@dahawa/hawa-cli-analysis 1.0.4

package/api-anthropic.js

@@ -0,0 +1,279 @@
+import LogManager from './logger-manager.js';
+
+const logger = LogManager.getSystemLogger();
+
+export function mergeAnthropic(all){
+   logger.debug("mergeAnthropic input: "+ all);
+   return mergeAnthropicChunks(all.split("\n"));
+}
+/**
+ * 将 Anthropic 流式 chunk（原始字符串数组）合并为完整响应消息体。
+ * 兼容 chunk 形态：
+ *  - "data: {...}" 或 "{...}"
+ *  - 事件类型：message_start, content_block_start, content_block_delta,
+ *    content_block_stop, message_delta, message_stop, ping
+ *  - 文本增量：delta.type === "text_delta", delta.text
+ *  - 工具增量：delta.partial_json / delta.input_json / delta.input_json_delta
+ *
+ * @param {string[]} rawChunks - 原始 chunk 文本数组，每个元素是一行 SSE data。
+ * @returns {{
+ *   message: {
+ *     id?: string,
+ *     type: "message",
+ *     role: "assistant",
+ *     model?: string,
+ *     stop_reason?: string | null,
+ *     stop_sequence?: string | null,
+ *     usage: { input_tokens: number, output_tokens: number },
+ *     content: Array<
+ *       | { type: "text", text: string }
+ *       | { type: "tool_use", id: string, name: string, input: any }
+ *     >
+ *   },
+ *   debug: { parsedCount: number, ignoredCount: number, errors: Array<{idx:number,error:string,raw:string}> }
+ * }}
+ */
+export function mergeAnthropicChunks(rawChunks) {
+  const message = {
+    type: "message",
+    role: "assistant",
+    content: [],
+    usage: { input_tokens: 0, output_tokens: 0 },
+    stop_reason: null,
+    stop_sequence: null,
+  };
+
+  // content block 临时态：按 index 管理；为工具输入保留 JSON 片段缓冲
+  const blockMap = new Map(); // index -> { ref, toolJsonBuffer }
+  const debug = { parsedCount: 0, ignoredCount: 0, errors: [] };
+
+  const parseMaybeJson = (line) => {
+    // 去掉 "data:" 前缀及空白/末尾 [DONE]
+    const trimmed = line.trim();
+    if (!trimmed || trimmed === "data: [DONE]") return null;
+    let jsonText = trimmed.startsWith("data:")
+      ? trimmed.slice(5).trim()
+      : trimmed;
+
+    if (!jsonText || jsonText === "[DONE]") return null;
+    try {
+      return JSON.parse(jsonText);
+    } catch (e) {
+      throw new Error(`JSON parse failed: ${e.message}`);
+    }
+  };
+
+  const ensureBlock = (index, starter) => {
+    if (blockMap.has(index)) return blockMap.get(index);
+
+    let ref;
+    let toolJsonBuffer = "";
+
+    if (starter?.type === "tool_use" || starter?.content_block?.type === "tool_use") {
+      // 工具块
+      const b = starter.type === "tool_use" ? starter : starter.content_block;
+      ref = { type: "tool_use", id: b.id, name: b.name, input: {} };
+      message.content.push(ref);
+      const slot = { ref, toolJsonBuffer };
+      blockMap.set(index, slot);
+      return slot;
+    } else {
+      // 文本块
+      ref = { type: "text", text: "" };
+      message.content.push(ref);
+      const slot = { ref, toolJsonBuffer };
+      blockMap.set(index, slot);
+      return slot;
+    }
+  };
+
+  const feedToolJson = (slot, piece) => {
+    // Anthropic 常见：delta.partial_json / delta.input_json 逐段拼接
+    if (typeof piece === "string") {
+      slot.toolJsonBuffer += piece;
+      return;
+    }
+    // 兜底：对象/数组直接合并为字符串再解析
+    try {
+      slot.toolJsonBuffer += JSON.stringify(piece);
+    } catch {
+      slot.toolJsonBuffer += String(piece);
+    }
+  };
+
+  const finalizeToolInputIfAny = (slot) => {
+    if (!slot || slot.ref?.type !== "tool_use") return;
+    if (!slot.toolJsonBuffer) return;
+    const raw = slot.toolJsonBuffer.trim();
+    if (!raw) return;
+    try {
+      // 尝试直接解析完整 JSON；若为裸字串片段，做一次修复
+      slot.ref.input = JSON.parse(raw);
+    } catch {
+      // 容错：尝试补齐 JSON（例如少量逗号/丢尾），否则作为字符串原样输出
+      // 这里不做激进修复，只保底为字符串，避免抛错中断
+      slot.ref.input = raw;
+    }
+    // 清空缓冲
+    slot.toolJsonBuffer = "";
+  };
+
+  for (let i = 0; i < rawChunks.length; i++) {
+    const raw = rawChunks[i];
+    let evt;
+    try {
+      evt = parseMaybeJson(raw);
+      if (!evt) {
+        debug.ignoredCount++;
+        continue;
+      }
+      debug.parsedCount++;
+    } catch (e) {
+      debug.errors.push({ idx: i, error: e.message, raw });
+      continue;
+    }
+
+    const t = evt.type;
+
+    switch (t) {
+      case "ping":
+        // 心跳，无内容
+        break;
+
+      case "message_start": {
+        // evt.message 里常携带 id/model/usage
+        const m = evt.message || {};
+        if (m.id) message.id = m.id;
