@schemat/parser-sql

SQL parser for Schemat: turns SQL DDL (CREATE TABLE ...) into the canonical Schemat IR.

Install

npm i @schemat/parser-sql

Usage

import { parseSql, sqlParser } from "@schemat/parser-sql";

// Parse a raw DDL string:
const ir = parseSql("CREATE TABLE users (id INT PRIMARY KEY);");

// Or use the SchemaParser interface (detect + parse a project dir / file):
const ir2 = await sqlParser.parse({ projectPath: ".", files: ["schema.sql"] });

Both return an IRSchema from @schemat/core.

What it maps

• CREATE TABLE → tables, columns (type, NOT NULL, DEFAULT, PRIMARY KEY, UNIQUE), plus inline REFERENCES and table-level FOREIGN KEY → relations
• CREATE TYPE ... AS ENUM (Postgres) → enums
• ALTER TABLE ... ADD CONSTRAINT → primary keys, uniques, and foreign keys applied back onto the tables
• CREATE UNIQUE INDEX on a single column → isUnique

Limitations (v1)

• Cardinality is always one-to-many for SQL foreign keys. Uniqueness of the owning column is not yet used to promote an FK to one-to-one.
• Composite and partial uniqueness is not represented. A multi-column UNIQUE index, or a CREATE UNIQUE INDEX ... WHERE ..., does not make its columns individually unique, so it is deliberately dropped rather than overstated.
• CHECK constraints are skipped (with a warning); they have no place in the IR.
• Views, functions, triggers, sequences, grants and similar statements are ignored on purpose. Anything else the parser cannot read is skipped and summarised as one warning per statement kind, so a large dump does not produce thousands of lines of noise.
• Detection only claims a project from schema.sql, db/schema.sql, sql/schema.sql, or a root *.sql that actually contains CREATE TABLE — a seed.sql of INSERTs will not hijack a Drizzle or Prisma project.

License

MIT © Ali Reza Hamid