package com.borqs.browser; import java.util.regex.Matcher; import java.util.regex.Pattern; import android.net.Uri; import android.util.Patterns; import android.webkit.URLUtil; /** * Utility methods for Url manipulation */ public class UrlUtils { static final Pattern ACCEPTED_URI_SCHEMA = Pattern.compile( "(?i)" + // switch on case insensitive matching "(" + // begin group for schema "(?:http|https|file):\\/\\/" + "|(?:inline|data|about|javascript):" + ")" + "(.*)" ); // Google search private final static String QUICKSEARCH_G = "http://www.google.com/m?q=%s"; private final static String QUERY_PLACE_HOLDER = "%s"; // Regular expression to strip http:// and optionally // the trailing slash private static final Pattern STRIP_URL_PATTERN = Pattern.compile("^http://(.*?)/?$"); private UrlUtils() { /* cannot be instantiated */ } /** * Strips the provided url of preceding "http://" and any trailing "/". Does not * strip "https://". If the provided string cannot be stripped, the original string * is returned. * * TODO: Put this in TextUtils to be used by other packages doing something similar. * * @param url a url to strip, like "http://www.google.com/" * @return a stripped url like "www.google.com", or the original string if it could * not be stripped */ public static String stripUrl(String url) { if (url == null) return null; Matcher m = STRIP_URL_PATTERN.matcher(url); if (m.matches()) { return m.group(1); } else { return url; } } protected static String smartUrlFilter(Uri inUri) { if (inUri != null) { return smartUrlFilter(inUri.toString()); } return null; } /** * Attempts to determine whether user input is a URL or search * terms. Anything with a space is passed to search. * * Converts to lowercase any mistakenly uppercased schema (i.e., * "Http://" converts to "http://" * * @return Original or modified URL * */ public static String smartUrlFilter(String url) { return smartUrlFilter(url, true); } /** * Attempts to determine whether user input is a URL or search * terms. Anything with a space is passed to search if canBeSearch is true. * * Converts to lowercase any mistakenly uppercased schema (i.e., * "Http://" converts to "http://" * * @param canBeSearch If true, will return a search url if it isn't a valid * URL. If false, invalid URLs will return null * @return Original or modified URL * */ public static String smartUrlFilter(String url, boolean canBeSearch) { String inUrl = url.trim(); boolean hasSpace = inUrl.indexOf(' ') != -1; Matcher matcher = ACCEPTED_URI_SCHEMA.matcher(inUrl); if (matcher.matches()) { // force scheme to lowercase String scheme = matcher.group(1); String lcScheme = scheme.toLowerCase(); if (!lcScheme.equals(scheme)) { inUrl = lcScheme + matcher.group(2); } if (hasSpace && Patterns.WEB_URL.matcher(inUrl).matches()) { inUrl = inUrl.replace(" ", "%20"); } return inUrl; } if (!hasSpace) { if (Patterns.WEB_URL.matcher(inUrl).matches()) { return URLUtil.guessUrl(inUrl); } } if (canBeSearch) { return URLUtil.composeSearchUrl(inUrl, QUICKSEARCH_G, QUERY_PLACE_HOLDER); } return null; } /* package */ static String fixUrl(String inUrl) { // FIXME: Converting the url to lower case // duplicates functionality in smartUrlFilter(). // However, changing all current callers of fixUrl to // call smartUrlFilter in addition may have unwanted // consequences, and is deferred for now. int colon = inUrl.indexOf(':'); boolean allLower = true; for (int index = 0; index < colon; index++) { char ch = inUrl.charAt(index); if (!Character.isLetter(ch)) { break; } allLower &= Character.isLowerCase(ch); if (index == colon - 1 && !allLower) { inUrl = inUrl.substring(0, colon).toLowerCase() + inUrl.substring(colon); } } if (inUrl.startsWith("http://") || inUrl.startsWith("https://")) return inUrl; if (inUrl.startsWith("http:") || inUrl.startsWith("https:")) { if (inUrl.startsWith("http:/") || inUrl.startsWith("https:/")) { inUrl = inUrl.replaceFirst("/", "//"); } else inUrl = inUrl.replaceFirst(":", "://"); } return inUrl; } // Returns the filtered URL. Cannot return null, but can return an empty string /* package */ static String filteredUrl(String inUrl) { if (inUrl == null) { return ""; } if (inUrl.startsWith("content:") || inUrl.startsWith("browser:")) { return ""; } return inUrl; } }