Fix kotlin extractor graph pass july 3rd 26 (#64)

* Fix Kotlin dead-code false positives: emit method-kind for members, receiver/property/callable-reference/outside-body call edges, and Groovy-namespace cross-module imports in the extractor, and exclude framework/DI/override/androidTest entry points in the orphan detector

* Kotlin/Java: make analyze_performance perf-facts precise — arity- and super-aware recursion detection, RxJava/Flow chains no longer counted as loops, and Retrofit/Room methods tagged performs_io (cacheVersion v57).

* fix(package-metrics): eliminate false positives on multi-module Kotlin/Java by resolving cross-module imports via a declared-package index, excluding test-support/DI/enum symbols from the population, and reclassifying data-holder packages instead of flagging them rigid

* Adding missing files

* Fixing lint

Author: dejo1307

internal/engine/cache.go

@@ -65,7 +65,42 @@ import (
 // v49: Swift models XcodeGen test-bundle targets (bundle.unit-test/bundle.ui-testing) as one module each, so a test bundle's files (e.g. Tests/Core/**) collapse into a single module instead of exploding into per-leaf-directory modules; and every module fact now carries a normalized `module_role` prop (production/test/tooling/unknown) — derived from the XcodeGen target type, the SPM target vs testTarget call, or a path heuristic for leaf-directory fallback — so package-metrics and other analyses can measure the production population without re-parsing manifests.
 // v50: the `module_role` prop is now emitted by the Ruby extractor too (packwerk packages → production; leaf-directory modules → path heuristic), and the path heuristic was hoisted to facts.ModuleRoleForPath and broadened to common cross-language conventions (spec/test/tests + scripts/bin/fastlane/ci_scripts), so Ruby build-tooling modules (fastlane/, Scripts/) are classified as tooling rather than defaulting to the production population.
 // v51: Swift no longer emits type-reference-derived module→module dependency edges for files that belong to a resolved SPM/XcodeGen target — those files' cross-module deps are captured completely by their `import X` statements plus the declared target graph, whereas the type-reference pass resolved bare short names through a collision-prone global index (Swift namespaces nested types, so names like Event/State/Coordinator recur across targets) and fabricated impossible back-edges (a Foundation-level target "importing" a feature target) that produced a false module cycle. For loose Swift projects (leaf-directory fallback, no target graph) the pass still runs but now skips any type name defined in more than one module. Fixes the false Swift dependency cycle.
-const cacheVersion = "v51"
+// v52: Kotlin emits SymbolMethod (not SymbolFunc) for functions declared inside a class/object (parity with Go/Java), keeping member functions out of the high-confidence orphan bucket; records a short-name RelCalls edge for every navigation expression (receiver method call `repo.getUser()` and property/field access `slot.uniqueId`), which the short-name-matching dead-code detector needs to see live members as used; and tags `override` methods and Dagger/Hilt `@Provides`/`@Binds` methods with `override`/`di_provider` props so framework/DI entry points are excluded from orphan reporting. Fixes the mass Kotlin/Android dead-code false positives (thousands of live Retrofit/lifecycle/interface members reported as high-confidence orphans).
+// v53: Kotlin base-package detection now matches Groovy build scripts (`namespace 'x'`, single quotes) in addition to Kotlin-DSL (`namespace = "x"`). A double-quote-only regex left the base package empty for `.gradle` (Groovy) projects, so every in-repo import resolved as external and bare calls to imported top-level/extension functions (`formatPrettyDate(...)`, `setMargin(...)`) emitted no call edge — reporting live utilities as high-confidence orphans.
+// v54: Kotlin now walks calls that live OUTSIDE a function body — function default-parameter values (`fun f(x = helper())`) and supertype constructor-delegation arguments (`class NpeId(id) : Enrichment(npeEntity(id))`, and the object equivalent). These were previously skipped, so a helper/factory referenced only from a default value or a base-class initializer was mis-reported as a high-confidence orphan (e.g. Snowplow entity builders, Compose default-arg providers).
