package com.custom.webui.service.impl; import com.baomidou.mybatisplus.core.toolkit.Wrappers; import com.custom.webui.config.props.ExportProperties; import com.custom.webui.dto.resp.MessageDTO; import com.custom.webui.dto.resp.StreamEvent; import com.custom.webui.entity.Chat; import com.custom.webui.entity.Message; import com.custom.webui.exception.BusinessException; import com.custom.webui.mapper.MessageMapper; import com.custom.webui.service.ChatExportService; import com.custom.webui.service.ChatService; import com.custom.webui.util.ChatPdfWriter; import com.custom.webui.util.MessageTreeUtil; import com.custom.webui.util.PdfFontSupport; import com.fasterxml.jackson.databind.ObjectMapper; import lombok.RequiredArgsConstructor; import lombok.extern.slf4j.Slf4j; import org.apache.pdfbox.pdmodel.PDDocument; import org.apache.pdfbox.pdmodel.PDDocumentInformation; import org.apache.pdfbox.pdfwriter.compress.CompressParameters; import org.springframework.stereotype.Service; import java.io.ByteArrayOutputStream; import java.nio.charset.StandardCharsets; import java.time.LocalDateTime; import java.time.format.DateTimeFormatter; import java.util.ArrayList; import java.util.LinkedHashMap; import java.util.List; import java.util.Map; /** * 会话导出实现。 * *

导出的是「当前分支」而不是整棵树

* 一个会话因为「重新生成 / 编辑消息」可能存在多条分支,而用户屏幕上同一时刻 * 只看到其中一条。导出的语义是「把我现在看到的这段对话存下来」, * 所以 TXT / PDF 只走当前活跃分支(与前端 {@code buildPath} 完全一致的规则: * 从最后一个根节点出发,沿 {@code activeChildId} 前进,缺失时回退到最后一个子节点)。 * *

但**分支本身就是这个项目的核心能力**,直接丢掉太可惜, * 因此 JSON 里额外给出 {@code allMessages}(扁平全量,带 parentId)—— * 想看全部内容或做二次处理的人可以用它还原整棵树。 */ @Slf4j @Service @RequiredArgsConstructor public class ChatExportServiceImpl implements ChatExportService { private static final DateTimeFormatter DATE_TIME = DateTimeFormatter.ofPattern("yyyy-MM-dd HH:mm:ss"); private static final DateTimeFormatter FILE_STAMP = DateTimeFormatter.ofPattern("yyyyMMdd-HHmmss"); /** 文件名里保留的最大字符数(中文文件名过长在部分系统上会被截断甚至失败) */ private static final int MAX_FILENAME_CHARS = 40; /** Windows / macOS 都不允许出现在文件名里的字符 */ private static final String ILLEGAL_FILENAME_CHARS = "[\\\\/:*?\"<>|\\r\\n\\t]"; private final ChatService chatService; private final MessageMapper messageMapper; private final PdfFontSupport pdfFontSupport; private final ExportProperties properties; private final ObjectMapper objectMapper; @Override public ExportFile export(Long userId, Long chatId, Format format) { Chat chat = chatService.requireOwnedChat(userId, chatId); List all = messageMapper.selectList( Wrappers.lambdaQuery() .eq(Message::getChatId, chatId) .orderByAsc(Message::getId)); // 树 → 当前分支。没有任何消息时给出空列表,仍然导出一份带标题的空文档, // 而不是报错 —— 「导出」是只读操作,没有理由因为对话为空就失败。 List tree = MessageTreeUtil.buildTree(all); List branch = activeBranch(tree); String stamp = LocalDateTime.now().format(FILE_STAMP); String filename = buildFilename(chat, stamp, format); byte[] data = switch (format) { case JSON -> toJson(chat, all, branch); case TXT -> toText(chat, branch).getBytes(StandardCharsets.UTF_8); case PDF -> toPdf(chat, branch); }; log.info("导出会话 userId={} chatId={} 格式={} 消息={}/{} 字节={}", userId, chatId, format, branch.size(), all.size(), data.length); return new ExportFile(data, filename, format.contentType()); } // ================================================================== // 当前分支 // ================================================================== /** * 沿活跃分支走出一条线性对话。 * *

规则与前端 {@code utils/treeBuild.js} 的 {@code buildPath} 保持一致 —— * 两处必须同源,否则会出现「屏幕上是这样、导出来是那样」的诡异问题: *

