#!/usr/bin/env python3
"""Pinned HF downloads; immutable raw archives; CRC-checked pilot extraction."""
import argparse
import hashlib
import io
import json
import random
import shutil
import struct
import subprocess
import time
import urllib.error
import urllib.request
import zipfile
import zlib
from pathlib import Path, PurePosixPath

REPO = 'ParityLab/Hand-Datasets-Raw'
REVISION = '8818d04d6cbdc0bf32b6ef949257b82310f16798'
HERE = Path(__file__).resolve().parent


def sha(path):
    with open(path, 'rb') as f:
        return hashlib.file_digest(f, 'sha256').hexdigest()


def safe_path(root, name):
    p = PurePosixPath(name)
    if p.is_absolute() or '..' in p.parts or '\\' in name:
        raise ValueError(f'Unsafe archive path: {name}')
    target = root.joinpath(*p.parts)
    if not target.resolve().is_relative_to(root.resolve()):
        raise ValueError(f'Path escapes extraction root: {name}')
    return target


class RemoteFile(io.RawIOBase):
    """Small seekable HTTP range reader, suitable for ZIP central directories."""
    def __init__(self, path, size):
        self.url = f'https://huggingface.co/datasets/{REPO}/resolve/{REVISION}/{path}'
        self.size, self.pos = size, 0
        self.cache_start, self.cache = 0, b''

    def seekable(self):
        return True

    def tell(self):
        return self.pos

    def seek(self, offset, whence=0):
        self.pos = offset + (0 if whence == 0 else self.pos if whence == 1 else self.size)
        if self.pos < 0:
            raise ValueError('Negative offset')
        return self.pos

    def read(self, n=-1):
        n = min(self.size - self.pos, n if n >= 0 else self.size)
        if n <= 0:
            return b''
        if not (self.cache_start <= self.pos and self.pos + n <= self.cache_start + len(self.cache)):
            end = min(self.size, self.pos + max(n, 262144)) - 1
            req = urllib.request.Request(self.url, headers={'Range': f'bytes={self.pos}-{end}'})
            for attempt in range(4):
                try:
                    with urllib.request.urlopen(req, timeout=30) as r:
                        expected = f'bytes {self.pos}-{end}/{self.size}'
                        if r.status != 206 or r.headers.get('Content-Range') != expected:
                            raise RuntimeError(f'Server did not honor byte range {expected}')
                        self.cache = r.read()
                        self.cache_start = self.pos
                        if len(self.cache) != end - self.pos + 1:
                            raise IOError('Truncated HTTP range')
                    break
                except (urllib.error.URLError, TimeoutError, OSError):
                    if attempt == 3:
                        raise
                    time.sleep(attempt + 1)
        start = self.pos - self.cache_start
        self.pos += n
        return self.cache[start:start+n]


def checksums(dataset):
    return json.loads((HERE / 'source_evidence' / f'{dataset}-archive_checksums.json').read_text())['files']


def extract_zip(archive, target, origin, selected=None):
    records = []
    with zipfile.ZipFile(archive) as z:
        for info in z.infolist():
            if info.is_dir() or (selected is not None and info.filename not in selected):
                continue
            # Do not create symlinks/devices. Treat regular members only.
            mode = info.external_attr >> 16
            if mode & 0o170000 not in (0, 0o100000):
                raise ValueError(f'Non-regular archive member: {info.filename}')
            p = safe_path(target, info.filename)
            p.parent.mkdir(parents=True, exist_ok=True)
            if shutil.disk_usage(p.parent).free < info.file_size + 256 * 1024**2:
                raise RuntimeError('Insufficient space for next uncompressed member')
            with z.open(info) as src, p.open('xb') as dst:
                shutil.copyfileobj(src, dst)  # ZipFile checks member CRC at EOF.
            records.append({'path': p.relative_to(target).as_posix(), 'sha256': sha(p),
                            'source_locator': dict(origin, member=info.filename)})
    return records


def download(dataset, root):
    from huggingface_hub import hf_hub_download
    entries = [e for e in checksums(dataset) if '#' not in e['path']]
    need = sum(e['size'] for e in entries if not (root / e['path']).exists())
    root.mkdir(parents=True, exist_ok=True)
    if shutil.disk_usage(root).free < need + 5 * 1024**3:
        raise RuntimeError(f'Insufficient disk; raw downloads need {need / 1e9:.1f} GB plus reserve')
    for e in entries:
        dst = root / e['path']
        if dst.exists():
            if dst.stat().st_size != e['size'] or sha(dst) != e['sha256']:
                raise RuntimeError(f'Existing raw file differs: {dst}')
            print(f'Already verified: {dst}', flush=True)
            continue
        remote = f'{dataset}/raw/{e["path"]}'
        p = Path(hf_hub_download(REPO, remote, repo_type='dataset', revision=REVISION,
                                local_dir=root / '_hf'))
        if p.stat().st_size != e['size'] or sha(p) != e['sha256']:
            raise RuntimeError(f'Archive checksum mismatch: {remote}')
        dst.parent.mkdir(parents=True, exist_ok=True)
        if not dst.exists():
            p.replace(dst)
        elif sha(dst) != e['sha256']:
            raise RuntimeError(f'Existing raw file differs: {dst}')
        print(dst, flush=True)


