• java使用aspose.words实现Word拼接的功能


    需求:需要将其他模块生成的各个word汇总成一个整的Word。

    参考官方介绍地址:https://apireference.aspose.com/java/words/com.aspose.words/DocumentBuilder

    public Document appendDocument(Document mainDoc, Document addDoc, boolean isPortrait) {
         //设置书签,指定文档拼接的位置 String bookmark
    = "插入的位置"; DocumentBuilder builder = null; try { builder = new DocumentBuilder(mainDoc); BookmarkCollection bms = mainDoc.getRange().getBookmarks(); Bookmark bm = bms.get(bookmark); if (bm != null) { builder.moveToBookmark(bookmark, true, false); builder.writeln(); Node insertAfterNode = builder.getCurrentParagraph().getPreviousSibling(); insertDocumentAfterNode(insertAfterNode, mainDoc, addDoc); }
           //设置纸张大小 builder.getPageSetup().setPaperSize(PaperSize.A4);
    if (isPortrait) { //纵向纸张, builder.getPageSetup().setOrientation(Orientation.PORTRAIT); builder.insertBreak(BreakType.SECTION_BREAK_NEW_PAGE); } else { //横向 builder.getPageSetup().setOrientation(Orientation.LANDSCAPE); builder.insertBreak(BreakType.SECTION_BREAK_NEW_PAGE); } //builder.insertBreak(BreakType.PAGE_BREAK); } catch (Exception e) { // TODO Auto-generated catch block e.printStackTrace(); } return mainDoc; }

     之前少了拼接的关键代码,现在补上(2019-10-8):

    /**
          * @Description 向书签后插入文档
          * @param mainDoc 主文档
          * @param tobeInserted 拼接的文档
          * @param bookmark 书签
          * @Return com.aspose.words.Document
          * @Author Mr.Walloce
          * @Date 2019/7/27 18:33
          */
        private static Document insertDocumentAfterBookMark(Document mainDoc, Document tobeInserted, String bookmark)
                throws Exception {
            if (mainDoc == null) {
                return null;
            } else if (tobeInserted == null) {
                return mainDoc;
            } else {
                //构建新文档
                DocumentBuilder mainDocBuilder = new DocumentBuilder(mainDoc);
                if (bookmark != null && bookmark.length() > 0) {
                    //获取到书签
                    BookmarkCollection bms = mainDoc.getRange().getBookmarks();
                    Bookmark bm = bms.get(bookmark);
                    if (bm != null) {
                        mainDocBuilder.moveToBookmark(bookmark, true, false);
                        mainDocBuilder.writeln();
                        //获取到插入的位置
                        Node insertAfterNode = mainDocBuilder.getCurrentParagraph().getPreviousSibling();
                        insertDocumentAfterNode(insertAfterNode, mainDoc, tobeInserted);
                    }
                } else {
                    appendDoc(mainDoc, tobeInserted, true);
                }
    
                return mainDoc;
            }
        }
    
        /**
          * @Description TODO
          * @param insertAfterNode 插入的位置
          * @param mainDoc
          * @param srcDoc
          * @Return void
          * @Author Mr.Walloce
          * @Date 2019/7/27 14:51
          */
        private static void insertDocumentAfterNode(Node insertAfterNode, Document mainDoc, Document srcDoc)
                throws Exception {
            if (insertAfterNode.getNodeType() != 8 & insertAfterNode.getNodeType() != 5) {
                throw new Exception("The destination node should be either a paragraph or table.");
            } else {
                CompositeNode dstStory = insertAfterNode.getParentNode();
    
                while (null != srcDoc.getLastSection().getBody().getLastParagraph()
                        && !srcDoc.getLastSection().getBody().getLastParagraph().hasChildNodes()) {
                    srcDoc.getLastSection().getBody().getLastParagraph().remove();
                }
    
                NodeImporter importer = new NodeImporter(srcDoc, mainDoc, 1);
                int sectCount = srcDoc.getSections().getCount();
    
                for (int sectIndex = 0; sectIndex < sectCount; ++sectIndex) {
                    Section srcSection = srcDoc.getSections().get(sectIndex);
                    int nodeCount = srcSection.getBody().getChildNodes().getCount();
    
                    for (int nodeIndex = 0; nodeIndex < nodeCount; ++nodeIndex) {
                        Node srcNode = srcSection.getBody().getChildNodes().get(nodeIndex);
                        Node newNode = importer.importNode(srcNode, true);
                        dstStory.insertAfter(newNode, insertAfterNode);
                        insertAfterNode = newNode;
                    }
                }
    
            }
        }
    
        /**
          * @Description 文档拼接
          * @param dstDoc
          * @param srcDoc
          * @param includeSection
          * @Return void
          * @Author Mr.Walloce
          * @Date 2019/7/27 14:53
          */
        private static void appendDoc(Document dstDoc, Document srcDoc, boolean includeSection) throws Exception {
            if (includeSection) {
                Iterator<Section> var3 = srcDoc.getSections().iterator();
                while (var3.hasNext()) {
                    Section srcSection = (Section) var3.next();
                    Node dstNode = dstDoc.importNode(srcSection, true, 0);
                    dstDoc.appendChild(dstNode);
                }
            } else {
                Node node = dstDoc.getLastSection().getBody().getLastParagraph();
                if (node == null) {
                    node = new Paragraph(srcDoc);
                    dstDoc.getLastSection().getBody().appendChild(node);
                }
    
                if (node.getNodeType() != 8 & node.getNodeType() != 5) {
                    throw new Exception("Use appendDoc(dstDoc, srcDoc, true) instead of appendDoc(dstDoc, srcDoc, false)");
                }
    
                insertDocumentAfterNode(node, dstDoc, srcDoc);
            }
    初心回归,时光已逝!
  • 相关阅读:
    rsync同步公网yum源搭建本地yum仓库
    为smokeping添加日志开启debug
    python中if __name__ == '__main__':的作用
    使用telnet访问百度
    RRDtool绘制lvs连接数图形
    HTML表格
    ip netns
    redis动态扩展内存
    route命令
    设置nginx日志滚动
  • 原文地址:https://www.cnblogs.com/yin1361866686/p/9561006.html
Copyright © 2020-2023  润新知