    *
  1. 起点取**最后一个**根节点(脏数据下可能有多个根)
  2. *
  3. 每步优先走 {@code activeChildId},它缺失或已失效时回退到最后一个子节点
  4. *
  5. 带步数上限,防止脏数据成环时死循环
  6. *
*/ private List activeBranch(List tree) { List path = new ArrayList<>(); if (tree == null || tree.isEmpty()) { return path; } MessageDTO current = tree.get(tree.size() - 1); int guard = 0; while (current != null && guard++ < 10000) { path.add(current); List children = current.getChildren(); if (children == null || children.isEmpty()) { break; } MessageDTO next = null; if (current.getActiveChildId() != null) { for (MessageDTO child : children) { if (child.getId() != null && child.getId().equals(current.getActiveChildId())) { next = child; break; } } } if (next == null) { next = children.get(children.size() - 1); } current = next; } return path; } // ================================================================== // JSON // ================================================================== private byte[] toJson(Chat chat, List all, List branch) { Map root = new LinkedHashMap<>(); root.put("exportedAt", LocalDateTime.now().format(DATE_TIME)); root.put("formatVersion", 1); root.put("source", "custom-webui"); Map chatNode = new LinkedHashMap<>(); chatNode.put("id", chat.getId()); chatNode.put("title", chat.getTitle()); chatNode.put("createdAt", chat.getCreatedAt()); chatNode.put("updatedAt", chat.getUpdatedAt()); chatNode.put("pinned", Boolean.TRUE.equals(chat.getPinned())); chatNode.put("archived", Boolean.TRUE.equals(chat.getArchived())); chatNode.put("messageCount", all.size()); root.put("chat", chatNode); // 当前分支:带引用来源、图片等展示所需的信息 List> messages = new ArrayList<>(); for (MessageDTO m : branch) { Map node = new LinkedHashMap<>(); node.put("id", m.getId()); node.put("parentId", m.getParentId()); node.put("role", m.getRole()); node.put("content", m.getContent()); node.put("model", m.getModel()); node.put("createdAt", m.getCreatedAt()); if (m.getSources() != null && !m.getSources().isEmpty()) { node.put("sources", m.getSources()); } if (m.getImageFileIds() != null && !m.getImageFileIds().isEmpty()) { node.put("imageFileIds", m.getImageFileIds()); } messages.add(node); } root.put("messages", messages); // 扁平全量:含其他分支,parentId 可还原整棵树 List> flat = new ArrayList<>(); for (Message m : all) { Map node = new LinkedHashMap<>(); node.put("id", m.getId()); node.put("parentId", m.getParentId()); node.put("role", m.getRole()); node.put("content", m.getContent()); node.put("model", m.getModel()); node.put("createdAt", m.getCreatedAt()); flat.add(node); } root.put("allMessages", flat); root.put("note", "messages 为当前显示的分支;allMessages 为扁平全量(含其他分支)," + "用 parentId 可还原完整消息树"); try { return objectMapper.writerWithDefaultPrettyPrinter().writeValueAsBytes(root); } catch (Exception e) { throw BusinessException.serverError("生成 JSON 失败:" + e.getMessage()); } } // ================================================================== // TXT // ================================================================== private String toText(Chat chat, List branch) { StringBuilder sb = new StringBuilder(4096); sb.append(chat.getTitle() == null ? "新对话" : chat.getTitle()).append('\n'); sb.append("=".repeat(48)).append('\n'); sb.append("导出时间:").append(LocalDateTime.now().format(DATE_TIME)).append('\n'); sb.append("创建时间:").append(formatTime(chat.getCreatedAt())).append('\n'); sb.append("消息条数:").append(branch.size()).append('\n'); sb.append('\n'); if (branch.isEmpty()) { sb.append("(这段对话还没有消息)\n"); return sb.toString(); } // 截断说明单独收集,最后再追加 —— 不能塞进 sb 让它在文档中间冒出来 StringBuilder truncationNote = new StringBuilder(); List visible = truncateBranch(branch, truncationNote); boolean first = true; for (MessageDTO m : visible) { if (!first) { sb.append('\n'); } first = false; sb.append('[').append(formatTime(m.getCreatedAt())).append("] ") .append(roleLabel(m)).append('\n'); sb.append("─".repeat(32)).append('\n'); sb.append(clip(m.getContent())).append('\n'); appendExtras(sb, m); } if (truncationNote.length() > 0) { sb.append('\n').append(truncationNote).append('\n'); } return sb.toString(); } // ================================================================== // PDF // ================================================================== private byte[] toPdf(Chat chat, List branch) { try (PDDocument document = new PDDocument(); ByteArrayOutputStream out = new ByteArrayOutputStream()) { PDDocumentInformation info = new PDDocumentInformation(); info.setTitle(chat.getTitle() == null ? "对话" : chat.getTitle()); info.setProducer("custom-webui"); document.setDocumentInformation(info); // 注意生命周期:embedSubset 是延迟到 save() 才读字体的, // 所以 handle 必须活到 save() 之后(见 PdfFontSupport 的类注释) try (PdfFontSupport.PdfFontHandle handle = pdfFontSupport.open(document)) { ChatPdfWriter writer = new ChatPdfWriter(document, handle); writer.writeTitle(chat.getTitle() == null ? "新对话" : chat.getTitle()); writer.writeMeta("导出时间:" + LocalDateTime.now().format(DATE_TIME) + " 创建时间:" + formatTime(chat.getCreatedAt()) + " 消息条数:" + branch.size()); writer.writeSpacing(6f); writer.writeRule(); if (branch.isEmpty()) { writer.writeSpacing(10f); writer.writeMeta("(这段对话还没有消息)"); } StringBuilder truncationNote = new StringBuilder(); for (MessageDTO m : truncateBranch(branch, truncationNote)) { writer.writeMessageHeader("[" + formatTime(m.getCreatedAt()) + "] " + roleLabel(m)); writer.writeParagraph(clip(m.getContent())); String extras = extrasText(m); if (!extras.isEmpty()) { writer.writeSpacing(2f); writer.writeMeta(extras); } writer.writeSpacing(6f); } if (truncationNote.length() > 0) { writer.writeSpacing(6f); writer.writeMeta(truncationNote.toString()); } writer.close(); // 压缩内容流:中文长对话不压缩时体积会明显偏大 document.save(out, CompressParameters.DEFAULT_COMPRESSION); } return out.toByteArray(); } catch (BusinessException e) { throw e; } catch (Exception e) { log.error("生成 PDF 失败 chatId={}", chat.getId(), e); throw BusinessException.serverError("生成 PDF 失败:" + e.getMessage()); } } // ================================================================== // 公共小工具 // ================================================================== /** * 按上限截断分支,并把「截断了」这件事写进 {@code note}。 * *

