Spaces:
Sleeping
Sleeping
/** | |
* SimplePie | |
* | |
* A PHP-Based RSS and Atom Feed Framework. | |
* Takes the hard work out of managing a complete RSS/Atom solution. | |
* | |
* Copyright (c) 2004-2016, Ryan Parman, Sam Sneddon, Ryan McCue, and contributors | |
* All rights reserved. | |
* | |
* Redistribution and use in source and binary forms, with or without modification, are | |
* permitted provided that the following conditions are met: | |
* | |
* * Redistributions of source code must retain the above copyright notice, this list of | |
* conditions and the following disclaimer. | |
* | |
* * Redistributions in binary form must reproduce the above copyright notice, this list | |
* of conditions and the following disclaimer in the documentation and/or other materials | |
* provided with the distribution. | |
* | |
* * Neither the name of the SimplePie Team nor the names of its contributors may be used | |
* to endorse or promote products derived from this software without specific prior | |
* written permission. | |
* | |
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS | |
* OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY | |
* AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDERS | |
* AND CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR | |
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR | |
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY | |
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR | |
* OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE | |
* POSSIBILITY OF SUCH DAMAGE. | |
* | |
* @package SimplePie | |
* @copyright 2004-2016 Ryan Parman, Sam Sneddon, Ryan McCue | |
* @author Ryan Parman | |
* @author Sam Sneddon | |
* @author Ryan McCue | |
* @link http://simplepie.org/ SimplePie | |
* @license http://www.opensource.org/licenses/bsd-license.php BSD License | |
*/ | |
/** | |
* Parses XML into something sane | |
* | |
* | |
* This class can be overloaded with {@see SimplePie::set_parser_class()} | |
* | |
* @package SimplePie | |
* @subpackage Parsing | |
*/ | |
class SimplePie_Parser | |
{ | |
var $error_code; | |
var $error_string; | |
var $current_line; | |
var $current_column; | |
var $current_byte; | |
var $separator = ' '; | |
var $namespace = array(''); | |
var $element = array(''); | |
var $xml_base = array(''); | |
var $xml_base_explicit = array(false); | |
var $xml_lang = array(''); | |
var $data = array(); | |
var $datas = array(array()); | |
var $current_xhtml_construct = -1; | |
var $encoding; | |
protected $registry; | |
public function set_registry(SimplePie_Registry $registry) | |
{ | |
$this->registry = $registry; | |
} | |
public function parse(&$data, $encoding, $url = '') | |
{ | |
if (class_exists('DOMXpath') && function_exists('Mf2\parse')) { | |
$doc = new DOMDocument(); | |
@$doc->loadHTML($data); | |
$xpath = new DOMXpath($doc); | |
// Check for both h-feed and h-entry, as both a feed with no entries | |
// and a list of entries without an h-feed wrapper are both valid. | |
$query = '//*[contains(concat(" ", @class, " "), " h-feed ") or '. | |
'contains(concat(" ", @class, " "), " h-entry ")]'; | |
$result = $xpath->query($query); | |
if ($result->length !== 0) { | |
return $this->parse_microformats($data, $url); | |
} | |
} | |
// Use UTF-8 if we get passed US-ASCII, as every US-ASCII character is a UTF-8 character | |
if (strtoupper($encoding) === 'US-ASCII') | |
