This release notes cover the major changes, but don't include every little bug fix and change. If you want to see everything, check out the [list of commits](https://github.com/rails/rails/commits/master) in the main Rails repository on GitHub.
If you're upgrading an existing application, it's a great idea to have good test coverage before going in. You should also first upgrade to Rails 3 in case you haven't and make sure your application still runs as expected before attempting to update to Rails 3.1. Then take heed of the following changes:
Rails 3.1 requires Ruby 1.8.7 or higher. Support for all of the previous Ruby versions has been dropped officially and you should upgrade as early as possible. Rails 3.1 is also compatible with Ruby 1.9.2.
TIP: Note that Ruby 1.8.7 p248 and p249 have marshaling bugs that crash Rails. Ruby Enterprise Edition have these fixed since release 1.8.7-2010.02 though. On the 1.9 front, Ruby 1.9.1 is not usable because it outright segfaults, so if you want to use 1.9.x jump on 1.9.2 for smooth sailing.
Rails now uses a `Gemfile` in the application root to determine the gems you require for your application to start. This `Gemfile` is processed by the [Bundler](https://github.com/carlhuda/bundler) gem, which then installs all your dependencies. It can even install all the dependencies locally to your application so that it doesn't depend on the system gems.
`Bundler` and `Gemfile` makes freezing your Rails application easy as pie with the new dedicated `bundle` command. If you want to bundle straight from the Git repository, you can pass the `--edge` flag:
The major change in Rails 3.1 is the Assets Pipeline. It makes CSS and JavaScript first-class code citizens and enables proper organization, including use in plugins and engines.
The assets pipeline is powered by [Sprockets](https://github.com/sstephenson/sprockets) and is covered in the [Asset Pipeline](asset_pipeline.html) guide.
HTTP Streaming is another change that is new in Rails 3.1. This lets the browser download your stylesheets and JavaScript files while the server is still generating the response. This requires Ruby 1.9.2, is opt-in and requires support from the web server as well, but the popular combo of nginx and unicorn is ready to take advantage of it.
Active Record has an Identity Map in Rails 3.1. An identity map keeps previously instantiated records and returns the object associated with the record if accessed again. The identity map is created on a per-request basis and is flushed at request completion.
Rails 3.1 comes with the identity map turned off by default.
* The application generator accepts an option `-j` which can be an arbitrary string. If passed "foo", the gem "foo-rails" is added to the `Gemfile`, and the application JavaScript manifest requires "foo" and "foo_ujs". Currently only "prototype-rails" and "jquery-rails" exist and provide those files via the asset pipeline.
* The controller and resource generators will now automatically produce asset stubs (this can be turned off with `--skip-assets`). These stubs will use CoffeeScript and Sass, if those libraries are available.
* Specify `force_ssl` in a controller to force the browser to transfer data via HTTPS protocol on that particular controller. To limit to specific actions, `:only` or `:except` can be used.
* Added `ActionController::ParamsWrapper` to wrap parameters into a nested hash, and will be turned on for JSON request in new applications by default. This can be customized in `config/initializers/wrap_parameters.rb`.
* Added `config.action_controller.include_all_helpers`. By default `helper :all` is done in `ActionController::Base`, which includes all the helpers by default. Setting `include_all_helpers` to `false` will result in including only application_helper and the helper corresponding to controller (like foo_helper for foo_controller).
You can restrict it to some actions by using `:only` or `:except`. Please read the docs at [`ActionController::Streaming`](http://api.rubyonrails.org/classes/ActionController/Streaming.html) for more information.
* The redirect route method now also accepts a hash of options which will only change the parts of the url in question, or an object which responds to call, allowing for redirects to be reused.
*`config.action_dispatch.x_sendfile_header` now defaults to `nil` and `config/environments/production.rb` doesn't set any particular value for it. This allows servers to set it through `X-Sendfile-Type`.
* Rely on `Rack::Session` stores API for more compatibility across the Ruby world. This is backwards incompatible since `Rack::Session` expects `#get_session` to accept four arguments and requires `#destroy_session` instead of simply `#destroy`.
* Allows `FormHelper#form_for` to specify the `:method` as a direct option instead of through the `:html` hash. `form_for(@post, remote: true, method: :delete)` instead of `form_for(@post, remote: true, html: { method: :delete })`.
* Provided `JavaScriptHelper#j()` as an alias for `JavaScriptHelper#escape_javascript()`. This supersedes the `Object#j()` method that the JSON gem adds within templates using the JavaScriptHelper.
* Added a class method `pluralize_table_names` to singularize/pluralize table names of individual models. Previously this could only be set globally for all models through `ActiveRecord::Base.pluralize_table_names`.
