Doesn't play nice with ActiveRecord::Dirty

[bryanp]

For example's sake, foo is secure:

o = SomeModel.first
o.foo
# => 'one'

o.foo = 'two'
o.foo
# => 'two'

o.changes
# => {"foo"=> ["jEZjjBbnjDDjbK4G3RR3eNMrQqWvW32vAVUssdQsn-yhZvroRTc6u7f-_K_PY9-r6FqUkinG9PDy9cjGRGq42w==|CjeZg4VWzalY5tLpkQ-q9A==|fa6e59b9614566bf29ac4a86b05e0bd885fbd67226b18271c3a6d03c4be8512b", 'two']}

o.foo_was
# => "jEZjjBbnjDDjbK4G3RR3eNMrQqWvW32vAVUssdQsn-yhZvroRTc6u7f-_K_PY9-r6FqUkinG9PDy9cjGRGq42w==|CjeZg4VWzalY5tLpkQ-q9A==|fa6e59b9614566bf29ac4a86b05e0bd885fbd67226b18271c3a6d03c4be8512b"

[neilmiddleton]

Ah, I see what you're saying - that the changes value isn't showing as encypted / decrypted?

PR's welcome ;)

[bryanp]

Honestly a bit unsure how to proceed. AR uses read_attribute to get the old value instead of calling the getter. This will always return the encrypted data. Ideas?

[dmathieu]

With a bit of refactoring, we might be able to achieve a similar behavior by overriding attribute_will_change!, which could store the unencrypted value and change it to the encrypted one in the attributes.
If run after ActiveModel::Dirty's attribute_will_change!, dirty would then store the unencrypted values.

[bryanp]

From the hour or so of digging I did, attribute_will_change! isn't called under normal circumstances. Unless I misunderstand what you're saying, the change would need to be in (or by overriding) ActiveRecord::AttributeMethods::Dirty#write_attribute.

To confirm it would work I changed this line:

old = clone_attribute_value(:read_attribute, attr)

to:

old = send(attr).clone

It does work for my test case. Obviously a real solution would need to go a bit further :)

[dmathieu]

Actually, we're already overriding read and write attribute.
https://github.com/neilmiddleton/attr_secure/blob/master/lib/attr_secure/adapters/active_record.rb#L11

However, we're calling write_attribute, while rails, internally, in the dynamic setter it defines for each attributes, is calling write_store_attribute.
Perhaps calling it instead would go through the full stack appropriately and solve this issue?

[bryanp]

Possible. Would take me some digging to understand the first arg to write_store_attribute though. Do you happen to have any insight there?

[dmathieu]

My bad, this wasn't the right way to look into. Feel free to keep investigating. I'll also take a look into it soon.