ci: drop EOL Rails 7.2 from the matrix

[Fivell]

Drops end-of-life Rails 7.2 from CI and removes two stale exclude: rules that were suppressing working coverage.

Why

• Rails 7.2 reached EOL on 2026-08-09 (endoflife.date/rails). CI should not gate on a release that receives no security fixes.
• Rails 8.0 goes EOL on 2026-11-07, five weeks out. It is kept for now as the oldest still-supported release, but the PostgreSQL jobs have been moved to 8.1 so the DB-backed coverage is not sitting on a branch that is about to die.

Ruby stays ['3.3', '3.4', '4.0'] and the gemspec is untouched (required_ruby_version is already >= 3.3.0; the activeadmin and activerecord-import runtime constraints are unchanged). No version bump, no release.

Matrix: before / after

	Before	After
ruby	3.3, 3.4, 4.0	unchanged
rails	7.2.0, 8.0.0, 8.1.0	8.0.0, 8.1.0
activeadmin	3.2.0, 3.3.0, 3.4.0, 3.5.1	unchanged
exclude	8.0.0 x 3.2.0, 8.1.0 x 3.2.0	none
include	AA 4.0.0.beta22 on Rails 7.2.0, 8.0.0, 8.1.0	AA 4.0.0.beta22 on Rails 8.0.0, 8.1.0
total legs	37	26

Standalone jobs:

Job	Before	After
test-mysql (MySQL 8.0, AA 3.5.1)	RAILS: '8.0.0'	RAILS: '8.1.0'
test-postgres (PostgreSQL 16, AA 3.5.1)	RAILS: '8.0.0'	RAILS: '8.1.0'
test-postgres-aa4 (PostgreSQL 16, AA 4.0.0.beta22)	RAILS: '8.0.0'	RAILS: '8.1.0'
coverage	standalone, no matrix	unchanged

No exclude: was added. In particular there is no Ruby 3.3 x Rails 8.1 exclude: the anonymous-parameter-forwarding SyntaxError that motivates one exists only in Ruby 3.3.0, the first 3.3 release, and ruby/setup-ruby with ruby-version: '3.3' installs the newest 3.3 patch. Ruby 3.3 x Rails 8.1 is a valid cell.

The ActiveAdmin 3.2.0 excludes were wrong — evidence

The two removed rules claimed "ActiveAdmin 3.2.0 predates Rails 8 support". The Gemfile pins ~> #{ENV['AA']}, so the 3.2.0 matrix entry resolves to activeadmin 3.2.5, which supports Rails 8. Resolved locally:

$ RAILS=8.0.0 AA=3.2.0 bundle install   # Bundle complete! 10 Gemfile dependencies, 108 gems
  * activeadmin (3.2.5)
  * rails (8.0.5.1)
  * railties (8.0.5.1)

$ RAILS=8.1.0 AA=3.2.0 bundle install   # Bundle complete! 10 Gemfile dependencies, 109 gems
  * activeadmin (3.2.5)
  * rails (8.1.4)
  * railties (8.1.4)

Both suites pass:

$ CI=true RBENV_VERSION=3.4.10 RAILS=8.0.0 AA=3.2.0 rbenv exec bundle exec rspec spec
66 examples, 0 failures        Line coverage: 285 / 287 (99.30%)

$ CI=true RBENV_VERSION=3.4.10 RAILS=8.1.0 AA=3.2.0 rbenv exec bundle exec rspec spec
66 examples, 0 failures        Line coverage: 285 / 287 (99.30%)

So the rules are deleted rather than widened — they were excluding six green legs (3 Rubies x 2 Rails).

Standalone DB jobs

test-postgres and test-postgres-aa4 moved to Rails 8.1.0, both verified locally against a real PostgreSQL 17 instance with the same DB/DB_HOST/DB_PORT/DB_USERNAME/DB_PASSWORD wiring the workflow uses. The 67th example is spec/import_spec.rb's if ENV['DB'] == 'postgres' native-upsert case, so the adapter-specific path really did run:

$ CI=true RBENV_VERSION=3.4.10 RAILS=8.1.0 AA=3.5.1 DB=postgres ... rbenv exec bundle exec rspec spec
67 examples, 0 failures        Line coverage: 285 / 287 (99.30%)

$ CI=true RBENV_VERSION=3.4.10 RAILS=8.1.0 AA=4.0.0.beta22 DB=postgres ... rbenv exec bundle exec rspec spec
67 examples, 0 failures        Line coverage: 285 / 287 (99.30%)

test-mysql moved to RAILS: '8.1.0' as well, so no job is left on Rails 8.0 (EOL 2026-11-07).

This one could not be verified locally — no MySQL server was reachable on the dev machine (Homebrew mysql installed but stopped, and standing up an isolated mysqld was not possible there). Rather than push it unverified, it was proved on the runner first, in a throwaway branch and draft PR (#229, closed without merging): job log — 66 examples, 0 failures, with mysql2 0.5.7 built with native extensions and the dummy app configured for MySQL through ENV['DB'] in spec/support/rails_template.rb. The 67-example counts in the PostgreSQL rows below are the ENV['DB'] == 'postgres' native-upsert spec, so the numbers line up.

Only after that result did the change land here.

The service blocks themselves (mysql:8.0, postgres:16, their env, ports and health-check options) are unchanged and are independent of the Rails version.

Other checks

• tasks/test.rake already passes --skip-javascript for the AA 3 dummy app, so the Rails 8.1 stale_when_importmap_changes trap is already handled at the generator; nothing to add.
• The coverage job is a standalone job with no matrix and no if: on matrix values, so it runs exactly once; deploy-coverage's if: only gates on master + push. Unchanged.
• README has no supported-versions table and no Rails-version badge — only build/coverage/Code Climate/gem/license badges. Nothing to update.
• No CHANGELOG file exists in this repo, so no changelog entry was added.

Local verification

All runs are CI=true (the dummy app's config.eager_load = ENV["CI"].present? makes it mandatory locally), Ruby via rbenv, SQLite unless noted.

Ruby	Rails	AA	DB	Result
3.4.10	8.1.0	3.5.1	sqlite	66 examples, 0 failures
3.4.10	8.0.0	3.2.0	sqlite	66 examples, 0 failures
3.4.10	8.1.0	3.2.0	sqlite	66 examples, 0 failures
4.0.6	8.1.0	3.5.1	sqlite	66 examples, 0 failures
3.4.10	8.1.0	4.0.0.beta22	sqlite	66 examples, 0 failures
3.4.10	8.1.0	3.5.1	PostgreSQL	67 examples, 0 failures
3.4.10	8.1.0	4.0.0.beta22	PostgreSQL	67 examples, 0 failures
3.4.10	8.1.0	3.5.1	MySQL	66 examples, 0 failures — on the runner, not locally (see above)

Required run, verbatim:

$ CI=true RBENV_VERSION=3.4.10 RAILS=8.1.0 AA=3.5.1 rbenv exec bundle install
Bundle complete! 10 Gemfile dependencies, 109 gems now installed.

$ CI=true RBENV_VERSION=3.4.10 RAILS=8.1.0 AA=3.5.1 rbenv exec bundle exec rspec spec
Finished in 3.19 seconds (files took 12.62 seconds to load)
66 examples, 0 failures

Coverage report generated for RSpec to coverage/index.html
Line coverage: 285 / 287 (99.30%)