@massu/core 1.10.7 → 1.12.0

package/src/detect/adapters/types.ts

@@ -1,191 +0,0 @@
-// Copyright (c) 2026 Massu. All rights reserved.
-// Licensed under BSL 1.1 - see LICENSE file for details.
-
-/**
- * Plan 3b — Phase 1: AST Adapter contract types.
- *
- * Lives at `packages/core/src/detect/adapters/types.ts`. All types are local —
- * NONE re-exported from `web-tree-sitter`.
- *
- * Adapter authors import from this module only; the runner (`runner.ts`)
- * orchestrates execution and the loader (`tree-sitter-loader.ts`) handles
- * grammar acquisition.
- *
- * Per-field confidence is enforced (NOT per-adapter): a single weak field
- * MUST NOT poison the rest. The runner consumes `confidence` per-adapter for
- * the moment, but the merge rule reads each `conventions[field]` against the
- * provenance trail to decide what survives.
- */
-
-// ============================================================
-// Languages enumerated for the AST adapter set (Phase 1 + 3c)
-// ============================================================
-
-/**
- * Closed-set of Tree-sitter grammars massu ships first-party adapters for.
- *
- * Note: this is a string-literal union, NOT re-exported from `web-tree-sitter`
- * (which exposes `Language` as a class, not a name list). Phase 1 ships
- * adapters for python/typescript/javascript/swift only — the remaining
- * languages are reserved for Plan 3c.
- */
-export type TreeSitterLanguage =
-  | 'python'
-  | 'typescript'
-  | 'javascript'
-  | 'swift'
-  | 'rust'
-  | 'go'
-  | 'ruby'
-  | 'php'
-  | 'java'
-  | 'kotlin'
-  | 'elixir'
-  | 'erlang'
-  | 'csharp'
-  | 'cpp'
-  | 'haskell'
-  | 'ocaml';
-
-// ============================================================
-// Inputs to adapter dispatch
-// ============================================================
-
-/**
- * Read-only signal bundle the runner builds BEFORE adapter dispatch.
- *
- * Adapters consume signals to answer `matches()` cheaply (no file IO inside
- * `matches()` — that's why the bundle is built up-front).
- */
-export interface DetectionSignals {
-  /** Parsed `package.json` (root or first workspace) — undefined if absent. */
-  packageJson?: Record<string, unknown>;
-  /** Parsed `pyproject.toml` — undefined if absent. */
-  pyprojectToml?: Record<string, unknown>;
-  /** Raw `Gemfile` text — undefined if absent. */
-  gemfile?: string;
-  /** Parsed `Cargo.toml` — undefined if absent. */
-  cargoToml?: Record<string, unknown>;
-  /** Raw `go.mod` text — undefined if absent. */
-  goMod?: string;
-  /** Raw `mix.exs` text — undefined if absent. Elixir/Mix manifest. */
-  mixExs?: string;
-  /**
-   * Raw text of the FIRST `.csproj` file found at the project root —
-   * undefined if absent. .NET project file (XML). Adapters that need to
-   * inspect more than one csproj (multi-project solutions) can re-scan from
-   * `presentFiles`.
-   */
-  csproj?: string;
-  /** Raw `pom.xml` text — undefined if absent. Maven build manifest. */
-  pomXml?: string;
-  /**
-   * Raw text of `build.gradle` OR `build.gradle.kts` — whichever exists at
-   * the project root, with `.kts` (Kotlin DSL) preferred when both are
-   * present (Kotlin DSL is the modern default per Gradle 7+ docs).
-   * Undefined if neither exists.
-   */
-  gradleBuild?: string;
-  /** Set of present directory names directly under the project root (one level). */
-  presentDirs: Set<string>;
-  /** Set of present file basenames directly under the project root (one level). */
-  presentFiles: Set<string>;
-}
-
-/**
- * A sampled source file the runner hands to the adapter.
