SchemaPlus

SchemaPlus is an ActiveRecord extension that provides enhanced capabilities for schema definition and querying, including: enhanced and more DRY index capabilities, support and automation for foreign key constraints, and support for views.

For added rails DRYness see also the gems schema_associations and schema_validations

Compatibility

SchemaPlus supports all combinations of:

• Rails/ActiveRecord 3.2, 4.0, and 4.1 (currently 4.1.1)
• PostgreSQL, MySQL (using mysql2 gem; mysql gem only supported with Rails 3.2), or SQLite3 (using sqlite3 >= 3.7.7 which has foreign key support)
• MRI Ruby 1.9.3, 2.0.0, or 2.1.0

And also supports:

• jruby with Rails/ActiveRecord 3.2 and PostgreSQL or MySQL

Installation

Install from http://rubygems.org via

$ gem install "schema_plus"

or in a Gemfile

gem "schema_plus"

Features

This README lists the major features, with examples of use. For full details see the RDoc documentation.

Indexes

With standard Rails 3 migrations, you specify indexes separately from the table definition:

# Standard Rails approach...
create_table :parts do |t|
  t.string :name
  t.string :product_code
end

add_index :parts, :name     # index repeats table and column names and is defined separately
add_index :parts, :product_code, unique: true

But with SchemaPlus you can specify your indexes when you define each column, with options as desired

# More DRY way...
create_table :parts do |t|
  t.string :name,           index: true
  t.string :product_code,   index: { unique: true }
end

The options hash can include an index name:

t.string :product_code,     index: { unique: true, name: "my_index_name" }

You can also create multi-column indexes, for example:

t.string :first_name
t.string :last_name,        index: { with: :first_name }

t.string :country_code
t.string :area_code
t.string :local_number,      index: { with: [:country_code, :area_code], unique: true }

And you can specify index orders:

t.string :first_name
t.string :last_name,        index: { with: :first_name, order: { first_name: :desc, last_name: :asc}}

As a convenient shorthand, the :unique option can be specified as

t.string :product_code,   index: :unique

which is equivalent to

t.string :product_code,   index: { unique: true }

If you're using PostgreSQL, SchemaPlus provides support for conditions, expressions, index methods, and case-insensitive indexes:

t.string :last_name,  index: { conditions: 'deleted_at IS NULL' }
t.string :last_name,  index: { expression: 'upper(last_name)' }
t.string :last_name,  index: { kind: 'hash' }
t.string :last_name,  index: { case_sensitive: false }        # shorthand for expression: 'lower(last_name)'

These features are available also in ActiveRecord::Migration.add_index. See doc for SchemaPlus::ActiveRecord::ConnectionAdapters::PostgresqlAdapter and SchemaPlus::ActiveRecord::ConnectionAdapters::IndexDefinition

When you query column information using ActiveRecord::Base#columns, SchemaPlus analogously provides index information relevant to each column: which indexes reference the column, whether the column must be unique, etc. See doc for SchemaPlus::ActiveRecord::ConnectionAdapters::Column.

SchemaPlus also tidies some index-related behavior:

• Rails' various db adapters have inconsistent behavior regarding an attempt to create a duplicate index: some quietly ignore the attempt, some raise an error. SchemaPlus regularizes the behavor to ignore the attempt for all db adapters.

• If you rename a table, indexes named using rails' automatic naming convention will be renamed correspondingly.

• remove_index now accepts an :if_exists option to prevent errors from attempting to remove non-existent indexes.

Foreign Key Constraints

SchemaPlus adds support for foreign key constraints. In fact, for the common convention that you name a column with suffix _id to indicate that it's a foreign key, SchemaPlus automatically defines the appropriate constraint.

SchemaPlus also creates foreign key constraints for rails' t.references or t.belongs_to, which take the singular of the referenced table name and implicitly create the column suffixed with _id.

