#!/usr/bin/env python3
"""libgen.vg downloader (the box-verified flow, 2026-07-14).

Pipeline (curl engine — Python's DNS is flaky on this box, curl is not):
  search:  tools/lg.py "query"                    -> edition ids
  ed:      json.php?object=e&ids=E                 -> title/author/publisher/year + files{f_id, md5}
  f:       json.php?object=f&addkeys=*&ids=F       -> filesize, extension
  dl:      ads.php?md5=M (UA+cookie+referer)       -> card HTML with
          libgen.bz/get.php?md5=M&key=KEY          -> one-time download URL
          curl that (same cookie jar)              -> real bytes

Usage:
    tools/lgdl.py ed   <edition_id> [more_ids...]
    tools/lgdl.py dl   <file_id> <out_dir> <target_name> [edition_id]
    tools/lgdl.py md5  <file_id>
    tools/lgdl.py precheck <file_id> <edition_id> [expected_pages]

precheck (no download, ~2 libgen calls): file record vs the target edition.
  The ONLY reliable ownership signal is the file `locator` (the importer's
  original Windows path — it names the real book). libgen's `file.editions`
  reverse-map uses internal IDs that do NOT resolve via object=e (they return
  unrelated books), so it is logged as a hint but never gates the verdict.
  RED   (exit 2) — locator non-empty and shares NO content word with the target
        title/author (locator names a different book), or a foreign-language
        batch path for a Cyrillic target.
  OK    (exit 0) — locator non-empty and shares content words with the target.
  SUSPECT(exit 1) — locator empty (no signal), too few target tokens to judge,
        or size implausible vs expected pages. ALWAYS post-verify content.
`dl` with the optional 4th arg runs precheck first and aborts on RED.

Verification on dl: final size must equal libgen filesize; magic-byte sniff
(fb2/pdf/epub/doc) must not be HTML. Card+key are re-fetched up to 3x.
Content-verify after dl (AGENTS.md lesson 8): pdfinfo pages vs RSL + first-page
text must contain the RU title/author. Politeness: ~4s between libgen requests.
"""
import json, os, re, subprocess, sys, time

UA = ("Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 "
      "(KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36")
BASE = "https://libgen.vg"
JAR = "/tmp/lgdl_cookies.txt"


def curl(url, out=None, referer=None, binary=False, timeout=240, tries=3, extra=None,
         connect_timeout=None):
    cmd = ["curl", "-s", "--fail", "--max-time", str(timeout),
           "-c", JAR, "-b", JAR,
           "-H", "User-Agent: " + UA,
           "-H", "Accept: */*",
           "-H", "Accept-Language: ru,en;q=0.9",
           "-H", "Referer: " + (referer or BASE + "/"),
           "-L"]
    if connect_timeout:
        cmd += ["--connect-timeout", str(connect_timeout)]
    if extra:
        cmd += extra
    if out:
        cmd.append("-o")
        cmd.append(out)
    cmd.append(url)
    last = None
    for i in range(tries):
        p = subprocess.run(cmd, capture_output=True, timeout=timeout + 30)
        if p.returncode == 0:
            if out is None:
                data = p.stdout
                if not binary:
                    try:
                        data = data.decode("utf-8")
                    except UnicodeDecodeError:
                        data = data.decode("utf-8", "replace")
                return data
            return p.returncode
        last = p.returncode
        time.sleep(5 * (i + 1))
    raise RuntimeError(f"curl failed (code {last}) after {tries} tries: {url}")


def jget(path):
    return json.loads(curl(BASE + "/" + path))


def file_info(f_id):
    d = jget(f"json.php?object=f&addkeys=*&ids={f_id}")
    return d[str(f_id)]


def edition(e_id):
    d = jget(f"json.php?object=e&ids={e_id}")
    e = d[str(e_id)]
    files = []
    for rel, f in (e.get("files") or {}).items():
        fi = file_info(f["f_id"])
        time.sleep(1)
        files.append({
            "f_id": f["f_id"], "md5": f["md5"],
            "ext": fi.get("extension"), "size": int(fi.get("filesize") or 0),
        })
    e["files"] = files
    return e


