docs(specs): FR2 plan-of-record + ADR-0008 (param-passing generated repo helpers)

FR2 rewrite folds in the brainstormed decisions:
- Option B scope: queries-side hard break in 0.7.0 + docs recipe; routesFile()
  unchanged; no runtime-ts split; no new generator factories.
- Docs deliverable at docs/recipes/wiring-generated-queries.md with three
  framework worked examples (Hono primary edge, Fastify, raw fetch),
  composing-with-custom-routes pattern, plugin-model walkthrough for
  consumer-owned routes generators, and the migration guide.
- Cross-language principle moved out to a separate ADR.

ADR-0008 captures the cross-language principle: every port's generated repo
helpers accept persistence-context as a parameter (or via the language-
idiomatic equivalent — DI for C#, session arg for Python, @repository
injection for Java). C# already conforms (EF Core DI). TS adopts in 0.7.0.
Java + Python adopt when their codegen ships. Module-level singletons are
not emitted by any port.

Roadmap pointer updated: FR2 is now plan-of-record (no longer "design
proposal"); ADR-0008 linked alongside.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: dmealing

docs/superpowers/specs/2026-05-25-fr-param-passing-generated-repo-helpers-design.md

@@ -1,163 +1,257 @@
-# FR: Parameter-passing for generated repo helpers (`db` as first arg)
+# FR: Parameter-passing for generated repo helpers
+
+**Status:** Design — implementation-ready (plan-of-record)
+**Date:** 2026-05-25 (revised after brainstorm)
+**Scope:** TypeScript implementation (`@metaobjectsdev/codegen-ts`) + a docs recipe; the
+cross-language design principle is captured separately in
+[ADR-0008](../../../spec/decisions/ADR-0008-parameter-passing-generated-repo-helpers.md).
+**Depends on:** existing `queriesFile()` codegen pipeline; nothing else.
+**Breaking change for:** 0.6.0 → 0.7.0 TS consumers — every call site that invokes a
+generated `<Entity>.queries.ts` helper updates from `findUserById(id)` to
+`findUserById(db, id)`. Migration is mechanical (search-and-replace at call sites; run
+`meta gen` to update the generated files).
 
-**Status:** Design proposal — needs brainstorm before implementation
-**Date:** 2026-05-25
-**Scope:** Cross-language design decision; TypeScript impl first (`@metaobjectsdev/codegen-ts`)
-**Origin:** Friction observed in a downstream consumer adopting 0.6.0 on Cloudflare Workers,
-where per-request DB bindings make the current module-level `db` import unworkable. The
-underlying design question — "does the generated CRUD helper take a `db` argument, or
-import it from a known module path?" — is cross-language: every port's codegen makes
-the same choice.
-
-## Current state
+## Goal
 
-Generated `<Entity>.queries.ts` (TS, see `codegen-ts/src/templates/queries.ts`) emits:
+Generated CRUD helpers in `<Entity>.queries.ts` accept their Drizzle persistence-context
+as the first parameter. The module-level `import { db } from "../db"` they emit today is
+removed.
 
 ```ts
+// Before (0.6.0)
 import { db } from "../db";
-// ...
-export async function findCouncilById(id: string) {
-  return db.select()...
+export async function findUserById(id: string): Promise<User | null> {
+  const [u] = await db.select().from(users).where(eq(users.id, id));
+  return u ?? null;
+}
+
+// After (0.7.0)
+import type { BaseSQLiteDatabase } from "drizzle-orm/sqlite-core";
+type Db = BaseSQLiteDatabase<"async", Record<string, never>>;
+
+export async function findUserById(db: Db, id: string): Promise<User | null> {
+  const [u] = await db.select().from(users).where(eq(users.id, id));
+  return u ?? null;
 }
 ```
 
