Fix ts extractor graph pass july 3rd 26

ARCHITECTURE.md

@@ -149,7 +149,7 @@ Repository
 
 Stage by stage:
 
-1. **File walker** — enumerates files under the repo, applying the `ignore` globs from config so build output, vendored code, tests, and generated files never reach the parsers. A few extractors (OpenAPI, and PHP's Symfony route config) deliberately scan specific config-format files (YAML/JSON) directly from disk, bypassing these globs — because the globs exist to suppress config/data noise, not to hide those architecturally meaningful files (see [Supported languages](#supported-languages)).
+1. **File walker** — enumerates files under the repo, applying the `ignore` globs from config so build output, vendored code, tests, and generated files never reach the parsers. Beyond the path globs, the TypeScript extractor additionally detects **minified/bundled** files by content (any line longer than ~2000 chars) and skips them, so a hash-named vendor bundle checked in outside a build directory (e.g. a minified third-party bundle served from a static assets dir) does not pollute the fact graph with obfuscated symbols and spurious complexity/hotspot findings. A few extractors (OpenAPI, and PHP's Symfony route config) deliberately scan specific config-format files (YAML/JSON) directly from disk, bypassing these globs — because the globs exist to suppress config/data noise, not to hide those architecturally meaningful files (see [Supported languages](#supported-languages)).
 2. **Extractors** — each enabled language extractor first *detects* whether it applies (e.g. Go runs when there's a `go.mod`), then *parses* the matching files and emits facts. This is pure parsing; see [Supported languages](#supported-languages) for what each one understands.
 3. **Fact store** — facts land in an in-memory store ([`internal/facts/store.go`](internal/facts/store.go)) indexed by kind, file, name, and repo for fast queries. In append mode, facts are tagged with a repo label and file paths are repo-prefixed.
 4. **Cross-repo linker** — only when two or more repos are loaded. It connects the per-repo graphs by matching HTTP client/server routes and shared-library imports, emitting `service` nodes and cross-repo dependency edges. The link set is recomputed from scratch on every append, so it always reflects exactly the repos currently loaded.
@@ -478,7 +478,7 @@ The bundled [`mcp-arch.yaml`](mcp-arch.yaml) ships a much fuller `ignore` list (
 | Field | Description | Default |
 |-------|-------------|---------|
 | `repo` | Repository root path | `"."` |
-| `ignore` | Glob patterns for files/dirs to skip | vendor, node_modules, .git, tests, build dirs, docs, config data, … |
+| `ignore` | Glob patterns for files/dirs to skip | vendor, node_modules, .git, tests, build dirs, minified JS (`*.min.js`/`*.bundle.js`), docs, config data, … |
 | `extractors` | Enabled extractors | `["cpp", "go", "java", "kotlin", "openapi", "python", "typescript", "swift", "ruby"]` |
 | `explainers` | Enabled explainers | `["cycles", "layers", "crossrepo", "coverage", "unused-routes", "god-class", "hotspots", "dependency-depth", "exported-surface", "complexity-outliers"]` |
 | `renderers` | Enabled renderers | `["llm_context"]` |
@@ -510,11 +510,11 @@ Each extractor is detected by characteristic project files and then parses what
 
 **Go** uses the standard-library parser directly, so symbols, methods, interfaces, imports, and call edges are exact.
 
-**TypeScript** (tree-sitter) includes Next.js route detection (App Router and Pages Router), monorepo detection one level deep, and parsing of `openapi-typescript`-generated client files — each operation is emitted as a `route` fact with `role:"client"`. App Router route groups like `(standard)` are stripped from URLs.
+**TypeScript** (tree-sitter) includes Next.js route detection (App Router and Pages Router), monorepo detection one level deep, and parsing of `openapi-typescript`-generated client files — each operation is emitted as a `route` fact with `role:"client"`. App Router route groups like `(standard)` are stripped from URLs. Like the other AST extractors it walks function/method bodies for the complexity metrics (`cyclomatic`, `loop_depth`, `loop_count`, `calls_in_loop`, `recursive_self`) the enterprise `analyze_performance` tool consumes, and — mirroring Swift — it tags a body `io_direct` when it directly invokes a network/file primitive (`fetch`, `axios`, `fs.readFile`, `navigator.sendBeacon`, `new WebSocket`/`XMLHttpRequest`/`EventSource`) **or** calls a binding imported from a network module (a known HTTP-client package, or any path with a `network` segment — e.g. a `request` helper from a `.../lib/network/request` module). A serial post-pass (`computeTSPerformsIO`) then propagates that flag transitively over the `calls` graph into a `performs_io` prop via a cycle-safe monotone fixpoint, so a function reaching the network only through wrapper helpers is still flagged — letting the analyzer catch a per-iteration network call (an N+1) hidden behind a wrapper. *Limitation:* default-imported internal wrappers aren't resolved by the call-edge pass, so the fixpoint does not cross that hop; in practice most wrappers call their I/O sink directly (so they are seeded `io_direct` without needing the edge), and the analyzer's short-name I/O index still matches the in-loop call.
 
