activerecord 4.2.9 → 6.1.4.1

data/CHANGELOG.md

@@ -1,2014 +1,1596 @@
-## Rails 4.2.9 (June 26, 2017) ##
+## Rails 6.1.4.1 (August 19, 2021) ##
 
-*   Fix regression caused by `collection_singular_ids=` ignoring different primary key on relationship.
-
-    *Nick Pezza*
-
-*   Fix `rake db:schema:load` with subdirectories.
-
-    *Ryuta Kamizono*
-
-*   Fix `rake db:migrate:status` with subdirectories.
-
-    *Ryuta Kamizono*
-
-*   Fix regression of #1969 with SELECT aliases in HAVING clause.
-
-    *Eugene Kenny*
-
-*   Fix `wait_timeout` to configurable for mysql2 adapter.
-
-    Fixes #26556.
-
-    *Ryuta Kamizono*
-
-*   Make `table_name=` reset current statement cache,
-    so queries are not run against the previous table name.
-
-    *namusyaka*
+*   No changes.
 
 
-## Rails 4.2.8 (February 21, 2017) ##
+## Rails 6.1.4 (June 24, 2021) ##
 
-*   Using a mysql2 connection after it fails to reconnect will now have an error message
-    saying the connection is closed rather than an undefined method error message.
+*   Do not try to rollback transactions that failed due to a `ActiveRecord::TransactionRollbackError`.
 
-    *Dylan Thacker-Smith*
+    *Jamie McCarthy*
 
-*   Bust Model.attribute_names cache when resetting column information
+*   Raise an error if `pool_config` is `nil` in `set_pool_config`.
 
-    *James Coleman*
+    *Eileen M. Uchitelle*
 
-*   Fix query caching when type information is reset
+*   Fix compatibility with `psych >= 4`.
 
-    Backports ancillary fix in 5.0.
+    Starting in Psych 4.0.0 `YAML.load` behaves like `YAML.safe_load`. To preserve compatibility
+    Active Record's schema cache loader and `YAMLColumn` now uses `YAML.unsafe_load` if available.
 
-    *James Coleman*
+    *Jean Boussier*
 
-*   Allow `joins` to be unscoped.
+*   Support using replicas when using `rails dbconsole`.
 
-    Fixes #13775.
+    *Christopher Thornton*
 
-    *Takashi Kokubun*
+*   Restore connection pools after transactional tests.
 
-*   Hashes can once again be passed to setters of `composed_of`, if all of the
-    mapping methods are methods implemented on `Hash`.
+    *Eugene Kenny*
 
-    Fixes #25978.
+*   Change `upsert_all` to fails cleanly for MySQL when `:unique_by` is used.
 
-    *Sean Griffin*
+    *Bastian Bartmann*
 
+*   Fix user-defined `self.default_scope` to respect table alias.
 
-## Rails 4.2.7 (July 12, 2016) ##
+    *Ryuta Kamizono*
 
-*   Inspecting an object with an associated array of over 10 elements no longer
-    truncates the array, preventing `inspect` from looping infinitely in some
-    cases.
+*   Clear `@cache_keys` cache after `update_all`, `delete_all`, `destroy_all`.
 
-    *Kevin McPhillips*
+    *Ryuta Kamizono*
 
-*   Ensure hashes can be assigned to attributes created using `composed_of`.
-    Fixes #25210.
+*   Changed Arel predications `contains` and `overlaps` to use
+    `quoted_node` so that PostgreSQL arrays are quoted properly.
 
-    *Sean Griffin*
+    *Bradley Priest*
 
-*   Queries such as `Computer.joins(:monitor).group(:status).count` will now be
-    interpreted as  `Computer.joins(:monitor).group('computers.status').count`
-    so that when `Computer` and `Monitor` have both `status` columns we don't
-    have conflicts in projection.
+*   Fix `merge` when the `where` clauses have string contents.
 
-    *Rafael Sales*
+    *Ryuta Kamizono*
 
-*   ActiveRecord::Relation#count: raise an ArgumentError when finder options
-    are specified or an ActiveRecord::StatementInvalid when an invalid type
-    is provided for a column name (e.g. a Hash).
+*   Fix rollback of parent destruction with nested `dependent: :destroy`.
 
-    Fixes #20434
+    *Jacopo Beschi*
 
-    *Konstantinos Rousis*
+*   Fix binds logging for `"WHERE ... IN ..."` statements.
 
-*   Correctly pass MySQL options when using structure_dump or structure_load
+    *Ricardo Díaz*
 
-    Specifically, it fixes an issue when using SSL authentication.
+*   Handle `false` in relation strict loading checks.
 
-    *Alex Coomans*
+    Previously when a model had strict loading set to true and then had a
+    relation set `strict_loading` to false the false wasn't considered when
+    deciding whether to raise/warn about strict loading.
 
+    ```
+    class Dog < ActiveRecord::Base
+      self.strict_loading_by_default = true
 
-## Rails 4.2.6 (March 07, 2016) ##
+      has_many :treats, strict_loading: false
+    end
+    ```
 
-*   Fix a bug where using `t.foreign_key` twice with the same `to_table` within
-    the same table definition would only create one foreign key.
+    In the example, `dog.treats` would still raise even though
+    `strict_loading` was set to false. This is a bug effecting more than
+    Active Storage which is why I made this PR superceeding #41461. We need
+    to fix this for all applications since the behavior is a little
+    surprising. I took the test from ##41461 and the code suggestion from #41453
+    with some additions.
 
-    *George Millo*
+    *Eileen M. Uchitelle*, *Radamés Roriz*
 
-*   Fix regression in dirty attribute tracking after #dup.  Changes to the
-    clone no longer show as changed attributes in the original object.
+*   Fix numericality validator without precision.
 
-    *Dominic Cleal*
+    *Ryuta Kamizono*
 
-*   Fix regression when loading fixture files with symbol keys.
+*   Fix aggregate attribute on Enum types.
 
-    Closes #22584.
+    *Ryuta Kamizono*
 
-    *Yves Senn*
+*   Fix `CREATE INDEX` statement generation for PostgreSQL.
 
-*   Fix `rake db:structure:dump` on Postgres when multiple schemas are used.
+    *eltongo*
 
-    Fixes #22346.
+*   Fix where clause on enum attribute when providing array of strings.
 
-    *Nick Muerdter*, *ckoenig*
+    *Ryuta Kamizono*
 
-*   Introduce `connection.data_sources` and `connection.data_source_exists?`.
-    These methods determine what relations can be used to back Active Record
-    models (usually tables and views).
+*   Fix `unprepared_statement` to work it when nesting.
 
-    *Yves Senn*, *Matthew Draper*
+    *Ryuta Kamizono*
 
 
-## Rails 4.2.5.2 (February 26, 2016) ##
+## Rails 6.1.3.2 (May 05, 2021) ##
 
 *   No changes.
 
 
-## Rails 4.2.5.1 (January 25, 2015) ##
+## Rails 6.1.3.1 (March 26, 2021) ##
 
 *   No changes.
 
 
-## Rails 4.2.5 (November 12, 2015) ##
-
-*   No longer pass deprecated option `-i` to `pg_dump`.
-
-    *Paul Sadauskas*
-
-*   Set `scope.reordering_value` to `true` if :reordering values are specified.
-
-    Fixes #21886.
-
-    *Hiroaki Izu*
-
-*   Avoid disabling errors on the PostgreSQL connection when enabling the
-    standard_conforming_strings setting. Errors were previously disabled because
-    the setting wasn't writable in Postgres 8.1 and didn't exist in earlier
-    versions. Now Rails only supports Postgres 8.2+ we're fine to assume the
-    setting exists. Disabling errors caused problems when using a connection
-    pooling tool like PgBouncer because it's not guaranteed to have the same
-    connection between calls to `execute` and it could leave the connection
-    with errors disabled.
-
-    Fixes #22101.
-
-    *Harry Marr*
-
-*   Includes HABTM returns correct size now. It's caused by the join dependency
-    only instantiates one HABTM object because the join table hasn't a primary key.
-
-    Fixes #16032.
-
-    Examples:
-
-        before:
-
-        Project.first.salaried_developers.size # => 3
-        Project.includes(:salaried_developers).first.salaried_developers.size # => 1
-
-        after:
-
-        Project.first.salaried_developers.size # => 3
-        Project.includes(:salaried_developers).first.salaried_developers.size # => 3
-
-    *Bigxiang*
-
-*   Descriptive error message when fixtures contain a missing column.
-
-    Closes #21201.
-
-    *Yves Senn*
+## Rails 6.1.3 (February 17, 2021) ##
 
-*   `bin/rake db:migrate` uses
-    `ActiveRecord::Tasks::DatabaseTasks.migrations_paths` instead of
-    `Migrator.migrations_paths`.
-
-    *Tobias Bielohlawek*
-
-*   Fix `rewhere` in a `has_many` association.
-
-    Fixes #21955.
-
-    *Josh Branchaud*, *Kal*
-
-*   Added run_cmd class method to ActiveRecord::Tasks::DatabaseTasks for
-    drying up Kernel.system() calls within this namespace and to avoid
-    shell expansion by using a paramter list instead of string as arguments
-    for Kernel.system(). Thanks to Nate Berkopec for supply patch to get
-    test units passing.
-
-    *Bryan Paxton*
-
-*   Avoid leaking the first relation we call `first` on, per model.
-
-    Fixes #21921.
-
-    *Matthew Draper*, *Jean Boussier*
-
-*   Allow deserialization of Active Record models that were YAML encoded prior
-    to Rails 4.2
-
-    *Sean Griffin*
-
-*   Correctly apply `unscope` when preloading through associations.
-
-    *Jimmy Bourassa*
-
-*   Ensure `select` quotes aliased attributes, even when using `from`.
-
-    Fixes #21488
-
-    *Sean Griffin & @johanlunds*
-
-*   Correct query for PostgreSQL 8.2 compatibility.
-
-    *Ben Murphy*, *Matthew Draper*
-
-*   Uniqueness validator raises descriptive error when running on a persisted
-    record without primary key.
-
-    Closes #21304.
-
-    *Yves Senn*
-
-
-## Rails 4.2.4 (August 24, 2015) ##
-
-*   Skip statement cache on through association reader.
-
-    If the through class has default scopes we should skip the statement
-    cache.
-
-    Closes #20745.
+*   Fix the MySQL adapter to always set the right collation and charset
+    to the connection session.
 
     *Rafael Mendonça França*
 
-*   Fixes #19420. When generating schema.rb using Postgres BigInt[] data type
-    the limit: 8 was not coming through. This caused it to become Int[] data type
-    after doing a rebuild off of schema.rb.
-
-    *Jake Waller*
-
-*   Fix state being carried over from previous transaction.
-
-    Considering the following example where `name` is a required attribute.
-    Before we had `new_record?` returning `true` for a persisted record:
-
-        author = Author.create! name: 'foo'
-        author.name = nil
-        author.save        # => false
-        author.new_record? # => true
-
-    Fixes #20824.
-
-    *Roque Pinel*
-
-*   Correctly ignore `mark_for_destruction` when `autosave` isn't set to `true`
-    when validating associations.
-
-    Fixes #20882.
-
-    *Sean Griffin*
-
-*   Fix through associations using scopes having the scope merged multiple
-    times.
-
-    Fixes #20721.
-    Fixes #20727.
-
-    *Sean Griffin*
-
-*   `ActiveRecord::Base.dump_schema_after_migration` applies migration tasks
-    other than `db:migrate`. (eg. `db:rollback`, `db:migrate:dup`, ...)
-
-    Fixes #20743.
-
-    *Yves Senn*
-
-*   Correctly raise `ActiveRecord::AssociationTypeMismatch` when assigning
-    a wrong type to a namespaced association.
-
-    Fixes #20545.
-
-    *Diego Carrion*
+*   Fix MySQL adapter handling of time objects when prepared statements
+    are enabled.
 
-*   Prevent error when using `force_reload: true` on an unassigned polymorphic
-    belongs_to association.
-
-    Fixes #20426.
-
-    *James Dabbs*
-
-
-## Rails 4.2.3 (June 25, 2015) ##
-
-*   Let `WITH` queries (Common Table Expressions) be explainable.
-
-    *Vladimir Kochnev*
-
-*   Fix n+1 query problem when eager loading nil associations (fixes #18312)
-
-    *Sammy Larbi*
-
-*   Fixed an error which would occur in dirty checking when calling
-    `update_attributes` from a getter.
-
-    Fixes #20531.
-
-    *Sean Griffin*
-
-*   Ensure symbols passed to `ActiveRecord::Relation#select` are always treated
-    as columns.
-
-    Fixes #20360.
-
-    *Sean Griffin*
-
-*   Clear query cache when `ActiveRecord::Base#reload` is called.
-
-    *Shane Hender*
-
-*   Pass `:extend` option for `has_and_belongs_to_many` associations to the
-    underlying `has_many :through`.
-
-    *Jaehyun Shin*
-
-*   Make `unscope` aware of "less than" and "greater than" conditions.
-
-    *TAKAHASHI Kazuaki*
-
-*   Revert behavior of `db:schema:load` back to loading the full
-    environment. This ensures that initializers are run.
-
-    Fixes #19545.
-
-    *Yves Senn*
+    *Rafael Mendonça França*
 
-*   Fix missing index when using `timestamps` with the `index` option.
+*   Fix scoping in enum fields using conditions that would generate
+    an `IN` clause.
 
