@feng0531/dsh-code-review 0.24.1 → 0.25.1

package/CHANGELOG.md

@@ -1,5 +1,16 @@
 # 变更记录
 
+## 0.25.1
+
+- 修复大仓库小提交历史快照Command output limit exceeded：用diff-tree raw NUL元数据取代两端全树ls-tree，不扩大整树上限。
+- 上下文/规则局部字面非递归对象查询，自动依赖512路径限，禁目录/通配符扩张；无工作树读取或fetch。
+- 真实11000文件超1MiB树的4 Shader提交回归、自动上下文规则及安全边界通过；全量266通过。
+
+## 0.25.0
+
+- 跨文件交互批次也制定风险假设，至少两文件精确editRefs，同一主调用signal及全局预算，不冒充证据。
+- 实际生成输入/单文件假设拒绝/调用耗尽回归，业务组集成更新额外规划调用。
+
 ## 0.24.1
 
 - 历史上下文关系分别对基线/目标文本推断，标记contextSide/contextVersion；缺少oldText不回退目标正文。

package/README.md

@@ -4,7 +4,9 @@
 
 ## 当前状态
 
-当前代码版本为 0.24.1
+当前代码版本为 0.25.1
+
+历史捕获用diff-tree精确端点变更元数据，不前置完整枚举文件树；显式上下文/规则和直接导入依赖按字面非递归路径查询。自动依赖最多512个待查路径，仍保留200变更/20上下文/4MiB快照限制，大树小提交不因无关文件元数据超限。
 
 历史自动上下文也导航旧侧import/include，仅未修改且两侧普通blob；上下文oldText/oldRevision和目标revision明确，检索context-old单独身份，仍共用20文件总量及4MiB快照。旧侧独有/删除上下文暂不捕获，调用者覆盖仍不保证。
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@feng0531/dsh-code-review",
-  "version": "0.24.1",
+  "version": "0.25.1",
   "private": false,
   "license": "SEE LICENSE IN LICENSE",
   "publishConfig": { "access": "public", "registry": "https://registry.npmjs.org/" },

package/src/context-candidates.mjs

@@ -1,7 +1,7 @@
 import path from 'node:path';
 import {relativePath} from './content.mjs';
 // Navigation only over caller-supplied tracked regular paths; no reads or authority.
-export function contextCandidates(files, trackedPaths, {maxCandidates=20}={}) {
+export function contextCandidates(files, trackedPaths, {maxCandidates=20,onProbe}={}) {
  if(!Array.isArray(files)||files.length>200||!Array.isArray(trackedPaths)||trackedPaths.length>10000||!Number.isSafeInteger(maxCandidates)||maxCandidates<1||maxCandidates>20)throw new Error('Invalid context candidate limit');
  const tracked=new Set();for(const p of trackedPaths){relativePath(p);if(tracked.has(p))throw new Error('Duplicate tracked path');tracked.add(p);}
  const changed=new Set(files.flatMap(f=>[f.path,...(f.oldPath?[f.oldPath]:[])]));
@@ -17,6 +17,7 @@ export function contextCandidates(files, trackedPaths, {maxCandidates=20}={}) {
    if(!match)continue;const spec=cpp?match[1]:match[2];if(!cpp&&!/^\.\.?\//.test(spec))continue;
    const target=path.posix.normalize(path.posix.join(path.posix.dirname(f.path),spec));try{relativePath(target);}catch{continue;}
    const candidates=js&&!path.posix.extname(target)?[target,...['.mjs','.cjs','.js','.jsx','.ts','.tsx'].flatMap(e=>[target+e,target+'/index'+e])]:[target];
+   onProbe?.(candidates);
    const matches=candidates.filter(p=>tracked.has(p));if(matches.length===1)add(matches[0],f.path,cpp?'literal-relative-include':'literal-relative-import');
   }
  }