-The `../db` module path is consumer-provided. The codegen assumes a singleton, module-level
-`db` instance exists at that path.
-
-This works in:
-- Long-lived Node.js / Bun processes (the `db` is created once at boot).
-- Server runtimes where the DB connection is process-scoped (Postgres pool, libsql client).
-
-This fails in:
-- **Cloudflare Workers / Vercel Edge / Deno Deploy.** The DB binding (`env.DB`) is
-  request-scoped — there is no module-level `db` to import. Consumers ship a
-  runtime-throwing stub purely to satisfy the typecheck:
-  ```ts
-  export const db = (() => {
-    throw new Error("type-only stub — use getDb(env) in the handler");
-  })() as unknown as BaseSQLiteDatabase<...>;
-  ```
-  Every Worker adopter pays this dead-code tax forever.
-- **Multi-tenant servers** where each tenant has its own DB connection.
-- **Test isolation** where each test wants its own in-memory DB.
-
-The idiomatic 2026 ORM-helper pattern (current Drizzle docs, Kysely, Prisma, TypeORM,
-Knex) is parameter-passing. We're the outlier.
+## Why
+
+The module-level singleton works in long-lived Node.js / Bun processes (the `db` is
+created once at boot). It fails on:
+
+- **Cloudflare Workers / Vercel Edge / Deno Deploy** — the DB binding (`env.DB`) is
+  request-scoped; no module-level `db` to import. Today's workers consumers ship a
+  runtime-throwing stub purely to satisfy the typecheck.
+- **Multi-tenant servers** — each tenant has its own DB connection.
+- **Test isolation** — each test wants an in-memory DB.
+
+The idiomatic 2026 ORM-helper pattern (Drizzle, Kysely, Prisma, TypeORM, Knex,
+SQLAlchemy 2.x, EF Core via DI) is parameter-passing. We are the outlier. ADR-0008
+documents this as the cross-language principle; this FR is the TypeScript implementation.
+
+## Design — minimal scope (Option B from brainstorm)
+
+The brainstorm considered four scopes (queries-only; queries + docs recipe; queries + a
+deferred Hono routes FR; queries + dual-target routes generator + runtime-ts split).
+Option B — **queries-only + docs recipe** — was chosen because:
+
+- Workers consumers were blocked by module-level `db`, not by the absence of generated
+  HTTP routes; they hand-write their HTTP layer regardless.
+- Pre-emptively adding a Hono routes generator forces a framework pick (Hono vs itty vs
+  raw-fetch vs whatever comes next) that the user base hasn't asked for. The
+  consumer-writable `Generator` plugin pattern already lets adopters ship their own.
+- `routesFile()` (Fastify) stays unchanged — well-served Node consumers see no
+  regression on the routes side.
+
+### What changes
+
+1. **`codegen-ts/src/templates/queries.ts`** — every renderXxxFn() helper
+   (`renderFindByIdFn`, `renderListFn`, `renderCreateFn`, `renderUpdateFn`,
+   `renderDeleteByIdFn`) gains `db: Db` as the first parameter. No helper has a
+   top-level `import { db } from "../db"`.
+
+2. **`codegen-ts/src/templates/queries-file.ts`** — the file composer emits a dialect-
+   correct `type Db = ...` alias at the top, then every helper signature references
+   `Db`:
+
+   ```ts
+   // sqlite/d1 (covers libsql, Turso, Cloudflare D1 — all async sqlite)
+   import type { BaseSQLiteDatabase } from "drizzle-orm/sqlite-core";
+   type Db = BaseSQLiteDatabase<"async", Record<string, never>>;
+
+   // postgres
+   import type { NodePgDatabase } from "drizzle-orm/node-postgres";
+   type Db = NodePgDatabase<Record<string, never>>;
+   ```
+
+   `<Record<string, never>>` for the schema parameter lets consumers narrow the type if
+   they want (`drizzle(connection, { schema })`); we don't impose a schema.
+
+3. **Goldens** — all `<Entity>.queries.ts` snapshots regenerated:
+   `test/golden/__snapshots__/{sqlite,postgres,package}/*.queries.ts`. The
+   `UPDATE_GOLDEN=1 bun test` mechanism (per `test/golden/golden-output.test.ts:5`)
+   handles regeneration.
+
+4. **Unit tests** — `codegen-ts/test/templates/queries.test.ts` updated. Each renderXxxFn
+   test asserts the rendered code (a) declares `db: Db` as first parameter, (b) does
+   NOT contain `import { db }`.
+
+### What doesn't change
+
+- `routesFile()` generator — still emits Fastify routes; consumers using it keep their
+  wiring exactly as today.
+- `@metaobjectsdev/runtime-ts` — untouched. No split into `/core`. Fastify helpers still
+  live at `/drizzle-fastify`.
+- `entity-file`, `barrel`, `payload-codegen`, `projection-decl`, all other generators —
+  no signature change.
+- `dbImport` codegen config field — stays (still useful for cross-target entity imports
+  unrelated to this change).
 