You can explicitly specify whether or not to generate a foreign key constraint, and specify or override automatic options, using the :foreign_key keyword

Here are some examples:

t.integer :author_id                              # automatically references table 'authors', key id
t.integer :parent_id                              # special name parent_id automatically references its own table (for tree nodes)
t.integer :author_id, foreign_key: true           # same as default automatic behavior
t.integer :author,    foreign_key: true           # non-conventional column name needs to force creation, table name is assumed to be 'authors'
t.integer :author_id, foreign_key: false          # don't create a constraint

t.integer :author_id, foreign_key: { references: :authors }        # same as automatic behavior
t.integer :author,    foreign_key: { reference: :authors}          # same default name
t.integer :author_id, foreign_key: { references: [:authors, :id] } # same as automatic behavior
t.integer :author_id, foreign_key: { references: :people }         # override table name
t.integer :author_id, foreign_key: { references: [:people, :ssn] } # override table name and key
t.integer :author_id, foreign_key: { references: nil }             # don't create a constraint
t.integer :author_id, foreign_key: { name: "my_fk" }               # override default generated constraint name
t.integer :author_id, foreign_key: { on_delete: :cascade }
t.integer :author_id, foreign_key: { on_update: :set_null }
t.integer :author_id, foreign_key: { deferrable: true }
t.integer :author_id, foreign_key: { deferrable: :initially_deferred }

Of course the options can be combined, e.g.

t.integer :author_id, foreign_key: { name: "my_fk", on_delete: :no_action }

As a shorthand, all options except :name can be specified without placing them in a hash, e.g.

t.integer :author_id, on_delete: :cascade
t.integer :author_id, references: nil

The foreign key behavior can be configured globally (see Config) or per-table (see create_table).

To examine your foreign key constraints, connection.foreign_keys returns a list of foreign key constraints defined for a given table, and connection.reverse_foreign_keys returns a list of foreign key constraints that reference a given table. See doc at SchemaPlus::ActiveRecord::ConnectionAdapters::ForeignKeyDefinition.

Foreign Key Issues

Foreign keys can cause issues for Rails utilities that delete or load data because referential integrity imposes a sequencing requirement that those utilities may not take into consideration. Monkey-patching may be required to address some of these issues. The Wiki article Making yaml_db work with foreign key constraints in PostgreSQL has some information that may be of assistance in resolving these issues.

Tables

SchemaPlus extends rails' drop_table method to accept these options:

drop_table :table_name                    # same as rails
drop_table :table_name, if_exists: true   # no error if table doesn't exist
drop_table :table_name, cascade: true     # delete dependencies

The :cascade option is particularly useful given foreign key constraints. For Postgresql it is implemented using DROP TABLE...CASCADE which deletes all dependencies. For MySQL, SchemaPlus implements the :cascade option to delete foreign key references, but does not delete any other dependencies. For Sqlite3, the :cascade option is ignored, but Sqlite3 always drops tables with cascade-like behavior.

SchemaPlus likewise extends create_table ... force: true to use :cascade

Views

SchemaPlus provides support for creating and dropping views. In a migration, a view can be created using a rails relation or literal sql:

create_view :posts_commented_by_staff,  Post.joins(comment: user).where(users: {role: 'staff'}).uniq
create_view :uncommented_posts,        "SELECT * FROM posts LEFT OUTER JOIN comments ON comments.post_id = posts.id WHERE comments.id IS NULL"

And can be dropped:

drop_view :posts_commented_by_staff
drop_view :uncommented_posts

ActiveRecord works with views the same as with ordinary tables. That is, for the above views you can define

class PostCommentedByStaff < ActiveRecord::Base
  table_name = "posts_commented_by_staff"
end

class UncommentedPost < ActiveRecord::Base
end

Note: In Postgres, all internal views (the ones with pg_ prefix) will be skipped.

Column Defaults: Expressions

SchemaPlus allows defaults to be set using expressions or constant values:

t.datetime :seen_at, default: { expr: 'NOW()' }
t.datetime :seen_at, default: { value: "2011-12-11 00:00:00" }

Note that in MySQL only the TIMESTAMP column data type accepts SQL column defaults and Rails uses DATETIME, so expressions can't be used with MySQL.

The standard syntax will still work as usual:

t.datetime :seen_at, default: "2011-12-11 00:00:00"

Also, as a convenience

t.datetime :seen_at, default: :now

resolves to:

NOW()                 # PostgreSQL
(DATETIME('now'))     # SQLite3
invalid               # MySQL

Column Defaults: Using

SchemaPlus introduces a constant ActiveRecord::DB_DEFAULT that you can use to explicitly instruct the database to use the column default value (or expression). For example:

Post.create(category: ActiveRecord::DB_DEFAULT)
post.update_attributes(category: ActiveRecord::DB_DEFAULT)

(Without ActiveRecord::DB_DEFAULT, you can update a value to NULL but not to its default value.)

