Define missing attribute methods from method_missing

[byroot]

Ref: mastodon/mastodon#27622

If applications don't eager load and use the old Rails 6.1 Marshal format, they can load Active Record instances from caches without calling init_internals hence attribute methods and aliases are nnot defined yet, leading to NoMethodError

Initially I was hopping to fully get rid of the define_attribute_methods call in init_internals in favor of this new method missing, as it would remove work from the happy path, unfortunately that isn't possible because if a generated method overrides a default method inherited from object, method_missing won't be called. e.g. Kernel#format

We may want to get rid of this extra code once we remove support for the 6.1 marshal format.

cc @ClearlyClaire
FYI: @nvasilevski

[byroot]

Backported to 7-1-stable.

[skipkayhil]

@byroot did the backport not get pushed up? I'm not seeing it on 7-1-stable

[byroot]

yeah I forgot to push 😫 . It's now at f7255ec

[rafaelfranca]

@byroot I think isn't backported yet. I don't see this code in the branch. https://github.com/rails/rails/commits/7-1-stable/activerecord/lib/active_record/model_schema.rb

There is a changelog entry for it, but no other code change went, and the commit message is wrong.

[byroot]

Hum, I must have borked my merge, I will have a look.

[stephankaag]

@byroot @rafaelfranca
This change changed / broke the following scenario where AR tries to load the schema of a non-existing table.

I noticed because motor-admin-rails uses this pattern with the audited gem.

Setup:

class Audit < ApplicationRecord
  belongs_to :auditable, polymorphic: true
end

class MotorAudit < Audit
  self.table_name = 'motor_audits'
end

class Article < ApplicationRecord
  has_many :audits, as: :auditable, class_name: "MotorAudit", inverse_of: :auditable
end

create_table :motor_audits do |t|
  t.column :auditable_id, :string
  t.column :auditable_type, :string
end

create_table :articles do |t|
  t.timestamps
end

Rails 7.1.3.2:

> article = Article.create
  TRANSACTION (0.1ms)  begin transaction
  Article Create (0.8ms)  INSERT INTO "articles" ("created_at", "updated_at") VALUES (?, ?) RETURNING "id"  [["created_at", "2024-02-28 09:09:01.693949"], ["updated_at", "2024-02-28 09:09:01.693949"]]
  TRANSACTION (0.1ms)  commit transaction
 => #<Article:0x0000000107e57588 id: 4, created_at: Wed, 28 Feb 2024 09:09:01.693949000 UTC +00:00, updated_at: Wed, 28 Feb 2024 09:09:01.693949000 UTC +00:00> 
> article.audits.create
  TRANSACTION (0.0ms)  begin transaction
  MotorAudit Create (0.2ms)  INSERT INTO "motor_audits" ("auditable_id", "auditable_type") VALUES (?, ?) RETURNING "id"  [["auditable_id", "4"], ["auditable_type", "Article"]]
  TRANSACTION (0.1ms)  commit transaction
 => #<MotorAudit:0x00000001078ffa40 id: 2, auditable_id: "4", auditable_type: "Article"> 

Rails 7.2.0.alpha:

> article = Article.create
  TRANSACTION (0.0ms)  begin transaction
  Article Create (0.8ms)  INSERT INTO "articles" ("created_at", "updated_at") VALUES (?, ?) RETURNING "id"  [["created_at", "2024-02-28 09:09:57.515359"], ["updated_at", "2024-02-28 09:09:57.515359"]]
  TRANSACTION (0.1ms)  commit transaction
 => #<Article:0x0000000105e944a8 id: 5, created_at: Wed, 28 Feb 2024 09:09:57.515359000 UTC +00:00, updated_at: Wed, 28 Feb 2024 09:09:57.515359000 UTC +00:00> 
> article.audits.create
ruby-3.2.2/bundler/gems/rails-649e99bf9267/activerecord/lib/active_record/connection_adapters/sqlite3_adapter.rb:484:in `table_structure': Could not find table 'audits' (ActiveRecord::StatementInvalid)

[byroot]

If the table doesn't exist, shouldn't Audit have self.abstract_class = true?

If that doesn't solve your problem, please open an issue with a repro script, and I'll figure a solution.

[stephankaag]

If the table doesn't exist, shouldn't Audit have self.abstract_class = true?

If that doesn't solve your problem, please open an issue with a repro script, and I'll figure a solution.

Not sure. But at least I wanted to mention that the behaviour is now changed.

[byroot]

But at least I wanted to mention that the behaviour is now changed.

Of course, but to be clear, if that behavior wasn't supported nor documented in the first place but just happened to work by accident, it's not a concern.

[stephankaag]

Pinging @omohokcoj as motor-admin-rails maintainer.