{ | |
$this->encoding = 'UTF-8'; | |
} | |
else | |
{ | |
$this->encoding = $encoding; | |
} | |
// Strip BOM: | |
// UTF-32 Big Endian BOM | |
if (substr($data, 0, 4) === "\x00\x00\xFE\xFF") | |
{ | |
$data = substr($data, 4); | |
} | |
// UTF-32 Little Endian BOM | |
elseif (substr($data, 0, 4) === "\xFF\xFE\x00\x00") | |
{ | |
$data = substr($data, 4); | |
} | |
// UTF-16 Big Endian BOM | |
elseif (substr($data, 0, 2) === "\xFE\xFF") | |
{ | |
$data = substr($data, 2); | |
} | |
// UTF-16 Little Endian BOM | |
elseif (substr($data, 0, 2) === "\xFF\xFE") | |
{ | |
$data = substr($data, 2); | |
} | |
// UTF-8 BOM | |
elseif (substr($data, 0, 3) === "\xEF\xBB\xBF") | |
{ | |
$data = substr($data, 3); | |
} | |
if (substr($data, 0, 5) === '<?xml' && strspn(substr($data, 5, 1), "\x09\x0A\x0D\x20") && ($pos = strpos($data, '?>')) !== false) | |
{ | |
$declaration = $this->registry->create('XML_Declaration_Parser', array(substr($data, 5, $pos - 5))); | |
if ($declaration->parse()) | |
{ | |
$data = substr($data, $pos + 2); | |
$data = '<?xml version="' . $declaration->version . '" encoding="' . $encoding . '" standalone="' . (($declaration->standalone) ? 'yes' : 'no') . '"?>' ."\n". $this->declare_html_entities() . $data; | |
} | |
else | |
{ | |
$this->error_string = 'SimplePie bug! Please report this!'; | |
return false; | |
} | |
} | |
$return = true; | |
static $xml_is_sane = null; | |
if ($xml_is_sane === null) | |
{ | |
$parser_check = xml_parser_create(); | |
xml_parse_into_struct($parser_check, '<foo>&</foo>', $values); | |
xml_parser_free($parser_check); | |
$xml_is_sane = isset($values[0]['value']); | |
} | |
// Create the parser | |
if ($xml_is_sane) | |
{ | |
$xml = xml_parser_create_ns($this->encoding, $this->separator); | |
xml_parser_set_option($xml, XML_OPTION_SKIP_WHITE, 1); | |
xml_parser_set_option($xml, XML_OPTION_CASE_FOLDING, 0); | |
xml_set_object($xml, $this); | |
xml_set_character_data_handler($xml, 'cdata'); | |
xml_set_element_handler($xml, 'tag_open', 'tag_close'); | |
// Parse! | |
$wrapper = @is_writable(sys_get_temp_dir()) ? 'php://temp' : 'php://memory'; | |
if (($stream = fopen($wrapper, 'r+')) && | |
fwrite($stream, $data) && | |
rewind($stream)) | |
{ | |
//Parse by chunks not to use too much memory | |
do | |
{ | |
$stream_data = fread($stream, 1048576); | |
if (!xml_parse($xml, $stream_data === false ? '' : $stream_data, feof($stream))) | |
{ | |
$this->error_code = xml_get_error_code($xml); | |
$this->error_string = xml_error_string($this->error_code); | |
$return = false; | |
break; | |
} | |
} while (!feof($stream)); | |
fclose($stream); | |
} | |
else | |
{ | |
$return = false; | |
} | |
$this->current_line = xml_get_current_line_number($xml); | |
$this->current_column = xml_get_current_column_number($xml); | |
$this->current_byte = xml_get_current_byte_index($xml); | |
xml_parser_free($xml); | |
return $return; | |
} | |
libxml_clear_errors(); | |
$xml = new XMLReader(); | |
$xml->xml($data); | |
while (@$xml->read()) | |
{ | |
switch ($xml->nodeType) | |
{ | |
case constant('XMLReader::END_ELEMENT'): | |
if ($xml->namespaceURI !== '') | |
{ | |
$tagName = $xml->namespaceURI . $this->separator . $xml->localName; | |
} | |
else | |
{ | |
$tagName = $xml->localName; | |
} | |
$this->tag_close(null, $tagName); | |
break; | |
case constant('XMLReader::ELEMENT'): | |
$empty = $xml->isEmptyElement; | |
if ($xml->namespaceURI !== '') | |
