@zerotal/orm 1.8.1 → 1.10.0

package/src/commands/_backup.ts

@@ -0,0 +1,356 @@
+/**
+ * Taking a copy of the one file that cannot be rebuilt.
+ *
+ * SQLite is the framework's default database, which makes the database a single
+ * file on disk — and makes `cp` look like a backup. It is not one. A live SQLite
+ * database has pages in flight and a write-ahead log beside it; copying the file
+ * while the server is serving can capture a half-written page, and the result is a
+ * file that looks like a backup, sits in the retention directory for months, and
+ * turns out to be a corrupt database on the one morning anybody opens it.
+ *
+ * `VACUUM INTO` is the answer SQLite ships. It takes a read lock, walks the
+ * b-tree, and writes a complete, defragmented database while the server keeps
+ * serving. It needs no external binary either — which removes the `apt install
+ * sqlite3` step and the failure mode where a backup silently stops working because
+ * a base image dropped the CLI.
+ *
+ * The rest of this module exists because **a backup nobody has restored is a
+ * hope.** Every snapshot is opened and integrity-checked the moment it is written,
+ * and `rehearse` performs the actual restore — copy the file, open the copy, read
+ * it — because that is the operation you will be doing at 3am and it is the one
+ * worth knowing works.
+ *
+ * Every failure path throws. A backup timer that reports success while writing
+ * nothing is worse than no timer at all: it buys the confidence without the file.
+ *
+ * @module
+ */
+import { ZerotalError } from "@zerotal/core";
+import { Database } from "bun:sqlite";
+import { copyFileSync, mkdirSync, readdirSync, rmSync, statSync } from "node:fs";
+import { basename, join } from "node:path";
+
+/** How a backup was asked for. */
+export interface BackupOptions {
+  /** Absolute or cwd-relative path of the live SQLite database. */
+  source: string;
+  /** Directory snapshots are written into. Created if absent. */
+  dir: string;
+  /** Snapshots to keep, newest first. `0` keeps every one. */
+  keep: number;
+  /**
+   * Tables that must contain at least one row in the snapshot.
+   *
+   * The check that turns "a file was written" into "the file has the business in
+   * it". An empty `bookings` table in a snapshot of a live system is not a small
+   * discrepancy, it is the whole failure — and it is invisible in a byte count.
+   */
+  requireRows?: string[] | undefined;
+  /**
+   * Perform the restore, not just a read of the snapshot: copy it to a scratch
+   * path, open the copy, and check it there. Exercises the operation an incident
+   * actually needs, rather than the one that is convenient to test.
+   */
+  rehearse?: boolean | undefined;
+  /** Clock, so the snapshot name is deterministic under test. */
+  now?: Date | undefined;
+}
+
+/** What a completed backup wrote and proved. */
+export interface BackupResult {
+  /** Full path of the snapshot. */
+  path: string;
+  /** Its size on disk. */
+  bytes: number;
+  /** User tables found in it. */
+  tables: string[];
+  /** Row counts for the tables named in `requireRows`. */
+  rows: Record<string, number>;
+  /** Snapshots deleted by retention. */
+  pruned: string[];
+  /** Whether a full restore round-trip ran. */
+  rehearsed: boolean;
+}
+
+/**
+ * Raised for every way a backup can fail to be a backup.
+ *
+ * `ZerotalError` rather than `Error`, like every other error the framework throws:
+ * it carries a stable `code` a wrapper can switch on, which matters here because
+ * the caller is usually a shell script reading an exit status and a log line.
+ */
+export class BackupError extends ZerotalError {
+  constructor(message: string) {
+    super(message, "E_BACKUP_FAILED", 500);
+  }
+}
+
+/** Snapshot names this module writes, and therefore the only ones it will delete. */
+const SNAPSHOT_PATTERN = /^(.+)-(\d{8}-\d{6})\.sqlite$/;
+
+/**
+ * A UTC timestamp that sorts lexically in the same order it sorts chronologically,
+ * so retention can order snapshots by filename without parsing dates.
+ */
+export function backupStamp(now: Date): string {
+  const pad = (n: number, width = 2): string => String(n).padStart(width, "0");
+  return (
+    `${now.getUTCFullYear()}${pad(now.getUTCMonth() + 1)}${pad(now.getUTCDate())}` +
+    `-${pad(now.getUTCHours())}${pad(now.getUTCMinutes())}${pad(now.getUTCSeconds())}`
+  );
+}
+
+/** The snapshot filename for a source database at a moment. */
+export function snapshotName(source: string, now: Date): string {
+  const stem = basename(source).replace(/\.(sqlite3?|db)$/i, "") || "database";
+  return `${stem}-${backupStamp(now)}.sqlite`;
