ImgScroll/src/main/java/com/rarchives/ripme/ripper/rippers/ModelxRipper.java
2018-03-25 09:42:38 +02:00

65 lines
1.6 KiB
Java

package com.rarchives.ripme.ripper.rippers;
import com.rarchives.ripme.ripper.AbstractHTMLRipper;
import com.rarchives.ripme.utils.Http;
import org.jsoup.nodes.Document;
import org.jsoup.nodes.Element;
import java.io.IOException;
import java.net.MalformedURLException;
import java.net.URL;
import java.util.ArrayList;
import java.util.List;
import java.util.regex.Matcher;
import java.util.regex.Pattern;
public class ModelxRipper extends AbstractHTMLRipper {
public ModelxRipper(URL url) throws IOException {
super(url);
}
@Override
public String getHost() {
return "modelx";
}
@Override
public String getDomain() {
return "modelx.org";
}
@Override
public String getGID(URL url) throws MalformedURLException {
Pattern p = Pattern.compile("^.*modelx.org/.*/(.+)$");
Matcher m = p.matcher(url.toExternalForm());
if (m.matches()) {
return m.group(1);
}
throw new MalformedURLException("Expected URL format: http://www.modelx.org/[category (one or more)]/xxxxx got: " + url);
}
@Override
public Document getFirstPage() throws IOException {
return Http.url(url).get();
}
@Override
public List<String> getURLsFromPage(Document page) {
List<String> result = new ArrayList<>();
for (Element el : page.select(".gallery-icon > a")) {
result.add(el.attr("href"));
}
return result;
}
@Override
public void downloadURL(URL url, int index) {
addURLToDownload(url, getPrefix(index));
}
}