DerManoMann / openapi-verifier / 35562429355

78%

master: 53%

push

github

Add rector, and split lint off test (#73)

* Split the lint step off the test script

Running the linters as part of `composer test` couples a local test run to tooling
no CI job invokes that way: code-style runs `composer lint`, static-analysis runs
`composer analyse`, and the adapter jobs call phpunit directly. `test` is phpunit
alone now.

* Add rector, and run it with static analysis rather than with lint

`rector.php` matches the repo's floor (PHP 8.1) and enables the prepared sets plus
TypedPropertyFromStrictConstructorRector; `composer rector` applies, and `composer
analyse` now ends with a dry run.

That placement is deliberate. Both frameworks are composer suggests, so rector only
sees the adapters when something installs them first — which static-analysis already
does for phpstan and code-style, where `lint` runs, does not. Run without Laravel it
reads `tests/Adapters/LaravelAdapterTest` against a base class it cannot resolve and
proposes to declare `$app` and delete `parent::setUp()`; neither is a change this
repo wants.

The `src` rewrites are that first run. One of its suggestions is dropped: typing the
loader property in `VerifiesOpenApi` also dropped its `= null`, which leaves a typed
property uninitialised and fatals on the trait's own lazy-init path — the one the
Laravel and Slim adapters reach and the test suite overrides.

* Move rector to 2.x, and keep its named-argument sorting off the fixtures

rector/rector ^1.2 requires phpstan ^1.11, which pulled the analyser back
to 1.12.34 while main resolves 2.2.14 — that, not the rewrites, is why
static-analysis reported six unmatched ignore patterns and one new error.

Rector 2's code-quality set also wants to sort named arguments
alphabetically, which would reorder the swagger-php attributes that decide
the generated specification's own order; skip those two rules. The two
remaining rewrites it proposes are applied.

The typed $openapiSpecificationLoader makes one of the baselined
"alway... (continued)

11 of 12 new or added lines in 4 files covered. (91.67%)

130 of 166 relevant lines covered (78.31%)

5.87 hits per line