"""Randomized read-only audit of repoinfo against local The Stack v2 data. The default run samples ten batches of ten repositories. It verifies source blob integrity, graph invariants, reference/edge consistency, SCC ordering and reports AST-vs-regex differences for manual review. """ from __future__ import annotations import argparse from collections import Counter, defaultdict from dataclasses import asdict import hashlib import json import os from pathlib import Path import random import time import pyarrow.parquet as pq from stackbuild import Repo, RepoFile, concatenation_dfs, scan ROOT = Path(__file__).resolve().parents[1] DATA_ROOT = Path(os.environ.get( "STACKGO_DATA_ROOT", "/home/ldj/tgai/data/pretrain-data" )) DEFAULT_DATA = DATA_ROOT / "the-stack-v2-contents-star-gt10" DEFAULT_REPORT = ROOT / "reports/repoinfo-stack-v2-audit.json" def sample_repositories(root: Path, batches: int, batch_size: int, seed: int): rng = random.Random(seed) shards = sorted((root / "shards").glob("*/repositories.parquet")) if not shards: raise FileNotFoundError(f"No repository metadata under {root}") selected = [] for batch_index in range(batches): metadata = shards[batch_index % len(shards)] parquet = pq.ParquetFile(metadata) indices = sorted(rng.sample(range(parquet.metadata.num_rows), batch_size)) wanted = set(indices) rows = [] offset = 0 for record_batch in parquet.iter_batches( columns=["repo_name", "revision_id", "files"], batch_size=256, use_threads=False ): for local_index, row in enumerate(record_batch.to_pylist()): if offset + local_index in wanted: rows.append(row) offset += record_batch.num_rows if len(rows) == batch_size: break selected.append((metadata.parent, rows)) return selected def load_bodies(shard: Path, rows: list[dict]) -> dict[str, bytes]: needed = {file["blob_id"] for row in rows for file in row.get("files") or []} bodies: dict[str, bytes] = {} for path in sorted((shard / "contents").glob("*.parquet")): for batch in pq.ParquetFile(path).iter_batches( columns=["blob_id", "content"], batch_size=2048, use_threads=False ): for item in batch.to_pylist(): if item["blob_id"] in needed: bodies[item["blob_id"]] = bytes(item["content"]) if len(bodies) == len(needed): break return bodies def make_repo(row: dict, bodies: dict[str, bytes]) -> tuple[Repo, list[str]]: files, errors = [], [] for item in row.get("files") or []: body = bodies.get(item["blob_id"]) if body is None: errors.append(f"missing body {item['blob_id']} {item['path']}") continue if hashlib.sha1(body).hexdigest() != item["blob_id"]: errors.append(f"sha1 mismatch {item['blob_id']} {item['path']}") continue files.append(RepoFile( item["path"], body, item.get("language"), item.get("src_encoding"), item["blob_id"], item.get("is_vendor"), item.get("is_generated"), )) return Repo(row["repo_name"], row["revision_id"], tuple(files)), errors def audit_info(repo: Repo, info) -> tuple[list[str], list[str]]: errors, warnings = [], [] paths = {file.path.lstrip("/") for file in repo.files} if len(paths) != len(repo.files): errors.append("duplicate normalized paths") components = {} for index, component in enumerate(info.strongly_connected_components): for path in component: components[path] = index positions = {path: index for index, path in enumerate(info.topological_order)} dfs_paths = concatenation_dfs(info) if dfs_paths != info.topological_order: errors.append("concatenation tree DFS differs from topological order") if len(dfs_paths) != len(set(dfs_paths)) or set(dfs_paths) != paths: errors.append("concatenation tree does not cover every file exactly once") if info.concatenation_tree is None or info.concatenation_tree.name != repo.repo_id: errors.append("concatenation tree repository root mismatch") edge_keys = {(edge.source, edge.target) for edge in info.edges if edge.status == "resolved"} for edge in info.edges: if edge.source not in paths or edge.target not in paths: errors.append(f"edge endpoint missing: {edge.source} -> {edge.target}") if edge.source == edge.target: