"""Apply the user's Excel A-class grouping without changing frozen annotations."""
from __future__ import annotations

import argparse
from collections import Counter, defaultdict
import csv
from datetime import datetime
import html
import io
import json
import os
from pathlib import Path, PurePosixPath
import re
import xml.etree.ElementTree as ET
import zipfile
from zoneinfo import ZoneInfo

from pallet.fork_face_anylabeling import digest, encoded, protected_write, safe_path


NS = {'s': 'http://schemas.openxmlformats.org/spreadsheetml/2006/main',
      'r': 'http://schemas.openxmlformats.org/officeDocument/2006/relationships'}
EXPECTED_A_CLASSES = {f'A{i:02d}' for i in range(44)}
DEFAULT_SOURCE = Path('outputs/pallet_target_clustering_v1/analysis/run_006')
DEFAULT_OUTPUT = Path('outputs/pallet_appearance_class_merge_v1')
DEFAULT_SEED = Path('outputs/pallet_cluster_annotation_seed_rectangles_v1/delivery/targets.jsonl')


def read_json(path):
    return json.loads(Path(path).read_text())


def read_jsonl(path):
    return [json.loads(line) for line in Path(path).read_text().splitlines() if line.strip()]


def column_number(name):
    result = 0
    for letter in name: result = result*26+ord(letter)-64
    return result


def column_name(number):
    name = ''
    while number:
        number, remainder = divmod(number-1, 26); name = chr(65+remainder)+name
    return name


def read_workbook(path, sheet_name=None):
    """Read OOXML shared/inline strings and cells; never execute workbook formulas."""
    with zipfile.ZipFile(path) as archive:
        if archive.testzip() is not None: raise ValueError('Workbook ZIP CRC failed')
        workbook = ET.fromstring(archive.read('xl/workbook.xml'))
        sheets = workbook.findall('s:sheets/s:sheet', NS)
        candidates = [s for s in sheets if sheet_name is None or s.get('name') == sheet_name]
        if not candidates: raise ValueError(f'Worksheet not found: {sheet_name}')
        if sheet_name is None and len(candidates) != 1:
            raise ValueError('Multiple worksheets: specify --sheet explicitly')
        sheet = candidates[0]
        relationships = ET.fromstring(archive.read('xl/_rels/workbook.xml.rels'))
        relationship = next(r for r in relationships if r.get('Id') == sheet.get('{'+NS['r']+'}id'))
        if relationship.get('TargetMode') == 'External': raise ValueError('External worksheet unsupported')
        target = relationship.get('Target')
        member = target.lstrip('/') if target.startswith('/') else str(PurePosixPath('xl')/target)
        if '..' in PurePosixPath(member).parts: raise ValueError('Unsafe workbook worksheet path')
        strings = []
        if 'xl/sharedStrings.xml' in archive.namelist():
            strings = [''.join(t.text or '' for t in item.findall('.//s:t', NS))
                       for item in ET.fromstring(archive.read('xl/sharedStrings.xml')).findall('s:si', NS)]
        cells = {}
        for cell in ET.fromstring(archive.read(member)).findall('s:sheetData/s:row/s:c', NS):
            address = cell.get('r')
            if address in cells: raise ValueError('Duplicate workbook cell address')
            if cell.find('s:f', NS) is not None: raise ValueError(f'Formula in grouping rules: {address}')
            value = cell.find('s:v', NS); kind = cell.get('t')
            if kind == 's': decoded = strings[int(value.text)]
            elif kind == 'inlineStr': decoded = ''.join(t.text or '' for t in cell.findall('.//s:t', NS))
            elif value is None or value.text is None: decoded = None
            elif kind in ('str','e'): decoded = value.text
            else:
                decoded = float(value.text)
                if decoded.is_integer(): decoded = int(decoded)
            cells[address] = decoded
        return sheet.get('name'), cells