def sniff(data: bytes) -> str:
    if data[:5] == b"<?xml" and b"FictionBook" in data[:500]:
        return "fb2"
    if data[:5] == b"%PDF-":
        return "pdf"
    if data[:4] == b"PK\x03\x04":
        return "epub/zip"
    if data[:8] == b"\xd0\xcf\x11\xe0\xa1\xb1\x1a\xe1":
        return "doc/ole"
    if data[:8] in (b"AT&TFORM", b"AT&ToC\01", b"AT&ToC\02"):
        return "djvu"
    if b"<html" in data[:2000].lower():
        return "HTML-ERROR"
    return "unknown"


def download(f_id, out_dir, name, tries=16):
    fi = file_info(f_id)
    md5, size, ext = fi["md5"], int(fi.get("filesize") or 0), fi.get("extension") or "bin"
    if not name.lower().endswith("." + ext.lower()):
        name = f"{name}.{ext}"
    os.makedirs(out_dir, exist_ok=True)
    out = os.path.join(out_dir, name)
    stall = 0
    for t in range(1, tries + 1):
        card = None
        print(f"  try {t}: card fetch start")
        for c in range(5):  # card fetch with backoff (libgen DB is flaky)
            try:
                card = curl(f"{BASE}/ads.php?md5={md5}",
                            referer=f"{BASE}/file.php?id={f_id}", tries=1,
                            timeout=90, connect_timeout=20)
                break
            except RuntimeError:
                print(f"  try {t}/{c+1}: card fetch failed — back off")
                time.sleep(8 * (c + 1))
        if card is None:
            print(f"  try {t}: card fetch failed 5x — back off harder")
            time.sleep(30)
            continue
        m = re.search(r'(https?://[^"\']*get\.php\?md5=' + md5 + r'&key=[^"\']+)', card)
        if not m:
            print(f"  try {t}: no get.php link in card ({len(card)}b) — retry")
            time.sleep(4)
            continue
        url = m.group(1)
        # speed probe: 6s range fetch on the fresh CDN edge; <15KB/s => this
        # edge is wedged, re-draw a new get.php link (libgen 302s to different
        # storage hosts per request)
        probe_ok = False
        for p in range(3):
            try:
                pr = subprocess.run(["curl", "-s", "--fail", "-o", "/dev/null",
                                     "-r", "0-300000", "--max-time", "6",
                                     "--connect-timeout", "10",
                                     "-H", "User-Agent: " + UA,
                                     "-H", "Referer: " + f"{BASE}/ads.php?md5={md5}",
                                     "-w", "%{size_download} %{speed_download}",
                                     url], capture_output=True, text=True, timeout=20)
                if pr.returncode == 0:
                    parts = pr.stdout.split()
                    spd = float(parts[1]) if len(parts) > 1 else 0
                    print(f"  try {t}: probe edge {p+1}: {spd/1024:.0f} KB/s")
                    if spd >= 15 * 1024:
                        probe_ok = True
                        break
            except Exception:
                pass
            print(f"  try {t}: edge {p+1} slow/wedged — re-drawing get.php link")
            time.sleep(3)
            try:
                card2 = curl(f"{BASE}/ads.php?md5={md5}",
                             referer=f"{BASE}/file.php?id={f_id}", tries=1,
                             timeout=90, connect_timeout=20)
                m2 = re.search(r'(https?://[^"\']*get\.php\?md5=' + md5 + r'&key=[^"\']+)', card2)
                if m2:
                    url = m2.group(1)
            except RuntimeError:
                pass
        if not probe_ok:
            print(f"  try {t}: all 3 edges slow — waiting out the CDN")
        tmp = out + ".part"
        prev = os.path.getsize(tmp) if os.path.exists(tmp) else 0
        print(f"  try {t}: data fetch start ({prev}b resumed)")
        try:
            rc = curl(url, out=tmp, referer=f"{BASE}/ads.php?md5={md5}", binary=True,
                      timeout=3600, connect_timeout=30,
                      # abort hung connections: <500B/s for 300s
                      extra=["-C", "-", "--speed-time", "300", "--speed-limit", "500"])
        except RuntimeError as e:
            print(f"  try {t}: fetch error ({e}) — back off")
            time.sleep(45)
            continue
        cur = os.path.getsize(tmp) if os.path.exists(tmp) else 0
        # resume-hang detection: zero-progress resume at the same offset =>
        # this CDN offset is wedged; wipe and start fresh (first stall is enough)
        if 0 < prev == cur:
            print(f"  try {t}: resume stalled at {prev}b — wiping part, fresh start")
            os.remove(tmp)
            stall = 0
            cur = 0
        data = open(tmp, "rb").read()
        kind = sniff(data)
        ok_size = (size == 0) or (len(data) == size)
        print(f"  try {t}: {len(data)}b (expected {size}) kind={kind}")
        if kind == "HTML-ERROR":
            os.remove(tmp)
        if kind == "HTML-ERROR" or not ok_size:
            time.sleep(4)
            continue
        os.replace(tmp, out)
        print(f"  OK -> {out} ({len(data)}b, {kind})")
        return out
    print(f"  FAILED after {tries} tries: file_id={f_id} md5={md5}")
    return None


