@cmach_socket/opencode-qq-bot 0.0.0-stage → 0.3.0

package/src/naplink/render.ts

@@ -0,0 +1,440 @@
+import type { MessageEvent, MessageSegment, OneBotApi } from "@naplink/naplink"
+import type { ChatTarget } from "../chat"
+import { targetKey } from "../chat"
+import type { Logger } from "../logger"
+import {
+  MAX_IMAGE_BYTES,
+  inlineImageBytes,
+  truncateCodePoints,
+  type MediaStore,
+  type PromptFile,
+  type StoredImage,
+} from "./media"
+
+export { MAX_IMAGE_BYTES, inlineImageBytes }
+export type { PromptFile } from "./media"
+
+/** 合并转发的解析上限，避免撑爆上下文。 */
+export const MAX_FORWARD_DEPTH = 3
+export const MAX_FORWARD_NODES = 50
+export const MAX_FORWARD_CHARS = 4000
+/** 整棵转发树最多解析的节点数，避免嵌套转发把 API 调用放大。 */
+export const MAX_FORWARD_TOTAL_NODES = 100
+/** 整棵转发树最多保留的字符数。 */
+export const MAX_FORWARD_TOTAL_CHARS = 8000
+/** 单轮提示里最多引用的图片数量，超出的图片只保留占位文本。 */
+export const MAX_PROMPT_IMAGES = 8
+/** 单轮提示里引用图片的原始字节总量上限，超出的图片只保留占位文本。 */
+export const MAX_PROMPT_IMAGE_BYTES = 16 * 1024 * 1024
+
+export interface RenderedMessage {
+  /** 纯文本版本，用于会话历史与提示词。 */
+  text: string
+  /** 图片附件，交给 OpenCode 作为多模态输入。 */
+  files: PromptFile[]
+}
+
+export interface RenderContext {
+  api: OneBotApi
+  target: ChatTarget
+  logger: Logger
+  /** 受控临时图片存储。 */
+  media: MediaStore
+  /** 用户显示名缓存，避免每条消息都请求一次 API。 */
+  names?: Map<string, { name: string; expires: number }>
+}
+
+/** NapCat 的 forward 段不在 SDK 类型里，这里补进判别联合。 */
+interface ForwardSegment {
+  type: "forward"
+  data: { id?: string }
+}
+
+type Segment = MessageSegment | ForwardSegment
+
+function segments(message: readonly MessageSegment[]): readonly Segment[] {
+  return message as readonly Segment[]
+}
+
+/** 判断消息是否含有值得交给模型的内容。 */
+export function shouldProcess(message: readonly MessageSegment[], userId?: string): boolean {
+  if (userId && mentionsUser(message, userId)) return true
+  return segments(message).some((segment) => {
+    switch (segment.type) {
+      case "text":
+        return segment.data.text.trim().length > 0
+      case "image":
+      case "record":
+      case "forward":
+      case "reply":
+        return true
+      default:
+        return false
+    }
+  })
+}
+
+/** 判断消息里是否 @ 了指定 QQ 号。 */
+export function mentionsUser(
+  message: readonly MessageSegment[],
+  userId: string | undefined,
+): boolean {
+  if (!userId) return false
+  return segments(message).some(
+    (segment) => segment.type === "at" && String(segment.data.qq) === userId,
+  )
+}
+
+/** 图片引用的预算：数量与原始字节总量，渲染过程中边用边扣。 */
+export interface ImageBudget {
+  images: number
+  bytes: number
+}
+
+/** 创建一份默认图片预算。 */
+export function createImageBudget(): ImageBudget {
+  return { images: MAX_PROMPT_IMAGES, bytes: MAX_PROMPT_IMAGE_BYTES }
+}
+
+/** 从预算里扣掉一张图片；预算不足或单图过大时不扣。 */
+function takeImage(budget: ImageBudget, file: PromptFile): boolean {
+  const bytes = file.bytes ?? inlineImageBytes(file.uri)
+  if (bytes > MAX_IMAGE_BYTES) return false
+  if (budget.images <= 0 || budget.bytes < bytes) return false
+  budget.images -= 1
+  budget.bytes -= bytes
+  return true
+}
+
+/** 按预算裁剪图片附件列表（数量与字节总量）。 */
+export function limitFiles(files: readonly PromptFile[]): PromptFile[] {
+  const budget = createImageBudget()
+  const kept: PromptFile[] = []
+  for (const file of files) {
+    if (!takeImage(budget, file)) continue
+    kept.push(file)
+  }
+  return kept
+}
+
+const USER_NAME_TTL_MS = 5 * 60 * 1000
+
+/** 查询用户显示名：群聊优先群名片，私聊使用 QQ 昵称。 */
