legex/scrapers/uk.py

from __future__ import annotations

import logging
import re
import threading
import time
import xml.etree.ElementTree as ET
from concurrent.futures import ThreadPoolExecutor, as_completed
from datetime import date

import requests
from dateutil import parser as dateutil_parser

from legex.models.base import Case
from legex.scrapers.base import BaseScraper

log = logging.getLogger(__name__)

# Find Case Law (caselaw.nationalarchives.gov.uk): UKSC + Court of Appeal Civil
# Division. BAILII is gated behind an Anubis proof-of-work and supremecourt.uk
# only covers post-2020; National Archives is the clean upstream. Bulk
# computational analysis requires a separate licence (held for legex).
BASE_URL = "https://caselaw.nationalarchives.gov.uk"
ATOM_URL = f"{BASE_URL}/atom.xml"
COURTS = ["uksc", "ewca/civ"]
PAGE_SIZE = 50
DELAY_SECONDS = 3
RETRY_DELAY_SECONDS = 30
XML_WORKERS = 8

# 1000 requests / 5 min rolling window per IP. Cap to ~2.85 req/sec globally.
_MIN_INTERVAL = 0.35
_RATE_LOCK = threading.Lock()
_LAST_REQ_AT = [0.0]

ATOM_NS = {
    "atom": "http://www.w3.org/2005/Atom",
    "tna": "https://caselaw.nationalarchives.gov.uk",
}
AKN = "{http://docs.oasis-open.org/legaldocml/ns/akn/3.0}"

# Non-civil title patterns. Find Case Law truncates the bracketed applicant in
# UKSC judicial review ("R (on the application of X) v ..." → "R v ...") so it
# is indistinguishable from criminal "R v <defendant>" by surface form. EWCA
# Civil instead inverts the form ("<applicant>, R (on the application of) v
# ..."). Both criminal and administrative are out of scope, so we drop:
#   - any "R v ..." or "R (..." opening
#   - any "..., R (on the application of) ..." (EWCA Civil's inverted JR)
#   - DPP / Attorney General's Reference / HM Advocate / Procurator Fiscal.
_NOT_CIVIL_RE = re.compile(
    r"^R\s+(?:v\b|\()"
    r"|\bR\s*\(on the application of\b"
    r"|^(?:Director of Public Prosecutions|DPP)\s+v\b"
    r"|^Attorney General(?:['’]s)?\s+Reference\b"
    r"|\b(?:Her Majesty(?:['’]s)?\s+Advocate|HM\s+Advocate|Procurator Fiscal)\b",
    re.IGNORECASE,
)


def _throttle() -> None:
    with _RATE_LOCK:
        wait = _MIN_INTERVAL - (time.monotonic() - _LAST_REQ_AT[0])
        if wait > 0:
            time.sleep(wait)
        _LAST_REQ_AT[0] = time.monotonic()


def _parse_date(raw: str | None) -> date | None:
    if not raw:
        return None
    try:
        return date.fromisoformat(raw[:10])
    except ValueError:
        pass
    try:
        return dateutil_parser.parse(raw, dayfirst=True).date()
    except (ValueError, OverflowError, dateutil_parser.ParserError):
        return None


def _fetch(session: requests.Session, url: str, params: dict | None = None) -> bytes | None:
    _throttle()
    try:
        resp = session.get(url, params=params, timeout=30)
        if resp.status_code == 429:
            log.warning("HTTP 429 on %s — sleeping %ds then retrying", url, RETRY_DELAY_SECONDS)
            time.sleep(RETRY_DELAY_SECONDS)
            _throttle()
            resp = session.get(url, params=params, timeout=30)
        if not resp.ok:
            log.warning("HTTP %d for %s", resp.status_code, url)
            return None
        return resp.content
    except requests.RequestException as exc:
        log.warning("Error fetching %s: %s", url, exc)
        return None


def _extract_full_text(xml_bytes: bytes) -> str | None:
    try:
        root = ET.fromstring(xml_bytes)
    except ET.ParseError:
        return None
    body = root.find(f".//{AKN}judgmentBody")
    if body is None:
        return None
    parts: list[str] = []
    for el in body.iter():
        for t in (el.text, el.tail):
            if t and (s := t.strip()):
                parts.append(s)
    return "\n".join(parts) if parts else None


def _entry_to_case(entry: ET.Element, court: str) -> Case | None:
    title_el = entry.find("atom:title", ATOM_NS)
    title = (title_el.text if title_el is not None else "") or ""