{ | |
$tagName = $xml->namespaceURI . $this->separator . $xml->localName; | |
} | |
else | |
{ | |
$tagName = $xml->localName; | |
} | |
$attributes = array(); | |
while ($xml->moveToNextAttribute()) | |
{ | |
if ($xml->namespaceURI !== '') | |
{ | |
$attrName = $xml->namespaceURI . $this->separator . $xml->localName; | |
} | |
else | |
{ | |
$attrName = $xml->localName; | |
} | |
$attributes[$attrName] = $xml->value; | |
} | |
$this->tag_open(null, $tagName, $attributes); | |
if ($empty) | |
{ | |
$this->tag_close(null, $tagName); | |
} | |
break; | |
case constant('XMLReader::TEXT'): | |
case constant('XMLReader::CDATA'): | |
$this->cdata(null, $xml->value); | |
break; | |
} | |
} | |
if ($error = libxml_get_last_error()) | |
{ | |
$this->error_code = $error->code; | |
$this->error_string = $error->message; | |
$this->current_line = $error->line; | |
$this->current_column = $error->column; | |
return false; | |
} | |
return true; | |
} | |
public function get_error_code() | |
{ | |
return $this->error_code; | |
} | |
public function get_error_string() | |
{ | |
return $this->error_string; | |
} | |
public function get_current_line() | |
{ | |
return $this->current_line; | |
} | |
public function get_current_column() | |
{ | |
return $this->current_column; | |
} | |
public function get_current_byte() | |
{ | |
return $this->current_byte; | |
} | |
public function get_data() | |
{ | |
return $this->data; | |
} | |
public function tag_open($parser, $tag, $attributes) | |
{ | |
list($this->namespace[], $this->element[]) = $this->split_ns($tag); | |
$attribs = array(); | |
foreach ($attributes as $name => $value) | |
{ | |
list($attrib_namespace, $attribute) = $this->split_ns($name); | |
$attribs[$attrib_namespace][$attribute] = $value; | |
} | |
if (isset($attribs[SIMPLEPIE_NAMESPACE_XML]['base'])) | |
{ | |
$base = $this->registry->call('Misc', 'absolutize_url', array($attribs[SIMPLEPIE_NAMESPACE_XML]['base'], end($this->xml_base))); | |
if ($base !== false) | |
{ | |
$this->xml_base[] = $base; | |
$this->xml_base_explicit[] = true; | |
} | |
} | |
else | |
{ | |
$this->xml_base[] = end($this->xml_base); | |
$this->xml_base_explicit[] = end($this->xml_base_explicit); | |
} | |
if (isset($attribs[SIMPLEPIE_NAMESPACE_XML]['lang'])) | |
{ | |
$this->xml_lang[] = $attribs[SIMPLEPIE_NAMESPACE_XML]['lang']; | |
} | |
else | |
{ | |
$this->xml_lang[] = end($this->xml_lang); | |
} | |
if ($this->current_xhtml_construct >= 0) | |
{ | |
$this->current_xhtml_construct++; | |
if (end($this->namespace) === SIMPLEPIE_NAMESPACE_XHTML) | |
{ | |
$this->data['data'] .= '<' . end($this->element); | |
if (isset($attribs[''])) | |
{ | |
foreach ($attribs[''] as $name => $value) | |
{ | |
$this->data['data'] .= ' ' . $name . '="' . htmlspecialchars($value, ENT_COMPAT, $this->encoding) . '"'; | |
} | |
} | |
$this->data['data'] .= '>'; | |
} | |
} | |
else | |
{ | |
$this->datas[] =& $this->data; | |
$this->data =& $this->data['child'][end($this->namespace)][end($this->element)][]; | |
$this->data = array('data' => '', 'attribs' => $attribs, 'xml_base' => end($this->xml_base), 'xml_base_explicit' => end($this->xml_base_explicit), 'xml_lang' => end($this->xml_lang)); | |
if ((end($this->namespace) === SIMPLEPIE_NAMESPACE_ATOM_03 && in_array(end($this->element), array('title', 'tagline', 'copyright', 'info', 'summary', 'content')) && isset($attribs['']['mode']) && $attribs['']['mode'] === 'xml') | |
