The current code base is not uniform. After some discussion,
we have chosen to go with double quotes by default.

`rack/test` is require at the top of file.
https://github.com/rails/rails/blob/master/railties/test/railties/engine_test.rb#L3

In addtion, `Rack::Test::Methods` is include in class, extend is unnecessary.
https://github.com/rails/rails/blob/master/railties/test/railties/engine_test.rb#L9

I dropped the version constraint in web-console with the idea it will be
easier to upgrade the console between Rails releases. However, issues
like #25899 started popping up.

I'm reintroducing the constraint, but this time, I don't set an upper
limit to the major version. This will keep the web-console in a version
that always works for the current Rails version and can be easily
upgraded to the last one with `bundle update`.

We may need to backport this for Rails 5.0.1.

Fixes #25899.

- By default the session store will be set to cookie store with
  application name as session key.
- Older apps are not affected as they will have the session store
  initializer generated by Rails in older versions, and Rails will not
  overwrite the session store if it is already set or disabled.
- But new apps will not have the initializer, instead the session store
  will be set to cookie store by default.
- Based on comment by DHH here - https://github.com/rails/rails/issues/25181#issuecomment-222312764.

A mailer preview test description misspelled the word configuration.
This commit updates the test description to spell the word correctly.

`class_variable_get` is public from Ruby 1.9.
Ref: http://ruby-doc.org/core-2.2.0/Module.html#method-i-class_variable_get

accidently set it to.

Made the change and test still passes.

Additional directories can be added using
SourceAnnotationExtractor::Annotation.register_directories("spec", "other_dir")

Result: rake notes will now extract notes from these directories.

- The `boot_rails` method from abstract_unit.rb is empty after 2abcdfd9.
- So let's remove it and its usage.

Since bundler 1.12.0, the gemspec is validated so the `bundle install`
command will fail just after the gem is created causing confusion to the
users. This change was a bug fix to correctly validate gemspecs.

- Cleaned up backtrace cleaner test code and removed instaces variables

The /rails/info routes were inaccessible in apps with a catch-all
globbing route, as they were being appended after the globbing route and
would never be matched.

See also ccc3ddb7.

- Railties tests related to fetching asset URL started failing after the
  release of sprockets 3.6.1 on Travis.
- This was due to the change in
  https://github.com/rails/sprockets/pull/311/files
  which changed the logic in `concat_javascript_sources` to add `;` at
  the end of file if the source did not end with semicolon.
- Bumped up sprockets minor version and fixed the failing tests.

In #25380 I moved an `active_support/core_ext/object/blank` in the
`activesupport/lib/active_support/testing/assertions.rb`, however it caused a
chain reaction, resulting in failing railties tests.

Moving the require to `railties/test/isolation/abstract_unit.rb` and
`activesupport/test/abstract_unit.rb`.

Previously model file was generated first, which resulted in
inheriting from `ActiveRecord::Base`, but since application_record.rb
is generated as well, it should already be used.

Forgotten followup to #23669 😬

If you went to an internal route (e.g. `/rails/info/routes`), you would
previously see the following in your logger:

```bash
Processing by Rails::InfoController#routes as HTML
  Parameters: {"internal"=>true}
  Rendering /Users/jon/code/rails/rails/railties/lib/rails/templates/rails/info/routes.html.erb within layouts/application
  Rendered collection of /Users/jon/code/rails/rails/actionpack/lib/action_dispatch/middleware/templates/routes/_route.html.erb [2 times] (10.5ms)
  Rendered /Users/jon/code/rails/rails/actionpack/lib/action_dispatch/middleware/templates/routes/_table.html.erb (2.5ms)
  Rendered /Users/jon/code/rails/rails/railties/lib/rails/templates/rails/info/routes.html.erb within layouts/application (23.5ms)
Completed 200 OK in 50ms (Views: 35.1ms | ActiveRecord: 0.0ms)
```

Now, with this change, you would see:

```bash
Processing by Rails::InfoController#routes as HTML
  Rendering /Users/jon/code/rails/rails/railties/lib/rails/templates/rails/info/routes.html.erb within layouts/application
  Rendered collection of /Users/jon/code/rails/rails/actionpack/lib/action_dispatch/middleware/templates/routes/_route.html.erb [2 times] (1.6ms)
  Rendered /Users/jon/code/rails/rails/actionpack/lib/action_dispatch/middleware/templates/routes/_table.html.erb (10.2ms)
  Rendered /Users/jon/code/rails/rails/railties/lib/rails/templates/rails/info/routes.html.erb within layouts/application (17.4ms)
Completed 200 OK in 44ms (Views: 28.0ms | ActiveRecord: 0.0ms)
```

- Fixes #25183.
- The `as: :json` feature was added in
  https://github.com/rails/rails/pull/21671 and recommended to use for
  JSON endpoints so let's use it by default for API controller tests.

- Adjusted tests instead.

Fix minor regression about old apps not getting per_form_csrf and request_forgery_protection configs

- Earlier per_form_csrf_tokens and request_forgery_protection config
  files were generated for old apps upgraded to Rails 5.
- But when we collapsed all initializers into one file, the entire file
  does not get created for old apps.
- This commit fixes it and also changes values for all new defaults for
  old apps so that they will not break.
- Also added a test for `rails app:update`.

Using strings for middleware class names is deprecated in 83b767ce.

- Adjusted tests also for this new behavior.
- Based on the discussion in
  https://github.com/rails/rails/pull/25184#issuecomment-222454583.

- These initializers are new defaults for Rails 5 and newly generated
  apps do not need to change them.
- Fixes #25181

Follow up to e45f388e.

When changing the generated integration tests to assert redirects with
an absolute path this redirection check in `assert_redirected_to` against
the absolute response location would now pass:
https://github.com/rails/rails/blob/af245aaf3a1c447752a1b5895adddc66e5f73c00/actionpack/lib/action_dispatch/testing/assertions/response.rb#L55

Thus we'd break early instead of hitting `assert_operator` and have
2 fewer assertions per `assert_redirected_to` as `assert_operator`
is composed of 2 assertions internally:
https://github.com/seattlerb/minitest/blob/4e146b1515b19a5c474e39c174b037510f6dbc6e/lib/minitest/assertions.rb#L254-L258

Deduct 2 assertions for 2 redirects taking the expected count down by 4
in total for the two failing tests.

- Fixes #25180

Sorry, I missed a few places in my last PR. This should be the last of
'em 😬