java 读取xlsx
(1)导入maven
<!-- 读取excel文件 -->
<dependency>
<groupId>org.apache.poi</groupId>
<artifactId>poi</artifactId>
<version>3.9</version>
</dependency>
(2)代码
public static void main(String[] args) {
// 解压Book1.xlsx
ZipFile xlsxFile;
try {
xlsxFile = new ZipFile(new File("/Users/chengtao/downloads/教练入职时间.xlsx"));
DocumentBuilderFactory dbf = DocumentBuilderFactory.newInstance();
// 先读取sharedStrings.xml这个文件备用
ZipEntry sharedStringXML = xlsxFile.getEntry("xl/sharedStrings.xml");
InputStream sharedStringXMLIS = xlsxFile.getInputStream(sharedStringXML);
Document sharedString;
sharedString = dbf.newDocumentBuilder().parse(sharedStringXMLIS);
NodeList str = sharedString.getElementsByTagName("t");
String sharedStrings[] = new String[str.getLength()];
for (int n = 0; n < str.getLength(); n++) {
Element element = (Element) str.item(n);
sharedStrings[n] = element.getTextContent();
}
// 找到解压文件夹里的workbook.xml,此文件中包含了这张工作表中有几个sheet
ZipEntry workbookXML = xlsxFile.getEntry("xl/workbook.xml");
InputStream workbookXMLIS = xlsxFile.getInputStream(workbookXML);
Document doc = dbf.newDocumentBuilder().parse(workbookXMLIS);
// 获取一共有几个sheet
NodeList nl = doc.getElementsByTagName("sheet");
for (int i = 0; i < nl.getLength(); i++) {
Element element = (Element) nl.item(i);// 将node转化为element,用来得到每个节点的属性
System.out.println(element.getAttribute("name"));// 输出sheet节点的name属性的值
// 接着就要到解压文件夹里找到对应的name值的xml文件,比如在workbook.xml中有<sheet name="Sheet1"
// sheetId="1" r:id="rId1" /> 节点
// 那么就可以在解压文件夹里的xl/worksheets下找到sheet1.xml,这个xml文件夹里就是包含的表格的内容
ZipEntry sheetXML = xlsxFile.getEntry("xl/worksheets/" + element.getAttribute("name").toLowerCase() + ".xml");
InputStream sheetXMLIS = xlsxFile.getInputStream(sheetXML);
Document sheetdoc = dbf.newDocumentBuilder().parse(sheetXMLIS);
NodeList rowdata = sheetdoc.getElementsByTagName("row");
HashMap<String, String> nameAndEntryTimeMap = new HashMap<String,String>();
for (int j = 0; j < rowdata.getLength(); j++) {
System.out.println(" 第" + j +"行 ------------- ");
// 得到每个行
// 行的格式:
/*
* <row r="1" spans="1:3">r表示第一行,spans表示有几列 <c r="A1"
* t="s">//r表示该列的列表
* ,t="s"个人认为是表示这个单元格的内容可以在sharedStrings.xml这个文件里找到,对应的节点
* 下标就是v节点的值,即0,若没有t属性,则v的值就是该单元格的内容 <v>0</v> </c> <c r="B1"
* t="s"> <v>1</v> </c> <c r="C1" t="s"> <v>2</v> </c> </row>
*/
Element row = (Element) rowdata.item(j); // 获取每一行
NodeList columndata = row.getElementsByTagName("c");// 根据行得到每个行中的列
String name_Key = null; //姓名
String time_value = null; //入职时间
for (int k = 0; k < columndata.getLength(); k++) {
Element column = (Element) columndata.item(k);
NodeList values = column.getElementsByTagName("v");
Element value = (Element) values.item(0);
if (column.getAttribute("t") != null & column.getAttribute("t").equals("s")) {
// 如果是共享字符串则在sharedstring.xml里查找该列的值
System.out.println(" 第" + k +"列 "+sharedStrings[Integer.parseInt(value.getTextContent())] + " ");
if(k ==0){
name_Key = sharedStrings[Integer.parseInt(value.getTextContent())];
}else if(k==2){
time_value = sharedStrings[Integer.parseInt(value.getTextContent())];
}
} else {
if (value != null) {
System.out.print(value.getTextContent() + " ");
}else {
System.out.println("j : " + j + " k : " + k + " null");
}
}
}
nameAndEntryTimeMap.put(name_Key, time_value);
System.out.println();
}
System.out.println(nameAndEntryTimeMap);
}
} catch (ZipException e) {
// TODO Auto-generated catch block
e.printStackTrace();
} catch (IOException e) {
// TODO Auto-generated catch block
e.printStackTrace();
} catch (SAXException e) {
// TODO Auto-generated catch block
e.printStackTrace();
} catch (ParserConfigurationException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}