+}
+
+/**
+ * A path as a SQLite string literal.
+ *
+ * `VACUUM INTO` takes an expression, and the driver layer between here and SQLite
+ * does not reliably bind a parameter into that position across versions — so the
+ * path is inlined, and inlining a path means escaping it. A directory with an
+ * apostrophe in it is unusual and is not a reason to write a broken statement.
+ */
+export function sqlLiteral(value: string): string {
+  return `'${value.replaceAll("'", "''")}'`;
+}
+
+/** Whether a database URL points at SQLite rather than a database server. */
+export function isSqliteUrl(raw: string): boolean {
+  return !/^(postgres|postgresql|mysql|mysql2):\/\//.test(raw);
+}
+
+/** The filesystem path inside a SQLite URL, with any scheme prefix removed. */
+export function sqlitePathFromUrl(raw: string): string {
+  if (raw.startsWith("sqlite://")) return raw.slice("sqlite://".length);
+  if (raw.startsWith("sqlite:")) return raw.slice("sqlite:".length);
+  if (raw.startsWith("file:")) return raw.slice("file:".length);
+  return raw;
+}
+
+/** User tables in an open database — `sqlite_%` internals excluded. */
+function userTables(database: Database): string[] {
+  const rows = database
+    .query<{ name: string }, []>(
+      `SELECT name FROM sqlite_master WHERE type = 'table' AND name NOT LIKE 'sqlite_%' ORDER BY name`,
+    )
+    .all();
+  return rows.map((r) => r.name);
+}
+
+/**
+ * Open a snapshot and satisfy yourself that it is a database.
+ *
+ * `PRAGMA integrity_check` is the whole point: it walks every page and every
+ * index and reports the single row `ok` when the file is sound. Anything else is
+ * a corrupt snapshot, and finding that out now — while the source database is
+ * still there — is the difference between a bad night and a lost business.
+ *
+ * @throws {@link BackupError} When the file will not open, fails its integrity
+ *   check, or is missing rows the caller said it must have.
+ */
+export function verifySnapshot(
+  path: string,
+  requireRows: string[] = [],
+): { tables: string[]; rows: Record<string, number> } {
+  let database: Database;
+  try {
+    database = new Database(path, { readonly: true });
+  } catch (error) {
+    throw new BackupError(
+      `The snapshot at ${path} will not open as a database: ${(error as Error).message}`,
+    );
+  }
+
+  try {
+    // The open above proves nothing — bun:sqlite defers the real work to the first
+    // statement, so a text file masquerading as a database opens cleanly and fails
+    // here instead. The driver's own error is the honest one; it is wrapped so
+    // every failure out of this module is a BackupError a caller can match on.
+    let integrity: string[];
+    try {
+      integrity = database
+        .query<{ integrity_check: string }, []>(`PRAGMA integrity_check`)
+        .all()
+        .map((r) => r.integrity_check);
+    } catch (error) {
+      throw new BackupError(
+        `The snapshot at ${path} is not a readable database: ${(error as Error).message}`,
+      );
+    }
+    if (integrity.length !== 1 || integrity[0] !== "ok") {
+      throw new BackupError(
+        `The snapshot at ${path} failed its integrity check: ${integrity.join("; ")}`,
+      );
+    }
+
+    const tables = userTables(database);
+    const rows: Record<string, number> = {};
+    for (const table of requireRows) {
+      if (!tables.includes(table)) {
+        throw new BackupError(
+          `The snapshot at ${path} has no \`${table}\` table, and --require-rows said it must.`,
+        );
+      }
+      // The table name came from an operator's own flag, not from a request, and
+      // it has just been checked against the snapshot's own schema.
+      const count =
+        database.query<{ n: number }, []>(`SELECT COUNT(*) AS n FROM "${table}"`).get()?.n ?? 0;
+      if (count === 0) {
+        throw new BackupError(
+          `The snapshot at ${path} has an empty \`${table}\` table. A backup of a live ` +
+            `system with nothing in it is the failure, not a small discrepancy.`,
+        );
+      }
+      rows[table] = count;
+    }
+    return { tables, rows };
+  } finally {
+    database.close();
+  }
+}
+
+/**
+ * Do the restore. Copy the snapshot the way an incident would, open the copy, and
+ * check it there.
+ *
+ * Reading the snapshot in place proves the file is sound. This proves the thing
+ * you will actually be asked to do with it works — and it costs one file copy.
+ *
+ * @throws {@link BackupError} When the restored copy will not open or does not check out.
+ */
+export function rehearseRestore(path: string, requireRows: string[] = []): void {
+  const scratch = `${path}.rehearsal`;
