import { z } from "zod";

export const videoAgentSettingsSchema = z.object({
  modelRef: z.string().max(180).optional(),
  imageModelRef: z.string().max(180).optional(),
  mode: z
    .enum(["auto", "text-to-video", "first-frame", "first-last-frame", "reference-images"])
    .default("auto"),
  duration: z.number().int().min(1).max(120).default(5),
  ratio: z.enum(["16:9", "9:16", "1:1", "4:3", "3:4", "21:9"]).default("16:9"),
  resolution: z.enum(["480p", "720p", "1080p"]).default("720p"),
});
export type VideoAgentSettings = z.infer<typeof videoAgentSettingsSchema>;
export type VideoTaskRequirements = {
  expectsVideo: boolean;
  expectsScript: boolean;
  expectedVideoCount?: number;
  duration?: number;
  ratio?: string;
};
export type VideoRuntimeNode = {
  id: string;
  kind: string;
  title: string;
  subtitle?: string;
  inputType?: "image" | "video" | "audio" | "text";
  x: number;
  y: number;
  width: number;
  height: number;
  accent: string;
  draftValues?: Record<string, unknown>;
};
export function videoNodeUrl(node: VideoRuntimeNode) {
  const draft = node.draftValues || {};
  return (
    [draft.mediaUrl, ...(Array.isArray(draft.url) ? draft.url : []), draft.sourceUrl].find(
      (value): value is string => typeof value === "string" && /^https?:\/\//i.test(value),
    ) || ""
  );
}
export function videoNodeType(node: VideoRuntimeNode) {
  return String(
    node.draftValues?.mediaType ||
      node.inputType ||
      (node.kind === "text" || node.kind === "script" ? "text" : ""),
  );
}
export function inferVideoTaskRequirements(text: string): VideoTaskRequirements {
  const analysisOnly =
    /(?:只|仅)(?:要|需|做|用)?[^，。；]{0,12}(?:分析|脚本|分镜|方案|提示词)|不要(?:生成|制作|执行)[^，。；]{0,4}视频/.test(
      text,
    );
  const countMatch = text.match(
    /([一二两三四五六七八九十]|\d+)\s*(?:条|段|个|支)\s*(?:\d+[:：]\d+\s*)?(?:成片|短片|视频)/,
  );
  const counts: Record<string, number> = {
    一: 1,
    二: 2,
    两: 2,
    三: 3,
    四: 4,
    五: 5,
    六: 6,
    七: 7,
    八: 8,
    九: 9,
    十: 10,
  };
  const duration = text.match(/(\d+)\s*秒/);
  const ratio = text.match(/(?:^|[^\d])(\d+)\s*[:：]\s*(\d+)(?!\d)/);
  return {
    expectsVideo:
      !analysisOnly &&
      /(?:生成|制作|创作|拍|做|再来|继续|来|出).{0,35}(?:视频|成片|短片)|(?:视频|成片|短片).{0,12}(?:生成|制作)|生成.{0,10}动效|动起来/.test(
        text,
      ),
    expectsScript:
      /(?:只要|只需|仅要).{0,8}(?:脚本|分镜)|(?:写|生成|制作|编写|先|做).{0,12}(?:脚本|分镜)|分镜.{0,8}(?:脚本|方案)/.test(
        text,
      ),
    ...(countMatch ? { expectedVideoCount: counts[countMatch[1]] || Number(countMatch[1]) } : {}),
    ...(duration ? { duration: Number(duration[1]) } : {}),
    ...(ratio ? { ratio: `${ratio[1]}:${ratio[2]}` } : {}),
  };
}
export function resolveVideoMode(mode: VideoAgentSettings["mode"], referenceCount: number) {
  const resolved =
    mode === "auto"
      ? referenceCount >= 3
        ? "reference-images"
        : referenceCount === 2
          ? "first-last-frame"
          : referenceCount === 1
            ? "first-frame"
