"""Offline, immutable, sourced interview cards and labels for the demo tenant.""" import hashlib import json from datetime import datetime, timezone from pathlib import Path from sqlalchemy import select from .models import KnowledgeBase, Grant, LibraryConfig, DocumentInfo, Document, Chunk, DocumentSource, KnowledgeError from .parsing import extract BASE_ID='kb-interviews' def corpus(): value=json.loads(Path(__file__).with_name('interview-corpus.json').read_text(encoding='utf-8')) if len({x['id'] for x in value['records']})!=len(value['records']):raise ValueError('duplicate interview IDs') return value def seed_interviews(service): data=corpus() with service.Session.begin() as db: if db.get(KnowledgeBase,('demo-tenant',BASE_ID)) is None: db.add(KnowledgeBase(tenant='demo-tenant',id=BASE_ID,name='科技企业秋招面试经验库',visibility='shared')) db.add(LibraryConfig(tenant='demo-tenant',base_id=BASE_ID,description=data['notice'],category='秋招与求职',status='enabled',revision=1)) for uid in ('u-admin','u-lin','u-zhou'):db.add(Grant(tenant='demo-tenant',base_id=BASE_ID,user_id=uid)) with service.Session() as db: existing={d.document_id:(d.sha256,db.get(DocumentInfo,d.id)) for d in db.scalars(select(Document).where(Document.tenant=='demo-tenant',Document.base_id==BASE_ID,Document.version=='1'))} prepared=[] for r in data['records']: text='\n'.join(['# '+r['title'],'公司:'+r['company'],'岗位:'+r['role'],'时期:'+r['season'], '主题:'+'、'.join(r['topics']),'内容核验:'+r['verification'],'导读:'+r['summary'],'原文:'+r['source_url'], '注意:个人历史经验,不是当前官方招聘政策;自动标注需人工复核。']) known=existing.get(r['id']) tags=list(dict.fromkeys([r['company'],r['role'],r['verification'],*r['topics']])) if known and known[0]==hashlib.sha256(text.encode()).hexdigest() and known[1] is not None:continue provenance={k:r[k] for k in ('source_url','verification','annotation_status','season','company','role')} provenance.update(origin_kind='public_interview_reference',origin_label='公开面经原创导读/索引', author='原文作者(见来源) / Work OS 导读',license='原文保留原作者权利;不收录全文', retrieved_at=data['retrieved_at'],version='1',currency='historical_reference', notice=data['notice'],topics=r['topics'],source_content_sha256=r.get('source_content_sha256')) prepared.append((r,text.encode(),tags,provenance)) # Commit a validated bundle once, rather than two disk syncs for every card. with service.lock,service.Session.begin() as db: for r,content,tags,provenance in prepared: sha=hashlib.sha256(content).hexdigest() key='doc-'+hashlib.sha256(json.dumps(['demo-tenant',r['id'],'1']).encode()).hexdigest() doc=db.get(Document,key) if doc is not None and doc.sha256!=sha:raise KnowledgeError(409,'DOCUMENT_VERSION_CONFLICT','面试资料同一版本不可替换。') chunks=extract(r['id']+'.txt',content) if doc is None: doc=Document(id=key,tenant='demo-tenant',document_id=r['id'],version='1',base_id=BASE_ID,title=r['title'], owner=None,workspace=None,source_type='interview_reference',sha256=sha,status='ready', valid_from=None,valid_to=None,error_code=None,created_at=datetime.now(timezone.utc).isoformat()) db.add(doc);db.flush() for ordinal,(locator,text,terms) in enumerate(chunks,1): db.add(Chunk(id='cite-'+hashlib.sha256((key+':'+str(ordinal)).encode()).hexdigest(), document_key=key,ordinal=ordinal,locator=locator,text=text,terms=terms)) service._info(db,doc,r['id']+'.txt',content,chunks) info=db.get(DocumentInfo,key);info.tags=tags;info.category='秋招面试经验' source=db.get(DocumentSource,key) body={**provenance,'content_sha256':sha,'sha256':sha} if source is None:db.add(DocumentSource(document_key=key,body=body)) else:source.body=body def stats(): data=corpus() return {k:data[k] for k in ('total','article_read_count','index_only_count','retrieved_at','notice')}