def parse_grouping(cells, expected=EXPECTED_A_CLASSES):
    headers = []
    for address, value in cells.items():
        if re.fullmatch(r'[A-Z]+1', address) and address != 'A1' and value is not None:
            match = re.fullmatch(r'Class\s+([1-9][0-9]*)', str(value).strip())
            if not match: raise ValueError(f'Invalid class header {address}: {value}')
            headers.append({'column': re.match('[A-Z]+', address)[0],
                            'merged_class_id': int(match[1]), 'merged_class': str(value).strip(),
                            'original_appearance_classes': []})
    headers.sort(key=lambda h: column_number(h['column']))
    if not headers or len({h['merged_class_id'] for h in headers}) != len(headers):
        raise ValueError('Missing or duplicate new class headers')
    mapping = {}; header_columns = {h['column'] for h in headers}
    row_numbers = sorted({int(re.search('[0-9]+$', address)[0]) for address in cells}- {1})
    for number in row_numbers:
        old = cells.get(f'A{number}')
        populated = [(address, value) for address, value in cells.items()
                     if re.search('[0-9]+$', address)[0] == str(number) and address != f'A{number}' and value not in (None, '', 0)]
        if old in (None, ''):
            if populated: raise ValueError(f'Grouping row {number} has no A class')
            continue
        old = str(old).strip()
        if old not in expected or old in mapping: raise ValueError(f'Unknown or duplicate A class: {old}')
        for address, value in populated:
            if re.match('[A-Z]+', address)[0] not in header_columns or value not in (1, '1'):
                raise ValueError(f'Invalid grouping indicator: {address}={value}')
        selected = [h for h in headers if cells.get(f'{h["column"]}{number}') in (1, '1')]
        if len(selected) != 1: raise ValueError(f'{old} must belong to exactly one new class')
        header = selected[0]; header['original_appearance_classes'].append(old)
        mapping[old] = {'original_appearance_class': old, 'merged_class_id': header['merged_class_id'],
                        'merged_class': header['merged_class'], 'worksheet_row': number,
                        'indicator_cell': f'{header["column"]}{number}'}
    if set(mapping) != set(expected): raise ValueError(f'Uncovered A classes: {sorted(set(expected)-set(mapping))}')
    if any(not h['original_appearance_classes'] for h in headers): raise ValueError('Empty new class group')
    for h in headers: h['original_appearance_classes'].sort()
    return headers, mapping


def merge_target(row, mapping):
    original = row['appearance_cluster']
    if isinstance(original, bool) or not isinstance(original, int) or original < -2:
        raise ValueError('Invalid source appearance cluster')
    old = f'A{original:02d}' if original >= 0 else None
    if old is not None and old not in mapping: raise ValueError(f'No mapping for {old}')
    result = dict(row)
    result.update(original_appearance_class=old, merged_class_id=mapping[old]['merged_class_id'] if old else None,
                  merged_class=mapping[old]['merged_class'] if old else 'unassigned',
                  class_merge_status='mapped_from_workbook' if old else (
                      'unassigned_appearance_noise' if original == -1 else 'unassigned_processing_failure'))
    return result


def group_statistics(headers, rows, seed_rows):
    by_class = defaultdict(list); seeds = defaultdict(list)
    for r in rows: by_class[r['merged_class_id']].append(r)
    for r in seed_rows: seeds[r['merged_class_id']].append(r)
    groups = []
    for header in headers:
        selected = by_class[header['merged_class_id']]; seed = seeds[header['merged_class_id']]
        groups.append({k: v for k, v in header.items() if k != 'column'} | {
            'original_class_count': len(header['original_appearance_classes']), 'targets': len(selected),
            'frames': len({r['frame_id'] for r in selected}), 'acquisition_groups': len({r['group_id'] for r in selected}),
            'splits': {s:sum(r['split']==s for r in selected) for s in ('train','val','test')},
            'domains': dict(sorted(Counter(r['domain'] for r in selected).items())),
            'original_class_sizes': dict(sorted(Counter(r['original_appearance_class'] for r in selected).items())),
            'seed_targets': len(seed), 'seed_splits':{s:sum(r['original_split']==s for r in seed) for s in ('train','val','test')}})
    return groups


def write_csv(path, rows):
    if not rows: return protected_write(path,b'')
    fields = list(dict.fromkeys(k for row in rows for k in row))
    stream = io.StringIO(newline=''); writer = csv.DictWriter(stream,fieldnames=fields);writer.writeheader()
    for row in rows:
        writer.writerow({k: json.dumps(v,ensure_ascii=False,separators=(',',':')) if isinstance(v,(dict,list)) else v
                         for k,v in row.items()})
    protected_write(path,stream.getvalue().encode('utf-8-sig'))