+// v55: Kotlin captures callable references (`::foo`, `Type::foo`, e.g. `onClick = ::doNothing`, `.map(::helper)`) as short-name RelCalls edges. A function referenced only as a method reference (never called directly) was previously mis-reported as a high-confidence orphan.
+// v56: Kotlin/Java perf-fact precision — (1) recursion (`recursive_self`) is now argument-count aware: a call sharing the enclosing function's name is only flagged recursion when its arg count matches the parameter count, so a call to a same-named overload (`updateItem(x)` → `updateItem(i, x)`, `onChangeStarted(2)` override → `onChangeStarted(3)`) is no longer read as self-recursion — the dominant Kotlin/Android recursion false positive (Conductor `onChange*` lifecycle). (2) Kotlin RxJava/coroutine-Flow chains no longer inflate loop_depth: in a reactive function (reactive return type Single/Observable/Maybe/Flowable/Completable/Flow, or body reactive operators subscribeOn/observeOn/applySchedulers/andThen/.subscribe/flowOn/.collect/…), the ambiguous operators (map/flatMap/filter/fold/reduce/onEach) are stream transforms, not per-element collection loops, so a `Single.flatMap { … .map { } }` is no longer a false O(n²)/O(n³). Fixes the analyze_performance false positives on this RxJava-heavy codebase.
+// v57: Kotlin/Java recursion also clears the arity-matched case where an `override` delegates to a same-name, same-arity overload declared in a parent (invisible here): a body that calls `super.<self>()` marks the sibling `<self>(…)` call as delegation, not recursion (fixes the residual Conductor `onChangeEnded` false positives). And Kotlin methods now carry `io_direct`/`performs_io` when annotated as a Retrofit endpoint (@GET/@POST/@PUT/@DELETE/@PATCH/@HEAD/@OPTIONS/@HTTP) or a Room DAO op (@Query/@Insert/@Update/@Upsert/@Delete/@RawQuery) — a precise per-method I/O identity that lets analyze_performance flag a per-iteration call to a real network/DB method as a genuine N+1 (ranked high) without relying on the cross-language keyword guess.
+// v58: Kotlin/Java multi-module import resolution — imports are now resolved via a
+// cross-language declared-package→directory index (built from every non-test .kt AND
+// .java file) instead of assuming a single global source root. In a multi-module
+// Gradle project where several modules root packages at the same prefix (app/, api/,
+// business/ all under de.foo.*), the old Kotlin resolver mapped every internal import
+// under the single most common source root (the app module), collapsing all
+// cross-module afferent coupling onto the app package (bogus Ca god-package) and
+// starving library modules of Ca (falsely "useless" in package-metrics). Now an
+// import resolves to the module that actually declares the package. The Java extractor
+// seeds its FQN resolver with the same cross-language index so Java→Kotlin imports no
+// longer drop. Kotlin & Java module facts now carry `module_role` (test for
+// src/test & src/androidTest, else production) so package-metrics excludes test source
+// sets; and Kotlin `sealed` classes are marked `abstract` so abstractness (A) counts
+// them (they are non-instantiable), fixing inflated Distance / false "rigid" findings.
+// v59: the package index now prefers a main source set (…/src/main/…) over a Gradle
+// build-variant source set (src/debug, src/release, src/staging) when both declare the
+// same package. An Android app's src/main and src/debug both declare the root
+// application package; the v58 lexicographic tie-break wrongly mapped it to src/debug
+// ('d' < 'm'), misrouting the whole app's afferent coupling onto the debug variant
+// (a bogus Ca god-package). Imports of the root package now resolve to the main module.
+// v60: package-metrics precision — (1) module_role now sub-token-matches compound test
+// module names (split each path segment on -/_ and match an exact `test`/`tests` token),
+// so Gradle test-automation modules that compile as src/main (release-tests, ui-test-utils,
+// test-lab) are classified test rather than leaking into the production population, without
+// misfiring on latest/contest/abtest. (2) Dagger/Hilt DI infrastructure is now tagged:
+// @Component/@Subcomponent interfaces get `di_component`, @Module classes get `di_module`
+// (Java & Kotlin); a Dagger @Component interface is no longer mislabeled a Spring component
+// (disambiguated by interface-vs-class). Lets package-metrics exclude DI wiring from
+// abstractness/type counts (a Dagger component package was falsely "useless").
+const cacheVersion = "v60"
 
