aboutsummaryrefslogtreecommitdiffstats
path: root/include/Scrape.php
diff options
context:
space:
mode:
authorfriendica <info@friendica.com>2013-03-10 20:39:27 -0700
committerfriendica <info@friendica.com>2013-03-10 20:39:27 -0700
commit26499981b6c18879cfb1d7ddb8321b965277a5fd (patch)
tree0268b6732aafe167e1036bc566c13a54d7854d99 /include/Scrape.php
parent952b2ef2ab256d46d616db70edc42ffe640a5e9d (diff)
downloadvolse-hubzilla-26499981b6c18879cfb1d7ddb8321b965277a5fd.tar.gz
volse-hubzilla-26499981b6c18879cfb1d7ddb8321b965277a5fd.tar.bz2
volse-hubzilla-26499981b6c18879cfb1d7ddb8321b965277a5fd.zip
cleanup
Diffstat (limited to 'include/Scrape.php')
-rw-r--r--include/Scrape.php712
1 files changed, 0 insertions, 712 deletions
diff --git a/include/Scrape.php b/include/Scrape.php
deleted file mode 100644
index b1138091c..000000000
--- a/include/Scrape.php
+++ /dev/null
@@ -1,712 +0,0 @@
-<?php /** @file */
-
-require_once('library/HTML5/Parser.php');
-require_once('include/crypto.php');
-
-if(! function_exists('scrape_dfrn')) {
-function scrape_dfrn($url) {
-
- $a = get_app();
-
- $ret = array();
-
- logger('scrape_dfrn: url=' . $url);
-
- $s = fetch_url($url);
-
- if(! $s)
- return $ret;
-
- $headers = $a->get_curl_headers();
- logger('scrape_dfrn: headers=' . $headers, LOGGER_DEBUG);
-
-
- $lines = explode("\n",$headers);
- if(count($lines)) {
- foreach($lines as $line) {
- // don't try and run feeds through the html5 parser
- if(stristr($line,'content-type:') && ((stristr($line,'application/atom+xml')) || (stristr($line,'application/rss+xml'))))
- return ret;
- }
- }
-
- try {
- $dom = HTML5_Parser::parse($s);
- } catch (DOMException $e) {
- logger('scrape_dfrn: parse error: ' . $e);
- }
-
- if(! $dom)
- return $ret;
-
- $items = $dom->getElementsByTagName('link');
-
- // get DFRN link elements
-
- foreach($items as $item) {
- $x = $item->getAttribute('rel');
- if(($x === 'alternate') && ($item->getAttribute('type') === 'application/atom+xml'))
- $ret['feed_atom'] = $item->getAttribute('href');
- if(substr($x,0,5) == "dfrn-") {
- $ret[$x] = $item->getAttribute('href');
- }
- if($x === 'lrdd') {
- $decoded = urldecode($item->getAttribute('href'));
- if(preg_match('/acct:([^@]*)@/',$decoded,$matches))
- $ret['nick'] = $matches[1];
- }
- }
-
- // Pull out hCard profile elements
-
- $largest_photo = 0;
-
- $items = $dom->getElementsByTagName('*');
- foreach($items as $item) {
- if(attribute_contains($item->getAttribute('class'), 'vcard')) {
- $level2 = $item->getElementsByTagName('*');
- foreach($level2 as $x) {
- if(attribute_contains($x->getAttribute('class'),'fn')) {
- $ret['fn'] = $x->textContent;
- }
- if((attribute_contains($x->getAttribute('class'),'photo'))
- || (attribute_contains($x->getAttribute('class'),'avatar'))) {
- $size = intval($x->getAttribute('width'));
- // dfrn prefers 175, so if we find this, we set largest_size so it can't be topped.
