How to test a custom validator?

Question

I have the following validator:

# Source: http://guides.rubyonrails.org/active_record_validations_callbacks.html#custom-validators
# app/validators/email_validator.rb

class EmailValidator < ActiveModel::EachValidator
  def validate_each(object, attribute, value)
    unless value =~ /^([^@\s]+)@((?:[-a-z0-9]+\.)+[a-z]{2,})$/i
      object.errors[attribute] << (options[:message] || "is not formatted properly") 
    end
  end
end

I would like to be able to test this in RSpec inside of my lib directory. The problem so far is I am not sure how to initialize an EachValidator.

Answer 1

I am not a huge fan of the other approach because it ties the test too close to the implementation. Also, it's fairly hard to follow. This is the approach I ultimately use. Please keep in mind that this is a gross oversimplification of what my validator actually did... just wanted to demonstrate it more simply. There are definitely optimizations to be made

class OmniauthValidator < ActiveModel::Validator
  def validate(record)
    if !record.omniauth_provider.nil? && !%w(facebook github).include?(record.omniauth_provider)
      record.errors[:omniauth_provider] << 'Invalid omniauth provider'
    end
  end
end

Associated Spec:

require 'spec_helper'

class Validatable
  include ActiveModel::Validations
  validates_with OmniauthValidator
  attr_accessor  :omniauth_provider
end

describe OmniauthValidator do
  subject { Validatable.new }

  context 'without provider' do
    it 'is valid' do
      expect(subject).to be_valid
    end
  end

  context 'with valid provider' do
    it 'is valid' do
      subject.stubs(omniauth_provider: 'facebook')

      expect(subject).to be_valid
    end
  end

  context 'with unused provider' do
    it 'is invalid' do
      subject.stubs(omniauth_provider: 'twitter')

      expect(subject).not_to be_valid
      expect(subject).to have(1).error_on(:omniauth_provider)
    end
  end
end

Basically my approach is to create a fake object "Validatable" so that we can actually test the results on it rather than have expectations for each part of the implementation