 **Vue** support is integrated within the TypeScript extractor. `.vue` Single File Components are handled natively — the extractor parses each SFC's `<script>` and `<script setup>` blocks (case-insensitive, with `lang` attribute detection for TypeScript vs JavaScript) and feeds them through the existing tree-sitter TypeScript pipeline. Detection checks for a `"vue"` dependency in `package.json` (TypeScript root first, then repo root fallback); **Nuxt** is additionally detected by `nuxt.config.js/ts/mjs` or a `"nuxt"` package dependency. Each `.vue` file emits a `symbol` fact for the component, named `<dir>.<ComponentName>` (kebab-case converted to PascalCase), carrying `web_component: "component"`, `framework: "vue"` or `"nuxt"`, and `vue_setup: true` when using `<script setup>`. Functions named `use*` anywhere in the project are automatically classified as composables (`web_component: "composable"`). **Nuxt file-based routing** emits one `route` fact per file under `pages/`, with the URL derived from the file path — index files resolve to `/`, dynamic segments like `[id].vue` are preserved — each with `method: "GET"` and `router: "pages"`. Files containing a `createRouter()` call are emitted as a route fact with `type: "router_config"`. Import statements in all script blocks become `dependency` facts, and call edges from method bodies participate in `traverse`, `find_path`, and `impact_analysis`. Vue detection runs automatically inside the `typescript` extractor — no separate entry is needed under `extractors:` in config.
 
-**JavaScript** (`.js`/`.jsx`) is handled by the TypeScript extractor. Tree-sitter's TypeScript parser natively parses JavaScript (JS is a subset of TS), so all extraction features — imports, declarations, call graphs, JSX component detection — work identically for `.js` and `.jsx` files. No separate configuration is needed; any project detected by the TypeScript extractor will have its JavaScript files processed automatically alongside TypeScript files.
+**JavaScript** (`.js`/`.jsx`) is handled by the TypeScript extractor. Tree-sitter's TypeScript parser natively parses JavaScript (JS is a subset of TS), so all extraction features — imports, declarations, call graphs, JSX component detection — work identically for `.js` and `.jsx` files. No separate configuration is needed; any project detected by the TypeScript extractor will have its JavaScript files processed automatically alongside TypeScript files. **Minified/bundled JS is skipped**: before parsing, any file with a line longer than ~2000 characters is treated as a generated artifact and produces no facts (a directory containing only such files emits no module either), so checked-in vendor bundles do not distort the graph or the enterprise complexity/performance analyses.
 
 **Svelte** support is integrated within the TypeScript extractor, following the same pattern as Vue. `.svelte` Single File Components are parsed by extracting `<script>` and `<script module>` blocks (Svelte 5 syntax; the older `<script context="module">` form from Svelte 4 is also supported) and feeding them through tree-sitter. Detection checks for a `"svelte"` dependency in `package.json`; **SvelteKit** is additionally detected by `svelte.config.js/ts/mjs` or a `"@sveltejs/kit"` package dependency. Each `.svelte` file emits a component fact with `web_component: "component"`, `framework: "svelte"` or `"sveltekit"`. **SvelteKit file-based routing** emits route facts for `+page.svelte`, `+layout.svelte`, `+error.svelte`, and `+server.ts` files under `src/routes/` — route groups in parentheses like `(groupName)` are stripped from URLs, dynamic segments like `[slug]` and catch-all `[...rest]` are preserved. Server-side load files (`+page.server.ts`, `+layout.server.ts`) are not emitted as routes. The SvelteKit `$lib` path alias is automatically resolved to `src/lib/`, ensuring imports like `$lib/utils` appear as internal dependency edges rather than unresolved externals.
 

