`Association` and `AssociationRelation` are implementation details, I do
not want to expose those internals in the API doc, and people basically
should not rely on those internals.

https://github.com/rails/rails/blob/8544c9c23687964ab754c06a7745215a5369a4e0/activerecord/lib/active_record/associations/association.rb
https://github.com/rails/rails/blob/8544c9c23687964ab754c06a7745215a5369a4e0/activerecord/lib/active_record/association_relation.rb

As a side note, `record.association(:name)` is also private API.

https://github.com/rails/rails/blob/8544c9c23687964ab754c06a7745215a5369a4e0/activerecord/lib/active_record/associations.rb#L228-L240

From #38142

[Rafael Mendonça França + Nick Soracco]

`ActionView::Base.annotate_template_file_names` only annotates HTML

Add attribute_names method on errors

Update the Active Model error treated as hash deprecation warning

This method replaces the `keys` method on `errors` as a way to get the error attribute names from the errors object without treating `errors` like a hash.

This message more clearly communicates how to access the attribute and message keys that you would expect to get when using the previous API.

before you might iterate over errors like,

errors.each do |attribute, message|
  # My error code here
end

This message helps the user find the methods on error that match the previous API.

In testing https://github.com/rails/rails/pull/38848 in the
GitHub monolith, we realized that we probably should only
be annotating HTML output with these comments, at least
in their current format. By passing `format` to
`erb_implementation`, we set ourselves up to eventually
support annotations for other formats as well.

While using perform_enqueued_jobs enqueued jobs must be stored as well

Clarify how broadcastings are named [ci skip]

Build Journey::Path::Pattern ast in a single loop

https://github.com/tzinfo/tzinfo/pull/113

Deprecate rendering templates with . in the name

Adding a route for Mandrill's url check.

Testing that when you set a symbol for an enum type, you get a string
back and that when you set a string you also get a string back.

This method is only used internally and it being public it was being
retorned in the `action_methods` list.

Mandrill's Inbound API checks to see if a URL exists before it creates
the webhook. It sends a HEAD request, to which we now return a 200 OK
response to indicate that the route exists.

Now we can generate inbound API calls with ease on Mandrill, without
having to shuffle around tokens in production.

Fixes #37609.

Fix random CI fail due to auto-updating timestamp

Example failure: https://buildkite.com/rails/rails/builds/68074#0fe7ca54-fcce-4a47-85db-a784275c8f51/1115-1125

Each time a job is serialized, `enqueued_at` is updated.  Thus, separate
serializations of the same job can have different `enqueued_at`
timestamps if the serializations do not occur within the same second.

* Fix EagerLoadPolyAssocsTest setup

  * EagerLoadPolyAssocsTest includes a Remembered module in multiple test
    ActiveRecord classes. The module is supposed to keep track of records
    created for each of the included classes individually, but it collects all
    records for every class. This happens because @@remembered is defined on the
    Remembered module and shared between the ActiveRecord classes.  This only
    becomes an issue for databases (like CockroachDB) that use random primary
    keys instead of sequential ones by default.
  * To fix the bug, we can make the remembered collection name unique per
    ActiveRecord class.

* Update EagerLoadPolyAssocsTest test setup

  * Instead of defining remembered as a class variable, we can define it as an
    instance variable that will be unique to every class that includes the
    Remembered module.

Heed config.force_ssl when building URL

Only compute :only and :except callback conditions once

Ref: https://github.com/rails/rails/issues/38679

`_normalize_callback_options` mutates the options hash, but
doesn't remove the `:only` and `:except` conditions.

So if you call `before_action` & al with the same option hash
you end up with multiple instance of identical `:if` / `:unless` procs

Updated Active Storage configuration guide for previewers default values [ci skip]

`url_for` will now use "https://" as the default protocol when
`Rails.application.config.force_ssl` is set to true.

Action Mailer already behaves this way, effectively.  This commit
extends that behavior application-wide.

Closes #23543.

The tasks are called db:rollback:[NAME], not db:migrate:rollback:[NAME].

Remove :rdoc Rake task in generated plugin

Clarify database.yml DATABASE_URL instructions

Add description to db:rollback command for multiple databases [ci skip]

rails -T command omits rake tasks that do not have a task description,
so I added a description to `db:rollback` command for multiple databases
and updated the documentation accordingly

From the instructions in database.yml, it is not clear that Rails will
automatically use ENV['DATABASE_URL'] if it is present.

This commit rewords the instructions to clarify that Rails will do so.

The `rdoc lib` command produces nearly the same output as `rake rdoc`.
The `rdoc lib` command also has the benefit of being standard, whereas
Rake subcommands can vary from project to project.  Since the purpose of
either command is to support local development, as opposed to generating
official docs for consumption by e.g. rubygems, the standard command
seems preferable.

Note that the `rdoc` command outputs to the doc/ directory by default,
so this commit also adds that directory to the plugin .gitignore file.

These were removed in 74201c38 when the
template for the 6.0 new framework defaults initializer was deleted.

While we no longer generate the file, upgrading applications will still
have it, so it's still important to check that these options can be set.

I introduced this pattern of referencing a constant to trigger lazy load
hooks in 458a5502, and it arrived at its
current form via c98a641f and
c24be369.

I now realise autoloading doesn't need to be involved at all; we can
require the files that trigger the lazy load hooks directly.

Merge pull request #38863 from eugeneius/utc_to_local_returns_utc_offset_times_new_framework_defaults

Allow utc_to_local_returns_utc_offset_times to be set in new_framework_defaults_6_1.rb

Enabling this option in new_framework_defaults_6_1.rb didn't work
before, as railtie initializers run before application initializers.