Struct CodeStore 

pub struct CodeStore { /* private fields */ }

Qdrant + SQLite dual-write store for code chunks.

CodeStore is the persistence layer for the indexing pipeline. It is cheaply cloneable (all fields are reference-counted) and can safely be shared across async tasks.

Lifecycle

1. Call CodeStore::with_ops to construct.
2. Call CodeStore::ensure_collection once at startup to create the Qdrant collection if it does not yet exist.
3. Use CodeStore::upsert_chunks_batch during indexing and CodeStore::search during retrieval.

Implementations

impl CodeStore

pub fn with_ops(ops: QdrantOps, pool: DbPool) -> Self

Create a CodeStore from a pre-built QdrantOps instance and a SQLite pool.

The Qdrant collection is not created here — call CodeStore::ensure_collection before performing any upserts.

Examples

use zeph_index::store::CodeStore;
use zeph_memory::QdrantOps;

let ops = QdrantOps::new("http://localhost:6334", None).unwrap();
let store = CodeStore::with_ops(ops, pool);
store.ensure_collection(1536).await?;

pub async fn ensure_collection(&self, vector_size: u64) -> Result<()>

Create collection with INT8 scalar quantization if it doesn’t exist.

Errors

Returns an error if Qdrant operations fail.

pub async fn upsert_chunk( &self, chunk: &ChunkInsert<'_>, vector: Vec<f32>, ) -> Result<String>

Upsert a code chunk into both Qdrant and SQLite.

Errors

Returns an error if Qdrant or SQLite operations fail.

pub async fn upsert_chunks_batch( &self, chunks: Vec<(ChunkInsert<'_>, Vec<f32>)>, ) -> Result<Vec<String>>

Upsert multiple chunks into both Qdrant and SQLite in a single batch.

All vector points are sent to Qdrant in one request and all metadata rows are inserted in a single SQLite transaction, reducing per-chunk overhead during full-project indexing.

Errors

Returns an error if Qdrant or SQLite operations fail.

pub async fn chunk_exists(&self, content_hash: &str) -> Result<bool>

Check if a chunk with this content hash already exists.

Errors

Returns an error if the SQLite query fails.

pub async fn existing_hashes(&self, hashes: &[&str]) -> Result<HashSet<String>>

Return the set of content hashes that already exist in the store.

Uses WHERE content_hash IN (...) with chunks of 900 to stay below SQLite’s default variable limit of 999.

Errors

Returns an error if the SQLite query fails.

pub async fn remove_file_chunks(&self, file_path: &str) -> Result<usize>

Remove all chunks for a given file path from both stores.

Errors

Returns an error if Qdrant or SQLite operations fail.

pub async fn search( &self, query_vector: EmbeddingVector<Normalized>, limit: usize, language_filter: Option<String>, ) -> Result<Vec<SearchHit>>

Search for similar code chunks.

The query_vector must be L2-normalized (use EmbeddingVector::<Unnormalized>::normalize or EmbeddingVector::<Normalized>::new_normalized before calling). Requiring Normalized at the type level prevents silent near-zero cosine scores that Qdrant gRPC returns for mismatched or unnormalized vectors.

Errors

Returns an error if Qdrant search fails.

pub async fn indexed_files(&self) -> Result<Vec<String>>

List all indexed file paths.

Errors

Returns an error if the SQLite query fails.

Trait Implementations

impl Clone for CodeStore

fn clone(&self) -> CodeStore

Returns a duplicate of the value.

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

Performs copy-assignment from source.