- if(($size > $largest_photo) || ($size == 175) || (! $largest_photo)) {
- $ret['photo'] = $x->getAttribute('src');
- $largest_photo = (($size == 175) ? 9999 : $size);
- }
- }
- if(attribute_contains($x->getAttribute('class'),'key')) {
- $ret['key'] = $x->textContent;
- }
- }
- }
- }
-
- return $ret;
-}}
-
-
-
-
-
-
-if(! function_exists('validate_dfrn')) {
-function validate_dfrn($a) {
- $errors = 0;
- if(! x($a,'key'))
- $errors ++;
- if(! x($a,'dfrn-request'))
- $errors ++;
- if(! x($a,'dfrn-confirm'))
- $errors ++;
- if(! x($a,'dfrn-notify'))
- $errors ++;
- if(! x($a,'dfrn-poll'))
- $errors ++;
- return $errors;
-}}
-
-if(! function_exists('scrape_meta')) {
-function scrape_meta($url) {
-
- $a = get_app();
-
- $ret = array();
-
- logger('scrape_meta: url=' . $url);
-
- $s = fetch_url($url);
-
- if(! $s)
- return $ret;
-
- $headers = $a->get_curl_headers();
- logger('scrape_meta: headers=' . $headers, LOGGER_DEBUG);
-
- $lines = explode("\n",$headers);
- if(count($lines)) {
- foreach($lines as $line) {
- // don't try and run feeds through the html5 parser
- if(stristr($line,'content-type:') && ((stristr($line,'application/atom+xml')) || (stristr($line,'application/rss+xml'))))
- return ret;
- }
- }
-
- try {
- $dom = HTML5_Parser::parse($s);
- } catch (DOMException $e) {
- logger('scrape_meta: parse error: ' . $e);
- }
-
- if(! $dom)
- return $ret;
-
- $items = $dom->getElementsByTagName('meta');
-
- // get DFRN link elements
-
- foreach($items as $item) {
- $x = $item->getAttribute('name');
- if(substr($x,0,5) == "dfrn-")
- $ret[$x] = $item->getAttribute('content');
- }
-
- return $ret;
-}}
-
-
-if(! function_exists('scrape_vcard')) {
-function scrape_vcard($url) {
-
- $a = get_app();
-
- $ret = array();
-
- logger('scrape_vcard: url=' . $url);
-
- $s = fetch_url($url);
-
- if(! $s)
- return $ret;
-
- $headers = $a->get_curl_headers();
- $lines = explode("\n",$headers);
- if(count($lines)) {
- foreach($lines as $line) {
- // don't try and run feeds through the html5 parser
- if(stristr($line,'content-type:') && ((stristr($line,'application/atom+xml')) || (stristr($line,'application/rss+xml'))))
- return ret;
- }
- }
-
- try {
- $dom = HTML5_Parser::parse($s);
- } catch (DOMException $e) {
- logger('scrape_vcard: parse error: ' . $e);
- }
-
- if(! $dom)
- return $ret;
-
- // Pull out hCard profile elements
-
- $largest_photo = 0;
-
- $items = $dom->getElementsByTagName('*');
- foreach($items as $item) {
- if(attribute_contains($item->getAttribute('class'), 'vcard')) {
- $level2 = $item->getElementsByTagName('*');
- foreach($level2 as $x) {
- if(attribute_contains($x->getAttribute('class'),'fn'))
- $ret['fn'] = $x->textContent;
- if((attribute_contains($x->getAttribute('class'),'photo'))
- || (attribute_contains($x->getAttribute('class'),'avatar'))) {
- $size = intval($x->getAttribute('width'));
- if(($size > $largest_photo) || (! $largest_photo)) {
- $ret['photo'] = $x->getAttribute('src');
- $largest_photo = $size;
- }
- }
- if((attribute_contains($x->getAttribute('class'),'nickname'))
- || (attribute_contains($x->getAttribute('class'),'uid'))) {
- $ret['nick'] = $x->textContent;
- }
- }
- }
- }
-
- return $ret;
-}}
-
-
-if(! function_exists('scrape_feed')) {
-function scrape_feed($url) {
-
- $a = get_app();
-
- $ret = array();
- $s = fetch_url($url);
-
- $headers = $a->get_curl_headers();
- $code = $a->get_curl_code();
-
- logger('scrape_feed: returns: ' . $code . ' headers=' . $headers, LOGGER_DEBUG);
-
- if(! $s) {
- logger('scrape_feed: no data returned for ' . $url);
- return $ret;
- }
-
-
- $lines = explode("\n",$headers);
- if(count($lines)) {
- foreach($lines as $line) {
- if(stristr($line,'content-type:')) {
- if(stristr($line,'application/atom+xml') || stristr($s,'<feed')) {
- $ret['feed_atom'] = $url;
- return $ret;
- }
- if(stristr($line,'application/rss+xml') || stristr($s,'<rss')) {
- $ret['feed_rss'] = $url;
- return $ret;
- }
- }
- }
- // perhaps an RSS version 1 feed with a generic or incorrect content-type?