-    The `index` option used with `timestamps` should be passed to both
-    `column` definitions for `created_at` and `updated_at` rather than just
-    the first.
+    *Ryuta Kamizono*
 
-    *Paul Mucur*
+*   Skip optimised #exist? query when #include? is called on a relation
+    with a having clause
 
-*   Rename `:class` to `:anonymous_class` in association options.
+    Relations that have aliased select values AND a having clause that
+    references an aliased select value would generate an error when
+    #include? was called, due to an optimisation that would generate
+    call #exists? on the relation instead, which effectively alters
+    the select values of the query (and thus removes the aliased select
+    values), but leaves the having clause intact. Because the having
+    clause is then referencing an aliased column that is no longer
+    present in the simplified query, an ActiveRecord::InvalidStatement
+    error was raised.
 
-    Fixes #19659.
+    An sample query affected by this problem:
 
-    *Andrew White*
+    ```ruby
+    Author.select('COUNT(*) as total_posts', 'authors.*')
+          .joins(:posts)
+          .group(:id)
+          .having('total_posts > 2')
+          .include?(Author.first)
+    ```
 
-*   Fixed a bug where uniqueness validations would error on out of range values,
-    even if an validation should have prevented it from hitting the database.
+    This change adds an addition check to the condition that skips the
+    simplified #exists? query, which simply checks for the presence of
+    a having clause.
 
-    *Andrey Voronkov*
+    Fixes #41417
 
-*   Foreign key related methods in the migration DSL respect
-    `ActiveRecord::Base.pluralize_table_names = false`.
+    *Michael Smart*
 
-    Fixes #19643.
+*   Increment postgres prepared statement counter before making a prepared statement, so if the statement is aborted
+    without Rails knowledge (e.g., if app gets kill -9d during long-running query or due to Rack::Timeout), app won't end
+    up in perpetual crash state for being inconsistent with Postgres.
 
-    *Mehmet Emin İNAÇ*
+    *wbharding*, *Martin Tepper*
 
-*   Reduce memory usage from loading types on pg.
 
-    Fixes #19578.
+## Rails 6.1.2.1 (February 10, 2021) ##
 
-    *Sean Griffin*
+*   Fix possible DoS vector in PostgreSQL money type
 
-*   Fix referencing wrong table aliases while joining tables of has many through
-    association (only when calling calculation methods).
+    Carefully crafted input can cause a DoS via the regular expressions used
+    for validating the money format in the PostgreSQL adapter.  This patch
+    fixes the regexp.
 
-    Fixes #19276.
+    Thanks to @dee-see from Hackerone for this patch!
 
-    *pinglamb*
+    [CVE-2021-22880]
 
-*   Don't attempt to update counter caches, when the column wasn't selected.
+    *Aaron Patterson*
 
-    Fixes #19437.
 
-    *Sean Griffin*
+## Rails 6.1.2 (February 09, 2021) ##
 
-*   Correctly persist a serialized attribute that has been returned to
-    its default value by an in-place modification.
+*   Fix timestamp type for sqlite3.
 
-    Fixes #19467.
+    *Eileen M. Uchitelle*
 
-    *Matthew Draper*
+*   Make destroy async transactional.
 
-*   Fix default `format` value in `ActiveRecord::Tasks::DatabaseTasks#schema_file`.
+    An active record rollback could occur while enqueuing a job. In this
+    case the job would enqueue even though the database deletion
+    rolledback putting things in a funky state.
 
-    *James Cox*
+    Now the jobs are only enqueued until after the db transaction has been committed.
 
-*   Dont enroll records in the transaction if they dont have commit callbacks.
-    That was causing a memory grow problem when creating a lot of records inside a transaction.
+    *Cory Gwin*
 
-    Fixes #15549.
+*   Fix malformed packet error in MySQL statement for connection configuration.
 
-    *Will Bryant*, *Aaron Patterson*
+    *robinroestenburg*
 
-*   Correctly create through records when created on a has many through
-    association when using `where`.
+*   Connection specification now passes the "url" key as a configuration for the
+    adapter if the "url" protocol is "jdbc", "http", or "https". Previously only
+    urls with the "jdbc" prefix were passed to the Active Record Adapter, others
+    are assumed to be adapter specification urls.
 
-    Fixes #19073.
+    Fixes #41137.
 
-    *Sean Griffin*
+    *Jonathan Bracy*
 
+*   Fix granular connection swapping when there are multiple abstract classes.
 
-## Rails 4.2.2 (June 16, 2015) ##
+    *Eileen M. Uchitelle*
 
-* No Changes *
+*   Fix `find_by` with custom primary key for belongs_to association.
 
+    *Ryuta Kamizono*
 
-## Rails 4.2.1 (March 19, 2015) ##
+*   Add support for `rails console --sandbox` for multiple database applications.
 
-*   Fixed ActiveRecord::Relation#becomes! and changed_attributes issues for type column
+    *alpaca-tc*
 
-    Fixes #17139.
+*   Fix `where` on polymorphic association with empty array.
 
-    *Miklos Fazekas*
+    *Ryuta Kamizono*
 
-*   `remove_reference` with `foreign_key: true` removes the foreign key before
-    removing the column. This fixes a bug where it was not possible to remove
-    the column on MySQL.
+*   Fix preventing writes for `ApplicationRecord`.
 
-    Fixes #18664.
+    *Eileen M. Uchitelle*
 
-    *Yves Senn*
 
-*   Add a `:foreign_key` option to `references` and associated migration
-    methods. The model and migration generators now use this option, rather than
-    the `add_foreign_key` form.
+## Rails 6.1.1 (January 07, 2021) ##
 
-    *Sean Griffin*
+*   Fix fixtures loading when strict loading is enabled for the association.
 
-*   Fix rounding problem for PostgreSQL timestamp column.
+    *Alex Ghiculescu*
 
-    If timestamp column have the precision, it need to format according to
-    the precision of timestamp column.
+*   Fix `where` with custom primary key for belongs_to association.
 
     *Ryuta Kamizono*
 
-*   Respect the database default charset for `schema_migrations` table.
-
-    The charset of `version` column in `schema_migrations` table is depend
-    on the database default charset and collation rather than the encoding
-    of the connection.
+*   Fix `where` with aliased associations.
 
     *Ryuta Kamizono*
 
-*   Respect custom primary keys for associations when calling `Relation#where`
-
-    Fixes #18813.
-
-    *Sean Griffin*
-
-*   Fixed several edge cases which could result in a counter cache updating
-    twice or not updating at all for `has_many` and `has_many :through`.
-
-    Fixes #10865.
-
-    *Sean Griffin*
-
-*   Foreign keys added by migrations were given random, generated names. This
-    meant a different `structure.sql` would be generated every time a developer
-    ran migrations on their machine.
-
-    The generated part of foreign key names is now a hash of the table name and
-    column name, which is consistent every time you run the migration.
-
-    *Chris Sinjakli*
-
-*   Fixed ActiveRecord::Relation#group method when argument is SQL reserved key word:
-
-      SplitTest.group(:key).count
-      Property.group(:value).count
-
-    *Bogdan Gusiev*
-
-*   Don't define autosave association callbacks twice from
-    `accepts_nested_attributes_for`.
-
-    Fixes #18704.
-
-    *Sean Griffin*
-
-*   Integer types will no longer raise a `RangeError` when assigning an
-    attribute, but will instead raise when going to the database.
-
-    Fixes several vague issues which were never reported directly. See the
-    commit message from the commit which added this line for some examples.
-
-    *Sean Griffin*
-
-*   Values which would error while being sent to the database (such as an
-    ASCII-8BIT string with invalid UTF-8 bytes on Sqlite3), no longer error on
-    assignment. They will still error when sent to the database, but you are
-    given the ability to re-assign it to a valid value.
-
-    Fixes #18580.
-
-    *Sean Griffin*
-
-*   Don't remove join dependencies in `Relation#exists?`
-
-    Fixes #18632.
-
-    *Sean Griffin*
-
-*   Invalid values assigned to a JSON column are assumed to be `nil`.
-
-    Fixes #18629.
-
-    *Sean Griffin*
-
-*   No longer issue deprecation warning when including a scope with extensions.
-    Previously every scope with extension methods was transformed into an
-    instance dependent scope. Including such a scope would wrongfully issue a
-    deprecation warning. This is no longer the case.
-
-    Fixes #18467.
-
-    *Yves Senn*
-
-*   Correctly use the type provided by `serialize` when updating records using
-    optimistic locking.
-
-    Fixes #18385.
-
-    *Sean Griffin*
-
-*   `attribute_will_change!` will no longer cause non-persistable attributes to
-    be sent to the database.
-
-    Fixes #18407.
-
-    *Sean Griffin*
-
-*   Format the datetime string according to the precision of the datetime field.
-
-    Incompatible to rounding behavior between MySQL 5.6 and earlier.
-
-    In 5.5, when you insert `2014-08-17 12:30:00.999999` the fractional part
-    is ignored. In 5.6, it's rounded to `2014-08-17 12:30:01`:
-
-    http://bugs.mysql.com/bug.php?id=68760
+*   Fix `composed_of` with symbol mapping.
 
     *Ryuta Kamizono*
 
-*   Allow precision option for MySQL datetimes.
+*   Don't skip money's type cast for pluck and calculations.
 
     *Ryuta Kamizono*
 
-*   Clear query cache on rollback.
-
-    *Florian Weingarten*
-
-*   Fixed setting of foreign_key for through associations while building of new record.
-
-    Fixes #12698.
-
-    *Ivan Antropov*
-
-*   Fixed automatic inverse_of for models nested in module.
-
-    *Andrew McCloud*
-
-*   Fix `reaping_frequency` option when the value is a string.
-
-    This usually happens when it is configured using `DATABASE_URL`.
-
-    *korbin*
-
-*   Fix error message when trying to create an associated record and the foreign
-    key is missing.
-
-    Before this fix the following exception was being raised:
-
-        NoMethodError: undefined method `val' for #<Arel::Nodes::BindParam:0x007fc64d19c218>
-
-    Now the message is:
-
-        ActiveRecord::UnknownAttributeError: unknown attribute 'foreign_key' for Model.
-
-    *Rafael Mendonça França*
-
-*   Fix change detection problem for PostgreSQL bytea type and
-    `ArgumentError: string contains null byte` exception with pg-0.18.
-
-    Fixes #17680.
-
-    *Lars Kanis*
-
-*   When a table has a composite primary key, the `primary_key` method for
-    SQLite3 and PostgreSQL adapters was only returning the first field of the key.
-    Ensures that it will return nil instead, as Active Record doesn't support
-    composite primary keys.
-
-    Fixes #18070.
-
-    *arthurnn*
-
-*   Ensure `first!` and friends work on loaded associations.
-
-    Fixes #18237.
-
-    *Sean Griffin*
-
-*   Dump the default `nil` for PostgreSQL UUID primary key.
+*   Fix `where` on polymorphic association with non Active Record object.
 
     *Ryuta Kamizono*
 
-*   Don't raise when writing an attribute with an out-of-range datetime passed
-    by the user.
-
-    *Grey Baker*
-
-*   Fixes bug with 'ActiveRecord::Type::Numeric' that causes negative values to
-    be marked as having changed when set to the same negative value.
-
-    Fixes #18161.
-
-    *Daniel Fox*
-
+*   Make sure `db:prepare` works even the schema file doesn't exist.
 
-## Rails 4.2.0 (December 20, 2014) ##
-
-*   Introduce `force: :cascade` option for `create_table`. Using this option
-    will recreate tables even if they have dependent objects (like foreign keys).
-    `db/schema.rb` now uses `force: :cascade`. This makes it possible to
-    reload the schema when foreign keys are in place.
-
-    *Matthew Draper*, *Yves Senn*
-
-*   `db:schema:load` and `db:structure:load` no longer purge the database
-    before loading the schema. This is left for the user to do.
-    `db:test:prepare` will still purge the database.
-
-    Fixes #17945.
-
-    *Yves Senn*
+    *Rafael Mendonça França*
 
-*   Fix undesirable RangeError by Type::Integer. Add Type::UnsignedInteger.
+*   Fix complicated `has_many :through` with nested where condition.
 
     *Ryuta Kamizono*
 
-*   Add `foreign_type` option to `has_one` and `has_many` association macros.
-
-    This option enables to define the column name of associated object's type for polymorphic associations.
-
-    *Ulisses Almeida, Kassio Borges*
-
-*   `add_timestamps` and `remove_timestamps` now properly reversible with
-    options.
-
-    *Noam Gagliardi-Rabinovich*
-
-*   Bring back `db:test:prepare` to synchronize the test database schema.
-
-    Manual synchronization using `bin/rake db:test:prepare` is required
-    when a migration is rolled-back, edited and reapplied.
-
-    `ActiveRecord::Base.maintain_test_schema` now uses `db:test:prepare`
-    to synchronize the schema. Plugins can use this task as a hook to
-    provide custom behavior after the schema has been loaded.
-
-    NOTE: `test:prepare` runs before the schema is synchronized.
-
-    Fixes #17171, #15787.
-
-    *Yves Senn*
-
-*   Change `reflections` public api to return the keys as String objects.
-
-    Fixes #16928.
-
-    *arthurnn*
-
-*   Renaming a table in pg also renames the primary key index.
-
-    Fixes #12856
-
-    *Sean Griffin*
-
-*   Make it possible to access fixtures excluded by a `default_scope`.
-
-    *Yves Senn*
-
-*   Fix preloading of associations with a scope containing joins along with
-    conditions on the joined association.
-
-    *Siddharth Sharma*
-
-*   Add `Table#name` to match `TableDefinition#name`.
-
-    *Cody Cutrer*
-
-*   Cache `CollectionAssociation#reader` proxies separately before and after
-    the owner has been saved so that the proxy is not cached without the
-    owner's id.
-
-    *Ben Woosley*
-
-*   `ActiveRecord::ReadOnlyRecord` now has a descriptive message.
-
-    *Franky W.*
-
-*   Fix preloading of associations which unscope a default scope.
+*   Handle STI models for `has_many dependent: :destroy_async`.
 
