We pretty frequently get bug reports that "dirty is broken inside of
after callbacks". Intuitively they are correct. You'd expect
`Model.after_save { puts changed? }; model.save` to do the same thing as
`model.save; puts model.changed?`, but it does not.

However, changing this goes much farther than just making the behavior
more intuitive. There are a _ton_ of places inside of AR that can be
drastically simplified with this change. Specifically, autosave
associations, timestamps, touch, counter cache, and just about anything
else in AR that works with callbacks have code to try to avoid "double
save" bugs which we will be able to flat out remove with this change.

We introduce two new sets of methods, both with names that are meant to
be more explicit than dirty. The first set maintains the old behavior,
and their names are meant to center that they are about changes that
occurred during the save that just happened. They are equivalent to
`previous_changes` when called outside of after callbacks, or once the
deprecation cycle moves.

The second set is the new behavior. Their names imply that they are
talking about changes from the database representation. The fact that
this is what we really care about became clear when looking at
`BelongsTo.touch_record` when tests were failing. I'm unsure that this
set of methods should be in the public API. Outside of after callbacks,
they are equivalent to the existing methods on dirty.

Dirty itself is not deprecated, nor are the methods inside of it. They
will only emit the warning when called inside of after callbacks. The
scope of this breakage is pretty large, but the migration path is
simple. Given how much this can improve our codebase, and considering
that it makes our API more intuitive, I think it's worth doing.

When the association is autosaved we were storing the details with
string keys. This was creating inconsistency with other details that are
added using the `Errors#add` method. It was also inconsistent with the
`Errors#messages` storage.

To fix this inconsistency we are always storing with symbols. This will
cause a small breaking change because in those cases the details could
be accessed as strings keys but now it can not.

The reason that we chose to do this breaking change is because `#details`
should be considered a low level object like `#messages` is.

Fix #26499.

[Rafael Mendonça França + Marcus Vieira]

Previously, if the the association was previously loaded and then
the foreign key changed by itself, a #save call would trigger a
load of the new associated record during autosave. This is unnecessary
and the autosave code (in that case) didn't use the loaded record
anyways.

The focus of this change is to make the API more accessible.
References to method and classes should be linked to make it easy to
navigate around.

This patch makes exzessiv use of `rdoc-ref:` to provide more readable
docs. This makes it possible to document `ActiveRecord::Base#save` even
though the method is within a separate module
`ActiveRecord::Persistence`. The goal here is to bring the API closer to
the actual code that you would write.

This commit only deals with Active Record. The other gems will be
updated accordingly but in different commits. The pass through Active
Record is not completely finished yet. A follow up commit will change
the spots I haven't yet had the time to update.

/cc @fxn

[ci skip]

As per the docs, `mark_for_destruction` should do nothing if `autosave`
is not set to true. We normally persist associations on a record no
matter what if the record is a new record, but we were always skipping
records which were `marked_for_destruction?`.

Fixes #20882

This code is so fucked. Things that cause this bug not to replicate:

- Defining the validation before the association (we end up calling
  `uniq!` on the errors in the autosave validation)
- Adding `accepts_nested_attributes_for` (I have no clue why. The only
  thing it does that should affect this is adds `autosave: true` to the
  inverse reflection, and doing that manually doesn't fix this).

This solution is a hack, and I'm almost certain there's a better way to
go about it, but this shouldn't cause a huge hit on validation times,
and is the simplest way to get it done.

Fixes #20874.