Note that after updating, you would need to reload a record to replace ActiveRecord::DB_DEFAULT with the value assigned by the database.

Note also that Sqlite3 does not support ActiveRecord::DB_DEFAULT; attempting to use it will raise ActiveRecord::StatementInvalid

Schema Dump and Load (schema.rb)

When dumping schema.rb, SchemaPlus orders the views and tables in the schema dump alphabetically, but subject to the requirement that each table or view be defined before those that depend on it. This allows all foreign key constraints to be defined within the scope of the table definition. (Unless there are cyclical dependencies, in which case some foreign keys constraints must be defined later.)

Also, when dumping schema.rb, SchemaPlus dumps explicit foreign key definition statements rather than relying on the auto-creation behavior, for maximum clarity and for independence from global config. And correspondingly, when loading a schema, i.e. with the context of ActiveRecord::Schema.define, SchemaPlus ensures that auto creation of foreign key constraints is turned off regardless of the global setting. But if for some reason you are creating your schema.rb file by hand, and would like to take advantage of auto-creation of foreign key constraints, you can re-enable it:

ActiveRecord::Schema.define do
    SchemaPlus.config.foreign_keys.auto_create = true
    SchemaPlus.config.foreign_keys.auto_index = true

    create_table ...etc...
end

History

• See CHANGELOG for per-version release notes.

• SchemaPlus is derived from several "Red Hill On Rails" plugins originally created by @harukizaemon

• SchemaPlus was created in 2011 by @mlomnicki and @ronen

• And lots of contributors since then

Development & Testing

Are you interested in contributing to schema_plus? Thanks!

Schema_plus has a full set of rspec tests. travis-ci runs the tests on the full matrix of supported versions of ruby, rails, and db adapters. But you can also test all or some part of the matrix locally before you push your changes. Here's what you need to know:

Required environment:

• You must have either rbenv or rvm installed and working, whichever you prefer. Within it, have available whichever ruby versions you want to test. The default set is MRI 1.9.3, 2.0.0, 2.1.0, and jruby

• Of course you must have installed whichever databases you want to test. The default set is: PostgreSQL, MySQL, and SQLite3.

• For PostgreSQL and MySQL the tests need a db user with permissions to create and access databases: The default username used by the specs is 'postgres' for Postgresql and 'schema_plus' for MySQL; you can change them via:

    $ export POSTGRES_DB_USER = pgusername
    $ export MYSQL_DB_USER = mysqlusername
  

• For PostgreSQL and MySQL you must explicitly create the databases used by the tests:

    $ rake create_databases  # creates both postgresql & mysql
       OR
    $ rake postgresql:create_databases
    $ rake mysql:create_databases
  

Running the tests

The tests are run via a script in the repo root. Its args are documented by

 $ ./runspecs --help

By default it runs on a matrix of postgresql, mysql2, and sqlite3, for all rubies and all versions of rails. But the matrix options --db, --ruby, and --rails options let you limit those. The --quick option runs on just one set: postgresql, rails 4.1 and ruby 2.1.0. The --full option adds the mysql adapter to the set (in addition to mysql2 adapter).

• Install gem dependencies for the sets you'll be testing:

     $ ./runspecs [matrix options] --install   # runs 'bundle install' for all sets
     	e.g.
     $ ./runspecs --db 'posgresql' --ruby '2.1.0' --rails '4.0 4.1' --install
  

• Run all the tests:

     $ ./runspecs [matrix options]  # runs rspec for each set in the matrix
        e.g.
     $ ./runspecs --db 'posgresql' --ruby '2.1.0' --rails '4.0 4.1'
  

  Code coverage information will be in coverage/index.html -- it should be at 100% coverage if you're running against all databases.

• To run rspec on just a limited set of specs, you can do:

    $ ./runspecs [matrix options] --rspec -- [rspec args]
       e.g.
    $ ./runspecs --quick --rspec -- spec/migration_spec.rb -e 'default name'
  

Contributions to making the testing process itself easier and better will also be gratefully accepted!