package/src/git-history.mjs

@@ -15,20 +15,28 @@ export async function captureGitHistory(cwd, options={}) {
  let base,target;
  if(commit!==undefined){if(baseRevision!==undefined||targetRevision!==undefined)throw new Error('Commit and range are mutually exclusive');target=await resolve(commit);const object=utf8(await git(root,['cat-file','commit',target]));const header=object.split('\n\n',1)[0];const parents=header.split('\n').filter(line=>line.startsWith('parent ')).map(line=>line.slice(7));if(parents.some(oid=>!/^(?:[a-f0-9]{40}|[a-f0-9]{64})$/.test(oid)))throw new Error('Invalid commit parent');if(parents.length>1)throw new Error('Merge commit requires explicit revision range');base=parents[0]??null;}
  else {base=await resolve(baseRevision);target=await resolve(targetRevision);}
- const tree=async oid=>{const map=new Map();if(!oid)return map;const records=utf8(await git(root,['ls-tree','-rz','--full-tree',oid])).split('\0').filter(Boolean);if(records.length>10000)throw new Error('Historical tree entry limit exceeded');for(const r of records){const m=/^(\d{6}) (blob|commit) ([a-f0-9]+)\t([\s\S]+)$/.exec(r);if(!m)throw new Error('Invalid historical tree');relativePath(m[4]);map.set(m[4],{mode:m[1],type:m[2],oid:m[3]});}return map;};
- const left=await tree(base),right=await tree(target),names=[...new Set([...left.keys(),...right.keys()])].filter(p=>JSON.stringify(left.get(p))!==JSON.stringify(right.get(p))).sort();
+ // Metadata scales with changed paths, not every tracked file in either tree.
+ const raw=utf8(await git(root,['diff-tree','--no-abbrev','--raw','-z','-r','--no-renames','--no-ext-diff','--no-textconv','--no-commit-id',...(base?[base,target]:['--root',target]),'--'])).split('\0');
+ if(raw.at(-1)==='')raw.pop();
+ const left=new Map(),right=new Map(),names=[];
+ if(raw.length%2)throw new Error('Invalid historical diff metadata');
+ for(let i=0;i<raw.length;i+=2){const m=/^:(\d{6}) (\d{6}) ([a-f0-9]{40}|[a-f0-9]{64}) ([a-f0-9]{40}|[a-f0-9]{64}) ([AMDT])$/.exec(raw[i]);if(!m)throw new Error('Invalid historical diff metadata');const name=relativePath(raw[i+1]);if(names.includes(name))throw new Error('Duplicate historical change');names.push(name);for(const [map,mode,oid] of [[left,m[1],m[3]],[right,m[2],m[4]]])if(mode!=='000000')map.set(name,{mode,type:mode==='160000'?'commit':'blob',oid});}
+ names.sort();
+ // Literal, nonrecursive lookups cannot enumerate a supplied directory or wildcard.
+ const lookup=async (oid,map,requested)=>{if(!oid)return;const pending=[...new Set(requested)].filter(p=>!map.has(p));for(let i=0;i<pending.length;i+=32){const paths=pending.slice(i,i+32);paths.forEach(relativePath);const records=utf8(await git(root,['--literal-pathspecs','ls-tree','-z','--full-tree',oid,'--',...paths])).split('\0').filter(Boolean);for(const r of records){const m=/^(\d{6}) (blob|commit|tree) ([a-f0-9]+)\t([\s\S]+)$/.exec(r);if(!m||!paths.includes(m[4]))throw new Error('Invalid historical path lookup');if(m[2]!=='tree')map.set(m[4],{mode:m[1],type:m[2],oid:m[3]});}}};
  if(names.length>maxFiles)throw new Error('Historical change count limit exceeded');if(selectedPaths?.some(p=>!names.includes(p)))throw new Error('Selected path is not a historical change');
  const files=[];for(const name of names){signal?.throwIfAborted();const l=left.get(name),r=right.get(name),item={id:hash(name),path:name,rawStatus:!l?'A':!r?'D':'M',eligibility:'reviewable',properties:[]};files.push(item);if(selectedPaths&&!selectedPaths.includes(name)){item.eligibility='excluded';item.reason='Not explicitly selected';continue;}try{for(const e of [l,r].filter(Boolean))if(e.type!=='blob'||!['100644','100755'].includes(e.mode))throw new Error('Historical link or submodule excluded');const read=async e=>decode(e?await git(root,['cat-file','blob',e.oid],maxFileBytes):Buffer.alloc(0));item.left=await read(l);item.right=await read(r);item.rightExists=!!r;if(l&&r&&item.left.hash===item.right.hash){item.eligibility='excluded';item.reason='Mode-only historical change not analyzed';}}catch(error){signal?.throwIfAborted();item.eligibility='blocked';item.reason=error.message;delete item.left;delete item.right;}if(Buffer.byteLength(JSON.stringify(files))>maxSnapshotBytes)throw new Error('Historical snapshot limit exceeded');}
  const rulePaths=options.rulePaths ?? [];
  const contextPaths=options.contextPaths ?? [], maxContextFiles=options.maxContextFiles ?? 20;
  if(!Array.isArray(contextPaths)||!Number.isSafeInteger(maxContextFiles)||maxContextFiles<1||maxContextFiles>20||contextPaths.length>maxContextFiles)throw new Error('Historical context file limit exceeded');
  if(options.autoContext!==undefined&&typeof options.autoContext!=='boolean')throw new Error('Invalid autoContext option');
- if(!Array.isArray(rulePaths)||rulePaths.some(p=>contextPaths.includes(p)))throw new Error('Rule paths must be separate from context paths');