 // extractorCache holds per-extractor facts keyed by a content hash of the files
 // the extractor depends on. It is loaded from disk at the start of a snapshot and
@@ -169,15 +204,27 @@ func computeExtractorKeys(all []extractors.Extractor, files []string, hashes map
 	}
 
 	// Partition files: per-owner owned lists + the shared (un-owned) remainder.
+	// keyFiles is owned ∪ AffectsKey — the full set whose contents feed the key,
+	// so a cross-language file that a KeyDependent extractor reads (but does not
+	// own) still invalidates its cache. Ownership (and thus the shared remainder)
+	// is decided purely by OwnsFile; AffectsKey only widens the key, never the
+	// ownership partition.
 	owned := map[string][]string{}
+	keyFiles := map[string][]string{}
 	var shared []string
 	for _, f := range files {
 		ownedByAny := false
 		for name, fo := range owners {
-			if fo.OwnsFile(f) {
+			owns := fo.OwnsFile(f)
+			if owns {
 				owned[name] = append(owned[name], f)
 				ownedByAny = true
 			}
+			if owns {
+				keyFiles[name] = append(keyFiles[name], f)
+			} else if kd, ok := fo.(plugin.KeyDependent); ok && kd.AffectsKey(f) {
+				keyFiles[name] = append(keyFiles[name], f)
+			}
 		}
 		if !ownedByAny {
 			shared = append(shared, f)
@@ -189,7 +236,7 @@ func computeExtractorKeys(all []extractors.Extractor, files []string, hashes map
 	for name := range owners {
 		h := sha256.New()
 		h.Write([]byte(cacheVersion + "\x00" + name + "\x00" + sharedHash + "\x00"))
-		h.Write([]byte(hashFileSet(owned[name], hashes)))
+		h.Write([]byte(hashFileSet(keyFiles[name], hashes)))
 		keys[name] = hex.EncodeToString(h.Sum(nil))
 	}
 	return keys

internal/extractors/javaextractor/java.go

@@ -7,6 +7,7 @@ import (
 	"path/filepath"
 	"strings"
 
+	"github.com/enola-labs/enola/internal/extractors/jvmsrc"
 	"github.com/enola-labs/enola/internal/facts"
 	"github.com/enola-labs/enola/internal/parallel"
 )
@@ -72,7 +73,12 @@ func (e *JavaExtractor) Extract(ctx context.Context, repoPath string, files []st
 		modules[filepath.Dir(javaFiles[i])] = true
 	}
 
-	canonicalizeTargets(allFacts)
+	// Cross-language package index (.kt AND .java) so a Java import of a Kotlin
+	// type resolves to the module that declares it, instead of being dropped as
+	// external. Java→Java imports still resolve via the in-facts FQN index below;
+	// this only fills the cross-language gap.
+	packageIndex := jvmsrc.BuildPackageIndex(repoPath, files)
+	canonicalizeTargets(allFacts, packageIndex)
 	resolveTableConstants(allFacts)
 
 	for dir := range modules {
@@ -81,7 +87,8 @@ func (e *JavaExtractor) Extract(ctx context.Context, repoPath string, files []st
 			Name: dir,
 			File: dir,
 			Props: map[string]any{
-				"language": "java",
+				"language":           "java",
+				facts.PropModuleRole: jvmsrc.ModuleRole(dir),
 			},
 		})
 	}
@@ -97,10 +104,16 @@ func (e *JavaExtractor) Extract(ctx context.Context, repoPath string, files []st
 //   - import dependency facts whose target FQN resolves to a declared type — or whose
 //     value names a known source package — are marked source="internal" and pointed at
 //     the owning module dir.
-func canonicalizeTargets(allFacts []facts.Fact) {
+func canonicalizeTargets(allFacts []facts.Fact, crossLangIndex map[string]string) {
 	typeIndex := make(map[string]string) // FQN -> "<dir>.<Type>" canonical name
 	typeDir := make(map[string]string)   // FQN -> dir
 	packageDir := make(map[string]string)
+	// Seed with cross-language packages (e.g. Kotlin modules) so a Java import of
+	// a package we didn't index from .java files still resolves. Java-declared
+	// packages below take precedence (they overwrite these entries).
+	for pkg, dir := range crossLangIndex {
+		packageDir[pkg] = dir
+	}
 	for _, f := range allFacts {
 		if f.Kind != facts.KindSymbol {
 			continue
@@ -240,6 +253,13 @@ func isJavaFile(path string) bool {
 // OwnsFile implements plugin.FileOwner for incremental caching.
 func (e *JavaExtractor) OwnsFile(relFile string) bool { return isJavaFile(relFile) }
 
