This caused a bug with the new associations implementation, because now
association conditions are represented as Arel nodes internally right up
to when the whole thing gets turned to SQL.

In Rails 3.2, association conditions get turned to raw SQL early on,
which prevents Relation#merge from interfering.

The current implementation was buggy when a default_scope existed on the
target model, since we would basically end up doing:

  default_scope.merge(association_scope)

If default_scope contained a where(foo: 'a') and association_scope
contained a where(foo: 'b').where(foo: 'c') then the merger would see
that the same column is representated on both sides of the merge and
collapse the wheres to all but the last: where(foo: 'c')

Now, the RHS of the merge is left alone.

Fixes #8990

because of an ambiguous column name. This happened if the association
model had a default scope that referenced a third table, and the third
table also referenced the original table (with an identical
foreign_key).

Mysql requires that ambiguous columns are deambiguated by using the full
table.column syntax. Postgresql and Sqlite use a different syntax for
updates altogether (and don't tolerate table.name syntax), so the fix
requires always including the full table.column and discarding it later
for Sqlite and Postgresql.

Previously, when `time_zone_aware_attributes` were enabled, after
changing a datetime or timestamp attribute and then changing it back
to the original value, `changed_attributes` still tracked the
attribute as changed. This caused `[attribute]_changed?` and
`changed?` methods to return true incorrectly.

Example:

    in_time_zone 'Paris' do
      order = Order.new
      original_time = Time.local(2012, 10, 10)
      order.shipped_at = original_time
      order.save
      order.changed? # => false

      # changing value
      order.shipped_at = Time.local(2013, 1, 1)
      order.changed? # => true

      # reverting to original value
      order.shipped_at = original_time
      order.changed? # => false, used to return true
    end

closes #6865

https://github.com/rails/rails/commit/4beb4dececcf10c642c74fbcb8548c833e921a86#commitcomment-2482869

This reverts commit 4e05bfb8.

Reason: BlankTopic#blank? should not be removed to check that dynamic finder with a bang can find a model that responds to `blank?`

this would give us some more clues in case a test silently dies inside Thread

Travis only has PostgreSQL 9.1.x but 9.2 is required for range datatypes.

The native JSON library bypasses the `to_json` overrides in
active_support/core_ext/object/to_json.rb by calling its native
implementation directly. However `ActiveRecord::Store` uses a
HWIA so `JSON.dump` will call our `to_json` instead with a
`State` object for options rather than a `Hash`. This generates
a warning when the `:encoding`, `:only` & `:except` keys are
accessed in `Hash#as_json` because the `State` object delegates
unknown keys to `instance_variable_get` in its `:[]` method.

Workaround this warning in the test by using a custom coder that
calls `ActiveSupport::JSON.encode` directly.

This reverts commit e9d2ad39.

Closes #8460

Conflicts:
	activerecord/lib/active_record/attribute_methods/time_zone_conversion.rb
	activerecord/test/cases/dirty_test.rb

Add a test case to ensure that fractional second updates are detected.

[ci skip]

because some tests were not resetting them, and thus the tests were order dependent

Refactored transaction state into its own object. Each transaction creates a new transaction state object upon initialization.

Cleanup change_schema tests to remove extra sorts on columns.

This reverts commit 637a7d9d, reversing
changes made to 5937bd02.

Suggested by @dhh.

It doesn't affect the generated SQL, so seems reasonable to continue to
allow it as an association option.

Fixes #8795

If we set encoding latin1 for a PostgreSQL database, it calls
PostgreSQLAdapter::create_database with options that have,
among other things:

  { 'encoding' => 'latin1' }

Then, we use reverse_merge(:encoding => "utf8") to setup the default
encoding. In the end, the hash looks like:

  { :encoding => 'utf8', 'encoding' => 'latin1' }

The call to options.symbolize_keys calls to_sym on each_key of this
Hash. It usually means that the encoding passed overwrites the default
utf8, but it's not guaranteed. So, we shouldn't rely on it.

The same was happening in ActiveRecord::ConnectionHandling.