The delegation was needed since passing `relation` with
`relation.bound_attributes`. It should use `relation.arel` in that case.

`test_middleware_caches` also failed same as #29600.

https://travis-ci.org/rails/rails/jobs/248017174#L487-L489

ActiveRecord query cache is available when a connection is connected.
Therefore, query cache is unavailable when entering the ActiveRecord::Base.cache block without being connected.

```ruby
ActiveRecord::Base.cache do
  Task.find(1) # access to database.
  Task.find(1) # access to database. unavailable query cache
end
```

If we use query cache with batch script etc, we need to connect before that.
```ruby
Task.connection

ActiveRecord::Base.cache do
  Task.find(1) # access to database.
  Task.find(1) # available query cache
end
```

Before version 3.1, query cache had been enabled if a configuration was set up.
In order to solve the `DATABASE_URL` issue(#8074), ActiveRecord has checked whether a connection is connected or not.

Today, ActiveRecord.configurations respect `DATABASE_URL`.
https://github.com/rails/rails/blob/master/activerecord/lib/active_record/core.rb#L46

Passing `klass.connection` is redundant because `AssociationScope` is
passed an association itself and an association has `klass`.

Unscoping `default_scope` in associations has already supported (#17360
for preloading, c9cf8b8d for eager loading).

Fixes #20679.
Closes #16531.

```ruby
  car = Car.create!(name: "Tofaş")

  # Before
  car.bulb_ids # => SELECT "bulbs".ID FROM "bulbs" WHERE "bulbs"."name" = $1 AND "bulbs"."car_id" = $2  [["name", "defaulty"], ["car_id", 3]]

  # After
  car.bulb_ids # => SELECT "bulbs"."ID" FROM "bulbs" WHERE "bulbs"."name" = $1 AND "bulbs"."car_id" = $2  [["name", "defaulty"], ["car_id", 3]]
```

And use `assert_same` instead of `assert_equal` and tiny fix assert
message s/#reload/#reset/.

Follow up of #29511.

If a record was built on a HasManyThroughAssociation, then removed, and
then the record was saved, the removed record would be created anyways.

`test_exceptional_middleware_clears_and_disables_cache_on_error` in
postgresql adapter sometime fails recently. Show the query cache keys to
investigte the cause.

https://travis-ci.org/rails/rails/jobs/246467252#L490-L493

Covering #27994 in tests.

Closes #27994.

This fixes the following warnings:

```
activerecord/test/cases/relation_test.rb:231: warning: assigned but unused variable - authors_with_commented_posts
```

- On Mysql, some command line options that affect option-file handling such as `--no-defaults` have to be passed before any other options
- Modified rails to pass them right after the `mysql` command
- Ref https://dev.mysql.com/doc/refman/5.7/en/option-file-options.html and https://bugs.mysql.com/bug.php?id=83386
- Ref #27437

Doing `Author.joins(:posts).merge(Post.joins(:comments))` does this
`SELECT ... INNER JOIN posts ON... LEFT OUTER JOIN comments ON...`
instead of doing
`SELECT ... INNER JOIN posts ON... INNER JOIN comments ON...`.

This behavior is unexpected and makes little sense as, basically, doing
`Post.joins(:comments)` means I want posts that have comments. Turning
it to a LEFT JOIN means I want posts and join the comments data, if
any.

We can see this problem directly in the existing tests.
The test_relation_merging_with_merged_joins_as_symbols only does joins
from posts to comments to ratings while the ratings fixture isn't
loaded, but the count is non-zero.

The `@offsets` cache is used by FinderMethods to cache records found by
find_nth. This cache is cleared in AR::Relation#reset, but not in
CollectionProxy#reset or CollectionProxy#reload.

Because of this, since #29098, calling #first/#find_nth/etc after
calling #reload or #reset on an association could return a stale record.
This is an issue both when the full association target is loaded and
when the item is loaded in #find_nth.

This commit solves the problem by clearing the `@offsets` cache in
CollectionProxy#reset and CollectionProxy#reload.

`ActiveRecord::Persistence#touch` does not work well when optimistic
locking enabled and `locking_column`, without default value, is null in
the database.

Fix destroying existing object does not work well when optimistic
locking enabled and `locking column` is null in the database.

Follow 22a822e5, #28914

Improves the performance from O(n) to O(1).
Previously it would require 50 queries to
insert 50 fixtures. Now it takes only one query.

Disabled on sqlite which doesn't support multiple inserts.

This debug print was added in 87f5e5e2 for finding the regression caused
by d314646c and the regression has already been fixed in 969339bd.

A part for JSON was splitted from #28416 as #29273 and #29273 has
already been merged.

Closes #28754.
Closes #27588.

[Ryuta Kamizono & Bradley Priest]

Unlike with other databses, MySQL doesn't let you remove the column
if there's a FK on this column.

For better developer experience we want to remove the FK together with
the column.

`id` column in `subscribers` was added as a primary key for ignorable in
INSERT. But it caused `NotNullViolation` for oracle-enhanced adapter.

https://github.com/rsim/oracle-enhanced/issues/1357

I changed the column to nullable to address the issue.

The `raw_write_attribute` method is used to update a record's attributes
to reflect the new state of the database in `update_columns`. The hash
provided to `update_columns` is turned into an UPDATE query directly,
which means passing an `id` key results in an update to the `id` column,
even if the model uses a different attribute as its primary key. When
updating the record, we don't want to apply the `id` column change to
the primary key attribute, since that's not what happened in the query.

Without the code to handle this case, `write_attribute_with_type_cast`
no longer contains any logic shared between `raw_write_attribute` and
`write_attribute`, so we can inline the code into those two methods.

Fixes #29460.

If a connection adapter overrides `select_*` methods, query caching will
doesn't work. This patch changes `select_value` to `select_one` in
`Relation#exists?` to ensure query caching.

Fixes #29449.

`object.id` is correctly restored since #29378 has merged.

Closes #28274, Closes #28395.

[Ryuta Kamizono & Eugene Kenny]

Running `bin/test` from the activerecord directory produces this error:

    test/cases/adapters/postgresql/uuid_test.rb:43:in `<class:PostgresqlUUIDTest>': undefined method `supports_pgcrypto_uuid?' for #<ActiveRecord::ConnectionAdapters::SQLite3Adapter:0x007fc405e72a68> (NoMethodError)

The test only actually runs on the PostgreSQL adapter; we can avoid
triggering the error on other adapters with this `respond_to?` guard.

If a relation has binds, `connection.to_sql(relation)` without binds
will generate invalid SQL. It should use `relation.to_sql` in that case.

Add test to ensure that locking_column cannot be updated explicitly.

Add test to prevent regression on creating new object with
explicit value locking_column.

Related to #28318

`belongs_to :developer` on `Comment` model was added in 431f8e01 but it
is unused.

A collection association will raise on `#create_association` when the parent is
unpersisted. A singular association should do the same. This addresses
issue #29219.

Currently the methods of `AttributeMethods::PrimaryKey` are overwritten
by `define_attribute_methods`. It will be broken if a table that
customized primary key has non primary key id column.
It should not be overwritten if a table has any primary key.

Fixes #29350.

- Rename test cases from  `unknown` to `unrecognized` since unknown OID
is one of possible unrecognized types by Rails

- Use "select 'pg_catalog.pg_class'::regclass" whose OID is 2205, which
will not be converted to recognized type in PostgreSQL 10.

activerecord_unittest=# select oid, typname from pg_type where oid in (2205, 2277);
 oid  | typname
------+----------
 2205 | regclass
 2277 | anyarray
(2 rows)

Addresses #28868