Datasets:

Modalities:
Text
Formats:
json
Languages:
English
ArXiv:
License:
reuse-qa / scripts /prepare_sql_api_bench.py
zan-nabil's picture
Publish REUSE-QA FinanceBench and SQL-API-Bench release
1c77159 verified
Raw History Blame Contribute Delete
11.7 kB
"""Prepare the Spider databases used by the bundled REUSE-QA SQL questions.
Adapted from the RECAP project's sql_api_bench_dataset.py. This keeps its Spider
Google Drive download and database-discovery workflow, but reads REUSE-QA's
existing JSONL records instead of downloading or regenerating question data.
Use an existing Spider distribution without downloading anything:
python scripts/prepare_sql_api_bench.py --spider-dir /path/to/spider
Or download and extract Spider into a separate working directory:
python scripts/prepare_sql_api_bench.py --work-dir ./sql_api_bench_data
Only the download path needs the optional gdown package. No package is installed
automatically. The complete archive, including upstream notices, is retained.
This script prepares databases, not the scalar tools or an agent environment.
"""
import argparse
from contextlib import closing
import json
import os
from pathlib import Path, PurePosixPath, PureWindowsPath
import re
import shutil
import sqlite3
import stat
import sys
import zipfile
SPIDER_DRIVE_ID = "1403EGqzIDoHMdQF4c9Bkyl7dZLZ5Wt6J"
QUESTION_FILE = Path(__file__).resolve().parents[1] / "data" / "sql_api_bench.jsonl"
def load_required_databases(question_file=QUESTION_FILE):
"""Read database names from the bundled questions without changing them."""
required = set()
question_count = 0
with question_file.open(encoding="utf-8-sig") as handle:
for line_number, line in enumerate(handle, 1):
if not line.strip():
raise ValueError(f"Blank JSONL line {line_number} in {question_file}")
try:
row = json.loads(line)
except json.JSONDecodeError as exc:
raise ValueError(
f"Invalid JSONL line {line_number} in {question_file}: {exc}"
) from exc
if not isinstance(row, dict) or row.get("dataset") != "sql_api_bench":
raise ValueError(f"Unexpected source on JSONL line {line_number}")
context = row.get("context")
db_name = context.get("db_name") if isinstance(context, dict) else None
if not isinstance(db_name, str) or not re.fullmatch(r"[A-Za-z0-9_]+", db_name):
raise ValueError(f"Invalid context.db_name on JSONL line {line_number}")
required.add(db_name)
question_count += 1
if not required:
raise ValueError(f"No SQL questions found in {question_file}")
return question_count, required
def database_path(database_dir, db_name):
"""Support Spider's name/name.sqlite layout and flat database directories."""
for path in (
database_dir / db_name / f"{db_name}.sqlite",
database_dir / f"{db_name}.sqlite",
):
if path.is_file():
return path
return None
def find_database_dir(spider_dir, required):
"""Find the directory containing the requested Spider databases."""
candidates = [
spider_dir,
spider_dir / "spider_data" / "database",
spider_dir / "database",
spider_dir / "spider" / "database",
]
if spider_dir.is_dir():
for root, dirs, _ in os.walk(spider_dir, followlinks=False):
dirs[:] = sorted(d for d in dirs if d != "__MACOSX")
if Path(root).name == "database":
candidates.append(Path(root))
best = None
best_count = 0
seen = set()
for candidate in candidates:
key = candidate.resolve()
if key in seen or not candidate.is_dir():
continue
seen.add(key)
count = sum(database_path(candidate, name) is not None for name in required)
if count == len(required):
return candidate
if count > best_count:
best, best_count = candidate, count
return best
def safe_extract(archive, destination):
"""Extract every archive member after rejecting unsafe paths and links."""
destination = destination.resolve()
if destination.exists() and (
not destination.is_dir() or any(destination.iterdir())
):
raise ValueError(
f"Extraction directory is not empty: {destination}. "
"Choose a fresh --work-dir or use --spider-dir for an existing copy."
)
with zipfile.ZipFile(archive) as zf:
members = []
targets = {}
for info in zf.infolist():
name = info.filename.replace("\\", "/")
relative = PurePosixPath(name)
windows_path = PureWindowsPath(name)
if (
not name
or relative.is_absolute()
or windows_path.drive
or ".." in relative.parts
or any(
part.endswith((" ", "."))
