tools/02_generate_metadata.py

#!/usr/bin/env -S uv run
"""
Collect relevant metadata from local UD directories:
- extracting a markdown Summary section and machine-readable metadata
  from README.{md,txt}
- using the UD directory name for collecting metadata from the
  codes_and_flags.yaml file
- collecting {dev,train,test}.conllu files.
"""

import json
import os
import xml.etree.ElementTree as ET
import argparse
import logging
import re
from collections import defaultdict
from pathlib import Path

import yaml
from dotenv import load_dotenv


TOOLS_DIR = Path(__file__).resolve().parent

load_dotenv(TOOLS_DIR / ".env")
UD_VER = os.getenv('UD_VER', "2.17")

# Parse command line arguments
parser = argparse.ArgumentParser(description=__doc__,
                                 formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument('-o', '--override', action='store_true',
                    help='override output file if it already exists')
parser.add_argument('--ud-ver', default=UD_VER,
                    help='UD version to process (default: UD_VER from env or 2.17)')
parser.add_argument('-v', '--verbose', action='count', default=0,
                    help='increase verbosity level')
args = parser.parse_args()
UD_VER = args.ud_ver

# Set up logging
logging.basicConfig(
    level = max(logging.DEBUG, logging.WARNING - args.verbose * 10),
    format='%(asctime)s [%(levelname)s] %(message)s',
    datefmt='%Y-%m-%d %H:%M:%S',
    force=True
)

GENUS_SEPARATED_FROM_UD_VERSION = (2, 15)
SUMMARY_HEADING_SCAN_LIMIT = 120
SUMMARY_HEADING_RE = re.compile(r"^#{1,2}\s+summary\s*$", re.IGNORECASE)
MARKDOWN_HEADING_RE = re.compile(r"^#{1,6}\s+\S")


def parse_ud_version(version: str) -> tuple[int, ...] | None:
    """
    Parse UD version strings like "2.14" or "2.15.1" into integer tuples.
    Returns None if parsing fails (e.g. "latest").
    """
    try:
        return tuple(int(part) for part in version.split("."))
    except ValueError:
        return None


def normalize_codes_and_flags_entry(entry: dict, ud_version: str) -> dict:
    """
    Normalize family/genus fields from codes_and_flags YAML across UD versions.

    Before UD 2.15, genus was often encoded in "family" as "Family, Genus".
    From UD 2.15 onward, genus is expected as a dedicated "genus" field.
    """
    normalized = dict(entry)
    version_tuple = parse_ud_version(ud_version)
    if version_tuple is None or version_tuple >= GENUS_SEPARATED_FROM_UD_VERSION:
        return normalized

    family = normalized.get("family")
    genus = normalized.get("genus")
    if isinstance(family, str) and (not genus) and "," in family:
        family_part, genus_part = family.split(",", 1)
        normalized["family"] = family_part.strip()
        genus_part = genus_part.strip()
        if genus_part:
            normalized["genus"] = genus_part

    return normalized


def find_summary_bounds(lines: list[str]) -> tuple[int, int] | None:
    """
    Find the markdown Summary section bounds.

    We look for an H1/H2 heading named "Summary" near the beginning
    of the file and collect lines until the next markdown heading.
    Returns (start_index, end_index_exclusive) for summary content lines.
    """
    scan_limit = min(len(lines), SUMMARY_HEADING_SCAN_LIMIT)
    summary_heading_idx = None

    for i in range(scan_limit):
        line = lines[i].strip()
        if SUMMARY_HEADING_RE.match(line):
            summary_heading_idx = i
            break

    if summary_heading_idx is None:
        return None

    start = summary_heading_idx + 1
    end = len(lines)
    for i in range(start, len(lines)):
        if MARKDOWN_HEADING_RE.match(lines[i].strip()):
            end = i
            break
    return (start, end)


def extract_metadata(file_path) -> {}:
    """
    Collect relevant metadata from UD directories.

    Args:
        file_path (str): The path to the README.{md,txt} file.

