Struct PostingList 

pub struct PostingList { /* private fields */ }

In-memory inverted index. See module-level doc.

Implementations

impl PostingList

pub fn new() -> Self

Empty index with no postings and no documents.

pub fn len(&self) -> usize

Number of indexed documents.

pub fn is_empty(&self) -> bool

True iff no document has been inserted (or all have been removed).

pub fn avg_doc_len(&self) -> f64

Average document length in tokens. Returns 0.0 when the index is empty so BM25 can guard cleanly without a div-by-zero.

pub fn serialize_doc_lengths(&self) -> Vec<(i64, u32)>

Phase 8c — emit (rowid, doc_len) pairs for every indexed doc, in ascending rowid order. The pager writes these into the FTS index’s doc-lengths sidecar cell; reload feeds them back to Self::from_persisted_postings.

pub fn serialize_postings(&self) -> Vec<(String, Vec<(i64, u32)>)>

Phase 8c — emit (term, [(rowid, term_freq)]) triples in lexicographic term order; per-term entries are in ascending rowid order (the underlying BTreeMap already guarantees this). One element per unique indexed term; pager writes one cell per element.

pub fn from_persisted_postings<I, J>(doc_lengths: I, postings: J) -> Self

where I: IntoIterator<Item = (i64, u32)>, J: IntoIterator<Item = (String, Vec<(i64, u32)>)>,

Phase 8c — rebuild a PostingList directly from the persisted doc-lengths sidecar + per-term postings. No tokenization runs; the resulting index is byte-equivalent to what was saved (assuming the input came from serialize_*).

doc_lengths is the full (rowid, doc_len) map written into the sidecar cell. postings is one (term, [(rowid, tf)]) element per term cell.

pub fn insert(&mut self, rowid: i64, text: &str)

Tokenize text and add its postings under rowid. If rowid is already indexed, its previous postings are removed first — i.e. insert is idempotent for re-indexing the same row.

A row whose tokenization yields zero tokens is still recorded (with doc_len = 0 and no posting entries). This keeps len() honest for “indexed but empty” rows; BM25 returns 0.0 for them.

pub fn remove(&mut self, rowid: i64)

Remove all postings for rowid. No-op if rowid was never inserted. Empty per-term posting lists left behind by the last referencing row are pruned to keep the BTreeMap tight.

pub fn matches(&self, rowid: i64, query: &str) -> bool

True iff rowid is indexed and at least one of its terms is in the (tokenized) query. Powers fts_match(col, 'q') in 8b without going through scoring.

pub fn score(&self, rowid: i64, query: &str, params: &Bm25Params) -> f64

BM25 score for a single (rowid, query) pair. Returns 0.0 if rowid is unknown or no query terms hit.

pub fn query(&self, query: &str, params: &Bm25Params) -> Vec<(i64, f64)>

Score every doc that contains at least one query term and return (rowid, score) sorted by score descending, ties broken by rowid ascending. Powers the bulk path used by 8b’s try_fts_probe optimizer hook.

Empty query → empty result. Empty index → empty result. Rows that don’t match any query term are not scored at all (they would score 0.0 — including them just bloats the result).

Trait Implementations

impl Clone for PostingList

fn clone(&self) -> PostingList

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for PostingList

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

Formats the value using the given formatter.

impl Default for PostingList

fn default() -> PostingList

Returns the “default value” for a type.