@datafn/core

Core types, schema validation, DFQL normalization, and shared utilities for the DataFn ecosystem. Every other DataFn package depends on @datafn/core.

Installation

npm install @datafn/core

Features

• Type Definitions — Complete TypeScript types for schemas, resources, fields, relations, events, signals, and plugins
• Const-Safe Field Builders — Reusable field definitions that retain literal names, defaults, and enum members
• Schema Validation — Runtime validation and normalization of DataFn schemas
• DFQL Types — Query, mutation, and transaction type definitions
• DFQL Normalization — Deterministic normalization for stable cache keys
• Envelope Pattern — Structured ok | error result types with helper functions
• KV Utilities — Built-in key-value resource helpers (ensureBuiltinKv, kvId)
• Error Codes — Enumerated error codes for consistent error handling
• Structural Resource Selectors — Parse request envelopes and collect protocol-level resource selectors for routing and authorization

Schema Definition

A DataFn schema describes your entire data model: resources (tables), their fields, and the relationships between them.

Use defineSchema for the full schema and field when fields need to be reused. Both preserve literal names and metadata for typed clients:

import { defineSchema, field } from "@datafn/core";

const id = field.string("id", {
  required: true,
  unique: true,
});

const sharedFields = [
  field.boolean("isStarred", { default: false }),
  field.string("status", {
    required: true,
    enum: ["NOT_STARTED", "IN_PROGRESS", "COMPLETED"],
  }),
];

export const schema = defineSchema({
  resources: [
    {
      name: "tasks",
      version: 1,
      fields: [id, ...sharedFields],
      indices: { base: ["id", "status"] },
    },
  ],
});

Field builders always return plain schema objects and emit required: false and nullable: false when those options are omitted. Requiredness controls whether a property must exist. Nullability independently controls whether its value may be null.

Inline object literals passed directly to defineSchema receive the same record inference. Avoid annotating a schema or reusable field collection as DatafnSchema or DatafnFieldSchema[] before calling defineSchema; those annotations widen literal field and resource names. Use satisfies when you only need an assignability check, or use the builders for reusable fields.

Builder options are specific to each field kind. Defaults and enum members must match that kind, and options cannot override name or type. Constraint metadata such as enum, bounds, patterns, uniqueness, encryption, volatility, and read-only flags is preserved for adapters and generated schemas. Core mutation validation currently enforces field type and nullability, not every constraint metadata value.

DatafnSchema

type DatafnSchema = {
  resources: DatafnResourceSchema[];
  relations?: DatafnRelationSchema[];
};

DatafnResourceSchema

Each resource maps to a table or collection in your database.

type DatafnResourceSchema = {
  /** Unique resource name (e.g. "todos", "users") */
  name: string;
  /** Schema version — increment when making breaking changes */
  version: number;
  /** Optional prefix for generated IDs (e.g. "todo" → "todo:uuid") */
  idPrefix?: string;
  /**
   * When true, the resource is never stored locally.
   * Queries always go to the remote server.
   */
  isRemoteOnly?: boolean;
  /** Field definitions */
  fields: DatafnFieldSchema[];
  /**
   * Index hints for optimisation.
   * Can be a simple string[] (treated as base indices) or a structured object.
   */
  indices?:
    | { base?: string[]; search?: string[]; vector?: string[] }
    | string[];
  /** Optional permissions policy for server-side authorization */
  permissions?: DatafnPermissionsPolicy;
};

DatafnFieldSchema

Every field has a name, type, and a rich set of optional validation constraints.

type DatafnFieldSchema = {
  name: string;
  type:
    | "string"
    | "number"
    | "boolean"
    | "object"
    | "array"
    | "date"
    | "file"
    | "json";
  required: boolean;
  /** Allow explicit null values */
  nullable?: boolean;
  /** Prevent mutation after initial insert */
  readonly?: boolean;
  /** Default value applied on insert when the field is omitted */
  default?: unknown;
  /** Restrict to a fixed set of allowed values */
  enum?: unknown[];
  /** Minimum numeric value or minimum array length */
  min?: number;
  /** Maximum numeric value or maximum array length */
  max?: number;
  /** Minimum string length */
  minLength?: number;
  /** Maximum string length */
  maxLength?: number;
  /** Regex pattern the string value must match */
  pattern?: string;
  /**
   * Uniqueness constraint.
   * - `true` — globally unique
   * - `string` — unique within a composite group
   */
  unique?: boolean | string;
  /** Encrypt the field value at rest */
  encrypt?: boolean;
  /** Volatile fields are excluded from sync and persistence */
  volatile?: boolean;
};