def write_workbook(sheets):
    """Small dependency-free OOXML workbook with frozen headers and auto filters."""
    buffer = io.BytesIO()
    def escape(value):return html.escape(str(value),quote=True)
    types = '<?xml version="1.0" encoding="UTF-8"?><Types xmlns="http://schemas.openxmlformats.org/package/2006/content-types"><Default Extension="rels" ContentType="application/vnd.openxmlformats-package.relationships+xml"/><Default Extension="xml" ContentType="application/xml"/><Override PartName="/xl/workbook.xml" ContentType="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet.main+xml"/><Override PartName="/xl/styles.xml" ContentType="application/vnd.openxmlformats-officedocument.spreadsheetml.styles+xml"/>'
    workbook = '<?xml version="1.0" encoding="UTF-8"?><workbook xmlns="'+NS['s']+'" xmlns:r="'+NS['r']+'"><sheets>'
    rels = '<?xml version="1.0" encoding="UTF-8"?><Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships">'
    members = {'_rels/.rels':'<?xml version="1.0" encoding="UTF-8"?><Relationships xmlns="http://schemas.openxmlformats.org/package/2006/relationships"><Relationship Id="rId1" Type="'+NS['r']+'/officeDocument" Target="xl/workbook.xml"/></Relationships>'}
    for index,(name,rows) in enumerate(sheets,1):
        types += f'<Override PartName="/xl/worksheets/sheet{index}.xml" ContentType="application/vnd.openxmlformats-officedocument.spreadsheetml.worksheet+xml"/>'
        workbook += f'<sheet name="{escape(name)}" sheetId="{index}" r:id="rId{index}"/>'
        rels += f'<Relationship Id="rId{index}" Type="{NS["r"]}/worksheet" Target="worksheets/sheet{index}.xml"/>'
        columns = max(map(len,rows)); last=f'{column_name(columns)}{len(rows)}'
        sheet = f'<?xml version="1.0" encoding="UTF-8"?><worksheet xmlns="{NS["s"]}"><dimension ref="A1:{last}"/><sheetViews><sheetView workbookViewId="0"><pane ySplit="1" topLeftCell="A2" activePane="bottomLeft" state="frozen"/></sheetView></sheetViews><cols>'
        sheet += ''.join(f'<col min="{c}" max="{c}" width="{52 if c==2 else 18}" customWidth="1"/>' for c in range(1,columns+1))+'</cols><sheetData>'
        for ri,row in enumerate(rows,1):
            sheet += f'<row r="{ri}">'
            for ci,value in enumerate(row,1):
                if value is None:continue
                address=f'{column_name(ci)}{ri}';style=' s="1"' if ri==1 else ''
                if isinstance(value,(int,float)) and not isinstance(value,bool):sheet+=f'<c r="{address}"{style}><v>{value}</v></c>'
                else:sheet+=f'<c r="{address}" t="inlineStr"{style}><is><t xml:space="preserve">{escape(value)}</t></is></c>'
            sheet+='</row>'
        members[f'xl/worksheets/sheet{index}.xml']=sheet+f'</sheetData><autoFilter ref="A1:{last}"/></worksheet>'
    members['[Content_Types].xml']=types+'</Types>'
    members['xl/workbook.xml']=workbook+'</sheets></workbook>'
    members['xl/_rels/workbook.xml.rels']=rels+f'<Relationship Id="rId{len(sheets)+1}" Type="{NS["r"]}/styles" Target="styles.xml"/></Relationships>'
    members['xl/styles.xml']='''<?xml version="1.0" encoding="UTF-8"?><styleSheet xmlns="http://schemas.openxmlformats.org/spreadsheetml/2006/main"><fonts count="2"><font><sz val="11"/><name val="Calibri"/></font><font><b/><sz val="11"/><color rgb="FFFFFFFF"/><name val="Calibri"/></font></fonts><fills count="3"><fill><patternFill patternType="none"/></fill><fill><patternFill patternType="gray125"/></fill><fill><patternFill patternType="solid"><fgColor rgb="FF173E59"/><bgColor indexed="64"/></patternFill></fill></fills><borders count="1"><border><left/><right/><top/><bottom/><diagonal/></border></borders><cellStyleXfs count="1"><xf numFmtId="0" fontId="0" fillId="0" borderId="0"/></cellStyleXfs><cellXfs count="2"><xf numFmtId="0" fontId="0" fillId="0" borderId="0" xfId="0"/><xf numFmtId="0" fontId="1" fillId="2" borderId="0" xfId="0" applyFont="1" applyFill="1"/></cellXfs><cellStyles count="1"><cellStyle name="Normal" xfId="0" builtinId="0"/></cellStyles></styleSheet>'''
    with zipfile.ZipFile(buffer,'w',compression=zipfile.ZIP_DEFLATED) as z:
        for name,content in sorted(members.items()):
            info=zipfile.ZipInfo(name,(2026,1,1,0,0,0));z.writestr(info,content,compress_type=zipfile.ZIP_DEFLATED)
    return buffer.getvalue()


