#!/usr/bin/env python3
"""Publish the new remaining-frame delivery only after full payload verification."""
import argparse
import hashlib
import json
from pathlib import Path
import zipfile
import numpy as np
from PIL import Image


def sha(path):
    h=hashlib.sha256()
    with Path(path).open('rb') as f:
        for chunk in iter(lambda:f.read(4*1024*1024),b''):h.update(chunk)
    return h.hexdigest()


def finalize(output):
    output=Path(output).resolve(); project=Path(__file__).resolve().parents[1]
    delivery=json.loads((output/'delivery_manifest.json').read_text())
    merge=json.loads((output/'merge_summary.json').read_text())
    archive=Path(delivery['archive']); workspace=output/'anylabeling'
    if delivery['predicted_frames']!=3788 or delivery['errors'] or delivery['test_labels_scored']:
        raise ValueError('Incomplete or incorrectly scoped delivery')
    if sha(archive)!=delivery['archive_sha256']:raise ValueError('Archive hash mismatch')
    files={str(p.relative_to(workspace)):p for p in workspace.rglob('*') if p.is_file()}
    prefix=archive.stem+'/'
    with zipfile.ZipFile(archive) as z:
        if z.testzip() is not None:raise ValueError('ZIP CRC failed')
        if set(z.namelist())!={prefix+name for name in files}:raise ValueError('ZIP member identity mismatch')
        for name,path in files.items():
            if hashlib.sha256(z.read(prefix+name)).hexdigest()!=sha(path):
                raise ValueError('ZIP payload differs from audited workspace')
    checked={json.loads(l)['frame_id'] for l in (output/'inputs/all_checked_annotations.jsonl').read_text().splitlines()}
    records=[json.loads(l) for l in (workspace/'frames.jsonl').read_text().splitlines()]
    ids={r['frame_id'] for r in records}
    if len(ids)!=3788 or ids & checked:raise ValueError('Manual frame overlap')
    black=0
    for r in records:
        with Image.open(workspace/r['image']) as image:
            black+=np.asarray(image).max()==0
    if black:raise ValueError('Retired black primary frame present')
    result={'state':'passed','zip_sha256':sha(archive),'zip_crc':'passed','zip_payloads_verified':len(files),
        'checked_frames_excluded':len(checked),'manual_overlap':0,'black_primary_frames':0,
        'remaining_frames':len(ids),'priority_override_count':merge['priority_overrides']}
    (output/'payload_audit.json').write_text(json.dumps(result,indent=2))
    marker=f'<!-- manual-update-complete:{output.name} -->'
    v=delivery['best_validation']; rel=archive.relative_to(project)
    body=(f'\n\n{marker}\n## 本轮训练与修正包已完成\n\n'
        f'本轮6,586张人工检查结果已归档；其中原train的6,552张按组隔离为5,569训练/983验证，'
        f'原val/test的34张仅保存修正结果并排除交付。新YOLO26s完成{delivery["completed_epochs"]}轮，'
        f'最佳epoch{delivery["best_epoch"]}（从0计），内部人工验证AP50:95={v["AP50_95"]:.5f}，'
        f'建议阈值{delivery["threshold"]:.2f}。未开展最终测试评分。\n\n'
        f'[下载剩余3,788帧修正包]({rel.as_posix()})：val1,439/test2,349，无train帧；'
        '已排除全部6,586张人工帧及不可恢复黑图。配对、原生坐标、待审标志、ZIP CRC及逐成员哈希均通过。'
        '空预测仍待审，val/test属于模型辅助人工工作版本。\n\n'
        f'ZIP SHA-256：`{delivery["archive_sha256"]}`。\n')
    for filename in ['docs/human_correction_6552_report.md','docs/human_correction_latest_report.md','README.md']:
        path=project/filename;text=path.read_text()
        if marker not in text:
            heading,rest=text.split('\n',1)
            # Links in documents resolve relative to docs, README resolves at the project root.
            current=body.replace(f']({rel.as_posix()})',f'](../{rel.as_posix()})') if filename.startswith('docs/') else body
            path.write_text(heading+'\n'+current+'\n'+rest)
    control=project/'configs/active_dataset.json';config=json.loads(control.read_text())
    config['human_training']=str(output);config['remaining_review_package']=str(archive)
    config['latest_assistant_checkpoint']=delivery['checkpoint'];config['latest_assistant_delivery']=str(output/'delivery_manifest.json')
    control.write_text(json.dumps(config,ensure_ascii=False,indent=2))
    return result


if __name__=='__main__':
    p=argparse.ArgumentParser();p.add_argument('--output',required=True);a=p.parse_args()
    print(json.dumps(finalize(a.output),ensure_ascii=False))