+export async function resolveUserName(ctx: RenderContext, userId: string): Promise<string> {
+  if (userId === "all") return "全体成员"
+  const key = `${targetKey(ctx.target)}:${userId}`
+  const cached = ctx.names?.get(key)
+  if (cached && cached.expires > Date.now()) return cached.name
+  try {
+    const info = (
+      ctx.target.kind === "group"
+        ? await ctx.api.getGroupMemberInfo(ctx.target.id, userId)
+        : await ctx.api.getStrangerInfo(userId)
+    ) as { card?: string; nickname?: string } | undefined
+    const name = info?.card?.trim() || info?.nickname?.trim() || "未知用户"
+    ctx.names?.set(key, { name, expires: Date.now() + USER_NAME_TTL_MS })
+    return name
+  } catch (error) {
+    ctx.logger.warn(`获取用户信息失败 user_id=${userId}`, error)
+    return "未知用户"
+  }
+}
+
+interface ForwardContent {
+  text: string
+  images: StoredImage[]
+}
+
+interface ForwardBudget {
+  nodes: number
+  chars: number
+}
+
+function createForwardBudget(): ForwardBudget {
+  return { nodes: MAX_FORWARD_TOTAL_NODES, chars: MAX_FORWARD_TOTAL_CHARS }
+}
+
+async function renderForwardSegments(
+  ctx: RenderContext,
+  message: readonly MessageSegment[],
+  depth: number,
+  budget: ImageBudget,
+  forwardBudget: ForwardBudget,
+): Promise<ForwardContent> {
+  const parts: string[] = []
+  const images: StoredImage[] = []
+  for (const segment of segments(message)) {
+    try {
+      switch (segment.type) {
+        case "text":
+          parts.push(segment.data.text)
+          break
+        case "at":
+          parts.push(`@${segment.data.qq} `)
+          break
+        case "image": {
+          parts.push("[图片]")
+          const url = segment.data.url || segment.data.file
+          if (!url || budget.images <= 0) break
+          const stored = await ctx.media.storeImage(url)
+          if (!stored || !takeImage(budget, stored)) break
+          images.push(stored)
+          break
+        }
+        case "record":
+          parts.push("[语音]")
+          break
+        case "video":
+          parts.push("[视频]")
+          break
+        case "audio":
+          parts.push("[音频]")
+          break
+        case "file":
+          parts.push(`[文件]${segment.data.name || segment.data.file}`)
+          break
+        case "face":
+          parts.push("[表情]")
+          break
+        case "json":
+          parts.push("[卡片消息]")
+          break
+        case "xml":
+          parts.push("[XML消息]")
+          break
+        case "markdown":
+          parts.push("[Markdown消息]")
+          break
+        case "reply":
+          // 转发节点里的引用没有上下文，跳过
+          break
+        case "forward": {
+          if (depth >= MAX_FORWARD_DEPTH) {
+            parts.push("[嵌套合并转发]")
+            break
+          }
+          const nested = await renderForward(ctx, segment.data.id, depth + 1, budget, forwardBudget)
+          parts.push(nested.text)
+          images.push(...nested.images)
+          break
+        }
+        default:
+          break
+      }
+    } catch (error) {
+      ctx.logger.warn(`解析转发消息段失败 type=${segment.type}`, error)
+      parts.push("[无法解析的消息]")
+    }
+  }
+  return { text: parts.join("").trim(), images }
+}
+
+/** 把合并转发解析成 ``<forward>`` 文本块，并附带其中的图片。 */
+export async function renderForward(
+  ctx: RenderContext,
+  id: string | undefined,
+  depth = 0,
+  budget: ImageBudget = createImageBudget(),
+  forwardBudget: ForwardBudget = createForwardBudget(),
