"""Runtime-owned source resolution. Files and retrieval evidence are separate inputs.""" import copy,hashlib from KnowledgeService.parsing import extract,ParseError from .models import RuntimeError ACCEPT=('.txt','.md','.docx','.pdf') def meeting_fields(): return [{'name':'meeting_material','type':'file_or_text','required':True,'description':'会议原始材料','accept':list(ACCEPT),'max_files':5},{'name':'meeting_date','type':'date','required':False,'description':'会议日期;留空则相对期限待确认'}] def prepare_material(workspace,user,wid,request,options=None,*,allow_inline=False): options=copy.deepcopy(options or {});ids=options.get('attachment_ids',[]) allowed={a['attachment_id']:a for a in workspace.attachments(user,wid)['items']} if not ids and not options.get('source_text','').strip() and options.get('auto_attach',True): seen=set();ids=[] for a in allowed.values(): if a['filename'].lower().endswith(ACCEPT) and a.get('parse_status')=='ready' and a['sha256'] not in seen:ids.append(a['attachment_id']);seen.add(a['sha256']) if allow_inline and not ids and len(request)>80 and '\n' in request:options['source_text']=request if len(ids)>5:raise RuntimeError(422,'MEETING_FILE_LIMIT','会议材料最多5个,请在启动卡片中选择文件。') unique=[];seen=set() for aid in ids: if aid not in allowed:raise RuntimeError(404,'RESOURCE_NOT_FOUND','会议附件不属于当前事项。') info=allowed[aid] if not info['filename'].lower().endswith(ACCEPT):raise RuntimeError(422,'MEETING_FILE_TYPE','请选择 TXT、Markdown、DOCX 或可提取文字的 PDF。') if info.get('parse_status')!='ready':raise RuntimeError(422,'MEETING_FILE_NOT_READY','会议文件未就绪或解析失败,请重新上传。') if info['sha256'] not in seen:unique.append(aid);seen.add(info['sha256']) if not unique and not options.get('source_text','').strip():raise RuntimeError(422,'WORKFLOW_INPUT_REQUIRED','请先在启动卡片中上传或选择会议文件,或粘贴会议原文;处理要求不能代替会议记录。',{'missing':['meeting_material'],'input_schema':meeting_fields()}) options['attachment_ids']=unique return {'meeting_options':options,'material':read_material(workspace,user,wid,request,options)} def read_material(workspace, user, wid, request, options): chunks = []; files = [] allowed = {a['attachment_id']: a for a in workspace.attachments(user, wid)['items']} for aid in options.get('attachment_ids', []): if aid not in allowed: raise RuntimeError(404, 'RESOURCE_NOT_FOUND', '附件不属于当前事项。') info = allowed[aid] if not info['filename'].lower().endswith(('.txt', '.md', '.docx', '.pdf')): raise RuntimeError(422, 'MEETING_FILE_TYPE', '会议节点支持 TXT、Markdown、Word DOCX 和可提取文字的 PDF。') data = workspace.attachment_content(user, aid) try: # Parsing chunks overlap; reconstruct the original text without duplicated passages. parts = extract(info['filename'], data) except ParseError as exc: raise RuntimeError(422, 'MEETING_PARSE_FAILED', '会议文件解析失败。') from exc grouped = {} for locator, text, terms in parts: group = locator.split(' · 字符 ')[0] offset = int(locator.split(' · 字符 ')[1].split('–')[0]) - 1 old = grouped.get(group, '') grouped[group] = old + text[max(0, len(old)-offset):] text = '\n'.join(grouped.values()) chunks.append(text); files.append({'attachment_id': aid, 'filename': info['filename'], 'sha256': hashlib.sha256(data).hexdigest(), 'characters': len(text)}) text = '\n\n'.join(chunks) if chunks else options.get('source_text','') if not text.strip() or len(text) > 160000: raise RuntimeError(422, 'MEETING_TEXT_LIMIT', '会议文字须为 1–160000 字。') return {'text': text, 'files': files, 'instructions': request, 'source_sha256': hashlib.sha256(text.encode()).hexdigest()}