+  try {
+    copyFileSync(path, scratch);
+    verifySnapshot(scratch, requireRows);
+  } catch (error) {
+    if (error instanceof BackupError) {
+      throw new BackupError(`Restore rehearsal failed. ${error.message}`);
+    }
+    throw new BackupError(`Restore rehearsal failed: ${(error as Error).message}`);
+  } finally {
+    rmSync(scratch, { force: true });
+  }
+}
+
+/**
+ * Delete snapshots beyond the newest `keep`.
+ *
+ * Only files matching the name this module writes are considered, so a retention
+ * setting can never reach anything a person put in the directory by hand. Names
+ * carry a sortable UTC stamp, which is why this can order them without opening one.
+ *
+ * @returns The filenames deleted.
+ */
+export function pruneSnapshots(dir: string, keep: number): string[] {
+  if (keep <= 0) return [];
+  let entries: string[];
+  try {
+    entries = readdirSync(dir);
+  } catch {
+    return [];
+  }
+  const snapshots = entries.filter((name) => SNAPSHOT_PATTERN.test(name)).sort();
+  const doomed = snapshots.slice(0, Math.max(0, snapshots.length - keep));
+  for (const name of doomed) rmSync(join(dir, name), { force: true });
+  return doomed;
+}
+
+/**
+ * Take a verified snapshot of a live SQLite database.
+ *
+ * @param run - Runs one statement against the live connection. Injected rather
+ *   than imported so this module stays testable without a booted application.
+ * @param options - Where the database is, where the snapshot goes, and what must
+ *   be true of it before this returns.
+ * @throws {@link BackupError} When anything at all goes wrong. There is no path
+ *   through this function that reports success without a checked file on disk.
+ */
+export async function takeBackup(
+  run: (sql: string) => Promise<unknown>,
+  options: BackupOptions,
+): Promise<BackupResult> {
+  const { source, dir, keep } = options;
+  const requireRows = options.requireRows ?? [];
+
+  if (source === ":memory:" || source === "") {
+    throw new BackupError(
+      "This app's database is in memory, so there is nothing on disk to back up. " +
+        "Point `database.url` at a file first.",
+    );
+  }
+  if (!isSqliteUrl(source)) {
+    throw new BackupError(
+      `db:backup takes SQLite snapshots, and this app is on ${source.split("://")[0]}. ` +
+        `Use that server's own tool — pg_dump or mysqldump — and keep its output somewhere ` +
+        `this command is not responsible for.`,
+    );
+  }
+
+  const path = sqlitePathFromUrl(source);
+  try {
+    statSync(path);
+  } catch {
+    throw new BackupError(`There is no database at ${path} to back up.`);
+  }
+
+  mkdirSync(dir, { recursive: true });
+  const target = join(dir, snapshotName(path, options.now ?? new Date()));
+
+  // Never write over a snapshot. Two runs inside the same second is the only way
+  // to get here, and silently replacing the first one loses a backup to a clock.
+  try {
+    statSync(target);
+    throw new BackupError(
+      `${target} already exists. A snapshot is never overwritten — wait a second and run again.`,
+    );
+  } catch (error) {
+    if (error instanceof BackupError) throw error;
+  }
+
+  try {
+    await run(`VACUUM INTO ${sqlLiteral(target)}`);
+  } catch (error) {
+    throw new BackupError(
+      `VACUUM INTO failed, so no snapshot was written: ${(error as Error).message}`,
+    );
+  }
+
+  let bytes: number;
+  try {
+    bytes = statSync(target).size;
+  } catch {
+    throw new BackupError(
+      `VACUUM INTO reported success but wrote no file at ${target}. Nothing has been backed up.`,
+    );
+  }
+  if (bytes === 0) {
+    rmSync(target, { force: true });
+    throw new BackupError(`The snapshot at ${target} was empty, and has been removed.`);
+  }
+
+  // A snapshot that fails its checks does not get to stay. Left on disk it is
+  // indistinguishable from a good one — and it is the *newest*, so the next run's
+  // retention would prune a verified older snapshot to make room for it. The
+  // directory must contain only files that passed.
+  let tables: string[];
+  let rows: Record<string, number>;
+  const rehearsed = options.rehearse === true;
+  try {
+    ({ tables, rows } = verifySnapshot(target, requireRows));
+    if (rehearsed) rehearseRestore(target, requireRows);
+  } catch (error) {
+    rmSync(target, { force: true });
+    throw error;
+  }
+
+  // Retention runs last, and only after the new snapshot has checked out. Pruning
+  // first would mean a failed backup that also deleted the oldest good one.
+  const pruned = pruneSnapshots(dir, keep);
+
+  return { path: target, bytes, tables, rows, pruned, rehearsed };
+}