`has_many` can now take `index_errors: true` as an
option.  When this is enabled, errors for nested models will be
returned alongside an index, as opposed to just the nested model name.
This option can also be enabled (or disabled) globally through
`ActiveRecord::Base.index_nested_attribute_errors`

E.X.

```ruby
class Guitar < ActiveRecord::Base
  has_many :tuning_pegs
  accepts_nested_attributes_for :tuning_pegs
end

class TuningPeg < ActiveRecord::Base
  belongs_to :guitar
  validates_numericality_of :pitch
end
```

 - Old style
 - `guitar.errors["tuning_pegs.pitch"] = ["is not a number"]`

 - New style (if defined globally, or set in has_many_relationship)
 - `guitar.errors["tuning_pegs[1].pitch"] = ["is not a number"]`

[Michael Probber, Terence Sun]

Collection associations would have already been validated, but singular
associations were not.

Fixes #18735.

When associations checked for autosave have a cycle, and
none of them is dirty, then changed_for_autosave? will be an
infinite loop. We now remember if we're in the check and
will short circuit the recursion.

Before this commit, returning `false` in an ActiveRecord `before_` callback
such as `before_create` would halt the callback chain.

After this commit, the behavior is deprecated: will still work until
the next release of Rails but will also display a deprecation warning.

The preferred way to halt a callback chain is to explicitly `throw(:abort)`.

After this commit, returning `false` in a callback will display a deprecation
warning to make developers aware of the fact that they need to explicitly
`throw(:abort)` if their intention is to halt a callback chain.

This commit also patches two internal uses of AS::Callbacks (inside
ActiveRecord and ActionDispatch) which sometimes return `false` but whose
returned value is not meaningful for the purpose of execution.

In both cases, the returned value is set to `true`, which does not affect the
execution of the callbacks but prevents unrequested deprecation warnings from
showing up.

068f092c registered autosave callbacks
as `after_save` callbacks. This caused the regression described in #17209.

Autosave callbacks should be registered as `after_update` and
`after_create` callbacks, just like before.

This is a partial revert of 068f092c.

Fixes #17209.

Callback order in Active Record objects are important. Users should not
define callbacks before the association definition or surprising
behaviours like the described at #3798 will happen. This callback order
dependency is documented at https://github.com/rails/rails/blob/31bfcdc77ca0d8cec9b5fe513bdc6f05814dd4f1/activerecord/lib/active_record/associations.rb#L1222-1227.

This reverts #15728.

Fixes #16620.

If the through record gets created in an `after_create` hook that is
defined before the association is defined (therefore after its
`after_create` hook) get saved twice. This ensures that the through
records are created only once, regardless of the order of the hooks.

Instead of checking for `macro == :has_one` throughout the
codebase we can create a `has_one?` method to match the `belongs_to?`,
`polymorphic?` and other methods.

After ad7b5efb, which changed how
has_an_belongs_to_many used to work, we start raising an error when
redefining the same has_an_belongs_to_many association. This commits fix
that regression.

[Fixes #14983]

Fix habtm reflection
Conflicts:
    activerecord/CHANGELOG.md
    activerecord/lib/active_record/counter_cache.rb
    activerecord/lib/active_record/reflection.rb
    activerecord/test/cases/reflection_test.rb

This reverts commit e94e6c27.

Conflicts:
	activerecord/CHANGELOG.md

The original commit was reverted only to be safe since #14407 were reported.
We don't have any proof we added a regression with the original commit
so reverting it now will give us more problem.

Closes #14407

associations were not being saved.

Fixes #13854.

[Eric Chahin, Aaron Nelson, & Kevin Casey]

[ci skip]

The examples are written in a way you expect them to be executable.
However one snippet assumed there to be two comments when only one
was created above.

The defined models did not extend `ActiveRecord::Base`

The example used `comments.last.mark_for_destruction`. This does no
longer load the whole collection but just the last record. It is
then refetcht on subsequent calls to `last`. This breaks the example.

This reverts commit 6e3ab3e1, reversing
changes made to 39e07b64.

Conflicts:
	activerecord/CHANGELOG.md
	activerecord/test/cases/autosave_association_test.rb

This reverts commit 5e3d466d.

Prevents save related callbacks such as `after_commit` being
triggered when `has_one` objects are already persisted and have no
changes.

Closes #11450