mirror of
https://github.com/friendica/friendica
synced 2024-11-10 03:02:54 +00:00
bb2diaspora: Horizontal ruler must have a leading new line for libertree
This commit is contained in:
parent
b1338548b0
commit
b01520babf
1 changed files with 5 additions and 274 deletions
|
@ -15,37 +15,24 @@ require_once("include/markdownify/markdownify.php");
|
|||
|
||||
function diaspora2bb($s) {
|
||||
|
||||
// for testing purposes: Collect raw markdown articles
|
||||
// $file = tempnam("/tmp/friendica/", "markdown");
|
||||
// file_put_contents($file, $s);
|
||||
|
||||
$s = html_entity_decode($s,ENT_COMPAT,'UTF-8');
|
||||
|
||||
// Too many new lines. So deactivated the following line
|
||||
// $s = str_replace("\r","\n",$s);
|
||||
// Simply remove cr.
|
||||
$s = str_replace("\r","",$s);
|
||||
|
||||
// <br/> is invalid. Replace it with the valid expression
|
||||
logger("diaspora2bb: 1: ".$s);
|
||||
$s = str_replace(array("<br/>", "</p>", "<p>"),array("<br />", "<br />", "<br />"),$s);
|
||||
logger("diaspora2bb: 2: ".$s);
|
||||
|
||||
$s = preg_replace('/\@\{(.+?)\; (.+?)\@(.+?)\}/','@[url=https://$3/u/$2]$1[/url]',$s);
|
||||
|
||||
// Escaping the hash tags - doesn't always seem to work
|
||||
// $s = preg_replace('/\#([^\s\#])/','\\#$1',$s);
|
||||
// This seems to work
|
||||
// Escaping the hash tags
|
||||
$s = preg_replace('/\#([^\s\#])/','#$1',$s);
|
||||
|
||||
$s = Markdown($s);
|
||||
|
||||
$s = str_replace('#','#',$s);
|
||||
// we seem to have double linebreaks
|
||||
// $s = str_replace("\n",'<br />',$s);
|
||||
|
||||
$s = html2bbcode($s);
|
||||
// $s = str_replace('*','*',$s);
|
||||
|
||||
// protect the recycle symbol from turning into a tag, but without unescaping angles and naked ampersands
|
||||
$s = str_replace('♲',html_entity_decode('♲',ENT_QUOTES,'UTF-8'),$s);
|
||||
|
@ -67,82 +54,12 @@ function diaspora2bb($s) {
|
|||
return $s;
|
||||
}
|
||||
|
||||
|
||||
function stripdcode_br_cb($s) {
|
||||
return '[code]' . str_replace('<br />', "\n\t", $s[1]) . '[/code]';
|
||||
}
|
||||
|
||||
|
||||
//////////////////////
|
||||
// The following "diaspora_ul" and "diaspora_ol" are only appropriate for the
|
||||
// pre-Markdownify conversion. If Markdownify isn't used, use the non-Markdownify
|
||||
// versions below
|
||||
//////////////////////
|
||||
/*
|
||||
function diaspora_ul($s) {
|
||||
// Replace "[*]" followed by any number (including zero) of
|
||||
// spaces by "* " to match Diaspora's list format
|
||||
if( strpos($s[0], "[list]") === 0 )
|
||||
return '<ul class="listbullet" style="list-style-type: circle;">' . preg_replace("/\[\*\]( *)/", "* ", $s[1]) . '</ul>';
|
||||
elseif( strpos($s[0], "[ul]") === 0 )
|
||||
return '<ul class="listbullet" style="list-style-type: circle;">' . preg_replace("/\[\*\]( *)/", "* ", $s[1]) . '</ul>';
|
||||
else
|
||||
return $s[0];
|
||||
}
|
||||
|
||||
|
||||
function diaspora_ol($s) {
|
||||
// A hack: Diaspora will create a properly-numbered ordered list even
|
||||
// if you use '1.' for each element of the list, like:
|
||||
// 1. First element
|
||||
// 1. Second element
|
||||
// 1. Third element
|
||||
if( strpos($s[0], "[list=1]") === 0 )
|
||||
return '<ul class="listdecimal" style="list-style-type: decimal;">' . preg_replace("/\[\*\]( *)/", "1. ", $s[1]) . '</ul>';
|
||||
elseif( strpos($s[0], "[list=i]") === 0 )
|
||||
return '<ul class="listlowerroman" style="list-style-type: lower-roman;">' . preg_replace("/\[\*\]( *)/", "1. ", $s[1]) . '</ul>';
|
||||
elseif( strpos($s[0], "[list=I]") === 0 )
|
||||
return '<ul class="listupperroman" style="list-style-type: upper-roman;">' . preg_replace("/\[\*\]( *)/", "1. ", $s[1]) . '</ul>';
|
||||
elseif( strpos($s[0], "[list=a]") === 0 )
|
||||
return '<ul class="listloweralpha" style="list-style-type: lower-alpha;">' . preg_replace("/\[\*\]( *)/", "1. ", $s[1]) . '</ul>';
|
||||
elseif( strpos($s[0], "[list=A]") === 0 )
|
||||
return '<ul class="listupperalpha" style="list-style-type: upper-alpha;">' . preg_replace("/\[\*\]( *)/", "1. ", $s[1]) . '</ul>';
|
||||
elseif( strpos($s[0], "[ol]") === 0 )
|
||||
return '<ul class="listdecimal" style="list-style-type: decimal;">' . preg_replace("/\[\*\]( *)/", "1. ", $s[1]) . '</ul>';
|
||||
else
|
||||
return $s[0];
|
||||
}
|
||||
*/
|
||||
|
||||
//////////////////////
|
||||
// Non-Markdownify versions of "diaspora_ol" and "diaspora_ul"
|
||||
//////////////////////
|
||||
function diaspora_ul($s) {
|
||||
// Replace "[\\*]" followed by any number (including zero) of
|
||||
// spaces by "* " to match Diaspora's list format
|
||||
return preg_replace("/\[\\\\\*\]( *)/", "* ", $s[1]);
|
||||
}
|
||||
|
||||
function diaspora_ol($s) {
|
||||
// A hack: Diaspora will create a properly-numbered ordered list even
|
||||
// if you use '1.' for each element of the list, like:
|
||||
// 1. First element
|
||||
// 1. Second element
|
||||
// 1. Third element
|
||||
return preg_replace("/\[\\\\\*\]( *)/", "1. ", $s[1]);
|
||||
}
|
||||
|
||||
|
||||
function bb2diaspora($Text,$preserve_nl = false, $fordiaspora = true) {
|
||||
|
||||
// Since Diaspora is creating a summary for links, this function removes them before posting
|
||||
if ($fordiaspora)
|
||||
$Text = bb_remove_share_information($Text);
|
||||
|
||||
// Re-enabling the converter again.