|| (end($this->namespace) === SIMPLEPIE_NAMESPACE_ATOM_10 && in_array(end($this->element), array('rights', 'subtitle', 'summary', 'info', 'title', 'content')) && isset($attribs['']['type']) && $attribs['']['type'] === 'xhtml') | |
|| (end($this->namespace) === SIMPLEPIE_NAMESPACE_RSS_20 && in_array(end($this->element), array('title'))) | |
|| (end($this->namespace) === SIMPLEPIE_NAMESPACE_RSS_090 && in_array(end($this->element), array('title'))) | |
|| (end($this->namespace) === SIMPLEPIE_NAMESPACE_RSS_10 && in_array(end($this->element), array('title')))) | |
{ | |
$this->current_xhtml_construct = 0; | |
} | |
} | |
} | |
public function cdata($parser, $cdata) | |
{ | |
if ($this->current_xhtml_construct >= 0) | |
{ | |
$this->data['data'] .= htmlspecialchars($cdata, ENT_QUOTES, $this->encoding); | |
} | |
else | |
{ | |
$this->data['data'] .= $cdata; | |
} | |
} | |
public function tag_close($parser, $tag) | |
{ | |
if ($this->current_xhtml_construct >= 0) | |
{ | |
$this->current_xhtml_construct--; | |
if (end($this->namespace) === SIMPLEPIE_NAMESPACE_XHTML && !in_array(end($this->element), array('area', 'base', 'basefont', 'br', 'col', 'frame', 'hr', 'img', 'input', 'isindex', 'link', 'meta', 'param'))) | |
{ | |
$this->data['data'] .= '</' . end($this->element) . '>'; | |
} | |
} | |
if ($this->current_xhtml_construct === -1) | |
{ | |
$this->data =& $this->datas[count($this->datas) - 1]; | |
array_pop($this->datas); | |
} | |
array_pop($this->element); | |
array_pop($this->namespace); | |
array_pop($this->xml_base); | |
array_pop($this->xml_base_explicit); | |
array_pop($this->xml_lang); | |
} | |
public function split_ns($string) | |
{ | |
static $cache = array(); | |
if (!isset($cache[$string])) | |
{ | |
if ($pos = strpos($string, $this->separator)) | |
{ | |
static $separator_length; | |
if (!$separator_length) | |
{ | |
$separator_length = strlen($this->separator); | |
} | |
$namespace = substr($string, 0, $pos); | |
$local_name = substr($string, $pos + $separator_length); | |
if (strtolower($namespace) === SIMPLEPIE_NAMESPACE_ITUNES) | |
{ | |
$namespace = SIMPLEPIE_NAMESPACE_ITUNES; | |
} | |
// Normalize the Media RSS namespaces | |
if ($namespace === SIMPLEPIE_NAMESPACE_MEDIARSS_WRONG || | |
$namespace === SIMPLEPIE_NAMESPACE_MEDIARSS_WRONG2 || | |
$namespace === SIMPLEPIE_NAMESPACE_MEDIARSS_WRONG3 || | |
$namespace === SIMPLEPIE_NAMESPACE_MEDIARSS_WRONG4 || | |
$namespace === SIMPLEPIE_NAMESPACE_MEDIARSS_WRONG5 ) | |
{ | |
$namespace = SIMPLEPIE_NAMESPACE_MEDIARSS; | |
} | |
$cache[$string] = array($namespace, $local_name); | |
} | |
else | |
{ | |
$cache[$string] = array('', $string); | |
} | |
} | |
return $cache[$string]; | |
} | |
private function parse_hcard($data, $category = false) { | |
$name = ''; | |
$link = ''; | |
// Check if h-card is set and pass that information on in the link. | |
if (isset($data['type']) && in_array('h-card', $data['type'])) { | |
if (isset($data['properties']['name'][0])) { | |
$name = $data['properties']['name'][0]; | |
} | |
if (isset($data['properties']['url'][0])) { | |
$link = $data['properties']['url'][0]; | |
if ($name === '') { | |
$name = $link; | |
} | |
else { | |
// can't have commas in categories. | |
$name = str_replace(',', '', $name); | |
} | |
$person_tag = $category ? '<span class="person-tag"></span>' : ''; | |
return '<a class="h-card" href="'.$link.'">'.$person_tag.$name.'</a>'; | |