- if(stristr($s,'</item>')) {
- $ret['feed_rss'] = $url;
- return $ret;
- }
- }
-
- try {
- $dom = HTML5_Parser::parse($s);
- } catch (DOMException $e) {
- logger('scrape_feed: parse error: ' . $e);
- }
-
- if(! $dom) {
- logger('scrape_feed: failed to parse.');
- return $ret;
- }
-
-
- $head = $dom->getElementsByTagName('base');
- if($head) {
- foreach($head as $head0) {
- $basename = $head0->getAttribute('href');
- break;
- }
- }
- if(! $basename)
- $basename = implode('/', array_slice(explode('/',$url),0,3)) . '/';
-
- $items = $dom->getElementsByTagName('link');
-
- // get Atom/RSS link elements, take the first one of either.
-
- if($items) {
- foreach($items as $item) {
- $x = $item->getAttribute('rel');
- if(($x === 'alternate') && ($item->getAttribute('type') === 'application/atom+xml')) {
- if(! x($ret,'feed_atom'))
- $ret['feed_atom'] = $item->getAttribute('href');
- }
- if(($x === 'alternate') && ($item->getAttribute('type') === 'application/rss+xml')) {
- if(! x($ret,'feed_rss'))
- $ret['feed_rss'] = $item->getAttribute('href');
- }
- }
- }
-
- // Drupal and perhaps others only provide relative URL's. Turn them into absolute.
-
- if(x($ret,'feed_atom') && (! strstr($ret['feed_atom'],'://')))
- $ret['feed_atom'] = $basename . $ret['feed_atom'];
- if(x($ret,'feed_rss') && (! strstr($ret['feed_rss'],'://')))
- $ret['feed_rss'] = $basename . $ret['feed_rss'];
-
- return $ret;
-}}
-
-
-/**
- *
- * Probe a network address to discover what kind of protocols we need to communicate with it.
- *
- * Warning: this function is a bit touchy and there are some subtle dependencies within the logic flow.
- * Edit with care.
- *
- */
-
-/**
- *
- * PROBE_DIASPORA has a bias towards returning Diaspora information
- * while PROBE_NORMAL has a bias towards dfrn/zot - in the case where
- * an address (such as a Friendica address) supports more than one type
- * of network.
- *
- */
-
-
-define ( 'PROBE_NORMAL', 0);
-
-function probe_url($url, $mode = PROBE_NORMAL) {
- require_once('include/email.php');
-
- $result = array();
-
- if(! $url)
- return $result;
-
- $network = null;
- $has_lrdd = false;
- $email_conversant = false;
-
- $twitter = ((strpos($url,'twitter.com') !== false) ? true : false);
- $lastfm = ((strpos($url,'last.fm/user') !== false) ? true : false);
-
- $at_addr = ((strpos($url,'@') !== false) ? true : false);
-
- if((! $twitter) && (! $lastfm)) {
-
- if(strpos($url,'mailto:') !== false && $at_addr) {
- $url = str_replace('mailto:','',$url);
- $links = array();
- }
- else
- $links = lrdd($url);
-
- if(count($links)) {
- $has_lrdd = true;
-
- logger('probe_url: found lrdd links: ' . print_r($links,true), LOGGER_DATA);
- foreach($links as $link) {
- if($link['@attributes']['rel'] === NAMESPACE_ZOT)
- $zot = unamp($link['@attributes']['href']);
- if($link['@attributes']['rel'] === NAMESPACE_DFRN)
- $dfrn = unamp($link['@attributes']['href']);
- if($link['@attributes']['rel'] === 'salmon')
- $notify = unamp($link['@attributes']['href']);
- if($link['@attributes']['rel'] === NAMESPACE_FEED)
- $poll = unamp($link['@attributes']['href']);
- if($link['@attributes']['rel'] === 'http://microformats.org/profile/hcard')
- $hcard = unamp($link['@attributes']['href']);
- if($link['@attributes']['rel'] === 'http://webfinger.net/rel/profile-page')
- $profile = unamp($link['@attributes']['href']);
- if($link['@attributes']['rel'] === 'http://portablecontacts.net/spec/1.0')
- $poco = unamp($link['@attributes']['href']);
-
- }
-
- // Status.Net can have more than one profile URL. We need to match the profile URL
- // to a contact on incoming messages to prevent spam, and we won't know which one
- // to match. So in case of two, one of them is stored as an alias. Only store URL's
- // and not webfinger user@host aliases. If they've got more than two non-email style
- // aliases, let's hope we're lucky and get one that matches the feed author-uri because
- // otherwise we're screwed.
