When dumping the database schema, Rails will dump foreign key names only
if those names were not generate by Rails. Currently this is determined
by checking if the foreign key name is `fk_rails_` followed by
a 10-character hash.

At [Cookpad](https://github.com/cookpad), we use
[Departure](https://github.com/departurerb/departure) (Percona's
pt-online-schema-change runner for ActiveRecord migrations) to run migrations.
Often, `pt-osc` will make a copy of a table in order to run a long migration
without blocking it. In this copy process, foreign keys are copied too,
but [their name is prefixed with an underscore to prevent name collision
](https://www.percona.com/doc/percona-toolkit/LATEST/pt-online-schema-change.html#cmdoption-pt-online-schema-change-alter-foreign-keys-method).

In the process described above, we often end up with a development
database that contains foreign keys which name starts with `_fk_rails_`.
That name does not match the ignore pattern, so next time Rails dumps
the database schema (eg. when running `rake db:migrate`), our
`db/schema.rb` file ends up containing those unwanted foreign key names.
This also produces an unwanted git diff that we'd prefer not to commit.

In this PR, I'd like to suggest a way to expose the foreign key name
ignore pattern to the Rails configuration, so that individual projects
can decide on a different pattern of foreign keys that will not get
their names dumped in `schema.rb`.

The current documentation explicitly mentions only PostgreSQL (hstore/json)
for use with `.store_accessor`, making it somewhat confusing what to choose on
a MySQL 5.7+ setup (which introduced a json data type).

If you had a three-tier config, the `establish_connection` that's called
in the Railtie on load can't figure out how to access the default
configuration.

This is because Rails assumes that the config is the first value in the
hash and always associated with the key from the environment. With a
three tier config however we need to go one level deeper.

This commit includes 2 changes. 1) removes a line from `resolve_all`
which was parsing out the the environment from the config so instead of
getting

```
{
  :development => {
    :primary => {
      :database => "whatever"
    }
  },
    :animals => {
      :database => "whatever-animals"
    }
  },
  etc with test / prod
}
```

We'd instead end up with a config that had no attachment to it's
envioronment.

```
{
  :primary => {
    :database => "whatever"
  }
  :animals => {
    :database => "whatever-animals"
  }
  etc - without test and prod
}
```

Not only did this mean that Active Record didn't know how to establish a
connection, it didn't have the other necessary configs along with it in
the configs list.

So fix this I removed the line that deletes these configs.

The second thing this commit changes is adding this line to
`establish_connection`

```
spec = spec[spec_name.to_sym] if spec[spec_name.to_sym]
```

When you have a three-tier config and don't pass any hash/symbol/env etc
to `establish_connection` the resolver will automatically return both
the primary and secondary (in this case animals db) configurations.
We'll get an `database configuration does not specify adapter` error
because AR will try to establish a connection on the `primary` key
rather than the `primary` key's config. It assumes that the
`development` or default env automatically will return a config hash,
but with a three-tier config we actually get a key and config `primary
=> config`.

This fix is a bit of a bandaid because it's not the "correct" way to
handle this situation, but it does solve our immediate problem. The new
code here is saying "if the config returned from the resolver (I know
it's called spec in here but we interchange our meanings a LOT and what
is returned is a three-tier config) has a key matching the "primary"
spec name, grab the config from the spec and pass that to the
estalbish_connection method".

This works because if we pass `:animals` or a hash, or `:primary` we'll
already have the correct configuration to connect with.

This fixes the case where we want Rail to connect with the default
connection.

Coming soon is a refactoring that should eliminate the need to do this
but I need this fix in order to write the multi-db rake tasks that I
promised in my RailsConf submission. `@tenderlove` and I are working on
the refactoring of the internals for connection management but it won't
be ready for a few weeks and this issue has been blocking progress.

[ci skip]

In #24542, quoted_time was introduced to strip the leading date
component for time columns because it was having a significant
effect in mariadb. However, it assumed that the date component
was always 2000-01-01 which isn't the case, especially if the
source wasn't another time column.

For legacy reasons Rails stores time columns on sqlite as full
timestamp strings. However because the date component wasn't being
normalized this meant that when they were read back they were being
prefixed with 2001-01-01 by ActiveModel::Type::Time. This had a
twofold result - first it meant that the fast code path wasn't being
used because the string was invalid and second it was corrupting the
second fractional component being read by the Date._parse code path.

Fix this by a combination of normalizing the timestamps on writing
and also changing Active Model to be more lenient when detecting
whether a string starts with a date component before creating the
dummy time value for parsing.

