archive_helpers.py

"""
Shared helpers for CatholicCorpus download scripts.

Why this exists:
  - Archive.org returns 503 under load; naive single-shot downloads fail.
  - Item IDs guessed from search results often point to redirected or
    renamed items; the /metadata/ API tells us the real PDF filename.

Public API:
  fetch_metadata(item)            -> dict | None
  resolve_pdf_url(item)           -> (url, server_filename) | (None, None)
  download_with_retry(url, dest)  -> bool   (exponential backoff on 5xx/429)
  archive_pdf_smart(item, dest)   -> bool   (metadata lookup + retry download)
"""
from __future__ import annotations

import atexit
import json
import time
from datetime import datetime, timezone
from pathlib import Path
from typing import List, Optional, Tuple

import requests

# Persistent lending-restricted log — shared across all tasks
_LENDING_LOG = Path(__file__).resolve().parent / "LENDING_RESTRICTED.json"

UA = {"User-Agent": "CatholicCorpus/1.0 (academic research; admin@catholiccorpus.org)"}
TIMEOUT = 180
CHUNK = 1 << 15

# Retry policy
RETRY_STATUSES = {429, 500, 502, 503, 504}
MAX_RETRIES = 3        # 3 attempts (not 5) — cuts wait from 124s to 28s per item
BACKOFF_BASE = 4       # 4, 8, 16 → 28s max total wait per item
BACKOFF_MAX = 30

# Circuit breaker: if archive.org fires N consecutive 5xx/metadata-nulls,
# pause for a cooldown period then try again. Prevents wasting time on a
# degraded server while still retrying later in the same run.
_consecutive_failures = 0
_CIRCUIT_BREAKER_THRESHOLD = 5
_circuit_open = False
_circuit_opened_at = 0.0
_CIRCUIT_COOLDOWN = 60  # seconds before auto-resetting the circuit


def _note_failure():
    global _consecutive_failures, _circuit_open, _circuit_opened_at
    _consecutive_failures += 1
    if _consecutive_failures >= _CIRCUIT_BREAKER_THRESHOLD and not _circuit_open:
        _circuit_open = True
        _circuit_opened_at = time.time()
        _log(f"    [!] Circuit breaker: {_consecutive_failures} consecutive archive.org failures")
        _log(f"        Likely causes: archive.org is degraded (5xx), IP is being rate-limited,")
        _log(f"        or the guessed item IDs don't exist.")
        _log(f"        Pausing for {_CIRCUIT_COOLDOWN}s cooldown, then retrying.")


def _note_success():
    global _consecutive_failures, _circuit_open
    _consecutive_failures = 0
    _circuit_open = False


def circuit_is_open() -> bool:
    global _circuit_open, _consecutive_failures
    if _circuit_open and (time.time() - _circuit_opened_at) >= _CIRCUIT_COOLDOWN:
        _log(f"    [circuit cooldown elapsed — retrying archive.org]")
        _circuit_open = False
        _consecutive_failures = 0
    return _circuit_open


def _log(msg: str) -> None:
    print(msg, flush=True)


# ---------------------------------------------------------------------------
# Lending-restricted item tracker
# ---------------------------------------------------------------------------

def _record_lending_restricted(item: str, title: str, task_hint: str = "") -> None:
    """Append a lending-restricted item to LENDING_RESTRICTED.json."""
    entries = {}
    if _LENDING_LOG.exists():
        try:
            entries = json.loads(_LENDING_LOG.read_text())
        except (json.JSONDecodeError, OSError):
            entries = {}

    if item not in entries:
        entries[item] = {
            "title": title,
            "task": task_hint,
            "url": f"https://archive.org/details/{item}",
            "discovered": datetime.now(timezone.utc).strftime("%Y-%m-%d"),
        }
        _LENDING_LOG.write_text(json.dumps(entries, indent=2, ensure_ascii=False) + "\n")
        _log(f"    → logged to LENDING_RESTRICTED.json")


def render_lending_restricted_md() -> None:
    """Generate LENDING_RESTRICTED.md from the JSON log."""
    if not _LENDING_LOG.exists():
        return
    try:
        entries = json.loads(_LENDING_LOG.read_text())
    except (json.JSONDecodeError, OSError):
        return
    if not entries:
        return

    md_path = _LENDING_LOG.with_suffix(".md")
    by_task: dict[str, list] = {}
    for item_id, info in entries.items():
        task = info.get("task", "unknown")
        by_task.setdefault(task, []).append((item_id, info))