+// AffectsKey implements plugin.KeyDependent: a .kt file's package declaration
+// feeds the cross-language package index used to resolve Java imports of Kotlin
+// types, so a change to any Kotlin source must invalidate the Java extractor's cache.
+func (e *JavaExtractor) AffectsKey(relFile string) bool {
+	return strings.HasSuffix(strings.ToLower(relFile), ".kt")
+}
+
 // containsJavaSource reports whether any .java file exists under root within
 // maxDepth directory levels. It returns on the first match and skips hidden and
 // common build/dependency directories so it stays cheap on large repos.

internal/extractors/javaextractor/java_ast.go

@@ -80,6 +80,10 @@ type astWalker struct {
 	loopDepth int
 	selfName  string
 	selfShort string
+	// selfParams is the enclosing method's declared parameter count. A resolved
+	// self-call is only genuine recursion when its argument count matches — otherwise
+	// it is a call to a same-named overload, not recursion.
+	selfParams int
 }
 
 // javaBodyMetrics accumulates per-method complexity signals during the single
@@ -88,9 +92,10 @@ type javaBodyMetrics struct {
 	loopDepth   int             // max loop nesting depth
 	loopCount   int             // number of loop constructs (syntactic + stream lambdas)
 	decisions   int             // decision points (cyclomatic = 1 + decisions)
-	callsInLoop []string        // distinct call targets invoked at loop depth >= 1
-	inLoopSeen  map[string]bool // dedup set for callsInLoop
-	recursive   bool            // body directly calls the enclosing method
+	callsInLoop  []string        // distinct call targets invoked at loop depth >= 1
+	inLoopSeen   map[string]bool // dedup set for callsInLoop
+	recursive    bool            // body directly calls the enclosing method
+	sawSuperSelf bool            // body calls super.<enclosingName>() (override delegation)
 }
 
 // javaIterators are Stream/Collection methods whose lambda argument runs once per
@@ -119,16 +124,53 @@ var javaCheapMethods = map[string]bool{
 
 // recordCallMetrics notes a resolved call target against the current method's
 // complexity metrics: flags direct recursion and records calls made inside loops.
-func (w *astWalker) recordCallMetrics(target string) {
+// argCount is the invocation's argument count; recursion is flagged only when it
+// matches the enclosing method's parameter count, so a call to a same-named overload
+// is not mistaken for self-recursion.
+func (w *astWalker) recordCallMetrics(target string, argCount int) {
 	if w.metrics == nil || target == "" {
 		return
 	}
-	if target == w.selfName || target == w.selfShort {
+	if (target == w.selfName || target == w.selfShort) && argCount == w.selfParams {
 		w.metrics.recursive = true
 	}
 	w.recordInLoop(target)
 }
 