This reverts commit a19e91f0.

When a class has a belongs_to or has_one relationship with dependent: :destroy
option enabled, objects of this class should not be deleted if it's dependents
cannot be deleted.

Example:

    class Parent
      has_one :child, dependent: :destroy
    end

    class Child
      belongs_to :parent, inverse_of: :child
      before_destroy { throw :abort }
    end

    c = Child.create
    p = Parent.create(child: c)

    p.destroy

    p.destroyed? # expected: false; actual: true;

Fixes #32022

This issue is caused by `@_trigger_update_callback` won't be updated due
to `_update_record` in `Locking::Optimistic` doesn't call `super` when
optimistic locking is enabled.

Now optimistic locking concern when updating is supported by
`_update_row` low level API, therefore overriding `_update_record` is no
longer necessary.

Removing the method just fix the issue.

Closes #29096.
Closes #29321.
Closes #30823.

Currently primary key value can not be updated if a record has a locking
column because of `_update_record` in `Locking::Optimistic` doesn't
respect `id_in_database` as primary key value unlike in `Persistence`.

And also, if a record has dirty primary key value, it may destroy any
other record by the lock version of dirty record itself.

When updating/destroying persisted records, it should identify
themselves by `id_in_database`, not by dirty primary key value.

This removes `|| id` which were added in #9963 and #23887 since it is no
longer necessary.

This reverts ignoring polymorphic error introduced at 02da8aea.

What the ignoring want to solve was caused by force eager loading
regardless of whether it is necessary, but it has been fixed by #29043.

The ignoring is now only causing a mismatch of `exists?` behavior with
`to_a`, `count`, etc. It should behave consistently.

This is an alternative of #29722, and follow up of #32048.

This does not change the current behavior, but makes it easier to modify
all polymorphic names consistently.

Ruby 2.4+ provides `Hash#compact` and `Hash#compact!` natively,
so `active_support/core_ext/hash/compact` is no longer necessary.

https://bugs.ruby-lang.org/issues/12752
https://ruby-doc.org/core-2.4.0/String.html#method-i-unpack1

This comment was added at 070dda28. That arguments has already been
changed since those are internal nodoc classes, but the comment does not
reflect the current state.

I decided to remove the staled comment since it is not useful for
understanding what the class does.

[ci skip]

Follow up of b988ecb9.

This was added for internal usage, it doesn't need to be public.

Duplicated method name list is no longer needed.

There is no reason `attributes=` doesn't take `assign_attributes`.

This was added in 9bfa13bb, but it is never used from the beginning.

Prevent `ActiveRecord::FinderMethods#limited_ids_for` from using correct primary
key values even if `ORDER BY` columns include other table's primary key.

Fixes #28364.

Actually `reflection.klass` should be valid AR model unless
`polymorphic?`. Previously it worked limitedly by ignoring `NameError`
even if `reflection.klass` is invalid, and our isolated testing depends
on the limited working.
Probably we should also check the klass validity in `check_validity!`
properly. Until that time, I restored the error suppression for now.

Closes #32113.

See 948b9319 for context

Currently the place where we limit what gets sent to the database is in
the implementation for `partial_writes`. We should also be restricting
it to column names when partial writes are turned off.

Note that we're using `&` instead of just defaulting to
`self.class.column_names`, as the instance version of `attribute_names`
does not include attributes which are uninitialized (were not included
in the select clause)

This is an alternative of #29722, and revert of #29601 and a1fcbd97.

Currently, association creation and normal association finding doesn't
respect `store_full_sti_class`. But eager loading and preloading respect
the setting. This means that if set `store_full_sti_class = false`
(`true` by default), eager loading and preloading can not find
created polymorphic records.

Association creation and finding should work consistently.

BC dates are supported by both date and datetime types.

https://www.postgresql.org/docs/current/static/datatype-datetime.html

Since #1097, new datetime allows year zero as 1 BC, but new date does
not. It should be allowed even in new date consistently.

The values infinity and -infinity are supported by both date and
timestamp types.

https://www.postgresql.org/docs/current/static/datatype-datetime.html#DATATYPE-DATETIME-SPECIAL-TABLE

And also, it can not be known whether a value is infinity correctly
unless cast a value.

I've added `QueryAttribute#infinity?` to handle that case.

Closes #27585.

Since #32028, Rails 6 requires Ruby 2.3+.
No longer needed workaround for Ruby 2.2 "private attribute?" warning.