+): Promise<ForwardContent> {
+  if (!id) return { text: "[合并转发]", images: [] }
+  if (forwardBudget.nodes <= 0 || forwardBudget.chars <= 0) {
+    return { text: "[合并转发（内容过多，已截断）]", images: [] }
+  }
+  let nodes: { name: string; segments: MessageSegment[] }[] = []
+  try {
+    const data = (await ctx.api.getForwardMessage(id)) as {
+      messages?: {
+        sender?: { user_id?: number; nickname?: string; card?: string }
+        message?: unknown[]
+      }[]
+    }
+    nodes = (data.messages ?? []).map((entry) => ({
+      name:
+        entry.sender?.card || entry.sender?.nickname || String(entry.sender?.user_id ?? "未知用户"),
+      segments: (entry.message ?? []) as MessageSegment[],
+    }))
+  } catch (error) {
+    ctx.logger.warn(`解析合并转发失败 id=${id}`, error)
+    return { text: "[合并转发（获取失败）]", images: [] }
+  }
+  if (nodes.length === 0) return { text: "[合并转发]", images: [] }
+
+  const body: string[] = []
+  const images: StoredImage[] = []
+  let consumed = 0
+  for (const node of nodes) {
+    if (consumed >= MAX_FORWARD_NODES || forwardBudget.nodes <= 0 || forwardBudget.chars <= 0) {
+      break
+    }
+    forwardBudget.nodes -= 1
+    consumed += 1
+    const rendered = await renderForwardSegments(ctx, node.segments, depth, budget, forwardBudget)
+    const line = `${node.name}：${rendered.text || "[空消息]"}`
+    const limited = truncateCodePoints(line, Math.max(0, forwardBudget.chars))
+    forwardBudget.chars -= Array.from(limited).length
+    body.push(limited)
+    images.push(...rendered.images)
+  }
+  if (consumed < nodes.length) {
+    body.push(`...（还有 ${nodes.length - consumed} 条）`)
+  }
+
+  let rendered = body.join("\n")
+  if (Array.from(rendered).length > MAX_FORWARD_CHARS) {
+    rendered = `${truncateCodePoints(rendered, MAX_FORWARD_CHARS)}\n...（合并转发过长，已截断）`
+  }
+  return { text: `<forward>\n${rendered}\n</forward>`, images }
+}
+
+/** 把语音消息转成文字，转写失败时返回空字符串。 */
+export async function fetchVoiceText(
+  ctx: RenderContext,
+  messageId: number | undefined,
+): Promise<string> {
+  if (messageId === undefined) return ""
+  try {
+    const result = await ctx.api.fetchPttText(messageId)
+    return result?.text?.trim() ?? ""
+  } catch (error) {
+    ctx.logger.warn(`语音转文字失败 message_id=${messageId}`, error)
+    return ""
+  }
+}
+
+/**
+ * 把一条 QQ 消息拆成 (纯文本, 图片附件)。
+ *
+ * 引用消息会展开成 ``<quote>`` 块，合并转发展开成 ``<forward>`` 块，
+ * 图片既留占位文本又内联成附件，语音先转写成文字。
+ */
+export async function renderMessage(
+  ctx: RenderContext,
+  event: MessageEvent,
+): Promise<RenderedMessage> {
+  const textBuffer: string[] = []
+  const files: PromptFile[] = []
+  const message = segments(event.message)
+  const budget = createImageBudget()
+  const forwardBudget = createForwardBudget()
+  const voiceCache = new Map<string, string>()
+  let imageIndex = 0
+
+  const replySegments = message.filter((segment) => segment.type === "reply")
+  if (replySegments.length > 0) {
+    const quoteLines = ["<quote>"]
+    for (const segment of replySegments) {
+      if (!segment.data.id) continue
+      try {
+        const data = (await ctx.api.getMessage(segment.data.id)) as
+          { sender?: { user_id?: number }; message?: MessageSegment[] } | undefined
+        if (!data) continue
+        const name = data.sender?.user_id
+          ? await resolveUserName(ctx, String(data.sender.user_id))
+          : "未知用户"
+        const quoted = await renderForwardSegments(
+          ctx,