STYLE='''@font-face{font-family:Pallet;src:url(assets/ReportSans.woff)}*{box-sizing:border-box}body{margin:0;background:#f0f4f7;color:#22384d;font:15px/1.75 Pallet,"Microsoft YaHei",sans-serif}main{max-width:1300px;margin:auto;padding:32px 24px}h1{font-size:30px}h2{margin-top:34px}a{color:#006e86}table{border-collapse:collapse;width:100%;background:white}th,td{border-bottom:1px solid #dbe4eb;padding:10px 13px;text-align:left}th{background:#173e59;color:white;white-space:nowrap}td{font-variant-numeric:tabular-nums}tr:nth-child(even){background:#f8fafb}.table{overflow:auto;border:1px solid #dbe4eb;border-radius:9px}.banner{padding:20px 24px;border-radius:12px;background:#173e59;color:#fff}.group{margin:28px 0;background:white;border:1px solid #dbe4eb;border-radius:12px;padding:20px}.panels{display:grid;grid-template-columns:repeat(auto-fit,minmax(min(300px,100%),1fr));gap:16px}.panel img{width:100%;height:auto;background:#172333}.panel p{margin:5px 0}.links{display:flex;gap:16px;flex-wrap:wrap}code,pre{font-family:Pallet,ui-monospace,monospace;overflow-wrap:anywhere}.muted{color:#687f93}select,input,button{font:inherit;padding:8px 12px;border:1px solid #c6d3df;border-radius:7px;background:white}.filters{display:flex;gap:12px;flex-wrap:wrap;margin:22px 0}.cards{display:grid;grid-template-columns:repeat(auto-fill,minmax(min(270px,100%),1fr));gap:16px}.card{padding:12px;border-radius:10px;background:white;border:1px solid #dbe4eb;cursor:pointer}.card img{height:120px;width:100%;object-fit:contain;background:#182936;image-rendering:pixelated}.card small{display:block;overflow-wrap:anywhere}.pages{display:flex;gap:16px;justify-content:center;align-items:center;margin:24px 0}dialog{border:0;border-radius:12px;width:min(1000px,95vw);max-height:94vh}dialog::backdrop{background:#102b3dbb}dialog canvas{max-width:100%;height:auto;image-rendering:pixelated}pre{white-space:pre-wrap;overflow-wrap:anywhere}@media(max-width:600px){main{padding:20px 14px}h1{font-size:25px}}'''


def relative(path, output):
    return Path(os.path.relpath(Path(path).resolve(),Path(output).resolve())).as_posix()