- *
- * `content` is pre-read; adapters MUST NOT re-read from disk inside
- * `introspect()`. `size` is in bytes (pre-read length).
- */
-export interface SourceFile {
-  path: string;
-  content: string;
-  language: TreeSitterLanguage;
-  size: number;
-}
-
-// ============================================================
-// Adapter contract
-// ============================================================
-
-/**
- * Trail entry produced for every captured field — the user can audit
- * `detected.<adapter>._provenance` to see exactly which file/line/query
- * produced a value.
- */
-export interface Provenance {
-  field: string;
-  sourceFile: string;
-  line: number;
-  query: string;
-}
-
-export interface AdapterResult {
-  /**
-   * Becomes `detected.<adapter.id>` in `massu.config.yaml`. Field names are
-   * adapter-defined; values are `unknown` so adapters can return strings,
-   * arrays, or nested records as needed.
-   */
-  conventions: Record<string, unknown>;
-  /**
-   * Per-field provenance trail. The runner writes this to
-   * `detected.<adapter.id>._provenance` so a downstream auditor can verify
-   * any extracted value.
-   */
-  provenance: Provenance[];
-  /**
-   * 'high'  : single canonical match, query produced exactly one result
-   * 'medium': multiple matches, all agree
-   * 'low'   : multiple matches with disagreement (still emitted, with warning)
-   * 'none'  : no matches, timed out, or threw — fields are dropped
-   */
-  confidence: 'high' | 'medium' | 'low' | 'none';
-}
-
-export interface CodebaseAdapter {
-  /** Stable adapter id, e.g. "python-fastapi". Becomes `detected.<id>` block. */
-  id: string;
-  /** Languages this adapter consumes. Used by the runner to skip work. */
-  languages: TreeSitterLanguage[];
-  /**
-   * Cheap signal check — must NOT do file IO. Returns true if any signal
-   * suggests this adapter should run.
-   */
-  matches(signals: DetectionSignals): boolean;
-  /**
-   * Sample N files (already read by the runner), run AST queries, return
-   * extracted conventions. May throw — the runner isolates failures.
-   */
-  introspect(files: SourceFile[], rootDir: string): Promise<AdapterResult>;
-}
-
-// ============================================================
-// Runner output
-// ============================================================
-
-/**
- * The runner's output: per-adapter id → its conventions block (with the
- * `_provenance` map merged in). The introspector then folds this into the
- * `detected.<adapter.id>` namespace alongside the existing
- * `detected.python` / `detected.swift` / `detected.typescript` regex blocks.
- */
-export interface MergedAdapterOutput {
-  /** Per-adapter id → resolved conventions. */
-  byAdapter: Record<string, AdapterResolved>;
-  /** Adapters that were skipped (didn't match) for diagnostic logging. */
-  skipped: string[];
-  /** Adapters that threw during introspect — runner isolates these. */
-  errored: Array<{ adapterId: string; error: string }>;
-}
-
-/**
- * Resolved-and-merged form of an `AdapterResult`. Provenance is folded into
- * `_provenance` (key per field, value = `path:line :: query`).
- */
-export interface AdapterResolved {
-  conventions: Record<string, unknown>;
-  /** field-name -> "relativePath:line :: queryName". Empty when no fields. */
-  _provenance: Record<string, string>;
-  confidence: 'high' | 'medium' | 'low' | 'none';
-}

package/src/detect/codebase-introspector.ts

@@ -1,210 +0,0 @@
-// Copyright (c) 2026 Massu. All rights reserved.
-// Licensed under BSL 1.1 - see LICENSE file for details.
-
-/**
- * Codebase Introspector — 2-tier dispatcher (Plan 3b Phase 1)
- * ============================================================
- *
- * Public signature: `introspect(detection, projectRoot): DetectedConventions`.
- * Byte-for-byte unchanged from Plan #2 — `detect/index.ts` calls this
- * function as before.
- *
- * Internal change (Plan 3b Phase 1): the function is now a 2-tier dispatcher.
