streams/include/text.php

2037 lines
53 KiB
PHP
Raw Normal View History

2013-02-26 01:09:40 +00:00
<?php /** @file */
require_once("include/template_processor.php");
2014-07-14 06:00:03 +00:00
require_once("include/smarty.php");
/**
* This is our template processor
*
* @param string|FriendicaSmarty $s the string requiring macro substitution,
* or an instance of FriendicaSmarty
* @param array $r key value pairs (search => replace)
* @return string substituted string
*/
function replace_macros($s,$r) {
2013-01-06 21:42:51 +00:00
$a = get_app();
$arr = array('template' => $s, 'params' => $r);
call_hooks('replace_macros', $arr);
$t = $a->template_engine();
$output = $t->replace_macros($arr['template'],$arr['params']);
2013-01-06 21:42:51 +00:00
return $output;
2013-02-27 02:26:33 +00:00
}
// random string, there are 86 characters max in text mode, 128 for hex
// output is urlsafe
define('RANDOM_STRING_HEX', 0x00 );
define('RANDOM_STRING_TEXT', 0x01 );
2013-02-27 02:26:33 +00:00
function random_string($size = 64,$type = RANDOM_STRING_HEX) {
// generate a bit of entropy and run it through the whirlpool
$s = hash('whirlpool', (string) rand() . uniqid(rand(),true) . (string) rand(),(($type == RANDOM_STRING_TEXT) ? true : false));
$s = (($type == RANDOM_STRING_TEXT) ? str_replace("\n","",base64url_encode($s,true)) : $s);
return(substr($s,0,$size));
2013-02-27 02:26:33 +00:00
}
/**
* This is our primary input filter.
*
* The high bit hack only involved some old IE browser, forget which (IE5/Mac?)
* that had an XSS attack vector due to stripping the high-bit on an 8-bit character
* after cleansing, and angle chars with the high bit set could get through as markup.
*
* This is now disabled because it was interfering with some legitimate unicode sequences
* and hopefully there aren't a lot of those browsers left.
*
* Use this on any text input where angle chars are not valid or permitted
* They will be replaced with safer brackets. This may be filtered further
* if these are not allowed either.
*
* @param string $string Input string
* @return string Filtered string
*/
2013-02-27 02:26:33 +00:00
function notags($string) {
return(str_replace(array("<",">"), array('[',']'), $string));
// High-bit filter no longer used
// return(str_replace(array("<",">","\xBA","\xBC","\xBE"), array('[',']','','',''), $string));
2013-02-27 02:26:33 +00:00
}
// use this on "body" or "content" input where angle chars shouldn't be removed,
// and allow them to be safely displayed.
2013-02-27 02:26:33 +00:00
/**
* use this on "body" or "content" input where angle chars shouldn't be removed,
* and allow them to be safely displayed.
* @param string $string
* @return string
*/
function escape_tags($string) {
2012-12-07 00:12:45 +00:00
return(htmlspecialchars($string, ENT_COMPAT, 'UTF-8', false));
2012-08-16 03:01:55 +00:00
2013-02-27 02:26:33 +00:00
}
function z_input_filter($channel_id,$s,$type = 'text/bbcode') {
if($type === 'text/bbcode')
return escape_tags($s);
if($type === 'text/markdown')
return escape_tags($s);
if($type == 'text/plain')
return escape_tags($s);
$r = q("select account_id, account_roles from account left join channel on channel_account_id = account_id where channel_id = %d limit 1",
intval($channel_id)
);
2013-09-02 23:01:29 +00:00
if($r && ($r[0]['account_roles'] & ACCOUNT_ROLE_ALLOWCODE)) {
if(local_user() && (get_account_id() == $r[0]['account_id'])) {
return $s;
}
}
if($type === 'text/html')
return purify_html($s);
return escape_tags($s);
}
function purify_html($s) {
require_once('library/HTMLPurifier.auto.php');
require_once('include/html2bbcode.php');
// FIXME this function has html output, not bbcode - so safely purify these
// $s = html2bb_video($s);
// $s = oembed_html2bbcode($s);
$config = HTMLPurifier_Config::createDefault();
$config->set('Cache.DefinitionImpl', null);
$config->set('Attr.EnableID', true);
$purifier = new HTMLPurifier($config);
return $purifier->purify($s);
}
2013-12-05 05:10:03 +00:00
// generate a string that's random, but usually pronounceable.
// used to generate initial passwords
2013-02-27 02:26:33 +00:00
/**
* generate a string that's random, but usually pronounceable.
* used to generate initial passwords
* @param int $len
* @return string
*/
function autoname($len) {
2012-04-07 08:04:45 +00:00
if($len <= 0)
2012-04-06 01:44:36 +00:00
return '';
$vowels = array('a','a','ai','au','e','e','e','ee','ea','i','ie','o','ou','u');
if(mt_rand(0,5) == 4)
$vowels[] = 'y';
$cons = array(
'b','bl','br',
'c','ch','cl','cr',
'd','dr',
'f','fl','fr',
'g','gh','gl','gr',
'h',
'j',
'k','kh','kl','kr',
'l',
'm',
'n',
'p','ph','pl','pr',
'qu',
'r','rh',
's','sc','sh','sm','sp','st',
't','th','tr',
'v',
'w','wh',
'x',
'z','zh'
);
$midcons = array('ck','ct','gn','ld','lf','lm','lt','mb','mm', 'mn','mp',
'nd','ng','nk','nt','rn','rp','rt');
$noend = array('bl', 'br', 'cl','cr','dr','fl','fr','gl','gr',
'kh', 'kl','kr','mn','pl','pr','rh','tr','qu','wh');
$start = mt_rand(0,2);
if($start == 0)
$table = $vowels;
else
$table = $cons;
$word = '';
for ($x = 0; $x < $len; $x ++) {
$r = mt_rand(0,count($table) - 1);
$word .= $table[$r];
if($table == $vowels)
$table = array_merge($cons,$midcons);
else
$table = $vowels;
}
$word = substr($word,0,$len);
foreach($noend as $noe) {
if((strlen($word) > 2) && (substr($word,-2) == $noe)) {
$word = substr($word,0,-1);
break;
}
}
if(substr($word,-1) == 'q')
$word = substr($word,0,-1);
return $word;
2013-02-27 02:26:33 +00:00
}
// escape text ($str) for XML transport
// returns escaped text.
2013-02-27 02:26:33 +00:00
/**
* escape text ($str) for XML transport
* @param string $str
* @return string Escaped text.
*/
function xmlify($str) {
$buffer = '';
$len = mb_strlen($str);
for($x = 0; $x < $len; $x ++) {
$char = mb_substr($str,$x,1);
switch( $char ) {
case "\r" :
break;
case "&" :
$buffer .= '&amp;';
break;
case "'" :
$buffer .= '&apos;';
break;
case "\"" :
$buffer .= '&quot;';
break;
case '<' :
$buffer .= '&lt;';
break;
case '>' :
$buffer .= '&gt;';
break;
case "\n" :
$buffer .= "\n";
break;
default :
$buffer .= $char;
break;
}
}
$buffer = trim($buffer);
return($buffer);
2013-02-27 02:26:33 +00:00
}
// undo an xmlify
// pass xml escaped text ($s), returns unescaped text
2013-02-27 02:26:33 +00:00
function unxmlify($s) {
$ret = str_replace('&amp;','&', $s);
$ret = str_replace(array('&lt;','&gt;','&quot;','&apos;'),array('<','>','"',"'"),$ret);
return $ret;
2013-02-27 02:26:33 +00:00
}
// convenience wrapper, reverse the operation "bin2hex"
// This is a built-in function in php >= 5.4
2013-02-27 02:26:33 +00:00
if(! function_exists('hex2bin')) {
function hex2bin($s) {
2011-10-21 02:11:34 +00:00
if(! (is_string($s) && strlen($s)))
return '';
if(strlen($s) & 1) {
logger('hex2bin: illegal hex string: ' . $s);
return $s;
}
if(! ctype_xdigit($s)) {
return($s);
}
return(pack("H*",$s));
}}
// Automatic pagination.
// To use, get the count of total items.
// Then call $a->set_pager_total($number_items);
// Optionally call $a->set_pager_itemspage($n) to the number of items to display on each page
// Then call paginate($a) after the end of the display loop to insert the pager block on the page
// (assuming there are enough items to paginate).