    Returns:
        dict: The extracted metadata.
    """
    metadata = {
        "summary": None,
        "license": None,
        "genre": None,
        "lemmas": None,
        "upos": None,
        "xpos": None,
        "language": None,   # German:
        "flag": None,       #   flag: DE
        "lcode": None,      #   lcode: de
        "iso3": None,       #   iso3: deu
        "family": None,     #   family: IE
        "genus": None,      #   genus: Germanic
        "blocked": None,    #   blocked: True|False
        # "splits": None,      # {"train":  {"files": []},
                               #  "dev":    {"files": []},
                               #  "test":   {"files": []},
                               #  "unknown":{"files": []}}
    }

    with open(file_path, 'r') as file:
        lines = [line.strip() for line in file.readlines()]
        summary_bounds = find_summary_bounds(lines)
        metadata_scan_start = 0
        if summary_bounds is not None:
            summary_start, summary_end = summary_bounds
            metadata_scan_start = summary_end
            summary = (' '.join(lines[summary_start:summary_end])).strip()
            if summary:
                metadata["summary"] = summary

        # This is a (quite hackish) approach inspired by:
        # https://github.com/UniversalDependencies/LICENSE/blob/master/generate_license_for_lindat.pl
        for line in lines[metadata_scan_start:]:
            if ":" in line:
                key, val = line.split(":", 1)
                if key.lower() in metadata.keys():
                    if key.lower() == "genre":
                        val = val.strip().split(" ")
                    else:
                        val = val.strip()
                    metadata[key.lower()] = val
    return metadata


def traverse_directory(directory):
    """
    Traverses the directory and its first-level subdirectories, finds the
    specified files, and extracts the summary from the README.{md,txt} file.

    Args:
        directory (str): The path to the directory.

    Returns:
        dict: A dictionary containing the extracted summaries for each file.
    """
    results = defaultdict(lambda: defaultdict(dict))

    codes_and_flags_path = TOOLS_DIR / "etc" / f"codes_and_flags-{UD_VER}.yaml"
    with codes_and_flags_path.open('r') as file:
        codes_and_flags = yaml.safe_load(file)
        logging.debug(codes_and_flags)

    for item in os.listdir(directory):
        if item.startswith("."):
            continue

        if os.path.isdir(os.path.join(directory, item)):
            dir_path = os.path.join(directory, item)
            logging.debug(dir_path)

            tag_fn = os.path.join(dir_path, f".tag-r{UD_VER}")
            has_version_tag = Path(tag_fn).exists()
            if not has_version_tag:
                # Snapshot layouts from CLARIN do not include per-treebank tag marker files.
                logging.debug(f"No tag file found (expected for snapshots): {tag_fn}")

            results[item]["splits"] = {
                "train": {"files": [], "num_bytes": 0},
                "dev": {"files": [], "num_bytes": 0},
                "test": {"files": [], "num_bytes": 0},
                "unknown": {"files": [], "num_bytes": 0}
            }
            for file in os.listdir(dir_path):
                if file.endswith(".conllu"):
                    file_path = os.path.join(dir_path, file)
                    if has_version_tag:
                        source_path = os.path.join(item, f"r{UD_VER}", file)
                    else:
                        source_path = os.path.join(item, file)
                    logging.debug(file_path)
                    match file:
                        case x if "dev" in file:
                            subset = "dev"
                        case x if "test" in file:
                            subset = "test"
                        case x if "train" in file:
                            subset = "train"
                        case _:
                            subset = "unknown"
                    results[item]["splits"][subset]["files"].append(source_path)

                    sum_bytes = os.stat(file_path).st_size
                    results[item]["splits"][subset]["num_bytes"] += sum_bytes

                elif file.startswith("README") and (file.endswith(
                        tuple(["md", "txt"]))):
                    results[item].update(
                        extract_metadata(os.path.join(dir_path, file)))
                    # print(results[item])

                elif file == "stats.xml":
                    # Extract size values:
                    # <treebank>
                    #  <size>
                    #    <total><sentences>15589</sentences><tokens>287708</tokens><words>292756</words><fused>5048</fused></total>
                    #    <train><sentences>13813</sentences><tokens>259167</tokens><words>263777</words><fused>4610</fused></train>
                    #    <dev><sentences>799</sentences><tokens>12316</tokens><words>12480</words><fused>164</fused></dev>
                    #    <test><sentences>977</sentences><tokens>16225</tokens><words>16499</words><fused>274</fused></test>
                    #  </size>
                    tree = ET.parse(os.path.join(dir_path, file))
                    root = tree.getroot()
                    size_node = root.find('.//size')
                    if size_node is None:
                        continue

                    for child_node_name in ["train", "dev", "test"]:

