"""Immutable class-modal standard and corrected unsaved M2 keypoint drafts."""
from collections import Counter
from contextlib import closing
import html
import json
import math
from pathlib import Path
import shutil
import sqlite3
import sys

import numpy as np
from PIL import Image,ImageDraw,ImageFont
from pycocotools import mask as mu

PROJECT=Path(__file__).resolve().parents[1];sys.path.insert(0,str(PROJECT))
from pallet.cluster_targets import digest
from pallet.mask_class_keypoints import modal_counts,convert,VERSION,SETTINGS
from pallet.mask_model_training_data import read,jsonl,write,sha,require
from pallet.mask_model_worklist import SCHEMA,FLAGS,CHECKPOINT_SHA
from pallet.mask_quota_saved_extension import file_inventory
from pallet.mask_quality_recovery import fingerprint,POSE,GALLERY
from pallet.pose_review_geometry import raster_mask

PARENT=PROJECT/'outputs/pallet_mask_model_requests_v1';ROOT=PROJECT/'outputs/pallet_mask_model_requests_v3'


def prepare():
    require(not ROOT.exists(),'Never overwrite an existing class-keypoint publication')
    catalog_path=PROJECT/'outputs/pallet_mask_continuation_sources_v16/private/source_catalog.jsonl'
    source=jsonl(catalog_path);records=[dict(target_key=r['target_key'],class_id=r['row']['final_class_id'],polygon=r['polygon'],
        current_revision=r['revision'],source_origin=r['origin'],source_event_raw_sha256=r['event_raw_sha256']) for r in source]
    training={r['target_key']:r for r in jsonl(PROJECT/'outputs/pallet_mask_annotation_full_v1/private/training_records.jsonl')}
    isolated=PROJECT/'outputs/pallet_mask_nonconforming_reviews_v1/mask_reviews.sqlite3'
    with closing(sqlite3.connect(isolated.resolve().as_uri()+'?mode=ro',uri=True)) as db:
        for key,raw in db.execute('SELECT target_key,payload FROM reviews'):
            p=json.loads(raw);records.append(dict(target_key=key,class_id=training[key]['class_id'],polygon=p['polygon'],
                current_revision=p['revision'],source_origin='saved_nonconforming_user_quality_approved7',
                source_event_raw_sha256=__import__('hashlib').sha256(raw.encode('utf-8')).hexdigest()))
    old_db=PROJECT/'outputs/pallet_mask_model_reviews_v1/mask_reviews.sqlite3'
    with closing(sqlite3.connect(old_db.resolve().as_uri()+'?mode=ro',uri=True)) as db:
        saved={r[0] for r in db.execute('SELECT target_key FROM reviews')}
    inputs=read(PARENT/'input_manifest.json')
    fps=dict(pose=fingerprint(POSE),gallery=fingerprint(GALLERY),isolated=fingerprint(isolated))
    require(fps==inputs['source_fingerprints'],'New human source changes require a new count snapshot')
    standard=modal_counts(records);require(len(records)==1559,'Count real distinct sources only')
    ROOT.mkdir(mode=0o700);(ROOT/'private').mkdir(mode=0o700)
    write(ROOT/'private/count_sources.jsonl',records,lines=True)
    write(ROOT/'class_keypoint_standard.json',dict(schema='user_authorized_class_modal_keypoint_counts_v1',
        authorization='User explicitly requested each class modal saved point count as its annotation standard and regeneration',
        method='unique mode by distinct latest saved manual target; not per-event voting; tied modes must stop',
        classes=standard,manual_sources=1559,count_sources_sha256=digest(ROOT/'private/count_sources.jsonl'),
        prior_minority_human_polygons_modified=False,applies_to='new regenerated machine drafts only',
        convention_not_confirmed_semantic_ontology=True))
    write(ROOT/'input_manifest.json',dict(parent_directory=str(PARENT.relative_to(PROJECT)),parent_static_inventory=file_inventory(PARENT),