-
- foreach($links as $link) {
- if($link['@attributes']['rel'] === 'alias') {
- if(strpos($link['@attributes']['href'],'@') === false) {
- if(isset($profile)) {
- if($link['@attributes']['href'] !== $profile)
- $alias = unamp($link['@attributes']['href']);
- }
- else
- $profile = unamp($link['@attributes']['href']);
- }
- }
- }
- }
- elseif($mode == PROBE_NORMAL) {
-
- // Check email
-
- $orig_url = $url;
- if((strpos($orig_url,'@')) && validate_email($orig_url)) {
- $x = q("SELECT `prvkey` FROM `user` WHERE `uid` = %d LIMIT 1",
- intval(local_user())
- );
- $r = q("SELECT * FROM `mailacct` WHERE `uid` = %d AND `server` != '' LIMIT 1",
- intval(local_user())
- );
- if(count($x) && count($r)) {
- $mailbox = construct_mailbox_name($r[0]);
- $password = '';
- openssl_private_decrypt(hex2bin($r[0]['pass']),$password,$x[0]['prvkey']);
- $mbox = email_connect($mailbox,$r[0]['user'],$password);
- if(! $mbox)
- logger('probe_url: email_connect failed.');
- unset($password);
- }
- if($mbox) {
- $msgs = email_poll($mbox,$orig_url);
- logger('probe_url: searching ' . $orig_url . ', ' . count($msgs) . ' messages found.', LOGGER_DEBUG);
- if(count($msgs)) {
- $addr = $orig_url;
- $network = NETWORK_MAIL;
- $name = substr($url,0,strpos($url,'@'));
- $phost = substr($url,strpos($url,'@')+1);
- $profile = 'http://' . $phost;
- // fix nick character range
- $vcard = array('fn' => $name, 'nick' => $name, 'photo' => avatar_img($url));
- $notify = 'smtp ' . random_string();
- $poll = 'email ' . random_string();
- $priority = 0;
- $x = email_msg_meta($mbox,$msgs[0]);
- if(stristr($x[0]->from,$orig_url))
- $adr = imap_rfc822_parse_adrlist($x[0]->from,'');
- elseif(stristr($x[0]->to,$orig_url))
- $adr = imap_rfc822_parse_adrlist($x[0]->to,'');
- if(isset($adr)) {
- foreach($adr as $feadr) {
- if((strcasecmp($feadr->mailbox,$name) == 0)
- &&(strcasecmp($feadr->host,$phost) == 0)
- && (strlen($feadr->personal))) {
-
- $personal = imap_mime_header_decode($feadr->personal);
- $vcard['fn'] = "";
- foreach($personal as $perspart)
- if ($perspart->charset != "default")
- $vcard['fn'] .= iconv($perspart->charset, 'UTF-8//IGNORE', $perspart->text);
- else
- $vcard['fn'] .= $perspart->text;
-
- $vcard['fn'] = notags($vcard['fn']);
- }
- }
- }
- }
- imap_close($mbox);
- }
- }
- }
- }
-
- if($mode == PROBE_NORMAL) {
- if(strlen($zot)) {
- $s = fetch_url($zot);
- if($s) {
- $j = json_decode($s);
- if($j) {
- $network = NETWORK_ZOT;
- $vcard = array(
- 'fn' => $j->fullname,
- 'nick' => $j->nickname,
- 'photo' => $j->photo
- );
- $profile = $j->url;
- $notify = $j->post;
- $pubkey = $j->pubkey;
- $poll = 'N/A';
- }
- }
- }
-
- if(strlen($dfrn)) {
- $ret = scrape_dfrn(($hcard) ? $hcard : $dfrn);
- if(is_array($ret) && x($ret,'dfrn-request')) {
- $network = NETWORK_DFRN;
- $request = $ret['dfrn-request'];
- $confirm = $ret['dfrn-confirm'];
- $notify = $ret['dfrn-notify'];
- $poll = $ret['dfrn-poll'];
-
- $vcard = array();
- $vcard['fn'] = $ret['fn'];
- $vcard['nick'] = $ret['nick'];
- $vcard['photo'] = $ret['photo'];
- }
- }
- }
-
- if($network !== NETWORK_ZOT && $network !== NETWORK_DFRN && $network !== NETWORK_MAIL) {
- if($has_lrdd)
- $network = NETWORK_OSTATUS;
- $priority = 0;
