tools/read_source_record.py

"""Convenience reader for the supplemental Hugging Face source bundle.

This script lives inside the published `huggingface/` folder so users can read
source-side data directly from the exported package without depending on the
rest of the repository layout.
"""

import argparse
import json
import re
import sys
from pathlib import Path


HF_ROOT = Path(__file__).resolve().parents[1]
ROOT = HF_ROOT.parent
SOURCE_RECORDS_DIR = HF_ROOT / "source_records"
SOURCE_RECORDS_BY_ID_DIR = SOURCE_RECORDS_DIR / "by_id"
SUMMARY_PATH = SOURCE_RECORDS_DIR / "summary.json"


def flatten_sample_ids(sample_id_args):
    sample_ids = []
    for item in sample_id_args or []:
        sample_ids.extend(item)
    return [str(sample_id) for sample_id in sample_ids]


def load_json(path: Path):
    with open(path, "r", encoding="utf-8") as handle:
        return json.load(handle)


def load_sample_ids_from_file(path: Path):
    sample_ids = []
    with open(path, "r", encoding="utf-8") as handle:
        for line in handle:
            sample_id = line.strip()
            if sample_id:
                sample_ids.append(sample_id)
    return sample_ids


def strip_html_tags(text: str):
    text = re.sub(r"<script.*?</script>", " ", text, flags=re.IGNORECASE | re.DOTALL)
    text = re.sub(r"<style.*?</style>", " ", text, flags=re.IGNORECASE | re.DOTALL)
    text = re.sub(r"<[^>]+>", " ", text)
    text = re.sub(r"\s+", " ", text)
    return text.strip()


def walk_path(value, dotted_path: str):
    current = value
    for part in dotted_path.split("."):
        if isinstance(current, dict):
            if part not in current:
                raise KeyError(part)
            current = current[part]
            continue
        if isinstance(current, list):
            index = int(part)
            current = current[index]
            continue
        raise KeyError(part)
    return current


def first_nonempty(values):
    for value in values:
        if value not in (None, "", [], {}):
            return value
    return None


def extract_preferred_umf(record):
    candidates = []
    for dataset_name in ("property_prediction", "image_generation"):
        dataset_block = record.get(dataset_name, {})
        for split_name in ("train", "test"):
            split_block = dataset_block.get(split_name)
            if not split_block:
                continue
            recipe = split_block.get("recipe") or {}
            metadata = split_block.get("metadata") or {}
            candidates.append(recipe.get("umf"))
            candidates.append(metadata.get("umf"))
    html_metadata = record.get("html_metadata") or {}
    if isinstance(html_metadata, dict):
        candidates.append(html_metadata.get("umf"))
    return first_nonempty(candidates)


def extract_shortcut(record, field_name: str):
    html_metadata = record.get("html_metadata") or {}
    if field_name == "title":
        return first_nonempty([
            html_metadata.get("title") if isinstance(html_metadata, dict) else None,
        ])
    if field_name == "author":
        return first_nonempty([
            html_metadata.get("author") if isinstance(html_metadata, dict) else None,
        ])
    if field_name == "description":
        return first_nonempty([
            html_metadata.get("description") if isinstance(html_metadata, dict) else None,
            html_metadata.get("notes") if isinstance(html_metadata, dict) else None,
        ])
    if field_name == "umf":
        return extract_preferred_umf(record)
    raise KeyError(field_name)


def parse_args():
    parser = argparse.ArgumentParser(
        description="Read supplemental GlazyBench source-side data from the Hugging Face export."
    )
    parser.add_argument(
        "--sample-id",
        nargs="+",
        action="append",
        help="One or more sample IDs to read from huggingface/source_records/by_id.",
    )
    parser.add_argument(
        "--sample-id-file",
        help="Optional text file with one sample ID per line for batch reading.",
    )
    parser.add_argument(
        "--field",
        choices=[
            "full",
            "memberships",
            "property_prediction",
            "image_generation",
            "html_metadata",
            "raw_html_path",
            "title",
            "author",
            "description",
            "umf",
        ],
        default="full",
        help="Which part of the merged record to print.",
    )
    parser.add_argument(
        "--dataset",
        choices=["property_prediction", "image_generation"],
        help="Optional dataset selector when printing dataset-specific content.",
    )
    parser.add_argument(
        "--split",
        choices=["train", "test"],
        help="Optional split selector when printing dataset-specific content.",
    )
    parser.add_argument(
        "--summary",
        action="store_true",
        help="Print source bundle summary instead of an individual sample.",
    )
    parser.add_argument(
        "--list-sample-ids",
        type=int,
        default=0,
        help="Print the first N sample IDs available in source_records/by_id.",
    )
    parser.add_argument(
        "--list-fields",
        action="store_true",
        help="Print the top-level fields available in a sample record.",
    )
    parser.add_argument(
        "--path",
        help="Optional dotted path lookup inside the selected record value, for example property_prediction.test.recipe.umf.",
    )
    parser.add_argument(
        "--html-preview-chars",
        type=int,
        default=0,
        help="Also print the first N characters of the raw HTML file for the sample.",
    )
    parser.add_argument(
        "--html-preview-strip-tags",
        action="store_true",
        help="When previewing HTML, strip tags into rough plain text.",
    )
    parser.add_argument(
        "--output-format",
        choices=["json", "jsonl", "text"],
        default="json",
        help="Output format for batch or single-sample exports.",
    )
    parser.add_argument(
        "--output",
        help="Optional output file path. If omitted, prints to stdout.",
    )
    parser.add_argument(
        "--compact",
        action="store_true",
        help="Print compact JSON instead of pretty JSON.",
    )
    return parser.parse_args()