def pilot_ego(target, count, seed):
    e = checksums('EgoHOS')[0]
    remote = f'EgoHOS/raw/{e["path"]}'
    f = RemoteFile(remote, e['size'])
    with zipfile.ZipFile(f) as z:
        images = [n for n in z.namelist() if '/image/' in n and n.lower().endswith(('.jpg', '.jpeg', '.png'))]
        chosen = []
        for split in ('train', 'val', 'test_indomain', 'test_outdomain'):
            pool = sorted(n for n in images if f'/{split}/image/' in n)
            chosen.extend(random.Random(seed).sample(pool, min(count, len(pool))))
        if not chosen:
            raise RuntimeError('Expected outer ZIP image layout not found')
        names = set(z.namelist())
        members = set(chosen)
        for name in chosen:
            label = str(PurePosixPath(name.replace('/image/', '/label/')).with_suffix('.png'))
            if label not in names:
                raise RuntimeError(f'Label missing: {label}')
            members.add(label)
    records = extract_zip(f, target, {'repo': REPO, 'revision': REVISION, 'file': remote,
                          'archive_sha256_expected': e['sha256'],
                          'verification': 'selected member CRC + SHA256; full archive NOT downloaded/verified'}, members)
    return records


def pilot_oak_rgb(target, count, seed):
    """Pilot-only: sample first 128 PNG local entries from volume z01, stratified by camera.

    This is NOT a whole-dataset random sample and cannot substitute for a full download.
    Known split ZIP uses local CRC/sizes; other layouts fail closed.
    """
    entry = next(e for e in checksums('OakInk-v1') if e['path'].endswith('.z01'))
    remote = f'OakInk-v1/raw/{entry["path"]}'
    f = RemoteFile(remote, entry['size'])
    if f.read(4) != b'PK\x07\x08':
        raise ValueError('Unexpected split ZIP signature')
    pool = {}
    seen = 0
    while seen < 128:
        header = f.read(30)
        if len(header) != 30 or header[:4] != b'PK\x03\x04':
            raise ValueError('Unsupported split ZIP local header; use full download + 7z')
        _, version, flags, method, _, _, crc, compressed, size, nlen, xlen = struct.unpack('<4s5H3I2H', header)
        name = f.read(nlen).decode('utf-8' if flags & 2048 else 'cp437')
        f.read(xlen)
        if flags & 9 or compressed == 0xffffffff or method not in (0, 8):
            raise ValueError('Unsupported encrypted/descriptor/ZIP64 local member; use full extraction')
        offset = f.tell()
        if offset + compressed > entry['size']:
            raise ValueError('Pilot member crosses volume; use full extraction')
        if name.endswith('.png'):
            seen += 1
            camera = PurePosixPath(name).stem.rsplit('_', 1)[0]
            pool.setdefault(camera, []).append((name, offset, compressed, size, method, crc))
        f.seek(offset + compressed)
    rng = random.Random(seed)
    records = []
    for camera in sorted(pool):
        for name, offset, compressed, size, method, crc in rng.sample(pool[camera], min(count, len(pool[camera]))):
            f.seek(offset)
            packed = f.read(compressed)
            content = packed if method == 0 else zlib.decompress(packed, -15)
            if len(content) != size or zlib.crc32(content) & 0xffffffff != crc:
                raise ValueError(f'Pilot member CRC/size mismatch: {name}')
            path = safe_path(target / 'image', name)
            path.parent.mkdir(parents=True, exist_ok=True)
            path.write_bytes(content)
            records.append({'path': path.relative_to(target).as_posix(), 'sha256': sha(path),
                'source_locator': {'repo': REPO, 'revision': REVISION, 'file': remote, 'member': name,
                    'compressed_member_offset': offset, 'archive_sha256_expected': entry['sha256'],
                    'verification': 'selected member CRC + SHA256; complete volume NOT verified',
                    'selection_scope': 'first 128 PNG entries in z01; camera-stratified, NOT global random'}})
    return records


def extract_local(dataset, raw, target):
    entries = [e for e in checksums(dataset) if '#' not in e['path']]
    for e in entries:
        p = raw / e['path']
        if not p.exists() or p.stat().st_size != e['size'] or sha(p) != e['sha256']:
            raise RuntimeError(f'Download/verify raw first: {p}')
    records = []
    for e in entries:
        p = raw / e['path']
        if p.suffix != '.zip':
            continue
        origin = {'repo': REPO, 'revision': REVISION, 'file': f'{dataset}/raw/{e["path"]}',
                  'archive_sha256': e['sha256']}
        if dataset == 'EgoHOS':
            with zipfile.ZipFile(p) as z:
                members = {n for n in z.namelist() if '/image/' in n or '/label/' in n or '/contact/' in n}
            records.extend(extract_zip(p, target, origin, members))
        elif '/stream_zipped/' in e['path']:
            tool = shutil.which('7zz') or shutil.which('7z')
            if not tool:
                raise RuntimeError('Install 7zip (7zz or 7z) for OakInk split ZIP volumes')
            # Official immutable archives were SHA256 checked above. Never concatenate volumes.
            dest = target / 'image'
            subprocess.run([tool, 'x', '-y', '-o' + str(dest), str(p)], check=True)
            volumes = [dict(file=f'{dataset}/raw/{v["path"]}', sha256=v['sha256'])
                       for v in entries if '/stream_zipped/' in v['path']]
            for image in sorted((dest / 'stream_release_v2').rglob('*.png')):
                records.append({'path': image.relative_to(target).as_posix(), 'sha256': sha(image),
                                'source_locator': dict(origin, volumes=volumes,
                                                       member=image.relative_to(dest).as_posix())})
        else:
            dest = target / ('image' if '/image/' in e['path'] else 'shape' if '/shape/' in e['path'] else '')
            batch = extract_zip(p, dest, origin)
            for r in batch:
                r['path'] = (dest.relative_to(target) / r['path']).as_posix()
            records.extend(batch)
    return records