package/src/config.ts

@@ -28,17 +28,19 @@ export interface DatabaseConfigShape {
    *   env('REPLICA_2_URL'),
    * ].filter((url) => url !== undefined)
    */
-  replicas?: string[];
+  replicas?: string[] | undefined;
   /**
    * Connection pool options (PostgreSQL and MySQL only).
    * Bun.sql manages the pool automatically - these tune its behaviour.
    */
-  pool?: {
-    /** Maximum number of connections in the pool. Default: 10 */
-    max?: number;
-    /** Seconds an idle connection is kept before being closed. Default: 30 */
-    idleTimeout?: number;
-  };
+  pool?:
+    | {
+        /** Maximum number of connections in the pool. Default: 10 */
+        max?: number;
+        /** Seconds an idle connection is kept before being closed. Default: 30 */
+        idleTimeout?: number;
+      }
+    | undefined;
   /** SQLite-specific options */
   sqlite: {
     /** Path to the SQLite file. Use ':memory:' for in-memory database. */
@@ -58,7 +60,7 @@ export interface DatabaseConfigShape {
    * synchronize: { enabled: true, disruptive: false }     // explicit, additive
    * synchronize: { enabled: true, disruptive: true }      // also drops removed columns
    */
-  synchronize?: boolean | { enabled: boolean; disruptive?: boolean };
+  synchronize?: boolean | { enabled: boolean; disruptive?: boolean } | undefined;
 }
 
 const defaults: DatabaseConfigShape = {

package/src/db/DB.ts

@@ -22,7 +22,11 @@ import {
 // Production code resolves the connection from the container instead.
 let _connection: SQLInstance | undefined;
 
-/** Test helper: inject a primary connection without going through the container/provider. */
+/**
+ * Test helper: inject a primary connection without going through the container/provider.
+ *
+ * @internal
+ */
 export function _setDbConnection(conn: SQLInstance | null): void {
   _connection = conn ?? undefined;
 }
@@ -34,6 +38,8 @@ export function _setDbConnection(conn: SQLInstance | null): void {
  * a helper that installs its own connection needs to restore the *override
  * slot* exactly as it found it, and cannot tell an absent override from a
  * container-resolved connection otherwise.
+ *
+ * @internal
  */
 export function _getDbConnectionOverride(): SQLInstance | null {
   return _connection ?? null;
@@ -42,6 +48,8 @@ export function _getDbConnectionOverride(): SQLInstance | null {
 /**
  * Test helper: inject a primary + replicas as a read/write router.
  * Resets to a plain primary when `replicas` is empty.
+ *
+ * @internal
  */
 export function _setReadReplicas(primary: SQLInstance, replicas: SQLInstance[]): void {
   _connection = replicas.length > 0 ? createReadWriteRouter(primary, replicas) : primary;
@@ -90,7 +98,11 @@ export function _getScopedDbConnection(): SQLInstance {
   return conn;
 }
 
-/** Alias used by migration command helpers. */
+/**
+ * Alias used by migration command helpers.
+ *
+ * @internal
+ */
 export function _getConnection(): SQLInstance {
   return _getDbConnection();
 }
@@ -132,6 +144,19 @@ function _runRaw(conn: SQLInstance, sql: string): Promise<unknown> {
   return conn(arr);
 }
 