    published_el = entry.find("atom:published", ATOM_NS)
    decision_date = _parse_date(published_el.text if published_el is not None else None)

    link_html = link_xml = link_pdf = None
    for link in entry.findall("atom:link", ATOM_NS):
        rel = link.attrib.get("rel")
        ltype = link.attrib.get("type")
        href = link.attrib.get("href")
        if not href:
            continue
        if rel == "alternate" and ltype is None and href.startswith(BASE_URL):
            link_html = href
        elif ltype == "application/akn+xml":
            link_xml = href
        elif ltype == "application/pdf":
            link_pdf = href

    if not link_html:
        return None

    cite_el = entry.find("tna:identifier[@type='ukncn']", ATOM_NS)
    citation = cite_el.text if cite_el is not None else None
    slug = cite_el.attrib.get("slug") if cite_el is not None else None

    return Case(
        case_id=citation,
        link=link_html,
        decision_date=decision_date,
        jurisdiction="uk",
        language="en",
        full_text=None,
        metadata={
            "title": title,
            "slug": slug,
            "court": court,
            "xml_url": link_xml,
            "pdf_url": link_pdf,
        },
    )


class UKScraper(BaseScraper):
    country = "United Kingdom"

    def scrape(
        self,
        start_date: date | None = None,
        end_date: date | None = None,
    ) -> list[Case]:
        session = requests.Session()
        session.headers.update({"Accept": "application/atom+xml, application/xml, */*"})

        cases: list[Case] = []
        seen: set[str] = set()

        for court in COURTS:
            cases.extend(self._scrape_court(session, court, seen, start_date, end_date))

        todo = [c for c in cases if (c.metadata or {}).get("xml_url")]
        if todo:
            log.info("Fetching full_text for %d Akoma Ntoso XML files", len(todo))
            done = failed = 0
            with ThreadPoolExecutor(max_workers=XML_WORKERS) as pool:
                futures = {
                    pool.submit(_fetch, session, c.metadata["xml_url"]): c for c in todo
                }
                for fut in as_completed(futures):
                    case = futures[fut]
                    xml_bytes = fut.result()
                    text = _extract_full_text(xml_bytes) if xml_bytes else None
                    if text:
                        case.full_text = text
                        done += 1
                    else:
                        failed += 1
                    if (done + failed) % 100 == 0:
                        log.info(
                            "XML progress %d/%d (failed=%d)",
                            done + failed, len(todo), failed,
                        )
            log.info("Hydrated %d XML files (failed=%d)", done, failed)

        return cases

    @staticmethod
    def _scrape_court(
        session: requests.Session,
        court: str,
        seen: set[str],
        start_date: date | None,
        end_date: date | None,
    ) -> list[Case]:
        out: list[Case] = []
        page = 1
        while True:
            log.info("[%s] atom page %d", court, page)
            data = _fetch(session, ATOM_URL, params={"court": court, "page": page})
            if not data:
                break
            try:
                root = ET.fromstring(data)
            except ET.ParseError as exc:
                log.warning("[%s] parse error on page %d: %s", court, page, exc)
                break

            entries = root.findall("atom:entry", ATOM_NS)
            if not entries:
                log.info("[%s] no entries on page %d — stopping", court, page)
                break

            all_before_start = True
            for entry in entries:
                case = _entry_to_case(entry, court)
                if case is None or not case.link or case.link in seen:
                    continue
                d = case.decision_date
                if end_date and d and d > end_date:
                    # Too new — feed is sorted desc, so older (in-window) pages are still ahead.
                    all_before_start = False
                    continue
                if start_date and d and d >= start_date:
                    all_before_start = False
                if start_date and d and d < start_date:
                    continue
                seen.add(case.link)
                out.append(case)

            if start_date and all_before_start:
                log.info("[%s] all on page %d predate start_date — stopping early", court, page)
                break
            if len(entries) < PAGE_SIZE:
                log.info("[%s] last page at %d (%d items)", court, page, len(entries))
                break

            time.sleep(DELAY_SECONDS)
            page += 1

        log.info("[%s] collected %d cases", court, len(out))
        return out

    @staticmethod
    def civil_filter(cases: list[Case]) -> list[Case]:
        return [
            c for c in cases
            if not _NOT_CIVIL_RE.search(((c.metadata or {}).get("title") or "").strip())
        ]