} | |
} | |
return isset($data['value']) ? $data['value'] : ''; | |
} | |
private function parse_microformats(&$data, $url) { | |
$feed_title = ''; | |
$feed_author = NULL; | |
$author_cache = array(); | |
$items = array(); | |
$entries = array(); | |
$mf = Mf2\parse($data, $url); | |
// First look for an h-feed. | |
$h_feed = array(); | |
foreach ($mf['items'] as $mf_item) { | |
if (in_array('h-feed', $mf_item['type'])) { | |
$h_feed = $mf_item; | |
break; | |
} | |
// Also look for h-feed or h-entry in the children of each top level item. | |
if (!isset($mf_item['children'][0]['type'])) continue; | |
if (in_array('h-feed', $mf_item['children'][0]['type'])) { | |
$h_feed = $mf_item['children'][0]; | |
// In this case the parent of the h-feed may be an h-card, so use it as | |
// the feed_author. | |
if (in_array('h-card', $mf_item['type'])) $feed_author = $mf_item; | |
break; | |
} | |
else if (in_array('h-entry', $mf_item['children'][0]['type'])) { | |
$entries = $mf_item['children']; | |
// In this case the parent of the h-entry list may be an h-card, so use | |
// it as the feed_author. | |
if (in_array('h-card', $mf_item['type'])) $feed_author = $mf_item; | |
break; | |
} | |
} | |
if (isset($h_feed['children'])) { | |
$entries = $h_feed['children']; | |
// Also set the feed title and store author from the h-feed if available. | |
if (isset($mf['items'][0]['properties']['name'][0])) { | |
$feed_title = $mf['items'][0]['properties']['name'][0]; | |
} | |
if (isset($mf['items'][0]['properties']['author'][0])) { | |
$feed_author = $mf['items'][0]['properties']['author'][0]; | |
} | |
} | |
else if (count($entries) === 0) { | |
$entries = $mf['items']; | |
} | |
for ($i = 0; $i < count($entries); $i++) { | |
$entry = $entries[$i]; | |
if (in_array('h-entry', $entry['type'])) { | |
$item = array(); | |
$title = ''; | |
$description = ''; | |
if (isset($entry['properties']['url'][0])) { | |
$link = $entry['properties']['url'][0]; | |
if (isset($link['value'])) $link = $link['value']; | |
$item['link'] = array(array('data' => $link)); | |
} | |
if (isset($entry['properties']['uid'][0])) { | |
$guid = $entry['properties']['uid'][0]; | |
if (isset($guid['value'])) $guid = $guid['value']; | |
$item['guid'] = array(array('data' => $guid)); | |
} | |
if (isset($entry['properties']['name'][0])) { | |
$title = $entry['properties']['name'][0]; | |
if (isset($title['value'])) $title = $title['value']; | |
$item['title'] = array(array('data' => $title)); | |
} | |
if (isset($entry['properties']['author'][0]) || isset($feed_author)) { | |
// author is a special case, it can be plain text or an h-card array. | |
// If it's plain text it can also be a url that should be followed to | |
// get the actual h-card. | |
$author = isset($entry['properties']['author'][0]) ? | |
$entry['properties']['author'][0] : $feed_author; | |
if (!is_string($author)) { | |
$author = $this->parse_hcard($author); | |
} | |
else if (strpos($author, 'http') === 0) { | |
if (isset($author_cache[$author])) { | |
$author = $author_cache[$author]; | |
} | |
else { | |
$mf = Mf2\fetch($author); | |
foreach ($mf['items'] as $hcard) { | |
// Only interested in an h-card by itself in this case. | |
if (!in_array('h-card', $hcard['type'])) { | |
continue; | |
} | |
// It must have a url property matching what we fetched. | |
if (!isset($hcard['properties']['url']) || | |
