cc-viewer 1.7.14 → 1.7.17

package/src/utils/sessionMerge.js

@@ -18,6 +18,10 @@ export { messageFingerprint };
  *    无 live-port 配置下"提问气泡请求时即显示"依赖这一行为。
  * AppBase 的 SSE 与批量两个 merge 入口、以及单测共用此谓词，防三处逻辑漂移。
  *
+ * 批量路径的降级例外见 shouldDegradeBrokenMerge：当 broken 条目恰好是它那个
+ * session 的唯一载体（slim 只留最后一条 main）时，其 messages 是可信前缀，
+ * 降级合并比整会话空白更可取（会话打 _partialData 标记，ChatView 显示提示条）。
+ *
  * @param {object} entry
  * @param {object} [options]
  * @param {boolean} [options.batch=false] - 批量（强刷/历史加载）路径
@@ -30,6 +34,36 @@ export function isMergeBlockedEntry(entry, options = {}) {
   return false;
 }
 
+/**
+ * 批量路径降级例外：broken 载体能否以"可信前缀"身份合并进 mainAgentSessions。
+ *
+ * 仅在两条安全路径放行（两条都只会走进 mergeMainAgentSessions 的创建分支，
+ * 不会触发 anchor/rebuild 等可能翻倍的对齐逻辑）：
+ *  - prevSessions 为空：会话直接创建，无对齐风险；
+ *  - 双方都有 _seqEpoch 且不同：确定性会话边界，走 epochChanged append 分支
+ *    （三部分守卫防 v1 无 epoch 的条目被误放行）。
+ *
+ * `_staleReorder` 明确不降级（流内稍后有权威副本自建会话）；inProgress 不降级
+ * （批量下其 messages 是裸切片，不是可信前缀）。v1 / 无 epoch 会话的 broken
+ * 载体也刻意不降级（放松守卫会让 broken 条目掉进同会话分支，错位前缀 → 翻倍）。
+ *
+ * @param {object} entry
+ * @param {Array} prevSessions - 当前已合并的 sessions
+ * @returns {boolean} true = 允许降级合并
+ */
+export function shouldDegradeBrokenMerge(entry, prevSessions) {
+  if (!entry || !entry._reconstructBroken) return false;
+  if (entry._staleReorder || entry.inProgress) return false;
+  if (!Array.isArray(prevSessions) || prevSessions.length === 0) return true;
+  const e = entry._seqEpoch;
+  if (!e) return false; // v1 / epoch-less carriers never degrade
+  // Reject when ANY already-merged session shares the entry's epoch (a same-
+  // session broken carrier preceded by a healthy one must not degrade into the
+  // same-session merge branch). Today batches are single-session per epoch, so
+  // this is a defensive closure for future non-monotonic epoch streams.
+  return prevSessions.every((s) => !s._seqEpoch || s._seqEpoch !== e);
+}
+
 /**
  * 增量合并 mainAgent sessions。
  *
@@ -39,6 +73,8 @@ export function isMergeBlockedEntry(entry, options = {}) {
  *  - 命中且 overlapLen <  newLen：push `newMessages[overlapLen..]`，引用稳定。
  *  - 未命中：newLen<curLen → rebuild（/compact summary）；newLen===curLen → 整段 append（Plan Mode 全新片段）；
  *           newLen>curLen → 严格前缀扩展语义（push tail），fp 加固后真正存在重叠的窗口必被 anchor 命中。
+ *  - 降级 partial 基底（lastSession._partialData）且 anchor 未命中：任何长度的全量条目都整段替换
+ *    （partial 前缀有中段缺口、不可信，push tail 会翻倍）。
  *
  * 顶部守卫（isPostClearCheckpoint / userId / transient filter）维持 1.6.245 行为不变。
  *
@@ -68,7 +104,13 @@ export function mergeMainAgentSessions(prevSessions, entry, options = {}) {
   const entryModel = getEffectiveModel(entry);
 