-## Goal
+## Docs recipe deliverable
 
-Generated CRUD helpers take a `db` instance as the first argument. Module-level `db`
-imports are no longer required:
+The other half of this FR is a consumer-facing docs page at
+`docs/recipes/wiring-generated-queries.md`. The page is paste-and-run, ~200-300 lines
+including code blocks. Outline:
 
-```ts
-// Generated <Entity>.queries.ts (after this FR)
-import type { BaseSQLiteDatabase } from "drizzle-orm/sqlite-core";
+1. **TL;DR** — generated queries take `db` as the first arg; pass any compatible
+   Drizzle instance.
 
-export async function findCouncilById(
-  db: BaseSQLiteDatabase<"async", { rowsAffected: number }>,
-  id: string,
-): Promise<Council | null> {
-  const [row] = await db.select().from(councils).where(eq(councils.id, id));
-  return row ?? null;
-}
-```
+2. **Setting up `db` per dialect** — three subsections:
+   - SQLite/libsql/Turso — `drizzle(libsqlClient)` at module init.
+   - D1 (Cloudflare Workers) — `drizzle(env.DB)` inside the request handler.
+   - Postgres — `drizzle(pool)` at module init or per-request (multi-tenant).
 
-Consumer call sites pass their request-scoped (or module-level) `db`:
+3. **Wiring in Hono** — a 30-line worked example: full CRUD for one entity,
+   request-scoped `getDb(c)`, custom auth middleware applied to a sub-app, the
+   sub-app mounted at `/api/users`. Hono is chosen as the primary edge example
+   because it is the de facto edge HTTP router in 2026, not because the framework
+   is blessed — the same pattern works against itty, hono v5, or raw `fetch()`
+   handlers with a different import line.
 
-```ts
-// Worker handler
-const db = drizzle(env.DB, { schema });
-const council = await findCouncilById(db, "abc123");
+4. **Wiring in Fastify** — equivalent example. Two variants:
+   (a) using the existing `routesFile()` generator (unchanged in 0.7.0);
+   (b) hand-rolling routes with the generated queries.
 
-// Node server (unchanged ergonomics)
-const db = drizzle(pool, { schema });
-const council = await findCouncilById(db, "abc123");
-```
+5. **Wiring with raw `fetch()` / itty-router** — short minimal example for
+   consumers avoiding HTTP framework deps. Shows how the generated queries
+   compose with any router.
 
-## Why this is cross-language
+6. **Composing with custom routes** — pattern showing generated queries +
+   hand-written endpoints living together (e.g. generated `findUserById` plus
+   custom `POST /users/:id/promote`). Demonstrates the all-or-nothing-per-verb
+   property of today's routes generator is non-binding when you use queries
+   directly.
 
-The same design choice exists in every port's codegen:
+7. **Wrapping into your own `Generator` factory** — for consumers who want
+   their routes generated, the ~30-line plugin-model factory pattern. This
+   doubles as the worked example for the codegen plugin system itself.
 
-| Port | Today's choice (probably) | Idiomatic 2026 |
-|---|---|---|
-| TypeScript (codegen-ts) | module-level `db` import | parameter-passing |
-| C# (`MetaObjects.Codegen`) | `AppDbContext` constructor-injected (EF Core idiom) | already parameter-passing-ish via DI |
-| Java (not yet shipped) | Spring `@Repository` injection | per-Spring convention |
-| Python (planned post-H3) | SQLAlchemy session factory or module-level | parameter or context |
+8. **Migration from 0.6.0** — three-step guide:
+   ```
+   bun add -E @metaobjectsdev/cli@0.7.0   # bump
+   meta gen                                # regenerate
+   # find/replace call sites: findX(args) → findX(db, args)
+   ```
 
-C# already gets this right via DI. TypeScript is the obvious gap. The cross-language
-spec should:
-1. Declare that generated repo helpers take their persistence-context as a parameter
-   (or via equivalent idiom: constructor injection for OO ports, `db` arg for functional).
-2. Document the per-port acceptable shapes.
-3. Let each port migrate at its own pace.
+9. **Why parameter-passing** — one-paragraph rationale linking to ADR-0008.
 
