[nz]

Two places in the Rails integration answered an if with a value that is always truthy. A converted model's note.pinned? was true for an explicit false, and a delegate_optional :enterprise? handed back Some(false) or None, both of which take the true branch. This release makes both answer true or false, the way the unconverted model and Rails' own delegate ..., allow_nil: true do. It is 0.11.0, and it asks the reviewer to accept one behavior change: a delegated ? reader returns a boolean where it returned an Option.

An application carrying an initializer that prepends its own query_attribute unwrap onto ActiveRecord::Base can delete that initializer once it bumps to 0.11.0. The concern does the unwrap now.

What changes

query_attribute, its private attribute? alias, and the attr? methods Rails generates from them answer what the same column answers on an unconverted model. note.pinned? is false for false, note.rank? is false for 0, and note.title? is false for "". Before, every present value answered true, including false itself. This closes #88.

delegate_optional with a method name ending in ? returns true or false, and reads an absent target as false. Before, it returned Some(true), Some(false) or None, and the last two branched an if the wrong way in silence. Every other method name is lifted into an Option as before. This closes #89.

How it works

For #88, the concern defines query_attribute beside the other unwrapping seams. It unwraps the reader's Option and passes the inner value to Rails' own query_cast_attribute, so the cast stays Rails'. It then aliases the private attribute? to it, because Rails copies query_attribute into attribute? at definition time and compiles a generated pinned? to self.attribute?(:pinned). An override of query_attribute alone never reaches a generated method.

Rails' cast previously received a Some. A Some matches neither the true arm nor the false, nil arm, so the cast fell through to !blank?, and Option#blank? is none?. Every Some was therefore present, and present answered true.

For #89, delegation_body generates receiver.to_option.some_and? { |target| target.m?(...) } for a name that ends in ?, and the previous and_then body for every other name. There is no opt-out keyword. A ? name asks for a verdict, and no Option can serve as one, because every Option is truthy.

The README's delegate_optional section and the method's YARD docstring both state the convention, and a doctest delegates a ? method to prove it.

Evidence

All gates ran locally on aarch64-darwin at the branch head.

• bundle exec rake test: 156 runs, 702 assertions, 0 failures, 0 errors, 0 skips
• bundle exec rake yard:doctest: 196 runs, 520 assertions, 0 failures, 0 errors, 0 skips
• bundle exec rubocop: 25 files inspected, 6 offenses detected, the same six that main reports, none of them in a file this branch changes for the fix
• nix build .#errgonomic: /nix/store/xndn6j09bnym4x0k9xif45gaqh4mmpr0-errgonomic-0.11.0
• nix flake check --all-systems: all checks passed!

Each fix is a red commit followed by a green one, so the failing state is in the history. The #88 reproduction fails on main with Expected #<WrappedStaffRecord id: nil, staff_access: false, ...> to not be staff_access?. The #89 reproduction fails on main with six failures.

The tests live beside the code they cover. test/rails_test.rb asserts every #88 case against an unconverted twin model on the same table, rather than against a literal, and covers a column that errgonomic_optional_except keeps unwrapped. It also holds the #89 cases beside the existing delegate_optional tests.

CI runs yard doctest and rake test on Ruby 3.4.x on ubuntu-latest. It does not run rubocop, nix build or nix flake check, so those three are local-only results.

What is left open

The gemspec declares no activerecord dependency and no version bound; its only runtime dependency is concurrent-ruby ~> 1.0. The #88 fix calls query_cast_attribute, a private ActiveRecord method, which was measured on 8.1.3.1, the version Gemfile.lock resolves. Recommendation: accept here. The gem already overrides private Rails API for serialization, and adding a version bound is a separate change with its own reasoning.

Object#to_option lifts whatever it is handed, so record.delegated?.to_option is now Some(false) for an absent target where it was None. That is the one path where the #89 change is silent rather than loud; some?, unwrap_or and unwrap! on a boolean all raise NoMethodError. Recommendation: accept. to_option lifting a boolean is the documented behavior of to_option, and a caller who wants the absent case back can ask the target directly.

A ? method that answers a value by design, such as a headline? that returns a string, now answers true. Recommendation: accept, and note it in the upgrading section, which it is. That method's name misstates its return; the escape is to declare the delegate without the ?, or to write some_and? by hand.

An unprefixed delegate_optional :valid? used to answer Some(true) and so disabled validation on the declaring model in silence. It now validates. That is a behavior change on top of the intended one, and it is an improvement rather than a regression. Recommendation: accept.

[lutzcc1]

🚀