-    Fixes #11036.
+    *Muhammad Usman*
 
-    *Byron Bischoff*
+*   Restore possibility of passing `false` to :polymorphic option of `belongs_to`.
 
-*   Added SchemaDumper support for tables with jsonb columns.
+    Previously, passing `false` would trigger the option validation logic
+    to throw an error saying :polymorphic would not be a valid option.
 
-    *Ted O'Meara*
+    *glaszig*
 
-*   Deprecate `sanitize_sql_hash_for_conditions` without replacement. Using a
-    `Relation` for performing queries and updates is the prefered API.
+*   Allow adding nonnamed expression indexes to be revertible.
 
-    *Sean Griffin*
+    Fixes #40732.
 
-*   Queries now properly type cast values that are part of a join statement,
-    even when using type decorators such as `serialize`.
+    Previously, the following code would raise an error, when executed while rolling back,
+    and the index name should be specified explicitly. Now, the index name is inferred
+    automatically.
 
-    *Melanie Gilman & Sean Griffin*
+    ```ruby
+    add_index(:items, "to_tsvector('english', description)")
+    ```
 
-*   MySQL enum type lookups, with values matching another type, no longer result
-    in an endless loop.
+    *fatkodima*
 
-    Fixes #17402.
 
-    *Yves Senn*
+## Rails 6.1.0 (December 09, 2020) ##
 
-*   Raise `ArgumentError` when the body of a scope is not callable.
+*   Only warn about negative enums if a positive form that would cause conflicts exists.
 
-    *Mauro George*
+    Fixes #39065.
 
-*   Use type column first in multi-column indexes created with `add-reference`.
+    *Alex Ghiculescu*
 
-    *Derek Prior*
+*   Change `attribute_for_inspect` to take `filter_attributes` in consideration.
 
-*   Fix `Relation.rewhere` to work with Range values.
-
-    *Dan Olson*
-
-*   `AR::UnknownAttributeError` now includes the class name of a record.
-
-        User.new(name: "Yuki Nishijima", project_attributes: {name: "kaminari"})
-        # => ActiveRecord::UnknownAttributeError: unknown attribute 'name' for User.
-
-    *Yuki Nishijima*
-
-*   Fix a regression causing `after_create` callbacks to run before associated
-    records are autosaved.
-
-    Fixes #17209.
-
-    *Agis Anastasopoulos*
-
-*   Honor overridden `rack.test` in Rack environment for the connection
-    management middleware.
-
-    *Simon Eskildsen*
-
-*   Add a truncate method to the connection.
-
-    *Aaron Patterson*
-
-*   Don't autosave unchanged has_one through records.
-
-    *Alan Kennedy*, *Steve Parrington*
-
-*   Do not dump foreign keys for ignored tables.
-
-    *Yves Senn*
+    *Rafael Mendonça França*
 
-*   PostgreSQL adapter correctly dumps foreign keys targeting tables
-    outside the schema search path.
+*   Fix odd behavior of inverse_of with multiple belongs_to to same class.
 
-    Fixes #16907.
+    Fixes #35204.
 
-    *Matthew Draper*, *Yves Senn*
+    *Tomoyuki Kai*
 
-*   When a thread is killed, rollback the active transaction, instead of
-    committing it during the stack unwind. Previously, we could commit half-
-    completed work. This fix only works for Ruby 2.0+; on 1.9, we can't
-    distinguish a thread kill from an ordinary non-local (block) return, so must
-    default to committing.
+*   Build predicate conditions with objects that delegate `#id` and primary key:
 
-    *Chris Hanks*
+    ```ruby
+    class AdminAuthor
+      delegate_missing_to :@author
 
-*   A `NullRelation` should represent nothing. This fixes a bug where
-    `Comment.where(post_id: Post.none)` returned a non-empty result.
+      def initialize(author)
+        @author = author
+      end
+    end
 
-    Fixes #15176.
+    Post.where(author: AdminAuthor.new(author))
+    ```
 
-    *Matthew Draper*, *Yves Senn*
+    *Sean Doyle*
 
-*   Include default column limits in schema.rb. Allows defaults to be changed
-    in the future without affecting old migrations that assumed old defaults.
+*   Add `connected_to_many` API.
 
-    *Jeremy Kemper*
+    This API allows applications to connect to multiple databases at once without switching all of them or implementing a deeply nested stack.
 
-*   MySQL: schema.rb now includes TEXT and BLOB column limits.
+    Before:
 
-    *Jeremy Kemper*
+        AnimalsRecord.connected_to(role: :reading) do
+          MealsRecord.connected_to(role: :reading) do
+            Dog.first # read from animals replica
+            Dinner.first # read from meals replica
+            Person.first # read from primary writer
+          end
+        end
 
-*   MySQL: correct LONGTEXT and LONGBLOB limits from 2GB to their true 4GB.
+    After:
 
-    *Jeremy Kemper*
+        ActiveRecord::Base.connected_to_many([AnimalsRecord, MealsRecord], role: :reading) do
+          Dog.first # read from animals replica
+          Dinner.first # read from meals replica
+          Person.first # read from primary writer
+        end
 
-*   SQLite3Adapter now checks for views in `table_exists?`. Fixes #14041.
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-    *Girish Sonawane*
+*   Add option to raise or log for `ActiveRecord::StrictLoadingViolationError`.
 
-*   Introduce `connection.supports_views?` to check whether the current adapter
-    has support for SQL views. Connection adapters should define this method.
+    Some applications may not want to raise an error in production if using `strict_loading`. This would allow an application to set strict loading to log for the production environment while still raising in development and test environments.
 
-    *Yves Senn*
+    Set `config.active_record.action_on_strict_loading_violation` to `:log` errors instead of raising.
 
-*   Allow included modules to override association methods.
+    *Eileen M. Uchitelle*
 
-    Fixes #16684.
+*   Allow the inverse of a `has_one` association that was previously autosaved to be loaded.
 
-    *Yves Senn*
+    Fixes #34255.
 
-*   Schema loading rake tasks (like `db:schema:load` and `db:setup`) maintain
-    the database connection to the current environment.
+    *Steven Weber*
 
-    Fixes #16757.
+*   Optimise the length of index names for polymorphic references by using the reference name rather than the type and id column names.
 
-    *Joshua Cody*, *Yves Senn*
+    Because the default behaviour when adding an index with multiple columns is to use all column names in the index name, this could frequently lead to overly long index names for polymorphic references which would fail the migration if it exceeded the database limit.
 
-*   MySQL: set the connection collation along with the charset.
+    This change reduces the chance of that happening by using the reference name, e.g. `index_my_table_on_my_reference`.
 
-    Sets the connection collation to the database collation configured in
-    database.yml. Otherwise, `SET NAMES utf8mb4` will use the default
-    collation for that charset (utf8mb4_general_ci) when you may have chosen
-    a different collation, like utf8mb4_unicode_ci.
+    Fixes #38655.
 
-    This only applies to literal string comparisons, not column values, so it
-    is unlikely to affect you.
+    *Luke Redpath*
 
-    *Jeremy Kemper*
+*   MySQL: Uniqueness validator now respects default database collation,
+    no longer enforce case sensitive comparison by default.
 
-*   `default_sequence_name` from the PostgreSQL adapter returns a `String`.
+    *Ryuta Kamizono*
 
-    *Yves Senn*
+*   Remove deprecated methods from `ActiveRecord::ConnectionAdapters::DatabaseLimits`.
 
-*   Fix a regression where whitespaces were stripped from DISTINCT queries in
-    PostgreSQL.
+    `column_name_length`
+    `table_name_length`
+    `columns_per_table`
+    `indexes_per_table`
+    `columns_per_multicolumn_index`
+    `sql_query_length`
+    `joins_per_query`
 
-    *Agis Anastasopoulos*
+    *Rafael Mendonça França*
 
-    Fixes #16623.
+*   Remove deprecated `ActiveRecord::ConnectionAdapters::AbstractAdapter#supports_multi_insert?`.
 
-*   Fix has_many :through relation merging failing when dynamic conditions are
-    passed as a lambda with an arity of one.
+    *Rafael Mendonça França*
 
-    Fixes #16128.
+*   Remove deprecated `ActiveRecord::ConnectionAdapters::AbstractAdapter#supports_foreign_keys_in_create?`.
 
-    *Agis Anastasopoulos*
+    *Rafael Mendonça França*
 
-*   Fix `Relation#exists?` to work with polymorphic associations.
+*   Remove deprecated `ActiveRecord::ConnectionAdapters::PostgreSQLAdapter#supports_ranges?`.
 
-    Fixes #15821.
+    *Rafael Mendonça França*
 
-    *Kassio Borges*
+*   Remove deprecated `ActiveRecord::Base#update_attributes` and `ActiveRecord::Base#update_attributes!`.
 
-*   Currently, Active Record rescues any errors raised within
-    `after_rollback`/`after_create` callbacks and prints them to the logs.
-    Future versions of Rails will not rescue these errors anymore and
-    just bubble them up like the other callbacks.
+    *Rafael Mendonça França*
 
-    This commit adds an opt-in flag to enable not rescuing the errors.
+*   Remove deprecated `migrations_path` argument in `ActiveRecord::ConnectionAdapter::SchemaStatements#assume_migrated_upto_version`.
 
-    Example:
+    *Rafael Mendonça França*
 
-        # Do not swallow errors in after_commit/after_rollback callbacks.
-        config.active_record.raise_in_transactional_callbacks = true
+*   Remove deprecated `config.active_record.sqlite3.represent_boolean_as_integer`.
 
-    Fixes #13460.
+    *Rafael Mendonça França*
 
-    *arthurnn*
+*   `relation.create` does no longer leak scope to class level querying methods
+    in initialization block and callbacks.
 
-*   Fix an issue where custom accessor methods (such as those generated by
-    `enum`) with the same name as a global method are incorrectly overridden
-    when subclassing.
+    Before:
 
-    Fixes #16288.
+        User.where(name: "John").create do |john|
+          User.find_by(name: "David") # => nil
+        end
 
-    *Godfrey Chan*
+    After:
 
-*   `*_was` and `changes` now work correctly for in-place attribute changes as
-    well.
+        User.where(name: "John").create do |john|
+          User.find_by(name: "David") # => #<User name: "David", ...>
+        end
 
-    *Sean Griffin*
+    *Ryuta Kamizono*
 
-*   Fix regression on `after_commit` that did not fire with nested transactions.
+*   Named scope chain does no longer leak scope to class level querying methods.
 
-    Fixes #16425.
+        class User < ActiveRecord::Base
+          scope :david, -> { User.where(name: "David") }
+        end
 
-    *arthurnn*
+    Before:
 
-*   Do not try to write timestamps when a table has no timestamps columns.
+        User.where(name: "John").david
+        # SELECT * FROM users WHERE name = 'John' AND name = 'David'
 
-    Fixes #8813.
+    After:
 
-    *Sergey Potapov*
+        User.where(name: "John").david
+        # SELECT * FROM users WHERE name = 'David'
 
-*   `index_exists?` with `:name` option does verify specified columns.
+    *Ryuta Kamizono*
 
-    Example:
+*   Remove deprecated methods from `ActiveRecord::DatabaseConfigurations`.
 
-        add_index :articles, :title, name: "idx_title"
+    `fetch`
+    `each`
+    `first`
+    `values`
+    `[]=`
 
-        # Before:
-        index_exists? :articles, :title, name: "idx_title" # => `true`
-        index_exists? :articles, :body, name: "idx_title" # => `true`
+    *Rafael Mendonça França*
 
-        # After:
-        index_exists? :articles, :title, name: "idx_title" # => `true`
-        index_exists? :articles, :body, name: "idx_title" # => `false`
+*   `where.not` now generates NAND predicates instead of NOR.
 
-    *Yves Senn*, *Matthew Draper*
+     Before:
 
-*   `add_timestamps` and `t.timestamps` now require you to pass the `:null` option.
-    Not passing the option is deprecated but the default is still `null: true`.
-    With Rails 5 this will change to `null: false`.
+        User.where.not(name: "Jon", role: "admin")
+        # SELECT * FROM users WHERE name != 'Jon' AND role != 'admin'
 
-    *Sean Griffin*
+     After:
 
-*   When calling `update_columns` on a record that is not persisted, the error
-    message now reflects whether that object is a new record or has been
-    destroyed.
+        User.where.not(name: "Jon", role: "admin")
+        # SELECT * FROM users WHERE NOT (name == 'Jon' AND role == 'admin')
 
-    *Lachlan Sylvester*
+    *Rafael Mendonça França*
 
-*   Define `id_was` to get the previous value of the primary key.
+*   Remove deprecated `ActiveRecord::Result#to_hash` method.
 
