Make "class_name" option for polymorphic has_many relation to STI required

[loqimean]

Steps to reproduce

Models:

class Product < ApplicationRecord
  has_many :requests, as: :requestable, dependent: :destroy
end

# STI tables
class Request < ApplicationRecord
  belongs_to :requestable, polymorphic: true

  validate :request_type, presence: true
end

class ProductRequest < Request
  belongs_to :user
end

Then running in your console:

Product.joins(requests: :user).where(requests: { user: { role: "admin" } })

Raises you an error:

/Users/some/.rvm/gems/ruby-3.1.1/gems/activerecord-7.0.4/lib/active_record/table_metadata.rb:22:in `has_column?': undefined method `key?' for nil:NilClass (NoMethodError)

Expected behavior

To raise a detailed error, like "provide a class_name option to use joins on STI table with polymorphic, or make required option class_name for polymorphic relation with STI", like:

class Product < ApplicationRecord
  has_many :requests, as: :requestable, class_name: "ProductRequest", dependent: :destroy
end

# STI tables
class Request < ApplicationRecord
  belongs_to :requestable, polymorphic: true

  validate :request_type, presence: true
end

class ProductRequest < Request
  belongs_to :user
end

Actual behavior

Unknown error 🙂

/Users/some/.rvm/gems/ruby-3.1.1/gems/activerecord-7.0.4/lib/active_record/table_metadata.rb:22:in `has_column?': undefined method `key?' for nil:NilClass (NoMethodError)

System configuration

Rails version: 7.0.4

Ruby version: 3.1.1

[rosanelli]

+1

[xfifix]

+1

i interest because i have the same problem on my projet ! But i don't want tell the class_name, i want a belongs_to or has_many on the Request and finds the right object on its own.

[xfifix]

@byroot do you have an idea ? :)

[matthewd]

It's definitely a bug; this should be using &.:

rails/activerecord/lib/active_record/table_metadata.rb

Line 22 in 8015c2c

klass&.columns_hash.key?(column_name)

I imagine I'll be happy with whatever behaviour that restores, though. "This might be pointing at an STI class and one of the STI children might have an association with the name you asked for, and you might know that all instances referencing this object are of that class" is clearly possible, but I don't foresee us encoding that level of detail into an error.

If anyone's up for an easy PR, ☝🏻 plus a test would be welcome.

[xfifix]

@matthewd i seems to work :D

[xfifix]

@matthewd i will to try do an PR ;)

[byroot]

Backported to 7-0-stable as c1150f4

[zzak]

plus a test would be welcome

I think we missed this but just incase anyone is looking for something to do 👀