    lines = [
        "# Lending-Restricted Items",
        "",
        "These archive.org items exist but are **borrow-only** (`access-restricted-item: true`).",
        "They cannot be bulk-downloaded without an Open Library lending session.",
        f"",
        f"Total: {len(entries)} items across {len(by_task)} tasks.",
        "",
    ]
    for task in sorted(by_task):
        lines.append(f"## {task}")
        lines.append("")
        for item_id, info in sorted(by_task[task], key=lambda x: x[0]):
            lines.append(f"- **{item_id}**")
            lines.append(f"  - Title: {info.get('title', '?')}")
            lines.append(f"  - URL: {info.get('url', '?')}")
            lines.append(f"  - Discovered: {info.get('discovered', '?')}")
        lines.append("")

    lines.append("---")
    lines.append("*Auto-generated by archive_helpers.py — do not edit manually.*")
    lines.append("")
    md_path.write_text("\n".join(lines))
    _log(f"Updated {md_path.name} ({len(entries)} items)")


# Auto-render the markdown when any script using archive_helpers exits
atexit.register(render_lending_restricted_md)


# ---------------------------------------------------------------------------

# Sentinel distinguishing "item exists but metadata API was unreachable"
# (fall back to URL guess) from "metadata said the item does not exist"
# (don't bother guessing — item literally is not on archive.org).
_META_UNREACHABLE = object()


def fetch_metadata(item: str) -> dict | None | object:
    """
    Return parsed /metadata/{item} payload.

    - dict: item exists and has files
    - None: item does NOT exist on archive.org (metadata returned {} or 404)
    - _META_UNREACHABLE: could not reach metadata API (5xx / network error)
      → caller may want to fall back to URL guessing
    """
    url = f"https://archive.org/metadata/{item}"
    for attempt in range(MAX_RETRIES):
        try:
            r = requests.get(url, headers=UA, timeout=30)
        except requests.RequestException as e:
            wait = min(BACKOFF_BASE * (2 ** attempt), BACKOFF_MAX)
            _log(f"    metadata {item}: {type(e).__name__}, retry in {wait}s")
            time.sleep(wait)
            continue
        if r.status_code == 200:
            try:
                data = r.json()
            except ValueError:
                return None
            # Archive.org returns {} for items that do not exist
            if not data or "files" not in data:
                return None
            return data
        if r.status_code in RETRY_STATUSES:
            wait = min(BACKOFF_BASE * (2 ** attempt), BACKOFF_MAX)
            _log(f"    metadata {item}: HTTP {r.status_code}, retry in {wait}s")
            time.sleep(wait)
            continue
        # 404 etc. — item doesn't exist
        return None
    # All retries exhausted on network error or 5xx — API was unreachable
    return _META_UNREACHABLE


def _pick_best_pdf(meta: dict, item: str) -> tuple[str | None, str | None]:
    """
    Given a metadata dict, find the best PDF file.
    Returns (download_url, server_filename). Returns (None, None) if nothing found.

    Strategy:
      1. Prefer files whose `format` is 'Text PDF' or 'Abbyy PDF'.
      2. Prefer the largest PDF.
      3. Fall back to any .pdf file.
    """
    pdfs = []
    for f in meta.get("files", []):
        name = f.get("name", "")
        if not name.lower().endswith(".pdf"):
            continue
        # Skip encrypted PDFs (lending artifacts)
        if "_encrypted" in name.lower():
            continue
        try:
            size = int(f.get("size", "0"))
        except ValueError:
            size = 0
        fmt = (f.get("format") or "").lower()
        pdfs.append((fmt, size, name))

    if not pdfs:
        return None, None

    # Prefer Text/Abbyy PDFs, then largest
    def _score(p):
        fmt, size, _ = p
        preferred = any(k in fmt for k in ("text pdf", "abbyy"))
        return (preferred, size)

    pdfs.sort(key=_score, reverse=True)
    _, _, best = pdfs[0]
    return f"https://archive.org/download/{item}/{best}", best


def resolve_pdf_url(item: str) -> tuple[str | None, str | None]:
    """
    Given an archive.org item, find its best PDF file via the metadata API.
    Returns (download_url, server_filename). Returns (None, None) if nothing found.
    """
    meta = fetch_metadata(item)
    if meta is _META_UNREACHABLE:
        return _META_UNREACHABLE, None
    if not meta:
        return None, None
    return _pick_best_pdf(meta, item)