- *
- *   Tier 1 — AST adapters (preferred). Run via `runner.ts` against the four
- *   first-party adapters: python-fastapi, python-django, nextjs-trpc,
- *   swift-swiftui. Each writes to its own `detected.<adapter.id>` block
- *   alongside the regex blocks. AST adapters use Tree-sitter S-expression
- *   queries — never regex. Confidence is per-field.
- *
- *   Tier 2 — Regex fallback. For fields the AST adapters returned 'none'
- *   confidence on, the regex helpers in `regex-fallback.ts` (verbatim moved
- *   from this file's previous incarnation) take over. AST-wins rule: when
- *   both tiers produce a value for the same `detected.<lang>.<field>` slot,
- *   AST wins; the runner records both in provenance.
- *
- *   Tier 3 — null. The template engine's `| default("...")` then takes over.
- *
- * The AST tier may degrade silently to regex-only when:
- *   - The Tree-sitter grammar is unavailable offline AND uncached
- *   - The adapter throws (per-adapter try/catch in `runner.ts`)
- *   - The grammar SHA-256 doesn't match the manifest (refused by loader)
- *
- * Exception: if a Tree-sitter query is malformed (developer bug), the loader
- * propagates `InvalidQueryError` so we don't silently mask it.
- */
-
-import type { DetectionResult } from './index.ts';
-import {
-  introspectPython,
-  introspectSwift,
-  introspectTypeScript,
-  type DetectedPython,
-  type DetectedSwift,
-  type DetectedTypeScript,
-} from './regex-fallback.ts';
-import { runAdapters, buildDetectionSignals } from './adapters/runner.ts';
-import { pythonFastApiAdapter } from './adapters/python-fastapi.ts';
-import { pythonDjangoAdapter } from './adapters/python-django.ts';
-import { nextjsTrpcAdapter } from './adapters/nextjs-trpc.ts';
-import { swiftSwiftUiAdapter } from './adapters/swift-swiftui.ts';
-// Plan 1.5.4 R-011 discovery: Phase 7 adapters were committed but never
-// added to FIRST_PARTY_ADAPTERS. The omission was masked pre-1.5.4 by
-// sampleFiles=[] (every adapter returned 'none' anyway). Now that the
-// sampler works, these 6 adapters MUST be in the dispatch list or
-// `npx massu init` against a Phase 7 project produces no detected.<id>:
-// block (verified via debug instrumentation 2026-05-08 against the
-// 1.5.4 published bundle).
-import { pythonFlaskAdapter } from './adapters/python-flask.ts';
-import { goChiAdapter } from './adapters/go-chi.ts';
-import { railsAdapter } from './adapters/rails.ts';
-import { phoenixAdapter } from './adapters/phoenix.ts';
-import { aspnetAdapter } from './adapters/aspnet.ts';
-import { springAdapter } from './adapters/spring.ts';
-import type { CodebaseAdapter, AdapterResolved } from './adapters/types.ts';
-
-// ============================================================
-// Public types — unchanged from Plan #2 to preserve consumers
-// ============================================================
-
-export type { DetectedPython, DetectedSwift, DetectedTypeScript };
-
-export interface DetectedConventions {
-  python?: DetectedPython;
-  swift?: DetectedSwift;
-  typescript?: DetectedTypeScript;
-  /**
-   * AST adapter blocks live here (Plan 3b). Keys are adapter ids
-   * (`python-fastapi`, etc.). Values include both extracted conventions and
-   * a `_provenance` map.
-   */
-  [adapterId: string]: unknown;
-}
-
-// ============================================================
-// Static adapter list (v1 — first-party only, per spec §6)
-// ============================================================
-
-const FIRST_PARTY_ADAPTERS: CodebaseAdapter[] = [
-  pythonFastApiAdapter,
-  pythonDjangoAdapter,
-  pythonFlaskAdapter,
-  nextjsTrpcAdapter,
-  swiftSwiftUiAdapter,
-  goChiAdapter,
-  railsAdapter,
-  phoenixAdapter,
-  aspnetAdapter,
-  springAdapter,
-];
-
-// ============================================================
-// Public entry point
-// ============================================================
-
-/**
- * Introspect the project's source files. Returns per-language conventions or
- * an empty object if nothing was extracted.
