logger to ORM::Base, which is set to Rails.logger if gem is used with Rails.
• The database cleaner will no longer destroy tables, only delete all of their contents between tests. Speed tests up a lot.
• known_attribute_names are now available as instance method as well.
• If you add a created_at attribute it will be maintained by the ORM with the time the object was created.
• An adapter (wrapper) row now has an updated_at attribute. ORM objects also responds to updated_at
• Bugfix: Database cleaner no longer tries to remove tables with same name twice.

v0.1.0

• Communication with Hbase via Thrift.
• Basic ORM capabilities.