def render_reports(output, source, dataset, groups, rows, summary):
    table=['| 新类别 | 原 A 类 | 全量目标 | train | val | test | 采集组 | 440种子集目标 |',
           '|---|---|---:|---:|---:|---:|---:|---:|']
    for g in groups:
        table.append(f'| {g["merged_class"]} | {", ".join(g["original_appearance_classes"])} | {g["targets"]} | {g["splits"]["train"]} | {g["splits"]["val"]} | {g["splits"]["test"]} | {g["acquisition_groups"]} | {g["seed_targets"]} |')
    lines=['# A 类合并分组结果','',
        f'按照 `类别统计.xlsx` 的指示，将 **44 个原 A 类合并为 {len(groups)} 个新类别**，沿用表格中的 Class 编号。所有 A00–A43 均恰好归属一组。','',
        f'正式聚类共 {summary["total_targets"]:,} 个目标，其中 **{summary["mapped_targets"]:,} 个目标映射到新组**；原外观离群 {summary["unassigned_targets"]:,} 个保留未分组（unassigned），没有自动分配到相近类别。原 440 帧种子集已同步生成映射，合并后每组数量取决于包含的 A 类数量。','',
        '[浏览合并类别图册](gallery.html) · [下载 Excel 统计](类别合并结果.xlsx) · [A 类映射 CSV](class_mapping.csv) · [分组统计 CSV](groups.csv) · [全量目标 CSV](targets.csv) · [440种子集映射 CSV](seed_targets.csv)','',
        '## 合并规则及数量','',*table,'',
        '## 数据口径与追溯','',
        '分组规则直接来自用户表格中值为1的单元格；矩阵为空或为0表示不属于该组。Class 编号沿用表格（从1开始），不是训练用的默认零起始类别编号。Excel统计包含新类统计、44条A类映射和原始规则矩阵。','',
        '所有逐目标原字段均保留，新增 `original_appearance_class`、`merged_class_id`、`merged_class` 与 `class_merge_status`。原 `appearance_cluster`、隶属分数、几何簇、框坐标、目标身份和数据划分保持不变。原外观隶属分数不代表合并类别的置信度。','',
        '合并属于用户指定的类别分组，不重新运行聚类，也不推断新的材质或物理规格。几何簇和历史候选结构描述作为原始信息保留，未用旧结构名称替换表格中的新组名。','',
        '新类总帧数和采集组数按类内去重，跨类可能共享同一帧或采集组，不能直接相加。train/val/test 列按目标计数。440 种子集清单是属性映射，原有分割标签类别和人工/机器复核状态保持原样。','',
        '来源固定为 `run_006`。输入表格、聚类结果及种子清单的哈希记录在 `run_manifest.json`；原正式标签与冻结元数据核验见 `validation.json`。原表格副本位于 `inputs/类别统计.xlsx`。','',
        '## 按原 A 类查看代表样本','',
        '下方保留各 A 类原代表/边界样本面板，便于检查合并关系。目标图册支持新类别、旧A类、配置域及原划分筛选，点击样本可查看原强度图定位。']
    markdown='\n'.join(lines)+'\n';protected_write(output/'report_zh.md',markdown.encode())
    from markdown_it import MarkdownIt
    body=MarkdownIt('commonmark').enable('table').render(markdown).replace('<table>','<div class="table"><table>').replace('</table>','</table></div>')
    panels=''
    for g in groups:
        panels+=f'<section class="group" id="class-{g["merged_class_id"]}"><h2>{html.escape(g["merged_class"])} · {g["targets"]:,} 个目标</h2><p>原类：{", ".join(g["original_appearance_classes"])} · 种子集 {g["seed_targets"]} 个目标</p><div class="panels">'
        for old in g['original_appearance_classes']:
            panels+=f'<div class="panel"><p><b>{old}</b> · {g["original_class_sizes"][old]} 个目标 · <a href="gallery.html?class={g["merged_class_id"]}&a={old}">查看该原类</a></p><img loading="lazy" src="panels/{old}.jpg" alt="{old} 代表及边界样本"></div>'
        panels+='</div></section>'
    document=f'<!doctype html><html lang="zh-CN"><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>A 类合并：22个新分组</title><style>{STYLE}</style><main>{body}{panels}</main></html>'
    protected_write(output/'report_zh.html',document.encode())
    gallery_rows=[]
    for r in rows:
        gallery_rows.append({k:r[k] for k in ['target_id','target_key','frame_id','box_index','group_id','domain','split',
            'image','bbox','appearance_cluster','geometry_cluster','original_appearance_class','merged_class_id',
            'merged_class','appearance_membership','class_merge_status']})
    template='''<!doctype html><html lang="zh-CN"><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>合并类别目标图册</title><style>__STYLE__</style><main><h1>A 类合并 · 目标图册</h1><p class="links"><a href="report_zh.html">22组报告与代表样本</a><a href="类别合并结果.xlsx">Excel统计</a><a href="targets.csv">逐目标CSV</a></p><div class="banner">44 个 A 类 → 22 个新组 · 6,982 个已分组目标 · 5,362 个原离群目标保留未分组</div><div class="filters"><select id="group" aria-label="新类别"><option value="">全部新类别及未分组</option></select><select id="old" aria-label="原A类别"><option value="">全部原A类别</option></select><select id="split" aria-label="原划分"><option value="">全部原划分</option><option>train</option><option>val</option><option>test</option></select><select id="domain" aria-label="配置域"><option value="">全部配置域</option></select><input id="search" placeholder="搜索目标 / 采集组" aria-label="搜索"><span id="count"></span></div><div id="cards" class="cards"></div><div class="pages"><button id="prev">上一页</button><span id="page"></span><button id="next">下一页</button></div></main><dialog id="detail"><button id="close">关闭</button><h2 id="title"></h2><canvas id="image"></canvas><p class="links"><a id="original" target="_blank">原始强度图</a><a id="old-gallery" target="_blank">原聚类详情</a></p><pre id="metadata"></pre></dialog><script>