- *
- * Synchronous signature is preserved — AST adapter execution is intentionally
- * fire-and-forget at this layer. Phase 1 wires the adapter pipeline behind
- * the existing sync function so `detect/index.ts` is byte-for-byte unchanged
- * (plan line 137-142). The async adapter orchestration lives entirely inside
- * `runIntrospect()`, which `introspect()` does NOT await — adapters either
- * have their grammars cached (fast path, no async needed at the JS level by
- * Phase 4 wiring) or degrade to regex.
- *
- * For Phase 1 Tier 1 to actually contribute values, callers must use the
- * async variant `introspectAsync()`. The sync `introspect()` runs the regex
- * tier ONLY for Phase 1; Phase 4 callers (LSP enrichment + adapter wiring)
- * will switch to async.
- */
-export function introspect(
-  detection: DetectionResult,
-  projectRoot: string,
-): DetectedConventions {
-  const out: DetectedConventions = {};
-  const languages = Array.from(
-    new Set(detection.manifests.map(m => m.language)),
-  );
-
-  // Tier 2 — regex fallback. Always runs. AST adapters in the async variant
-  // (`introspectAsync`) populate `detected.<adapter-id>` alongside; for the
-  // sync entry point, only the regex tier participates so Plan #2 callers
-  // see no behavior change.
-  if (languages.includes('python')) {
-    const python = introspectPython(detection, projectRoot);
-    if (python !== null) out.python = python;
-  }
-
-  if (languages.includes('swift')) {
-    const swift = introspectSwift(detection, projectRoot);
-    if (swift !== null) out.swift = swift;
-  }
-
-  if (languages.includes('typescript') || languages.includes('javascript')) {
-    const ts = introspectTypeScript(detection, projectRoot);
-    if (ts !== null) out.typescript = ts;
-  }
-
-  return out;
-}
-
-// ============================================================
-// Async variant — used by callers who want AST tier participation
-// ============================================================
-
-/**
- * Async introspect: runs AST adapters first (Tier 1), then regex fallback
- * (Tier 2) for fields the adapters returned 'none' on.
- *
- * Returns the same `DetectedConventions` shape as the sync `introspect()`,
- * plus per-adapter blocks under their ids.
- *
- * Callers who can `await` (CLI commands, tests, etc.) should prefer this
- * variant. The session-start hook keeps using sync `introspect()` for its
- * 5s budget reason (P4-006).
- */
-export async function introspectAsync(
-  detection: DetectionResult,
-  projectRoot: string,
-): Promise<DetectedConventions> {
-  const out: DetectedConventions = introspect(detection, projectRoot);
-
-  // Build signals + run AST adapters with the real file sampler (Plan
-  // 1.5.4 §3). Pre-1.5.4 this was a placeholder returning [], which kept
-  // every AST adapter at 'none' confidence in the init flow. The
-  // adapters themselves always worked (verified by adapter-grammar-
-  // strict.test.ts); the gap was purely the seam between detection +
-  // adapter execution. The sampler reuses EXTENSIONS / TEST_FILE_PATTERNS
-  // from source-dir-detector.ts so adding a new language doesn't require
-  // a parallel map (CR-46 self-attest #3).