// When using with SQL, the setting LIMIT %d, %d => $a->pager['start'],$a->pager['itemspage']
// will limit the results to the correct items for the current page.
// The actual page handling is then accomplished at the application layer.
2013-02-27 02:26:33 +00:00
function paginate(&$a) {
$o = '';
$stripped = preg_replace('/(&page=[0-9]*)/','',$a->query_string);
2012-04-02 07:45:45 +00:00
2012-11-09 03:07:19 +00:00
// $stripped = preg_replace('/&zid=(.*?)([\?&]|$)/ism','',$stripped);
2012-04-02 07:45:45 +00:00
$stripped = str_replace('q=','',$stripped);
$stripped = trim($stripped,'/');
$pagenum = $a->pager['page'];
$url = $a->get_baseurl() . '/' . $stripped;
if($a->pager['total'] > $a->pager['itemspage']) {
$o .= '<div class="pager">';
if($a->pager['page'] != 1)
$o .= '<span class="pager_prev">'."<a href=\"$url".'&page='.($a->pager['page'] - 1).'">' . t('prev') . '</a></span> ';
$o .= "<span class=\"pager_first\"><a href=\"$url"."&page=1\">" . t('first') . "</a></span> ";
$numpages = $a->pager['total'] / $a->pager['itemspage'];
$numstart = 1;
$numstop = $numpages;
if($numpages > 14) {
$numstart = (($pagenum > 7) ? ($pagenum - 7) : 1);
$numstop = (($pagenum > ($numpages - 7)) ? $numpages : ($numstart + 14));
}
for($i = $numstart; $i <= $numstop; $i++){
if($i == $a->pager['page'])
$o .= '<span class="pager_current">'.(($i < 10) ? '&nbsp;'.$i : $i);
else
$o .= "<span class=\"pager_n\"><a href=\"$url"."&page=$i\">".(($i < 10) ? '&nbsp;'.$i : $i)."</a>";
$o .= '</span> ';
}
if(($a->pager['total'] % $a->pager['itemspage']) != 0) {
if($i == $a->pager['page'])
$o .= '<span class="pager_current">'.(($i < 10) ? '&nbsp;'.$i : $i);
else
$o .= "<span class=\"pager_n\"><a href=\"$url"."&page=$i\">".(($i < 10) ? '&nbsp;'.$i : $i)."</a>";
$o .= '</span> ';
}
$lastpage = (($numpages > intval($numpages)) ? intval($numpages)+1 : $numpages);
$o .= "<span class=\"pager_last\"><a href=\"$url"."&page=$lastpage\">" . t('last') . "</a></span> ";
if(($a->pager['total'] - ($a->pager['itemspage'] * $a->pager['page'])) > 0)
$o .= '<span class="pager_next">'."<a href=\"$url"."&page=".($a->pager['page'] + 1).'">' . t('next') . '</a></span>';
$o .= '</div>'."\r\n";
}
return $o;
2013-02-27 02:26:33 +00:00
}
function alt_pager(&$a, $i, $more = '', $less = '') {
$o = '';
if(! $more)
$more = t('older');
if(! $less)
$less = t('newer');
2012-07-15 03:39:46 +00:00
$stripped = preg_replace('/(&page=[0-9]*)/','',$a->query_string);
$stripped = str_replace('q=','',$stripped);
$stripped = trim($stripped,'/');
$pagenum = $a->pager['page'];
$url = $a->get_baseurl() . '/' . $stripped;
2012-07-15 03:39:46 +00:00
2013-06-04 01:16:00 +00:00
return replace_macros(get_markup_template('alt_pager.tpl'),array(
'$has_less' => (($a->pager['page'] > 1) ? true : false),
'$has_more' => (($i > 0 && $i >= $a->pager['itemspage']) ? true : false),
2013-06-04 01:16:00 +00:00
'$less' => $less,
'$more' => $more,
'$url' => $url,
'$prevpage' => $a->pager['page'] - 1,
'$nextpage' => $a->pager['page'] + 1,
));
2012-07-15 03:39:46 +00:00
2013-02-27 02:26:33 +00:00
}
2012-07-15 03:39:46 +00:00
// Turn user/group ACLs stored as angle bracketed text into arrays
2013-02-27 02:26:33 +00:00
function expand_acl($s) {
// turn string array of angle-bracketed elements into string array
// e.g. "<123xyz><246qyo><sxo33e>" => array(123xyz,246qyo,sxo33e);
$ret = array();
if(strlen($s)) {
$t = str_replace('<','',$s);
$a = explode('>',$t);
foreach($a as $aa) {
if($aa)
$ret[] = $aa;
}
}
return $ret;
2013-02-27 02:26:33 +00:00
}
// Used to wrap ACL elements in angle brackets for storage
2013-02-27 02:26:33 +00:00
function sanitise_acl(&$item) {
if(strlen($item))
$item = '<' . notags(trim($item)) . '>';
else
unset($item);
2013-02-27 02:26:33 +00:00
}
// Convert an ACL array to a storable string
2013-02-27 02:26:33 +00:00
function perms2str($p) {
$ret = '';
2012-11-12 03:12:20 +00:00
if(is_array($p))
$tmp = $p;
else
$tmp = explode(',',$p);
if(is_array($tmp)) {
array_walk($tmp,'sanitise_acl');
$ret = implode('',$tmp);
}
return $ret;
2013-02-27 02:26:33 +00:00
}
// generate a guaranteed unique (for this domain) item ID for ATOM
// safe from birthday paradox
2013-02-27 02:26:33 +00:00
function item_message_id() {
do {
$dups = false;
$hash = random_string();
$mid = $hash . '@' . get_app()->get_hostname();
$r = q("SELECT id FROM item WHERE mid = '%s' LIMIT 1",
dbesc($mid));
if(count($r))
$dups = true;
} while($dups == true);
return $mid;
2013-02-27 02:26:33 +00:00
}
// Generate a guaranteed unique photo ID.
// safe from birthday paradox
2013-02-27 02:26:33 +00:00
function photo_new_resource() {
do {
$found = false;
$resource = hash('md5',uniqid(mt_rand(),true));
$r = q("SELECT id FROM photo WHERE resource_id = '%s' LIMIT 1",
dbesc($resource)
);
if(count($r))
$found = true;
} while($found == true);
return $resource;
2013-02-27 02:26:33 +00:00
}
// for html,xml parsing - let's say you've got
// an attribute foobar="class1 class2 class3"
// and you want to find out if it contains 'class3'.
// you can't use a normal sub string search because you
// might match 'notclass3' and a regex to do the job is
// possible but a bit complicated.
// pass the attribute string as $attr and the attribute you
// are looking for as $s - returns true if found, otherwise false
function attribute_contains($attr,$s) {
$a = explode(' ', $attr);
if(count($a) && in_array($s,$a))
return true;
return false;
2013-02-27 02:26:33 +00:00
}
function logger($msg,$level = 0) {
// turn off logger in install mode
global $a;
2012-02-12 22:18:32 +00:00
global $db;
if(($a->module == 'install') || (! ($db && $db->connected))) return;
$debugging = get_config('system','debugging');
$loglevel = intval(get_config('system','loglevel'));
$logfile = get_config('system','logfile');
if((! $debugging) || (! $logfile) || ($level > $loglevel))
return;
$where = '';
if(version_compare(PHP_VERSION,'5.4.0') >= 0) {
$stack = debug_backtrace(DEBUG_BACKTRACE_IGNORE_ARGS,2);
$where = basename($stack[0]['file']) . ':' . $stack[0]['line'] . ':' . $stack[1]['function'] . ': ';
}
@file_put_contents($logfile, datetime_convert() . ':' . session_id() . ' ' . $where . $msg . "\n", FILE_APPEND);
return;
2013-02-27 02:26:33 +00:00
}
// This is a special logging facility for developers. It allows one to target specific things to trace/debug
// and is identical to logger() with the exception of the log filename. This allows one to isolate specific
// calls while allowing logger() to paint a bigger picture of overall activity and capture more detail.
// If you find dlogger() calls in checked in code, you are free to remove them - so as to provide a noise-free
// development environment which responds to events you are targetting personally.
