In CI, these are unnecessary because done by preprocessing.
To test locally, you need to do it manually. If assets not exist, it will
fail when a run specific test, so force to build those only when necessary.

This change adds a new method that loads the YAML for the database
config without parsing the ERB. This may seem odd but bear with me:

When we added the ability to have rake tasks for multiple databases we
started looping through the configurations to collect the namespaces so
we could do `rake db:create:my_second_db`. See #32274.

This caused a problem where if you had `Rails.config.max_threads` set in
your database.yml it will blow up because the environment that defines
`max_threads` isn't loaded during `rake -T`. See #35468.

We tried to fix this by adding the ability to just load the YAML and
ignore ERB all together but that caused a bug in GitHub's YAML loading
where if you used multi-line ERB the YAML was invalid. That led us to
reverting some changes in #33748.

After trying to resolve this a bunch of ways `@tenderlove` came up with
replacing the ERB values so that we don't need to load the environment
but we also can load the YAML.

This change adds a DummyCompiler for ERB that will replace all the
values so we can load the database yaml and create the rake tasks.
Nothing else uses this method so it's "safe".

DO NOT use this method in your application.

Fixes #35468

* Add `ActiveRecord::Base.connection.truncate` for SQLite3 adapter.

SQLite doesn't support `TRUNCATE TABLE`, but SQLite3 adapter can support
`ActiveRecord::Base.connection.truncate` by using `DELETE FROM`.

`DELETE` without `WHERE` uses "The Truncate Optimization",
see https://www.sqlite.org/lang_delete.html.

* Add `rails db:seed:replant` that truncates database tables and loads the seeds

Closes #34765

Ref: https://travis-ci.org/rails/rails/jobs/501220262#L1194

Similarly to https://github.com/rails/webpacker/pull/1976

This is needed to fix #35278.

We test the inflections for both autoloaders, but we can
also autoload the constant as a sort of integration test.

[Harry Brundage & Xavier Noria]

Possible thanks to Zeitwerk 1.3.0.

Possible thanks to Zeitwerk 1.3.0.

Offenses:

railties/lib/rails/autoloaders.rb:1:1: C: [Corrected] Style/FrozenStringLiteralComment: Missing magic comment # frozen_string_literal: true.
module Rails
^
actionmailer/test/base_test.rb:917:1: C: [Corrected] Layout/EmptyLinesAroundBlockBody: Extra empty line detected at block body beginning.
actionmailer/test/base_test.rb:917:1: C: [Corrected] Layout/TrailingWhitespace: Trailing whitespace detected.
actionmailer/test/base_test.rb:917:5: C: [Corrected] Style/RedundantBegin: Redundant begin block detected.
    begin
    ^^^^^
actionmailer/test/base_test.rb:918:3: C: [Corrected] Layout/IndentationWidth: Use 2 (not 4) spaces for indentation.
      events = []
  ^^^^
actionmailer/test/base_test.rb:930:1: C: [Corrected] Layout/EmptyLinesAroundBlockBody: Extra empty line detected at block body end.
actionmailer/test/base_test.rb:930:1: C: [Corrected] Layout/TrailingWhitespace: Trailing whitespace detected.

Rails.autoloader and Rails.once_autoloader was just tentative API good
enough for a first patch. Rails.autoloader is singular and does not
convey in its name that there is another autoloader. That might be
confusing, for example if you set a logger and miss traces. On the other
hand, the name `once_autoloader` is very close to being horrible.

Rails.autoloaders.main and Rails.autoloaders.once read better for my
taste, and have a nice symmetry. Also, both "main" and "once" are four
letters long, short and same length.

They are tagged as "rails.main" and "rails.once", respectively.

References #35235.

- If you have hashes inside array, the hashes were getting initialized
  as regular HWIA wereas we want them to be
  NonSymbolAccessDeprecatedHash in order to trigger a deprecation
  warning when keys are accessed with string.

  This patch fixes that by overwriting the `[]=` to to the same
  as what HWIA does (with the difference that we don't call
  `convert_key` to not trigger a deprecation when setting value).

  I also took the liberty to extract `hash.nested_under_indifferent_access`,
  into a separate method to allow subclasses to return whatever
  they want.
  Inheriting HWIA is not common, but I think it's useful for cases
  like this one where we want to preprocess reading and writing values
  in the hash (for deprecation purposes or other reasons).

Zeitwerk is a strong dependency, planned to replace AS::Dependencies. A
line in the generated Gemfile does not convey this as much.

If generate application without specified options,`actioncable` and
`activestorage` loads by default.

The assertion from the previous PR had the expected and the actual
values in the wrong order, so when a test failed the error message was
confusing.

This commit fixes the problem by switching the order.

A change to `Rails::Application.config_for` in
https://github.com/rails/rails/pull/33815 and
https://github.com/rails/rails/pull/33882 has altered the behaviour of
the returned object in a breaking manner. Before that change, nested
hashes returned from `config_for` could be accessed using non-symbol keys.
After the change, all keys are recursively symbolized so non-symbol access
fails to read the expected values.

This is a breaking change for any app that might be relying on the
nested hashes returned from `config_for` calls, and thus should be
deprecated before being removed from the codebase.

This commit introduces a temporary `NonSymbolAccessDeprecatedHash` class
that recursively wraps any nested hashes inside the `OrderedOptions`
object returned from `config_for` and issues a deprecation notice when a
non-symbol based access is performed.

This way, apps that are still relying on the ability to access these
nested hashes using non-symbol keys will be able to observe the
deprecation notices and have time to implement changes before non-symbol
access is removed for good.

A CHANGELOG entry is also added to note that non-symbol access to nested
`config_for` hashes is deprecated.

Some tests are running yarn install during the test.
The directory used for isolation test is not subject to yarn workspace,
and it occurs because the required package is not installed.
In order to avoid this, I fixed all necessary packages to be installed
before run test and use symlink to `node_modules`.

This is a bit complicated, as `yarn install` needs to be run in a specific
directory before running the test.
However, running `yarn install` every time run the test is expensive
when testing locally and should be avoided.

This reverts commit 38f9e41f, reversing
changes made to 5e493c3b.

Reason: The ajv@6.9.1 was released that fixes issue.

Because ajv@6.9.0 does not work with webpack.
Ref: https://github.com/epoberezkin/ajv/issues/941

Revert "Remove finalizer and configuration"

This reverts commit 9e7b4a3173788ea43b11e74a4d2f69a5f1565daa.

To avoid running require individually.

For avoid to affect tests. Also, `action_text:install` task execute
`yarn add`. This is an expensive and should be avoided if it is not needed.

If they're not set we'll still fall back to localhost, but this makes it
possible to run the tests against a remote Postgres / Redis / whatever.

During the development of #33145, I have named a few concepts in the
code as `whitelisted`. We decided to stay away from the term and I
adjusted most of the code afterwards, but here are the cases I forgot to
change.

I also found a case in the API guide that we could have cleaned up as
well.

[ci skip]