Supported field types:

DatafnRelationSchema

Relations describe how resources are connected.

type DatafnRelationSchema = {
  /** Source resource name(s) */
  from: string | string[];
  /** Target resource name(s) */
  to: string | string[];
  /** Relation cardinality */
  type: "one-many" | "many-one" | "many-many" | "htree";
  /** Forward relation name (e.g. "tags") */
  relation?: string;
  /** Inverse relation name (e.g. "todos") */
  inverse?: string;
  /** Cache relation data for faster reads */
  cache?: boolean;
  /** Extra metadata fields on the join row (many-many only) */
  metadata?: Array<{
    name: string;
    type: "string" | "number" | "boolean" | "date" | "object";
  }>;
  /** Foreign key field name (many-one / one-many) */
  fkField?: string;
  /** Materialized path field (htree) */
  pathField?: string;
};

Relation types:

DatafnPermissionsPolicy

type DatafnPermissionsPolicy = {
  read?: { fields: string[] };
  write?: { fields: string[] };
  ownerField?: string;
};

Events

Events represent lifecycle notifications for mutations and sync operations.

DatafnEvent

interface DatafnEvent {
  type:
    | "mutation_applied" // A mutation was successfully applied
    | "mutation_rejected" // A mutation was rejected (validation, conflict, etc.)
    | "sync_applied" // A sync operation completed successfully
    | "sync_failed"; // A sync operation failed
  resource?: string; // Affected resource name
  ids?: string[]; // Affected record IDs
  mutationId?: string; // Mutation identifier
  clientId?: string; // Originating client identifier
  timestampMs: number; // Event timestamp in milliseconds
  context?: unknown; // Arbitrary context data
  action?: string; // Mutation action (insert, merge, delete, etc.)
  fields?: string[]; // Changed fields
}

DatafnEventFilter

Filter which events you receive when subscribing.

type DatafnEventFilter = Partial<{
  type: DatafnEvent["type"] | Array<DatafnEvent["type"]>;
  resource: string | string[];
  ids: string | string[];
  mutationId: string | string[];
  action: string | string[];
  fields: string | string[];
  contextKeys: string[];
  context: Record<string, unknown>;
}>;

Signals

Signals are reactive data containers that represent live query results. They are the bridge between DataFn and your UI framework.

DatafnSignal<T>

interface DatafnSignal<T> {
  /** Get the current value synchronously */
  get(): T;
  /** Subscribe to value changes. Returns an unsubscribe function. */
  subscribe(handler: (value: T) => void): () => void;
  /** True while the initial fetch is in progress */
  readonly loading: boolean;
  /** Non-null if the last fetch/refresh failed */
  readonly error: DatafnError | null;
  /** True while a background refresh is in progress (after initial load) */
  readonly refreshing: boolean;
}

Plugins

Plugins intercept and extend queries, mutations, and sync operations on both client and server.

DatafnPlugin

interface DatafnPlugin {
  name: string;
  runsOn: Array<"client" | "server">;

  /** Intercept queries before execution. Return modified query or throw to reject. */
  beforeQuery?: (
    ctx: DatafnHookContext,
    q: unknown,
  ) => Promise<unknown> | unknown;
  /** Process query results. Return modified result. */
  afterQuery?: (
    ctx: DatafnHookContext,
    q: unknown,
    result: unknown,
  ) => Promise<unknown> | unknown;

  /** Intercept mutations before execution. Return modified mutation or throw to reject. */
  beforeMutation?: (
    ctx: DatafnHookContext,
    m: unknown | unknown[],
  ) => Promise<unknown> | unknown;
  /** React to mutation results. */
  afterMutation?: (
    ctx: DatafnHookContext,
    m: unknown | unknown[],
    result: unknown,
  ) => Promise<void> | void;

  /** Intercept sync operations before execution. Return modified payload or throw to reject. */
  beforeSync?: (
    ctx: DatafnHookContext,
    phase: "seed" | "clone" | "pull" | "push" | "cloneUp" | "reconcile",
    payload: unknown,
  ) => Promise<unknown> | unknown;
  /** React to sync results. */
  afterSync?: (
    ctx: DatafnHookContext,
    phase: "seed" | "clone" | "pull" | "push" | "cloneUp" | "reconcile",
    payload: unknown,
    result: unknown,
  ) => Promise<void> | void;
}

DatafnHookContext

type DatafnHookContext = {
  env: "client" | "server";
  schema: DatafnSchema;
  context?: unknown;
};

DFQL Types

DFQL (DataFn Query Language) defines the structure of queries, mutations, and transactions.