2013-02-27 02:26:33 +00:00
function dlogger($msg,$level = 0) {
// turn off logger in install mode
global $a;
global $db;
if(($a->module == 'install') || (! ($db && $db->connected))) return;
$debugging = get_config('system','debugging');
$loglevel = intval(get_config('system','loglevel'));
$logfile = get_config('system','dlogfile');
if((! $debugging) || (! $logfile) || ($level > $loglevel))
return;
@file_put_contents($logfile, datetime_convert() . ':' . session_id() . ' ' . $msg . "\n", FILE_APPEND);
return;
2013-02-27 02:26:33 +00:00
}
function profiler($t1,$t2,$label) {
if(file_exists('profiler.out') && $t1 && t2)
@file_put_contents('profiler.out', sprintf('%01.4f %s',$t2 - $t1,$label) . "\n", FILE_APPEND);
}
2013-02-27 02:26:33 +00:00
function activity_match($haystack,$needle) {
if(($haystack === $needle) || ((basename($needle) === $haystack) && strstr($needle,NAMESPACE_ACTIVITY_SCHEMA)))
return true;
return false;
2013-02-27 02:26:33 +00:00
}
// Pull out all #hashtags and @person tags from $s;
// We also get @person@domain.com - which would make
// the regex quite complicated as tags can also
// end a sentence. So we'll run through our results
// and strip the period from any tags which end with one.
// Returns array of tags found, or empty array.
2013-02-27 02:26:33 +00:00
function get_tags($s) {
$ret = array();
// ignore anything in a code block
$s = preg_replace('/\[code\](.*?)\[\/code\]/sm','',$s);
2014-02-05 13:34:25 +00:00
// ignore anything in [style= ]
$s = preg_replace('/\[style=(.*?)\]/sm','',$s);
// match any double quoted tags
if(preg_match_all('/([@#]\&quot\;.*?\&quot\;)/',$s,$match)) {
foreach($match[1] as $mtch) {
$ret[] = $mtch;
}
}
// Match full names against @tags including the space between first and last
// We will look these up afterward to see if they are full names or not recognisable.
2011-11-03 04:07:17 +00:00
if(preg_match_all('/(@[^ \x0D\x0A,:?]+ [^ \x0D\x0A@,:?]+)([ \x0D\x0A@,:?]|$)/',$s,$match)) {
foreach($match[1] as $mtch) {
if(strstr($mtch,"]")) {
// we might be inside a bbcode color tag - leave it alone
continue;
}
if(substr($mtch,-1,1) === '.')
$ret[] = substr($mtch,0,-1);
else
$ret[] = $mtch;
}
}
// Otherwise pull out single word tags. These can be @nickname, @first_last
// and #hash tags.
2011-10-26 02:59:57 +00:00
if(preg_match_all('/([@#][^ \x0D\x0A,;:?]+)([ \x0D\x0A,;:?]|$)/',$s,$match)) {
foreach($match[1] as $mtch) {
if(strstr($mtch,"]")) {
// we might be inside a bbcode color tag - leave it alone
continue;
}
if(substr($mtch,-1,1) === '.')
$mtch = substr($mtch,0,-1);
// ignore strictly numeric tags like #1
2014-02-04 03:38:15 +00:00
if((strpos($mtch,'#') === 0) && ( ctype_digit(substr($mtch,1)) || substr($mtch,1,1) === '^'))
continue;
2011-11-26 06:41:50 +00:00
// try not to catch url fragments
if(strpos($s,$mtch) && preg_match('/[a-zA-z0-9\/]/',substr($s,strpos($s,$mtch)-1,1)))
continue;
2014-04-13 00:04:54 +00:00
// or quote remnants from the quoted strings we already picked out earlier
if(strpos($mtch,'&quot'))
continue;
$ret[] = $mtch;
}
}
2014-02-04 03:38:15 +00:00
// bookmarks
2014-02-05 03:39:56 +00:00
if(preg_match_all('/#\^\[(url|zrl)(.*?)\](.*?)\[\/(url|zrl)\]/',$s,$match,PREG_SET_ORDER)) {
2014-02-04 03:38:15 +00:00
foreach($match as $mtch) {
$ret[] = $mtch[0];
}
}
// make sure the longer tags are returned first so that if two or more have common substrings
// we'll replace the longest ones first. Otherwise the common substring would be found in
// both strings and the string replacement would link both to the shorter strings and
// fail to link the longer string. RedMatrix github issue #378
usort($ret,'tag_sort_length');
2014-02-04 03:38:15 +00:00
2014-04-13 06:57:19 +00:00
// logger('get_tags: ' . print_r($ret,true));
2014-02-04 03:38:15 +00:00
return $ret;
2013-02-27 02:26:33 +00:00
}
function tag_sort_length($a,$b) {
if(mb_strlen($a) == mb_strlen($b))
return 0;
return((mb_strlen($b) < mb_strlen($a)) ? (-1) : 1);
}
function strip_zids($s) {
return preg_replace('/[\?&]zid=(.*?)(&|$)/ism','$2',$s);
}
// quick and dirty quoted_printable encoding
2013-02-27 02:26:33 +00:00
function qp($s) {
return str_replace ("%","=",rawurlencode($s));
2013-02-27 02:26:33 +00:00
}
function get_mentions($item,$tags) {
$o = '';
if(! count($tags))
return $o;
foreach($tags as $x) {
if($x['type'] == TERM_MENTION) {
$o .= "\t\t" . '<link rel="mentioned" href="' . $x['url'] . '" />' . "\r\n";
$o .= "\t\t" . '<link rel="ostatus:attention" href="' . $x['url'] . '" />' . "\r\n";
}
}
return $o;
2013-02-27 02:26:33 +00:00
}
function contact_block() {
$o = '';
$a = get_app();
if(! $a->profile['uid'])
return;
if(! perm_is_allowed($a->profile['uid'],get_observer_hash(),'view_contacts'))
return;
$shown = get_pconfig($a->profile['uid'],'system','display_friend_count');
2012-11-14 03:32:59 +00:00
if($shown === false)
$shown = 24;
if($shown == 0)
return;
$is_owner = ((local_user() && local_user() == $a->profile['uid']) ? true : false);
$abook_flags = ABOOK_FLAG_PENDING|ABOOK_FLAG_SELF;
$xchan_flags = XCHAN_FLAGS_ORPHAN|XCHAN_FLAGS_DELETED;
if(! $is_owner) {
$abook_flags = $abook_flags | ABOOK_FLAG_HIDDEN;
$xchan_flags = $xchan_flags | XCHAN_FLAGS_HIDDEN;
}
2012-08-31 01:17:38 +00:00
if((! is_array($a->profile)) || ($a->profile['hide_friends']))
return $o;
2014-04-15 00:29:19 +00:00
$r = q("SELECT COUNT(abook_id) AS total FROM abook left join xchan on abook_xchan = xchan_hash WHERE abook_channel = %d and not ( abook_flags & %d ) and not (xchan_flags & %d)",
intval($a->profile['uid']),
intval($abook_flags),
intval($xchan_flags)
);
if(count($r)) {
$total = intval($r[0]['total']);
}
if(! $total) {
2013-02-02 19:07:28 +00:00
$contacts = t('No connections');
2011-08-31 15:31:44 +00:00
$micropro = Null;
2011-08-31 15:31:44 +00:00
} else {
2014-04-15 00:29:19 +00:00
$r = q("SELECT abook.*, xchan.* FROM abook left join xchan on abook.abook_xchan = xchan.xchan_hash WHERE abook_channel = %d AND not ( abook_flags & %d) and not (xchan_flags & %d ) ORDER BY RAND() LIMIT %d",
2011-08-31 15:31:44 +00:00
intval($a->profile['uid']),
intval($abook_flags|ABOOK_FLAG_ARCHIVED),
intval($xchan_flags),
2011-08-31 15:31:44 +00:00
intval($shown)
);
2011-08-31 15:31:44 +00:00
if(count($r)) {
2012-12-07 02:17:43 +00:00
$contacts = sprintf( tt('%d Connection','%d Connections', $total),$total);
2011-08-31 15:31:44 +00:00
$micropro = Array();
foreach($r as $rr) {
$rr['archived'] = (($rr['abook_flags'] & ABOOK_FLAG_ARCHIVED) ? true : false);
