Function _compute_summaries

code_review_graph/tools/build.py:133–358

Populate community_summaries, flow_snapshots, and risk_index tables. Uses batched aggregate queries and in-memory grouping instead of per-community/per-node loops. On graphs with ~100k edges this reduces the work from ``O(nodes + communities)`` SQLite round trips each doing their ow

(store: Any)

def _compute_summaries(store: Any) -> None:
    """Populate community_summaries, flow_snapshots, and risk_index tables.

    Uses batched aggregate queries and in-memory grouping instead of
    per-community/per-node loops. On graphs with ~100k edges this
    reduces the work from ``O(nodes + communities)`` SQLite round trips
    each doing their own B-tree scan to a handful of ``GROUP BY``
    queries, turning what used to be an effective hang into a few
    seconds.

    Each summary block (community_summaries, flow_snapshots, risk_index)
    is wrapped in an explicit transaction so the DELETE + INSERT sequence
    is atomic.  If a table doesn't exist yet the block is silently skipped.
    """
    import json as _json
    from collections import defaultdict
    from os.path import commonprefix

    conn = store._conn

    # -- community_summaries --
    try:
        conn.execute("BEGIN IMMEDIATE")
        conn.execute("DELETE FROM community_summaries")

        # Pre-compute per-qualified_name edge counts once. Previously
        # this section ran a per-community triple-JOIN aggregate query
        # (nodes LEFT JOIN edges LEFT JOIN edges), which on graphs with
        # thousands of communities was the second-biggest hang.
        edge_counts: dict[str, int] = defaultdict(int)
        for row in conn.execute(
            "SELECT source_qualified, COUNT(*) FROM edges GROUP BY source_qualified"
        ):
            edge_counts[row[0]] += row[1]
        for row in conn.execute(
            "SELECT target_qualified, COUNT(*) FROM edges GROUP BY target_qualified"
        ):
            edge_counts[row[0]] += row[1]

        # Group non-File nodes per community for top-symbol selection.
        nodes_by_comm: dict[int, list[tuple[str, int]]] = defaultdict(list)
        for row in conn.execute(
            "SELECT community_id, name, qualified_name FROM nodes "
            "WHERE community_id IS NOT NULL AND kind != 'File'"
        ):
            cid, name, qn = row[0], row[1], row[2]
            nodes_by_comm[cid].append((name, edge_counts.get(qn, 0)))

        # Group distinct file paths per community (preserving first-seen
        # order for stable output, same as DISTINCT in the old query).
        files_by_comm: dict[int, list[str]] = defaultdict(list)
        seen_files: dict[int, set[str]] = defaultdict(set)
        for row in conn.execute(
            "SELECT community_id, file_path FROM nodes WHERE community_id IS NOT NULL"
        ):
            cid, fp = row[0], row[1]
            if fp not in seen_files[cid]:
                seen_files[cid].add(fp)