[aitorlb]

Ticket: CASE-4447

Draft. Releases 2.1.2 with #82 and #83, and makes master's CI green first: both failing jobs fail on master too, independently of those two PRs.

Problem

master fails two CI jobs:

• rubocop: three offenses already on master. database_validations.gemspec:12 (Layout/ExtraSpacing, Layout/SpaceAroundOperators) and the rubocop:disable comment in spec/validations/db_belongs_to_spec.rb (Lint/CopDirectiveSyntax, a trailing : after the cop name).

• test (4.0, railsmaster): 62 failures, all in spec/validations/validates_db_uniqueness_of_spec.rb (has (almost) the same errors and does not break transaction: 22 on PostgreSQL, 18 on MySQL, 22 on SQLite), every one of them the same difference. Rails main (8.2.0.alpha) now adds the conflicting record's id to the uniqueness error details (rails/rails@03a3ed8397, "Include record ID in error when uniqueness validation fails"):

  expected {field: [{error: :taken, existing_id: 1, value: "persisted"}]} to include {field: [{error: :taken, value: "persisted"}]}
  

Fix

• The two RuboCop offenses are fixed.
• The save-path comparison leaves existing_id out of Rails' details. On save the error comes from the database's unique violation, which reports the key's values but not the existing row's id, so the gem cannot supply it without the query it exists to avoid. The valid? path still compares exactly: there the gem runs Rails' own query, so both sides carry existing_id.
• 2.1.2: CHANGELOG moves the two Unreleased entries under ## [2.1.2] - 2026-10-27, and version.rb is bumped, as in the 2.1.0 and 2.1.1 releases. A patch version: [CASE-4454] Unique violation without key detail is handled on PostgreSQL #82 fixes a NoMethodError and [CASE-4447] Handle a violated composite foreign key in db_belongs_to on PostgreSQL #83 makes db_belongs_to add the validation error it already promises for a composite key; neither adds API. [CASE-4447] Handle a violated composite foreign key in db_belongs_to on PostgreSQL #83's entry now starts with "Fix …" like [CASE-4454] Unique violation without key detail is handled on PostgreSQL #82's.

One behaviour to know about on Rails 8.2: a uniqueness error the gem adds on save has no existing_id, while Rails' own validator's has. This PR does not change that.

Tests

• RuboCop 1.92.1 with rubocop-rspec 3.10.2 (CI's versions): no offenses.
• PostgreSQL on Active Record 8.1.3: 192 examples, 0 failures.
• PostgreSQL on Rails main (83ddc9f, 8.2.0.alpha): validates_db_uniqueness_of_spec.rb has 22 failures with master's spec and 0 with this one; the PostgreSQL examples under spec/validations and spec/adapters pass (192).

Not run

MySQL and SQLite: no MySQL server here and no sqlite3 in this environment's bundle. CI covers them; their failures in the railsmaster job are the same existing_id difference, and the fix is in the adapter-independent shared example.

Reviewed by a separate agent before opening; its two wording points (the spec comment, the CHANGELOG entry) are applied.

After merging, the release itself (tag v2.1.2 on the merged master commit, as for v2.1.0 and v2.1.1, and the gem push) is for a maintainer. career-services-backend then bumps to 2.1.2 for toptal/career-services-backend#2195.

🤖 Generated with Claude Code