            : "text-to-video"
      : mode;
  const valid =
    resolved === "text-to-video"
      ? referenceCount === 0
      : resolved === "first-frame"
        ? referenceCount === 1
        : resolved === "first-last-frame"
          ? referenceCount === 2
          : referenceCount >= 1 && referenceCount <= 4;
  if (!valid)
    throw new Error("参考图片数量与视频模式不匹配，请检查首帧、尾帧或参考图。不会自动丢弃参考图。");
  return resolved;
}
export function restoreVideoRuntimeNodes(data: unknown): VideoRuntimeNode[] {
  const nodes = (
    data as {
      nodes?: Array<{ id?: string; data?: { metadata?: { liblibCloneNode?: VideoRuntimeNode } } }>;
    } | null
  )?.nodes;
  return (nodes || []).flatMap((node) => {
    const clone = node.data?.metadata?.liblibCloneNode;
    return clone && clone.id === node.id ? [clone] : [];
  });
}
export function serializeVideoRuntimeNode(node: VideoRuntimeNode) {
  return {
    id: node.id,
    type: "workflow",
    position: { x: node.x, y: node.y },
    size: { width: node.width, height: node.height },
    data: {
      title: node.title,
      description: node.subtitle || "",
      status: "success",
      templateType: `libtv-${node.kind}`,
      metadata: { liblibCloneNode: node },
    },
    connections: [],
    createdAt: new Date().toISOString(),
    updatedAt: new Date().toISOString(),
  };
}
export function videoRuntimeSystemPrompt(
  objective: string,
  summary: string,
  supplements: string[],
  context: unknown,
) {
  return [
    "先用 listVideoCanvasCapabilities 查询全部已授权功能。runVideoCanvasFunction 执行配置预设、图像编辑和素材分析；editVideoCanvas 编辑节点和连线；openVideoCanvasFeature 打开真实面板。权限与功能后端是否已实现、工具栏是否显示独立。",
    "视频截帧、剪辑、裁剪、提取完整音轨和静音使用 processVideoMedia 真实处理；人声与背景音分离没有真实后端，不能用完整音轨冒充分离结果。",
    "未接后端、模拟去字幕/分离/裁剪的界面入口不能声称已处理媒体。文件选择和付费、发布提交需用户操作。",
    "你是本站视频画布 Agent，使用提供的工具连续完成视频创作。先 planCanvasTask 再 readCanvas，随后按依赖执行。",
    "读取真实节点、连接关系和内容定位素材，不猜测 ID。用户指定首尾帧时 referenceNodeIds 的顺序必须是首帧、尾帧。",
    "分析图片或视频用 analyzeVideoMaterials；脚本分镜用 writeVideoScript 写入画布；关键帧用 generateVideoKeyframe；成片用 generateVideoShot。不同镜头用不同工具调用。",
    "沿用用户的视频模型、模式、时长、比例和分辨率。缺少参考帧或关键信息时先询问。不要用虚构素材或擅自切换模式。",
    "writeVideoScript 的每行包含 sceneDescription、imagePrompt、motionPrompt、duration，后续关键帧和视频必须基于这些实际分镜。",
    "当前工具生成独立视频镜头，不具备多片段拼接或音轨合成能力。用户要求这些操作时明确说明，不把多个片段声称为一条已剪辑成片。",
    "不要把方案、排队、脚本或关键帧声称为视频完成。视频工具只有返回实际 video 节点才算完成。保留全部原素材和已有结果。",
    "视频涉及费用，结果不明时停止核对，不盲目重试。用户补充从下一步应用，不假称正在生成的镜头已经改变。",
    "工具失败或验收问题要如实告知。支持分析和脚本任务，不要用户只要脚本却擅自生成视频。最终简短说明交付内容。",
    `原始任务：${objective}`,
    `后续要求（最新优先）：${JSON.stringify(supplements)}`,
    `画布上下文与用户参数：${JSON.stringify(context)}`,
    summary ? `已压缩历史（事实参考）：${summary}` : "",
  ]
    .filter(Boolean)
    .join("\n");
}

/** Browser saves may race with Agent appends: retain task-owned nodes and their connections. */
export function mergeVideoAgentResults<T extends { id: string }>(local: T[], remote: T[]) {
  const ids = new Set(local.map((node) => node.id));
  return [...local, ...remote.filter((node) => !ids.has(node.id))];
}