const rows=__DATA__,groups=__GROUPS__,media=__MEDIA__,dataset=__DATASET__,oldGallery=__OLD_GALLERY__,$=id=>document.getElementById(id);let page=0,current=null;const pageSize=48;
function option(select,value,label){const o=document.createElement('option');o.value=value;o.textContent=label;$(select).append(o)}
for(const g of groups)option('group',String(g.merged_class_id),`${g.merged_class} · ${g.targets} 个目标`);option('group','unassigned','未分组（原离群）');
for(const a of [...new Set(rows.map(r=>r.original_appearance_class).filter(Boolean))].sort())option('old',a,a);for(const d of [...new Set(rows.map(r=>r.domain))].sort())option('domain',d,d);
function render(){const term=$('search').value.trim().toLowerCase();const result=rows.filter(r=>(!$('group').value||($('group').value==='unassigned'?r.merged_class_id===null:String(r.merged_class_id)===$('group').value))&&(!$('old').value||r.original_appearance_class===$('old').value)&&(!$('split').value||r.split===$('split').value)&&(!$('domain').value||r.domain===$('domain').value)&&(!term||(r.target_id+' '+r.group_id).toLowerCase().includes(term)));const pages=Math.max(1,Math.ceil(result.length/pageSize));page=Math.min(page,pages-1);$('count').textContent=`${result.length.toLocaleString()} 个目标`;$('cards').replaceChildren();for(const r of result.slice(page*pageSize,(page+1)*pageSize)){const c=document.createElement('article');c.className='card';const image=document.createElement('img');image.src=media+'/'+r.target_key+'.jpg';image.alt=r.target_id;image.loading='lazy';const title=document.createElement('b');title.textContent=(r.merged_class_id===null?'未分组':r.merged_class)+' · '+(r.original_appearance_class||'原离群');const meta=document.createElement('small');meta.textContent=r.split+' · '+r.domain;const frame=document.createElement('small');frame.textContent=r.frame_id+' · 框 '+r.box_index;c.append(image,title,meta,frame);c.onclick=()=>open(r);$('cards').append(c)}$('page').textContent=`${page+1} / ${pages}`;$('prev').disabled=page===0;$('next').disabled=page+1===pages}
function open(r){current=r.target_id;$('title').textContent=(r.merged_class_id===null?'未分组':r.merged_class)+' · '+(r.original_appearance_class||'原离群');$('detail').showModal();const image=new Image();image.onload=()=>{if(current!==r.target_id)return;const canvas=$('image');canvas.width=image.width;canvas.height=image.height;const ctx=canvas.getContext('2d');ctx.drawImage(image,0,0);ctx.strokeStyle='#ff4c45';ctx.lineWidth=Math.max(1,image.width/250);ctx.strokeRect(r.bbox.x,r.bbox.y,r.bbox.width,r.bbox.height)};image.src=dataset+'/'+r.image;$('original').href=image.src;$('old-gallery').href=oldGallery+'?target='+encodeURIComponent(r.target_key);$('metadata').textContent=JSON.stringify({新类别:r.merged_class,原类别:r.original_appearance_class,目标:r.target_id,采集组:r.group_id,原划分:r.split,原几何簇:r.geometry_cluster,原外观隶属分数:r.appearance_membership,框:r.bbox},null,2)}
for(const id of ['group','old','split','domain','search'])$(id).addEventListener(id==='search'?'input':'change',()=>{page=0;render()});$('prev').onclick=()=>{page--;render()};$('next').onclick=()=>{page++;render()};$('close').onclick=()=>$('detail').close();const query=new URLSearchParams(location.search);if(query.has('class'))$('group').value=query.get('class');if(query.has('a'))$('old').value=query.get('a');render();
</script></html>'''
    replacements={'__STYLE__':STYLE,'__DATA__':json.dumps(gallery_rows,ensure_ascii=False,separators=(',',':')).replace('<','\\u003c'),
        '__GROUPS__':json.dumps(groups,ensure_ascii=False,separators=(',',':')).replace('<','\\u003c'),
        '__MEDIA__':json.dumps(relative(source.parent.parent/'media',output)),
        '__DATASET__':json.dumps(relative(dataset,output)), '__OLD_GALLERY__':json.dumps(relative(source/'gallery.html',output))}
    gallery=re.sub('|'.join(map(re.escape,replacements)),lambda m:replacements[m[0]],template)
    protected_write(output/'gallery.html',gallery.encode())


