Struct GraphStore 

pub struct GraphStore { /* private fields */ }

Persistent graph store backed by Kuzu.

Implementations

impl GraphStore

pub fn open(path: &Path) -> Result<Self>

Open or create a Kuzu database at the given path.

pub fn write_lock(&self) -> Result<WriteLock>

Acquire exclusive write lock. Blocks until available.

pub fn try_write_lock(&self) -> Result<Option<WriteLock>>

Try to acquire write lock without blocking. Returns None if already held.

pub fn connection(&self) -> Result<Connection<'_>>

pub fn remove_file(&self, file: &str) -> Result<()>

Remove all graph data for a deleted file.

pub fn remove_file_conn(&self, conn: &Connection<'_>, file: &str) -> Result<()>

Caller must hold WriteLock.

pub fn get_file_hashes(&self) -> Result<HashMap<String, String>>

Return map of file path -> content_hash for all indexed modules. Used by incremental indexing to skip unchanged files.

pub fn get_all_symbols(&self) -> Result<Vec<(String, String, String, String)>>

Return all symbols as (name, id, file, kind) tuples – used by resolve_calls.

pub fn derive_tested_by_edges(&self) -> Result<usize>

Get total counts for stats.

pub fn stats(&self) -> Result<GraphStats>

impl GraphStore

pub fn test_parquet_quality(&self) -> Result<()>

Test: DELETE + COPY FROM parquet produces identical data to MERGE/UNWIND. Covers edge cases: <>, quotes, unicode, empty strings, backslashes, newlines.

pub fn benchmark_bulk_write(&self, n: usize) -> Result<()>

Benchmark: compare COPY FROM CSV vs UNWIND for bulk symbol inserts. Creates isolated test tables, measures both approaches, prints results.

pub fn benchmark_parquet_vs_csv(&self) -> Result<()>

Benchmark: CSV vs Parquet vs UNWIND – apple-to-apple with real symbol data. Tests performance AND data integrity (docstrings with <, >, quotes, unicode).

impl GraphStore

pub fn upsert_all_bulk( &self, conn: &Connection<'_>, extractions: &[FileExtraction], ) -> Result<()>

Bulk insert all extractions in minimal queries – one UNWIND per node/edge type. Much faster than calling upsert_file_conn_no_delete per file. Caller must hold WriteLock.

impl GraphStore

pub fn upsert_folders_bulk(&self, file_paths: &[&str]) -> Result<()>

Create Folder nodes and edges for a set of file paths in bulk. More efficient than per-file upsert_folder_hierarchy calls.

pub fn upsert_folders_bulk_conn( &self, conn: &Connection<'_>, file_paths: &[&str], ) -> Result<()>

Caller must hold WriteLock.

pub fn upsert_all_parquet(&self, extractions: &[FileExtraction]) -> Result<()>

Bulk write all extractions using COPY FROM Parquet – binary format eliminates escaping issues. Used for –full index. Incremental index still uses upsert_file_conn_no_delete.

pub fn upsert_all_parquet_conn( &self, conn: &Connection<'_>, extractions: &[FileExtraction], ) -> Result<()>

Caller must hold WriteLock.

impl GraphStore

pub fn upsert_file(&self, extraction: &FileExtraction) -> Result<()>

Insert a file extraction into the graph. Removes old data for the file first (incremental update).

pub fn upsert_file_conn( &self, conn: &Connection<'_>, extraction: &FileExtraction, ) -> Result<()>

Caller must hold WriteLock.

pub fn upsert_file_conn_no_delete( &self, conn: &Connection<'_>, extraction: &FileExtraction, ) -> Result<()>

Caller must hold WriteLock.