DfqlQuery

type DfqlQuery = {
  resource: string;
  version: number;
  select?: string[]; // Fields to include
  omit?: string[]; // Fields to exclude
  filters?: Record<string, unknown>; // Where clause
  search?: Record<string, unknown>; // Full-text search
  sort?: DfqlSort; // Ordering (e.g. ["-createdAt", "name"])
  limit?: number; // Max results
  offset?: number; // Skip N results
  cursor?: DfqlCursor; // Cursor-based pagination
  count?: boolean; // Return count only
  groupBy?: string[]; // Group by fields
  aggregations?: Record<string, unknown>; // Aggregate functions
  having?: Record<string, unknown>; // Having clause for groups
};

DfqlQueryFragment

Omits resource and version — used with the Table API where those are implicit.

type DfqlQueryFragment = Omit<DfqlQuery, "resource" | "version">;

DfqlMutation

type DfqlMutation = {
  resource: string;
  version: number;
  operation: string; // "insert" | "merge" | "replace" | "delete" | "relate" | "unrelate" | "modifyRelation"
  id?: string | string[]; // Target record ID(s)
  record?: Record<string, unknown>;
  records?: Array<Record<string, unknown>>;
  clientId?: string; // For idempotency
  mutationId?: string; // For idempotency
  timestamp?: number | string;
  context?: unknown;
  relations?: Record<string, unknown>;
  if?: Record<string, unknown>; // Optimistic concurrency guards
  cascade?: unknown;
};

DfqlMutationFragment

Omits resource and version — used with the Table API.

type DfqlMutationFragment = Omit<DfqlMutation, "resource" | "version">;

DfqlTransact

type DfqlTransact = {
  transactionId?: string;
  atomic?: boolean;
  steps: Array<{ query?: DfqlQuery; mutation?: DfqlMutation }>;
};

Structural resource selectors

Gateways and authorization plugins must not recursively inspect request JSON for fields named resource or resources. Application records and filters may legitimately use those names.

import { extractStructuralResourceSelectors } from "@datafn/core";

const extracted = extractStructuralResourceSelectors("query", payload);
if (extracted.ok) {
  extracted.result.selectors; // protocol-level resources only
  extracted.result.protocolVersion; // "1"
}

parseDatafnRequest plus collectStructuralResourceSelectors is the composable form. Unsupported protocolVersion values return DATAFN_UNSUPPORTED_PROTOCOL_VERSION.

Sort & Cursor

type DfqlSort = string[];
// e.g. ["name", "-createdAt"]  → name ASC, createdAt DESC (prefix "-" = descending)

type DfqlCursor = {
  after?: Record<string, unknown>;
  before?: Record<string, unknown>;
};

Functions

validateSchema

Validates and normalizes a DataFn schema. Returns an envelope.

import { validateSchema, unwrapEnvelope } from "@datafn/core";

const schema = unwrapEnvelope(
  validateSchema({
    resources: [
      {
        name: "user",
        version: 1,
        fields: [
          { name: "email", type: "string", required: true, unique: true },
          { name: "name", type: "string", required: true },
        ],
      },
    ],
  }),
);

Normalization applied:

• Converts indices: string[] → { base: string[], search: [], vector: [] }
• Defaults relations to [] if omitted
• Validates unique resource names, field names, and required properties

normalizeDfql

Recursively normalizes a value for deterministic comparison: sorts object keys, removes undefined values, preserves primitives, arrays, and null.

import { normalizeDfql } from "@datafn/core";

normalizeDfql({ b: 2, a: 1, c: undefined });
// → { a: 1, b: 2 }

dfqlKey

Returns a stable JSON string for a DFQL value. Used as cache keys for signals.

import { dfqlKey } from "@datafn/core";

const key = dfqlKey({ resource: "user", filters: { id: "user:1" } });
// Deterministic JSON string suitable for Map/Set keys

unwrapEnvelope

Unwraps a DatafnEnvelope: returns result on success, throws the DatafnError on failure.

