Summary
AGENTS.md states "Use data providers as much as possible" under Unit Tests. The convergence pass behind #138 measured how far the suite follows that, and the gap is concentrated in the biggest files. Split out as a follow-up so it does not obscure the convention work in that pull request.
The suite has 147 data providers overall and they are 100% compliant with the naming rule: every one is dataProvider-prefixed and placed after the test it feeds. The gap is adoption, not naming.
Details
Where the gap is
tests/phpunit/Unit/Render/PanelControllerTest.php - 80 tests, 1602 lines, zero data providers. It repeats the same Enter/Enter drill-then-edit key sequence in roughly 25 test bodies. Explicit provider-shaped twins: the modal Cancel-button versus Escape pair at lines 805-821 and 823-839, the fullscreen too-small guard variants at 992-1015, and the textarea editor-hint shown/hidden pair at 681-701.tests/phpunit/Unit/Builder/FormTest.php - 52 tests, only 3 providers, with 20 structurally identical one-off ...Throws methods of about 8 lines each. These sit directly beside dataProviderLayoutMismatchThrows in the same file, which already demonstrates the provider-of-closures alternative. This is the clearest single win.tests/phpunit/Unit/Schema/AgentHelpTest.php - 23 tests, 0 providers.tests/phpunit/Unit/Resolver/InputResolverTest.php - 21 tests, 0 providers. The six env-alias tests and seven coercion tests are provider-shaped.tests/phpunit/Unit/Schema/SchemaGeneratorTest.php - 16 tests, 0 providers. Its four closure-default tests duplicate scenarios already covered verbatim in DefaultResolverTest.tests/phpunit/Unit/Widget/WidgetFactoryTest.php - testCreatesByType is 17 sequential assertInstanceOf lines in a single body.
Suggested order of work, smallest and safest first
FormTest's 20 ...Throws methods, following the provider-of-closures pattern already present in that file.PanelControllerTest's explicit twins listed above.- The Schema and Resolver suites.
Note
This is a test-only refactor with no production code change, so the existing suite is its own safety net: coverage should stay equivalent while the number of test methods drops.
Summary
AGENTS.md states "Use data providers as much as possible" under Unit Tests. The convergence pass behind #138 measured how far the suite follows that, and the gap is concentrated in the biggest files. Split out as a follow-up so it does not obscure the convention work in that pull request.
The suite has 147 data providers overall and they are 100% compliant with the naming rule: every one is
dataProvider-prefixed and placed after the test it feeds. The gap is adoption, not naming.Details
Where the gap is
tests/phpunit/Unit/Render/PanelControllerTest.php- 80 tests, 1602 lines, zero data providers. It repeats the same Enter/Enter drill-then-edit key sequence in roughly 25 test bodies. Explicit provider-shaped twins: the modal Cancel-button versus Escape pair at lines 805-821 and 823-839, the fullscreen too-small guard variants at 992-1015, and the textarea editor-hint shown/hidden pair at 681-701.tests/phpunit/Unit/Builder/FormTest.php- 52 tests, only 3 providers, with 20 structurally identical one-off...Throwsmethods of about 8 lines each. These sit directly besidedataProviderLayoutMismatchThrowsin the same file, which already demonstrates the provider-of-closures alternative. This is the clearest single win.tests/phpunit/Unit/Schema/AgentHelpTest.php- 23 tests, 0 providers.tests/phpunit/Unit/Resolver/InputResolverTest.php- 21 tests, 0 providers. The six env-alias tests and seven coercion tests are provider-shaped.tests/phpunit/Unit/Schema/SchemaGeneratorTest.php- 16 tests, 0 providers. Its four closure-default tests duplicate scenarios already covered verbatim inDefaultResolverTest.tests/phpunit/Unit/Widget/WidgetFactoryTest.php-testCreatesByTypeis 17 sequentialassertInstanceOflines in a single body.Suggested order of work, smallest and safest first
FormTest's 20...Throwsmethods, following the provider-of-closures pattern already present in that file.PanelControllerTest's explicit twins listed above.Note
This is a test-only refactor with no production code change, so the existing suite is its own safety net: coverage should stay equivalent while the number of test methods drops.