-    Currently when we call `id_was` and we have a custom primary key name,
-    Active Record will return the current value of the primary key. This
-    makes it impossible to correctly do an update operation if you change the
-    id.
+    *Rafael Mendonça França*
 
-    Fixes #16413.
+*   Deprecate `ActiveRecord::Base.allow_unsafe_raw_sql`.
 
     *Rafael Mendonça França*
 
-*   Deprecate `DatabaseTasks.load_schema` to act on the current connection.
-    Use `.load_schema_current` instead. In the future `load_schema` will
-    require the `configuration` to act on as an argument.
+*   Remove deprecated support for using unsafe raw SQL in `ActiveRecord::Relation` methods.
 
-    *Yves Senn*
-
-*   Fix automatic maintaining test schema to properly handle sql structure
-    schema format.
+    *Rafael Mendonça França*
 
-    Fixes #15394.
+*   Allow users to silence the "Rails couldn't infer whether you are using multiple databases..."
+    message using `config.active_record.suppress_multiple_database_warning`.
 
-    *Wojciech Wnętrzak*
+    *Omri Gabay*
 
-*   Fix type casting to Decimal from Float with large precision.
+*   Connections can be granularly switched for abstract classes when `connected_to` is called.
 
-    *Tomohiro Hashidate*
+    This change allows `connected_to` to switch a `role` and/or `shard` for a single abstract class instead of all classes globally. Applications that want to use the new feature need to set `config.active_record.legacy_connection_handling` to `false` in their application configuration.
 
-*   Deprecate `Reflection#source_macro`
+    Example usage:
 
-    `Reflection#source_macro` is no longer needed in Active Record
-    source so it has been deprecated. Code that used `source_macro`
-    was removed in #16353.
+    Given an application we have a `User` model that inherits from `ApplicationRecord` and a `Dog` model that inherits from `AnimalsRecord`. `AnimalsRecord` and `ApplicationRecord` have writing and reading connections as well as shard `default`, `one`, and `two`.
 
-    *Eileen M. Uchtitelle*, *Aaron Patterson*
+    ```ruby
+    ActiveRecord::Base.connected_to(role: :reading) do
+      User.first # reads from default replica
+      Dog.first # reads from default replica
 
-*   No verbose backtrace by `db:drop` when database does not exist.
+      AnimalsRecord.connected_to(role: :writing, shard: :one) do
+        User.first # reads from default replica
+        Dog.first # reads from shard one primary
+      end
 
-    Fixes #16295.
+      User.first # reads from default replica
+      Dog.first # reads from default replica
 
-    *Kenn Ejima*
+      ApplicationRecord.connected_to(role: :writing, shard: :two) do
+        User.first # reads from shard two primary
+        Dog.first # reads from default replica
+      end
+    end
+    ```
 
-*   Add support for PostgreSQL JSONB.
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-    Example:
+*   Allow double-dash comment syntax when querying read-only databases
 
-        create_table :posts do |t|
-          t.jsonb :meta_data
-        end
+    *James Adam*
 
-    *Philippe Creux*, *Chris Teague*
+*   Add `values_at` method.
 
-*   `db:purge` with MySQL respects `Rails.env`.
+    Returns an array containing the values associated with the given methods.
 
-    *Yves Senn*
+    ```ruby
+    topic = Topic.first
+    topic.values_at(:title, :author_name)
+    # => ["Budget", "Jason"]
+    ```
 
-*   `change_column_default :table, :column, nil` with PostgreSQL will issue a
-    `DROP DEFAULT` instead of a `DEFAULT NULL` query.
+    Similar to `Hash#values_at` but on an Active Record instance.
 
-    Fixes #16261.
+    *Guillaume Briday*
 
-    *Matthew Draper*, *Yves Senn*
+*   Fix `read_attribute_before_type_cast` to consider attribute aliases.
 
-*   Allow to specify a type for the foreign key column in `references`
-    and `add_reference`.
+    *Marcelo Lauxen*
 
-    Example:
+*   Support passing record to uniqueness validator `:conditions` callable:
 
-        change_table :vehicle do |t|
-          t.references :station, type: :uuid
-        end
+    ```ruby
+    class Article < ApplicationRecord
+      validates_uniqueness_of :title, conditions: ->(article) {
+        published_at = article.published_at
+        where(published_at: published_at.beginning_of_year..published_at.end_of_year)
+      }
+    end
+    ```
 
-    *Andrey Novikov*, *Łukasz Sarnacki*
+    *Eliot Sykes*
 
-*   `create_join_table` removes a common prefix when generating the join table.
-    This matches the existing behavior of HABTM associations.
+*   `BatchEnumerator#update_all` and `BatchEnumerator#delete_all` now return the
+    total number of rows affected, just like their non-batched counterparts.
 
-    Fixes #13683.
+    ```ruby
+    Person.in_batches.update_all("first_name = 'Eugene'") # => 42
+    Person.in_batches.delete_all # => 42
+    ```
 
-    *Stefan Kanev*
+    Fixes #40287.
 
-*   Do not swallow errors on `compute_type` when having a bad `alias_method` on
-    a class.
+    *Eugene Kenny*
 
-    *arthurnn*
+*   Add support for PostgreSQL `interval` data type with conversion to
+    `ActiveSupport::Duration` when loading records from database and
+    serialization to ISO 8601 formatted duration string on save.
+    Add support to define a column in migrations and get it in a schema dump.
+    Optional column precision is supported.
 
-*   PostgreSQL invalid `uuid` are convert to nil.
+    To use this in 6.1, you need to place the next string to your model file:
 
-    *Abdelkader Boudih*
+        attribute :duration, :interval
 
-*   Restore 4.0 behavior for using serialize attributes with `JSON` as coder.
+    To keep old behavior until 6.2 is released:
 
-    With 4.1.x, `serialize` started returning a string when `JSON` was passed as
-    the second attribute. It will now return a hash as per previous versions.
+        attribute :duration, :string
 
     Example:
 
-        class Post < ActiveRecord::Base
-          serialize :comment, JSON
+        create_table :events do |t|
+          t.string   :name
+          t.interval :duration
         end
 
-        class Comment
-          include ActiveModel::Model
-          attr_accessor :category, :text
+        class Event < ApplicationRecord
+          attribute :duration, :interval
         end
 
-        post = Post.create!
-        post.comment = Comment.new(category: "Animals", text: "This is a comment about squirrels.")
-        post.save!
-
-        # 4.0
-        post.comment # => {"category"=>"Animals", "text"=>"This is a comment about squirrels."}
-
-        # 4.1 before
-        post.comment # => "#<Comment:0x007f80ab48ff98>"
-
-        # 4.1 after
-        post.comment # => {"category"=>"Animals", "text"=>"This is a comment about squirrels."}
-
-    When using `JSON` as the coder in `serialize`, Active Record will use the
-    new `ActiveRecord::Coders::JSON` coder which delegates its `dump/load` to
-    `ActiveSupport::JSON.encode/decode`. This ensures special objects are dumped
-    correctly using the `#as_json` hook.
+        Event.create!(name: 'Rock Fest', duration: 2.days)
+        Event.last.duration # => 2 days
+        Event.last.duration.iso8601 # => "P2D"
+        Event.new(duration: 'P1DT12H3S').duration # => 1 day, 12 hours, and 3 seconds
+        Event.new(duration: '1 day') # Unknown value will be ignored and NULL will be written to database
 
-    To keep the previous behaviour, supply a custom coder instead
-    ([example](https://gist.github.com/jenncoop/8c4142bbe59da77daa63)).
-
-    Fixes #15594.
-
-    *Jenn Cooper*
-
-*   Do not use `RENAME INDEX` syntax for MariaDB 10.0.
-
-    Fixes #15931.
-
-    *Jeff Browning*
-
-*   Calling `#empty?` on a `has_many` association would use the value from the
-    counter cache if one exists.
+    *Andrey Novikov*
 
-    *David Verhasselt*
+*   Allow associations supporting the `dependent:` key to take `dependent: :destroy_async`.
 
-*   Fix the schema dump generated for tables without constraints and with
-    primary key with default value of custom PostgreSQL function result.
+    ```ruby
+    class Account < ActiveRecord::Base
+        belongs_to :supplier, dependent: :destroy_async
+    end
+    ```
 
-    Fixes #16111.
+    `:destroy_async` will enqueue a job to destroy associated records in the background.
 
-    *Andrey Novikov*
+    *DHH*, *George Claghorn*, *Cory Gwin*, *Rafael Mendonça França*, *Adrianna Chang*
 
-*   Fix the SQL generated when a `delete_all` is run on an association to not
-    produce an `IN` statements.
+*   Add `SKIP_TEST_DATABASE` environment variable to disable modifying the test database when `rails db:create` and `rails db:drop` are called.
 
-    Before:
+    *Jason Schweier*
 
-      UPDATE "categorizations" SET "category_id" = NULL WHERE
-      "categorizations"."category_id" = 1 AND "categorizations"."id" IN (1, 2)
-
-    After:
+*   `connects_to` can only be called on `ActiveRecord::Base` or abstract classes.
 
-      UPDATE "categorizations" SET "category_id" = NULL WHERE
-      "categorizations"."category_id" = 1
+    Ensure that `connects_to` can only be called from `ActiveRecord::Base` or abstract classes. This protects the application from opening duplicate or too many connections.
 
-    *Eileen M. Uchitelle, Aaron Patterson*
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-*   Avoid type casting boolean and `ActiveSupport::Duration` values to numeric
-    values for string columns. Otherwise, in some database, the string column
-    values will be coerced to a numeric allowing false or 0.seconds match any
-    string starting with a non-digit.
+*   All connection adapters `execute` now raises `ActiveRecord::ConnectionNotEstablished` rather than
+    `ActiveRecord::StatementInvalid` when they encounter a connection error.
 
-    Example:
+    *Jean Boussier*
 
-        App.where(apikey: false) # => SELECT * FROM users WHERE apikey = '0'
+*   `Mysql2Adapter#quote_string` now raises `ActiveRecord::ConnectionNotEstablished` rather than
+    `ActiveRecord::StatementInvalid` when it can't connect to the MySQL server.
 
-    *Dylan Thacker-Smith*
+    *Jean Boussier*
 
-*   Add a `:required` option to singular associations, providing a nicer
-    API for presence validations on associations.
+*   Add support for check constraints that are `NOT VALID` via `validate: false` (PostgreSQL-only).
 
-    *Sean Griffin*
+    *Alex Robbin*
 
-*   Fix an error in `reset_counters` when associations have `select` scope.
-    (Call to `count` generated invalid SQL.)
+*   Ensure the default configuration is considered primary or first for an environment
 
-    *Cade Truitt*
+    If a multiple database application provides a configuration named primary, that will be treated as default. In applications that do not have a primary entry, the default database configuration will be the first configuration for an environment.
 
-*   After a successful `reload`, `new_record?` is always false.
+    *Eileen M. Uchitelle*
 
-    Fixes #12101.
+*   Allow `where` references association names as joined table name aliases.
 
-    *Matthew Draper*
+    ```ruby
+    class Comment < ActiveRecord::Base
+      enum label: [:default, :child]
+      has_many :children, class_name: "Comment", foreign_key: :parent_id
+    end
 
-*   PostgreSQL renaming table doesn't attempt to rename non existent sequences.
+    # ... FROM comments LEFT OUTER JOIN comments children ON ... WHERE children.label = 1
+    Comment.includes(:children).where("children.label": "child")
+    ```
 
-    *Abdelkader Boudih*
+    *Ryuta Kamizono*
 
-*   Move 'dependent: :destroy' handling for `belongs_to`
-    from `before_destroy` to `after_destroy` callback chain
+*   Support storing demodulized class name for polymorphic type.
 
-    Fixes #12380.
+    Before Rails 6.1, storing demodulized class name is supported only for STI type
+    by `store_full_sti_class` class attribute.
 
-    *Ivan Antropov*
+    Now `store_full_class_name` class attribute can handle both STI and polymorphic types.
 
-*   Detect in-place modifications on String attributes.
+    *Ryuta Kamizono*
 
-    Before this change, an attribute modified in-place had to be marked as
-    changed in order for it to be persisted in the database. Now it is no longer
-    required.
+*   Deprecate `rails db:structure:{load, dump}` tasks and extend
+    `rails db:schema:{load, dump}` tasks to work with either `:ruby` or `:sql` format,
+    depending on `config.active_record.schema_format` configuration value.
 
-    Before:
+    *fatkodima*
 
-        user = User.first
-        user.name << ' Griffin'
-        user.name_will_change!
-        user.save
-        user.reload.name # => "Sean Griffin"
+*   Respect the `select` values for eager loading.
 
-    After:
+    ```ruby
+    post = Post.select("UPPER(title) AS title").first
+    post.title # => "WELCOME TO THE WEBLOG"
+    post.body  # => ActiveModel::MissingAttributeError
 
-        user = User.first
-        user.name << ' Griffin'
-        user.save
-        user.reload.name # => "Sean Griffin"
+    # Rails 6.0 (ignore the `select` values)
+    post = Post.select("UPPER(title) AS title").eager_load(:comments).first
+    post.title # => "Welcome to the weblog"
+    post.body  # => "Such a lovely day"
 
-    *Sean Griffin*
+    # Rails 6.1 (respect the `select` values)
+    post = Post.select("UPPER(title) AS title").eager_load(:comments).first
+    post.title # => "WELCOME TO THE WEBLOG"
+    post.body  # => ActiveModel::MissingAttributeError
+    ```
 
-*   Add `ActiveRecord::Base#validate!` that raises `RecordInvalid` if the record
-    is invalid.
+    *Ryuta Kamizono*
 
-    *Bogdan Gusiev*, *Marc Schütz*
+*   Allow attribute's default to be configured but keeping its own type.
 