def render_json(value, compact: bool):
    if compact:
        return json.dumps(value, ensure_ascii=False, separators=(",", ":"))
    return json.dumps(value, ensure_ascii=False, indent=2)


def write_output(text: str, output_path: str | None):
    if output_path:
        path = Path(output_path)
        path.parent.mkdir(parents=True, exist_ok=True)
        with open(path, "w", encoding="utf-8") as handle:
            handle.write(text)
        return
    print(text)


def list_sample_ids(limit: int):
    sample_ids = sorted(path.stem for path in SOURCE_RECORDS_BY_ID_DIR.glob("*.json"))
    if limit > 0:
        sample_ids = sample_ids[:limit]
    return sample_ids


def get_record(sample_id: str):
    path = SOURCE_RECORDS_BY_ID_DIR / f"{sample_id}.json"
    if not path.exists():
        raise FileNotFoundError(f"Sample record not found: {path}")
    return load_json(path), path


def select_value(record, args):
    if args.field == "full":
        value = record
    elif args.field == "memberships":
        value = record.get("memberships", [])
    elif args.field == "html_metadata":
        value = record.get("html_metadata")
    elif args.field == "raw_html_path":
        raw_relpath = record.get("raw_html_relpath")
        value = None if raw_relpath is None else str(HF_ROOT / raw_relpath)
    elif args.field in {"title", "author", "description", "umf"}:
        value = extract_shortcut(record, args.field)
    else:
        dataset_block = record.get(args.field, {})
        if args.split:
            value = dataset_block.get(args.split)
        elif args.dataset and args.dataset != args.field:
            value = record.get(args.dataset, {})
        else:
            value = dataset_block
    if args.path:
        value = walk_path(value, args.path)
    return value


def maybe_get_html_preview(record, chars: int, strip_tags: bool):
    if chars <= 0:
        return None
    raw_relpath = record.get("raw_html_relpath")
    if raw_relpath is None:
        return "<missing raw html>"
    html_path = HF_ROOT / raw_relpath
    if not html_path.exists():
        return f"<file missing: {html_path}>"
    with open(html_path, "r", encoding="utf-8", errors="ignore") as handle:
        preview = handle.read(chars)
    if strip_tags:
        preview = strip_html_tags(preview)
    return preview


def build_sample_payload(sample_id: str, args):
    record, record_path = get_record(str(sample_id))

    if args.list_fields:
        payload = {
            "sample_id": str(sample_id),
            "record_path": str(record_path),
            "top_level_fields": sorted(record.keys()),
            "memberships": record.get("memberships", []),
        }
    else:
        value = select_value(record, args)
        payload = {
            "sample_id": str(sample_id),
            "record_path": str(record_path),
            "selected_field": args.field,
            "value": value,
        }

    html_preview = maybe_get_html_preview(record, args.html_preview_chars, args.html_preview_strip_tags)
    if html_preview is not None:
        payload["raw_html_preview"] = html_preview
    return payload


def render_payloads(payloads, args):
    if args.output_format == "jsonl":
        return "\n".join(render_json(payload, compact=True) for payload in payloads)
    if args.output_format == "text":
        lines = []
        for payload in payloads:
            lines.append(f"sample_id={payload.get('sample_id')}")
            lines.append(f"record_path={payload.get('record_path')}")
            if "selected_field" in payload:
                lines.append(f"selected_field={payload['selected_field']}")
                lines.append(f"value={payload.get('value')}")
            else:
                lines.append(f"top_level_fields={payload.get('top_level_fields')}")
            if "raw_html_preview" in payload:
                lines.append("raw_html_preview=")
                lines.append(str(payload["raw_html_preview"]))
            lines.append("")
        return "\n".join(lines).rstrip()
    if len(payloads) == 1:
        return render_json(payloads[0], args.compact)
    return render_json(payloads, args.compact)


def main():
    args = parse_args()

    if args.summary:
        write_output(render_json(load_json(SUMMARY_PATH), args.compact), args.output)
        return

    if args.list_sample_ids:
        payload = {
            "count": len(list_sample_ids(0)),
            "sample_ids": list_sample_ids(args.list_sample_ids),
        }
        write_output(render_json(payload, args.compact), args.output)
        return

    sample_ids = flatten_sample_ids(args.sample_id)
    if args.sample_id_file:
        sample_ids.extend(load_sample_ids_from_file(Path(args.sample_id_file)))
    if not sample_ids:
        raise SystemExit("Provide --sample-id, --sample-id-file, --summary, or --list-sample-ids.")

    payloads = [build_sample_payload(sample_id, args) for sample_id in sample_ids]
    write_output(render_payloads(payloads, args), args.output)


if __name__ == "__main__":
    try:
        main()
    except FileNotFoundError as exc:
        print(str(exc), file=sys.stderr)
        raise SystemExit(1)