        parent_manifest_sha256=digest(PARENT/'manifest.json'),raw_predictions_sha256=digest(PARENT/'private/predictions.jsonl'),
        parent_model_replay_sha256=digest(PARENT/'private/independent_model_replay.json'),source_fingerprints=fps,
        previous_model_review_store_fingerprint=fingerprint(old_db),excluded_real_saved_parent_keys=sorted(saved),
        configuration_sha256=file_inventory(PROJECT/'configs'),formal_protected=inputs['formal_protected'],
        class_standard_sha256=digest(ROOT/'class_keypoint_standard.json'),checkpoint_sha256=CHECKPOINT_SHA,
        requested_source_scope='only original588 previously-issued unsaved drafts; reserves and protected old work not reopened',
        failed_v2_inventory=file_inventory(PROJECT/'outputs/pallet_mask_model_requests_v2'),
        algorithm=VERSION,conversion_settings=SETTINGS,new_model_training_authorized=False,
        classes_are_user_authorized_postprocessing_constraint_not_network_features=True))
    print(json.dumps({c:v['standard_count'] for c,v in standard.items()}),flush=True)


def generate():
    require(not (ROOT/'private/conversions.jsonl').exists(),'Preserve completed conversions and abstentions')
    standard=read(ROOT/'class_keypoint_standard.json');inputs=read(ROOT/'input_manifest.json')
    require(digest(ROOT/'class_keypoint_standard.json')==inputs['class_standard_sha256'],'Class standard changed')
    predictions={r['target_key']:r for r in jsonl(PARENT/'private/predictions.jsonl')};outcomes=[]
    for old in jsonl(PARENT/'proposals.jsonl'):
        key=old['target_key'];raw=predictions[key];x1,y1,x2,y2=raw['roi']
        k=standard['classes'][str(old['class_id'])]['standard_count']
        if key in inputs['excluded_real_saved_parent_keys']:
            p,info=None,dict(reason='existing_real_human_save_protected')
        else:
            rle=raw['raw_native_prediction_rle'];mask=mu.decode(dict(size=rle['size'],counts=rle['counts'].encode())).astype(bool)
            dense=[[x-x1,y-y1] for x,y in raw['polygon']]
            p,info=convert(mask,dense,k,old['source']['bbox']['height'],origin=(x1,y1),native_size=old['source']['size'])
        outcomes.append(dict(target_key=key,class_id=old['class_id'],standard_keypoint_count=k,polygon=p,conversion=info,
            parent_proposal_sha256=sha(old),raw_native_model_mask_rle_sha256=sha(raw['raw_native_prediction_rle'])))
    write(ROOT/'private/conversions.jsonl',outcomes,lines=True)
    write(ROOT/'private/conversion_summary.json',dict(attempted=len(outcomes),exact_K_candidates=sum(bool(r['polygon']) for r in outcomes),
        candidates_by_class=dict(Counter(r['class_id'] for r in outcomes if r['polygon'])),
        withheld_reasons=dict(Counter(r['conversion']['reason'] for r in outcomes if not r['polygon'])),
        raw_predictions_reused_not_retrained=True,quality_certified=False))
    print(read(ROOT/'private/conversion_summary.json'),flush=True)


def contacts():
    out=ROOT/'private/contact_sheets';out.mkdir()
    font=ImageFont.truetype('/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf',13)
    old={p['target_key']:p for p in jsonl(PARENT/'proposals.jsonl')};outcomes=jsonl(ROOT/'private/conversions.jsonl');pages=[]
    for cls in range(1,23):
        records=[r for r in outcomes if r['class_id']==cls and r['polygon']]
        for start in range(0,len(records),24):
            chunk=records[start:start+24];sheet=Image.new('RGB',(1760,math.ceil(len(chunk)/4)*155),'#111a24');draw=ImageDraw.Draw(sheet)
            for i,r in enumerate(chunk):
                key=r['target_key'];row=old[key]['source'];b=row['bbox'];l=max(0,math.floor(b['x']-8));t=max(0,math.floor(b['y']-12))