静默截断是不可接受的:用户会以为导出的就是全部内容。 */ private List truncateBranch(List branch, StringBuilder note) { int max = properties.getMaxMessages(); if (max <= 0 || branch.size() <= max) { return branch; } note.append("注意:该对话共 ").append(branch.size()) .append(" 条消息,已按配置上限只导出前 ").append(max) .append(" 条(可调整 webui.export.max-messages)。"); return branch.subList(0, max); } private String clip(String content) { if (content == null) { return ""; } int max = properties.getMaxCharsPerMessage(); if (max > 0 && content.length() > max) { return content.substring(0, max) + "\n…(本条消息过长,已截断 " + (content.length() - max) + " 字)"; } return content; } /** TXT 里附在消息下方的补充信息:引用来源、图片数量 */ private void appendExtras(StringBuilder sb, MessageDTO m) { String extras = extrasText(m); if (!extras.isEmpty()) { sb.append(extras).append('\n'); } } private String extrasText(MessageDTO m) { List parts = new ArrayList<>(); if (m.getSources() != null && !m.getSources().isEmpty()) { List refs = new ArrayList<>(); int index = 1; for (StreamEvent.Source s : m.getSources()) { StringBuilder one = new StringBuilder(); one.append('[').append(index++).append("] "); one.append(s.getFilename() == null ? "(未知文件)" : s.getFilename()); if (s.getSection() != null && !s.getSection().isBlank()) { one.append(" · ").append(s.getSection()); } refs.add(one.toString()); } parts.add("引用来源:" + String.join(";", refs)); } if (m.getImageFileIds() != null && !m.getImageFileIds().isEmpty()) { // 导出文件不带图片本体:图片存在服务端,且 PDF 里嵌图会让体积和实现复杂度都陡增。 // 但必须**明确写出来**,否则用户会以为图片内容被丢掉了。 parts.add("[本条消息附带 " + m.getImageFileIds().size() + " 张图片," + "导出文件不含图片本体,可在网页里查看]"); } return String.join("\n", parts); } private String roleLabel(MessageDTO m) { String role = m.getRole() == null ? "" : m.getRole(); return switch (role) { case "user" -> "我"; case "assistant" -> m.getModel() == null || m.getModel().isBlank() ? "AI" : "AI(" + m.getModel() + ")"; case "system" -> "系统"; default -> role; }; } private String formatTime(LocalDateTime time) { return time == null ? "-" : time.format(DATE_TIME); } /** * 生成下载文件名:{标题}_{时间戳}.{后缀} * *

标题来自用户输入(早期版本甚至是模型生成的),里面完全可能出现 * {@code /}、{@code :}、换行等字符 —— 直接拼进 Content-Disposition * 轻则文件名错乱,重则触发响应头注入,所以必须白名单化清洗。 */ private String buildFilename(Chat chat, String stamp, Format format) { String title = chat.getTitle() == null ? "" : chat.getTitle().trim(); title = title.replaceAll(ILLEGAL_FILENAME_CHARS, " ") .replaceAll("\\s+", " ") .trim(); if (title.isEmpty()) { title = "对话" + chat.getId(); } if (title.length() > MAX_FILENAME_CHARS) { title = title.substring(0, MAX_FILENAME_CHARS); } return title + "_" + stamp + "." + format.extension(); } }