   if (prevSessions.length === 0) {
-    return [{ userId, messages: newMessages, response: newResponse, entryTimestamp, model: entryModel, _seqEpoch: seqEpoch }];
+    return [{
+      userId, messages: newMessages, response: newResponse, entryTimestamp, model: entryModel, _seqEpoch: seqEpoch,
+      // Degraded-broken carrier: truthful prefix with a mid-session hole. Only
+      // the create branches carry it (a same-session merge onto a partial base
+      // is defused by the whole-replace below); ChatView renders the banner.
+      ...(entry._partialData === true && { _partialData: true }),
+    }];
   }
 
   const lastSession = prevSessions[prevSessions.length - 1];
@@ -90,7 +132,10 @@ export function mergeMainAgentSessions(prevSessions, entry, options = {}) {
     for (let i = 0; i < newMessages.length; i++) {
       if (!newMessages[i]._timestamp) newMessages[i]._timestamp = entryTimestamp;
     }
-    return [...prevSessions, { userId, messages: newMessages, response: newResponse, entryTimestamp, model: entryModel, _seqEpoch: seqEpoch }];
+    return [...prevSessions, {
+      userId, messages: newMessages, response: newResponse, entryTimestamp, model: entryModel, _seqEpoch: seqEpoch,
+      ...(entry._partialData === true && { _partialData: true }),
+    }];
   }
 
   if (!options.skipTransientFilter && isNewConversation && newMessages.length <= 4 && prevMsgCount > 4) {
@@ -119,6 +164,17 @@ export function mergeMainAgentSessions(prevSessions, entry, options = {}) {
           lastSession.messages.push(newMessages[i]);
         }
       }
+      // Anchor 对齐成功：新全量条目与 partial 前缀吻合，数据已恢复完整。
+      delete lastSession._partialData;
+    } else if (lastSession._partialData) {
+      // 降级合并的 partial 会话（中段缺口，前缀不可信）：anchor 未命中时任何
+      // 同长/更长的全量条目都是权威真值，整段替换而非前缀扩展——否则会 push 出
+      // 尾部重复（partial 里已含 416..418，新全量 419 会被 append 一遍）。
+      for (let i = 0; i < newLen; i++) {
+        if (!newMessages[i]._timestamp) newMessages[i]._timestamp = entryTimestamp;
+      }
+      lastSession.messages = newMessages;
+      delete lastSession._partialData;
     } else if (newLen < curLen) {
       // /compact summary 等真重建：替换 messages 引用。
       for (let i = 0; i < newLen; i++) {
@@ -175,6 +231,9 @@ export function mergeMainAgentSessions(prevSessions, entry, options = {}) {
     if (entryModel) lastSession.model = entryModel; // latest wins; model-less entries keep the stamp
     return [...prevSessions];
   } else {
-    return [...prevSessions, { userId, messages: newMessages, response: newResponse, entryTimestamp, model: entryModel, _seqEpoch: seqEpoch }];
+    return [...prevSessions, {
+      userId, messages: newMessages, response: newResponse, entryTimestamp, model: entryModel, _seqEpoch: seqEpoch,
+      ...(entry._partialData === true && { _partialData: true }),
+    }];
   }
 }

package/src/utils/toolResultBuilder.js

@@ -188,8 +188,26 @@ export function buildSubAgentResultMap(req, globalIndex) {
   return { ...localState.toolResultMap, ...filled };
 }
 