!(in_array($author, $hcard['properties']['url']))) { | |
continue; | |
} | |
// Save parse_hcard the trouble of finding the correct url. | |
$hcard['properties']['url'][0] = $author; | |
// Cache this h-card for the next h-entry to check. | |
$author_cache[$author] = $this->parse_hcard($hcard); | |
$author = $author_cache[$author]; | |
break; | |
} | |
} | |
} | |
$item['author'] = array(array('data' => $author)); | |
} | |
if (isset($entry['properties']['photo'][0])) { | |
// If a photo is also in content, don't need to add it again here. | |
$content = ''; | |
if (isset($entry['properties']['content'][0]['html'])) { | |
$content = $entry['properties']['content'][0]['html']; | |
} | |
$photo_list = array(); | |
for ($j = 0; $j < count($entry['properties']['photo']); $j++) { | |
$photo = $entry['properties']['photo'][$j]; | |
if (!empty($photo) && strpos($content, $photo) === false) { | |
$photo_list[] = $photo; | |
} | |
} | |
// When there's more than one photo show the first and use a lightbox. | |
// Need a permanent, unique name for the image set, but don't have | |
// anything unique except for the content itself, so use that. | |
$count = count($photo_list); | |
if ($count > 1) { | |
$image_set_id = preg_replace('/[[:^alnum:]]/', '', $photo_list[0]); | |
$description = '<p>'; | |
for ($j = 0; $j < $count; $j++) { | |
$hidden = $j === 0 ? '' : 'class="hidden" '; | |
$description .= '<a href="'.$photo_list[$j].'" '.$hidden. | |
'data-lightbox="image-set-'.$image_set_id.'">'. | |
'<img src="'.$photo_list[$j].'"></a>'; | |
} | |
$description .= '<br><b>'.$count.' photos</b></p>'; | |
} | |
else if ($count == 1) { | |
$description = '<p><img src="'.$photo_list[0].'"></p>'; | |
} | |
} | |
if (isset($entry['properties']['content'][0]['html'])) { | |
// e-content['value'] is the same as p-name when they are on the same | |
// element. Use this to replace title with a strip_tags version so | |
// that alt text from images is not included in the title. | |
if ($entry['properties']['content'][0]['value'] === $title) { | |
$title = strip_tags($entry['properties']['content'][0]['html']); | |
$item['title'] = array(array('data' => $title)); | |
} | |
$description .= $entry['properties']['content'][0]['html']; | |
if (isset($entry['properties']['in-reply-to'][0])) { | |
$in_reply_to = ''; | |
if (is_string($entry['properties']['in-reply-to'][0])) { | |
$in_reply_to = $entry['properties']['in-reply-to'][0]; | |
} | |
else if (isset($entry['properties']['in-reply-to'][0]['value'])) { | |
$in_reply_to = $entry['properties']['in-reply-to'][0]['value']; | |
} | |
if ($in_reply_to !== '') { | |
$description .= '<p><span class="in-reply-to"></span> '. | |
'<a href="'.$in_reply_to.'">'.$in_reply_to.'</a><p>'; | |
} | |
} | |
$item['description'] = array(array('data' => $description)); | |
} | |
if (isset($entry['properties']['category'])) { | |
$category_csv = ''; | |
// Categories can also contain h-cards. | |
foreach ($entry['properties']['category'] as $category) { | |
if ($category_csv !== '') $category_csv .= ', '; | |
if (is_string($category)) { | |
// Can't have commas in categories. | |
$category_csv .= str_replace(',', '', $category); | |
} | |
else { | |
$category_csv .= $this->parse_hcard($category, true); | |
} | |
} | |
$item['category'] = array(array('data' => $category_csv)); | |
} | |
if (isset($entry['properties']['published'][0])) { | |
$timestamp = strtotime($entry['properties']['published'][0]); | |
