lezer-turtle

An incremental Lezer grammar for RDF 1.2 Turtle — with a trig dialect adding TriG's graph blocks — plus an RDF/JS quad emitter that tracks source provenance on the side.

Being a Lezer grammar, it parses incrementally and recovers from errors, so it slots directly into CodeMirror 6 for highlighting, folding and lint, and keeps producing quads (with span-carrying diagnostics) while a document is mid-edit.

parsing to quads with provenance

import N3 from "n3";
import {parseTurtle} from "lezer-turtle/emit";

const {quads, provenance, diagnostics, prefixes, tree} =
  parseTurtle(text, {
    factory: N3.DataFactory,      // any RDF/JS DataFactory
    baseIRI: "http://x.example/",
    dialect: "trig",              // omit for plain Turtle
  });

provenance is a ProvenanceIndex: a multiset of quad utterances layered over the set of quads. Quads stay ordinary RDF/JS objects; each utterance records where each position came from:

provenance.get(quad)  // → [{quad, subject: Range[], predicate: Range[],
                      //     object: Range[], graph: Range[]}, …]

Range is {start, end} in character offsets. Ranges are arrays because provenance can be split or synthetic: a [ … ] property list's subject is its bracket span, collection cell quads point at their item, and annotation-derived rdf:reifies quads point at the ~reifier/{| … |} syntax. A quad uttered twice has two utterances — the multiset lives here, not in your store.

The index keys on quad value (an N-Quads-shaped canonical form), never on object identity: RDF/JS only promises .equals(), and stores like N3.Store reconstruct quads on read — reconstructed quads still find their utterances.

RDF 1.2

Reified triples (<< s p o ~ r >>), triple terms (<<( s p o )>>), annotation blocks ({| … |}), base-direction language tags ("…"@en--ltr) and version directives (VERSION "1.2" / @version) are parsed and emitted per the 1.2 drafts.

Known v0 limitation: prefixed-name characters are the BMP subset of PN_CHARS_BASE (astral-plane ranges explode Lezer's DFA construction).

tests

test/corpus.json adopts the Turtle/TriG subset of N3.js's parser tests (test/extract-n3-corpus.js regenerates it from a neighboring checkout): released n3 is the oracle for the 1.1 buckets; the 1.2 bucket was verified isomorphic against N3.js's (1.2-capable) main at extraction and frozen. test/prov-test.js covers the utterance semantics.

npm install && npm test