-*   Support for adding and removing foreign keys. Foreign keys are now
-    a part of `schema.rb`. This is supported by Mysql2Adapter, MysqlAdapter
-    and PostgreSQLAdapter.
+    ```ruby
+    class Post < ActiveRecord::Base
+      attribute :written_at, default: -> { Time.now.utc }
+    end
 
-    Many thanks to *Matthew Higgins* for laying the foundation with his work on
-    [foreigner](https://github.com/matthuhiggins/foreigner).
+    # Rails 6.0
+    Post.type_for_attribute(:written_at) # => #<Type::Value ... precision: nil, ...>
 
-    Example:
+    # Rails 6.1
+    Post.type_for_attribute(:written_at) # => #<Type::DateTime ... precision: 6, ...>
+    ```
 
-        # within your migrations:
-        add_foreign_key :articles, :authors
-        remove_foreign_key :articles, :authors
+    *Ryuta Kamizono*
 
-    *Yves Senn*
+*   Allow default to be configured for Enum.
 
-*   Fix subtle bugs regarding attribute assignment on models with no primary
-    key. `'id'` will no longer be part of the attributes hash.
+    ```ruby
+    class Book < ActiveRecord::Base
+      enum status: [:proposed, :written, :published], _default: :published
+    end
 
-    *Sean Griffin*
+    Book.new.status # => "published"
+    ```
 
-*   Deprecate automatic counter caches on `has_many :through`. The behavior was
-    broken and inconsistent.
+    *Ryuta Kamizono*
 
-    *Sean Griffin*
+*   Deprecate YAML loading from legacy format older than Rails 5.0.
 
-*   `preload` preserves readonly flag for associations.
+    *Ryuta Kamizono*
 
-    See #15853.
+*   Added the setting `ActiveRecord::Base.immutable_strings_by_default`, which
+    allows you to specify that all string columns should be frozen unless
+    otherwise specified. This will reduce memory pressure for applications which
+    do not generally mutate string properties of Active Record objects.
 
-    *Yves Senn*
+    *Sean Griffin*, *Ryuta Kamizono*
 
-*   Assume numeric types have changed if they were assigned to a value that
-    would fail numericality validation, regardless of the old value. Previously
-    this would only occur if the old value was 0.
+*   Deprecate `map!` and `collect!` on `ActiveRecord::Result`.
 
-    Example:
+    *Ryuta Kamizono*
 
-        model = Model.create!(number: 5)
-        model.number = '5wibble'
-        model.number_changed? # => true
+*   Support `relation.and` for intersection as Set theory.
 
-    Fixes #14731.
+    ```ruby
+    david_and_mary = Author.where(id: [david, mary])
+    mary_and_bob   = Author.where(id: [mary, bob])
 
-    *Sean Griffin*
+    david_and_mary.merge(mary_and_bob) # => [mary, bob]
 
-*   `reload` no longer merges with the existing attributes.
-    The attribute hash is fully replaced. The record is put into the same state
-    as it would be with `Model.find(model.id)`.
+    david_and_mary.and(mary_and_bob) # => [mary]
+    david_and_mary.or(mary_and_bob)  # => [david, mary, bob]
+    ```
 
-    *Sean Griffin*
+    *Ryuta Kamizono*
 
-*   The object returned from `select_all` must respond to `column_types`.
-    If this is not the case a `NoMethodError` is raised.
+*   Merging conditions on the same column no longer maintain both conditions,
+    and will be consistently replaced by the latter condition in Rails 6.2.
+    To migrate to Rails 6.2's behavior, use `relation.merge(other, rewhere: true)`.
 
-    *Sean Griffin*
+    ```ruby
+    # Rails 6.1 (IN clause is replaced by merger side equality condition)
+    Author.where(id: [david.id, mary.id]).merge(Author.where(id: bob)) # => [bob]
 
-*   Detect in-place modifications of PG array types
+    # Rails 6.1 (both conflict conditions exists, deprecated)
+    Author.where(id: david.id..mary.id).merge(Author.where(id: bob)) # => []
 
-    *Sean Griffin*
+    # Rails 6.1 with rewhere to migrate to Rails 6.2's behavior
+    Author.where(id: david.id..mary.id).merge(Author.where(id: bob), rewhere: true) # => [bob]
 
-*   Add `bin/rake db:purge` task to empty the current database.
+    # Rails 6.2 (same behavior with IN clause, mergee side condition is consistently replaced)
+    Author.where(id: [david.id, mary.id]).merge(Author.where(id: bob)) # => [bob]
+    Author.where(id: david.id..mary.id).merge(Author.where(id: bob)) # => [bob]
+    ```
 
-    *Yves Senn*
+    *Ryuta Kamizono*
 
-*   Deprecate `serialized_attributes` without replacement.
+*   Do not mark Postgresql MAC address and UUID attributes as changed when the assigned value only varies by case.
 
-    *Sean Griffin*
+    *Peter Fry*
 
-*   Correctly extract IPv6 addresses from `DATABASE_URI`: the square brackets
-    are part of the URI structure, not the actual host.
+*   Resolve issue with insert_all unique_by option when used with expression index.
 
-    Fixes #15705.
+    When the `:unique_by` option of `ActiveRecord::Persistence.insert_all` and
+    `ActiveRecord::Persistence.upsert_all` was used with the name of an expression index, an error
+    was raised. Adding a guard around the formatting behavior for the `:unique_by` corrects this.
 
-    *Andy Bakun*, *Aaron Stone*
+    Usage:
 
-*   Ensure both parent IDs are set on join records when both sides of a
-    through association are new.
+    ```ruby
+    create_table :books, id: :integer, force: true do |t|
+      t.column :name, :string
+      t.index "lower(name)", unique: true
+    end
 
-    *Sean Griffin*
+    Book.insert_all [{ name: "MyTest" }], unique_by: :index_books_on_lower_name
+    ```
 
-*   `ActiveRecord::Dirty` now detects in-place changes to mutable values.
-    Serialized attributes on ActiveRecord models will no longer save when
-    unchanged.
+    Fixes #39516.
 
-    Fixes #8328.
+    *Austen Madden*
 
-    *Sean Griffin*
+*   Add basic support for CHECK constraints to database migrations.
 
-*   `Pluck` now works when selecting columns from different tables with the same
-    name.
+    Usage:
 
-    Fixes #15649.
+    ```ruby
+    add_check_constraint :products, "price > 0", name: "price_check"
+    remove_check_constraint :products, name: "price_check"
+    ```
 
-    *Sean Griffin*
+    *fatkodima*
 
-*   Remove `cache_attributes` and friends. All attributes are cached.
+*   Add `ActiveRecord::Base.strict_loading_by_default` and `ActiveRecord::Base.strict_loading_by_default=`
+    to enable/disable strict_loading mode by default for a model. The configuration's value is
+    inheritable by subclasses, but they can override that value and it will not impact parent class.
 
-    *Sean Griffin*
+    Usage:
 
-*   Remove deprecated method `ActiveRecord::Base.quoted_locking_column`.
+    ```ruby
+    class Developer < ApplicationRecord
+      self.strict_loading_by_default = true
 
-    *Akshay Vishnoi*
+      has_many :projects
+    end
 
-*   `ActiveRecord::FinderMethods.find` with block can handle proc parameter as
-    `Enumerable#find` does.
+    dev = Developer.first
+    dev.projects.first
+    # => ActiveRecord::StrictLoadingViolationError Exception: Developer is marked as strict_loading and Project cannot be lazily loaded.
+    ```
 
-    Fixes #15382.
+    *bogdanvlviv*
 
-    *James Yang*
+*   Deprecate passing an Active Record object to `quote`/`type_cast` directly.
 
-*   Make timezone aware attributes work with PostgreSQL array columns.
+    *Ryuta Kamizono*
 
-    Fixes #13402.
+*   Default engine `ENGINE=InnoDB` is no longer dumped to make schema more agnostic.
 
-    *Kuldeep Aggarwal*, *Sean Griffin*
+    Before:
 
-*   `ActiveRecord::SchemaMigration` has no primary key regardless of the
-    `primary_key_prefix_type` configuration.
+    ```ruby
+    create_table "accounts", options: "ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_0900_ai_ci", force: :cascade do |t|
+    end
+    ```
 
-    Fixes #15051.
+    After:
 
-    *JoseLuis Torres*, *Yves Senn*
+    ```ruby
+    create_table "accounts", charset: "utf8mb4", collation: "utf8mb4_0900_ai_ci", force: :cascade do |t|
+    end
+    ```
 
-*   `rake db:migrate:status` works with legacy migration numbers like `00018_xyz.rb`.
+    *Ryuta Kamizono*
 
-    Fixes #15538.
+*   Added delegated type as an alternative to single-table inheritance for representing class hierarchies.
+    See ActiveRecord::DelegatedType for the full description.
 
-    *Yves Senn*
+    *DHH*
 
-*   Baseclass becomes! subclass.
+*   Deprecate aggregations with group by duplicated fields.
 
-    Before this change, a record which changed its STI type, could not be
-    updated.
+    To migrate to Rails 6.2's behavior, use `uniq!(:group)` to deduplicate group fields.
 
-    Fixes #14785.
+    ```ruby
+    accounts = Account.group(:firm_id)
 
-    *Matthew Draper*, *Earl St Sauver*, *Edo Balvers*
+    # duplicated group fields, deprecated.
+    accounts.merge(accounts.where.not(credit_limit: nil)).sum(:credit_limit)
+    # => {
+    #   [1, 1] => 50,
+    #   [2, 2] => 60
+    # }
 
-*   Remove deprecated `ActiveRecord::Migrator.proper_table_name`. Use the
-    `proper_table_name` instance method on `ActiveRecord::Migration` instead.
+    # use `uniq!(:group)` to deduplicate group fields.
+    accounts.merge(accounts.where.not(credit_limit: nil)).uniq!(:group).sum(:credit_limit)
+    # => {
+    #   1 => 50,
+    #   2 => 60
+    # }
+    ```
 
-    *Akshay Vishnoi*
+    *Ryuta Kamizono*
 
-*   Fix regression on eager loading association based on SQL query rather than
-    existing column.
+*   Deprecate duplicated query annotations.
 
-    Fixes #15480.
+    To migrate to Rails 6.2's behavior, use `uniq!(:annotate)` to deduplicate query annotations.
 
-    *Lauro Caetano*, *Carlos Antonio da Silva*
+    ```ruby
+    accounts = Account.where(id: [1, 2]).annotate("david and mary")
 
-*   Deprecate returning `nil` from `column_for_attribute` when no column exists.
-    It will return a null object in Rails 5.0
+    # duplicated annotations, deprecated.
+    accounts.merge(accounts.rewhere(id: 3))
+    # SELECT accounts.* FROM accounts WHERE accounts.id = 3 /* david and mary */ /* david and mary */
 
-    *Sean Griffin*
+    # use `uniq!(:annotate)` to deduplicate annotations.
+    accounts.merge(accounts.rewhere(id: 3)).uniq!(:annotate)
+    # SELECT accounts.* FROM accounts WHERE accounts.id = 3 /* david and mary */
+    ```
 
-*   Implemented `ActiveRecord::Base#pretty_print` to work with PP.
+    *Ryuta Kamizono*
 
-    *Ethan*
+*   Resolve conflict between counter cache and optimistic locking.
 
-*   Preserve type when dumping PostgreSQL point, bit, bit varying and money
-    columns.
+    Bump an Active Record instance's lock version after updating its counter
+    cache. This avoids raising an unnecessary `ActiveRecord::StaleObjectError`
+    upon subsequent transactions by maintaining parity with the corresponding
+    database record's `lock_version` column.
 
-    *Yves Senn*
+    Fixes #16449.
 
-*   New records remain new after YAML serialization.
+    *Aaron Lipman*
 
-    *Sean Griffin*
+*   Support merging option `:rewhere` to allow mergee side condition to be replaced exactly.
 
-*   PostgreSQL support default values for enum types. Fixes #7814.
+    ```ruby
+    david_and_mary = Author.where(id: david.id..mary.id)
 
-    *Yves Senn*
+    # both conflict conditions exists
+    david_and_mary.merge(Author.where(id: bob)) # => []
 
-*   PostgreSQL `default_sequence_name` respects schema. Fixes #7516.
+    # mergee side condition is replaced by rewhere
+    david_and_mary.merge(Author.rewhere(id: bob)) # => [bob]
 
-    *Yves Senn*
+    # mergee side condition is replaced by rewhere option
+    david_and_mary.merge(Author.where(id: bob), rewhere: true) # => [bob]
+    ```
 
-*   Fix `columns_for_distinct` of PostgreSQL adapter to work correctly
-    with orders without sort direction modifiers.
+    *Ryuta Kamizono*
 
-    *Nikolay Kondratyev*
+*   Add support for finding records based on signed ids, which are tamper-proof, verified ids that can be
+    set to expire and scoped with a purpose. This is particularly useful for things like password reset
+    or email verification, where you want the bearer of the signed id to be able to interact with the
+    underlying record, but usually only within a certain time period.
 