* Added `ActiveRecord::Base.attribute_names` to return a list of attribute names. This will return an empty array if the model is abstract or the table does not exist.
*`ActiveRecord#new`, `ActiveRecord#create` and `ActiveRecord#update_attributes` all accept a second hash as an option that allows you to specify which role to consider when assigning attributes. This is built on top of Active Model's new mass assignment capabilities:
* Default scopes are now evaluated at the latest possible moment, to avoid problems where scopes would be created which would implicitly contain the default scope, which would then be impossible to get rid of via Model.unscoped.
* PostgreSQL adapter only supports PostgreSQL version 8.2 and higher.
* Added an `update_column` method on Active Record. This new method updates a given attribute on an object, skipping validations and callbacks. It is recommended to use `update_attributes` or `update_attribute` unless you are sure you do not want to execute any callback, including the modification of the `updated_at` column. It should not be called on new records.
* Associations with a `:through` option can now use any association as the through or source association, including other associations which have a `:through` option and `has_and_belongs_to_many` associations.
*`ActiveRecord::Associations::AssociationProxy` has been split. There is now an `Association` class (and subclasses) which are responsible for operating on associations, and then a separate, thin wrapper called `CollectionProxy`, which proxies collection associations. This prevents namespace pollution, separates concerns, and will allow further refactorings.
* Singular associations (`has_one`, `belongs_to`) no longer have a proxy and simply returns the associated record or `nil`. This means that you should not use undocumented methods such as `bob.mother.create` - use `bob.create_mother` instead.
* Support the `:dependent` option on `has_many :through` associations. For historical and practical reasons, `:delete_all` is the default deletion strategy employed by `association.delete(*records)`, despite the fact that the default strategy is `:nullify` for regular has_many. Also, this only works at all if the source reflection is a belongs_to. For other situations, you should directly modify the through association.
* The behavior of `association.destroy` for `has_and_belongs_to_many` and `has_many :through` is changed. From now on, 'destroy' or 'delete' on an association will be taken to mean 'get rid of the link', not (necessarily) 'get rid of the associated records'.
* Previously, `has_and_belongs_to_many.destroy(*records)` would destroy the records themselves. It would not delete any records in the join table. Now, it deletes the records in the join table.
* Previously, `has_many_through.destroy(*records)` would destroy the records themselves, and the records in the join table. [Note: This has not always been the case; previous version of Rails only deleted the records themselves.] Now, it destroys only the records in the join table.
* Note that this change is backwards-incompatible to an extent, but there is unfortunately no way to 'deprecate' it before changing it. The change is being made in order to have consistency as to the meaning of 'destroy' or 'delete' across the different types of associations. If you wish to destroy the records themselves, you can do `records.association.each(&:destroy)`.
* Migrations are now reversible, meaning that Rails will figure out how to reverse your migrations. To use reversible migrations, just define the `change` method.
* Some things cannot be automatically reversed for you. If you know how to reverse those things, you should define `up` and `down` in your migration. If you define something in change that cannot be reversed, an `IrreversibleMigration` exception will be raised when going down.
* Migration files generated from model and constructive migration generators (for example, add_name_to_users) use the reversible migration's `change` method instead of the ordinary `up` and `down` methods.
Inside the proc, `self` is the object which is the owner of the association, unless you are eager loading the association, in which case `self` is the class which the association is within.
* Previously `:insert_sql` and `:delete_sql` on `has_and_belongs_to_many` association allowed you to call 'record' to get the record being inserted or deleted. This is now passed as an argument to the proc.
* Added `ActiveRecord::Base#has_secure_password` (via `ActiveModel::SecurePassword`) to encapsulate dead-simple password usage with BCrypt encryption and salting.
* Calling `ActiveRecord::Base#dup` will duplicate the record, including calling after initialize hooks. Frozen state will not be copied, and all associations will be cleared. A duped record will return `true` for `new_record?`, have a `nil` id field, and is saveable.
*`InclusionValidator`, `ExclusionValidator`, and `FormatValidator` now accepts an option which can be a proc, a lambda, or anything that respond to `call`. This option will be called with the current record as an argument and returns an object which respond to `include?` for `InclusionValidator` and `ExclusionValidator`, and returns a regular expression object for `FormatValidator`.
* The default format has been changed to JSON for all requests. If you want to continue to use XML you will need to set `self.format = :xml` in the class. For example,
See the [full list of contributors to Rails](http://contributors.rubyonrails.org/) for the many people who spent many hours making Rails, the stable and robust framework it is. Kudos to all of them.