| |
| """Download the design sources referenced by this dataset from their upstream repositories |
| and verify them against the recorded SHA-256. No design source is redistributed here; every |
| design remains under the licence of its upstream repository. |
| |
| python3 scripts/fetch_sources.py corpus/test.jsonl compositional/designs.jsonl --out sources/ |
| |
| Each record's `source` gives the repository, the commit the dataset was built from (when known), |
| the path inside the repository and the SHA-256 of the file. The script downloads each repository |
| archive at that commit (or the default branch when no commit is recorded), extracts the referenced |
| files, and writes sources/<sha256>.v for every file whose content matches. Records without a |
| `source` field are ignored. Files are decoded as UTF-8 (undecodable bytes dropped), as when the |
| dataset was built, and the checksum is computed on that text. Paths that lie inside git |
| submodules of their repository are not part of the archive and are reported as not found.""" |
| import argparse, collections, hashlib, json, os, sys, tarfile, tempfile, urllib.request |
| ap = argparse.ArgumentParser(); ap.add_argument("files", nargs="+"); ap.add_argument("--out", default="sources") |
| a = ap.parse_args(); os.makedirs(a.out, exist_ok=True) |
| need = collections.defaultdict(dict) |
| for fn in a.files: |
| for line in open(fn): |
| if not line.strip(): |
| continue |
| s = json.loads(line).get("source") |
| if s and (s.get("repository") or "").startswith("https://github.com/"): |
| need[(s["repository"], s.get("commit") or "HEAD")][s["path"]] = s["sha256"] |
| ok = changed = missing = failed = 0 |
| for (repo, ref), files in sorted(need.items()): |
| owner_repo = repo.split("github.com/")[1] |
| print(f"{owner_repo}@{ref[:10]}: {len(files)} files", file=sys.stderr) |
| with tempfile.NamedTemporaryFile(suffix=".tar.gz") as tmp: |
| try: |
| with urllib.request.urlopen(f"https://codeload.github.com/{owner_repo}/tar.gz/{ref}", timeout=300) as r: |
| while True: |
| chunk = r.read(1 << 20) |
| if not chunk: break |
| tmp.write(chunk) |
| except OSError as e: |
| print(f" could not download {owner_repo}@{ref[:10]}: {e}", file=sys.stderr) |
| failed += 1; missing += len(files); continue |
| tmp.flush() |
| found = set() |
| try: |
| with tarfile.open(tmp.name, mode="r:gz") as t: |
| for m in t: |
| rel = m.name.split("/", 1)[1] if "/" in m.name else "" |
| if m.isfile() and rel in files: |
| text = t.extractfile(m).read().decode("utf-8", errors="ignore") |
| h = hashlib.sha256(text.encode()).hexdigest(); found.add(rel) |
| if h == files[rel]: |
| with open(os.path.join(a.out, h + ".v"), "w", encoding="utf-8", newline="") as o: o.write(text) |
| ok += 1 |
| else: |
| changed += 1 |
| except (tarfile.TarError, EOFError, OSError) as e: |
| print(f" could not read the archive of {owner_repo}: {e}", file=sys.stderr) |
| failed += 1 |
| missing += len(set(files) - found) |
| print(f"verified {ok} files; {changed} differ from the recorded checksum; {missing} not found" |
| f" ({failed} repositories could not be downloaded)", file=sys.stderr) |
| sys.exit(1 if failed or missing or changed else 0) |
|
|