-## Brainstorm topics (open)
+The recipe is implementation work; the actual prose is produced by the plan, not this
+spec.
 
-These need a brainstorm before implementation:
+## Tests + verification
 
-### 1. Migration path for existing 0.6.0 TypeScript consumers
+### Unit tests (codegen-ts)
 
-Three options:
+- `test/templates/queries.test.ts` updated:
+  - `renderFindByIdFn` test asserts `db: Db` is the first parameter; `db` symbol is
+    typed.
+  - Same for `renderListFn`, `renderCreateFn`, `renderUpdateFn`, `renderDeleteByIdFn`.
+  - File-level test on `queries-file.ts` asserts the `type Db = ...` alias is emitted
+    at the top and the dialect choice produces the correct Drizzle import path.
 
-**(a) Hard break in 0.7.0.** Consumers update every call site. Clean, predictable, but
-forces a single coordinated cut for any active codebase. Documented in CHANGELOG.
+### Goldens (codegen-ts)
 
-**(b) Soft transition: emit both shapes for one release.** Generated code includes:
-```ts
-// Legacy module-level import (deprecated; remove in 0.8.0)
-import { db as _db } from "../db";
-export async function findCouncilById(db: BaseSQLiteDatabase = _db, id: string) { ... }
-```
-Consumers migrate at leisure. Cleanup is automated in 0.8.0.
+- Regenerate with `UPDATE_GOLDEN=1 bun test packages/codegen-ts/test/golden/`. Affected
+  files: ~20 `<Entity>.queries.ts` snapshots across `sqlite`, `postgres`, and the
+  `package` layout fixture.
 
-**(c) Codegen config flag: `queriesShape: "module-db" | "param-db"`.** Default flips to
-`param-db` in 0.7.0; consumers can opt back briefly. Aligns with the existing
-`columnNamingStrategy` precedent.
+### Runtime integration (runtime-ts)
 
-Recommendation hypothesis: **(c)** — least disruptive, follows existing config-flag idiom,
-default flip in 0.7.0 with a CHANGELOG entry. But brainstorm should confirm.
+- `runtime-ts/test/`'s end-to-end suite already exercises a live libsql + a live
+  postgres against generated queries. Update the test plumbing so the test creates
+  its `db` and threads it through the helper calls (a few lines of test-side change).
+  Test bodies confirming behaviour are unchanged.
 
-### 2. Type-only signature for the `db` parameter
+### Docs validation
 
-For Drizzle specifically: `BaseSQLiteDatabase<"async", { rowsAffected: number }>` is the
-narrowest shape covering both libsql and D1. For Postgres it's `NodePgDatabase` (or its
-generic). Codegen needs to emit the dialect-correct type — already known from
-`config.dialect`. Worth a small design note on how this type is imported (deep import
-from `drizzle-orm/*-core` vs. re-exported through a shim).
+- Type-check the recipe code blocks via a small CI helper that extracts the TypeScript
+  fenced blocks and runs `tsc --noEmit` against them. Optional if extraction is fiddly;
+  human review is acceptable for a v1 recipe.
 
-### 3. Cross-port write-up timing
+## Migration story
 
-TS implementation in 0.7.0 is concrete and we have the consumer pain to motivate it.
-Should the cross-language design land alongside, or as a follow-up after TS proves the
-shape? Recommendation: **TS first, cross-language doc in `spec/decisions/ADR-XXXX-...md`
-as TS lands**, so other ports inherit the rationale without blocking TS work.
+`bun add @metaobjectsdev/cli@0.7.0` → run `meta gen` → search-and-replace call sites:
 
-## Out of scope (until brainstorm)
+| Before | After |
+|---|---|
+| `findUserById("abc")` | `findUserById(db, "abc")` |
+| `listUsers({ ... })` | `listUsers(db, { ... })` |
+| `createUser({ ... })` | `createUser(db, { ... })` |
+| `updateUserById("abc", { ... })` | `updateUserById(db, "abc", { ... })` |
+| `deleteUserById("abc")` | `deleteUserById(db, "abc")` |
 