# stop-ish / generic tokens to ignore when matching locator vs target
_STOP = set("the of and in to a an for with from on by edition new second first "
           "vol volume pdf epub fb2 doc djvu mobi txt part pt series book "
           "translated translation revised updated complete full original"
           .split())

def _tokens(s):
    return {w for w in re.findall(r"[a-z\u0400-\u04ff]{4,}", (s or "").lower())
            if w not in _STOP}

def _loc_tokens(s):
    # locator words: mask alphanumeric runs containing digits (IDs/hashes/ISBNs)
    # so "d1e2f4...aaae..." does not yield fake word 'aaae'
    masked = re.sub(r"[a-z0-9]+\d[a-z0-9]*", " ", (s or "").lower())
    return _tokens(masked)

def precheck(f_id, e_id, expected_pages=None):
    """Mislinked-file gate. Returns 0=OK, 1=SUSPECT, 2=BAD (see module doc).

    CRITICAL: pass the edition files-map VALUE (`f_id`), not the map key —
    the key is a different (often stale/mislinked) file record (verified
    2026-09-25: Solomon — key 92709688 = Russian book, f_id 93253044 = real book).

    Signals:
    1) reverse map: for a real f_id record the `editions` values ARE the public
       edition IDs — value == target is a strong OK signal; for a key's record
       it contains other numbers (no match => weak negative, never hard BAD).
    2) locator: words in the locator vs target title/author words.
       - no content words (hash/ISBN/number) => no signal (SUSPECT)
       - >=2 shared words (or 1 for a short target) => OK
       - words present, none shared => BAD (names a different book)
    ALWAYS content-verify after download (ground truth).
    """
    fi = file_info(f_id)
    time.sleep(2)
    ext, size = fi.get("extension"), int(fi.get("filesize") or 0)
    locator = (fi.get("locator") or "").strip()
    reasons = []

    rev = {k: v.get("e_id") for k, v in (fi.get("editions") or {}).items()}
    rev_match = int(e_id) in {int(v) for v in rev.values() if v}
    if rev:
        reasons.append(f"reverse-map: {rev} {'MATCHES target' if rev_match else '(no target)'}")

    try:
        ed = jget(f"json.php?object=e&addkeys=title,author&ids={e_id}").get(str(e_id), {})
        time.sleep(2)
        target = f"{ed.get('title','')} {ed.get('author','')}"
    except Exception:
        target = ""
    t_tok, l_tok = _tokens(target), _loc_tokens(locator)
    shared = t_tok & l_tok

    if not locator or not l_tok:
        loc_signal = "none"
        reasons.append(f"locator has no content words (hash/ISBN/empty) — no signal: {locator[:90]}")
    elif len(shared) >= 2 or (len(shared) == 1 and len(t_tok) <= 3):
        loc_signal = "match"
        reasons.append(f"locator names the target (shared: {sorted(shared)}): {locator[:120]}")
    else:
        loc_signal = "mismatch"
        low = locator.lower()
        target_cyr = re.search(r"[\u0400-\u04FF]", target)
        cyrillic_loc = re.search(r"[\u0400-\u04FF]", locator)
        foreign = re.search(r"(spa\\|spanish|english|epublibre|0day|\\fiction\\)", low)
        if target_cyr and not cyrillic_loc:
            reasons.append(f"Cyrillic target but locator is non-Cyrillic: {locator[:120]}")
        elif foreign:
            reasons.append(f"locator is a foreign batch naming another book: {locator[:120]}")
        else:
            reasons.append(f"locator names a DIFFERENT book (no shared words): {locator[:120]}")

