Removed uses of internal JDK classes.

Removed @author tag.
Dealt with exceprions properly.
General reformatting.

git-svn-id: https://svn.apache.org/repos/asf/poi/trunk@827905 13f79535-47bb-0310-9956-ffa450edef68
This commit is contained in:
Ugo Cei 2009-10-21 07:11:00 +00:00
parent ebc7f3cb26
commit 3e82ec8035

View File

@ -21,6 +21,7 @@ import java.io.IOException;
import java.io.StringReader;
import java.util.Iterator;
import java.util.List;
import javax.xml.namespace.NamespaceContext;
import javax.xml.parsers.DocumentBuilder;
import javax.xml.parsers.DocumentBuilderFactory;
@ -38,172 +39,138 @@ import org.apache.poi.xssf.usermodel.XSSFMap;
import org.apache.poi.xssf.usermodel.XSSFRow;
import org.apache.poi.xssf.usermodel.helpers.XSSFSingleXmlCell;
import org.apache.poi.xssf.usermodel.helpers.XSSFXmlColumnPr;
import org.apache.xml.utils.PrefixResolver;
import org.apache.xml.utils.PrefixResolverDefault;
import org.w3c.dom.Document;
import org.w3c.dom.Node;
import org.w3c.dom.NodeList;
import org.xml.sax.InputSource;
import org.xml.sax.SAXException;
import com.sun.org.apache.xml.internal.utils.PrefixResolver;
import com.sun.org.apache.xml.internal.utils.PrefixResolverDefault;
/**
*
* Imports data from an external XML to an XLSX to an XML according to one of the mapping defined.
*
* The output XML Schema must respect this limitations:
*
* - the input XML must be valid according to the XML Schema used in the mapping
* - denormalized table mapping is not supported (see OpenOffice part 4: chapter 3.5.1.7)
* - all the namespaces used in the document must be declared in the root node
*
*
* @author Roberto Manicardi
*
* Imports data from an external XML to an XLSX according to one of the mappings
* defined.The output XML Schema must respect this limitations:
* <ul>
* <li>the input XML must be valid according to the XML Schema used in the mapping</li>
* <li>denormalized table mapping is not supported (see OpenOffice part 4: chapter 3.5.1.7)</li>
* <li>all the namespaces used in the document must be declared in the root node</li>
* </ul>
*/
public class XSSFImportFromXML {
private XSSFMap map;
private static POILogger logger = POILogFactory.getLogger(XSSFImportFromXML.class);
public XSSFImportFromXML(XSSFMap map){
this.map = map;
}
/**
* Imports an XML into the XLSX using the Custom XML mapping defined
*
* @param xmlInputString the XML to import
* @throws SAXException raised if error occurs during XML parsing
* @throws XPathExpressionException raised if error occurs during XML navigation
*/
public void importFromXML(String xmlInputString) throws SAXException, XPathExpressionException{
try{
DocumentBuilderFactory factory = DocumentBuilderFactory.newInstance();
factory.setNamespaceAware(true);
DocumentBuilder builder = factory.newDocumentBuilder();
Document doc = builder.parse(new InputSource(new StringReader(xmlInputString.trim())));
List<XSSFSingleXmlCell> singleXmlCells = map.getRelatedSingleXMLCell();
List<Table> tables = map.getRelatedTables();
XPathFactory xpathFactory = XPathFactory.newInstance();
XPath xpath = xpathFactory.newXPath();
// Setting namespace context to XPath
// Assuming that the namespace prefix in the mapping xpath is the same as the one used in the document
final PrefixResolver resolver = new PrefixResolverDefault(doc.getDocumentElement());
NamespaceContext ctx = new NamespaceContext() {
public String getNamespaceURI(String prefix) {
return resolver.getNamespaceForPrefix(prefix);
}
// Dummy implementation - not used!
public Iterator getPrefixes(String val) {
return null;
}
// Dummy implemenation - not used!
public String getPrefix(String uri) {
return null;
}
};
xpath.setNamespaceContext(ctx);
private XSSFMap map;
for(XSSFSingleXmlCell singleXmlCell :singleXmlCells ){
String xpathString = singleXmlCell.getXpath();
Node result = (Node) xpath.evaluate(xpathString,doc, XPathConstants.NODE);
String textContent = result.getTextContent();
logger.log(POILogger.DEBUG,"Extracting with xpath "+xpathString+" : value is '"+textContent+"'");
XSSFCell cell = singleXmlCell.getReferencedCell();
logger.log(POILogger.DEBUG,"Setting '"+textContent+"' to cell "+cell.getColumnIndex()+"-"+cell.getRowIndex()+" in sheet "+cell.getSheet().getSheetName());
cell.setCellValue(textContent);
}
for(Table table : tables){
String commonXPath = table.getCommonXpath();
NodeList result = (NodeList) xpath.evaluate(commonXPath,doc, XPathConstants.NODESET);
int rowOffset = table.getStartCellReference().getRow()+1;//the first row contains the table header
int columnOffset = table.getStartCellReference().getCol()-1;
for(int i = 0; i< result.getLength();i++){
// TODO: implement support for denormalized XMLs (see OpenOffice part 4: chapter 3.5.1.7)
for(XSSFXmlColumnPr xmlColumnPr: table.getXmlColumnPrs()){
int localColumnId = (int)xmlColumnPr.getId();
int rowId = rowOffset+i;
int columnId = columnOffset+localColumnId;
String localXPath = xmlColumnPr.getLocalXPath();