2011-08-31 15:31:44 +00:00
$micropro[] = micropro($rr,true,'mpfriend');
}
}
}
2011-08-31 15:31:44 +00:00
$tpl = get_markup_template('contact_block.tpl');
$o = replace_macros($tpl, array(
'$contacts' => $contacts,
2013-02-19 08:20:47 +00:00
'$nickname' => $a->profile['channel_address'],
'$viewconnections' => t('View Connections'),
2011-08-31 15:31:44 +00:00
'$micropro' => $micropro,
));
$arr = array('contacts' => $r, 'output' => $o);
call_hooks('contact_block_end', $arr);
return $o;
2013-02-27 02:26:33 +00:00
}
2012-12-07 02:17:43 +00:00
function chanlink_hash($s) {
return z_root() . '/chanview?f=&hash=' . urlencode($s);
}
function chanlink_url($s) {
return z_root() . '/chanview?f=&url=' . urlencode($s);
}
function chanlink_cid($d) {
return z_root() . '/chanview?f=&cid=' . intval($d);
}
function magiclink_url($observer,$myaddr,$url) {
return (($observer)
? z_root() . '/magic?f=&dest=' . $url . '&addr=' . $myaddr
: $url
);
}
2012-12-07 02:17:43 +00:00
2013-02-27 02:26:33 +00:00
function micropro($contact, $redirect = false, $class = '', $textmode = false) {
if($contact['click'])
$url = '#';
else
$url = chanlink_hash($contact['xchan_hash']);
2012-12-07 02:17:43 +00:00
return replace_macros(get_markup_template(($textmode)?'micropro_txt.tpl':'micropro_img.tpl'),array(
'$click' => (($contact['click']) ? $contact['click'] : ''),
'$class' => $class . (($contact['archived']) ? ' archived' : ''),
2012-12-07 02:17:43 +00:00
'$url' => $url,
'$photo' => $contact['xchan_photo_s'],
'$name' => $contact['xchan_name'],
'$title' => $contact['xchan_name'] . ' [' . $contact['xchan_addr'] . ']',
));
2013-02-27 02:26:33 +00:00
}
2011-09-05 00:35:06 +00:00
function search($s,$id='search-box',$url='/search',$save = false) {
$a = get_app();
$o = '<div id="' . $id . '">';
$o .= '<form action="' . $a->get_baseurl((stristr($url,'network')) ? true : false) . $url . '" method="get" >';
$o .= '<input type="text" class="icon-search" name="search" id="search-text" placeholder="&#xf002;" value="' . $s .'" onclick="this.submit();" />';
2014-05-23 04:24:41 +00:00
$o .= '<input class="search-submit btn btn-default" type="submit" name="submit" id="search-submit" value="' . t('Search') . '" />';
if(feature_enabled(local_user(),'savedsearch'))
2014-05-23 04:24:41 +00:00
$o .= '<input class="search-save btn btn-default" type="submit" name="save" id="search-save" value="' . t('Save') . '" />';
$o .= '</form></div>';
return $o;
2013-02-27 02:26:33 +00:00
}
2013-12-10 05:20:55 +00:00
function searchbox($s,$id='search-box',$url='/search',$save = false) {
$a = get_app();
$o = '<div id="' . $id . '">';
$o .= '<form action="' . z_root() . '/' . $url . '" method="get" >';
$o .= '<input type="hidden" name="f" value="" />';
$o .= '<input type="text" class="icon-search" name="search" id="search-text" placeholder="&#xf002;" value="' . $s .'" onclick="this.submit();" />';
$o .= '<input type="submit" name="submit" class="btn btn-default" id="search-submit" value="' . t('Search') . '" />';
2013-12-10 05:20:55 +00:00
if(feature_enabled(local_user(),'savedsearch'))
$o .= '<input type="submit" name="searchsave" class="btn btn-default" id="search-save" value="' . t('Save') . '" />';
2013-12-10 05:20:55 +00:00
$o .= '</form></div>';
return $o;
}
function valid_email($x){
2012-07-02 01:56:00 +00:00
if(get_config('system','disable_email_validation'))
return true;
2011-12-16 22:31:39 +00:00
if(preg_match('/^[_a-zA-Z0-9\-\+]+(\.[_a-zA-Z0-9\-\+]+)*@[a-zA-Z0-9-]+(\.[a-zA-Z0-9-]+)+$/',$x))
return true;
return false;
2013-02-27 02:26:33 +00:00
}
/**
*
* Function: linkify
*
* Replace naked text hyperlink with HTML formatted hyperlink
*
*/
2013-02-27 02:26:33 +00:00
function linkify($s) {
$s = preg_replace("/(https?\:\/\/[a-zA-Z0-9\:\/\-\?\&\;\.\=\_\@\~\#\'\%\$\!\+]*)/", ' <a href="$1" >$1</a>', $s);
2011-09-15 03:47:49 +00:00
$s = preg_replace("/\<(.*?)(src|href)=(.*?)\&amp\;(.*?)\>/ism",'<$1$2=$3&$4>',$s);
return($s);
2013-02-27 02:26:33 +00:00
}
/**
* @function sslify($s)
* Replace media element using http url with https to a local redirector if using https locally
* @param string $s
*
* Looks for HTML tags containing src elements that are http when we're viewing an https page
* Typically this throws an insecure content violation in the browser. So we redirect them
* to a local redirector which uses https and which redirects to the selected content
*
* @returns string
*/
function sslify($s) {
if(strpos(z_root(),'https:') === false)
return $s;
$matches = null;
$cnt = preg_match_all("/\<(.*?)src=\"(http\:.*?)\"(.*?)\>/",$s,$matches,PREG_SET_ORDER);
if($cnt) {
foreach($matches as $match) {
$s = str_replace($match[2],z_root() . '/sslify?f=&url=' . urlencode($match[2]),$s);
}
}
return $s;
}
2012-07-20 01:53:26 +00:00
function get_poke_verbs() {
// index is present tense verb
// value is array containing past tense verb, translation of present, translation of past
$arr = array(
'poke' => array( 'poked', t('poke'), t('poked')),
2012-07-21 10:48:59 +00:00
'ping' => array( 'pinged', t('ping'), t('pinged')),
2012-07-20 01:53:26 +00:00
'prod' => array( 'prodded', t('prod'), t('prodded')),
'slap' => array( 'slapped', t('slap'), t('slapped')),
2012-07-21 10:48:59 +00:00
'finger' => array( 'fingered', t('finger'), t('fingered')),
'rebuff' => array( 'rebuffed', t('rebuff'), t('rebuffed')),
2012-07-20 01:53:26 +00:00
);
call_hooks('poke_verbs', $arr);
return $arr;
}
2012-08-24 03:00:10 +00:00
function get_mood_verbs() {
$arr = array(
'happy' => t('happy'),
'sad' => t('sad'),
'mellow' => t('mellow'),
'tired' => t('tired'),
'perky' => t('perky'),
'angry' => t('angry'),
'stupefied' => t('stupified'),
'puzzled' => t('puzzled'),
'interested' => t('interested'),
'bitter' => t('bitter'),
'cheerful' => t('cheerful'),
'alive' => t('alive'),
'annoyed' => t('annoyed'),
'anxious' => t('anxious'),
'cranky' => t('cranky'),
'disturbed' => t('disturbed'),
'frustrated' => t('frustrated'),
2014-04-27 02:33:39 +00:00
'depressed' => t('depressed'),
2012-08-24 03:00:10 +00:00
'motivated' => t('motivated'),
'relaxed' => t('relaxed'),
'surprised' => t('surprised'),
);
call_hooks('mood_verbs', $arr);
return $arr;
}
/**
*
* Function: smilies
*
* Description:
* Replaces text emoticons with graphical images
*
* @Parameter: string $s
*
* Returns string
2012-02-22 09:50:06 +00:00
*
* It is expected that this function will be called using HTML text.
* We will escape text between HTML pre and code blocks, and HTML attributes
* (such as urls) from being processed.