    if rev_match and loc_signal != "mismatch":
        verdict = "OK"
    elif loc_signal == "match":
        verdict = "OK"
    elif loc_signal == "mismatch" and not rev_match:
        verdict = "BAD"
    else:
        verdict = "SUSPECT"  # none/none, or contradiction (rev match vs locator mismatch)
        if rev_match and loc_signal == "mismatch":
            reasons.append("CONTRADICTION: reverse-map matches but locator names another book — verify content")

    # size sanity vs expected pages (text ~200B/pp; scans 50KB..15MB/pp)
    if expected_pages and size:
        per_lo = 200 if ext in ("fb2", "epub", "txt", "html") else 10_000
        lo, hi = expected_pages * per_lo, expected_pages * 15_000_000
        if size < lo:
            if verdict == "OK":
                verdict = "SUSPECT"
            reasons.append(f"size {size}b < {lo}b for ~{expected_pages} pp (too small — fragment?)")
        elif size > hi:
            if verdict == "OK":
                verdict = "SUSPECT"
            reasons.append(f"size {size}b > {hi}b for ~{expected_pages} pp (too big — scan mismatch?)")

    print(f"precheck f_id={f_id} vs e_id={e_id}: {verdict} [{ext}, {size}b]")
    for r in reasons:
        print(f"  - {r}")
    return {"OK": 0, "SUSPECT": 1, "BAD": 2}[verdict]


def main():
    if len(sys.argv) < 2:
        sys.exit(__doc__)
    cmd = sys.argv[1]
    if cmd == "ed":
        for e_id in sys.argv[2:]:
            e = edition(e_id)
            keep = {k: e[k] for k in ("title", "title_add", "author", "publisher",
                                      "year", "pages", "series_name", "libgen_topic") if e.get(k)}
            print(f"[{e_id}] {json.dumps(keep, ensure_ascii=False)}")
            for f in e["files"]:
                print(f"    f_id={f['f_id']} {f['size']:>10}b {f['ext']:<6} md5={f['md5']}")
            time.sleep(1)
    elif cmd == "dl":
        f_id, out_dir, name = sys.argv[2], sys.argv[3], sys.argv[4]
        if len(sys.argv) > 5 and sys.argv[5].isdigit():
            rc = precheck(f_id, sys.argv[5])
            if rc == 2:
                print("  ABORT: precheck RED (mislinked file) — pick another f_id/edition or flib")
                sys.exit(2)
            if rc == 1:
                print("  precheck SUSPECT — downloading, MUST post-verify content")
            time.sleep(3)
        download(f_id, out_dir, name)
    elif cmd == "precheck":
        f_id, e_id = sys.argv[2], sys.argv[3]
        pages = int(sys.argv[4]) if len(sys.argv) > 4 else None
        sys.exit(precheck(f_id, e_id, pages))
    elif cmd == "md5":
        print(file_info(sys.argv[2])["md5"])
    elif cmd == "bymd5":
        # md5 -> f_id + metadata (for files found via lg.py file search)
        d = jget(f"json.php?object=f&addkeys=*&md5={sys.argv[2]}")
        if not d:
            print("not found"); sys.exit(1)
        fid = list(d.keys())[0]
        f = d[fid]
        print(f"f_id={fid} {f.get('extension')} {int(f.get('filesize') or 0)}b pages={f.get('pages')} ocr={f.get('ocr')} added={f.get('time_added')}")
        print(f"locator: {f.get('locator','')}")
        rev = {k: v.get('e_id') for k, v in (f.get('editions') or {}).items()}
        if rev: print(f"editions (hint): {rev}")
    else:
        sys.exit(__doc__)


if __name__ == "__main__":
    main()