def run(args):
    workbook,source,output=args.workbook.resolve(),args.clustering.resolve(),args.output.resolve()
    if output.is_relative_to(source) or source.is_relative_to(output) or workbook.is_relative_to(output):
        raise ValueError('Output must be separate from grouping inputs and frozen clustering results')
    sheet,cells=read_workbook(workbook,args.sheet);headers,mapping=parse_grouping(cells)
    if {g['merged_class_id'] for g in headers}!=set(range(1,len(headers)+1)):
        raise ValueError('Class numbers must be contiguous and start at 1')
    formal_path=source.parent.parent/'run_manifest.json';formal=read_json(formal_path)
    dataset=Path(formal['binding']['dataset'])
    if output.is_relative_to(dataset) or dataset.is_relative_to(output):raise ValueError('Output must be outside formal dataset')
    report_font=Path(__file__).parent/'assets/AppearanceMergeSans.woff'
    inputs={str(p):digest(p) for p in [workbook,formal_path,Path(__file__),report_font.resolve(),*(p for p in source.rglob('*') if p.is_file())]}
    seed_source=args.seed_manifest.resolve() if args.seed_manifest else None
    if seed_source:
        if seed_source.is_relative_to(output):raise ValueError('Seed input must be outside output')
        inputs[str(seed_source)]=digest(seed_source)
    binding={'kind':'user_excel_appearance_class_merge','version':1,'workbook':str(workbook),
             'worksheet':sheet,'clustering':str(source),'seed_manifest':str(seed_source) if seed_source else None,
             'source_hashes':inputs,'rule':'exactly one value=1 per A row','new_class_numbering':'worksheet Class numbers; 1-based',
             'noise_policy':'unassigned; no automatic propagation','labels_modified':False}
    output.mkdir(parents=True,exist_ok=True);protected_write(output/'run_manifest.json',encoded(binding))
    protected_write(output/'inputs/类别统计.xlsx',workbook.read_bytes())
    raw_rows=read_jsonl(source/'targets.jsonl');rows=[merge_target(r,mapping) for r in raw_rows]
    if len({r['target_id'] for r in rows})!=len(rows):raise ValueError('Duplicate source target identity')
    summary_source=read_json(source/'summary.json')
    if len(rows)!=summary_source['validation']['actual_targets']:raise ValueError('Source target count mismatch')
    sizes=Counter(r['original_appearance_class'] for r in rows if r['merged_class_id'] is not None)
    expected={f'A{int(k):02d}':v for k,v in summary_source['clustering']['appearance']['cluster_sizes'].items()}
    if dict(sizes)!=expected or set(expected)!=set(mapping):raise ValueError('Source A sizes or coverage changed')
    by_id={r['target_id']:r for r in rows};seed_rows=[]
    for seed in read_jsonl(seed_source) if seed_source else []:
        target=by_id[seed['target_id']]
        if seed['appearance_class']!=target['original_appearance_class'] or seed['frame_id']!=target['frame_id']:
            raise ValueError('440 seed identities or A classes disagree with run_006')
        seed_rows.append(seed|{k:target[k] for k in ('original_appearance_class','merged_class_id','merged_class','class_merge_status')})
    if len({r['target_id'] for r in seed_rows})!=len(seed_rows) or len({r['frame_id'] for r in seed_rows})!=len(seed_rows):
        raise ValueError('Duplicate seed target or frame identity')
    groups=group_statistics(headers,rows,seed_rows)
    summary={'original_classes':len(mapping),'merged_classes':len(groups),'total_targets':len(rows),
             'mapped_targets':sum(g['targets'] for g in groups),'unassigned_targets':sum(r['merged_class_id'] is None for r in rows),
             'unassigned_splits':{s:sum(r['merged_class_id'] is None and r['split']==s for r in rows) for s in ('train','val','test')},
             'seed_targets':len(seed_rows),'class_mapping_source':'user_workbook','labels_modified':False,
             'reclustered':False,'geometry_clusters_modified':False}
    mappings=[mapping[old] for old in sorted(mapping)]
    for name,data in [('class_mapping.json',{'groups':groups,'a_to_class':mapping}),('groups.json',groups),('summary.json',summary)]:
        protected_write(output/name,encoded(data))
    for name,data in [('targets',rows),('seed_targets',seed_rows)]:
        protected_write(output/(name+'.jsonl'),b''.join(encoded(r) for r in data));write_csv(output/(name+'.csv'),data)