2012-02-22 09:50:06 +00:00
*
* At a higher level, the bbcode [nosmile] tag can be used to prevent this
* function from being executed by the prepare_text() routine when preparing
* bbcode source for HTML display
*
*/
2013-02-27 02:26:33 +00:00
function smilies($s, $sample = false) {
2012-03-19 10:18:39 +00:00
$a = get_app();
2012-03-19 10:18:39 +00:00
if(intval(get_config('system','no_smilies'))
|| (local_user() && intval(get_pconfig(local_user(),'system','no_smilies'))))
return $s;
2014-02-02 13:52:00 +00:00
$s = preg_replace_callback('{<(pre|code)>.*?</\1>}ism','smile_shield',$s);
$s = preg_replace_callback('/<[a-z]+ .*?>/ism','smile_shield',$s);
$texts = array(
'&lt;3',
'&lt;/3',
'&lt;\\3',
':-)',
';-)',
':-(',
':-P',
2012-03-22 04:45:45 +00:00
':-p',
':-"',
':-&quot;',
':-x',
':-X',
':-D',
'8-|',
'8-O',
':-O',
'\\o/',
'o.O',
'O.o',
2012-04-02 05:57:50 +00:00
'o_O',
'O_o',
":'(",
":-!",
":-/",
":-[",
"8-)",
':beer',
':homebrew',
':coffee',
2012-02-16 23:15:28 +00:00
':facepalm',
2012-05-31 00:22:51 +00:00
':like',
':dislike',
'red#matrix',
'red#',
'r#'
);
$icons = array(
2012-06-28 23:42:40 +00:00
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-heart.gif" alt="<3" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-brokenheart.gif" alt="</3" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-brokenheart.gif" alt="<\\3" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-smile.gif" alt=":-)" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-wink.gif" alt=";-)" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-frown.gif" alt=":-(" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-tongue-out.gif" alt=":-P" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-tongue-out.gif" alt=":-p" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-kiss.gif" alt=":-\"" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-kiss.gif" alt=":-\"" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-kiss.gif" alt=":-x" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-kiss.gif" alt=":-X" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-laughing.gif" alt=":-D" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-surprised.gif" alt="8-|" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-surprised.gif" alt="8-O" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-surprised.gif" alt=":-O" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-thumbsup.gif" alt="\\o/" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-Oo.gif" alt="o.O" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-Oo.gif" alt="O.o" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-Oo.gif" alt="o_O" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-Oo.gif" alt="O_o" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-cry.gif" alt=":\'(" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-foot-in-mouth.gif" alt=":-!" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-undecided.gif" alt=":-/" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-embarassed.gif" alt=":-[" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-cool.gif" alt="8-)" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/beer_mug.gif" alt=":beer" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/beer_mug.gif" alt=":homebrew" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/coffee.gif" alt=":coffee" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-facepalm.gif" alt=":facepalm" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/like.gif" alt=":like" />',
'<img class="smiley" src="' . $a->get_baseurl() . '/images/dislike.gif" alt=":dislike" />',
'<a href="http://getzot.com"><strong>red<img class="smiley" src="' . $a->get_baseurl() . '/images/rm-16.png" alt="red#matrix" />matrix</strong></a>',
'<a href="http://getzot.com"><strong>red<img class="smiley" src="' . $a->get_baseurl() . '/images/rm-16.png" alt="red#" />matrix</strong></a>',
'<a href="http://getzot.com"><strong>red<img class="smiley" src="' . $a->get_baseurl() . '/images/rm-16.png" alt="r#" />matrix</strong></a>'
);
$params = array('texts' => $texts, 'icons' => $icons, 'string' => $s);
call_hooks('smilie', $params);
2011-09-12 10:21:39 +00:00
if($sample) {
$s = '<div class="smiley-sample">';
for($x = 0; $x < count($params['texts']); $x ++) {
$s .= '<dl><dt>' . $params['texts'][$x] . '</dt><dd>' . $params['icons'][$x] . '</dd></dl>';
}
}
else {
$params['string'] = preg_replace_callback('/&lt;(3+)/','preg_heart',$params['string']);
$s = str_replace($params['texts'],$params['icons'],$params['string']);
}
2014-02-02 13:52:00 +00:00
$s = preg_replace_callback('/<!--base64:(.*?)-->/ism', 'smile_unshield', $s);
2011-09-12 10:21:39 +00:00
return $s;
2013-02-27 02:26:33 +00:00
}
2014-02-02 13:52:00 +00:00
function smile_shield($m) {
return '<!--base64:' . base64url_encode($m[0]) . '-->';
}
2014-02-02 13:52:00 +00:00
function smile_unshield($m) {
return base64url_decode($m[1]);
}
// expand <3333 to the correct number of hearts
function preg_heart($x) {
$a = get_app();
if(strlen($x[1]) == 1)
return $x[0];
$t = '';
for($cnt = 0; $cnt < strlen($x[1]); $cnt ++)
2012-06-28 23:42:40 +00:00
$t .= '<img class="smiley" src="' . $a->get_baseurl() . '/images/smiley-heart.gif" alt="<3" />';
$r = str_replace($x[0],$t,$x[0]);
return $r;
}
2013-02-27 02:26:33 +00:00
function day_translate($s) {
$ret = str_replace(array('Monday','Tuesday','Wednesday','Thursday','Friday','Saturday','Sunday'),
array( t('Monday'), t('Tuesday'), t('Wednesday'), t('Thursday'), t('Friday'), t('Saturday'), t('Sunday')),
$s);
$ret = str_replace(array('January','February','March','April','May','June','July','August','September','October','November','December'),
array( t('January'), t('February'), t('March'), t('April'), t('May'), t('June'), t('July'), t('August'), t('September'), t('October'), t('November'), t('December')),
$ret);
return $ret;
2013-02-27 02:26:33 +00:00
}
function normalise_link($url) {
$ret = str_replace(array('https:','//www.'), array('http:','//'), $url);
return(rtrim($ret,'/'));
2013-02-27 02:26:33 +00:00
}
/**
*
* Compare two URLs to see if they are the same, but ignore
* slight but hopefully insignificant differences such as if one
* is https and the other isn't, or if one is www.something and
* the other isn't - and also ignore case differences.
*
* Return true if the URLs match, otherwise false.
*
*/
2013-02-27 02:26:33 +00:00
function link_compare($a,$b) {
if(strcasecmp(normalise_link($a),normalise_link($b)) === 0)
return true;
return false;
2013-02-27 02:26:33 +00:00
}
// Given an item array, convert the body element from bbcode to html and add smilie icons.
// If attach is true, also add icons for item attachments
function unobscure(&$item) {
if(array_key_exists('item_flags',$item) && ($item['item_flags'] & ITEM_OBSCURED)) {
$key = get_config('system','prvkey');
if($item['title'])
$item['title'] = crypto_unencapsulate(json_decode_plus($item['title']),$key);
if($item['body'])
$item['body'] = crypto_unencapsulate(json_decode_plus($item['body']),$key);
}
}
2013-10-14 04:14:04 +00:00
function theme_attachments(&$item) {
$arr = json_decode_plus($item['attach']);
2013-10-14 04:14:04 +00:00
if(is_array($arr) && count($arr)) {
$attaches = array();
foreach($arr as $r) {
$icon = '';
$icontype = substr($r['type'],0,strpos($r['type'],'/'));
2013-10-14 22:34:47 +00:00
// FIXME This should probably be a giant "if" statement in the template so that we don't have icon names