-*   PostgreSQL `reset_pk_sequence!` respects schemas. Fixes #14719.
+    ```ruby
+    signed_id = User.first.signed_id expires_in: 15.minutes, purpose: :password_reset
 
-    *Yves Senn*
+    User.find_signed signed_id # => nil, since the purpose does not match
 
-*   Keep PostgreSQL `hstore` and `json` attributes as `Hash` in `@attributes`.
-    Fixes duplication in combination with `store_accessor`.
+    travel 16.minutes
+    User.find_signed signed_id, purpose: :password_reset # => nil, since the signed id has expired
 
-    Fixes #15369.
+    travel_back
+    User.find_signed signed_id, purpose: :password_reset # => User.first
 
-    *Yves Senn*
+    User.find_signed! "bad data" # => ActiveSupport::MessageVerifier::InvalidSignature
+    ```
 
-*   `rake railties:install:migrations` respects the order of railties.
+    *DHH*
 
-    *Arun Agrawal*
+*   Support `ALGORITHM = INSTANT` DDL option for index operations on MySQL.
 
-*   Fix redefine a `has_and_belongs_to_many` inside inherited class
-    Fixing regression case, where redefining the same `has_and_belongs_to_many`
-    definition into a subclass would raise.
+    *Ryuta Kamizono*
 
-    Fixes #14983.
+*   Fix index creation to preserve index comment in bulk change table on MySQL.
 
-    *arthurnn*
+    *Ryuta Kamizono*
 
-*   Fix `has_and_belongs_to_many` public reflection.
-    When defining a `has_and_belongs_to_many`, internally we convert that to two has_many.
-    But as `reflections` is a public API, people expect to see the right macro.
+*   Allow `unscope` to be aware of table name qualified values.
 
-    Fixes #14682.
+    It is possible to unscope only the column in the specified table.
 
-    *arthurnn*
+    ```ruby
+    posts = Post.joins(:comments).group(:"posts.hidden")
+    posts = posts.where("posts.hidden": false, "comments.hidden": false)
 
-*   Fix serialization for records with an attribute named `format`.
+    posts.count
+    # => { false => 10 }
 
-    Fixes #15188.
+    # unscope both hidden columns
+    posts.unscope(where: :hidden).count
+    # => { false => 11, true => 1 }
 
-    *Godfrey Chan*
+    # unscope only comments.hidden column
+    posts.unscope(where: :"comments.hidden").count
+    # => { false => 11 }
+    ```
 
-*   When a `group` is set, `sum`, `size`, `average`, `minimum` and `maximum`
-    on a NullRelation should return a Hash.
+    *Ryuta Kamizono*, *Slava Korolev*
 
-    *Kuldeep Aggarwal*
+*   Fix `rewhere` to truly overwrite collided where clause by new where clause.
 
-*   Fix serialized fields returning serialized data after being updated with
-    `update_column`.
+    ```ruby
+    steve = Person.find_by(name: "Steve")
+    david = Author.find_by(name: "David")
 
-    *Simon Hørup Eskildsen*
+    relation = Essay.where(writer: steve)
 
-*   Fix polymorphic eager loading when using a String as foreign key.
+    # Before
+    relation.rewhere(writer: david).to_a # => []
 
-    Fixes #14734.
+    # After
+    relation.rewhere(writer: david).to_a # => [david]
+    ```
 
-    *Lauro Caetano*
+    *Ryuta Kamizono*
 
-*   Change belongs_to touch to be consistent with timestamp updates
+*   Inspect time attributes with subsec and time zone offset.
 
-    If a model is set up with a belongs_to: touch relationship the parent
-    record will only be touched if the record was modified. This makes it
-    consistent with timestamp updating on the record itself.
+    ```ruby
+    p Knot.create
+    => #<Knot id: 1, created_at: "2016-05-05 01:29:47.116928000 +0000">
+    ```
 
-    *Brock Trappitt*
+    *akinomaeni*, *Jonathan Hefner*
 
-*   Fix the inferred table name of a `has_and_belongs_to_many` auxiliary
-    table inside a schema.
+*   Deprecate passing a column to `type_cast`.
 
-    Fixes #14824.
+    *Ryuta Kamizono*
 
-    *Eric Chahin*
+*   Deprecate `in_clause_length` and `allowed_index_name_length` in `DatabaseLimits`.
 
-*   Remove unused `:timestamp` type. Transparently alias it to `:datetime`
-    in all cases. Fixes inconsistencies when column types are sent outside of
-    `ActiveRecord`, such as for XML Serialization.
+    *Ryuta Kamizono*
 
-    *Sean Griffin*
+*   Support bulk insert/upsert on relation to preserve scope values.
 
-*   Fix bug that added `table_name_prefix` and `table_name_suffix` to
-    extension names in PostgreSQL when migrating.
+    *Josef Šimánek*, *Ryuta Kamizono*
 
-    *Joao Carlos*
+*   Preserve column comment value on changing column name on MySQL.
 
-*   The `:index` option in migrations, which previously was only available for
-    `references`, now works with any column types.
+    *Islam Taha*
 
-    *Marc Schütz*
+*   Add support for `if_exists` option for removing an index.
 
-*   Add support for counter name to be passed as parameter on `CounterCache::ClassMethods#reset_counters`.
+    The `remove_index` method can take an `if_exists` option. If this is set to true an error won't be raised if the index doesn't exist.
 
-    *jnormore*
+    *Eileen M. Uchitelle*
 
-*   Restrict deletion of record when using `delete_all` with `uniq`, `group`, `having`
-    or `offset`.
+*   Remove ibm_db, informix, mssql, oracle, and oracle12 Arel visitors which are not used in the code base.
 
-    In these cases the generated query ignored them and that caused unintended
-    records to be deleted.
+    *Ryuta Kamizono*
 
-    Fixes #11985.
+*   Prevent `build_association` from `touching` a parent record if the record isn't persisted for `has_one` associations.
 
-    *Leandro Facchinetti*
+    Fixes #38219.
 
-*   Floats with limit >= 25 that get turned into doubles in MySQL no longer have
-    their limit dropped from the schema.
+    *Josh Brody*
 
-    Fixes #14135.
+*   Add support for `if_not_exists` option for adding index.
 
-    *Aaron Nelson*
+    The `add_index` method respects `if_not_exists` option. If it is set to true
+    index won't be added.
 
-*   Fix how to calculate associated class name when using namespaced `has_and_belongs_to_many`
-    association.
+    Usage:
 
-    Fixes #14709.
+    ```ruby
+      add_index :users, :account_id, if_not_exists: true
+    ```
 
-    *Kassio Borges*
+    The `if_not_exists` option passed to `create_table` also gets propagated to indexes
+    created within that migration so that if table and its indexes exist then there is no
+    attempt to create them again.
 
-*   `ActiveRecord::Relation::Merger#filter_binds` now compares equivalent symbols and
-    strings in column names as equal.
+    *Prathamesh Sonpatki*
 
-    This fixes a rare case in which more bind values are passed than there are
-    placeholders for them in the generated SQL statement, which can make PostgreSQL
-    throw a `StatementInvalid` exception.
+*   Add `ActiveRecord::Base#previously_new_record?` to show if a record was new before the last save.
 
-    *Nat Budin*
+    *Tom Ward*
 
-*   Fix `stored_attributes` to correctly merge the details of stored
-    attributes defined in parent classes.
+*   Support descending order for `find_each`, `find_in_batches`, and `in_batches`.
 
-    Fixes #14672.
+    Batch processing methods allow you to work with the records in batches, greatly reducing memory consumption, but records are always batched from oldest id to newest.
 
-    *Brad Bennett*, *Jessica Yao*, *Lakshmi Parthasarathy*
+    This change allows reversing the order, batching from newest to oldest. This is useful when you need to process newer batches of records first.
 
-*   `change_column_default` allows `[]` as argument to `change_column_default`.
+    Pass `order: :desc` to yield batches in descending order. The default remains `order: :asc`.
 
-    Fixes #11586.
+    ```ruby
+    Person.find_each(order: :desc) do |person|
+      person.party_all_night!
+    end
+    ```
 
-    *Yves Senn*
+    *Alexey Vasiliev*
 
-*   Handle `name` and `"char"` column types in the PostgreSQL adapter.
+*   Fix `insert_all` with enum values.
 
-    `name` and `"char"` are special character types used internally by
-    PostgreSQL and are used by internal system catalogs. These field types
-    can sometimes show up in structure-sniffing queries that feature internal system
-    structures or with certain PostgreSQL extensions.
+    Fixes #38716.
 
-    *J Smith*, *Yves Senn*
+    *Joel Blum*
 
-*   Fix `PostgreSQLAdapter::OID::Float#type_cast` to convert Infinity and
-    NaN PostgreSQL values into a native Ruby `Float::INFINITY` and `Float::NAN`
+*   Add support for `db:rollback:name` for multiple database applications.
 
-    Before:
+    Multiple database applications will now raise if `db:rollback` is call and recommend using the `db:rollback:[NAME]` to rollback migrations.
 
-        Point.create(value: 1.0/0)
-        Point.last.value # => 0.0
+    *Eileen M. Uchitelle*
 
-    After:
+*   `Relation#pick` now uses already loaded results instead of making another query.
 
-        Point.create(value: 1.0/0)
-        Point.last.value # => Infinity
+    *Eugene Kenny*
 
-    *Innokenty Mikhailov*
+*   Deprecate using `return`, `break` or `throw` to exit a transaction block after writes.
 
-*   Allow the PostgreSQL adapter to handle bigserial primary key types again.
+    *Dylan Thacker-Smith*
 
-    Fixes #10410.
+*   Dump the schema or structure of a database when calling `db:migrate:name`.
 
-    *Patrick Robertson*
+    In previous versions of Rails, `rails db:migrate` would dump the schema of the database. In Rails 6, that holds true (`rails db:migrate` dumps all databases' schemas), but `rails db:migrate:name` does not share that behavior.
 
-*   Deprecate joining, eager loading and preloading of instance dependent
-    associations without replacement. These operations happen before instances
-    are created. The current behavior is unexpected and can result in broken
-    behavior.
+    Going forward, calls to `rails db:migrate:name` will dump the schema (or structure) of the database being migrated.
 
-    Fixes #15024.
+    *Kyle Thompson*
 
-    *Yves Senn*
+*   Reset the `ActiveRecord::Base` connection after `rails db:migrate:name`.
 
-*   Fix `has_and_belongs_to_many` CollectionAssociation size calculations.
+    When `rails db:migrate` has finished, it ensures the `ActiveRecord::Base` connection is reset to its original configuration. Going forward, `rails db:migrate:name` will have the same behavior.
 
-    `has_and_belongs_to_many` should fall back to using the normal CollectionAssociation's
-    size calculation if the collection is not cached or loaded.
+    *Kyle Thompson*
 
-    Fixes #14913, #14914.
+*   Disallow calling `connected_to` on subclasses of `ActiveRecord::Base`.
 
-    *Fred Wu*
+    Behavior has not changed here but the previous API could be misleading to people who thought it would switch connections for only that class. `connected_to` switches the context from which we are getting connections, not the connections themselves.
 
-*   Return a non zero status when running `rake db:migrate:status` and migration table does
-    not exist.
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-    *Paul B.*
+*   Add support for horizontal sharding to `connects_to` and `connected_to`.
 
-*   Add support for module-level `table_name_suffix` in models.
+    Applications can now connect to multiple shards and switch between their shards in an application. Note that the shard swapping is still a manual process as this change does not include an API for automatic shard swapping.
 
-    This makes `table_name_suffix` work the same way as `table_name_prefix` when
-    using namespaced models.
+    Usage:
 
-    *Jenner LaFave*
+    Given the following configuration:
 
-*   Revert the behaviour of `ActiveRecord::Relation#join` changed through 4.0 => 4.1 to 4.0.
+    ```yaml
+    # config/database.yml
+    production:
+      primary:
+        database: my_database
+      primary_shard_one:
+        database: my_database_shard_one
+    ```
 
-    In 4.1.0 `Relation#join` is delegated to `Arel#SelectManager`.
-    In 4.0 series it is delegated to `Array#join`.
+    Connect to multiple shards:
 
-    *Bogdan Gusiev*
+    ```ruby
+    class ApplicationRecord < ActiveRecord::Base
+      self.abstract_class = true
 
-*   Log nil binary column values correctly.
+      connects_to shards: {
+        default: { writing: :primary },
+        shard_one: { writing: :primary_shard_one }
+      }
+    ```
 
-    When an object with a binary column is updated with a nil value
-    in that column, the SQL logger would throw an exception when trying
-    to log that nil value. This only occurs when updating a record
-    that already has a non-nil value in that column since an initial nil
-    value isn't included in the SQL anyway (at least, when dirty checking
-    is enabled.) The column's new value will now be logged as `<NULL binary data>`
-    to parallel the existing `<N bytes of binary data>` for non-nil values.
+    Swap between shards in your controller / model code:
 
-    *James Coleman*
+    ```ruby
+    ActiveRecord::Base.connected_to(shard: :shard_one) do
+      # Read from shard one
+    end
+    ```
 
-*   Rails will now pass a custom validation context through to autosave associations
-    in order to validate child associations with the same context.
+    The horizontal sharding API also supports read replicas. See guides for more details.
 
