Otherwise random CI failure will be caused.

https://travis-ci.org/rails/rails/jobs/265848046#L777

Since 213796fb, it was lost the ability that SQL with binds for `insert`,
`update`, and `delete` (like `select_all`). This restores the ability
because `insert`, `update`, and `delete` are public API, so it should
not be removed without deprecation.

I tried to change the expectation in #29976, but it is expected behavior
at least for now. So I added the test cases to prevent anyone change the
expectation.

Because `to_sql` is public API. I introduced `to_sql_and_binds` internal
API to return SQL and binds.

`where.not` with multiple conditions is still unexpected behavior. But
`where.not` with only polymorphic association has already been fixed in
213796fb.

Closes #14161.
Closes #16983.
Closes #17010.
Closes #26207.

Move common tests to `SharedGeneratorTests`

Faster and more readable implementation of Hash#deep_merge

The initial commit (efaa6e4f) that changed this
behavior was intended to be a minor change, but ended up becoming a
large-ish breaking change within Active Record.

This is because instead of only JSON encoding `Hash`es or `Array`s in `#serialize`,
we now encode all values passed in. This is an issue if you're passing in a `String`,
that has already been transformed from a `Hash` to a `String`, since your data
is now being double encoded.

Unfortunately, the change was included in one of the v5.0.0 beta
releases, and it is too late to revert without huge ripple effects.
Thus, all we can do is update the documentation (via this commit), and
add some test coverage (coming soon in a PR) for the new behavior.

Please note that in the documentation I talk about deserialization, not
about serialization, where the actual change occurred. This is because
you won't notice any changes in serialized data until you try and
deserialize it. Also to make the change itself (confusing until you
read through everything multiple times) easier to understand.

Related #27788, #25594, #26101, #24234, #28292, #28285, #28285, and
probably others.

[ci skip]

Temporarily point to a fork of SDoc

Merge pull request #30294 from yahonda/move_test_index_is_created_for_both_timestamps_to_without_transaction

Move `test_index_is_created_for_both_timestamps` to `TimestampsWithou…

Add missing test for Blob#purge

Use nil Logger for tests

Remove obsolete explanation about passing string to `:if` and `:unless` [ci skip]

[ci skip]

about passing string to `:if` and `:unless`.
This is a follow up to https://github.com/rails/rails/pull/28058 [ci
skip]

This commit addresses these failures when the backend RDBMS executes implicit commit for DDL like MySQL and Oracle.

```ruby
$ ARCONN=mysql2 bin/test test/cases/integration_test.rb test/cases/timestamp_test.rb --seed 58225 -n '/^(?:TimestampTest#(?:test_index_is_created_for_both_timestamps)|IntegrationTest#(?:test_cache_key_for_newer_updated_at|test_cache_key_format_for_existing_record_with_updated_at|test_cache_key_format_for_existing_record_with_updated_at_and_custom_cache_timestamp_format))$/' -v
Using mysql2
Run options: --seed 58225 -n "/^(?:TimestampTest#(?:test_index_is_created_for_both_timestamps)|IntegrationTest#(?:test_cache_key_for_newer_updated_at|test_cache_key_format_for_existing_record_with_updated_at|test_cache_key_format_for_existing_record_with_updated_at_and_custom_cache_timestamp_format))$/" -v

TimestampTest#test_index_is_created_for_both_timestamps = 0.19 s = .
IntegrationTest#test_cache_key_format_for_existing_record_with_updated_at = 0.05 s = F
IntegrationTest#test_cache_key_format_for_existing_record_with_updated_at_and_custom_cache_timestamp_format = 0.02 s = F
IntegrationTest#test_cache_key_for_newer_updated_at = 0.01 s = F

Finished in 0.272880s, 14.6585 runs/s, 14.6585 assertions/s.

  1) Failure:
IntegrationTest#test_cache_key_format_for_existing_record_with_updated_at [/home/yahonda/git/rails/activerecord/test/cases/integration_test.rb:111]:
--- expected
+++ actual
@@ -1 +1 @@
-"developers/1-20170717135609430848"
+"developers/1-20170817135609283207"

  2) Failure:
IntegrationTest#test_cache_key_format_for_existing_record_with_updated_at_and_custom_cache_timestamp_format [/home/yahonda/git/rails/activerecord/test/cases/integration_test.rb:116]:
--- expected
+++ actual
@@ -1 +1 @@
-"cached_developers/1-20170717135609"
+"cached_developers/1-20170817135609"

  3) Failure:
IntegrationTest#test_cache_key_for_newer_updated_at [/home/yahonda/git/rails/activerecord/test/cases/integration_test.rb:147]:
--- expected
+++ actual
@@ -1 +1 @@
-"developers/1-20170717145609430848"
+"developers/1-20170817135609283207"

4 runs, 4 assertions, 3 failures, 0 errors, 0 skips
$
```

`ActiveRecord::TestCase::TimestampTest#test_index_is_created_for_both_timestamps`
calls`ActiveRecord::TestCase::TimestampTest#setup` which updates `updated_at` column value.

```ruby
    @developer = Developer.first
    ... snip ...
    @developer.update_columns(updated_at: Time.now.prev_month)
```

This transaction expected to be rolled back, but if the backend RDBMS like MySQL perfomes
implicit commit when DDL executed, this transacion is committed by `create table` statement
inside `ActiveRecord::TestCase::TimestampTest#test_index_is_created_for_both_timestamps`.

It causes these failures above which expect `update_at` column value are not changed(rollbacked).

Makes tests much more quietly, as opposed to the enormous amount of
logging that appears right now. This setting is used in AJ, as well as
other frameworks.

Output from test run:

```

.........................................................

Finished in 3.003355s, 18.9788 runs/s, 45.2827 assertions/s.
57 runs, 136 assertions, 0 failures, 0 errors, 0 skips
```

This fork brings:

* A tiny refresh of the default theme.
* SEO tags to class files.
* The removal of HTML tags from search results.
* Some general template clean-up (HTML 5, removal of the jQuery
  effect library, etc.).
* A speed up of the generation time (by ~30 seconds).

The value isn't actually secret, so I've kept it as a comment.

Use `ActiveJob::Base.queue_adapter_name` to get adapter name

Bump `thor` and `blade`

For test with latest thor.

add missing newline

call #deep_merge instead of #dup.deep_merge!

make variable and parameter naming more consistent

change `_key` to `key`

faster implementation of Hash#deep_merge

Since 673606a9, it holds adapter name.

Allow `table_name_prefix` and `table_name_suffix` have `$`

Pass missing name attribute to execute_hook

Fixes commit 10bf93ef created to solve issue #30025

Remove unused `action` arg for `generate_routing_code`

thor 0.20.0 was released so we can test with the released version.

[ci skip]

Changed sentence to be more clear

[ci skip]

`it` and `has` are for singular, not plural

[ci skip]

Periods should be outside of the <a> tags

[ci skip]

Change namespace of Active Storage rake task

Fix RuboCop offenses

And enable `context_dependent` of Style/BracesAroundHashParameters cop.

Since c51f9b61 changed the "_before_type_cast" expectation for enum.