// embedded in php code
switch($icontype) {
case 'video':
2013-10-14 04:14:04 +00:00
$icon = 'icon-facetime-video';
break;
case 'audio':
2013-10-14 04:14:04 +00:00
$icon = 'icon-volume-up';
break;
case 'image':
$icon = 'icon-picture';
2013-10-14 04:14:04 +00:00
break;
case 'text':
2013-10-14 04:14:04 +00:00
$icon = 'icon-align-justify';
break;
default:
2013-10-14 04:14:04 +00:00
$icon = 'icon-question';
break;
}
$title = htmlspecialchars($r['title'], ENT_COMPAT,'UTF-8');
if(! $title)
$title = t('unknown.???');
$title .= ' ' . $r['length'] . ' ' . t('bytes');
2013-10-14 04:14:04 +00:00
$url = z_root() . '/magic?f=&hash=' . $item['author_xchan'] . '&dest=' . $r['href'] . '/' . $r['revision'];
$s .= '<a href="' . $url . '" title="' . $title . '" class="attachlink" >' . $icon . '</a>';
2013-10-14 04:14:04 +00:00
$attaches[] = array('title' => $title, 'url' => $url, 'icon' => $icon );
}
2013-10-14 04:14:04 +00:00
}
$s = replace_macros(get_markup_template('item_attach.tpl'), array(
'$attaches' => $attaches
));
return $s;
}
function format_categories(&$item,$writeable) {
$s = '';
$terms = get_terms_oftype($item['term'],TERM_CATEGORY);
if($terms) {
$categories = array();
foreach($terms as $t) {
$term = htmlspecialchars($t['term'],ENT_COMPAT,'UTF-8',false) ;
2013-10-14 22:34:47 +00:00
if(! trim($term))
continue;
$removelink = (($writeable) ? z_root() . '/filerm/' . $item['id'] . '?f=&cat=' . urlencode($t['term']) : '');
2013-12-02 23:15:02 +00:00
$categories[] = array('term' => $term, 'writeable' => $writeable, 'removelink' => $removelink, 'url' => zid($t['url']));
}
}
$s = replace_macros(get_markup_template('item_categories.tpl'),array(
'$remove' => t('remove category'),
'$categories' => $categories
));
return $s;
}
// Add any hashtags which weren't mentioned in the message body, e.g. community tags
function format_hashtags(&$item) {
$s = '';
$terms = get_terms_oftype($item['term'],TERM_HASHTAG);
if($terms) {
$categories = array();
foreach($terms as $t) {
$term = htmlspecialchars($t['term'],ENT_COMPAT,'UTF-8',false) ;
if(! trim($term))
continue;
if(strpos($item['body'], $t['url']))
continue;
if($s)
$s .= '&nbsp';
$s .= '#<a href="' . zid($t['url']) . '" >' . $term . '</a>';
}
}
return $s;
}
function format_mentions(&$item) {
$s = '';
$terms = get_terms_oftype($item['term'],TERM_MENTION);
if($terms) {
$categories = array();
foreach($terms as $t) {
$term = htmlspecialchars($t['term'],ENT_COMPAT,'UTF-8',false) ;
if(! trim($term))
continue;
if(strpos($item['body'], $t['url']))
continue;
if($s)
$s .= '&nbsp';
$s .= '@<a href="' . zid($t['url']) . '" >' . $term . '</a>';
}
}
return $s;
}
function format_filer(&$item) {
$s = '';
$terms = get_terms_oftype($item['term'],TERM_FILE);
if($terms) {
$categories = array();
foreach($terms as $t) {
$term = htmlspecialchars($t['term'],ENT_COMPAT,'UTF-8',false) ;
2013-10-14 22:34:47 +00:00
if(! trim($term))
continue;
$removelink = z_root() . '/filerm/' . $item['id'] . '?f=&term=' . urlencode($t['term']);
$categories[] = array('term' => $term, 'removelink' => $removelink);
}
}
$s = replace_macros(get_markup_template('item_filer.tpl'),array(
'$remove' => t('remove from file'),
'$categories' => $categories
));
return $s;
}
2013-10-14 04:14:04 +00:00
function prepare_body(&$item,$attach = false) {
$a = get_app();
call_hooks('prepare_body_init', $item);
unobscure($item);
$s = prepare_text($item['body'],$item['mimetype']);
$prep_arr = array('item' => $item, 'html' => $s);
call_hooks('prepare_body', $prep_arr);
$s = $prep_arr['html'];
if(! $attach) {
return $s;
}
2012-07-10 05:08:25 +00:00
2013-10-14 04:14:04 +00:00
$s .= theme_attachments($item);
2013-03-04 07:38:08 +00:00
$writeable = ((get_observer_hash() == $item['owner_xchan']) ? true : false);
$s .= format_hashtags($item);
if($item['resource_type'])
$s .= format_mentions($item);
$s .= format_categories($item,$writeable);
if(local_user() == $item['uid'])
$s .= format_filer($item);
2011-10-24 22:47:17 +00:00
2013-12-25 12:06:26 +00:00
$s = sslify($s);
// Look for spoiler
$spoilersearch = '<blockquote class="spoiler">';
// Remove line breaks before the spoiler
while ((strpos($s, "\n".$spoilersearch) !== false))
$s = str_replace("\n".$spoilersearch, $spoilersearch, $s);
while ((strpos($s, "<br />".$spoilersearch) !== false))
$s = str_replace("<br />".$spoilersearch, $spoilersearch, $s);
while ((strpos($s, $spoilersearch) !== false)) {
$pos = strpos($s, $spoilersearch);
$rnd = random_string(8);
$spoilerreplace = '<br /> <span id="spoiler-wrap-'.$rnd.'" style="white-space:nowrap;" class="fakelink" onclick="openClose(\'spoiler-'.$rnd.'\');">'.sprintf(t('Click to open/close')).'</span>'.
'<blockquote class="spoiler" id="spoiler-'.$rnd.'" style="display: none;">';
$s = substr($s, 0, $pos).$spoilerreplace.substr($s, $pos+strlen($spoilersearch));
}
// Look for quote with author
$authorsearch = '<blockquote class="author">';
while ((strpos($s, $authorsearch) !== false)) {
$pos = strpos($s, $authorsearch);
$rnd = random_string(8);
$authorreplace = '<br /> <span id="author-wrap-'.$rnd.'" style="white-space:nowrap;" class="fakelink" onclick="openClose(\'author-'.$rnd.'\');">'.sprintf(t('Click to open/close')).'</span>'.
'<blockquote class="author" id="author-'.$rnd.'" style="display: block;">';
$s = substr($s, 0, $pos).$authorreplace.substr($s, $pos+strlen($authorsearch));
}
2011-10-24 22:47:17 +00:00
$prep_arr = array('item' => $item, 'html' => $s);
call_hooks('prepare_body_final', $prep_arr);
2011-10-24 22:47:17 +00:00
return $prep_arr['html'];
2013-02-27 02:26:33 +00:00
}
// Given a text string, convert from bbcode to html and add smilie icons.
2013-02-27 02:26:33 +00:00
function prepare_text($text,$content_type = 'text/bbcode') {
switch($content_type) {
case 'text/plain':
$s = escape_tags($text);
break;
case 'text/html':
$s = $text;
break;
case 'text/markdown':
require_once('library/markdown.php');
$s = Markdown($text);
break;
// No security checking is done here at display time - so we need to verify
// that the author is allowed to use PHP before storing. We also cannot allow
// importation of PHP text bodies from other sites. Therefore this content
// type is only valid for web pages (and profile details).
// It may be possible to provide a PHP message body which is evaluated on the
// sender's site before sending it elsewhere. In that case we will have a
// different content-type here.
case 'application/x-php':
ob_start();
eval($text);
$s = ob_get_contents();
ob_end_clean();
break;
case 'text/bbcode':
case '':
default:
require_once('include/bbcode.php');
if(stristr($text,'[nosmile]'))
$s = bbcode($text);
else
$s = smilies(bbcode($text));
$s = zidify_links($s);
break;
}
2013-09-02 23:37:54 +00:00
//logger('prepare_text: ' . $s);
return $s;
2013-02-27 02:26:33 +00:00
}
/**
* zidify_callback() and zidify_links() work together to turn any HTML a tags with class="zrl" into zid links
* These will typically be generated by a bbcode '[zrl]' tag. This is done inside prepare_text() rather than bbcode()
* because the latter is used for general purpose conversions and the former is used only when preparing text for
* immediate display.
*
* Issues: Currently the order of HTML parameters in the text is somewhat rigid and inflexible.
* We assume it looks like <a class="zrl" href="xxxxxxxxxx"> and will not work if zrl and href appear in a different order.
