scripts/normalize_sources.py

#!/usr/bin/env python3
"""
scripts/normalize_sources.py

Normalize or remove local filesystem paths in the cleaned dataset artifacts.
- Reads files from a `final_cleaned_data` directory (or other path you pass).
- Produces normalized outputs with source paths replaced by a compact identifier.
- Writes a `source_map.json` mapping original -> normalized for traceability.

Usage examples (from project root):

# Dry-run: show 10 mappings
python scripts/normalize_sources.py --input-dir final_cleaned_data --dry-run --show 10

# Apply changes and write normalized files
python scripts/normalize_sources.py --input-dir final_cleaned_data --apply

# Use a custom output dir
python scripts/normalize_sources.py --input-dir final_cleaned_data --apply --out-dir final_cleaned_data/normalized

Behavior / heuristics
- If a source contains an .onion hostname, normalized id will be the onion hostname (e.g. 222222222hsoeiok.onion).
- Else, we use the basename of `source_path` (filename).
- For concatenated records with `::original_filename`, the part after `::` is used when present.
- The script preserves the original files by writing new outputs (normalized_*).
"""

from pathlib import Path
import argparse
import json
import csv
import re
from urllib.parse import urlparse

ONION_RE = re.compile(r"([a-z2-7]{16,}\.onion)", flags=re.I)


def extract_identifier(src: str) -> str:
    if not src:
        return "unknown"
    # handle concatenated style: path::inner
    if "::" in src:
        src = src.split("::", 1)[1]
    # try to find onion host
    m = ONION_RE.search(src)
    if m:
        return m.group(1).lower()
    # try to parse as URL
    if src.startswith("http://") or src.startswith("https://"):
        try:
            p = urlparse(src)
            host = p.netloc
            if host:
                return host
        except Exception:
            pass
    # fallback: use basename
    try:
        return Path(src).name
    except Exception:
        return src.replace('\\', '/').split('/')[-1]


def normalize_jsonl(in_path: Path, out_path: Path, source_map: dict):
    out_path.parent.mkdir(parents=True, exist_ok=True)
    written = 0
    with in_path.open('r', encoding='utf-8') as inf, out_path.open('w', encoding='utf-8') as outf:
        for line in inf:
            line = line.strip()
            if not line:
                continue
            try:
                obj = json.loads(line)
            except Exception:
                continue
            orig = obj.get('source_path') or obj.get('source') or ''
            nid = extract_identifier(orig)
            source_map.setdefault(orig, nid)
            # update fields
            if 'source_path' in obj:
                obj['source_path'] = nid
            if 'source' in obj:
                obj['source'] = nid
            # also update site_folder if it looks like a path
            if 'site_folder' in obj:
                sf = obj.get('site_folder')
                if sf and ('\\' in sf or '/' in sf or ':' in sf):
                    obj['site_folder'] = nid
            outf.write(json.dumps(obj, ensure_ascii=False) + '\n')
            written += 1
    return written


def normalize_csv(in_path: Path, out_path: Path, source_map: dict):
    out_path.parent.mkdir(parents=True, exist_ok=True)
    written = 0
    with in_path.open('r', encoding='utf-8', newline='') as inf:
        reader = csv.DictReader(inf)
        fieldnames = reader.fieldnames
        rows = list(reader)
    # update rows
    for r in rows:
        orig = r.get('source_path') or ''
        nid = extract_identifier(orig)
        source_map.setdefault(orig, nid)
        r['source_path'] = nid
        # if there is site_folder field, normalize it too
        if 'site_folder' in r and (not r['site_folder'] or ('\\' in r['site_folder'] or '/' in r['site_folder'] or ':' in r['site_folder'])):
            r['site_folder'] = nid
    # write out
    with out_path.open('w', encoding='utf-8', newline='') as outf:
        writer = csv.DictWriter(outf, fieldnames=fieldnames)
        writer.writeheader()
        for r in rows:
            writer.writerow(r)
            written += 1
    return written


def main():
    p = argparse.ArgumentParser()
    p.add_argument('--input-dir', type=Path, default=Path('final_cleaned_data'))
    p.add_argument('--out-dir', type=Path, default=None)
    p.add_argument('--apply', action='store_true')
    p.add_argument('--show', type=int, default=0, help='Number of mappings to print in dry-run')
    p.add_argument('--dry-run', action='store_true')
    args = p.parse_args()

    inp = args.input_dir
    if not inp.exists():
        print('Input dir not found:', inp)
        return
    out_root = Path(args.out_dir) if args.out_dir else inp / 'normalized'
    out_root.mkdir(parents=True, exist_ok=True)

    source_map = {}

    # files to normalize (if present)
    jsonl_in = inp / 'cleaned_extracted_text.jsonl'
    csv_in = inp / 'cleaned_labeled_dataset.csv'
    instr_in = inp / 'cleaned_instruction_tuning.jsonl'

    if jsonl_in.exists():
        written = normalize_jsonl(jsonl_in, out_root / 'normalized_extracted_text.jsonl', source_map)
        print('Processed JSONL:', written)
    else:
        print('No cleaned_extracted_text.jsonl at', jsonl_in)

    if csv_in.exists():
        written = normalize_csv(csv_in, out_root / 'normalized_labeled_dataset.csv', source_map)
        print('Processed CSV:', written)
    else:
        print('No cleaned_labeled_dataset.csv at', csv_in)

    if instr_in.exists():
        written = normalize_jsonl(instr_in, out_root / 'normalized_instruction_tuning.jsonl', source_map)
        print('Processed instruction JSONL:', written)
    else:
        print('No cleaned_instruction_tuning.jsonl at', instr_in)

    # write mapping
    map_path = out_root / 'source_map.json'
    with map_path.open('w', encoding='utf-8') as f:
        json.dump(source_map, f, indent=2, ensure_ascii=False)

    if args.dry_run or not args.apply:
        print('\nDRY RUN - no files overwritten in originals. To write outputs add --apply')
        # print sample mappings
        cnt = 0
        for orig, nid in list(source_map.items())[:args.show or 20]:
            print(f'"{orig}" -> "{nid}"')
            cnt += 1
        print(f'Printed {cnt} mappings. Mapping written to', map_path)
        return

    # If apply, we've already written normalized files to out_root
    print('Wrote normalized outputs to', out_root)
    print('Source map written to', map_path)

if __name__ == '__main__':
    main()