diff options
author | Friendika <info@friendika.com> | 2011-01-17 19:50:18 -0800 |
---|---|---|
committer | Friendika <info@friendika.com> | 2011-01-17 19:50:18 -0800 |
commit | fa1f2ce8f38962b97c4a6a327c5f50e238eabb9b (patch) | |
tree | 6ff4478f2ab38ad446e5d436ed284b006786bf48 /include/items.php | |
parent | 050618a2b690ad2e2d8e84afbada8b12f9da0f01 (diff) | |
download | volse-hubzilla-fa1f2ce8f38962b97c4a6a327c5f50e238eabb9b.tar.gz volse-hubzilla-fa1f2ce8f38962b97c4a6a327c5f50e238eabb9b.tar.bz2 volse-hubzilla-fa1f2ce8f38962b97c4a6a327c5f50e238eabb9b.zip |
fix some linebreak issues
Diffstat (limited to 'include/items.php')
-rw-r--r-- | include/items.php | 23 |
1 files changed, 14 insertions, 9 deletions
diff --git a/include/items.php b/include/items.php index c076e0fbf..beed024a1 100644 --- a/include/items.php +++ b/include/items.php @@ -381,7 +381,7 @@ function get_atom_elements($feed,$item) { // It isn't certain at this point whether our content is plaintext or html and we'd be foolish to trust // the content type. Our own network only emits text normally, though it might have been converted to - // html if we used a pubsubhubbub transport. But if we see even one html open tag in our text, we will + // html if we used a pubsubhubbub transport. But if we see even one html tag in our text, we will // have to assume it is all html and needs to be purified. // It doesn't matter all that much security wise - because before this content is used anywhere, we are @@ -390,7 +390,7 @@ function get_atom_elements($feed,$item) { // html. - if(strpos($res['body'],'<')) { + if((strpos($res['body'],'<')) || (strpos($res['body'],'>'))) { $res['body'] = preg_replace('#<object[^>]+>.+?' . 'http://www.youtube.com/((?:v|cp)/[A-Za-z0-9\-_=]+).+?</object>#s', '[youtube]$1[/youtube]', $res['body']); @@ -404,11 +404,12 @@ function get_atom_elements($feed,$item) { $purifier = new HTMLPurifier($config); $res['body'] = $purifier->purify($res['body']); - } + $res['body'] = html2bbcode($res['body']); + } + else + $res['body'] = escape_tags($res['body']); - $res['body'] = html2bbcode($res['body']); - $allow = $item->get_item_tags(NAMESPACE_DFRN,'comment-allow'); if($allow && $allow[0]['data'] == 1) @@ -496,7 +497,7 @@ function get_atom_elements($feed,$item) { $body = $rawobj[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['summary'][0]['data']; // preserve a copy of the original body content in case we later need to parse out any microformat information, e.g. events $res['object'] .= '<orig>' . xmlify($body) . '</orig>' . "\n"; - if(strpos($body,'<')) { + if((strpos($body,'<')) || (strpos($body,'>'))) { $body = preg_replace('#<object[^>]+>.+?' . 'http://www.youtube.com/((?:v|cp)/[A-Za-z0-9\-_=]+).+?</object>#s', '[youtube]$1[/youtube]', $body); @@ -506,9 +507,11 @@ function get_atom_elements($feed,$item) { $purifier = new HTMLPurifier($config); $body = $purifier->purify($body); + $body = html2bbcode($body); } + else + $body = escape_tags($body); - $body = html2bbcode($body); $res['object'] .= '<content>' . $body . '</content>' . "\n"; } @@ -535,7 +538,7 @@ function get_atom_elements($feed,$item) { $body = $rawobj[0]['child'][SIMPLEPIE_NAMESPACE_ATOM_10]['summary'][0]['data']; // preserve a copy of the original body content in case we later need to parse out any microformat information, e.g. events $res['object'] .= '<orig>' . xmlify($body) . '</orig>' . "\n"; - if(strpos($body,'<')) { + if((strpos($body,'<')) || (strpos($body,'>'))) { $body = preg_replace('#<object[^>]+>.+?' . 'http://www.youtube.com/((?:v|cp)/[A-Za-z0-9\-_=]+).+?</object>#s', '[youtube]$1[/youtube]', $body); @@ -545,9 +548,11 @@ function get_atom_elements($feed,$item) { $purifier = new HTMLPurifier($config); $body = $purifier->purify($body); + $body = html2bbcode($body); } + else + $body = escape_tags($body); - $body = html2bbcode($body); $res['target'] .= '<content>' . $body . '</content>' . "\n"; } |