    write_csv(output/'class_mapping.csv',mappings)
    flat=[{k:v for k,v in g.items() if k not in ('splits','seed_splits')}|
          {s:g['splits'][s] for s in ('train','val','test')}|
          {'seed_'+s:g['seed_splits'][s] for s in ('train','val','test')} for g in groups]
    write_csv(output/'groups.csv',flat)
    statistics=[['新类别','包含A类','原A类数量','目标数','train','val','test','帧数(类内去重)','采集组数','种子集目标数','种子train','种子val','种子test']]
    statistics += [[g['merged_class'],', '.join(g['original_appearance_classes']),g['original_class_count'],g['targets'],
        *[g['splits'][s] for s in ('train','val','test')],g['frames'],g['acquisition_groups'],g['seed_targets'],
        *[g['seed_splits'][s] for s in ('train','val','test')]] for g in groups]
    statistics += [['未分组（原离群）','',0,summary['unassigned_targets'],*[summary['unassigned_splits'][s] for s in ('train','val','test')]],
                   ['全部目标','',44,len(rows),*[sum(r['split']==s for r in rows) for s in ('train','val','test')]]]
    matrix=[['原A类',*[h['merged_class'] for h in headers]]]
    matrix += [[old,*[1 if mapping[old]['merged_class_id']==h['merged_class_id'] else None for h in headers]] for old in sorted(mapping)]
    workbook_data=write_workbook([('新类别统计',statistics),('A类映射',
        [['原A类','新类别','新类别编号','原工作表行','标记单元格']]+[[m['original_appearance_class'],m['merged_class'],m['merged_class_id'],m['worksheet_row'],m['indicator_cell']] for m in mappings]),
        ('合并规则矩阵',matrix)])
    protected_write(output/'类别合并结果.xlsx',workbook_data)
    _,generated=read_workbook(output/'类别合并结果.xlsx','合并规则矩阵')
    _,remapping=parse_grouping(generated)
    if {k:v['merged_class_id'] for k,v in remapping.items()}!={k:v['merged_class_id'] for k,v in mapping.items()}:
        raise ValueError('Exported Excel grouping roundtrip failed')
    for old in sorted(mapping):protected_write(output/f'panels/{old}.jpg',(source/f'panels/{old}.jpg').read_bytes())
    protected_write(output/'assets/ReportSans.woff',report_font.read_bytes())
    protected_write(output/'assets/FONT_LICENSE.txt',(source/'FONT_LICENSE.txt').read_bytes())
    render_reports(output,source,dataset,groups,rows,summary)
    for path,sha in inputs.items():
        if digest(path)!=sha:raise ValueError(f'Frozen merge input changed: {path}')
    for name,sha in formal['protected_files'].items():
        if digest(safe_path(dataset,name))!=sha:raise ValueError(f'Frozen formal label/metadata changed: {name}')
    if any(any(merged[k]!=v for k,v in raw.items()) for raw,merged in zip(raw_rows,rows)):
        raise ValueError('An original target field changed')
    validation={'state':'passed','source_a_classes':len(mapping),'merged_classes':len(groups),
        'a_class_membership_exactly_one':True,'original_target_fields_unchanged':True,
        'target_identities_unique':True,'source_targets':len(rows),'mapped_targets':summary['mapped_targets'],
        'unassigned_targets':summary['unassigned_targets'],'seed_targets':len(seed_rows),
        'seed_frames_unique':True,'exported_workbook_mapping_roundtrip':True,'source_files_unchanged':len(inputs),
        'protected_formal_files_unchanged':len(formal['protected_files']),'labels_modified':False}
    protected_write(output/'validation.json',encoded(validation))
    outputs={str(p.relative_to(output)):digest(p) for p in output.rglob('*') if p.is_file() and p.name not in ('checksums.sha256','output_manifest.json')}
    protected_write(output/'output_manifest.json',encoded({'files':outputs,'summary':summary,'validation':validation}))
    protected_write(output/'checksums.sha256',''.join(f'{sha}  {name}\n' for name,sha in sorted(outputs.items())).encode())
    print(json.dumps({'output':str(output),**summary,'validation':validation},ensure_ascii=False))
    return groups


def main(argv=None):
    parser=argparse.ArgumentParser(description=__doc__)
    parser.add_argument('--workbook',type=Path,default=Path('类别统计.xlsx'))
    parser.add_argument('--sheet',help='Sheet name; required only when the input has multiple sheets')
    parser.add_argument('--clustering',type=Path,default=DEFAULT_SOURCE)
    parser.add_argument('--seed-manifest',type=Path,default=DEFAULT_SEED)
    parser.add_argument('--output',type=Path,default=DEFAULT_OUTPUT)
    run(parser.parse_args(argv))


if __name__=='__main__':main()