*/
function zidify_callback($match) {
$is_zid = ((feature_enabled(local_user(),'sendzid')) || (strpos($match[1],'zrl')) ? true : false);
$replace = '<a' . $match[1] . ' href="' . (($is_zid) ? zid($match[2]) : $match[2]) . '"';
$x = str_replace($match[0],$replace,$match[0]);
return $x;
}
function zidify_img_callback($match) {
$is_zid = ((feature_enabled(local_user(),'sendzid')) || (strpos($match[1],'zrl')) ? true : false);
$replace = '<img' . $match[1] . ' src="' . (($is_zid) ? zid($match[2]) : $match[2]) . '"';
$x = str_replace($match[0],$replace,$match[0]);
return $x;
}
function zidify_links($s) {
$s = preg_replace_callback('/\<a(.*?)href\=\"(.*?)\"/ism','zidify_callback',$s);
$s = preg_replace_callback('/\<img(.*?)src\=\"(.*?)\"/ism','zidify_img_callback',$s);
return $s;
}
/**
* return atom link elements for all of our hubs
*/
2013-02-27 02:26:33 +00:00
function feed_hublinks() {
$hub = get_config('system','huburl');
$hubxml = '';
if(strlen($hub)) {
$hubs = explode(',', $hub);
if(count($hubs)) {
foreach($hubs as $h) {
$h = trim($h);
if(! strlen($h))
continue;
$hubxml .= '<link rel="hub" href="' . xmlify($h) . '" />' . "\n" ;
}
}
}
return $hubxml;
2013-02-27 02:26:33 +00:00
}
/* return atom link elements for salmon endpoints */
2013-02-27 02:26:33 +00:00
function feed_salmonlinks($nick) {
$a = get_app();
$salmon = '<link rel="salmon" href="' . xmlify($a->get_baseurl() . '/salmon/' . $nick) . '" />' . "\n" ;
// old style links that status.net still needed as of 12/2010
$salmon .= ' <link rel="http://salmon-protocol.org/ns/salmon-replies" href="' . xmlify($a->get_baseurl() . '/salmon/' . $nick) . '" />' . "\n" ;
$salmon .= ' <link rel="http://salmon-protocol.org/ns/salmon-mention" href="' . xmlify($a->get_baseurl() . '/salmon/' . $nick) . '" />' . "\n" ;
return $salmon;
2013-02-27 02:26:33 +00:00
}
2014-01-10 00:23:58 +00:00
function get_plink($item,$conversation_mode = true) {
$a = get_app();
2014-01-10 00:23:58 +00:00
if($conversation_mode)
$key = 'plink';
else
$key = 'llink';
if(x($item,$key)) {
return array(
'href' => zid($item[$key]),
'title' => t('Link to Source'),
);
}
else {
return false;
}
2013-02-27 02:26:33 +00:00
}
function unamp($s) {
return str_replace('&amp;', '&', $s);
2013-02-27 02:26:33 +00:00
}
2013-09-03 03:25:33 +00:00
function layout_select($channel_id, $current = '') {
$r = q("select mid,sid from item left join item_id on iid = item.id where service = 'PDL' and item.uid = item_id.uid and item_id.uid = %d and (item_restrict & %d)",
intval($channel_id),
intval(ITEM_PDL)
);
if($r) {
$o = t('Select a page layout: ');
$o .= '<select name="layout_mid" id="select-layout_mid" >';
$empty_selected = (($current === '') ? ' selected="selected" ' : '');
$o .= '<option value="" ' . $empty_selected . '>' . t('default') . '</option>';
foreach($r as $rr) {
$selected = (($rr['mid'] == $current) ? ' selected="selected" ' : '');
$o .= '<option value="' . $rr['mid'] . '"' . $selected . '>' . $rr['sid'] . '</option>';
}
$o .= '</select>';
}
return $o;
}
function mimetype_select($channel_id, $current = 'text/bbcode') {
$x = array(
'text/bbcode',
2014-08-26 19:13:27 +00:00
'text/html',
'text/markdown',
'text/plain'
);
$r = q("select account_id, account_roles from account left join channel on account_id = channel_account_id where
channel_id = %d limit 1",
intval($channel_id)
);
if($r) {
if($r[0]['account_roles'] & ACCOUNT_ROLE_ALLOWCODE) {
2014-08-26 19:13:27 +00:00
if(local_user() && get_account_id() == $r[0]['account_id'])
$x[] = 'application/x-php';
}
}
$o = t('Page content type: ');
$o .= '<select name="mimetype" id="mimetype-select">';
foreach($x as $y) {
$select = (($y == $current) ? ' selected="selected" ' : '');
$o .= '<option name="' . $y . '"' . $select . '>' . $y . '</option>';
}
$o .= '</select>';
return $o;
}
function lang_selector() {
global $a;
$langs = glob('view/*/strings.php');
$lang_options = array();
$selected = "";
if(is_array($langs) && count($langs)) {
$langs[] = '';
if(! in_array('view/en/strings.php',$langs))
$langs[] = 'view/en/';
asort($langs);
foreach($langs as $l) {
if($l == '') {
$lang_options[""] = t('default');
continue;
}
$ll = substr($l,5);
$ll = substr($ll,0,strrpos($ll,'/'));
$selected = (($ll === $a->language && (x($_SESSION, 'language'))) ? $ll : $selected);
$lang_options[$ll]=$ll;
}
}
$tpl = get_markup_template("lang_selector.tpl");
$o = replace_macros($tpl, array(
'$title' => t('Select an alternate language'),
'$langs' => array($lang_options, $selected),
));
return $o;
2013-02-27 02:26:33 +00:00
}
function return_bytes ($size_str) {
switch (substr ($size_str, -1))
{
case 'M': case 'm': return (int)$size_str * 1048576;
case 'K': case 'k': return (int)$size_str * 1024;
case 'G': case 'g': return (int)$size_str * 1073741824;
default: return $size_str;
}
2013-02-27 02:26:33 +00:00
}
2012-08-27 06:05:00 +00:00
function base64url_encode($s, $strip_padding = true) {
$s = strtr(base64_encode($s),'+/','-_');
if($strip_padding)
$s = str_replace('=','',$s);
return $s;
}
function base64url_decode($s) {
if(is_array($s)) {
logger('base64url_decode: illegal input: ' . print_r(debug_backtrace(), true));
return $s;
}
return base64_decode(strtr($s,'-_','+/'));
}
2011-08-17 03:05:02 +00:00
function cleardiv() {
return '<div class="clear"></div>';
}
function bb_translate_video($s) {
$matches = null;
$r = preg_match_all("/\[video\](.*?)\[\/video\]/ism",$s,$matches,PREG_SET_ORDER);
if($r) {
foreach($matches as $mtch) {
if((stristr($mtch[1],'youtube')) || (stristr($mtch[1],'youtu.be')))
$s = str_replace($mtch[0],'[youtube]' . $mtch[1] . '[/youtube]',$s);
elseif(stristr($mtch[1],'vimeo'))
$s = str_replace($mtch[0],'[vimeo]' . $mtch[1] . '[/vimeo]',$s);
}
}
return $s;
}
function html2bb_video($s) {
$s = preg_replace('#<object[^>]+>(.*?)https?://www.youtube.com/((?:v|cp)/[A-Za-z0-9\-_=]+)(.*?)</object>#ism',
'[youtube]$2[/youtube]', $s);
$s = preg_replace('#<iframe[^>](.*?)https?://www.youtube.com/embed/([A-Za-z0-9\-_=]+)(.*?)</iframe>#ism',
'[youtube]$2[/youtube]', $s);
$s = preg_replace('#<iframe[^>](.*?)https?://player.vimeo.com/video/([0-9]+)(.*?)</iframe>#ism',
'[vimeo]$2[/vimeo]', $s);
return $s;
2011-10-27 08:54:52 +00:00
}
/**
* apply xmlify() to all values of array $val, recursively
*/
function array_xmlify($val){
if (is_bool($val)) return $val?"true":"false";
if (is_array($val)) return array_map('array_xmlify', $val);
return xmlify((string) $val);
}
function reltoabs($text, $base)
{
if (empty($base))
2013-02-27 02:26:33 +00:00
return $text;
$base = rtrim($base,'/');
$base2 = $base . "/";
// Replace links
$pattern = "/<a([^>]*) href=\"(?!http|https|\/)([^\"]*)\"/";
$replace = "<a\${1} href=\"" . $base2 . "\${2}\"";
$text = preg_replace($pattern, $replace, $text);
$pattern = "/<a([^>]*) href=\"(?!http|https)([^\"]*)\"/";
$replace = "<a\${1} href=\"" . $base . "\${2}\"";
$text = preg_replace($pattern, $replace, $text);
// Replace images
$pattern = "/<img([^>]*) src=\"(?!http|https|\/)([^\"]*)\"/";
$replace = "<img\${1} src=\"" . $base2 . "\${2}\"";
$text = preg_replace($pattern, $replace, $text);
$pattern = "/<img([^>]*) src=\"(?!http|https)([^\"]*)\"/";
$replace = "<img\${1} src=\"" . $base . "\${2}\"";
$text = preg_replace($pattern, $replace, $text);
// Done
return $text;
}
function item_post_type($item) {
2012-07-10 05:08:25 +00:00
switch($item['resource_type']) {
case 'photo':
$post_type = t('photo');
break;
case 'event':
$post_type = t('event');
break;
default:
$post_type = t('status');
if($item['mid'] != $item['parent_mid'])
$post_type = t('comment');
break;
}
2012-07-10 05:08:25 +00:00
if(strlen($item['verb']) && (! activity_match($item['verb'],ACTIVITY_POST)))
$post_type = t('activity');
2012-07-10 05:08:25 +00:00
return $post_type;
2012-03-12 04:32:11 +00:00
}
2012-03-22 23:17:10 +00:00
function undo_post_tagging($s) {
$matches = null;
$cnt = preg_match_all('/([@#])(\!*)\[zrl=(.*?)\](.*?)\[\/zrl\]/ism',$s,$matches,PREG_SET_ORDER);
2012-03-22 23:17:10 +00:00
if($cnt) {
foreach($matches as $mtch) {
$s = str_replace($mtch[0], $mtch[1] . $mtch[2] . str_replace(' ','_',$mtch[4]),$s);
2012-03-22 23:17:10 +00:00
}
}
return $s;
}
2012-04-10 23:31:49 +00:00
function fix_mce_lf($s) {
$s = str_replace("\r\n","\n",$s);
// $s = str_replace("\n\n","\n",$s);
2012-04-10 23:31:49 +00:00
return $s;
}
function protect_sprintf($s) {
return(str_replace('%','%%',$s));
}
function is_a_date_arg($s) {
$i = intval($s);
if($i > 1900) {
$y = date('Y');
if($i <= $y+1 && strpos($s,'-') == 4) {
$m = intval(substr($s,5));
if($m > 0 && $m <= 12)
return true;
}
}
return false;
}
function legal_webbie($s) {
if(! strlen($s))
return '';
$x = $s;
do {
$s = $x;
$x = preg_replace('/^([^a-z])(.*?)/',"$2",$s);
} while($x != $s);
return preg_replace('/([^a-z0-9\-\_])/','',$x);
}
function check_webbie($arr) {
$str = '';
$taken = array();
if(count($arr)) {
foreach($arr as $x) {
$y = legal_webbie($x);
if(strlen($y)) {
if($str)
$str .= ',';
$str .= "'" . dbesc($y) . "'";
}
}
if(strlen($str)) {
$r = q("select channel_address from channel where channel_address in ( $str ) ");
if(count($r)) {
foreach($r as $rr) {
$taken[] = $rr['channel_address'];
}
}
foreach($arr as $x) {
$y = legal_webbie($x);
if(! in_array($y,$taken)) {
return $y;
}
}
}
}
return '';
}
2012-08-30 06:03:03 +00:00
function ids_to_querystr($arr,$idx = 'id') {
$t = array();
foreach($arr as $x)
$t[] = $x[$idx];
return(implode(',', $t));
2012-10-08 04:44:11 +00:00
}
// Fetches xchan and hubloc data for an array of items with only an
// author_xchan and owner_xchan. If $abook is true also include the abook info.