localXPath = localXPath.substring(localXPath.substring(1).indexOf('/')+1);
// Build an XPath to select the right node (assuming that the commonXPath != "/")
String nodeXPath = commonXPath+"["+(i+1)+"]"+localXPath;
// TODO: convert the data to the cell format
String value = (String) xpath.evaluate(nodeXPath,result.item(i), XPathConstants.STRING);
logger.log(POILogger.DEBUG,"Extracting with xpath "+nodeXPath+" : value is '"+value+"'");
XSSFRow row = table.getXSSFSheet().getRow(rowId);
if(row==null){
row = table.getXSSFSheet().createRow(rowId);
}
XSSFCell cell = row.getCell(columnId);
if(cell==null){
cell = row.createCell(columnId);
}
logger.log(POILogger.DEBUG,"Setting '"+value+"' to cell "+cell.getColumnIndex()+"-"+cell.getRowIndex()+" in sheet "+table.getXSSFSheet().getSheetName());
cell.setCellValue(value.trim());
}
}
}
}catch(IOException e){
//Thrown by StringReader
e.printStackTrace();
}catch(ParserConfigurationException e){
//Thrown by DocumentBuilderFactory
e.printStackTrace();
}
}
private static POILogger logger = POILogFactory.getLogger(XSSFImportFromXML.class);
public XSSFImportFromXML(XSSFMap map) {
this.map = map;
}
/**
* Imports an XML into the XLSX using the Custom XML mapping defined
*
* @param xmlInputString the XML to import
* @throws SAXException if error occurs during XML parsing
* @throws XPathExpressionException if error occurs during XML navigation
* @throws ParserConfigurationException if there are problems with XML parser configuration
* @throws IOException if there are problems reading the input string
*/
public void importFromXML(String xmlInputString) throws SAXException, XPathExpressionException, ParserConfigurationException, IOException {
DocumentBuilderFactory factory = DocumentBuilderFactory.newInstance();
factory.setNamespaceAware(true);
DocumentBuilder builder = factory.newDocumentBuilder();
Document doc = builder.parse(new InputSource(new StringReader(xmlInputString.trim())));
List<XSSFSingleXmlCell> singleXmlCells = map.getRelatedSingleXMLCell();
List<Table> tables = map.getRelatedTables();
XPathFactory xpathFactory = XPathFactory.newInstance();
XPath xpath = xpathFactory.newXPath();
// Setting namespace context to XPath
// Assuming that the namespace prefix in the mapping xpath is the
// same as the one used in the document
final PrefixResolver resolver = new PrefixResolverDefault(doc.getDocumentElement());
NamespaceContext ctx = new NamespaceContext() {
public String getNamespaceURI(String prefix) {
return resolver.getNamespaceForPrefix(prefix);
}
// Dummy implementation - not used!
public Iterator getPrefixes(String val) {
return null;
}
// Dummy implemenation - not used!
public String getPrefix(String uri) {
return null;
}
};
xpath.setNamespaceContext(ctx);
for (XSSFSingleXmlCell singleXmlCell : singleXmlCells) {
String xpathString = singleXmlCell.getXpath();
Node result = (Node) xpath.evaluate(xpathString, doc, XPathConstants.NODE);
String textContent = result.getTextContent();
logger.log(POILogger.DEBUG, "Extracting with xpath " + xpathString + " : value is '" + textContent + "'");
XSSFCell cell = singleXmlCell.getReferencedCell();
logger.log(POILogger.DEBUG, "Setting '" + textContent + "' to cell " + cell.getColumnIndex() + "-" + cell.getRowIndex() + " in sheet "
+ cell.getSheet().getSheetName());
cell.setCellValue(textContent);
}
for (Table table : tables) {
String commonXPath = table.getCommonXpath();
NodeList result = (NodeList) xpath.evaluate(commonXPath, doc, XPathConstants.NODESET);
int rowOffset = table.getStartCellReference().getRow() + 1;// the first row contains the table header
int columnOffset = table.getStartCellReference().getCol() - 1;
for (int i = 0; i < result.getLength(); i++) {
// TODO: implement support for denormalized XMLs (see
// OpenOffice part 4: chapter 3.5.1.7)
for (XSSFXmlColumnPr xmlColumnPr : table.getXmlColumnPrs()) {
int localColumnId = (int) xmlColumnPr.getId();
int rowId = rowOffset + i;
int columnId = columnOffset + localColumnId;
String localXPath = xmlColumnPr.getLocalXPath();
localXPath = localXPath.substring(localXPath.substring(1).indexOf('/') + 1);
// Build an XPath to select the right node (assuming
// that the commonXPath != "/")
String nodeXPath = commonXPath + "[" + (i + 1) + "]" + localXPath;
// TODO: convert the data to the cell format
String value = (String) xpath.evaluate(nodeXPath, result.item(i), XPathConstants.STRING);
logger.log(POILogger.DEBUG, "Extracting with xpath " + nodeXPath + " : value is '" + value + "'");
XSSFRow row = table.getXSSFSheet().getRow(rowId);
if (row == null) {
row = table.getXSSFSheet().createRow(rowId);
}
XSSFCell cell = row.getCell(columnId);
if (cell == null) {
cell = row.createCell(columnId);
}
logger.log(POILogger.DEBUG, "Setting '" + value + "' to cell " + cell.getColumnIndex() + "-" + cell.getRowIndex() + " in sheet "
+ table.getXSSFSheet().getSheetName());
cell.setCellValue(value.trim());
}
}
}
}
}