+/**
+ * Run a raw, parameter-less statement on the ambient connection.
+ *
+ * The tagged-template path rather than `DB.raw(string)`, because that one splits
+ * its input on `?` to find bindings — fine for SQL somebody typed, wrong for a
+ * statement carrying a filesystem path that the caller has already made safe.
+ *
+ * @internal Used by `db:backup` for `VACUUM INTO`.
+ */
+export function _rawStatement(sql: string): Promise<unknown> {
+  return _runRaw(_getConnection(), sql);
+}
+
 /** Run a parameterised statement (`?` placeholders) on a specific connection. */
 function _runParams(conn: SQLInstance, sql: string, params: unknown[]): Promise<unknown> {
   const parts = sql.split("?");

package/src/db/NPlusOneDetector.ts

@@ -104,12 +104,12 @@ let _forced = false; // true when explicitly configured via preventNPlusOne()
 
 export interface NPlusOneOptions {
   /** How many identical query shapes in one request triggers a violation. Default: 5. */
-  threshold?: number;
+  threshold?: number | undefined;
   /**
    * 'warn'  → console.warn (default — non-breaking, still visible)
    * 'throw' → throws NPlusOneError (good for CI / strict mode)
    */
-  mode?: "warn" | "throw";
+  mode?: "warn" | "throw" | undefined;
 }
 
 /**

package/src/db/ReadWriteRouter.ts

@@ -24,6 +24,8 @@ import type { SQLInstance } from "./sql-types.ts";
  *
  * // Drop-in: pass router wherever you'd pass a SQL connection
  * DB.table('users')  // SELECT → replica, mutating → primary
+ *
+ * @internal
  */
 export function createReadWriteRouter(primary: SQLInstance, replicas: SQLInstance[]): SQLInstance {
   if (replicas.length === 0) return primary;

package/src/db/dialects/MysqlDialect.ts

@@ -3,6 +3,8 @@ import type { DatePart, DialectQuery, SqlDialect } from "./types.ts";
 /**
  * MySQL strategy — INFORMATION_SCHEMA introspection, DAY()/MONTH()/YEAR()
  * date parts, GET_LOCK()/RELEASE_LOCK() named advisory locks.
+ *
+ * @internal
  */
 export class MysqlDialect implements SqlDialect {
   readonly name = "mysql" as const;

package/src/db/dialects/PostgresDialect.ts

@@ -3,6 +3,8 @@ import type { DatePart, DialectQuery, SqlDialect } from "./types.ts";
 /**
  * PostgreSQL strategy — information_schema introspection, EXTRACT()/casts for
  * date parts, pg_advisory_lock for advisory locks.
+ *
+ * @internal
  */
 export class PostgresDialect implements SqlDialect {
   readonly name = "postgres" as const;

package/src/db/dialects/SqliteDialect.ts

@@ -3,6 +3,8 @@ import type { DatePart, DialectQuery, SqlDialect } from "./types.ts";
 /**
  * SQLite strategy — sqlite_master / pragma_table_info introspection,
  * strftime() date parts, no advisory-lock primitive.
+ *
+ * @internal
  */
 export class SqliteDialect implements SqlDialect {
   readonly name = "sqlite" as const;

package/src/db/dialects/index.ts

@@ -14,6 +14,8 @@ const _dialects: Record<DialectName, SqlDialect> = {
  *
  * @example
  * getDialect(dialectFor(conn)).dateExpr('day', 'created_at')
+ *
+ * @internal
  */
 export function getDialect(name: DialectName): SqlDialect {
   return _dialects[name];

package/src/db/dialects/types.ts

@@ -7,10 +7,18 @@
 // consult the active dialect instead of hard-coding SQLite syntax behind the
 // multi-dialect facade.
 
-/** Supported database engines. Mirrors `Dialect` in QueryBuilder.ts. */
+/**
+ * Supported database engines. Mirrors `Dialect` in QueryBuilder.ts.
+ *
+ * @internal
+ */
 export type DialectName = "sqlite" | "postgres" | "mysql";
 
-/** A parameterised statement: `sql` uses `?` placeholders bound from `params`. */
+/**
+ * A parameterised statement: `sql` uses `?` placeholders bound from `params`.
+ *
+ * @internal
+ */
 export interface DialectQuery {
   sql: string;
   params: unknown[];
@@ -25,6 +33,8 @@ export type DatePart = "date" | "time" | "day" | "month" | "year";
  * @example
  * const d = getDialect("postgres");
  * const { sql, params } = d.hasTableSql("users");
+ *
+ * @internal
  */
 export interface SqlDialect {
   readonly name: DialectName;

