Inconsistency between running application callback execution and test results

[aprescott]

Consider an Event model created by the following migration:

class AddEvents < ActiveRecord::Migration
  def change
    create_table :events do |t|
      t.string :event_classification
      t.integer :status
    end
  end
end

Any Event of a given event_classification is intended to have one and only one entry with status = 1, and all other Events should have status = 2 to represent that they are old.

This is achieved with an after_commit callback:

class Event < ActiveRecord::Base
  after_commit :invalidate_existing_events, on: [:update, :create]

  private

  def invalidate_existing_events
    Event.where(event_classification: "foo", status: 1).where.not(id: self.id).each do |event|
      event.status = 2
      event.save!
    end
  end
end

To verify the behaviour, open up a Rails console and run the following:

Event.create(event_classification: "foo", status: 1)
Event.create(event_classification: "foo", status: 1)

The outcome is that there are two Events:

[3] pry(main)> Event.find(1)
  Event Load (0.3ms)  SELECT `events`.* FROM `events` WHERE `events`.`id` = 1 LIMIT 1
=> #<Event id: 1, event_classification: "foo", status: 2>
[4] pry(main)> Event.find(2)
  Event Load (0.3ms)  SELECT `events`.* FROM `events` WHERE `events`.`id` = 2 LIMIT 1
=> #<Event id: 2, event_classification: "foo", status: 2>

This makes sense since

1. id=2 saves and
2. updates id=1 to set status=2, which then
3. fires its own callback and
4. updates id=2 to also have status=2.

However, the following test case passes:

require "spec_helper"

describe Event do
  specify do
    Event.create(event_classification: "foo", status: 1)
    Event.create(event_classification: "foo", status: 1)

    expect(Event.where(event_classification: "foo", status: 2).count).to eq(1)
    expect(Event.where(event_classification: "foo", status: 1).count).to eq(1)
  end
end

test_after_commit is in use, and spec_helper.rb sets config.use_transactional_fixtures = true.

That is: the spec does not arrive at the same result as the development console. This caused an issue with an application since the above test passes when it should fail.

I think this is a bug with test_after_commit since it claims to enable callbacks for testing but there is this discrepancy.

Version info:

• Rails 4.0.4.
• test_after_commit 0.2.3.
• rspec and rspec-rails 2.14.1 (but also reproduced on later 0.0.x versions too).

[grosser]

any patch or pending test-case welcome