Example:

    create_table :barcodes, primary_key: ["region", "code"] do |t|
      t.string :region
      t.integer :code
    end

Some test cases are testing only mysql adapter. We should test mysql2
adapter also.

`@connection` in `StatementPool` is only used for PG adapter.
No need for abstract `StatementPool` class.

`restrict_with_error` message will now respect owner’s human name in any locale [kuboon & Ronak Jangir]

The last call site of `last_version` was removed with:

838e1832

This reverts commit d5ba9a42, reversing
changes made to 30c50339.

Reason: This generate the dynalic finders more than one time

Basically view tests for MySQL are same with
`test/cases/adapters/postgresql/view_test.rb`.

So move `test/cases/adapters/postgresql/view_test.rb` to
`test/cases/view_test.rb` and make them only run if
`current_adapter` supports writable view.

Example:

```ruby
class Person
  include ActiveModel::Validations

  attr_reader :name, :title
  validates_presence_of :name, on: :create
  validates_presence_of :title, on: :update
end

person = Person.new
person.valid?([:create, :update])    # => true
person.errors.messages               # => {:name=>["can't be blank"], :title=>["can't be blank"]}
```

This reverts commit 51dd2588, reversing
changes made to ecb4e4b2.

This broke Active Record tests

`lengh` should be `length`  

This change allows to instantiate all ActiveRecordError descendant
execption classes without arguments, which might be useful in testing
and is far less surprising than mandatory arguments.

Apparently I managed to forget how similar the "tests passing" and
"no status reported" merge indicators look.

Note that the previous `stubs` in test_add_index wasn't working:
the method was still called, and just happened to return false.

See http://www.postgresql.org/docs/9.4/static/sql-dropindex.html
for more details.

Current implementation has a lot of utility methods that accept
reflection call a lot of methods on it and exit.

E.g. has_counter_cache?(reflection)

It causes confusion and inability to cache result of the method even
through it always returns the same result for the same reflection
object.
It can be done easier without access to the association context
by moving code into reflection itself.

e.g. reflection.has_counter_cache?

Reflection is less complex object than association so moving code there
automatically makes it simplier to understand.

[Rafael Mendonça França + Jean Boussier]

This is fix of #11826 which miss to add `disable_extension` to
`ReversibleAndIrreversibleMethods`. So `CommandRecorder#method_missing`
catches `change_column_default` and @delegate's method is called.

Closes #21418.

Previously schema names were not quoted. This leads to issues when a
schema names contains a ".". Methods in `schema_statements.rb` should
quote user input.

This fixes the following warning.

```ruby
warning: assigned but unused variable - index_definition
```

This is fix of #20018 which removes `change_column_default` from
array, so `CommandRecorder#method_missing` catches
`change_column_default` and @delegate's method is called.

This PR

* fix this bug
* define `ReversibleAndIrreversibleMethods` const making clear
  which this array means to prevent these miss

Since after 87d1aba3 `dependent: :destroy` callbacks on has_one
assocations run *after* destroy, it is possible that a nullification is
attempted on an already destroyed target:

    class Car < ActiveRecord::Base
      has_one :engine, dependent: :nullify
    end

    class Engine < ActiveRecord::Base
      belongs_to :car, dependent: :destroy
    end

    > car = Car.create!
    > engine = Engine.create!(car: car)
    > engine.destroy! # => ActiveRecord::ActiveRecordError: cannot update a
    >   destroyed record

In the above case, `engine.destroy!` deletes `engine` and *then* triggers the
deletion of `car`, which in turn triggers a nullification of `engine.car_id`.
However, `engine` is already destroyed at that point.

Fixes #21223.

Remove `MigrationTest#connection` and write `ActiveRecord::Base.connection`
directly to test, because `MigrationTest#connection` is only used in
`test_migration_instance_has_connection`.

Closes #21304.

While we can validate uniqueness for record without primary key on
creation, there is no way to exclude the current record when
updating. (The update itself will need a primary key to work correctly).

As of MySQL 5.7.8, MySQL supports a native JSON data type.

Example:

    create_table :json_data_type do |t|
      t.json :settings
    end

Closes #21201.

As described in the "Follow Coding Conventions" section in our
contribution guide (http://edgeguides.rubyonrails.org/contributing_to_ruby_on_rails.html#follow-the-coding-conventions)
we favor `assert_not` over `refute`.

While we don't usually make stylistic changes on it's own I opted to do
it in this case. The reason being that test cases are usually copied as
a starting point for new tests. This results in a spread of `refute` in
files that have been using it already.

If the through class has default scopes we should skip the statement
cache.

Closes #20745.