internal/config/config.go

@@ -68,6 +68,12 @@ func Default() *Config {
 			"__pycache__/**",
 			"**/Pods/**",
 			"**/.gradle/**",
+			// Minified / bundled JS by name. The extractor also detects minified
+			// content heuristically (very long lines), but these globs cheaply skip
+			// the common named cases before a file is ever read. Keep in sync with
+			// the bundled mcp-arch.yaml ignore list.
+			"**/*.min.js",
+			"**/*.bundle.js",
 		},
 		// TestGlobs identify test/spec files. They stay ignored for normal indexing
 		// (still listed in Ignore above) — production architecture facts must not

internal/engine/cache.go

@@ -100,7 +100,36 @@ import (
 // (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"
+// v61: TS/JS extractor now emits file-scope reference facts (KindFileRef) for JSX
+// component rendering (<Foo/>), imported-identifier values (route configs like
+// `{ component: Foo }`), namespace member access (`ns.foo`), require()-bound names,
+// and `export … from` re-exports — plus require()/dynamic-import() dependency edges.
+// Fixes massive dead-code false positives on React/CommonJS codebases, where a
+// component used only via JSX or a route table previously had no incoming edge.
+// v62: the TS/JS file-scope reference pass now also records same-module use
+// positions — a bare call callee and identifier call arguments — so a function used
+// only at module scope (`startSession()` at file top level) or passed as a value to
+// an HOC (`connect(mapStateToProps)`) is no longer falsely reported dead.
+// v63: a default import now also references the target module's default-export symbol
+// (resolved via the known-files set + fileSymbolName), so an anonymous folder-index
+// default like `export default connect(...)(X)` — named "<Folder>Index" — is no longer
+// falsely reported dead when imported by the component's own name.
+// v64: a `this.<member>` reference inside a class method now records a use of that
+// member, so a React class-component event handler bound as a prop value
+// (onClick={this.handleClick}) — never called by name — is no longer falsely dead.
+// v65: the TypeScript extractor now skips minified/bundled files (any line longer
+// than ~2000 chars), so checked-in vendor bundles emit no facts — invalidates
+// caches that still hold the obfuscated symbols.
+// v66: the TypeScript extractor now emits io_direct (body calls a network/file
+// primitive or a network-module import binding) and a transitively-propagated
+// performs_io prop, so cached TS facts must be re-extracted to carry them.
+// v67: tightened TS io_direct — only DEFAULT/NAMESPACE network-module imports are
+// I/O bindings (not named imports), and types/utils submodules are excluded, so pure
+// helpers (e.g. `resolved` from network/types) no longer mislabel callers.
+// v68: the TypeScript extractor now sets abstract:true on `abstract class`
+// declarations (was previously indistinguishable from a concrete class), so
+// package-metrics abstractness for TS must re-extract to pick up the flag.
+const cacheVersion = "v68"
 
