Stop entity autodetection from picking up a top-level ::Entity constant

[ericproulx]

Problem

Grape::DSL::Entity#entity_class_for_obj auto-detects a presenter with:

return unless klass.const_defined?(:Entity)
entity = klass.const_get(:Entity)

const_defined? with the default inherit: true walks the full ancestor chain — including Object, which is where Ruby stores top-level constants. So if an application defines any top-level class named Entity (a natural name in apps using grape-entity or similar), it is silently treated as Klass::Entity for every present-ed object that doesn't have its own nested Entity:

class Entity < Grape::Entity  # app-level base class
  expose :id
end

get '/thing' do
  present({ name: 'plain hash' })   # rendered through ::Entity, not returned as-is
end

The same path is reachable for empty collections: object_class([]) resolves to NilClass ([].first.class), whose ancestor chain also reaches Object.

This is the same failure family as #2773, where a top-level ::Options constant broke middleware building via an over-eager constant lookup.

Fix

Walk the ancestors explicitly and ask each one for its own Entity constant (const_defined?(:Entity, false)), skipping Object:

owner = klass.ancestors.detect { |ancestor| ancestor != Object && ancestor.const_defined?(:Entity, false) }
return unless owner

entity = owner.const_get(:Entity, false)
entity if entity.respond_to?(:represent)

Documented behavior is preserved: Klass::Entity on the object's class or any genuine ancestor (including module ancestors) is still found, nearest-ancestor first — which is exactly what the previous inherited lookup resolved. Only the accidental fallback to the global namespace is gone.

The regression spec stubs a top-level ::Entity (a Grape::Entity subclass, so it passes the respond_to?(:represent) check) and asserts it is not used for an object with no nested entity.

🤖 Generated with Claude Code

[github-actions]

Danger Report

No issues found.

View run