+        if (m.model) message.model = m.model;
+        if (m.usage?.input_tokens) message.usage.input_tokens += Number(m.usage.input_tokens) || 0;
+        if (m.usage?.output_tokens) message.usage.output_tokens += Number(m.usage.output_tokens) || 0;
+        break;
+      }
+
+      case "content_block_start": {
+        const idx = evt.index ?? 0;
+        const block = evt.content_block;
+        // 文本块或工具块都在这里创建
+        ensureBlock(idx, { content_block: block });
+        break;
+      }
+
+      case "content_block_delta": {
+        const idx = evt.index ?? 0;
+        const slot = ensureBlock(idx);
+        const d = evt.delta || evt; // 兼容有的实现把字段拍平
+
+        // 文本增量
+        if (d.type === "text_delta" && typeof d.text === "string") {
+          if (slot.ref.type !== "text") {
+            // 如果之前误判为工具，则转为文本（极少见）
+            slot.ref = { type: "text", text: "" };
+            message.content.push(slot.ref);
+          }
+          slot.ref.text += d.text;
+          break;
+        }
+
+        // 工具输入增量：partial_json / input_json / input_json_delta（都当作字符串片段拼接）
+        if (slot.ref.type === "tool_use") {
+          if (typeof d.partial_json === "string") {
+            feedToolJson(slot, d.partial_json);
+          } else if (typeof d.input_json === "string") {
+            feedToolJson(slot, d.input_json);
+          } else if (typeof d.input_json_delta === "string") {
+            feedToolJson(slot, d.input_json_delta);
+          } else if (d.json !== undefined) {
+            // 有些实现可能用 json 字段携带片段
+            feedToolJson(slot, d.json);
+          }
+        }
+        break;
+      }
+
+      case "content_block_stop": {
+        const idx = evt.index ?? 0;
+        const slot = blockMap.get(idx);
+        // 若是工具块，尝试把缓冲片段解析为最终 input
+        finalizeToolInputIfAny(slot);
+        break;
+      }
+
+      case "message_delta": {
+        // 累计 usage 增量、停止原因等
+        if (evt.delta?.stop_reason !== undefined) {
+          message.stop_reason = evt.delta.stop_reason;
+        }
+        if (evt.delta?.stop_sequence !== undefined) {
+          message.stop_sequence = evt.delta.stop_sequence;
+        }
+        if (evt.usage) {
+          if (evt.usage.output_tokens) {
+            message.usage.output_tokens += Number(evt.usage.output_tokens) || 0;
+          }
+          if (evt.usage.input_tokens) {
+            message.usage.input_tokens += Number(evt.usage.input_tokens) || 0;
+          }
+        }
+        break;
+      }
+
+      case "message_stop": {
+        // 终止事件，通常无需额外处理
+        break;
+      }
+
+      default: {
+        // 未识别事件，忽略
+        debug.ignoredCount++;
+      }
+    }
+  }
+
+  // 收尾：确保所有工具块的 JSON 缓冲被解析
+  for (const slot of blockMap.values()) {
+    finalizeToolInputIfAny(slot);
+  }
+  message.content = message.content.filter(
+	b => !(b.type === 'text' && b.text.trim() === '')
+  );
+  
+  if(message.content.length == 0){
+	  for(let c in rawChunks){
+		    logger.debug(c);
+	  }
+  }
+
+
+  return  message;
+}
+
+/* =========================
+ * 使用示例
+ * =========================
+const chunks = [
+  'data: {"type":"message_start","message":{"id":"msg_1","model":"claude-3-5-sonnet","usage":{"input_tokens":42}}}',
+  'data: {"type":"content_block_start","index":0,"content_block":{"type":"text"}}',
+  'data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"你好，"}}',
+  'data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"这里是 Claude。"}}',
+  'data: {"type":"content_block_stop","index":0}',
+
+  'data: {"type":"content_block_start","index":1,"content_block":{"type":"tool_use","id":"toolu_1","name":"search"}}',
+  'data: {"type":"content_block_delta","index":1,"delta":{"partial_json":"{\\"q\\":\\"Node.js\\","}}',
+  'data: {"type":"content_block_delta","index":1,"delta":{"partial_json":"\\"limit\\":5}"}}',
+  'data: {"type":"content_block_stop","index":1}',
+
+  'data: {"type":"message_delta","delta":{"stop_reason":"end_turn"},"usage":{"output_tokens":18}}',
+  'data: {"type":"message_stop"}'
+];
+
+const { message, debug } = mergeAnthropicChunks(chunks);
+logger.debug(JSON.stringify(message, null, 2), debug);
+*/
+
+
+export default mergeAnthropicChunks;
+
+// ---------------- 示例 ----------------
+// const chunks = [...]; // 这里放从 SSE 收集的 chunk
+// const merged = mergeAnthropicChunks(chunks);
+// console.log(JSON.stringify(merged, null, 2));