def main():
    p = argparse.ArgumentParser(description=__doc__)
    p.add_argument('command', choices=['download', 'extract', 'pilot-egohos', 'pilot-oakink-rgb', 'extract-annotations', 'fetch-oakink-objects'])
    p.add_argument('--dataset', choices=['EgoHOS', 'OakInk-v1'], default='EgoHOS')
    p.add_argument('--raw', type=Path)
    p.add_argument('--out', type=Path)
    p.add_argument('--archive', type=Path, help='Official anno_v2_1.zip for extract-annotations')
    p.add_argument('--per-split', type=int, default=4)
    p.add_argument('--seed', type=int, default=20260928)
    a = p.parse_args()
    if a.command == 'fetch-oakink-objects':
        if not a.out or not (a.out / 'sources.jsonl').is_file():
            p.error('Existing extracted --out with sources.jsonl required')
        entry = next(e for e in checksums('OakInk-v1') if e['path'] == 'data/image/obj.zip')
        remote = 'OakInk-v1/raw/' + entry['path']
        archive = a.out.parent / 'raw-pilot' / remote
        archive.parent.mkdir(parents=True, exist_ok=True)
        if not archive.exists():
            archive.write_bytes(RemoteFile(remote, entry['size']).read())
        if sha(archive) != entry['sha256']:
            raise ValueError('Object archive checksum mismatch')
        rows = extract_zip(archive, a.out / 'image', {'repo': REPO, 'revision': REVISION,
            'file': remote, 'archive_sha256': entry['sha256']})
        with (a.out / 'sources.jsonl').open('a') as f:
            for r in rows:
                r['path'] = 'image/' + r['path']
                f.write(json.dumps(r, ensure_ascii=False) + '\n')
        print(f'Imported {len(rows)} object assets; original zip retained at {archive}')
        return
    if a.command == 'extract-annotations':
        if not a.archive or not a.out or not (a.out / 'sources.jsonl').is_file():
            p.error('--archive and existing extracted --out with sources.jsonl required')
        with zipfile.ZipFile(a.archive) as z:
            regular = [n for n in z.namelist() if not n.endswith('/')]
            if not regular or not all(n.startswith('anno/') for n in regular):
                p.error('Expected archive rooted at anno/; inspect actual layout before importing')
            if any(safe_path(a.out / 'image', n).exists() for n in regular):
                p.error('Annotation target already exists; refusing to overwrite')
        rows = extract_zip(a.archive, a.out / 'image', {'file': str(a.archive.resolve()),
            'archive_sha256': sha(a.archive), 'origin': 'user-supplied official annotation archive; not present in pinned HF raw'})
        with (a.out / 'sources.jsonl').open('a') as f:
            for r in rows:
                r['path'] = 'image/' + r['path']
                f.write(json.dumps(r, ensure_ascii=False) + '\n')
        print(f'Imported {len(rows)} annotation files with provenance')
        return
    if a.command == 'download':
        if not a.raw:
            p.error('--raw required')
        download(a.dataset, a.raw)
        return
    if not a.out or a.out.exists():
        p.error('--out must be a new directory (raw is never modified)')
    if a.command == 'extract' and not a.raw:
        p.error('--raw required')
    if a.per_split < 1:
        p.error('--per-split must be positive')
    a.out.mkdir(parents=True)
    if a.command == 'pilot-egohos':
        a.dataset = 'EgoHOS'
        records = pilot_ego(a.out, a.per_split, a.seed)
    elif a.command == 'pilot-oakink-rgb':
        a.dataset = 'OakInk-v1'
        records = pilot_oak_rgb(a.out, a.per_split, a.seed)
    else:
        records = extract_local(a.dataset, a.raw, a.out)
    with (a.out / 'sources.jsonl').open('w') as f:
        for r in records:
            f.write(json.dumps(r, ensure_ascii=False) + '\n')
    (a.out / 'acquisition.json').write_text(json.dumps({'dataset': a.dataset, 'revision': REVISION,
        'command': a.command, 'seed': a.seed, 'files': len(records)}, indent=2))
    print(f'{len(records)} extracted files -> {a.out}')


if __name__ == '__main__':
    main()