My goal is Rails 4.2 support and it works well with Rails 4.2. I’m
leaving this bug not fixed for Rails 4.1 for the time being.

This is a workaround actually. Since
`ActiveRecord::CounterCache.each_counter_cached_associations` is
private. We shouldn’t use it. Proposals for the proper fix are welcome.

The reason the previous version stopped working is `affected_rows`
returns `0` at
[active_record/counter_cache.rb#L142](https://github.com/rails/rails/blo
b/ef99d4cd3ecc58a8c1484740b2fb5447dbda23ab/activerecord/lib/active_recor
d/counter_cache.rb#L142). If you follow the method call you’ll find
that it’s called at
[active_record/persistence.rb#L486](https://github.com/rails/rails/blob/
ef99d4cd3ecc58a8c1484740b2fb5447dbda23ab/activerecord/lib/active_record/
persistence.rb#L486). Probably we’d better override the return value in
`ActiveRecord::Relation.delete_all` at
[active_record/relation.rb#L481](https://github.com/rails/rails/blob/ef9
9d4cd3ecc58a8c1484740b2fb5447dbda23ab/activerecord/lib/active_record/rel
ation.rb#L481).

https://github.com/radar/paranoia/blame/1b051ae78793f3b62297b5adbaa9c1d0
0e6726e0/README.md#L154

http://edgeguides.rubyonrails.org/4_2_release_notes.html#active-record-d
eprecations
> Deprecated passing Active Record objects to .find or .exists?. Call
id on the objects first.
Commits in Rails:
https://github.com/rails/rails/commit/d92ae6ccca3bcfd73546d612efaea01127
0bd270
https://github.com/rails/rails/commit/d35f0033c7dec2b8d8b52058fb8db495d4
9596f7
https://github.com/rails/rails/commit/c0609dd0f0f0de604ac1ffeffdf5e3e0e7
356b35

Update README about rails3 branch

Update README.md

Add a mention for https://github.com/radar/paranoia/issues/109#issuecomment-74455252

I think this advice is misleading for most datasets and most RDBMS this
is not necessary, and as per #206 doesn't even work.

I have moved it to the wiki
https://github.com/radar/paranoia/wiki/Custom-sentinel-values

This is a test for a case when a model has a has_one association with an
after_validation callback, and we have failed validation. Paranoia will
attempt to touch the deleted_at column on the child record that has
been built but not saved.

Fixes #180

This allows testing rails 4.0 on ruby 2.2

We no longer have different behaviour between AR versions.

We shouldn't be overriding this. soft-deleted objects are still
persisted.

The default implementation is already destroy || raise

Thanks @kidlab

There was a change in ActiveRecord#update_columns on latest rails
so now it checks for `destroyed?` instead of `persisted?`. Paranoia
can't override `destoyed?` like this anymore otherwise
`update_columns` will always raise:

        ActiveRecord::ActiveRecordError: cannot update a destroyed record

see https://github.com/rails/rails/commit/0f99aa615e11c50cc64b567f54cf64f59108c5e5

To ensure that we are using the association _type in our scope.

If the association object is a has_one relationship with an :as option,
it will have a type attribute (see
https://github.com/rails/docrails/blob/master/activerecord/lib/active_record/reflection.rb#L280).

If this type attributes is present, that will be the type column on the
polymorphic model. The foreign key can be found as an attribute on
the association object.

Restoring polymorphic has_one relationships errored because paranoia was
not correctly looking up the foreign_key.

Output from failing test -

ActiveRecord::StatementInvalid: SQLite3::SQLException: no such column: parent_model_id: SELECT  "polymorphic_models".* FROM "polymorphic_models" WHERE ("polymorphic_models"."deleted_at" IS NOT NULL) AND (parent_model_id)  ORDER BY "polymorphic_models"."id" ASC LIMIT 1

The test sets up a PolymorphicModel, which has_many parents. The
ParentModel then has a has_one relationship with PolymorphicModel. When
restoring, the foreign key is set as - `self.class.name.to_s.underscore_id`
which will be parent_model_id, instead of the :as option.