package/src/db/resolver.ts

@@ -11,6 +11,8 @@ let _resolver: ConnectionResolver | null = null;
 /**
  * Register the callback used to look up the base connection from the container.
  * Pass `null` to clear it. Set once by the `DatabaseProvider` at boot.
+ *
+ * @internal
  */
 export function setConnectionResolver(fn: ConnectionResolver | null): void {
   _resolver = fn;
@@ -19,6 +21,8 @@ export function setConnectionResolver(fn: ConnectionResolver | null): void {
 /**
  * Resolve the container's base connection via the registered resolver, swallowing
  * any resolver error. Returns `undefined` when no resolver is set or it fails.
+ *
+ * @internal
  */
 export function resolveContainerConnection(): SQLInstance | undefined {
   if (!_resolver) return undefined;

package/src/db/types.ts

@@ -159,7 +159,7 @@ function _pageUrl(baseUrl: string, page: number, extra: Record<string, string> =
 /** Options accepted by the pagination helper factories. */
 export interface PaginationHelperOptions {
   /** Base path used by no-argument URL helpers and `meta.path`. Default: `''`. */
-  path?: string;
+  path?: string | undefined;
 }
 
 /** Attach URL helper methods + metadata to a plain paginate result object. */
@@ -276,13 +276,13 @@ export interface KeysetOptions {
    * Opaque cursor string from the previous page's `nextCursor`.
    * `null` or omitted = first page.
    */
-  cursor?: string | null;
+  cursor?: string | null | undefined;
   /** Column used for keyset ordering. Must be a safe SQL identifier. Default: `'id'`. */
-  column?: string;
+  column?: string | undefined;
   /** Sort direction. Default: `'asc'`. */
-  direction?: "asc" | "desc";
+  direction?: "asc" | "desc" | undefined;
   /** Items per page. Default: `15`. */
-  limit?: number;
+  limit?: number | undefined;
 }
 
 /** Result of {@link QueryBuilder.keysetPaginate} — opaque-cursor keyset pagination. */

package/src/diagnostics/pendingMigrationsCheck.ts

@@ -0,0 +1,60 @@
+/**
+ * `zt doctor` — are there migrations on disk that have not run?
+ *
+ * The same question the dev error overlay answers, asked before anything breaks.
+ * The overlay is reactive by nature: it needs a request to have already failed
+ * with `no such table: assets`, which means somebody already lost the thread of
+ * what they were doing. `doctor` is where that finding belongs when nothing has
+ * gone wrong yet — it already boots the app, already holds a connection, and
+ * already exists to report the silent things.
+ *
+ * A warning, never a failure. Pending migrations are the *normal* state of a
+ * checkout that just pulled, and a doctor that fails there would be one people
+ * learn to ignore. The deploy pipeline is the place where they must be applied,
+ * and it runs `migrate` as a step rather than asking about it.
+ *
+ * @module
+ */
+import type { DoctorCheck, DoctorCheckResult } from "@zerotal/core";
+import { pendingMigrations } from "./missingRelation.ts";
+
+/** How many migration names to print before summarising the rest. */
+const NAMED = 5;
+
+/**
+ * The check.
+ *
+ * Silent when nothing is pending, silent when there is no database to ask, and
+ * specific when there is something to say — the names, because "3 pending" sends
+ * you to `migrate:status` to find out which, and this already knows.
+ */
+export const pendingMigrationsCheck: DoctorCheck = {
+  id: "pending-migrations",
+  label: "Migrations",
+  async run(): Promise<DoctorCheckResult> {
+    let pending: string[];
+    try {
+      pending = await pendingMigrations();
+    } catch {
+      // No connection, no migrations directory, or a driver that will not answer.
+      // None of those is a finding about migrations, and guessing would make this
+      // check the noisiest thing in the report on every app that has no database.
+      return { status: "ok", message: "no migration state to read." };
+    }
+
+    if (pending.length === 0) {
+      return { status: "ok", message: "every migration on disk has run." };
+    }
+
+    const shown = pending.slice(0, NAMED).join(", ");
+    const rest = pending.length - NAMED;
+    return {
+      status: "warn",
+      message:
+        `${pending.length} migration(s) have not run: ${shown}${rest > 0 ? `, and ${rest} more` : ""}. ` +
+        `Until they do, any query touching what they create fails with \`no such table\` — ` +
+        `an error whose stack is entirely framework frames.`,
+      fix: "bun zt migrate",
+    };
+  },
+};

package/src/implicitBinding.ts

@@ -68,6 +68,7 @@ export function resolverForParam(paramName: string): ModelBindingResolver | unde
       : M.findOrFail(value);
 }
 
+/** @internal */
 export function registerImplicitBinding(): void {
   setImplicitModelResolver(resolverForParam);
 }