+          data.message ?? [],
+          MAX_FORWARD_DEPTH,
+          budget,
+          forwardBudget,
+        )
+        quoteLines.push(`${name}：${quoted.text || "[空消息]"}`)
+        for (const image of quoted.images) {
+          imageIndex += 1
+          files.push({ ...image, name: `qq-quote-${imageIndex}.${image.ext}` })
+        }
+      } catch (error) {
+        ctx.logger.warn(`获取引用消息失败 id=${segment.data.id}`, error)
+      }
+    }
+    quoteLines.push("</quote>")
+    textBuffer.push(`${quoteLines.join("\n")}\n`)
+  }
+
+  for (const segment of message) {
+    try {
+      switch (segment.type) {
+        case "text":
+          textBuffer.push(segment.data.text)
+          break
+        case "at":
+          textBuffer.push(`@${await resolveUserName(ctx, String(segment.data.qq))} `)
+          break
+        case "image": {
+          textBuffer.push("[图片] ")
+          const url = segment.data.url || segment.data.file
+          if (!url || budget.images <= 0) break
+          const stored = await ctx.media.storeImage(url)
+          if (!stored || !takeImage(budget, stored)) break
+          imageIndex += 1
+          files.push({ ...stored, name: `qq-image-${imageIndex}.${stored.ext}` })
+          break
+        }
+        case "record": {
+          const rawId = (segment.data as { message_id?: number | string }).message_id
+          const messageId =
+            rawId === undefined
+              ? event.message_id
+              : Number.isFinite(Number(rawId))
+                ? Number(rawId)
+                : undefined
+          const key = String(messageId ?? "unknown")
+          let voiceText = voiceCache.get(key)
+          if (voiceText === undefined) {
+            voiceText = await fetchVoiceText(ctx, messageId)
+            voiceCache.set(key, voiceText)
+          }
+          textBuffer.push(voiceText ? `[语音] ${voiceText}` : "[语音]")
+          break
+        }
+        case "reply":
+          // 已在外层 quote 块处理过
+          break
+        case "forward": {
+          const forward = await renderForward(ctx, segment.data.id, 0, budget, forwardBudget)
+          const prefix =
+            textBuffer.length > 0 && !textBuffer[textBuffer.length - 1]?.endsWith("\n") ? "\n" : ""
+          textBuffer.push(`${prefix}${forward.text}\n`)
+          for (const image of forward.images) {
+            imageIndex += 1
+            files.push({
+              ...image,
+              name: `qq-forward-${imageIndex}.${image.ext}`,
+            })
+          }
+          break
+        }
+        case "video":
+          textBuffer.push("[视频]")
+          break
+        case "audio":
+          textBuffer.push("[音频]")
+          break
+        case "file":
+          textBuffer.push(`[文件]${segment.data.name || segment.data.file}`)
+          break
+        case "face":
+          textBuffer.push("[表情]")
+          break
+        case "json":
+          textBuffer.push("[卡片消息]")
+          break
+        case "xml":
+          textBuffer.push("[XML消息]")
+          break
+        case "markdown":
+          textBuffer.push("[Markdown消息]")
+          break
+        default:
+          ctx.logger.debug(`跳过未知消息段 type=${(segment as { type: string }).type}`)
+          break
+      }
+    } catch (error) {
+      ctx.logger.warn(`解析消息段失败 type=${segment.type}`, error)
+      textBuffer.push("[无法解析的消息]")
+    }
+  }
+
+  return { text: textBuffer.join(""), files }
+}