-export function appendToolResultMap(state, messages, startIndex) {
+export function appendToolResultMap(state, messages, startIndex, extraToolUses) {
   const { toolUseMap, toolResultMap, readContentMap, editSnapshotMap, askAnswerMap, planApprovalMap, _fileState } = state;
+  // V2 live: tool_use blocks merged into an already-scanned message by a later
+  // split row are surfaced on the synthetic entry's _toolUses. Register them so
+  // their tool_result pairs with a real matchedTool (label / input / Read/Edit
+  // state). Side-effect tracking (Write/Edit/Ask/ExitPlanMode) stays in the
+  // message loop below — these blocks were already seen there when first merged.
+  if (Array.isArray(extraToolUses)) {
+    for (const block of extraToolUses) {
+      if (!block || !block.id || block.id in toolUseMap) continue;
+      let parsed = block;
+      if (typeof block.input === 'string') {
+        try {
+          const cleaned = block.input.replace(/^\[object Object\]/, '');
+          parsed = { ...block, input: JSON.parse(cleaned) };
+        } catch {}
+      }
+      toolUseMap[parsed.id] = parsed;
+    }
+  }
   for (let i = startIndex; i < messages.length; i++) {
     const msg = messages[i];
     if (msg.role === 'assistant' && Array.isArray(msg.content)) {

package/src/utils/toolResultCore.js

@@ -165,3 +165,30 @@ export function compactResultPreview(entry, opts = {}) {
   if (!hasImages && !text) return null;
   return { text, images: hasImages ? images : null };
 }
+
+/**
+ * 简化模式是否应在消息流内联 tool_result 图片。
+ * 与 compactResultPreview 的跳过条件（isPermissionDenied / isInputValidationError）
+ * 保持同一语义；另外要求至少一张「可渲染」的图（有 src 的非 oversized），
+ * 纯 oversized 占位不内联（否则消息流里全是虚线占位噪音）。
+ *
+ * @param {object|null} entry - toolResultMap 条目（可能为 undefined：SubAgent 末轮
+ *   tool_use 的 result 未到位 / WebSearch / 历史计数缺口等窗口）
+ * @returns {boolean}
+ */
+export function shouldInlineToolImages(entry) {
+  if (!entry || typeof entry !== 'object') return false;
+  if (entry.isPermissionDenied || entry.isInputValidationError) return false;
+  if (!Array.isArray(entry.images) || entry.images.length === 0) return false;
+  return entry.images.some((img) => img && img.src && !img.oversized);
+}
+
+/**
+ * oversized 图片占位文案（3 处复用：ToolResultView / ChatMessage Popover / 内联块）。
+ * `[image png · 512 KB · too large to preview]`
+ */
+export function formatOversizedImagePlaceholder(img) {
+  const mediaType = (img && img.mediaType ? String(img.mediaType).replace('image/', '') : 'image');
+  const sizeKB = img && typeof img.sizeBytes === 'number' ? Math.round(img.sizeBytes / 1024) : 0;
+  return `[image ${mediaType} · ${sizeKB} KB · too large to preview]`;
+}

package/src/utils/v3Assembler.js

@@ -80,8 +80,20 @@ export function createV3Assembler() {
         _seqEpoch: `v2:${row.sessionId}`,
         _totalMessageCount: messages.length,
         _v3Assembled: true,
+        // The SOURCE row — lets chat classification read the server-side
+        // typeTag directly (same source the request list uses), instead of
+        // re-deriving from a body stripped of system/tools. Known limitation:
+        // AppBase._applyV3Delta dedups by (sessionId, seq, inProgress), so a
+        // classification-correction re-send (Preflight/Plan lookahead) does NOT
+        // rebuild this entry — the row ref stays the FIRST one until a cold
+        // reload. Only lookahead-dependent rows are affected (Teammate/SubAgent
+        // tags are lookahead-free); tracked as backlog.
+        _v3Row: row,
       };
       if (row.teammate) entry.teammate = row.teammate;
+      // Wire agent identity (x-claude-code-agent-id) — display name for native
+      // teammates, persisted server-side (see agent-id.js).
+      if (row.agent) entry.agent = row.agent;
       if (row.proxyUrl) entry.proxyUrl = row.proxyUrl;
       if (isSnapshot) entry._isCheckpoint = true;
       if (row.inProgress) {