-  const signals = buildDetectionSignals(projectRoot);
-  const { sampleFilesForAdapter } = await import('./adapters/file-sampler.ts');
-  let merged;
-  try {
-    merged = await runAdapters(FIRST_PARTY_ADAPTERS, projectRoot, signals, {
-      sampleFiles: async (adapter, root) => {
-        return sampleFilesForAdapter(adapter, root, detection);
-      },
-    });
-  } catch {
-    return out;
-  }
-
-  for (const [adapterId, resolved] of Object.entries(merged.byAdapter)) {
-    if (resolved.confidence === 'none') continue;
-    out[adapterId] = serializeAdapterBlock(resolved);
-  }
-
-  return out;
-}
-
-function serializeAdapterBlock(r: AdapterResolved): Record<string, unknown> {
-  const block: Record<string, unknown> = { ...r.conventions };
-  if (Object.keys(r._provenance).length > 0) {
-    block._provenance = r._provenance;
-  }
-  block._confidence = r.confidence;
-  return block;
-}

package/src/detect/domain-inferrer.ts

@@ -1,197 +0,0 @@
-// Copyright (c) 2026 Massu. All rights reserved.
-// Licensed under BSL 1.1 - see LICENSE file for details.
-
-/**
- * Domain Inferrer (P1-006)
- * ========================
- *
- * Suggests `DomainConfig[]` entries based on monorepo + source-dir discovery.
- * Each workspace package (apps/*, packages/*, services/*, libs/*, modules/*)
- * becomes one suggested domain. In single-package repos, top-level
- * `src/<subdir>/` candidates are suggested as domains.
- *
- * Output matches the existing `DomainConfig` type from `config.ts` so init
- * and refresh can write it directly into `massu.config.yaml`. Suggested
- * `allowedImportsFrom` is always empty — the user fills relationships.
- *
- * Deterministic ordering: alphabetical by domain name.
- *
- * Usage:
- * ```ts
- * import { inferDomains } from './detect/domain-inferrer.ts';
- * const domains = inferDomains('/repo', monorepo, sourceDirs);
- * ```
- */
-
-import { existsSync, readdirSync } from 'fs';
-import { join } from 'path';
-import type { DomainConfig } from '../config.ts';
-import type { MonorepoInfo, WorkspacePackage } from './monorepo-detector.ts';
-import type { SourceDirMap } from './source-dir-detector.ts';
-
-const IGNORED_SUBDIRS = new Set([
-  'node_modules',
-  '__pycache__',
-  'dist',
-  'build',
-  '.build',
-  'target',
-  '.next',
-  '.git',
-  '.massu',
-  'coverage',
-  'tests',
-  'test',
-  '__tests__',
-]);
-
-function titleCase(s: string): string {
-  if (!s) return s;
-  return s
-    .split(/[-_\s]+/)
-    .filter(Boolean)
-    .map((p) => p.charAt(0).toUpperCase() + p.slice(1))
-    .join(' ');
-}
-
-function domainFromWorkspace(pkg: WorkspacePackage): DomainConfig {
-  // Prefer the explicit package name; fall back to the final path segment.
-  const pathTail = pkg.path.split('/').pop() ?? pkg.path;
-  const name = pkg.name ?? titleCase(pathTail);
-  return {
-    name,
-    routers: [],
-    pages: [],
-    tables: [],
-    allowedImportsFrom: [],
-  };
-}
-
-/**
- * Enumerate domain candidates under each detected source directory.
- *
- * The `sourceDirs` argument is the flattened, unique list of relative
- * source paths produced upstream by the source-dir detector
- * (`detectSourceDirs` in `source-dir-detector.ts`). For each path that
- * exists under `root`, this function lists immediate subdirectories as
- * candidate domain names. Hardcoded `src/` lookup was removed (plan
- * `plan-1.7.0-cohesive-cleanup` P-B-002) — the function now consumes
- * the detection pipeline's output verbatim, so projects whose source
- * lives at non-`src/` paths (e.g. `lib/`, `apps/<x>/src/`) are no
- * longer silently dropped.
- *
- * Empty `sourceDirs` is treated as a legacy single-repo `src/` lookup
- * to preserve behavior for callers that pre-date the source-dir
- * pipeline (CLI / test harnesses that hand-wire `inferDomains`).
- *
- * Returns deduplicated subdir names sorted alphabetically.