+ if(!Array.isArray(rulePaths)||rulePaths.length>4||rulePaths.some(p=>contextPaths.includes(p)))throw new Error('Rule paths must be separate from context paths');
+ await lookup(target,right,[...contextPaths,...rulePaths]);await lookup(base,left,contextPaths);
  const context=[], regular=e=>e?.type==='blob'&&['100644','100755'].includes(e.mode);
  const captureContext=async name=>{relativePath(name);const changed=files.find(f=>f.path===name);if(changed){if(changed.eligibility!=='reviewable'||changed.rightExists===false)throw new Error('Historical context is excluded blocked or deleted');return {path:name,...changed.right};}const entry=right.get(name);if(!regular(entry))throw new Error('Historical context not a target regular blob');const content=decode(await git(root,['cat-file','blob',entry.oid],maxFileBytes));const previous=left.get(name);return {path:name,...content,revision:target,...(regular(previous)?{oldText:decode(await git(root,['cat-file','blob',previous.oid],maxFileBytes)).text,oldRevision:base}:{})};};
  for(const name of [...new Set(contextPaths)].sort())context.push(await captureContext(name));
  const rules=await captureProjectRules(root,rulePaths,{signal,files,read:async (_root,name,{maxFileBytes:limit})=>{const entry=right.get(name);if(!regular(entry))throw new Error('Historical rule not a target regular blob');return decode(await git(root,['cat-file','blob',entry.oid],limit));}});
  let autoContext;
- if(options.autoContext===true){const plan=contextCandidates(files,[...right].filter(([,e])=>regular(e)).map(([p])=>p));const oldPlan=contextCandidates(files.map(f=>f.eligibility==='reviewable'?{...f,right:f.left,rightExists:true}:f),[...left].filter(([p,e])=>regular(e)&&regular(right.get(p))&&!names.includes(p)).map(([p])=>p));const combined=new Map(plan.candidates.map(c=>[c.path,{...c,sourceSides:['new']}]));for(const c of oldPlan.candidates){if(combined.has(c.path))combined.get(c.path).sourceSides.push('old');else combined.set(c.path,{...c,sourceSides:['old']});}plan.candidates=[...combined.values()].sort((a,b)=>a.path<b.path?-1:a.path>b.path?1:0);plan.truncated ||= oldPlan.truncated;autoContext={...plan,notice:'Direct baseline/target-tree import/include navigation only; no caller coverage, binding or working-tree reads.',target,base,candidates:plan.candidates.filter(c=>!contextPaths.includes(c.path)&&!rulePaths.includes(c.path)),capturedPaths:[]};for(const c of autoContext.candidates){signal?.throwIfAborted();if(context.length>=maxContextFiles){c.status='blocked';c.reason='Context file limit exceeded';continue;}try{context.push(await captureContext(c.path));c.status='captured';autoContext.capturedPaths.push(c.path);}catch(error){signal?.throwIfAborted();c.status='blocked';c.reason=error.message;}}}
+ if(options.autoContext===true){const probes=new Set();const oldFiles=files.map(f=>f.eligibility==='reviewable'?{...f,right:f.left,rightExists:true}:f);const onProbe=paths=>{for(const p of paths)probes.add(p);if(probes.size>512)throw new Error('Historical context lookup limit exceeded');};contextCandidates(files,[],{onProbe});contextCandidates(oldFiles,[],{onProbe});await lookup(target,right,[...probes]);await lookup(base,left,[...probes]);const plan=contextCandidates(files,[...right].filter(([,e])=>regular(e)).map(([p])=>p));const oldPlan=contextCandidates(files.map(f=>f.eligibility==='reviewable'?{...f,right:f.left,rightExists:true}:f),[...left].filter(([p,e])=>regular(e)&&regular(right.get(p))&&!names.includes(p)).map(([p])=>p));const combined=new Map(plan.candidates.map(c=>[c.path,{...c,sourceSides:['new']}]));for(const c of oldPlan.candidates){if(combined.has(c.path))combined.get(c.path).sourceSides.push('old');else combined.set(c.path,{...c,sourceSides:['old']});}plan.candidates=[...combined.values()].sort((a,b)=>a.path<b.path?-1:a.path>b.path?1:0);plan.truncated ||= oldPlan.truncated;autoContext={...plan,notice:'Direct baseline/target-tree import/include navigation only; no caller coverage, binding or working-tree reads.',target,base,candidates:plan.candidates.filter(c=>!contextPaths.includes(c.path)&&!rulePaths.includes(c.path)),capturedPaths:[]};for(const c of autoContext.candidates){signal?.throwIfAborted();if(context.length>=maxContextFiles){c.status='blocked';c.reason='Context file limit exceeded';continue;}try{context.push(await captureContext(c.path));c.status='captured';autoContext.capturedPaths.push(c.path);}catch(error){signal?.throwIfAborted();c.status='blocked';c.reason=error.message;}}}
  const snapshot={context,...(rules.length?{rules}:{}),...(autoContext?{autoContext}:{}),schemaVersion:1,vcs:'git',root,baseline:base,history:{mode:commit!==undefined?'commit':'range',base,target,semantics:'exact-tree-endpoints; renames represented as add/delete'},files};const serialized=JSON.stringify(snapshot);if(Buffer.byteLength(serialized)>maxSnapshotBytes)throw new Error('Historical snapshot limit exceeded');return {...snapshot,id:hash(serialized)};
 }

