@schemat/parser-dbml

DBML parser for Schemat: turns a DBML schema (the format used by dbdiagram.io) into the canonical Schemat IR.

Install

npm i @schemat/parser-dbml

Usage

import { dbmlParser } from "@schemat/parser-dbml";

// Detect + parse a project's DBML file into an IRSchema:
const ir = await dbmlParser.parse({ projectPath: "." });

// Or point at a specific file:
const ir2 = await dbmlParser.parse({ projectPath: ".", files: ["docs/schema.dbml"] });

dbmlParser implements the SchemaParser interface (detect + parse) from @schemat/core and returns an IRSchema.

Detection

The parser looks for a DBML file at (in order): schema.dbml, database.dbml, dbml/schema.dbml, docs/schema.dbml, db/schema.dbml, or any *.dbml file in the project root.

What it maps

• Tables → tables (with table Note as the comment)
• Fields → columns (type, pk, unique, not null, defaults, field note)
• Composite pk indexes → primary-key columns
• Enums → enums
• Refs (inline [ref: > ...] and standalone Ref:) → relations, with cardinality inferred from the DBML relation symbols (> / < = one-to-many, - = one-to-one, <> = many-to-many)

Limitations (v1)

• TablePartial is not supported. The parser runs @dbml/core in its dbml mode, which rejects the newer TablePartial block and the ~partial injection syntax outright — a schema using them fails to parse rather than losing a few fields. Inline the shared columns into each table until upstream support lands.
• Composite uniques are not represented. A composite pk index promotes its columns to primary keys, but a composite unique index is dropped: marking each column individually unique would be a stronger claim than the schema makes.
• Refs with unsupported endpoints are skipped, with a warning naming the ref and its schema, rather than emitted as a guessed edge.
• Table groups, project blocks, and sticky notes carry no IR meaning and are ignored. Table and field Notes are kept, as comments.

License

MIT © Ali Reza Hamid