import javax.swing.text.*;
import javax.swing.text.html.*;
import javax.swing.text.html.parser.*;
import java.io.*;
import java.net.*;
import java.util.*;
public class Outliner extends HTMLEditorKit.ParserCallback {
private Writer out;
private int level = 0;
private boolean inHeader=false;
private static String lineSeparator
= System.getProperty("line.separator", "\r\n");
public Outliner(Writer out) {
this.out = out;
}
public void handleStartTag(HTML.Tag tag,
MutableAttributeSet attributes, int position) {
int newLevel = 0;
if (tag == HTML.Tag.H1) newLevel = 1;
else if (tag == HTML.Tag.H2) newLevel = 2;
else if (tag == HTML.Tag.H3) newLevel = 3;
else if (tag == HTML.Tag.H4) newLevel = 4;
else if (tag == HTML.Tag.H5) newLevel = 5;
else if (tag == HTML.Tag.H6) newLevel = 6;
else return;
this.inHeader = true;
try {
if (newLevel > this.level) {
for (int i =0; i < newLevel-this.level; i++) {
out.write("<ul>" + lineSeparator + "<li>");
}
}
else if (newLevel < this.level) {
for (int i =0; i < this.level-newLevel; i++) {
out.write(lineSeparator + "</ul>" + lineSeparator);
}
out.write(lineSeparator + "<li>");
}
else {
out.write(lineSeparator + "<li>");
}
this.level = newLevel;
out.flush();
}
catch (IOException ex) {
System.err.println(ex);
}
}
public void handleEndTag(HTML.Tag tag, int position) {
if (tag == HTML.Tag.H1 || tag == HTML.Tag.H2
|| tag == HTML.Tag.H3 || tag == HTML.Tag.H4
|| tag == HTML.Tag.H5 || tag == HTML.Tag.H6) {
inHeader = false;
}
// work around bug in the parser that fails to call flush
if (tag == HTML.Tag.HTML) this.flush();
}
public void handleText(char[] text, int position) {
if (inHeader) {
try {
out.write(text);
out.flush();
}
catch (IOException ex) {
System.err.println(ex);
}
}
}
public void flush() {
try {
while (this.level-- > 0) {
out.write(lineSeparator + "</ul>");
}
out.flush();
}
catch (IOException e) {
System.err.println(e);
}
}
private static void parse(URL url, String encoding) throws IOException {
ParserGetter kit = new ParserGetter();
HTMLEditorKit.Parser parser = kit.getParser();
InputStream in = url.openStream();
InputStreamReader r = new InputStreamReader(in, encoding);
HTMLEditorKit.ParserCallback callback = new Outliner
(new OutputStreamWriter(System.out));
parser.parse(r, callback, true);
}
public static void main(String[] args) {
ParserGetter kit = new ParserGetter();
HTMLEditorKit.Parser parser = kit.getParser();
String encoding = "ISO-8859-1";
URL url = null;
try {
url = new URL(args[0]);
InputStream in = url.openStream();
InputStreamReader r = new InputStreamReader(in, encoding);
// parse once just to detect the encoding
HTMLEditorKit.ParserCallback doNothing
= new HTMLEditorKit.ParserCallback();
parser.parse(r, doNothing, false);
}
catch (MalformedURLException ex) {
System.out.println("Usage: java Outliner url");
return;
}
catch (ChangedCharSetException ex) {
String mimeType = ex.getCharSetSpec();
encoding = mimeType.substring(mimeType.indexOf("=") + 1).trim();
}
catch (IOException ex) {
System.err.println(ex);
}
catch (ArrayIndexOutOfBoundsException ex) {
System.out.println("Usage: java Outliner url");
return;
}
try {
parse(url, encoding);
}
catch(IOException ex) {
System.err.println(ex);
}
}
}
|