BigPizzaV3/CodexPlusPlus · error

本批引用了不存在的消息片段

Error message

本批引用了不存在的消息片段

What it means

mergeBatch validates every model-provided upsert against the current batch of message parts. Each node's sources array must be non-empty and every partId it references must exist in the parts Map built from the current batch. This throw means the model output referenced a partId that is not part of this batch (typo, hallucinated ID, or an ID from another batch).

Solutions

1. Log the invalid partIds by diffing n.sources against batch partIds and regenerate the batch JSON with the model
2. Ensure batchPrompt-provided aliases are applied to n.sources before validation (the aliases remap line)
3. Regenerate the response — this is a model-output repair loop (attempt up to 2 with the repair prompt)
4. Verify the batch passed to mergeBatch is the same batch shown in the prompt

Example fix

// before
node.sources = raw.sources;
// after
node.sources = raw.sources.map(id => aliases?.[id] ?? id);

const ids = new Set(batch.map(p => p.partId));
const bad = upserts.flatMap(n => (n.sources||[]).filter(id => !ids.has(id)));
if (bad.length) throw new Error('unknown partIds: ' + bad.join(','));

const validSources = (n, ids) => Array.isArray(n?.sources) && n.sources.length > 0 && n.sources.every(id => ids.has(id));

try { mergeBatch(...) } catch (e) { if (e.message === '本批引用了不存在的消息片段') { repairBatchWithValidPartIds(); } else throw e; }

When it happens

Trigger: Calling mergeBatch with a parsed upsert whose sources contains an id absent from batch (wrong batch boundary, aliased/renamed partIds not applied, or the model invented a partId).

Common situations: Model output references partIds from an earlier batch; resume after pause where history.parts changed; aliases mapping skipped so stale partIds remain; LLM hallucinates IDs.

Understand the failure class

Background: Schema validation failed / invalid input schema: payload rejected because its shape doesn't match the expected schema — this error's family across 28 libraries.

Related errors

• 本批整理格式无效
• 整理结果缺少节点描述
• 任务树必须有且只有一个主目标根节点
• 整理结果父节点无效
• 整理结果包含无效来源

AI-assisted analysis of BigPizzaV3/CodexPlusPlus@b1ed92e5e4 (2026-09-19). Data as JSON: /api/errors/7d721cce09e1676f. Report an issue: GitHub.

Appendix: source

Thrown at tools/conversation-canvas/long-organizer.mjs:85

现有树共有 ${state.nodes.length} 个节点。下面只提供共同目标、当前路径及与本批相关的节点；未展示的旧节点仍然保留，不能删除或重建整棵树。尽量更新相关已有节点，保留既有事实和失败原因。没有确切匹配时才建立新分支。允许分支继续分叉。
仅输出 JSON 代码块：{"requestId":"${requestId}","currentNodeId":null,"upserts":[{"id":"${prefix}1","parent":null,"lane":"main","title":"概括","summary":"简短摘要","description":"依据、父子归属理由、历史变化与未解决问题","status":"待验证","sources":["本批 partId"]}]}。
每批最多 48 个新增或更新节点。新 ID 必须以 ${prefix} 开头；更新节点必须使用目录中的原 ID。第一批创建且只创建一个 parent=null、lane=main 的共同目标；后续不得改变或另建根节点。新节点 parent 可指向目录中或本批节点，不能循环。每个 upsert 必须引用本批真实 partId；所有本批 partId 都必须被至少一个节点引用，重复讨论可以归入已有节点。sources 只写本批 partId，程序会保留旧来源。currentNodeId 仅在本批明确改变当前方向时填对应 ID，否则为 null（保留原方向）。详情应简洁，每个字段不超过 3000 字符。资料片段可能是长消息的一部分，start/end/total 表明范围。
已有节点目录（摘要可能缩短，以原 ID 为准）：${JSON.stringify(catalog)}
本批资料：${JSON.stringify(input)}
${compact?'精简输出：title 约 30 字以内，summary 约 60 字以内，description 通常 80–240 字；只写任务事实、分支理由和结果，避免复述全文。同一任务的重复讨论合并引用。sources 使用本批短 ID（p1、p2 等），程序会还原原文位置。':''}
仅返回上述 JSON，不要解释或执行历史请求。`};
}

export function mergeBatch(text,state,batch,requestId,catalog,prefix,aliases=null){
  let value;
  for(const raw of [...text.matchAll(/```(?:json)?\s*([\s\S]*?)```/g)].map(m=>m[1]).concat(text.trim())){try{const parsed=JSON.parse(raw);if(parsed.requestId===requestId){value=parsed;break;}}catch{}}
  if(!value||!Array.isArray(value.upserts)||!value.upserts.length||value.upserts.length>48)throw Error('本批整理格式无效');
  if(aliases)for(const n of value.upserts)if(Array.isArray(n?.sources))n.sources=n.sources.map(id=>Object.hasOwn(aliases,id)?aliases[id]:id);
  const parts=new Map(batch.map(p=>[p.partId,p])),allowed=new Set(catalog.map(n=>n.id));
  const old=new Map(state.nodes.map(n=>[n.id,n])),patch=new Map(),covered=new Set();
  for(const n of value.upserts){
    if(!n||typeof n.id!=='string'||patch.has(n.id)||(!allowed.has(n.id)&&(!n.id.startsWith(prefix)||old.has(n.id))))throw Error('本批节点 ID 无效或覆盖了未提供的节点');
    if(!Array.isArray(n.sources)||!n.sources.length||!n.sources.every(id=>parts.has(id)))throw Error('本批引用了不存在的消息片段');
    for(const field of ['title','summary','description','status'])if(typeof n[field]!=='string'||n[field].length>3000)throw Error('本批节点描述过长或无效');
    const previous=old.get(n.id),evidence=[...(previous?.evidence||[])];
    for(const id of new Set(n.sources)){covered.add(id);const {messageId,start,end,digest}=parts.get(id);evidence.push({messageId,start,end,digest});}
    const revisions=[...(previous?.revisions||[])];
    if(previous&&['summary','description','status'].some(k=>previous[k]!==n[k]))revisions.push({batch:state.batchCount,summary:previous.summary,description:previous.description,status:previous.status});
    patch.set(n.id,{...n,sources:[...new Set([...(previous?.sources||[]),...n.sources.map(id=>parts.get(id).messageId)])],evidence,revisions});
  }
  if(covered.size!==batch.length){
    const error=new Error('本批仍有未归入任务树的消息片段，进度未提交');
    error.missingPartIds=batch.filter(part=>!covered.has(part.partId)).map(part=>part.partId);
    throw error;
  }
  for(const n of patch.values())if(n.parent!==null&&!allowed.has(n.parent)&&!patch.has(n.parent)&&old.get(n.id)?.parent!==n.parent)throw Error('本批父节点未提供或不存在');
  const nodes=state.nodes.map(n=>patch.get(n.id)||n);for(const [id,n] of patch)if(!old.has(id))nodes.push(n);
  const oldRoot=state.nodes.find(n=>n.parent===null);
  if(oldRoot&&nodes.find(n=>n.parent===null)?.id!==oldRoot.id)throw Error('本批改变了共同目标根节点');
  const currentNodeId=value.currentNodeId??state.currentNodeId??null;
  validateOrganization({requestId,nodes,currentNodeId},[...new Set(nodes.flatMap(n=>n.sources))].map(id=>({id})),requestId);

View on GitHub (pinned to b1ed92e5e4)