MM-IssueLocBench / scripts /download_repos.py
Jasaxion's picture
Initial upload: MM-IssueLoc Bench v1.0 (canonical + function_level parquet, VCE+ preview, repo downloader)
41e2b8e verified
Raw
History Blame Contribute Delete
11.7 kB
"""Download repository snapshots for MM-IssueLoc Bench at each PR base_commit.
For every entry in `commit_cache.json`, this script fetches a source-code
tarball at the exact `base_commit` via the GitHub REST API and extracts it
into `repos/<dir_name>/`. No `.git` history is retained.
Requirements
------------
- Python 3.10+ (stdlib only — no external packages).
- A GitHub personal access token with `public_repo` scope is strongly
recommended: the anonymous rate limit is 60 req/hour, vs 5000/hour with
a token. Pass it via `--token` or set `GITHUB_TOKEN` / `GH_TOKEN`.
Generate one at https://github.com/settings/tokens (no extra permissions
needed for public repositories).
Example
-------
export GITHUB_TOKEN=ghp_xxx
python3 scripts/download_repos.py # ~10 min w/ token
python3 scripts/download_repos.py --workers 8
python3 scripts/download_repos.py --only 5 # dry-run small slice
python3 scripts/download_repos.py --retry-failed # retry failures
Resumability
------------
Already-extracted directories are skipped by default. Partial downloads
use a `<dir>.tmp` staging directory that is renamed atomically on success,
so a Ctrl+C never leaves a half-extracted repo under `repos/`.
Disk / bandwidth
----------------
Full download is ~5–20 GB across 653 repos; size is dominated by a few
large upstream projects. Individual tarballs are streamed in memory
(typical peak RSS < 500 MB) and extracted immediately.
"""
from __future__ import annotations
import argparse
import concurrent.futures as cf
import json
import os
import shutil
import sys
import tarfile
import time
import urllib.error
import urllib.request
from dataclasses import dataclass
from io import BytesIO
from pathlib import Path
API_ROOT = "https://api.github.com"
UA = "mm-issueloc-bench-downloader/1.0"
# ---------------------------------------------------------------------------
# Types
# ---------------------------------------------------------------------------
@dataclass
class Job:
instance_id: str
repo: str # "owner/name"
sha: str # full commit SHA
dir_name: str # target directory name
@property
def owner(self) -> str:
return self.repo.split("/", 1)[0]
@property
def name(self) -> str:
return self.repo.split("/", 1)[1]
@dataclass
class Result:
job: Job
ok: bool
files: int = 0
error: str = ""
# ---------------------------------------------------------------------------
# GitHub API
# ---------------------------------------------------------------------------
def _request(url: str, token: str | None, timeout: int = 180) -> bytes:
"""GET a URL, returning the raw body. Follows redirects (tarball → codeload)."""
headers = {"Accept": "application/vnd.github+json", "User-Agent": UA}
if token:
headers["Authorization"] = f"Bearer {token}"
req = urllib.request.Request(url, headers=headers)
with urllib.request.urlopen(req, timeout=timeout) as resp:
return resp.read()
def fetch_tarball(job: Job, token: str | None, max_retries: int = 5) -> bytes:
"""Fetch the repo as a gzipped tarball at the target commit.
Retries on 403 rate-limit and transient network errors with exponential backoff.
"""
url = f"{API_ROOT}/repos/{job.repo}/tarball/{job.sha}"
last_err: Exception | None = None
for attempt in range(max_retries):
try:
return _request(url, token)
except urllib.error.HTTPError as e:
body = e.read().decode("utf-8", errors="replace") if hasattr(e, "read") else ""
if e.code == 403 and ("rate limit" in body.lower() or "secondary" in body.lower()):
wait = min(60 * (2 ** attempt), 900)
print(f" rate limited ({e.code}); sleeping {wait}s (attempt {attempt + 1}/{max_retries})",
file=sys.stderr, flush=True)
time.sleep(wait)
last_err = e
continue
# 404 / 451 / 410 etc — repo is gone, don't retry
raise RuntimeError(f"HTTP {e.code}{body[:200]}") from e
except (urllib.error.URLError, TimeoutError, ConnectionError) as e:
wait = 2 ** attempt
print(f" network error ({e}); retrying in {wait}s", file=sys.stderr, flush=True)
time.sleep(wait)
last_err = e
raise RuntimeError(f"exhausted retries: {last_err}")
# ---------------------------------------------------------------------------
# Extraction
# ---------------------------------------------------------------------------
_TAR_KW = {"filter": "data"} if sys.version_info >= (3, 12) else {}
def extract_tarball(data: bytes, dest: Path) -> int:
"""Extract into dest/, stripping the GitHub tarball's top-level dir.
GitHub tarballs have a top-level directory like `owner-name-shortsha/`;
we strip it so that dest/ holds the repo contents directly.
