v2.5.0

What's Changed

Added

• The multilingual sitemap branch finally has tests — FrontPageSitemapLinksAlterKernelTest installs only drupal_kit and system, so all four of its cases ran the monolingual else-branch. The branch that reads a per-language page.front override had no coverage at all, which is exactly how it came to call a method that does not exist on the interface it was typed against: nothing ever executed the line. Every project this library serves is multilingual, so that untested branch is the one that runs in production.

  Four kernel cases with the language module installed and a Czech override: each language drops its own front page and keeps the other's (the same path survives in one language and vanishes in the other), a language with no override inherits the stored value, a link with no langcode is dropped on any match, and a surviving link loses only the alternates that point at a front page. Mutation-checked — reading $stored instead of the override fails three of the four, and disabling the multilingual branch fails all four.

Changed

• PHPStan now reads the test tree too, behind a baseline (#132) — phpcs has always scanned tests/; PHPStan had not, and two tools disagreeing on what counts as code is how the interface defect in the previous entry survived. The tree reports 573 findings at level 8, so it lands as a ratchet: the existing count is baselined, and a new finding fails CI.

  A baseline is the right call here and was the wrong call for the hook files. There, 20 findings were a morning's work and one of them was a real defect a baseline would have buried. Here the bulk is mock-typing friction — 128 phpunit.coversMethod rows that are docblock shape, and method.notFound on mocks typed as the concrete class — which is cleanup, not a bug hunt.

  The cheap half is already done: five unit-test classes now type their mocked service properties as MockObject&<Interface>, which is what makes ->expects() legible to the analyser. A mock typed as the concrete class hides a signature change behind a test that still compiles, so this is worth more than the count suggests.

• The hooks are now statically analysed, and one of them was calling a method its type does not have — phpstan.neon listed paths: [src], so drupal_kit.module and drupal_kit.install had never been read by PHPStan. Those eight hooks are the code that touches core API most directly, which makes them the code most likely to go stale when core deprecates something, and nothing was looking at them.

  With them in scope, the analysis found a latent defect: drupal_kit_simple_sitemap_links_alter() called getLanguageConfigOverride() on LanguageManagerInterface, which has no such method — it belongs to ConfigurableLanguageManagerInterface, provided only by the language module. The guard was isMultilingual(), which agrees with that condition on every real site but is not the same statement, so the call was correct by coincidence rather than by type. It now narrows with instanceof.

  The hooks also gain scalar parameter and return types. Array value types are deliberately not added: PHPStan asks for @param array<string, mixed>, and Drupal's own standard forbids exactly that on a hook implementation ("Hook implementations should not duplicate @PARAM documentation"), because the canonical documentation lives in core's *.api.php. The standard wins — core's own hooks carry no such annotation — and missingType.iterableValue is ignored for those two files with the reason written down.

  phpstan/phpstan-deprecation-rules is now installed, so a deprecated core call becomes a CI failure rather than something discovered at the next major. It reports nothing today: the module uses no deprecated API.

Pull Requests

• #131 — chore(phpstan): analyse the hooks, install deprecation rules, cover the multilingual branch
• #133 — chore(phpstan): analyse tests behind a baseline

Full Changelog: v2.4.0...v2.5.0