per_attribute_iv is ignoring the iv for -gcm algorithms

[trak3r]

attr_encrypted (2.0.0)

The same data encrypted with per_attribute_iv and different iv should produce different results.

require 'test_helper'

class Customer
  extend AttrEncrypted

  attr_encrypted :email,
    mode:        :per_attribute_iv,
    algorithm:   'aes-256-gcm',
    key:         'MyYBGxPZhjJaPTMtxE8muP9bBNxmRpkc'
end

class CustomerTest < ActiveSupport::TestCase
  test "two matching emails should NOT encrypt into the same string" do
    email = 'thing@thing.com'

    customer1 = Customer.new
    customer1.email = email

    customer2 = Customer.new
    customer2.email = email

    assert_not_nil   customer1.encrypted_email_iv
    assert_not_equal customer1.encrypted_email_iv, customer2.encrypted_email_iv,
      "Two different customers should NOT have received the same IV"

    assert_not_nil   customer1.encrypted_email
    assert_not_equal customer1.encrypted_email,    customer2.encrypted_email,
      "Two matching emails should NOT have encrypted into the same string"
  end
end

Failure:
CustomerTest#test_two_matching_emails_should_NOT_encrypt_into_the_same_string [/Users/.../test/models/customer_test.rb:27]
Minitest::Assertion: Two matching emails should NOT have encrypted into the same string.
Expected "\xC5\x96\xB9]b\xF9\f\xC9\xDD\n\xB8\xF0\x8C\xD0\x00\xB7\x9Eu\x12\xA7\x94+\xEB\x883\xC8\xCC\xCC\x18\xA0A" to not be equal to "\xC5\x96\xB9]b\xF9\f\xC9\xDD\n\xB8\xF0\x8C\xD0\x00\xB7\x9Eu\x12\xA7\x94+\xEB\x883\xC8\xCC\xCC\x18\xA0A".

[trak3r]

looks like this bug stems from the Encryptor gem attr-encrypted/encryptor#22

[trak3r]

This was fixed in the Encryptor gem and they did a major version increase (link above). Please update the dependencies and you should probably put a nice big warning on the README explaining what happened.

[saghaulor]

@trak3r Thanks for re-opening this. I'm working on bumping attr_encrypted to v3 but I've hit a speed bump because the bug in Encryptor v2 incorrectly caused a few tests in attr_encrypted to pass. So I'm trying to work fix those issues and as soon as I have everything working again I'll push attr_encrypted v3.

Specifically, the new AR dirty methods aren't working correctly and there were a few failures related to incorrect test setup. Most of it has been resolved but I'm having trouble getting `"#{attribute}_changed?" working.

Thank you for your patience.

[saghaulor]

Please use https://github.com/attr-encrypted/attr_encrypted/tree/3.0.0

[trak3r]

Thank you for everything.