+// javaArgCount returns the number of arguments of a method_invocation node.
+func javaArgCount(node *sitter.Node) int {
+	args := node.ChildByFieldName("arguments")
+	if args == nil {
+		return 0
+	}
+	n := 0
+	for i := uint(0); i < uint(args.ChildCount()); i++ {
+		if args.Child(i).IsNamed() {
+			n++
+		}
+	}
+	return n
+}
+
+// javaParamCount returns the declared parameter count of a method declaration node.
+func javaParamCount(node *sitter.Node) int {
+	params := node.ChildByFieldName("parameters")
+	if params == nil {
+		params = findChildByKind(node, "formal_parameters")
+	}
+	if params == nil {
+		return 0
+	}
+	n := 0
+	for i := uint(0); i < uint(params.ChildCount()); i++ {
+		switch params.Child(i).Kind() {
+		case "formal_parameter", "spread_parameter":
+			n++
+		}
+	}
+	return n
+}
+
 // recordInLoop adds a target to calls_in_loop (deduped) when inside a loop, without
 // the recursion check — used for raw instance-method names.
 func (w *astWalker) recordInLoop(target string) {
@@ -525,10 +567,12 @@ func (w *astWalker) handleMethod(node *sitter.Node) {
 	// may be invalidated if the body walk grows w.out).
 	savedMetrics, savedDepth := w.metrics, w.loopDepth
 	savedName, savedShort := w.selfName, w.selfShort
+	savedParams := w.selfParams
 	w.metrics = &javaBodyMetrics{}
 	w.loopDepth = 0
 	w.selfName = f.Name
 	w.selfShort = name
+	w.selfParams = javaParamCount(node)
 	if body := node.ChildByFieldName("body"); body != nil {
 		w.walkForCalls(body)
 	}
@@ -544,11 +588,14 @@ func (w *astWalker) handleMethod(node *sitter.Node) {
 	if len(m.callsInLoop) > 0 {
 		props["calls_in_loop"] = m.callsInLoop
 	}
-	if m.recursive {
+	// A body that calls super.<self>() is an override delegating to a same-named
+	// overload, not genuine recursion — clear the arity-matched self-call flag.
+	if m.recursive && !m.sawSuperSelf {
 		props["recursive_self"] = true
 	}
 	w.metrics, w.loopDepth = savedMetrics, savedDepth
 	w.selfName, w.selfShort = savedName, savedShort
+	w.selfParams = savedParams
 	w.popOwner()
 }
 
@@ -807,14 +854,19 @@ func (w *astWalker) handleInvocation(node *sitter.Node) {
 	// own methods. Calls on other receivers are left unresolved (the receiver's
 	// type is not tracked), matching the Kotlin extractor's conservative model.
 	isThis := obj != nil && nodeText(obj, w.src) == "this"
+	if obj != nil && w.metrics != nil && name == w.selfShort && nodeText(obj, w.src) == "super" {
+		// super.<self>() — an override delegating to its supertype. Note it so an
+		// arity-matched bare <self>(…) call is read as overload delegation, not recursion.
+		w.metrics.sawSuperSelf = true
+	}
 	if obj == nil || isThis {
 		if methods := w.currentMethods(); methods[name] {
 			target := w.dir + "." + w.enclosingType() + "." + name
 			owner.Relations = append(owner.Relations, facts.Relation{
 				Kind:   facts.RelCalls,
 				Target: target,
 			})
-			w.recordCallMetrics(target)
+			w.recordCallMetrics(target, javaArgCount(node))
 		}
 	} else if w.metrics != nil && w.loopDepth > 0 && !javaCheapMethods[name] {
 		// Method call on a non-this receiver inside a loop (repo.findById(), …). No

internal/extractors/javaextractor/spring.go

@@ -115,6 +115,21 @@ func findAnnotation(anns []javaAnnotation, name string) *javaAnnotation {
 // classifyComponent tags a type symbol fact with framework/component props for
 // Spring stereotypes and Dubbo SPI. It mutates f.Props in place.
 func classifyComponent(f *facts.Fact, name string, annotations []javaAnnotation, supertypes []string) {
+	// Dagger/Hilt DI infrastructure. Dagger @Component/@Subcomponent are declared
+	// on INTERFACES, whereas Spring stereotypes are always concrete classes — so a
+	// @Component on an interface is Dagger, not Spring. This disambiguates the
+	// simple-name collision between dagger.Component and springframework…Component
+	// (annotations are matched by simple name) and keeps DI wiring out of the
+	// domain-architecture metrics. @Module classes are DI wiring regardless of kind.
+	if hasAnnotation(annotations, "Module") {
+		f.Props["di_module"] = true
+	}
+	if f.Props["symbol_kind"] == facts.SymbolInterface &&
+		(hasAnnotation(annotations, "Component") || hasAnnotation(annotations, "Subcomponent")) {
+		f.Props["di_component"] = true
+		return // do NOT fall through to the Spring stereotype switch (avoids mislabel)
+	}
+
 	switch {
 	case hasAnnotation(annotations, "RestController"):
 		f.Props["framework"] = "spring"