                        child_node = size_node.find(child_node_name)
                        if child_node is None:
                            continue

                        for child_child_node_name in ["sentences", "tokens", "words", "fused"]:
                            value_node = child_node.find(child_child_node_name)
                            if value_node is None:
                                continue
                            value = value_node.text
                            # print(f"Item:{item} {child_node_name}-{child_child_node_name}: {value}")

                            if value and int(value) > 0:
                                results[item]["splits"][child_node_name][f"num_{child_child_node_name}"] = value
                                # print(f"key: {child_node_name}")
                                # print(f"value: {value}")

            # 'Read' language name from the dirname (UD_Abaza-ATB/), extract
            # the relevant metadata from 'codes_and_flags.yaml' and add it to
            # the metadata.
            # ```
            # Abaza:
            #   flag: RU-ABAZA
            #   lcode: abq
            #   iso3: abq
            #   family: Northwest Caucasian
            # ```
            language = item[3:].rsplit("-", 1)[0].replace("_", " ")
            results[item]["language"] = language
            #print(f"item: {item}, language: {language}")
            normalized_codes_entry = normalize_codes_and_flags_entry(
                codes_and_flags[language], UD_VER
            )
            results[item].update(normalized_codes_entry)

            # 'Read' the corpus name from the dirname (UD_Abaza-ATB/) and
            # construct a short name for the data set:
            # abq_atb
            corpus_name = item[3:].rsplit("-", 1)[1].lower()
            name = f"{results[item]["lcode"]}_{corpus_name}"
            results[item]["name"] = name

            # Delete empty
            #   1. file subsets
            for fileset_k,fileset_v in list(results[item]["splits"].items()):
                if not fileset_v["files"]:
                    del results[item]["splits"][fileset_k]

            #   2. sets without a summary, files or license
            missing_summary = not results[item]["summary"]
            missing_files = not any(value["files"] for value in results[item]["splits"].values())
            missing_license = not results[item]["license"]

            if missing_summary or missing_files or missing_license:
                if missing_summary:
                    logging.warning("ITEM DELETED - no summary: %s", item)
                if missing_files:
                    logging.debug(results[item]["splits"])
                    logging.warning("ITEM DELETED - no files: %s", item)
                if missing_license:
                    logging.warning("ITEM DELETED - no license: %s", item)
                del results[item]

    # Swap keys and key.value()["name"]; add old key as key.value()["dirname"]
    for key in list(results.keys()):
        name = results[key]["name"]
        del results[key]["name"]
        results[key]["dirname"] = key
        results[name] = results.pop(key)

    return results

if __name__ == '__main__':
    directory = TOOLS_DIR / 'UD_repos'
    results = traverse_directory(directory)

    # Load blocked treebanks and add "blocked" property
    blocked_treebanks_file = TOOLS_DIR / 'blocked_treebanks.yaml'
    if blocked_treebanks_file.exists():
        with open(blocked_treebanks_file, 'r') as fh:
            blocked_data = yaml.safe_load(fh)
            if blocked_data:
                blocked_treebanks = {k: v for k, v in blocked_data.items() if v is not None}
                for name in results.keys():
                    if name in blocked_treebanks:
                        results[name]["blocked"] = True
                        logging.info(f"Marked {name} as blocked: {blocked_treebanks[name]['reason']}")
                    else:
                        results[name]["blocked"] = False
            else:
                # No blocked treebanks, mark all as not blocked
                for name in results.keys():
                    results[name]["blocked"] = False
    else:
        logging.warning(f"Blocked treebanks file not found: {blocked_treebanks_file}")
        # Mark all as not blocked if file doesn't exist
        for name in results.keys():
            results[name]["blocked"] = False

    for name, metadata in results.items():
        logging.debug("Directory: %s", metadata["dirname"])

    # Write the metadata to json
    output_path = TOOLS_DIR / f"metadata-{UD_VER}.json"
    should_write = args.override or (not output_path.exists()) or output_path.stat().st_size == 0
    if should_write:
        with output_path.open('w') as fh:
            json.dump(results, fh, ensure_ascii=False, sort_keys=True,
                      indent=4, separators=(',', ': '))
            print(f"{output_path.name} written")
    else:
        logging.info("Output %s already exists: Not overriding.", output_path.name)