-    Fixes #13854.
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-    *Eric Chahin*, *Aaron Nelson*, *Kevin Casey*
+*   Deprecate `spec_name` in favor of `name` on database configurations.
 
-*   Stringify all variables keys of MySQL connection configuration.
+    The accessors for `spec_name` on `configs_for` and `DatabaseConfig` are deprecated. Please use `name` instead.
 
-    When `sql_mode` variable for MySQL adapters set in configuration as `String`
-    was ignored and overwritten by strict mode option.
+    Deprecated behavior:
 
-    Fixes #14895.
+    ```ruby
+    db_config = ActiveRecord::Base.configurations.configs_for(env_name: "development", spec_name: "primary")
+    db_config.spec_name
+    ```
 
-    *Paul Nikitochkin*
+    New behavior:
 
-*   Ensure SQLite3 statements are closed on errors.
+    ```ruby
+    db_config = ActiveRecord::Base.configurations.configs_for(env_name: "development", name: "primary")
+    db_config.name
+    ```
 
-    Fixes #13631.
+    *Eileen M. Uchitelle*
 
-    *Timur Alperovich*
+*   Add additional database-specific rake tasks for multi-database users.
 
-*   Give `ActiveRecord::PredicateBuilder` private methods the privacy they deserve.
+    Previously, `rails db:create`, `rails db:drop`, and `rails db:migrate` were the only rails tasks that could operate on a single
+    database. For example:
 
-    *Hector Satre*
+    ```
+    rails db:create
+    rails db:create:primary
+    rails db:create:animals
+    rails db:drop
+    rails db:drop:primary
+    rails db:drop:animals
+    rails db:migrate
+    rails db:migrate:primary
+    rails db:migrate:animals
+    ```
 
-*   When using a custom `join_table` name on a `habtm`, rails was not saving it
-    on Reflections. This causes a problem when rails loads fixtures, because it
-    uses the reflections to set database with fixtures.
+    With these changes, `rails db:schema:dump`, `rails db:schema:load`, `rails db:structure:dump`, `rails db:structure:load` and
+    `rails db:test:prepare` can additionally operate on a single database. For example:
 
-    Fixes #14845.
+    ```
+    rails db:schema:dump
+    rails db:schema:dump:primary
+    rails db:schema:dump:animals
+    rails db:schema:load
+    rails db:schema:load:primary
+    rails db:schema:load:animals
+    rails db:structure:dump
+    rails db:structure:dump:primary
+    rails db:structure:dump:animals
+    rails db:structure:load
+    rails db:structure:load:primary
+    rails db:structure:load:animals
+    rails db:test:prepare
+    rails db:test:prepare:primary
+    rails db:test:prepare:animals
+    ```
 
-    *Kassio Borges*
+    *Kyle Thompson*
 
-*   Reset the cache when modifying a Relation with cached Arel.
-    Additionally display a warning message to make the user aware.
+*   Add support for `strict_loading` mode on association declarations.
 
-    *Yves Senn*
+    Raise an error if attempting to load a record from an association that has been marked as `strict_loading` unless it was explicitly eager loaded.
 
-*   PostgreSQL should internally use `:datetime` consistently for TimeStamp. Assures
-    different spellings of timestamps are treated the same.
+    Usage:
 
-    Example:
+    ```ruby
+    class Developer < ApplicationRecord
+      has_many :projects, strict_loading: true
+    end
 
-        mytimestamp.simplified_type('timestamp without time zone')
-        # => :datetime
-        mytimestamp.simplified_type('timestamp(6) without time zone')
-        # => also :datetime (previously would be :timestamp)
+    dev = Developer.first
+    dev.projects.first
+    # => ActiveRecord::StrictLoadingViolationError: The projects association is marked as strict_loading and cannot be lazily loaded.
+    ```
 
-    See #14513.
+    *Kevin Deisz*
 
-    *Jefferson Lai*
+*   Add support for `strict_loading` mode to prevent lazy loading of records.
 
-*   `ActiveRecord::Base.no_touching` no longer triggers callbacks or start empty transactions.
+    Raise an error if a parent record is marked as `strict_loading` and attempts to lazily load its associations. This is useful for finding places you may want to preload an association and avoid additional queries.
 
-    Fixes #14841.
+    Usage:
 
-    *Lucas Mazza*
+    ```ruby
+    dev = Developer.strict_loading.first
+    dev.audit_logs.to_a
+    # => ActiveRecord::StrictLoadingViolationError: Developer is marked as strict_loading and AuditLog cannot be lazily loaded.
+    ```
 
-*   Fix name collision with `Array#select!` with `Relation#select!`.
+    *Eileen M. Uchitelle*, *Aaron Patterson*
 
-    Fixes #14752.
+*   Add support for PostgreSQL 11+ partitioned indexes when using `upsert_all`.
 
-    *Earl St Sauver*
+    *Sebastián Palma*
 
-*   Fix unexpected behavior for `has_many :through` associations going through
-    a scoped `has_many`.
+*   Adds support for `if_not_exists` to `add_column` and `if_exists` to `remove_column`.
 
-    If a `has_many` association is adjusted using a scope, and another
-    `has_many :through` uses this association, then the scope adjustment is
-    unexpectedly neglected.
+    Applications can set their migrations to ignore exceptions raised when adding a column that already exists or when removing a column that does not exist.
 
-    Fixes #14537.
+    Example Usage:
 
-    *Jan Habermann*
+    ```ruby
+    class AddColumnTitle < ActiveRecord::Migration[6.1]
+      def change
+        add_column :posts, :title, :string, if_not_exists: true
+      end
+    end
+    ```
 
-*   `@destroyed` should always be set to `false` when an object is duped.
+    ```ruby
+    class RemoveColumnTitle < ActiveRecord::Migration[6.1]
+      def change
+        remove_column :posts, :title, if_exists: true
+      end
+    end
+    ```
 
-    *Kuldeep Aggarwal*
+    *Eileen M. Uchitelle*
 
-*   Enable `has_many` associations to support irregular inflections.
+*   Regexp-escape table name for MS SQL Server.
 
-    Fixes #8928.
+    Add `Regexp.escape` to one method in ActiveRecord, so that table names with regular expression characters in them work as expected. Since MS SQL Server uses "[" and "]" to quote table and column names, and those characters are regular expression characters, methods like `pluck` and `select` fail in certain cases when used with the MS SQL Server adapter.
 
-    *arthurnn*, *Javier Goizueta*
+    *Larry Reid*
 
-*   Fix `count` used with a grouping not returning a Hash.
+*   Store advisory locks on their own named connection.
 
-    Fixes #14721.
+    Previously advisory locks were taken out against a connection when a migration started. This works fine in single database applications but doesn't work well when migrations need to open new connections which results in the lock getting dropped.
 
-    *Eric Chahin*
+    In order to fix this we are storing the advisory lock on a new connection with the connection specification name `AdvisoryLockBase`. The caveat is that we need to maintain at least 2 connections to a database while migrations are running in order to do this.
 
-*   `sanitize_sql_like` helper method to escape a string for safe use in an SQL
-    LIKE statement.
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-    Example:
+*   Allow schema cache path to be defined in the database configuration file.
 
-        class Article
-          def self.search(term)
-            where("title LIKE ?", sanitize_sql_like(term))
-          end
-        end
+    For example:
 
-        Article.search("20% _reduction_")
-        # => Query looks like "... title LIKE '20\% \_reduction\_' ..."
+    ```yaml
+    development:
+      adapter: postgresql
+      database: blog_development
+      pool: 5
+      schema_cache_path: tmp/schema/main.yml
+    ```
 
-    *Rob Gilson*, *Yves Senn*
+    *Katrina Owen*
 
-*   Do not quote uuid default value on `change_column`.
+*   Deprecate `#remove_connection` in favor of `#remove_connection_pool` when called on the handler.
 
-    Fixes #14604.
+    `#remove_connection` is deprecated in order to support returning a `DatabaseConfig` object instead of a `Hash`. Use `#remove_connection_pool`, `#remove_connection` will be removed in 6.2.
 
-    *Eric Chahin*
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-*   The comparison between `Relation` and `CollectionProxy` should be consistent.
+*   Deprecate `#default_hash` and it's alias `#[]` on database configurations.
 
-    Example:
+    Applications should use `configs_for`. `#default_hash` and `#[]` will be removed in 6.2.
 
-        author.posts == Post.where(author_id: author.id)
-        # => true
-        Post.where(author_id: author.id) == author.posts
-        # => true
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-    Fixes #13506.
+*   Add scale support to `ActiveRecord::Validations::NumericalityValidator`.
 
-    *Lauro Caetano*
+    *Gannon McGibbon*
 
-*   Calling `delete_all` on an unloaded `CollectionProxy` no longer
-    generates an SQL statement containing each id of the collection:
+*   Find orphans by looking for missing relations through chaining `where.missing`:
 
     Before:
 
-        DELETE FROM `model` WHERE `model`.`parent_id` = 1
-        AND `model`.`id` IN (1, 2, 3...)
+    ```ruby
+    Post.left_joins(:author).where(authors: { id: nil })
+    ```
 
     After:
 
-        DELETE FROM `model` WHERE `model`.`parent_id` = 1
-
-    *Eileen M. Uchitelle*, *Aaron Patterson*
-
-*   Fix invalid SQL when aggregate methods (`empty?`, `any?`, `count`) used
-    with `select`.
-
-    Fixes #13648.
-
-    *Simon Woker*
-
-*   PostgreSQL adapter only warns once for every missing OID per connection.
+    ```ruby
+    Post.where.missing(:author)
+    ```
 
-    Fixes #14275.
+    *Tom Rossi*
 
-    *Matthew Draper*, *Yves Senn*
+*   Ensure `:reading` connections always raise if a write is attempted.
 
-*   PostgreSQL adapter automatically reloads it's type map when encountering
-    unknown OIDs.
+    Now Rails will raise an `ActiveRecord::ReadOnlyError` if any connection on the reading handler attempts to make a write. If your reading role needs to write you should name the role something other than `:reading`.
 
-    Fixes #14678.
+    *Eileen M. Uchitelle*
 
-    *Matthew Draper*, *Yves Senn*
+*   Deprecate `"primary"` as the `connection_specification_name` for `ActiveRecord::Base`.
 
-*   Fix insertion of records via `has_many :through` association with scope.
+    `"primary"` has been deprecated as the `connection_specification_name` for `ActiveRecord::Base` in favor of using `"ActiveRecord::Base"`. This change affects calls to `ActiveRecord::Base.connection_handler.retrieve_connection` and `ActiveRecord::Base.connection_handler.remove_connection`. If you're calling these methods with `"primary"`, please switch to `"ActiveRecord::Base"`.
 
-    Fixes #3548.
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-    *Ivan Antropov*
+*   Add `ActiveRecord::Validations::NumericalityValidator` with
+    support for casting floats using a database columns' precision value.
 
-*   Auto-generate stable fixture UUIDs on PostgreSQL.
+    *Gannon McGibbon*
 
-    Fixes #11524.
+*   Enforce fresh ETag header after a collection's contents change by adding
+    ActiveRecord::Relation#cache_key_with_version. This method will be used by
+    ActionController::ConditionalGet to ensure that when collection cache versioning
+    is enabled, requests using ConditionalGet don't return the same ETag header
+    after a collection is modified.
 
-    *Roderick van Domburg*
+    Fixes #38078.
 
-*   Fix a problem where an enum would overwrite values of another enum with the
-    same name in an unrelated class.
+    *Aaron Lipman*
 
-    Fixes #14607.
+*   Skip test database when running `db:create` or `db:drop` in development
+    with `DATABASE_URL` set.
 
-    *Evan Whalen*
+    *Brian Buchalter*
 
-*   PostgreSQL and SQLite string columns no longer have a default limit of 255.
+*   Don't allow mutations on the database configurations hash.
 
-    Fixes #13435, #9153.
+    Freeze the configurations hash to disallow directly changing it. If applications need to change the hash, for example to create databases for parallelization, they should use the `DatabaseConfig` object directly.
 
-    *Vladimir Sazhin*, *Toms Mikoss*, *Yves Senn*
+    Before:
 
-*   Make possible to have an association called `records`.
+    ```ruby
+    @db_config = ActiveRecord::Base.configurations.configs_for(env_name: "test", spec_name: "primary")
+    @db_config.configuration_hash.merge!(idle_timeout: "0.02")
+    ```
 
-    Fixes #11645.
+    After:
 
-    *prathamesh-sonpatki*
+    ```ruby
+    @db_config = ActiveRecord::Base.configurations.configs_for(env_name: "test", spec_name: "primary")
+    config = @db_config.configuration_hash.merge(idle_timeout: "0.02")
+    db_config = ActiveRecord::DatabaseConfigurations::HashConfig.new(@db_config.env_name, @db_config.spec_name, config)
+    ```
 
-*   `to_sql` on an association now matches the query that is actually executed, where it
-    could previously have incorrectly accrued additional conditions (e.g. as a result of
-    a previous query). `CollectionProxy` now always defers to the association scope's
-    `arel` method so the (incorrect) inherited one should be entirely concealed.
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-    Fixes #14003.
+*   Remove `:connection_id` from the `sql.active_record` notification.
 
-    *Jefferson Lai*
+    *Aaron Patterson*, *Rafael Mendonça França*
 
-*   Block a few default Class methods as scope name.
+*   The `:name` key will no longer be returned as part of `DatabaseConfig#configuration_hash`. Please use `DatabaseConfig#owner_name` instead.
 