import { unwrapEnvelope } from "@datafn/core";

const result = unwrapEnvelope(someEnvelope);
// Throws DatafnError if envelope.ok === false

ok / err

Helpers to create envelopes.

import { ok, err } from "@datafn/core";

const success = ok({ data: [1, 2, 3] });
// { ok: true, result: { data: [1, 2, 3] } }

const failure = err("NOT_FOUND", "User not found", { path: "users" });
// { ok: false, error: { code: "NOT_FOUND", message: "User not found", details: { path: "users" } } }

KV Utilities

The built-in KV (key-value) resource provides a schemaless store that syncs alongside your typed resources.

ensureBuiltinKv

Ensures the schema includes the built-in kv resource. If it already exists, validates its shape; otherwise appends it.

import { ensureBuiltinKv } from "@datafn/core";

const schemaWithKv = ensureBuiltinKv(mySchema);
// Now includes { name: "kv", version: 1, fields: [{ name: "value", type: "object" }] }

kvId

Converts a plain key string to the canonical KV record ID.

import { kvId } from "@datafn/core";

kvId("theme"); // → "kv:theme"
kvId("user:prefs"); // → "kv:user:prefs"

KV_RESOURCE_NAME

The canonical resource name for the built-in KV store: "kv".

Error Handling

DatafnError

A plain object (not a class) with a structured shape.

type DatafnError = {
  code: DatafnErrorCode;
  message: string;
  details?: unknown;
};

DatafnErrorCode

type DatafnErrorCode =
  | "SCHEMA_INVALID"
  | "DFQL_INVALID"
  | "DFQL_UNKNOWN_RESOURCE"
  | "DFQL_UNKNOWN_FIELD"
  | "DFQL_UNKNOWN_RELATION"
  | "DFQL_UNSUPPORTED"
  | "LIMIT_EXCEEDED"
  | "FORBIDDEN"
  | "NOT_FOUND"
  | "CONFLICT"
  | "INTERNAL";

DatafnEnvelope<T>

type DatafnEnvelope<T> =
  | { ok: true; result: T }
  | { ok: false; error: DatafnError };

Full Schema Example

import { defineSchema, field } from "@datafn/core";

const schema = defineSchema({
  resources: [
    {
      name: "project",
      version: 1,
      idPrefix: "proj",
      fields: [
        field.string("id", { required: true, unique: true }),
        field.string("name", { required: true, maxLength: 200 }),
        field.string("description"),
        field.string("ownerId", { required: true }),
        field.date("createdAt", { required: true }),
      ],
      indices: { base: ["ownerId", "createdAt"] },
    },
    {
      name: "task",
      version: 1,
      idPrefix: "task",
      fields: [
        field.string("id", { required: true, unique: true }),
        field.string("title", { required: true, minLength: 1 }),
        field.boolean("completed", { required: true, default: false }),
        field.number("priority", { min: 1, max: 5 }),
        field.date("dueDate"),
        field.string("projectId", { required: true }),
      ],
      indices: { base: ["projectId", "completed"] },
    },
    {
      name: "tag",
      version: 1,
      idPrefix: "tag",
      fields: [
        field.string("id", { required: true, unique: true }),
        field.string("name", { required: true, unique: true }),
        field.string("color"),
      ],
    },
  ],
  relations: [
    {
      from: "task",
      to: "project",
      type: "many-one",
      relation: "project",
      fkField: "projectId",
    },
    {
      from: "task",
      to: "tag",
      type: "many-many",
      relation: "tags",
      inverse: "tasks",
    },
    {
      from: "project",
      to: "task",
      type: "one-many",
      relation: "tasks",
      inverse: "project",
    },
  ],
});

License

MIT

For pull requests, pass trusted application schema metadata as the third argument: extractStructuralResourceSelectors("pull", payload, { schema }) (or parseDatafnRequest("pull", payload, { schema })). The parser translates known many-to-many join cursor keys into their endpoint resources. Without schema metadata, cursor names remain selectors; unknown join_ names are never silently ignored. Gateways handling relational sync must supply the same schema as the server and authorize every returned endpoint. The server preflight supplies its validated schema automatically; custom authorization callbacks should also pass their trusted schema to the helper.