package/src/review.mjs

@@ -157,12 +157,14 @@ export async function reviewSnapshot(snapshot, model, options = {}) {
         for (const batch of batches) {
         enclosingRequest.signal.throwIfAborted();
         let batchPayload = batch.payload;
-        if (options.enableRiskPlanning === true && !batch.synthesis && !batch.interaction) {
+        if (options.enableRiskPlanning === true && !batch.synthesis) {
           state.riskPlans ??= [];
-          const entry = { status: 'pending', windowIds: batch.windowIds, risks: [] };
+          const entry = { status: 'pending', windowIds: batch.windowIds, ...(batch.interaction ? {groupId:batch.interaction.id,sourceMode:'file-interaction'} : {}), risks: [] };
           state.riskPlans.push(entry);
           const response = await executor({ ...enclosingRequest, input: batch.payload, instructions: RISK_PLAN_INSTRUCTIONS });
-          const risks = validateRiskPlan(response, [{ fileId: file.id, edits: changes.edits ?? [] }], scope?.catalog() ?? []);
+          const riskFiles = batch.interaction ? JSON.parse(batch.payload).files : [{ fileId: file.id, edits: changes.edits ?? [] }];
+          const risks = validateRiskPlan(response, riskFiles, scope?.catalog() ?? []);
+          if (batch.interaction && risks.some(r=>new Set(r.editRefs.map(ref=>ref.fileId)).size<2)) throw new Error('Interaction risk requires edits across at least two files');
           entry.status = 'completed'; entry.risks = risks;
           batchPayload = JSON.stringify({ ...JSON.parse(batch.payload), riskPlan: risks, riskPlanNotice: '未可信假设而非发现或证据；须验证及寻找反证，不增加读取权限。' });
           if (!initialInputBudget({ instructions: REVIEW_INSTRUCTIONS, input: batchPayload }, scope, maxInputBytes).fits) throw new Error('Risk-enriched review input exceeds budget');