Detect correct digestible attributes

[alpaca-tc]

RailsShim.digestible_attributes_in should return only digestible attributes when AR class has _confirmation writer method.

Example:

class UserSession < ApplicationRecord
  has_one :xxx_confirmation
end

record = UserSession.new
RailsShim.digestible_attributes_in(record) #=> [:xxx]  ooops :(

# spec
it { is_expected.to validate_uniqueness_of(:bar) }

# result
Failure/Error: fit { is_expected.to validate_uniqueness_of(:bar) }

NoMethodError:
 undefined method `xxx=' for #<UserSession:0x00007feb23b822a0>

This bug introduced in #1193.

[mcmire]

Hey @alpaca-tc, just want to let you know that I saw this earlier, but we need tests for this in order to merge it. If you want to take a crack at it that's fine but I'll add those when I merge this in.

[mcmire]

Hey there — so I looked more closely into this. There are two places where RailsShim.digestible_attributes_in is used. One is in validate_presence_of, but I did a quick check at what it was doing and there doesn't seem to be any issues there. The second, as you point out, is validate_uniqueness_of. I looked further into its use of RailsShim.digestible_attributes_in, and did some digging around in the Git history to figure out why it was added. It turns out that has_secure_password makes some sanity checks on the password attribute, and back in the day, it used to use before_create, which broke the uniqueness matcher. However, since then it's been changed to use validations. So, instead of changing RailsShim.digestible_attributes_in, I think we can just remove it completely from validate_uniqueness_of. I've done this in 9f0def1. Do you want to try this commit and see if it unbreaks your test?

[alpaca-tc]

9f0def1 works perfectly🙏🙏🙏Thank you for the fixing bug 😭