Open-source manufacturing ERP, MES and QMS. Quoting, MRP, inventory, shop floor, quality and lot/serial traceability on one Postgres schema, with a REST API and MCP server. Self-host or use Carbon Cloud.
TypeScript ai-agentsbill-of-materialserpinventory-managementleanmanufacturingmanufacturing-execution-systemmcp-servermesmrpopen-source-erppostgresqlproduction-planningqmsquality-management-systemreact-routerself-hostedsupabasetraceabilitytypescript
// SPDX-License-Identifier: AGPL-3.0-only
// Carbon (github.com/crbnos/carbon). Modified or adapted versions of this file,
// including ports, remain AGPLv3; serving them over a network requires releasing their source.
import * as pg from "pg";
import { describe, expect, it } from "vitest";
/**
* Importing the postgres client registers node-postgres' type parsers as a
* module-load side effect. The import is the subject of these tests, not a
* dependency of them — hence the bare side-effect import.
*/
import "./client.ts";
const NUMERIC_OID = 1700;
const DATE_OID = 1082;
const TIMESTAMPTZ_OID = 1184;
const TIMESTAMP_OID = 1114;
type TypeRegistry = {
getTypeParser: (oid: number, format?: string) => (value: string) => unknown;
};
/**
* `pg.types`, reached the same way the production registration reaches it. Both
* the namespace and its `default` carry it depending on interop, and the
* production code uses `?.` — so a miss would make the registration a SILENT
* no-op. Asserting a real decode below is what proves it actually took.
*/
const types = ((pg as unknown as { types?: TypeRegistry }).types ??
(pg as unknown as { default?: { types?: TypeRegistry } }).default
?.types) as TypeRegistry;
/**
* These pin the ONE invariant the generated types cannot: that what the driver
* hands back matches what `KyselyDatabase` declares. A `date` column is
* `string | null` to the compiler, so a `Date` at runtime type-checks clean all
* the way to the wire — which is how a `.slice` crash (Rillet payments) and
* `422 "Not a valid date"` (every Ramp draft bill) both shipped.
*/
describe("postgres type parsers", () => {
it("decodes DATE as the raw YYYY-MM-DD string, not a JS Date", () => {
const parsed = types.getTypeParser(DATE_OID)("2026-09-15");
expect(parsed).toBe("2026-09-15");
expect(parsed).not.toBeInstanceOf(Date);
});
it("does not shift the day in a timezone behind UTC", () => {
// The regression this exists for: `postgres-date` builds LOCAL midnight, so
// a naive `toISOString().slice(0, 10)` reported 2026-09-14 anywhere west of
// Greenwich. A raw string has no timezone to get wrong.
expect(types.getTypeParser(DATE_OID)("2026-01-01")).toBe("2026-01-01");
expect(types.getTypeParser(DATE_OID)("2026-12-31")).toBe("2026-12-31");
});
it("matches what PostgREST returns for the same column", () => {
// Kysely and the Supabase client must agree, or a value that round-trips
// through one breaks through the other.
expect(types.getTypeParser(DATE_OID)("2026-09-15")).toBe("2026-09-15");
});
it("still decodes NUMERIC as a number", () => {
expect(types.getTypeParser(NUMERIC_OID)("12.34567")).toBe(12.34567);
});
it("leaves the timestamp OIDs alone", () => {
// Deliberate: Postgres' timestamptz wire text is space-separated with a
// `+00` offset, where PostgREST emits ISO-8601 with `T` and `+00:00`. An
// identity parser here would make the two clients disagree on the SHAPE of
// the string. Changing these needs a normalizing parser, not this one.
expect(
types.getTypeParser(TIMESTAMPTZ_OID)("2026-09-15 16:36:52.677+00")
).toBeInstanceOf(Date);
expect(
types.getTypeParser(TIMESTAMP_OID)("2026-09-15 16:36:52.677")
).toBeInstanceOf(Date);
});
});