$pub_date = date('F j Y g:ia', $timestamp).' GMT'; | |
$item['pubDate'] = array(array('data' => $pub_date)); | |
} | |
// The title and description are set to the empty string to represent | |
// a deleted item (which also makes it an invalid rss item). | |
if (isset($entry['properties']['deleted'][0])) { | |
$item['title'] = array(array('data' => '')); | |
$item['description'] = array(array('data' => '')); | |
} | |
$items[] = array('child' => array('' => $item)); | |
} | |
} | |
// Mimic RSS data format when storing microformats. | |
$link = array(array('data' => $url)); | |
$image = ''; | |
if (!is_string($feed_author) && | |
isset($feed_author['properties']['photo'][0])) { | |
$image = array(array('child' => array('' => array('url' => | |
array(array('data' => $feed_author['properties']['photo'][0])))))); | |
} | |
// Use the name given for the h-feed, or get the title from the html. | |
if ($feed_title !== '') { | |
$feed_title = array(array('data' => htmlspecialchars($feed_title))); | |
} | |
else if ($position = strpos($data, '<title>')) { | |
$start = $position < 200 ? 0 : $position - 200; | |
$check = substr($data, $start, 400); | |
$matches = array(); | |
if (preg_match('/<title>(.+)<\/title>/', $check, $matches)) { | |
$feed_title = array(array('data' => htmlspecialchars($matches[1]))); | |
} | |
} | |
$channel = array('channel' => array(array('child' => array('' => | |
array('link' => $link, 'image' => $image, 'title' => $feed_title, | |
'item' => $items))))); | |
$rss = array(array('attribs' => array('' => array('version' => '2.0')), | |
'child' => array('' => $channel))); | |
$this->data = array('child' => array('' => array('rss' => $rss))); | |
return true; | |
} | |
private function declare_html_entities() { | |
// This is required because the RSS specification says that entity-encoded | |
// html is allowed, but the xml specification says they must be declared. | |
return '<!DOCTYPE html [ <!ENTITY nbsp " "> <!ENTITY iexcl "¡"> <!ENTITY cent "¢"> <!ENTITY pound "£"> <!ENTITY curren "¤"> <!ENTITY yen "¥"> <!ENTITY brvbar "¦"> <!ENTITY sect "§"> <!ENTITY uml "¨"> <!ENTITY copy "©"> <!ENTITY ordf "ª"> <!ENTITY laquo "«"> <!ENTITY not "¬"> <!ENTITY shy "­"> <!ENTITY reg "®"> <!ENTITY macr "¯"> <!ENTITY deg "°"> <!ENTITY plusmn "±"> <!ENTITY sup2 "²"> <!ENTITY sup3 "³"> <!ENTITY acute "´"> <!ENTITY micro "µ"> <!ENTITY para "¶"> <!ENTITY middot "·"> <!ENTITY cedil "¸"> <!ENTITY sup1 "¹"> <!ENTITY ordm "º"> <!ENTITY raquo "»"> <!ENTITY frac14 "¼"> <!ENTITY frac12 "½"> <!ENTITY frac34 "¾"> <!ENTITY iquest "¿"> <!ENTITY Agrave "À"> <!ENTITY Aacute "Á"> <!ENTITY Acirc "Â"> <!ENTITY Atilde "Ã"> <!ENTITY Auml "Ä"> <!ENTITY Aring "Å"> <!ENTITY AElig "Æ"> <!ENTITY Ccedil "Ç"> <!ENTITY Egrave "È"> <!ENTITY Eacute "É"> <!ENTITY Ecirc "Ê"> <!ENTITY Euml "Ë"> <!ENTITY Igrave "Ì"> <!ENTITY Iacute "Í"> <!ENTITY Icirc "Î"> <!ENTITY Iuml "Ï"> <!ENTITY ETH "Ð"> <!ENTITY Ntilde "Ñ"> <!ENTITY Ograve "Ò"> <!ENTITY Oacute "Ó"> <!ENTITY Ocirc "Ô"> <!ENTITY Otilde "Õ"> <!ENTITY Ouml "Ö"> <!ENTITY times "×"> <!ENTITY Oslash "Ø"> <!ENTITY Ugrave "Ù"> <!ENTITY Uacute "Ú"> <!ENTITY Ucirc "Û"> <!ENTITY Uuml "Ü"> <!ENTITY Yacute "Ý"> <!ENTITY THORN "Þ"> <!ENTITY szlig "ß"> <!ENTITY agrave "à"> <!ENTITY aacute "á"> <!ENTITY acirc "â"> <!ENTITY atilde "ã"> <!ENTITY auml "ä"> <!ENTITY aring "å"> <!ENTITY aelig "æ"> <!ENTITY ccedil "ç"> <!ENTITY egrave "è"> <!ENTITY eacute "é"> <!ENTITY ecirc "ê"> <!ENTITY euml "ë"> <!ENTITY igrave "ì"> <!ENTITY iacute "í"> <!ENTITY icirc "î"> <!ENTITY iuml "ï"> <!ENTITY eth "ð"> <!ENTITY ntilde "ñ"> <!ENTITY ograve "ò"> <!ENTITY oacute "ó"> <!ENTITY ocirc "ô"> <!ENTITY otilde "õ"> <!ENTITY ouml "ö"> <!ENTITY divide "÷"> <!ENTITY oslash "ø"> <!ENTITY ugrave "ù"> <!ENTITY uacute "ú"> <!ENTITY ucirc "û"> <!ENTITY uuml "ü"> <!ENTITY yacute "ý"> <!ENTITY thorn "þ"> <!ENTITY yuml "ÿ"> <!ENTITY OElig "Œ"> <!ENTITY oelig "œ"> <!ENTITY Scaron "Š"> <!ENTITY scaron "š"> <!ENTITY Yuml "Ÿ"> <!ENTITY fnof "ƒ"> <!ENTITY circ "ˆ"> <!ENTITY tilde "˜"> <!ENTITY Alpha "Α"> <!ENTITY Beta "Β"> <!ENTITY Gamma "Γ"> <!ENTITY Epsilon "Ε"> <!ENTITY Zeta "Ζ"> <!ENTITY Eta "Η"> <!ENTITY Theta "Θ"> <!ENTITY Iota "Ι"> <!ENTITY Kappa "Κ"> <!ENTITY Lambda "Λ"> <!ENTITY Mu "Μ"> <!ENTITY Nu "Ν"> <!ENTITY Xi "Ξ"> <!ENTITY Omicron "Ο"> <!ENTITY Pi "Π"> <!ENTITY Rho "Ρ"> <!ENTITY Sigma "Σ"> <!ENTITY Tau "Τ"> <!ENTITY Upsilon "Υ"> <!ENTITY Phi "Φ"> <!ENTITY Chi "Χ"> <!ENTITY Psi "Ψ"> <!ENTITY Omega "Ω"> <!ENTITY alpha "α"> <!ENTITY beta "β"> <!ENTITY gamma "γ"> <!ENTITY delta "δ"> <!ENTITY epsilon "ε"> <!ENTITY zeta "ζ"> <!ENTITY eta "η"> <!ENTITY theta "θ"> <!ENTITY iota "ι"> <!ENTITY kappa "κ"> <!ENTITY lambda "λ"> <!ENTITY mu "μ"> <!ENTITY nu "ν"> <!ENTITY xi "ξ"> <!ENTITY omicron "ο"> <!ENTITY pi "π"> <!ENTITY rho "ρ"> <!ENTITY sigmaf "ς"> <!ENTITY sigma "σ"> <!ENTITY tau "τ"> <!ENTITY upsilon "υ"> <!ENTITY phi "φ"> <!ENTITY chi "χ"> <!ENTITY psi "ψ"> <!ENTITY omega "ω"> <!ENTITY thetasym "ϑ"> <!ENTITY upsih "ϒ"> <!ENTITY piv "ϖ"> <!ENTITY ensp " "> <!ENTITY emsp " "> <!ENTITY thinsp " "> <!ENTITY zwnj "‌"> <!ENTITY zwj "‍"> <!ENTITY lrm "‎"> <!ENTITY rlm "‏"> <!ENTITY ndash "–"> <!ENTITY mdash "—"> <!ENTITY lsquo "‘"> <!ENTITY rsquo "’"> <!ENTITY sbquo "‚"> <!ENTITY ldquo "“"> <!ENTITY rdquo "”"> <!ENTITY bdquo "„"> <!ENTITY dagger "†"> <!ENTITY Dagger "‡"> <!ENTITY bull "•"> <!ENTITY hellip "…"> <!ENTITY permil "‰"> <!ENTITY prime "′"> <!ENTITY Prime "″"> <!ENTITY lsaquo "‹"> <!ENTITY rsaquo "›"> <!ENTITY oline "‾"> <!ENTITY frasl "⁄"> <!ENTITY euro "€"> <!ENTITY image "ℑ"> <!ENTITY weierp "℘"> <!ENTITY real "ℜ"> <!ENTITY trade "™"> <!ENTITY alefsym "ℵ"> <!ENTITY larr "←"> <!ENTITY uarr "↑"> <!ENTITY rarr "→"> <!ENTITY darr "↓"> <!ENTITY harr "↔"> <!ENTITY crarr "↵"> <!ENTITY lArr "⇐"> <!ENTITY uArr "⇑"> <!ENTITY rArr "⇒"> <!ENTITY dArr "⇓"> <!ENTITY hArr "⇔"> <!ENTITY forall "∀"> <!ENTITY part "∂"> <!ENTITY exist "∃"> <!ENTITY empty "∅"> <!ENTITY nabla "∇"> <!ENTITY isin "∈"> <!ENTITY notin "∉"> <!ENTITY ni "∋"> <!ENTITY prod "∏"> <!ENTITY sum "∑"> <!ENTITY minus "−"> <!ENTITY lowast "∗"> <!ENTITY radic "√"> <!ENTITY prop "∝"> <!ENTITY infin "∞"> <!ENTITY ang "∠"> <!ENTITY and "∧"> <!ENTITY or "∨"> <!ENTITY cap "∩"> <!ENTITY cup "∪"> <!ENTITY int "∫"> <!ENTITY there4 "∴"> <!ENTITY sim "∼"> <!ENTITY cong "≅"> <!ENTITY asymp "≈"> <!ENTITY ne "≠"> <!ENTITY equiv "≡"> <!ENTITY le "≤"> <!ENTITY ge "≥"> <!ENTITY sub "⊂"> <!ENTITY sup "⊃"> <!ENTITY nsub "⊄"> <!ENTITY sube "⊆"> <!ENTITY supe "⊇"> <!ENTITY oplus "⊕"> <!ENTITY otimes "⊗"> <!ENTITY perp "⊥"> <!ENTITY sdot "⋅"> <!ENTITY lceil "⌈"> <!ENTITY rceil "⌉"> <!ENTITY lfloor "⌊"> <!ENTITY rfloor "⌋"> <!ENTITY lang "〈"> <!ENTITY rang "〉"> <!ENTITY loz "◊"> <!ENTITY spades "♠"> <!ENTITY clubs "♣"> <!ENTITY hearts "♥"> <!ENTITY diams "♦"> ]>'; | |
} | |
} | |