-- Touching `meta init` defaults to remove the consumer's expected `src/db/db.ts` from
-  scaffold (will follow once the codegen change is in).
-- Removing the `dbImport` codegen config field (still useful for cross-target imports
-  unrelated to this change).
-- Routes codegen (separate Generator). The `routesFile()` generator imports the same
-  `db` via the queries module — once queries take `db` as a param, routes accept the
-  same and forward it. Mechanical fallout.
+CHANGELOG entry covered by the existing
+[release-notes FR](./2026-05-25-fr-release-notes-and-naming-convention-docs.md).
+
+Node-server consumers who prefer the old call-site shape write a one-line shim:
 
-## Tests required when this lands
+```ts
+import { findUserById as _findUserById } from "./generated/User.queries";
+import { db as myDb } from "./db";
+export const findUserById = (id: string) => _findUserById(myDb, id);
+```
 
-- `codegen-ts/test/templates/queries.test.ts` — every generated CRUD helper has `db` as
-  its first parameter.
-- Golden snapshot updates in `codegen-ts/test/golden/__snapshots__/{sqlite,postgres}/`.
-- A new conformance check (or extension of existing one) verifying the parameter shape
-  across dialects.
+This is the inverse of the dead-code stub Workers consumers ship today.
+
+## Out of scope
+
+- **New routes generator (Hono, Express, raw-fetch).** A separate follow-on FR ships
+  if user demand surfaces. The docs recipe + plugin model are the answer in 0.7.0.
+- **Splitting `@metaobjectsdev/runtime-ts` into universal + framework-specific
+  subpaths.** Not needed without a new framework target shipping.
+- **Per-verb skip mechanism** (`@routes: { create: false }`). YAGNI; today's routes
+  generator has the same all-or-nothing property and consumers haven't asked.
+- **A new edge-runtime package** (`@metaobjectsdev/runtime-edge`). Workers consumers
+  compose generated queries with their framework of choice; no new package needed.
+- **Codegen config flag** (`queriesShape: "module-db" | "param-db"`). Rejected
+  during brainstorm: dual code paths + dead-code risk on the deprecated branch.
+- **Soft transition** (emit both shapes for one release). Rejected during brainstorm:
+  generated code uglier than either pure form; doubles snapshot count; defers cleanup.
+
+## File-level change summary
+
+New files:
+
+- `docs/recipes/wiring-generated-queries.md` — the consumer recipe (~200-300 lines).
+- `spec/decisions/ADR-0008-parameter-passing-generated-repo-helpers.md` — cross-language
+  ADR (already landed alongside this FR).
+
+Modified files:
+
+- `server/typescript/packages/codegen-ts/src/templates/queries.ts` — all renderXxxFn()
+  helpers updated to accept `db: Db` as first parameter.
+- `server/typescript/packages/codegen-ts/src/templates/queries-file.ts` — emits
+  `type Db = ...` alias at top; drops the `import { db }` line.
+- `server/typescript/packages/codegen-ts/test/templates/queries.test.ts` — assertions
+  updated.
+- `server/typescript/packages/codegen-ts/test/golden/__snapshots__/{sqlite,postgres,package}/**/*.queries.ts` —
+  regenerated via `UPDATE_GOLDEN=1`.
+- `server/typescript/packages/runtime-ts/test/**` — test plumbing creates `db` and
+  passes it to helpers (test-side only, no production code changes).
+- `CHANGELOG.md` — 0.7.0 "Breaking" entry under "Changed" describing the migration.
+- `server/typescript/packages/codegen-ts/README.md` — link the new recipe.
 
 ## Open questions
 
-1. What is the public commitment around backward compat between 0.6.x and 0.7.0? Hard
-   break or soft transition?
-2. Should `@metaobjectsdev/runtime-ts` repo helpers (Kysely-based) follow the same
-   pattern? Currently they take `db` already, so the runtime side is consistent. The
-   gap is purely in codegen.
-3. Should this design extend to the routes generator (Fastify route registrar) so
-   request-scoped DBs flow through the entire generated stack? Probably yes — Workers
-   adopters need this end-to-end.
+None — all design decisions settled during brainstorm.