package com.rarchives.ripme.ripper.rippers; import java.io.IOException; import java.net.MalformedURLException; import java.net.URL; import java.util.ArrayList; import java.util.Arrays; import java.util.List; import java.util.regex.Matcher; import java.util.regex.Pattern; import org.jsoup.nodes.Document; import org.jsoup.nodes.Element; import org.jsoup.select.Elements; import com.rarchives.ripme.ripper.AbstractHTMLRipper; import com.rarchives.ripme.utils.Http; public class CheveretoRipper extends AbstractHTMLRipper { public CheveretoRipper(URL url) throws IOException { super(url); } public static List explicit_domains_1 = Arrays.asList("www.ezphotoshare.com", "hushpix.com"); @Override public String getHost() { String host = url.toExternalForm(); return host; } @Override public String getDomain() { String host = url.toExternalForm(); return host; } @Override public boolean canRip(URL url) { String url_name = url.toExternalForm(); if (explicit_domains_1.contains(url_name.split("/")[2]) == true) { return true; } return false; } @Override public String getGID(URL url) throws MalformedURLException { Pattern p = Pattern.compile("(?:https?://)?(?:www\\.)?[a-z1-9]*\\.[a-z1-9]*/album/([a-zA-Z1-9]*)/?$"); Matcher m = p.matcher(url.toExternalForm()); if (m.matches()) { return m.group(1); } else if (m.matches() == false) { Pattern pa = Pattern.compile("(?:https?://)?(?:www\\.)?[a-z1-9]*\\.[a-z1-9]*/([a-zA-Z1-9_-]*)/albums/?$"); Matcher ma = pa.matcher(url.toExternalForm()); if (ma.matches()) { return ma.group(1); } } throw new MalformedURLException("Expected chevereto URL format: " + "site.domain/album/albumName or site.domain/username/albums- got " + url + " instead"); } @Override public Document getFirstPage() throws IOException { // "url" is an instance field of the superclass return Http.url(url).get(); } @Override public Document getNextPage(Document doc) throws IOException { // Find next page String nextUrl = ""; Element elem = doc.select("li.pagination-next > a").first(); String nextPage = elem.attr("href"); if (nextUrl == "") { throw new IOException("No more pages"); } // Sleep for half a sec to avoid getting IP banned sleep(500); return Http.url(nextUrl).get(); } @Override public List getURLsFromPage(Document doc) { List result = new ArrayList(); Document userpage_doc; // We check for the following string to see if this is a user page or not if (doc.toString().contains("content=\"gallery\"")) { for (Element elem : doc.select("a.image-container")) { String link = elem.attr("href"); logger.info("Grabbing album " + link); try { userpage_doc = Http.url(link).get(); } catch(IOException e){ logger.warn("Failed to log link in Jsoup"); userpage_doc = null; e.printStackTrace(); } for (Element element : userpage_doc.select("a.image-container > img")) { String imageSource = element.attr("src"); logger.info("Found image " + link); // We remove the .md from images so we download the full size image // not the medium ones imageSource = imageSource.replace(".md", ""); result.add(imageSource); } } } else { for (Element el : doc.select("a.image-container > img")) { String imageSource = el.attr("src"); // We remove the .md from images so we download the full size image // not the medium ones imageSource = imageSource.replace(".md", ""); result.add(imageSource); } } return result; } @Override public void downloadURL(URL url, int index) { addURLToDownload(url, getPrefix(index)); } }