"""
dest.mkdir(parents=True, exist_ok=True)
n = 0
with tarfile.open(fileobj=BytesIO(data), mode="r:gz") as tf:
members = tf.getmembers()
if not members:
return 0
root = members[0].name.split("/", 1)[0] + "/"
for m in members:
if not m.name.startswith(root):
continue
m.name = m.name[len(root):]
if not m.name:
continue
try:
tf.extract(m, path=dest, **_TAR_KW)
n += 1
except Exception as e:
# Skip problematic members (broken symlinks, etc.) but keep going
print(f" skipped member {m.name!r}: {e}", file=sys.stderr)
return n
# ---------------------------------------------------------------------------
# Per-job driver
# ---------------------------------------------------------------------------
def run_one(job: Job, out_dir: Path, token: str | None) -> Result:
dest = out_dir / job.dir_name
tmp = dest.with_suffix(".tmp")
if tmp.exists():
shutil.rmtree(tmp)
try:
data = fetch_tarball(job, token)
n = extract_tarball(data, tmp)
tmp.rename(dest)
return Result(job, ok=True, files=n)
except Exception as e:
if tmp.exists():
shutil.rmtree(tmp, ignore_errors=True)
return Result(job, ok=False, error=str(e))
# ---------------------------------------------------------------------------
# Main
# ---------------------------------------------------------------------------
def _load_jobs(cache_path: Path, out_dir: Path, *, skip_existing: bool) -> tuple[list[Job], int]:
cache: dict[str, dict] = json.loads(cache_path.read_text())
jobs: list[Job] = []
skipped = 0
for iid, entry in cache.items():
job = Job(
instance_id=iid,
repo=entry["repo"],
sha=entry["sha"],
dir_name=entry["dir_name"],
)
dest = out_dir / job.dir_name
if skip_existing and dest.exists() and any(dest.iterdir()):
skipped += 1
continue
jobs.append(job)
return jobs, skipped
def _load_failed(failures_path: Path, out_dir: Path) -> list[Job]:
records = json.loads(failures_path.read_text())
return [
Job(instance_id=r["instance_id"], repo=r["repo"], sha=r["sha"], dir_name=r["dir_name"])
for r in records
]
def main() -> int:
ap = argparse.ArgumentParser(description=__doc__.split("\n")[0])
ap.add_argument("--output-dir", type=Path, default=Path("repos"),
help="Directory to place `<dir_name>/` snapshots into (default: repos)")
ap.add_argument("--cache-file", type=Path, default=Path("commit_cache.json"),
help="Path to commit_cache.json (default: ./commit_cache.json)")
ap.add_argument("--workers", type=int, default=4,
help="Parallel download workers (default: 4)")
ap.add_argument("--only", type=int, default=0,
help="Download only the first N jobs (for testing)")
ap.add_argument("--token", default=os.environ.get("GITHUB_TOKEN") or os.environ.get("GH_TOKEN"),
help="GitHub token; falls back to $GITHUB_TOKEN / $GH_TOKEN")
ap.add_argument("--no-skip-existing", action="store_true",
help="Re-download even if the target directory already exists")
ap.add_argument("--retry-failed", action="store_true",
help="Retry entries from download_failures.json instead of the full cache")
args = ap.parse_args()
if not args.cache_file.exists():
print(f"error: {args.cache_file} not found — run from the dataset root.", file=sys.stderr)
return 2
if not args.token:
print("WARNING: no token provided. Unauthenticated rate limit is 60 req/hour.\n"
" Set GITHUB_TOKEN or pass --token for a 5000/hour limit.",
file=sys.stderr)
args.output_dir.mkdir(parents=True, exist_ok=True)
failures_path = args.output_dir.parent / "download_failures.json"
if args.retry_failed:
if not failures_path.exists():
print(f"error: no {failures_path} — nothing to retry.", file=sys.stderr)
return 2
jobs = _load_failed(failures_path, args.output_dir)
skipped = 0
print(f"Retrying {len(jobs)} previously failed entries from {failures_path}")
else:
jobs, skipped = _load_jobs(
args.cache_file, args.output_dir, skip_existing=not args.no_skip_existing,
)
if args.only:
jobs = jobs[: args.only]
if not jobs:
print(f"Nothing to do — {skipped} already present in {args.output_dir}/.")
return 0
print(f"Downloading {len(jobs)} repos into {args.output_dir}/ ({skipped} already present) "
f"with {args.workers} workers.")
failures: list[Result] = []
start = time.time()
with cf.ThreadPoolExecutor(max_workers=args.workers) as ex:
futures = {ex.submit(run_one, j, args.output_dir, args.token): j for j in jobs}
try:
for i, fut in enumerate(cf.as_completed(futures), 1):
res = fut.result()
tag = "ok " if res.ok else "FAIL"
short_sha = res.job.sha[:12]
extra = f"{res.files:>5} files" if res.ok else res.error[:80]
print(f" [{i:>4}/{len(jobs)}] {tag} {res.job.repo:<55s} @ {short_sha} {extra}",
flush=True)
if not res.ok:
failures.append(res)
except KeyboardInterrupt:
print("\nInterrupted — waiting for in-flight downloads to finish cleanly...",
file=sys.stderr)
ex.shutdown(cancel_futures=True)
return 130
elapsed = time.time() - start
n_ok = len(jobs) - len(failures)
print(f"\n{n_ok}/{len(jobs)} repos fetched in {elapsed:.0f}s.")
if failures:
failures_path.write_text(json.dumps(
[
{
"instance_id": r.job.instance_id,
"repo": r.job.repo,
"sha": r.job.sha,
"dir_name": r.job.dir_name,
"error": r.error,
}
for r in failures
],
indent=2,
))
print(f"{len(failures)} failures logged to {failures_path}\n"
f"Retry with: python3 scripts/download_repos.py --retry-failed", file=sys.stderr)
return 1
return 0
if __name__ == "__main__":
sys.exit(main())