                rr=min(row['size'][0],math.ceil(b['x']+b['width']+8));bt=min(row['size'][1],math.ceil(b['y']+b['height']+12))
                image=Image.open(PARENT/'images'/f'{key}.png').convert('RGB').crop((l,t,rr,bt));overlay=image.copy();d=ImageDraw.Draw(overlay)
                points=[(x-l,y-t) for x,y in r['polygon']];d.line(points+[points[0]],fill='#55d5ff',width=1)
                for x,y in points:d.ellipse((x-1,y-1,x+1,y+1),fill='#ffbe43')
                scale=min(216/image.width,112/image.height);size=(max(1,round(image.width*scale)),max(1,round(image.height*scale)))
                x=(i%4)*440;y=(i//4)*155
                for j,img in enumerate((image,overlay)):sheet.paste(img.resize(size,Image.Resampling.NEAREST),(x+j*220,y+24))
                draw.text((x+3,y+2),f'C{cls} #{start+i+1} {key[:10]} K{len(points)} IoU{r["conversion"]["model_mask_fidelity_not_human_accuracy"]["iou"]:.2f}',font=font,fill='white')
                draw.text((x+3,y+136),'RAW                         CLASS-K DRAFT (not GT)',font=font,fill='#c5e1f4')
            name=f'C{cls:02d}_{start//24+1:02d}.png';sheet.save(out/name)
            pages.append(dict(path=str((out/name).relative_to(PROJECT)),sha256=digest(out/name),class_id=cls,target_keys=[r['target_key'] for r in chunk]))
    write(out/'inventory.json',dict(conversions_sha256=digest(ROOT/'private/conversions.jsonl'),pages=pages,
        scope='coarse visual screen only; matching K does not prove semantic corners or accuracy'))
    print('Created',len(pages),'contact sheets',flush=True)


def build():
    from pallet.mask_class_keypoint_review import editor_js,validate_keypoints
    from scripts.build_mask_remaining_direct_requests import build_font
    require(not (ROOT/'manifest.json').exists(),'Never overwrite a sealed class-keypoint publication')
    standard=read(ROOT/'class_keypoint_standard.json');standard_sha=digest(ROOT/'class_keypoint_standard.json')
    inputs=read(ROOT/'input_manifest.json');old={r['target_key']:r for r in jsonl(PARENT/'proposals.jsonl')}
    outcomes=jsonl(ROOT/'private/conversions.jsonl');contact=read(ROOT/'private/contact_sheets/inventory.json');references={}
    # All22 paired RAW/fixed-K contact sheets were actually inspected by the AI.
    # Extra turns on these candidates may be model jaggies, not physical corners.
    withheld_indices={9:{1,2,3},15:{1,2,3},18:{7},20:{12}}
    indices=Counter()
    for sheet in contact['pages']:
        for key in sheet['target_keys']:
            cls=old[key]['class_id'];indices[cls]+=1
            references[key]=dict(contact_sheet=sheet['path'],contact_sheet_sha256=sheet['sha256'],class_candidate_index=indices[cls])
    decisions={};selected=[];withheld=[]
    for r in outcomes:
        key=r['target_key'];cls=r['class_id'];ref=references.get(key);reject=ref and ref['class_candidate_index'] in withheld_indices.get(cls,set())
        decision=dict(publish=bool(r['polygon'] and not reject),AI_coarse_crop_inspection=bool(ref),human_quality_approved=False,
            reason=r['conversion']['reason'] or ('extra_corner_or_face_boundary_not_visually_supported' if reject else 'fixed_K_draft_requires_human_vertex_review'),
            entire_native_image_individually_inspected=False,**(ref or {}));decisions[key]=decision
        if not decision['publish']:
            withheld.append(dict(target_key=key,class_id=cls,reason=decision['reason']));continue
        p=json.loads(json.dumps(old[key]));p['polygon']=r['polygon'];p['vertex_count']=len(r['polygon'])
        p['class_keypoint_constraint']=dict(class_id=cls,standard_count=standard['classes'][str(cls)]['standard_count'],
            standard_sha256=standard_sha,method='unique_class_saved_polygon_vertex_count_mode',
