I was working on another branch and as part of my commit, Rubocop
removed a validation (but not the spec that asserted the validation).
Below is the "non-updating" rubocop offense on the other branch.
```shell
❯ rubocop ./app/models/notification_subscription.rb
Inspecting 1 file
C
Offenses:
app/models/notification_subscription.rb:13:29: C: [Correctable]
Rails/RedundantPresenceValidationOnBelongsTo: Remove explicit presence
validation for notifiable_id.
validates :notifiable_id, presence: true
^^^^^^^^^^^^^^
1 file inspected, 1 offense detected, 1 offense auto-correctable
```
To remediate, I ran:
```shell
> rubocop --only "Rails/RedundantPresenceValidationOnBelongsTo" \
--auto-correct
```
This resolved the `app/models`. Then did some regex magic and removed
the assertions from `spec/models`.
For Forem folks, I wrote a [forem.team post][1] discuss if this is how
we want to proceed.
[1]:https://forem.team/jeremy/rubocop-auto-updating-mayhem-33a6
* Refactoring poll/vote/skip logic
Favor asking the poll if the user has already voted instead of relying
on the poll vote or skip to know about the implementation
details (e.g. should a poll vote know that there's a related model call
poll skipped? likely not)
This refactor also tidies up the specs, for which there's quite a bit of
chatter.
Found via the `flay` gem.
* Adding method documentation
Also need to bump gitpod's build as it halted for the prior commit.
* Favoring dependent: :delete_all over :destroy
The `dependent: :destroy` callback is slower than `dependent: :delete`
and `dependent: :delete_all`. We need only favor the `dependent:
:destroy` when there's callbacks that happen.
In the case of :destroy, ActiveRecord instantiates each object and then
runs destroy. Whereas in the case of :delete, ActiveRecord issues a SQL
command to delete the related files.
It is often "safer" to use :destroy, as it guarantees that you'll
instantiate the record and run it's callbacks. But sometimes you have
to go with the speed of SQL.
This relates to #14140. Note there is still more to consider, but given
that we're looking at moving from :destroy on all of an article's page
views to :delete, we might buy enough time in the callback.
* Removing redundancy of article destroy
Prior to this commit, `before_destroy_actions` called the `bust_cache`
method which in turn called `touch_actor_latest_article_updated_at` but
`bust_cache` did not pass the destroying parameter. Then
`before_destroy_actions` immediately called
`touch_actor_latest_article_updated_at` with `destroying: true`.
With this change, we remove one of those duplicate calls.
* Fixing specs regarding relationships
* Fixing specs regarding relationships
* Favoring delete_all on user relationships
Prior to this commit, several of the user's "has_many" were marked as
`depenedent: :destroy`.
In the case of :destroy, ActiveRecord instantiates each object and then
runs destroy. Whereas in the case of :delete, ActiveRecord issues a SQL
command to delete the related files.
It is often "safer" to use :destroy, as it guarantees that you'll
instantiate the record and run it's callbacks. But sometimes you have
to go with the speed of SQL.
And while not directly related to #15424 it is representative of our
callback ecosystem creating some unexpected computational loads.
Related to #15442 and #15424
* Noting models that user no longer cascade destroys
* Enable new Rails/* cops and use autocorrect on them
* Fixed Rails/PluckInWhere leftovers
* Fix Rails/DefaultScope
* Enable and fix Rails/PluckId
* Fix manual mistake with forcing autocorrection on Rails/PluckId
* Apply PR feedback to remove Rails/PluckId inline disables
* Apply PR feedback to get rid of Rails/PluckInWhere inline