# ---------------------------------------------------------------------------

def _validate(path: Path, min_bytes: int, markers: list[bytes] | None) -> bool:
    if not path.exists():
        return False
    size = path.stat().st_size
    if size < min_bytes:
        _log(f"    ✗ too small ({size} bytes < {min_bytes})")
        return False
    with path.open("rb") as fh:
        head = fh.read(1024)
    if head.startswith(b"<!") or head.lstrip().lower().startswith(b"<html"):
        _log("    ✗ looks like HTML landing page")
        return False
    if not head.startswith(b"%PDF") and not head.startswith(b"PK") and not head.startswith(b"\x1f\x8b"):
        # Not PDF, not zip, not gzip — may be djvu.txt which is fine
        if markers:
            # fall through to marker check
            pass
        else:
            _log("    ✗ unknown file format")
            return False
    if markers:
        # Read more for marker search
        with path.open("rb") as fh:
            blob = fh.read(1 << 20)
        low = blob.lower()
        if not any(m.lower() in low for m in markers):
            _log("    ✗ expected markers not found")
            return False
    return True


def download_with_retry(
    url: str,
    dest: Path,
    min_bytes: int = 200_000,
    markers: list[bytes] | None = None,
) -> bool:
    """Stream-download url → dest, validate, retry on 5xx/429."""
    for attempt in range(MAX_RETRIES):
        try:
            r = requests.get(url, headers=UA, timeout=TIMEOUT, stream=True, allow_redirects=True)
        except requests.RequestException as e:
            wait = min(BACKOFF_BASE * (2 ** attempt), BACKOFF_MAX)
            _log(f"    network {type(e).__name__}: {e!s}, retry in {wait}s")
            time.sleep(wait)
            continue

        if r.status_code == 200:
            try:
                with dest.open("wb") as fh:
                    for chunk in r.iter_content(CHUNK):
                        if chunk:
                            fh.write(chunk)
            except Exception as e:
                _log(f"    write failed: {e}")
                if dest.exists():
                    dest.unlink()
                return False
            if _validate(dest, min_bytes, markers):
                return True
            if dest.exists():
                dest.unlink()
            # Not retrying on validation failure — the server gave a bad file
            return False

        if r.status_code in RETRY_STATUSES:
            wait = min(BACKOFF_BASE * (2 ** attempt), BACKOFF_MAX)
            _log(f"    HTTP {r.status_code}, retry in {wait}s (attempt {attempt + 1}/{MAX_RETRIES})")
            time.sleep(wait)
            continue

        _log(f"    HTTP {r.status_code}, giving up")
        return False

    _log("    retries exhausted")
    return False


def archive_pdf_smart(
    item: str,
    dest: Path,
    min_bytes: int = 200_000,
) -> bool:
    """
    Resolve the item's real PDF via the metadata API, then download with retry.
    Falls back to the conventional /download/{item}/{item}.pdf URL if metadata is unreachable.

    Circuit breaker: after N consecutive failures, returns False immediately
    without hitting the network. Reset between script runs.
    """
    if circuit_is_open():
        _log(f"    [circuit open] skipping {item}")
        return False

    meta = fetch_metadata(item)

    # Metadata API is unreachable (5xx/network). We don't know if the item
    # exists, so fall back to the conventional URL as a last resort.
    if meta is _META_UNREACHABLE:
        _log(f"    [metadata unreachable, trying convention] /download/{item}/{item}.pdf")
        ok = download_with_retry(
            f"https://archive.org/download/{item}/{item}.pdf",
            dest,
            min_bytes=min_bytes,
        )
        if ok:
            _note_success()
        else:
            _note_failure()
        return ok

    if not meta:
        _log(f"    [not on archive.org] {item}")
        return False

    # Check if the item is lending-restricted (borrow-only).
    # These always return 401/403 on direct download — skip without
    # counting as a failure (it's not a server problem).
    item_meta = meta.get("metadata", {})
    restricted = item_meta.get("access-restricted-item", "false")
    if restricted == "true" or restricted is True:
        title = item_meta.get("title", "?")
        # Infer task from dest path (e.g. .../13_mystics/emmerich/file.pdf)
        task_hint = ""
        for parent in dest.parents:
            if parent.name and parent.name[:2].isdigit():
                task_hint = parent.name
                break
        _log(f"    [lending-restricted] {item} — borrow-only, skipping")
        _record_lending_restricted(item, title, task_hint)
        return False

    # Find best PDF from metadata
    url, fname = _pick_best_pdf(meta, item)

    if url:
        _log(f"    [metadata] {item} → {fname}")
        if download_with_retry(url, dest, min_bytes=min_bytes):
            _note_success()
            return True
        _note_failure()
        return False

    _log(f"    [no PDF found] {item}")
    return False