            physical_semantic_slots_confirmed=False,ordered_clockwise=True)
        p['model_provenance'].update(corner_conversion=r['conversion'],class_standard_sha256=standard_sha,
            parent_worklist_manifest_sha256=inputs['parent_manifest_sha256'],parent_proposal_sha256=r['parent_proposal_sha256'],
            visual_inspection=decision,postprocessing='image-mask geometric corners subject to user-authorized class K, not raw dense model polygon')
        validate_keypoints(p,standard,standard_sha);selected.append(p)
    for name in ('images','cases','masks','labelme'):(ROOT/name).mkdir()
    for p in selected:
        key=p['target_key'];row=p['source'];polygon=p['polygon'];k=len(polygon)
        shutil.copyfile(PARENT/'images'/f'{key}.png',ROOT/'images'/f'{key}.png')
        image=Image.open(ROOT/'images'/f'{key}.png').convert('RGB');draw=ImageDraw.Draw(image);points=[tuple(v) for v in polygon]
        draw.line(points+[points[0]],fill='#55d5ff',width=1)
        b=row['bbox'];l=max(0,math.floor(b['x']-12));t=max(0,math.floor(b['y']-16));rr=min(row['size'][0],math.ceil(b['x']+b['width']+12));bt=min(row['size'][1],math.ceil(b['y']+b['height']+16))
        crop=image.crop((l,t,rr,bt)).resize(((rr-l)*3,(bt-t)*3),Image.Resampling.NEAREST);d=ImageDraw.Draw(crop)
        for i,(x,y) in enumerate(polygon,1):
            xx=(x-l)*3;yy=(y-t)*3;d.ellipse((xx-2,yy-2,xx+2,yy+2),fill='#ffbe43');d.text((xx+3,yy-9),str(i),fill='white',stroke_width=1,stroke_fill='black')
        crop.save(ROOT/'cases'/f'{key}.png');Image.fromarray(raster_mask(polygon,*row['size']).astype(np.uint8)*255).save(ROOT/'masks'/f'{key}.png')
        flags=dict(model_preannotation=True,human_confirmed=False,ground_truth_eligible=False)
        shapes=[dict(label='pallet',shape_type='polygon',points=polygon,group_id=row['box_index'],flags=flags)]
        shapes.extend(dict(label=f'kp_{i:02d}',shape_type='point',points=[point],group_id=row['box_index'],flags=flags) for i,point in enumerate(polygon,1))
        write(ROOT/'labelme'/f'{key}.json',dict(version='5.0.0',imagePath=f'../images/{key}.png',imageData=None,imageWidth=row['size'][0],imageHeight=row['size'][1],
            flags=flags,shapes=shapes,keypoints_xy=polygon,num_keypoints=k,class_id=p['class_id'],class_keypoint_constraint=p['class_keypoint_constraint'],
            keypoint_order_semantics='clockwise geometric slots, physical semantic correspondence not confirmed',
            model_checkpoint_sha256=CHECKPOINT_SHA,annotation_scope='single_target_pending_human_review_not_GT'))
    write(ROOT/'private/visual_decisions.json',dict(conversions_sha256=digest(ROOT/'private/conversions.jsonl'),
        contact_inventory_sha256=digest(ROOT/'private/contact_sheets/inventory.json'),actual_sheets_viewed=22,decisions=decisions))
    write(ROOT/'private/withheld.json',withheld);write(ROOT/'proposals.jsonl',selected,lines=True)
    write(ROOT/'requests.jsonl',[dict(target_key=p['target_key'],class_id=p['class_id'],keypoint_count=p['vertex_count'],status='machine_class_keypoints_pending_human_review') for p in selected],lines=True)
    counts=Counter(p['class_id'] for p in selected);before=read(PARENT/'manifest.json')['actual_human_saved_before']
    page='<!doctype html><html lang="zh-CN"><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>按类别标准点数 · M2关键点复核v3</title><style>@font-face{font-family:Review;src:url("report-font.otf")}*{box-sizing:border-box}body{font:16px/1.6 Review,sans-serif;background:#101720;color:#e7edf5;max-width:1100px;margin:auto;padding:12px}article{border:1px solid #536479;padding:12px;margin:18px 0}p,h2,td{overflow-wrap:anywhere}img{display:block;max-width:100%;height:auto}a{color:#80ceff}table{width:100%;border-collapse:collapse}td,th{padding:5px;border:1px solid #536479}@media(max-width:480px){h1{font-size:23px}body{padding:8px}}</style>'