-
- if($hcard && ! $vcard) {
- $vcard = scrape_vcard($hcard);
-
- // Google doesn't use absolute url in profile photos
-
- if((x($vcard,'photo')) && substr($vcard['photo'],0,1) == '/') {
- $h = @parse_url($hcard);
- if($h)
- $vcard['photo'] = $h['scheme'] . '://' . $h['host'] . $vcard['photo'];
- }
-
- logger('probe_url: scrape_vcard: ' . print_r($vcard,true), LOGGER_DATA);
- }
-
- if($diaspora && $addr) {
- // Diaspora returns the name as the nick. As the nick will never be updated,
- // let's use the Diaspora nickname (the first part of the handle) as the nick instead
- $addr_parts = explode('@', $addr);
- $vcard['nick'] = $addr_parts[0];
- }
-
- if($twitter) {
- logger('twitter: setup');
- $tid = basename($url);
- $tapi = 'https://api.twitter.com/1/statuses/user_timeline.rss';
- if(intval($tid))
- $poll = $tapi . '?user_id=' . $tid;
- else
- $poll = $tapi . '?screen_name=' . $tid;
- $profile = 'http://twitter.com/#!/' . $tid;
- //$vcard['photo'] = 'https://api.twitter.com/1/users/profile_image/' . $tid;
- $vcard['photo'] = 'https://api.twitter.com/1/users/profile_image?screen_name=' . $tid . '&size=bigger';
- $vcard['nick'] = $tid;
- $vcard['fn'] = $tid;
- }
-
- if($lastfm) {
- $profile = $url;
- $poll = str_replace(array('www.','last.fm/'),array('','ws.audioscrobbler.com/1.0/'),$url) . '/recenttracks.rss';
- $vcard['nick'] = basename($url);
- $vcard['fn'] = $vcard['nick'] . t(' on Last.fm');
- $network = NETWORK_FEED;
- }
-
- if(! x($vcard,'fn'))
- if(x($vcard,'nick'))
- $vcard['fn'] = $vcard['nick'];
-
- $check_feed = false;
-
- if($twitter || ! $poll)
- $check_feed = true;
- if((! isset($vcard)) || (! x($vcard,'fn')) || (! $profile))
- $check_feed = true;
- if(($at_addr) && (! count($links)))
- $check_feed = false;
-
- if($check_feed) {
-
- $feedret = scrape_feed(($poll) ? $poll : $url);
- logger('probe_url: scrape_feed ' . (($poll)? $poll : $url) . ' returns: ' . print_r($feedret,true), LOGGER_DATA);
- if(count($feedret) && ($feedret['feed_atom'] || $feedret['feed_rss'])) {
- $poll = ((x($feedret,'feed_atom')) ? unamp($feedret['feed_atom']) : unamp($feedret['feed_rss']));
- if(! x($vcard))
- $vcard = array();
- }
-
- if(x($feedret,'photo') && (! x($vcard,'photo')))
- $vcard['photo'] = $feedret['photo'];
- require_once('library/simplepie/simplepie.inc');
- $feed = new SimplePie();
- $xml = fetch_url($poll);
-
- logger('probe_url: fetch feed: ' . $poll . ' returns: ' . $xml, LOGGER_DATA);
- $a = get_app();
-
- logger('probe_url: scrape_feed: headers: ' . $a->get_curl_headers(), LOGGER_DATA);
-
- $feed->set_raw_data($xml);
-
- $feed->init();
- if($feed->error())
- logger('probe_url: scrape_feed: Error parsing XML: ' . $feed->error());
-
-
- if(! x($vcard,'photo'))
- $vcard['photo'] = $feed->get_image_url();
- $author = $feed->get_author();
-
- if($author) {
- $vcard['fn'] = unxmlify(trim($author->get_name()));
- if(! $vcard['fn'])
- $vcard['fn'] = trim(unxmlify($author->get_email()));
- if(strpos($vcard['fn'],'@') !== false)
- $vcard['fn'] = substr($vcard['fn'],0,strpos($vcard['fn'],'@'));
- $email = unxmlify($author->get_email());
- if(! $profile && $author->get_link())
- $profile = trim(unxmlify($author->get_link()));