// This is needed in the API to save extra per item lookups there.
function xchan_query(&$items,$abook = true) {
2012-10-08 04:44:11 +00:00
$arr = array();
if($items && count($items)) {
foreach($items as $item) {
if($item['owner_xchan'] && (! in_array($item['owner_xchan'],$arr)))
$arr[] = "'" . dbesc($item['owner_xchan']) . "'";
if($item['author_xchan'] && (! in_array($item['author_xchan'],$arr)))
$arr[] = "'" . dbesc($item['author_xchan']) . "'";
}
}
if(count($arr)) {
if($abook) {
$chans = q("select * from xchan left join hubloc on hubloc_hash = xchan_hash left join abook on abook_xchan = xchan_hash and abook_channel = %d
where xchan_hash in (" . implode(',', $arr) . ") and ( hubloc_flags & " . intval(HUBLOC_FLAGS_PRIMARY) . " )",
intval($item['uid'])
);
}
else {
$chans = q("select xchan.*,hubloc.* from xchan left join hubloc on hubloc_hash = xchan_hash
where xchan_hash in (" . implode(',', $arr) . ") and ( hubloc_flags & " . intval(HUBLOC_FLAGS_PRIMARY) . " )");
}
2014-09-01 03:51:05 +00:00
$xchans = q("select * from xchan where xchan_hash in (" . implode(',',$arr) . ") and xchan_network in ('rss','unknown')");
if(! $chans)
$chans = $xchans;
else
$chans = array_merge($xchans,$chans);
2012-10-08 04:44:11 +00:00
}
if($items && count($items) && $chans && count($chans)) {
for($x = 0; $x < count($items); $x ++) {
$items[$x]['owner'] = find_xchan_in_array($items[$x]['owner_xchan'],$chans);
$items[$x]['author'] = find_xchan_in_array($items[$x]['author_xchan'],$chans);
}
}
}
2012-12-06 00:44:07 +00:00
function xchan_mail_query(&$item) {
$arr = array();
$chans = null;
if($item) {
if($item['from_xchan'] && (! in_array($item['from_xchan'],$arr)))
$arr[] = "'" . dbesc($item['from_xchan']) . "'";
if($item['to_xchan'] && (! in_array($item['to_xchan'],$arr)))
$arr[] = "'" . dbesc($item['to_xchan']) . "'";
}
if(count($arr)) {
$chans = q("select xchan.*,hubloc.* from xchan left join hubloc on hubloc_hash = xchan_hash
where xchan_hash in (" . implode(',', $arr) . ") and ( hubloc_flags & " . intval(HUBLOC_FLAGS_PRIMARY) . " )");
}
if($chans) {
$item['from'] = find_xchan_in_array($item['from_xchan'],$chans);
$item['to'] = find_xchan_in_array($item['to_xchan'],$chans);
}
}
2012-10-08 04:44:11 +00:00
function find_xchan_in_array($xchan,$arr) {
if(count($arr)) {
foreach($arr as $x) {
if($x['xchan_hash'] === $xchan) {
return $x;
}
}
}
return array();
}
2012-11-14 03:32:59 +00:00
function get_rel_link($j,$rel) {
if(count($j))
foreach($j as $l)
2012-11-14 03:32:59 +00:00
if($l['rel'] === $rel)
return $l['href'];
return '';
}
// Lots of code to write here
function magic_link($s) {
return $s;
}
2012-11-15 01:02:30 +00:00
// if $escape is true, dbesc() each element before adding quotes
function stringify_array_elms(&$arr,$escape = false) {
2012-11-15 01:02:30 +00:00
for($x = 0; $x < count($arr); $x ++)
$arr[$x] = "'" . (($escape) ? dbesc($arr[$x]) : $arr[$x]) . "'";
2012-11-15 01:02:30 +00:00
}
/**
* Indents a flat JSON string to make it more human-readable.
*
* @param string $json The original JSON string to process.
*
* @return string Indented version of the original JSON string.
*/
function jindent($json) {
2013-02-27 02:26:33 +00:00
$result = '';
$pos = 0;
$strLen = strlen($json);
$indentStr = ' ';
$newLine = "\n";
$prevChar = '';
$outOfQuotes = true;
2013-02-27 02:26:33 +00:00
for ($i=0; $i<=$strLen; $i++) {
2013-02-27 02:26:33 +00:00
// Grab the next character in the string.
$char = substr($json, $i, 1);
2013-02-27 02:26:33 +00:00
// Are we inside a quoted string?
if ($char == '"' && $prevChar != '\\') {
$outOfQuotes = !$outOfQuotes;
// If this character is the end of an element,
// output a new line and indent the next line.
} else if(($char == '}' || $char == ']') && $outOfQuotes) {
$result .= $newLine;
$pos --;
for ($j=0; $j<$pos; $j++) {
$result .= $indentStr;
}
}
// Add the character to the result string.
$result .= $char;
// If the last character was the beginning of an element,
// output a new line and indent the next line.
if (($char == ',' || $char == '{' || $char == '[') && $outOfQuotes) {
$result .= $newLine;
if ($char == '{' || $char == '[') {
$pos ++;
}
for ($j = 0; $j < $pos; $j++) {
$result .= $indentStr;
}
}
$prevChar = $char;
}
2013-02-27 02:26:33 +00:00
return $result;
}
function json_decode_plus($s) {
$x = json_decode($s,true);
if(! $x)
$x = json_decode(str_replace(array('\\"','\\\\'),array('"','\\'),$s),true);
return $x;
}
function design_tools() {
2013-12-19 10:16:14 +00:00
$channel = get_app()->get_channel();
$who = $channel['channel_address'];
return replace_macros(get_markup_template('design_tools.tpl'), array(
'$title' => t('Design'),
'$who' => $who,
'$blocks' => t('Blocks'),
'$menus' => t('Menus'),
'$layout' => t('Layouts'),
'$pages' => t('Pages')
));
}
/* case insensitive in_array() */
function in_arrayi($needle, $haystack) {
return in_array(strtolower($needle), array_map('strtolower', $haystack));
}
function normalise_openid($s) {
return trim(str_replace(array('http://','https://'),array('',''),$s),'/');
}
2014-08-26 19:13:27 +00:00