full-text-search

sparq full-text search (sparq-text)

sparq-text is an opt-in, separate crate that adds full-text search over the string literals of a sparq Graph. It gives you two surfaces: a low-level owned BM25 inverted index (TextIndex) keyed by dictionary term id, and text: magic predicates that you write inside ordinary SPARQL and that query_text rewrites into inline VALUES over the index's hits. The engine, planner, and the lean sparq-wasm bundle carry zero text-search code — full-text support exists only when you depend on sparq-text (or load its dedicated tier-b sparq-text-wasm browser bundle; see below).

Quickstart

Cargo.toml:

[dependencies]
sparq-core = { path = "../sparq-core" }
sparq-text = { path = "../sparq-text" }   # default features: ["engine", "parallel"]

use sparq_core::Graph;
use sparq_text::{query_text, TextIndex};

let g = Graph::load_str(r#"
    <http://ex/post1> <http://ex/title> "The quick brown fox" .
    <http://ex/post2> <http://ex/title> "Fox hunting banned" .
"#, "ntriples").unwrap();

// 1. Library-level: build the BM25 index (one dict scan), search returns Hit { id, score }.
let index = TextIndex::build(&g);
for hit in index.search("quick fox") {          // AND of tokens, BM25 best-first
    let literal = g.dict.term(hit.id);           // hit.id IS the dict id of the literal
    println!("{literal} (bm25 {})", hit.score);
}

// 2. SPARQL-level: text: magic predicates, rewritten + executed by the engine.
let r = query_text(&g, r#"
    PREFIX text: <http://sparq.dev/text#>
    SELECT ?post ?s WHERE {
      ?post  <http://ex/title> ?title .
      ?title text:matches "fox" .
      ?title text:score   ?s .
    } ORDER BY DESC(?s)"#, &index).unwrap();
assert_eq!(r.rows.len(), 2);                     // r.rows: Vec<Vec<Option<oxrdf::Term>>>, r.vars

The dictionary term id of a string literal is its document id, so search results join back to subjects/predicates through the store's ordinary permutation indexes — that join is what the SPARQL surrounding the magic pattern does.