-    For instance, this will raise:
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-        scope :public, -> { where(status: 1) }
+*   ActiveRecord's `belongs_to_required_by_default` flag can now be set per model.
 
-    *arthurnn*
+    You can now opt-out/opt-in specific models from having their associations required
+    by default.
 
-*   Fix error when using `with_options` with lambda.
+    This change is meant to ease the process of migrating all your models to have
+    their association required.
 
-    Fixes #9805.
+    *Edouard Chin*
 
-    *Lauro Caetano*
+*   The `connection_config` method has been deprecated, please use `connection_db_config` instead which will return a `DatabaseConfigurations::DatabaseConfig` instead of a `Hash`.
 
-*   Switch `sqlite3:///` URLs (which were temporarily
-    deprecated in 4.1) from relative to absolute.
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-    If you still want the previous interpretation, you should replace
-    `sqlite3:///my/path` with `sqlite3:my/path`.
+*   Retain explicit selections on the base model after applying `includes` and `joins`.
 
-    *Matthew Draper*
+    Resolves #34889.
 
-*   Treat blank UUID values as `nil`.
+    *Patrick Rebsch*
 
-    Example:
+*   The `database` kwarg is deprecated without replacement because it can't be used for sharding and creates an issue if it's used during a request. Applications that need to create new connections should use `connects_to` instead.
 
-        Sample.new(uuid_field: '') #=> <Sample id: nil, uuid_field: nil>
+    *Eileen M. Uchitelle*, *John Crepezzi*
 
-    *Dmitry Lavrov*
+*   Allow attributes to be fetched from Arel node groupings.
 
-*   Enable support for materialized views on PostgreSQL >= 9.3.
+    *Jeff Emminger*, *Gannon McGibbon*
 
-    *Dave Lee*
+*   A database URL can now contain a querystring value that contains an equal sign. This is needed to support passing PostgreSQL `options`.
 
-*   The PostgreSQL adapter supports custom domains. Fixes #14305.
+    *Joshua Flanagan*
 
-    *Yves Senn*
+*   Calling methods like `establish_connection` with a `Hash` which is invalid (eg: no `adapter`) will now raise an error the same way as connections defined in `config/database.yml`.
 
-*   PostgreSQL `Column#type` is now determined through the corresponding OID.
-    The column types stay the same except for enum columns. They no longer have
-    `nil` as type but `enum`.
+    *John Crepezzi*
 
-    See #7814.
+*   Specifying `implicit_order_column` now subsorts the records by primary key if available to ensure deterministic results.
 
-    *Yves Senn*
+    *Paweł Urbanek*
 
-*   Fix error when specifying a non-empty default value on a PostgreSQL array
-    column.
+*   `where(attr => [])` now loads an empty result without making a query.
 
-    Fixes #10613.
+    *John Hawthorn*
 
-    *Luke Steensen*
+*   Fixed the performance regression for `primary_keys` introduced MySQL 8.0.
 
-*   Fix error where `.persisted?` throws SystemStackError for an unsaved model with a
-    custom primary key that did not save due to validation error.
+    *Hiroyuki Ishii*
 
-    Fixes #14393.
+*   Add support for `belongs_to` to `has_many` inversing.
 
-    *Chris Finne*
+    *Gannon McGibbon*
 
-*   Introduce `validate` as an alias for `valid?`.
+*   Allow length configuration for `has_secure_token` method. The minimum length
+    is set at 24 characters.
 
-    This is more intuitive when you want to run validations but don't care about the return value.
+    Before:
 
-    *Henrik Nyh*
+    ```ruby
+    has_secure_token :auth_token
+    ```
 
-*   Create indexes inline in CREATE TABLE for MySQL.
+    After:
 
-    This is important, because adding an index on a temporary table after it has been created
-    would commit the transaction.
+    ```ruby
+    has_secure_token :default_token             # 24 characters
+    has_secure_token :auth_token, length: 36    # 36 characters
+    has_secure_token :invalid_token, length: 12 # => ActiveRecord::SecureToken::MinimumLengthError
+    ```
 
-    It also allows creating and dropping indexed tables with fewer queries and fewer permissions
-    required.
+    *Bernardo de Araujo*
 
-    Example:
+*   Deprecate `DatabaseConfigurations#to_h`. These connection hashes are still available via `ActiveRecord::Base.configurations.configs_for`.
 
-        create_table :temp, temporary: true, as: "SELECT id, name, zip FROM a_really_complicated_query" do |t|
-          t.index :zip
-        end
-        # => CREATE TEMPORARY TABLE temp (INDEX (zip)) AS SELECT id, name, zip FROM a_really_complicated_query
+    *Eileen Uchitelle*, *John Crepezzi*
 
-    *Cody Cutrer*, *Steve Rice*, *Rafael Mendonça Franca*
+*   Add `DatabaseConfig#configuration_hash` to return database configuration hashes with symbol keys, and use all symbol-key configuration hashes internally. Deprecate `DatabaseConfig#config` which returns a String-keyed `Hash` with the same values.
 
-*   Use singular table name in generated migrations when
-    `ActiveRecord::Base.pluralize_table_names` is `false`.
+    *John Crepezzi*, *Eileen Uchitelle*
 
-    Fixes #13426.
+*   Allow column names to be passed to `remove_index` positionally along with other options.
 
-    *Kuldeep Aggarwal*
+    Passing other options can be necessary to make `remove_index` correctly reversible.
 
-*   `touch` accepts many attributes to be touched at once.
+    Before:
 
-    Example:
+        add_index    :reports, :report_id               # => works
+        add_index    :reports, :report_id, unique: true # => works
+        remove_index :reports, :report_id               # => works
+        remove_index :reports, :report_id, unique: true # => ArgumentError
 
-        # touches :signed_at, :sealed_at, and :updated_at/on attributes.
-        Photo.last.touch(:signed_at, :sealed_at)
+    After:
 
-    *James Pinto*
+        remove_index :reports, :report_id, unique: true # => works
 
-*   `rake db:structure:dump` only dumps schema information if the schema
-    migration table exists.
+    *Eugene Kenny*
 
-    Fixes #14217.
+*   Allow bulk `ALTER` statements to drop and recreate indexes with the same name.
 
-    *Yves Senn*
+    *Eugene Kenny*
 
-*   Reap connections that were checked out by now-dead threads, instead
-    of waiting until they disconnect by themselves. Before this change,
-    a suitably constructed series of short-lived threads could starve
-    the connection pool, without ever having more than a couple alive at
-    the same time.
+*   `insert`, `insert_all`, `upsert`, and `upsert_all` now clear the query cache.
 
-    *Matthew Draper*
+    *Eugene Kenny*
 
-*   `pk_and_sequence_for` now ensures that only the pg_depend entries
-    pointing to pg_class, and thus only sequence objects, are considered.
+*   Call `while_preventing_writes` directly from `connected_to`.
 
-    *Josh Williams*
+    In some cases application authors want to use the database switching middleware and make explicit calls with `connected_to`. It's possible for an app to turn off writes and not turn them back on by the time we call `connected_to(role: :writing)`.
 
-*   `where.not` adds `references` for `includes` like normal `where` calls do.
+    This change allows apps to fix this by assuming if a role is writing we want to allow writes, except in the case it's explicitly turned off.
 
-    Fixes #14406.
+    *Eileen M. Uchitelle*
 
-    *Yves Senn*
+*   Improve detection of ActiveRecord::StatementTimeout with mysql2 adapter in the edge case when the query is terminated during filesort.
 
-*   Extend fixture `$LABEL` replacement to allow string interpolation.
+    *Kir Shatrov*
 
-    Example:
+*   Stop trying to read yaml file fixtures when loading Active Record fixtures.
 
-        martin:
-          email: $LABEL@email.com
+    *Gannon McGibbon*
 
-        users(:martin).email # => martin@email.com
+*   Deprecate `.reorder(nil)` with `.first` / `.first!` taking non-deterministic result.
 
-    *Eric Steele*
+    To continue taking non-deterministic result, use `.take` / `.take!` instead.
 
-*   Add support for `Relation` be passed as parameter on `QueryCache#select_all`.
+    *Ryuta Kamizono*
 
-    Fixes #14361.
+*   Preserve user supplied joins order as much as possible.
 
-    *arthurnn*
+    Fixes #36761, #34328, #24281, #12953.
 
-*   Passing an Active Record object to `find` or `exists?` is now deprecated.
-    Call `.id` on the object first.
+    *Ryuta Kamizono*
 
-    *Aaron Patterson*
+*   Allow `matches_regex` and `does_not_match_regexp` on the MySQL Arel visitor.
 
-*   Only use BINARY for MySQL case sensitive uniqueness check when column
-    has a case insensitive collation.
+    *James Pearson*
 
-    *Ryuta Kamizono*
+*   Allow specifying fixtures to be ignored by setting `ignore` in YAML file's '_fixture' section.
 
-*   Support for MySQL 5.6 fractional seconds.
+    *Tongfei Gao*
 
-    *arthurnn*, *Tatsuhiko Miyagawa*
+*   Make the DATABASE_URL env variable only affect the primary connection. Add new env variables for multiple databases.
 
-*   Support for PostgreSQL `citext` data type enabling case-insensitive
-   `where` values without needing to wrap in UPPER/LOWER sql functions.
+    *John Crepezzi*, *Eileen Uchitelle*
 
-    *Troy Kruthoff*, *Lachlan Sylvester*
+*   Add a warning for enum elements with 'not_' prefix.
 
-*   Only save has_one associations if record has changes.
-    Previously after save related callbacks, such as `#after_commit`, were triggered when the has_one
-    object did not get saved to the db.
+        class Foo
+          enum status: [:sent, :not_sent]
+        end
 
-    *Alan Kennedy*
+    *Edu Depetris*
 
-*   Allow strings to specify the `#order` value.
+*   Make currency symbols optional for money column type in PostgreSQL.
 
-    Example:
+    *Joel Schneider*
 
-        Model.order(id: 'asc').to_sql == Model.order(id: :asc).to_sql
+*   Add support for beginless ranges, introduced in Ruby 2.7.
 
-    *Marcelo Casiraghi*, *Robin Dupret*
+    *Josh Goodall*
 
-*   Dynamically register PostgreSQL enum OIDs. This prevents "unknown OID"
-    warnings on enum columns.
+*   Add `database_exists?` method to connection adapters to check if a database exists.
 
-    *Dieter Komendera*
+    *Guilherme Mansur*
 
-*   `includes` is able to detect the right preloading strategy when string
-    joins are involved.
+*   Loading the schema for a model that has no `table_name` raises a `TableNotSpecified` error.
 
-    Fixes #14109.
+    *Guilherme Mansur*, *Eugene Kenny*
 
-    *Aaron Patterson*, *Yves Senn*
+*   PostgreSQL: Fix GROUP BY with ORDER BY virtual count attribute.
 
-*   Fix error with validation with enum fields for records where the value for
-    any enum attribute is always evaluated as 0 during uniqueness validation.
+    Fixes #36022.
 
-    Fixes #14172.
+    *Ryuta Kamizono*
 
-    *Vilius Luneckas* *Ahmed AbouElhamayed*
+*   Make ActiveRecord `ConnectionPool.connections` method thread-safe.
 
-*   `before_add` callbacks are fired before the record is saved on
-    `has_and_belongs_to_many` associations *and* on `has_many :through`
-    associations.  Before this change, `before_add` callbacks would be fired
-    before the record was saved on `has_and_belongs_to_many` associations, but
-    *not* on `has_many :through` associations.
+    Fixes #36465.
 
-    Fixes #14144.
+    *Jeff Doering*
 
-*   Fix STI classes not defining an attribute method if there is a conflicting
-    private method defined on its ancestors.
+*   Add support for multiple databases to `rails db:abort_if_pending_migrations`.
 
-    Fixes #11569.
+    *Mark Lee*
 
-    *Godfrey Chan*
+*   Fix sqlite3 collation parsing when using decimal columns.
 
-*   Coerce strings when reading attributes. Fixes #10485.
+    *Martin R. Schuster*
 
-    Example:
+*   Fix invalid schema when primary key column has a comment.
 
-        book = Book.new(title: 12345)
-        book.save!
-        book.title # => "12345"
+    Fixes #29966.
 
-    *Yves Senn*
+    *Guilherme Goettems Schneider*
 
-*   Deprecate half-baked support for PostgreSQL range values with excluding beginnings.
-    We currently map PostgreSQL ranges to Ruby ranges. This conversion is not fully
-    possible because the Ruby range does not support excluded beginnings.
+*   Fix table comment also being applied to the primary key column.
 
-    The current solution of incrementing the beginning is not correct and is now
-    deprecated. For subtypes where we don't know how to increment (e.g. `#succ`
-    is not defined) it will raise an `ArgumentException` for ranges with excluding
-    beginnings.
+    *Guilherme Goettems Schneider*
 
-    *Yves Senn*
+*   Allow generated `create_table` migrations to include or skip timestamps.
 
-*   Support for user created range types in PostgreSQL.
+    *Michael Duchemin*
 
-    *Yves Senn*
 
-Please check [4-1-stable](https://github.com/rails/rails/blob/4-1-stable/activerecord/CHANGELOG.md) for previous changes.
+Please check [6-0-stable](https://github.com/rails/rails/blob/6-0-stable/activerecord/CHANGELOG.md) for previous changes.