    page+=f'<h1>类别标准点数 · M2关键点预标注v3</h1><p>按用户授权，以1559张已保存人工轮廓的每类众数作为标准点数。本版{len(selected)}张，全部满足所属类别点数；不再展示密集边界。拖动现有关键点，点数固定，不增删点。起点及顺序采用左侧上端起点、图像顺时针，仅为几何编号，不代表物理语义映射已确认。</p><p>本批是原588份尚未保存机器草稿的纠正版，旧预测和人工历史保留。无法保留结构并达到标准点数者暂缓，不用插点、共线点或删凹口凑数。Class9/15本轮转换后仍有额外角点语义疑点，暂不交付；Class19仍无合格新图。</p><p>机器草稿未保存不计完成；点数一致不是边界准确率证明，也不自动获得GT、训练、传播或机器人资格。沿用原访问码，保存进入独立v3复核库；未改原位姿、框、图像或PCD。</p><p>筛选 <select id="request-filter"><option value="">全部</option>'+''.join(f'<option value="{c}">Class {c} · {standard["classes"][str(c)]["standard_count"]}点</option>' for c in range(1,23))+'</select></p><table><tr><th>类别</th><th>标准点数</th><th>众数支持</th><th>本批草稿</th></tr>'
    for cls in range(1,23):
        s=standard['classes'][str(cls)];page+=f'<tr><td>Class {cls}</td><td>{s["standard_count"]}</td><td>{s["modal_support"]}/{s["total_saved"]}</td><td>{counts[cls]}</td></tr>'
    page+='</table>'
    for p in selected:
        key=p['target_key'];page+=f'<article id="target-{key}" data-class="{p["class_id"]}"><h2>Class {p["class_id"]} · 固定{p["vertex_count"]}个关键点 · 待人工复核</h2><img loading="lazy" src="cases/{key}.png" alt="固定类别点数的机器关键点草稿"><p>逐点核对梁缘、脚底、凹口和观察面；点数一致不代表角点位置已正确。</p><a href="images/{key}.png">原图</a> · <a href="labelme/{key}.json">关键点LabelMe</a> · <a href="masks/{key}.png">草稿掩膜</a></article>'
    page+='<script>const f=document.getElementById("request-filter");f.onchange=()=>document.querySelectorAll("article").forEach(a=>a.hidden=!!f.value&&a.dataset.class!==f.value);</script></html>'
    (ROOT/'report_zh.html').write_text(page);build_font(ROOT,page);(ROOT/'model-editor.js').write_text(editor_js())
    assets={name:digest(ROOT/name) for name in ('report_zh.html','report-font.otf','FONT_LICENSE.txt','model-editor.js')}
    for p in selected:
        for folder,ext in [('images','png'),('cases','png'),('masks','png'),('labelme','json')]:
            name=f'{folder}/{p["target_key"]}.{ext}';assets[name]=digest(ROOT/name)
    write(ROOT/'manifest.json',dict(schema='class_modal_keypoint_model_worklist_v2',request_count=len(selected),class_counts=dict(counts),
        actual_human_saved_before=before,target_per_class=100,filter_classes=list(range(1,23)),
        class_standard_sha256=standard_sha,standard_counts={c:v['standard_count'] for c,v in standard['classes'].items()},
        input_manifest_sha256=digest(ROOT/'input_manifest.json'),proposals_sha256=digest(ROOT/'proposals.jsonl'),requests_sha256=digest(ROOT/'requests.jsonl'),
        conversions_sha256=digest(ROOT/'private/conversions.jsonl'),visual_decisions_sha256=digest(ROOT/'private/visual_decisions.json'),
        checkpoint_sha256=CHECKPOINT_SHA,public_assets=assets,source_fingerprints=inputs['source_fingerprints'],source_namespace='model-class-keypoints-v3',
        regenerated_attempts=588,withheld_count=len(withheld),points_padded_or_GT_created=0,models_retrained=False,
        conditional_gap_after_all_drafts_saved={str(c):max(0,100-before[str(c)]-counts[c]) for c in range(1,23)},quality_not_certified=True))
    print(json.dumps(dict(state='built_class_modal_fixed_K_drafts_pending_verification',count=len(selected),classes=dict(counts),withheld=len(withheld))),flush=True)


if __name__=='__main__':
    import argparse
    parser=argparse.ArgumentParser(description=__doc__);parser.add_argument('stage',choices=('prepare','generate','contacts','build'));args=parser.parse_args()
    {'prepare':prepare,'generate':generate,'contacts':contacts,'build':build}[args.stage]()
