[timtebeek]

Option keys in declarative YAML that a recipe can't bind were silently dropped. Reflective construction only reads constructor parameters, and the Jackson fallback has FAIL_ON_UNKNOWN_PROPERTIES disabled. As a result, a typo or a stale option passed validateAll() and the recipe ran without it:

recipeList:
  - org.openrewrite.text.ChangeText:
      toTxt: Hello   # org.openrewrite.text.ChangeText.toTxt: is not an option of this recipe; valid options are: toText

#4377 originally made tests fail on unknown keys, through the FAIL_ON_UNKNOWN_PROPERTIES mapper customizer in RecipeSpec. #5050 then moved recipe construction onto RecipeLoader's own mapper, so that customizer no longer reached recipes, and RecipeLifecycleTest was flipped to canCall…WithUnnecessaryArgs. This PR restores the original intent for every caller, not just tests.

YamlResourceLoader now checks each key after the recipe is built. A key is known if it is a public constructor parameter name, an @Option field, or a property Jackson can set on the recipe class itself, including @JsonAlias aliases. @Option fields count even when Jackson can't set them, because Recipe.withOptions (the marketplace path) binds fields at any visibility. The check is skipped for:

• RpcRecipe, whose options the remote validates;
• recipes declaring @JsonIgnoreProperties(ignoreUnknown = true) or a @JsonAnySetter;
• recipes whose class introspection fails, for example a missing optional dependency in a member signature, so a recipe that loaded fine is not reported invalid.

A DeclarativeRecipe reached through the marketplace takes no options. Keys are matched case-sensitively, like the reflective path, although Jackson accepts any case.

Merge after the downstream fixes

RewriteTest asserts validateAll(), and downstream CI picks up rewrite snapshots within minutes. These recipes pass unknown keys today, so their repos' tests break as soon as this merges:

• Restrict AddManualTrigger to workflow files rewrite-github-actions#240: AddManualTrigger MergeYaml fileMatcher → filePattern
• Pass configuration, not scope, to Gradle DependencyInsight in SpringWebDependency rewrite-spring#1227: SpringWebDependency Gradle DependencyInsight scope → configuration
• Use excludedBounds in CommonDeclarationSiteTypeVariances rewrite-static-analysis#1131: CommonDeclarationSiteTypeVariances excludeBounds → excludedBounds
• Drop Maven-only changeManagedDependency from Gradle ChangeDependencyClassifier rewrite-migrate-java#1363: UpdateApacheShiroDependencies drops the Maven-only changeManagedDependency from the Gradle recipe
• moderneinc/rewrite-hibernate#80: MigrateToHibernate70 AddDependency newVersion → version
• moderneinc/rewrite-spring#504: MigrateToJsf_2_3 drops fileMatcher from ChangeTagAttribute, which has no file filter

In each, the recipe silently ignored the misspelled option. The github-actions, rewrite-spring, static-analysis and hibernate ones were actual behaviour bugs.

Verification

• ./gradlew :rewrite-core:test :rewrite-test:test :rewrite-yaml:test passes, as do the inline-YAML recipe tests in rewrite-gradle, hcl, java-test, json, kotlin and maven. AppendToSequenceTest.appendTwice still passed fileMatcher, removed from AppendToSequence in Improved support for large repositories #3118, so I dropped that key.
• I ran validateAll() over 842 declarative recipes: this repo's YAML plus 23 published recipe modules (spring, migrate-java, static-analysis, testing-frameworks, etc.). The only new failures were the four openrewrite ones above. With those fixes swapped into the jars, the failure count matches the unpatched baseline. Load time is unchanged within noise.
• I ran the same check over the 1284 declarative recipes reachable from moderne-recipe-bom, covering every published io.moderne.recipe module plus the openrewrite modules it manages. The baseline had 0 failures. After this change, the only findings are the two moderneinc PRs above.

[timtebeek]

It seems strong to fail to load if we had ignored this up to now; should we introduce a grace period where we only validate this in tests? Or would that risk too many untested recipes continuing to ignore invalid options?

[timtebeek]

As discussed on Slack: we've already cleared out the issues on openrewrite/; now doing the same for moderneinc/; those downstream PRs can land already; we'll hold off on this one until after the next release is out.

[timtebeek]

With the 8.94.0 release now out we'll merge this one to avoid further potential issues in recipes. There's one more fix coming for rewrite-spring-to-quarkus.