Struct EmbeddingIndex 

pub struct EmbeddingIndex {
    pub version: u32,
    pub dimensions: usize,
    pub model_id: Option<String>,
    pub entries: Vec<EmbeddingEntry>,
    pub file_hashes: HashMap<String, String>,
}

Fields

version: u32

dimensions: usize

model_id: Option<String>

Model identifier that generated these embeddings. Used for mismatch detection when the user switches models.

entries: Vec<EmbeddingEntry>

file_hashes: HashMap<String, String>

Implementations

impl EmbeddingIndex

pub fn new(dimensions: usize) -> Self

pub fn new_with_model(dimensions: usize, model_id: &str) -> Self

Create a new index tagged with a specific model identity.

pub fn model_mismatch<'a>( &'a self, current_model: &'a str, ) -> Option<(&'a str, &'a str)>

Check if the index was built with a different model than currently selected. Returns Some((stored_model, current_model)) on mismatch, None if compatible.

pub fn dimension_mismatch(&self, engine_dimensions: usize) -> bool

Check if index dimensions are incompatible with the current engine.

pub fn memory_usage_bytes(&self) -> usize

Approximate heap memory used by this index in bytes.

pub fn unload(&mut self)

Drops all in-memory data to free heap. Index can be re-loaded from disk.

pub fn load_or_new(root: &Path, dimensions: usize) -> Self

Load a previously saved index, or create a new empty one.

pub fn files_needing_update(&self, chunks: &[CodeChunk]) -> Vec<String>

Determine which files need re-embedding based on content hashes.

When the index is empty (no prior embeddings), skips hash computation entirely by returning all unique file paths from chunks directly.

pub fn pending_chunk_count(&self, chunks: &[CodeChunk]) -> usize

Number of chunks a re-embed pass would have to embed right now — i.e. the chunks belonging to files flagged by Self::files_needing_update.

Used by the hybrid/dense cold-start guard (#512): on a server that came up before the on-disk index existed, the first query would otherwise embed the entire corpus inline under the request watchdog, producing a runaway the watchdog abandons but cannot cancel. Counting the pending chunks up front lets the caller fall back instead of starting that embed.

pub fn update( &mut self, chunks: &[CodeChunk], new_embeddings: &[(usize, Vec<f32>)], changed_files: &[String], precomputed_hashes: Option<HashMap<String, String>>, )

Update the index with new embeddings for changed files. Preserves existing embeddings for unchanged files.

precomputed_hashes can be passed to avoid re-computing file hashes when the caller already has them (e.g. from files_needing_update). When None, hashes are computed from chunks.

pub fn get_aligned_flat(&self, chunks: &[CodeChunk]) -> Option<FlatEmbeddings>

Get all embeddings in chunk order (aligned with BM25Index.chunks) as a single contiguous FlatEmbeddings allocation. Returns None if the index doesn’t cover all chunks.

The flat layout (_n_vectors × dim in row-major order) gives sequential memory access during dot-product scoring — one dereference instead of the two-level indirection of Arc<[Vec<f32>]>.

pub fn coverage(&self, total_chunks: usize) -> f64

pub fn save(&self, root: &Path) -> Result<()>

pub fn load(root: &Path) -> Option<Self>

Trait Implementations

impl Clone for EmbeddingIndex

fn clone(&self) -> EmbeddingIndex

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for EmbeddingIndex

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for EmbeddingIndex

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for EmbeddingIndex

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.