Class LadybugGraphStore

ui/src/store/ladybugStore.ts:486–2861

// ---- Store implementation ----

export class LadybugGraphStore implements GraphStore {
  private engine: LbugEngine;
  private ready: Promise<void> | null = null;
  private embedder: Embedder | null = null;
  private sourceCache = new Map<
    string,
    { compressed: Uint8Array; path: string; binary?: boolean }
  >();

  // --- Write buffer ---
  private pendingNodes: ImportBatchRequest['nodes'] = [];
  private pendingRels: ImportBatchRequest['relationships'] = [];
  private totalNodesBuffered = 0;
  private totalRelsBuffered = 0;

  // --- Source text for FTS indexing (populated by storeSource, consumed by flush) ---
  private sourceSnippets = new Map<string, string>();

  // --- JS-side indexes ---
  private bm25Index = new BM25Index(1.5, 0.75, { name: 2.0 }, 1);
  /** Whether the LadybugDB vector index has been created for the current session. */
  private hasVectorIndex = false;

  /** Maps node ID → typed table name. Populated eagerly during importBatch. */
  private nodeTypeMap = new Map<string, string>();

  /** LRU node cache: id → {type, name, properties}. Serves getNode and
   *  fetchNodesByIds from JS memory without WASM round-trips. Capped at
   *  NODE_CACHE_MAX entries — Map insertion order acts as the eviction queue. */
  private nodeCache = new Map<
    string,
    { type: string; name: string; properties: Record<string, unknown> }
  >();
  private static readonly NODE_CACHE_MAX = 10_000;

  /** Package node IDs already written to LadybugDB. Packages are shared across
   *  repos so the same ID can arrive from multiple pipeline runs — skip
   *  duplicates to avoid LadybugDB COPY FROM primary-key violations. */
  private flushedPackageIds = new Set<string>();
  private flushedSourceIds = new Set<string>();

  // --- Visualization limits ---
  private maxVisNodes = 20000;
  private maxVisEdges = 20000;

  // --- Serialization queue (lbug-wasm wraps single-threaded C++ engine) ---
  private queue: Promise<void> = Promise.resolve();

  /** Monotonic counter for COPY-FROM temp filenames. The engine's virtual
   *  filesystem is process-wide, so a fixed path (e.g. /nodes_Function.csv)
   *  would collide if two import/flush operations ran in close succession —
   *  one COPY could read the other's bytes, or an unlink could remove a file
   *  before the other's COPY runs. Unique names make the paths collision-proof
   *  regardless of caller serialization. */
  private copySeq = 0;

  /** A unique temp path for a CSV about to be COPY'd in. */
  private tmpCsvPath(label: string): string {