- if(! $vcard['photo']) {
- $rawtags = $feed->get_feed_tags( SIMPLEPIE_NAMESPACE_ATOM_10, 'author');
- if($rawtags) {
- $elems = $rawtags[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10];
- if((x($elems,'link')) && ($elems['link'][0]['attribs']['']['rel'] === 'photo'))
- $vcard['photo'] = $elems['link'][0]['attribs']['']['href'];
- }
- }
- }
- else {
- $item = $feed->get_item(0);
- if($item) {
- $author = $item->get_author();
- if($author) {
- $vcard['fn'] = trim(unxmlify($author->get_name()));
- if(! $vcard['fn'])
- $vcard['fn'] = trim(unxmlify($author->get_email()));
- if(strpos($vcard['fn'],'@') !== false)
- $vcard['fn'] = substr($vcard['fn'],0,strpos($vcard['fn'],'@'));
- $email = unxmlify($author->get_email());
- if(! $profile && $author->get_link())
- $profile = trim(unxmlify($author->get_link()));
- }
- if(! $vcard['photo']) {
- $rawmedia = $item->get_item_tags('http://search.yahoo.com/mrss/','thumbnail');
- if($rawmedia && $rawmedia[0]['attribs']['']['url'])
- $vcard['photo'] = unxmlify($rawmedia[0]['attribs']['']['url']);
- }
- if(! $vcard['photo']) {
- $rawtags = $item->get_item_tags( SIMPLEPIE_NAMESPACE_ATOM_10, 'author');
- if($rawtags) {
- $elems = $rawtags[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10];
- if((x($elems,'link')) && ($elems['link'][0]['attribs']['']['rel'] === 'photo'))
- $vcard['photo'] = $elems['link'][0]['attribs']['']['href'];
- }
- }
- }
- }
-
- if((! $vcard['photo']) && strlen($email))
- $vcard['photo'] = avatar_img($email);
- if($poll === $profile)
- $lnk = $feed->get_permalink();
- if(isset($lnk) && strlen($lnk))
- $profile = $lnk;
-
- if(! (x($vcard,'fn')))
- $vcard['fn'] = notags($feed->get_title());
- if(! (x($vcard,'fn')))
- $vcard['fn'] = notags($feed->get_description());
-
- if(strpos($vcard['fn'],'Twitter / ') !== false) {
- $vcard['fn'] = substr($vcard['fn'],strpos($vcard['fn'],'/')+1);
- $vcard['fn'] = trim($vcard['fn']);
- }
- if(! x($vcard,'nick')) {
- $vcard['nick'] = strtolower(notags(unxmlify($vcard['fn'])));
- if(strpos($vcard['nick'],' '))
- $vcard['nick'] = trim(substr($vcard['nick'],0,strpos($vcard['nick'],' ')));
- }
- if(! $network)
- $network = NETWORK_FEED;
- if(! $priority)
- $priority = 2;
- }
- }
-
- if(! x($vcard,'photo')) {
- $a = get_app();
- $vcard['photo'] = $a->get_baseurl() . '/images/person-175.jpg' ;
- }
-
- if(! $profile)
- $profile = $url;
-
- // No human could be associated with this link, use the URL as the contact name
-
- if(($network === NETWORK_FEED) && ($poll) && (! x($vcard,'fn')))
- $vcard['fn'] = $url;
-
- $vcard['fn'] = notags($vcard['fn']);
- $vcard['nick'] = str_replace(' ','',notags($vcard['nick']));
-
- $result['name'] = $vcard['fn'];
- $result['nick'] = $vcard['nick'];
- $result['url'] = $profile;
- $result['addr'] = $addr;
- $result['batch'] = $batch;
- $result['notify'] = $notify;
- $result['poll'] = $poll;
- $result['request'] = $request;
- $result['confirm'] = $confirm;
- $result['poco'] = $poco;
- $result['photo'] = $vcard['photo'];
- $result['priority'] = $priority;
- $result['network'] = $network;
- $result['alias'] = $alias;
- $result['pubkey'] = $pubkey;
-
- logger('probe_url: ' . print_r($result,true), LOGGER_DEBUG);
-
- return $result;
-}