errors.append(f"self edge: {edge.source}") if (components.get(edge.source) != components.get(edge.target) and positions.get(edge.target, -1) > positions.get(edge.source, -1)): errors.append(f"dependency ordered after source: {edge.source} -> {edge.target}") for reference in info.references: if reference.source_path not in paths: errors.append(f"reference source missing: {reference.source_path}") if reference.status == "internal": if not reference.target_path or reference.target_path not in paths: errors.append(f"internal target missing: {reference.source_path} {reference.name}") elif (reference.source_path, reference.target_path) not in edge_keys: errors.append(f"internal reference has no edge: {reference.source_path} -> {reference.target_path}") if reference.status == "external" and reference.candidates: errors.append(f"external reference has candidates: {reference.source_path} {reference.name}") if reference.status == "unresolved": tail = reference.name.replace("::", "/").replace(".", "/").rstrip("/").split("/")[-1] obvious = sorted(path for path in paths if path != reference.source_path and Path(path).stem.lower() == tail.lower())[:5] if obvious: warnings.append( f"unresolved with same-stem candidates: {reference.source_path} {reference.name} -> {obvious}" ) return errors, warnings def edge_set(info) -> set[tuple[str, str]]: return {(edge.source, edge.target) for edge in info.edges if edge.status == "resolved"} def main() -> None: parser = argparse.ArgumentParser(description=__doc__) parser.add_argument("--data-root", type=Path, default=DEFAULT_DATA) parser.add_argument("--batches", type=int, default=10) parser.add_argument("--batch-size", type=int, default=10) parser.add_argument("--seed", type=int, default=20261030) parser.add_argument("--report", type=Path, default=DEFAULT_REPORT) args = parser.parse_args() sampled = sample_repositories(args.data_root, args.batches, args.batch_size, args.seed) report = {"seed": args.seed, "batches": [], "totals": {}} totals = Counter() started = time.perf_counter() for batch_index, (shard, rows) in enumerate(sampled, 1): bodies = load_bodies(shard, rows) batch_report = {"batch": batch_index, "shard": shard.name, "repositories": []} for row in rows: repo, source_errors = make_repo(row, bodies) start = time.perf_counter() info = scan(repo) regex_info = scan(repo, use_treesitter=False) errors, warnings = audit_info(repo, info) errors[:0] = source_errors ast_edges, regex_edges = edge_set(info), edge_set(regex_info) ast_only, regex_only = sorted(ast_edges - regex_edges), sorted(regex_edges - ast_edges) if ast_only or regex_only: warnings.append(f"backend edge delta ast_only={len(ast_only)} regex_only={len(regex_only)}") record = { "repo_id": repo.repo_id, "revision_id": repo.revision_id, "elapsed_seconds": time.perf_counter() - start, "stats": info.stats, "errors": errors, "warnings": warnings[:50], "ast_only_edges": ast_only[:20], "regex_only_edges": regex_only[:20], } batch_report["repositories"].append(record) totals.update(repositories=1, files=info.stats["files"], edges=info.stats["resolved_edges"], errors=len(errors), warnings=len(warnings), ast_only=len(ast_only), regex_only=len(regex_only), missing_bodies=len(source_errors)) report["batches"].append(batch_report) print(f"batch {batch_index}/{args.batches} {shard.name}: " f"repos={len(rows)} errors={sum(len(x['errors']) for x in batch_report['repositories'])} " f"warnings={sum(len(x['warnings']) for x in batch_report['repositories'])}", flush=True) totals["elapsed_seconds"] = round(time.perf_counter() - started, 3) report["totals"] = dict(totals) args.report.parent.mkdir(parents=True, exist_ok=True) temporary = args.report.with_suffix(args.report.suffix + ".part") temporary.write_text(json.dumps(report, ensure_ascii=False, indent=2) + "\n") temporary.replace(args.report) print(json.dumps(report["totals"], indent=2), flush=True) print(f"report: {args.report}", flush=True) if totals["errors"]: raise SystemExit(1) if __name__ == "__main__": main()