Struct Bm25Index 

pub struct Bm25Index<'a> { /* private fields */ }

The BM25 index: postings with term frequencies plus per-document lengths and corpus statistics.

Implementations

impl<'a> Bm25Index<'a>

pub fn new(shards: usize, max_bytes: usize) -> Result<Self, Error>

Creates an empty index; shards per the engine config (shards_postings), max_bytes bounds each underlying pool.

pub fn index_doc( &mut self, fact: FactId, term_tfs: &[(u32, u8)], ) -> Result<(), Error>

Indexes one document given its (term, tf) pairs (the caller tokenizes and counts; pairs may arrive in any order, terms must be unique). Documents must arrive in ascending fact-id order.

Errors

Error::Arena when a pool hits its byte ceiling; the index may then hold the document partially — the engine treats that as fatal for the whole operation (journal replay rebuilds consistently).

pub fn df(&self, term: u32) -> u32

Document frequency of a term.

pub fn docs(&self) -> u64

Number of indexed documents.

pub fn idf(&self, df: u32) -> f32

Robertson idf for a term with document frequency df in this corpus (monotonically decreasing in df, always positive).

pub fn search( &self, (k1, b): (f32, f32), terms: &[u32], k: usize, live: &mut dyn FnMut(FactId) -> bool, scratch: &mut Bm25Scratch, out: &mut Vec<(FactId, f32)>, )

Scores terms against the corpus and writes the top k live documents into out (descending score, ties by ascending id). live filters candidates (tombstones, as-of, tag allow-sets) — filtered documents cost their posting decode but never rank.

Duplicate query terms are the caller’s choice: each occurrence accumulates again (a term repeated in the query weighs more).

pub fn pool_bytes(&self) -> usize

Bytes held by the underlying pools.

pub fn total_len(&self) -> u64

Total token count across the corpus (persisted in the engine state).

Trait Implementations

impl<'a> Debug for Bm25Index<'a>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.