Struct SemanticIndex 

pub struct SemanticIndex { /* private fields */ }

The semantic index — stores embeddings for all symbols in a project. Borrow-only roots retain only a root path plus an Arc to immutable relative data.

Implementations

impl SemanticIndex

pub fn new(project_root: PathBuf, dimension: usize) -> Self

pub fn entry_count(&self) -> usize

Number of embedded symbol entries.

pub fn estimated_memory(&self) -> MemoryEstimate

Estimate resident semantic-index bytes from the vectors and metadata actually held by each entry. This intentionally excludes allocator and hash-table bucket overhead, which are not cheaply observable.

pub fn indexed_file_count(&self) -> usize

Number of files currently tracked by the semantic index.

pub fn status_label(&self) -> &'static str

Human-readable status label for the index.

pub fn build<F>( project_root: &Path, files: &[PathBuf], embed_fn: &mut F, max_batch_size: usize, ) -> Result<Self, String>

where F: FnMut(Vec<String>) -> Result<Vec<Vec<f32>>, String>,

Build the semantic index from a set of files using the provided embedding function. embed_fn takes a batch of texts and returns a batch of embedding vectors.

pub fn build_with_progress<F, P>( project_root: &Path, files: &[PathBuf], embed_fn: &mut F, max_batch_size: usize, progress: &mut P, ) -> Result<Self, String>

where F: FnMut(Vec<String>) -> Result<Vec<Vec<f32>>, String>, P: FnMut(usize, usize),

Build the semantic index and report embedding progress using entry counts.

pub fn refresh_stale_files<F, P>( &mut self, project_root: &Path, current_files: &[PathBuf], embed_fn: &mut F, max_batch_size: usize, progress: &mut P, ) -> Result<RefreshSummary, String>

where F: FnMut(Vec<String>) -> Result<Vec<Vec<f32>>, String>, P: FnMut(usize, usize),

Incrementally refresh entries for changed/new files only, preserving cached embeddings for unchanged files. Used when loading the index from disk and finding that a small fraction of files have moved on, deleted, or appeared.

Returns RefreshSummary describing what changed. On success, self is mutated in place and remains a valid index.

current_files is the full set of files the project considers indexable (typically walk_project_files(...)). Files in the cache that are no longer in this set are treated as deleted.

pub fn refresh_invalidated_files<F, P>( &mut self, project_root: &Path, paths: &[PathBuf], embed_fn: &mut F, max_batch_size: usize, max_files: usize, progress: &mut P, ) -> Result<InvalidatedFilesRefresh, String>

where F: FnMut(Vec<String>) -> Result<Vec<Vec<f32>>, String>, P: FnMut(usize, usize),

Refresh exactly the files invalidated by the live watcher, without treating the provided path list as the whole project. This is the watcher-side counterpart to refresh_stale_files: it drops any stale entries for the requested paths from this in-memory index, re-extracts whatever still exists on disk, embeds those chunks, and returns the delta needed for another in-memory index to apply the same update.

pub fn apply_refresh_update( &mut self, added_entries: Vec<EmbeddingEntry>, updated_metadata: Vec<(PathBuf, FileFreshness)>, completed_paths: &[PathBuf], )

pub fn search(&self, query_vector: &[f32], top_k: usize) -> Vec<SemanticResult>

Search the index with a query embedding, returning top-K results sorted by relevance

pub fn len(&self) -> usize

Number of indexed entries

pub fn is_file_stale(&self, file: &Path) -> bool

Check if a file needs re-indexing based on mtime/size

pub fn remove_file(&mut self, file: &Path)

Remove entries for a specific file

pub fn invalidate_file(&mut self, file: &Path)

pub fn dimension(&self) -> usize

Get the embedding dimension

pub fn fingerprint(&self) -> Option<&SemanticIndexFingerprint>

pub fn backend_label(&self) -> Option<&str>

pub fn model_label(&self) -> Option<&str>

pub fn set_fingerprint(&mut self, fingerprint: SemanticIndexFingerprint)

pub fn write_to_disk(&self, storage_dir: &Path, project_key: &str) -> bool

Write the semantic index to disk using atomic temp+rename pattern. Empty indexes are persisted too so a completed rebuild cannot leave an older non-empty snapshot visible to the next process.

pub fn read_from_disk( storage_dir: &Path, project_key: &str, current_canonical_root: &Path, is_worktree_bridge: bool, expected_fingerprint: Option<&str>, ) -> Option<Self>

Read the semantic index from disk

pub fn to_bytes(&self) -> Vec<u8>

Serialize the index to bytes for disk persistence

pub fn from_bytes( data: &[u8], current_canonical_root: &Path, ) -> Result<Self, String>

Deserialize the index from bytes

Trait Implementations

impl Clone for SemanticIndex

fn clone(&self) -> SemanticIndex

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SemanticIndex

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

Formats the value using the given formatter.