 // 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

internal/engine/testdata/golden/ts_sample.facts.jsonl

@@ -1,9 +1,11 @@
 {"kind":"dependency","name":"src -\u003e src/repo","file":"src/index.ts","line":1,"repo":"ts_sample","props":{"language":"typescript","source":"internal"},"relations":[{"kind":"imports","target":"src/repo"}]}
 {"kind":"dependency","name":"src -\u003e src/repo","file":"src/svc.ts","line":1,"repo":"ts_sample","props":{"language":"typescript","source":"internal"},"relations":[{"kind":"imports","target":"src/repo"}]}
 {"kind":"dependency","name":"src -\u003e src/svc","file":"src/index.ts","line":2,"repo":"ts_sample","props":{"language":"typescript","source":"internal"},"relations":[{"kind":"imports","target":"src/svc"}]}
+{"kind":"file_ref","name":"src/index.ts","file":"src/index.ts","line":1,"repo":"ts_sample","props":{"language":"typescript"},"relations":[{"kind":"calls","target":"src.Repo"},{"kind":"calls","target":"src.Service"}]}
+{"kind":"file_ref","name":"src/svc.ts","file":"src/svc.ts","line":1,"repo":"ts_sample","props":{"language":"typescript"},"relations":[{"kind":"calls","target":"src.Repo"}]}
 {"kind":"module","name":"src","file":"src","repo":"ts_sample","props":{"language":"typescript"}}
 {"kind":"symbol","name":"src.Repo","file":"src/repo.ts","line":2,"repo":"ts_sample","props":{"exported":true,"language":"typescript","symbol_kind":"class"},"relations":[{"kind":"declares","target":"src"}]}
 {"kind":"symbol","name":"src.Repo.all","file":"src/repo.ts","line":3,"repo":"ts_sample","props":{"cyclomatic":1,"exported":true,"language":"typescript","receiver":"Repo","symbol_kind":"method"},"relations":[{"kind":"declares","target":"src"}]}
 {"kind":"symbol","name":"src.Service","file":"src/svc.ts","line":4,"repo":"ts_sample","props":{"exported":true,"language":"typescript","symbol_kind":"class"},"relations":[{"kind":"declares","target":"src"}]}
-{"kind":"symbol","name":"src.Service.list","file":"src/svc.ts","line":7,"repo":"ts_sample","props":{"cyclomatic":1,"exported":true,"language":"typescript","receiver":"Service","symbol_kind":"method"},"relations":[{"kind":"declares","target":"src"}]}
+{"kind":"symbol","name":"src.Service.list","file":"src/svc.ts","line":7,"repo":"ts_sample","props":{"cyclomatic":1,"exported":true,"language":"typescript","receiver":"Service","symbol_kind":"method"},"relations":[{"kind":"calls","target":"src.Service.repo"},{"kind":"declares","target":"src"}]}
 {"kind":"symbol","name":"src.main","file":"src/index.ts","line":5,"repo":"ts_sample","props":{"cyclomatic":1,"exported":true,"language":"typescript","symbol_kind":"function"},"relations":[{"kind":"declares","target":"src"}]}

internal/extractors/tsextractor/minified_test.go

@@ -0,0 +1,56 @@
+package tsextractor
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/enola-labs/enola/internal/facts"
+)
+
+func TestIsMinifiedSource(t *testing.T) {
+	longLine := "var x = \"" + strings.Repeat("z", minifiedLineThreshold+100) + "\";"
+	if !isMinifiedSource([]byte(longLine)) {
+		t.Errorf("isMinifiedSource(one very long line) = false, want true")
+	}
+	// A build banner on line 1 followed by a huge minified chunk (the common shape).
+	bundle := "/* Build */\n" + longLine + "\nfunction f(){}\n"
+	if !isMinifiedSource([]byte(bundle)) {
+		t.Errorf("isMinifiedSource(banner + long line) = false, want true")
+	}
+
+	ordinary := "export function add(a, b) {\n  return a + b;\n}\n"
+	if isMinifiedSource([]byte(ordinary)) {
+		t.Errorf("isMinifiedSource(ordinary source) = true, want false")
+	}
+	// Many short lines, none over the threshold, even if the file is large overall.
+	manyLines := strings.Repeat("const x = compute();\n", 500)
+	if isMinifiedSource([]byte(manyLines)) {
+		t.Errorf("isMinifiedSource(many short lines) = true, want false")
+	}
+}
+
+func TestExtract_SkipsMinifiedBundle(t *testing.T) {
+	longLine := "var bundledLibrary = \"" + strings.Repeat("z", minifiedLineThreshold+100) + "\";"
+	files := map[string]string{
+		"src/util.ts":             "export function realHelper() {\n  return 1;\n}\n",
+		"assets/vendor/bundle.js": longLine,
+	}
+	got := extractAll(t, files, false)
+
+	// The hand-written symbol is extracted.
+	if _, ok := findFact(got, "src.realHelper"); !ok {
+		t.Errorf("expected symbol fact for src.realHelper; got %+v", got)
+	}
+	// The minified bundle contributes no facts at all — no symbols and no module
+	// fact for its directory.
+	for _, f := range got {
+		if strings.Contains(f.File, "assets/vendor") || strings.Contains(f.Name, "assets/vendor") {
+			t.Errorf("minified bundle produced a fact it should have been skipped: %+v", f)
+		}
+	}
+	for _, m := range findFactsByKind(got, facts.KindModule) {
+		if m.Name == "assets/vendor" {
+			t.Errorf("minified-only directory should not emit a module fact; got %+v", m)
+		}
+	}
+}