or any(ch in '<>:"|?*' or ord(ch) < 32 for ch in part)
for part in relative.parts
)
):
raise ValueError(f"Unsafe archive member path: {info.filename!r}")
mode = info.external_attr >> 16
file_type = stat.S_IFMT(mode)
if stat.S_ISLNK(mode) or file_type not in (0, stat.S_IFREG, stat.S_IFDIR):
raise ValueError(f"Unsupported archive link or special file: {info.filename!r}")
target = destination.joinpath(*relative.parts).resolve()
if not target.is_relative_to(destination) or target == destination:
raise ValueError(f"Unsafe archive member path: {info.filename!r}")
key = target.as_posix().casefold()
is_dir = info.is_dir() or file_type == stat.S_IFDIR
if key in targets:
if is_dir and targets[key]:
continue
raise ValueError(f"Conflicting archive member path: {info.filename!r}")
targets[key] = is_dir
members.append((info, target, is_dir))
# A regular file cannot also be a parent directory of another member.
for _, target, _ in members:
for parent in target.parents:
if parent == destination:
break
if targets.get(parent.as_posix().casefold()) is False:
raise ValueError(f"Archive file/directory conflict: {target}")
destination.mkdir(parents=True, exist_ok=True)
for info, target, is_dir in members:
if is_dir:
target.mkdir(parents=True, exist_ok=True)
else:
target.parent.mkdir(parents=True, exist_ok=True)
with zf.open(info) as source, target.open("xb") as output:
shutil.copyfileobj(source, output)
def download_spider(work_dir, required):
"""Download the original Spider archive and keep its complete contents."""
work_dir = work_dir.resolve()
spider_dir = work_dir / "spider"
if spider_dir.exists():
database_dir = find_database_dir(spider_dir, required)
if database_dir and all(database_path(database_dir, n) for n in required):
print(f"Using existing extracted Spider copy: {spider_dir}")
return database_dir
raise ValueError(
f"Existing extraction is incomplete: {spider_dir}. "
"Choose a fresh --work-dir or provide a complete --spider-dir."
)
archive = work_dir / "spider_data.zip"
if not archive.is_file():
try:
import gdown
except ImportError as exc:
raise RuntimeError(
"Downloading Spider requires gdown. Install it explicitly with:\n"
f' "{sys.executable}" -m pip install gdown\n'
"Then rerun, or provide --spider-dir for an existing distribution."
) from exc
work_dir.mkdir(parents=True, exist_ok=True)
partial = work_dir / "spider_data.zip.part"
print("Downloading the Spider archive from Google Drive...")
try:
result = gdown.download(
f"https://drive.google.com/uc?id={SPIDER_DRIVE_ID}",
str(partial),
quiet=False,
)
except Exception as exc:
raise RuntimeError(f"Spider download failed: {exc}") from exc
if not result or not partial.is_file() or not zipfile.is_zipfile(partial):
raise RuntimeError(
"Spider download did not produce a ZIP archive. "
"Check the download source or use --spider-dir."
)
partial.replace(archive)
print(f"Extracting the full Spider archive to: {spider_dir}")
safe_extract(archive, spider_dir)
print(f"Archive retained, including upstream notices: {archive}")
database_dir = find_database_dir(spider_dir, required)
if database_dir is None:
raise FileNotFoundError(f"No requested Spider databases found in {spider_dir}")
return database_dir
def verify_databases(database_dir, required):
"""Check required files and their SQLite integrity without writing to them."""
missing = [name for name in sorted(required) if database_path(database_dir, name) is None]
if missing:
raise FileNotFoundError(
f"Missing {len(missing)} required databases in {database_dir}: "
+ ", ".join(missing)
)
details = []
for name in sorted(required):
path = database_path(database_dir, name).resolve()
uri = path.as_uri() + "?mode=ro&immutable=1"
with closing(sqlite3.connect(uri, uri=True)) as connection:
connection.execute("PRAGMA query_only=ON")
check = [row[0] for row in connection.execute("PRAGMA integrity_check")]
if check != ["ok"]:
raise ValueError(f"SQLite integrity check failed for {path}: {check}")
table_count = connection.execute(
"SELECT COUNT(*) FROM sqlite_master "
"WHERE type='table' AND name NOT LIKE 'sqlite_%'"
).fetchone()[0]
details.append((name, path, table_count))
return details
def main(argv=None):
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
"--spider-dir", type=Path,
help="Existing Spider root or database directory; read in place without downloading.",
)
parser.add_argument(
"--work-dir", type=Path, default=Path("sql_api_bench_data"),
help="Download/extraction directory when --spider-dir is omitted (default: %(default)s).",
)
args = parser.parse_args(argv)
try:
question_count, required = load_required_databases()
print(f"Bundled questions: {question_count}; required databases: {len(required)}")
if args.spider_dir is not None:
spider_dir = args.spider_dir.resolve()
if not spider_dir.is_dir():
raise FileNotFoundError(f"Existing Spider directory not found: {spider_dir}")
database_dir = find_database_dir(spider_dir, required)
if database_dir is None:
raise FileNotFoundError(f"No requested Spider databases found in {spider_dir}")
else:
database_dir = download_spider(args.work_dir, required)
details = verify_databases(database_dir, required)
print(f"Ready: {len(details)} databases for {question_count} bundled questions.")
print(f"Database directory: {database_dir.resolve()}")
for name, path, tables in details:
print(f" {name}: {path} ({tables} tables)")
return 0
except (OSError, ValueError, RuntimeError, sqlite3.Error, zipfile.BadZipFile) as exc:
print(f"Preparation failed: {exc}", file=sys.stderr)
return 1
if __name__ == "__main__":
sys.exit(main())