- */
-function topLevelSrcSubdirs(root: string, sourceDirs: readonly string[]): string[] {
-  const effective = sourceDirs.length > 0 ? sourceDirs : ['src'];
-  const seen = new Set<string>();
-  for (const rel of effective) {
-    const abs = join(root, rel);
-    if (!existsSync(abs)) continue;
-    try {
-      for (const e of readdirSync(abs, { withFileTypes: true })) {
-        if (!e.isDirectory()) continue;
-        if (IGNORED_SUBDIRS.has(e.name)) continue;
-        seen.add(e.name);
-      }
-    } catch {
-      // skip directories that cannot be read; do not throw.
-    }
-  }
-  return Array.from(seen).sort();
-}
-
-/**
- * Flatten a `SourceDirMap` into a unique, deduplicated list of relative
- * source paths across all detected languages.
- *
- * Drops the root sentinels `.` and `''` — those are emitted by the
- * source-dir-detector when source files live directly at the project
- * root (e.g. Django's `manage.py` or Swift's `Package.swift`). Treating
- * them as enumerable source dirs causes spurious top-level directory
- * inclusion (Tests/, Sources/, etc.), which collides with the
- * language-fallback path in `inferDomains`. Root-source repos rely on
- * the language-fallback path to emit `{Python}` / `{Swift}` domains,
- * NOT a fan-out of every root subdirectory.
- *
- * Order is not guaranteed — callers that need determinism must sort.
- */
-function flattenSourceDirs(sourceDirs: SourceDirMap): string[] {
-  const flat = new Set<string>();
-  for (const entry of Object.values(sourceDirs)) {
-    if (!entry) continue;
-    for (const dir of entry.source_dirs) {
-      if (dir === '.' || dir === '') continue;
-      flat.add(dir);
-    }
-  }
-  return Array.from(flat);
-}
-
-/**
- * Produce a suggested `DomainConfig[]`.
- *
- * @param projectRoot absolute path
- * @param monorepo    output of P1-004 detectMonorepo
- * @param sourceDirs  output of P1-003 detectSourceDirs
- */
-export function inferDomains(
-  projectRoot: string,
-  monorepo: MonorepoInfo,
-  sourceDirs: SourceDirMap
-): DomainConfig[] {
-  const domains: DomainConfig[] = [];
-
-  if (monorepo.type !== 'single' && monorepo.packages.length > 0) {
-    // Monorepo: one domain per workspace package.
-    for (const pkg of monorepo.packages) {
-      domains.push(domainFromWorkspace(pkg));
-    }
-  } else {
-    // Single repo: suggest one domain per top-level <sourceDir>/<subdir>/ for
-    // every detected source dir (formerly hardcoded to `src/` only — see
-    // P-B-002 in plan-1.7.0-cohesive-cleanup).
-    const flat = flattenSourceDirs(sourceDirs);
-    const subdirs = topLevelSrcSubdirs(projectRoot, flat);
-    for (const s of subdirs) {
-      domains.push({
-        name: titleCase(s),
-        routers: [],
-        pages: [],
-        tables: [],
-        allowedImportsFrom: [],
-      });
-    }
-    // If no src/ subdirs, emit a single-language-based domain when sourceDirs has entries.
-    if (domains.length === 0) {
-      const langs = Object.keys(sourceDirs);
-      for (const lang of langs.sort()) {
-        domains.push({
-          name: titleCase(lang),
-          routers: [],
-          pages: [],
-          tables: [],
-          allowedImportsFrom: [],
-        });
-      }
-    }
-  }
-
-  // Deterministic alphabetical order by name.
-  domains.sort((a, b) => a.name.localeCompare(b.name));
-
-  // Dedup by name (monorepo workspaces may coincidentally share a name).
-  const seen = new Set<string>();
-  const dedup: DomainConfig[] = [];
-  for (const d of domains) {
-    if (seen.has(d.name)) continue;
-    seen.add(d.name);
-    dedup.push(d);
-  }
-
-  return dedup;
-}