Struct SearchIndex 

pub struct SearchIndex { /* private fields */ }

Inverted word index for fast entity search.

For each entity, we tokenize its name, type, and observations, store each token → set of matching entity indices.

Uses a flat Vec<(StrId, u32)> sorted by (token, entity_idx) for cache-friendly lookups via binary search.

Implementations

impl SearchIndex

pub fn new() -> Self

pub fn clear(&mut self)

pub const fn len(&self) -> usize

pub const fn is_empty(&self) -> bool

pub fn index_entity( &mut self, interner: &mut StringInterner, entity_idx: u32, name: StrId, entity_type: StrId, observations: &[StrId], )

Index a single entity by its name, type, and observations. All strings must already be interned. entity_idx is the position in the entity storage vec.

pub fn index_additional( &mut self, interner: &mut StringInterner, entity_idx: u32, texts: &[StrId], )

Incrementally index additional strings (e.g. newly added observations) for an entity that is already indexed, without removing and rebuilding its existing entries (P3). Token entries that already exist are deduped during the merge, so calling this with text that overlaps existing tokens is safe.

pub fn remove_entity(&mut self, entity_idx: u32)

Remove all entries for a given entity (before re-indexing).

pub fn search(&self, query: &str, interner: &StringInterner) -> Vec<u32>

Search for entities whose name/type/observation tokens match query case-insensitively by prefix ("cof" matches "coffee").

Note: this is an O(n) scan over every index entry. The binary-search step below only narrows exact-token hits, but the subsequent prefix scan already covers those (an exact match is also a prefix match), so the scan dominates — do not read the binary search as making this sublinear.

pub fn search_ranked( &self, query: &str, interner: &StringInterner, ) -> Vec<(u32, u32)>

Like [search], but returns (entity_idx, score) pairs sorted by descending score (then ascending idx for stability). score is the number of indexed-token hits the entity accumulated for the query — a cheap relevance proxy so callers can surface the best matches first.

The scan is a single linear pass over the flat entries vec (no per-entity allocation until the final compaction), keeping it cache-friendly. A small Vec<(idx, score)> is gathered then sorted.

Trait Implementations

impl Clone for SearchIndex

fn clone(&self) -> SearchIndex

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Default for SearchIndex

fn default() -> Self

Returns the “default value” for a type.