Cache key behaviour for alliased timestamps

[alexandru-calinoiu]

Steps to reproduce

I a simple model

class Epixcms::Movies::Movie < ActiveRecord::Base
  alias_attribute :updated_at, :modified_date
  alias_attribute :created_at, :create_date
end

movie = Epixcms::Movies::Movie.new
movie.modified_date = Time.zone.now
movie.cache_key

Expected behavior

It should respect the defined alias and include the modified_date into the cache_key

Actual behavior

It does not include the modified_date

System configuration

Rails 4.2.7.1
Ruby 2.3.0

[prathamesh-sonpatki]

Shouldn't it be alias_attribute :modified_date, :updated_at ? new attribute name comes first.

[alexandru-calinoiu]

In this case the :updated_at is the new attribute, as it does not exist the legacy database has modified_date column.

[alexandru-calinoiu]

There is also a subtle difference in behavior for this method that is annoying.

Supposedly I work around my issue by writing this:

movie.cache_key(:modified_date)

When modified_date is nil it will throw an exception. But

movie.cache_key

When updated_at is nil does not fail it returns a cache key with the id only.

I will expect to have have the same behavior for nil modified_date.

[prathamesh-sonpatki]

Can you please post what exception you got?

[alexandru-calinoiu]

It will be a nil exception on this line

rails/activerecord/lib/active_record/integration.rb

Line 61 in b326e82

timestamp = timestamp.utc.to_s(cache_timestamp_format)

[matthewd]

Given the surrounding conditionals, the nil exception definitely sounds like a bug.

Can you reproduce the alias-ignoring behaviour on 5.0 or master? I think it might be fixed there.

[prathamesh-sonpatki]

Opened #26425 to fix the issue of getting an error when named timestamp column has value nil.

[prathamesh-sonpatki]

Found out that the issue related to aliased attribute exists on master as well. Reproduction script here - https://gist.github.com/prathamesh-sonpatki/1c82c0a86d1ca171b086324d41af8491

I fixed it here - prathamesh-sonpatki@78c8235

but would like to get some feedback cc @sgrif @matthewd

Also should we fix this issue for write_attribute method as well? It also does not honor aliased attributes as of now.