|
||||
// The bbcode parser now handles youtube-links (and the other stuff) correctly.
|
||||
// Additionally the html code is now fixed so that lists are now working.
|
||||
|
||||
/**
|
||||
* Transform #tags, strip off the [url] and replace spaces with underscore
|
||||
*/
|
||||
|
@ -154,21 +71,14 @@ function bb2diaspora($Text,$preserve_nl = false, $fordiaspora = true) {
|
|||
// Converting images with size parameters to simple images. Markdown doesn't know it.
|
||||
$Text = preg_replace("/\[img\=([0-9]*)x([0-9]*)\](.*?)\[\/img\]/ism", '[img]$3[/img]', $Text);
|
||||
|
||||
// the following was added on 10-January-2012 due to an inability of Diaspora's
|
||||
// new javascript markdown processor to handle links with images as the link "text"
|
||||
// It is not optimal and may be removed if this ability is restored in the future
|
||||
//if ($fordiaspora)
|
||||
// $Text = preg_replace("/\[url\=([^\[\]]*)\]\s*\[img\](.*?)\[\/img\]\s*\[\/url\]/ism",
|
||||
// "[url]$1[/url]\n[img]$2[/img]", $Text);
|
||||
|
||||
// Remove the avatar picture since that isn't looking good on the other side
|
||||
//$Text = preg_replace("/\[share(.*?)avatar\s?=\s?'.*?'\s?(.*?)\]\s?(.*?)\s?\[\/share\]\s?/ism","\n[share$1$2]$3[/share]",$Text);
|
||||
|
||||
// Convert it to HTML - don't try oembed
|
||||
if ($fordiaspora)
|
||||
$Text = bbcode($Text, $preserve_nl, false, 3);
|
||||
else
|
||||
else {
|
||||
$Text = bbcode($Text, $preserve_nl, false, 4);
|
||||
// Libertree doesn't convert a harizontal rule if there isn't a linefeed
|
||||
$Text = str_replace("<hr />", "\n<hr />", $Text);
|
||||
}
|
||||
|
||||
// Now convert HTML to Markdown
|
||||
$md = new Markdownify(false, false, false);
|
||||
|
@ -199,185 +109,6 @@ function bb2diaspora($Text,$preserve_nl = false, $fordiaspora = true) {
|
|||
// Remove all unconverted tags
|
||||
$Text = strip_tags($Text);
|
||||
|
||||
|
||||
/* Old routine
|
||||
|
||||
$ev = bbtoevent($Text);
|
||||
|
||||
// Replace any html brackets with HTML Entities to prevent executing HTML or script
|
||||
// Don't use strip_tags here because it breaks [url] search by replacing & with amp
|
||||
|
||||
$Text = str_replace("<", "<", $Text);
|
||||
$Text = str_replace(">", ">", $Text);
|
||||
|
||||
// If we find any event code, turn it into an event.
|
||||
// After we're finished processing the bbcode we'll
|
||||
// replace all of the event code with a reformatted version.
|
||||
|
||||
if($preserve_nl)
|
||||
$Text = str_replace(array("\n","\r"), array('',''),$Text);
|
||||
else
|
||||
// Remove the "return" character, as Diaspora uses only the "newline"
|
||||
// character, so having the "return" character can cause signature
|
||||
// failures
|
||||
$Text = str_replace("\r", "", $Text);
|
||||
|
||||
|
||||
// Set up the parameters for a URL search string
|
||||
$URLSearchString = "^\[\]";
|
||||
// Set up the parameters for a MAIL search string
|
||||
$MAILSearchString = $URLSearchString;
|
||||
|
||||
// Perform URL Search
|
||||
|
||||
// [img]pathtoimage[/img]
|
||||
|
||||
// the following was added on 10-January-2012 due to an inability of Diaspora's
|
||||
// new javascript markdown processor to handle links with images as the link "text"
|
||||
// It is not optimal and may be removed if this ability is restored in the future
|
||||
|
||||
$Text = preg_replace("/\[url\=([$URLSearchString]*)\]\[img\](.*?)\[\/img\]\[\/url\]/ism",
|
||||
'![' . t('image/photo') . '](' . '$2' . ')' . "\n" . '[' . t('link') . '](' . '$1' . ')', $Text);
|
||||
|
||||
$Text = preg_replace("/\[bookmark\]([$URLSearchString]*)\[\/bookmark\]/ism", '[$1]($1)', $Text);
|
||||
$Text = preg_replace("/\[bookmark\=([$URLSearchString]*)\](.*?)\[\/bookmark\]/ism", '[$2]($1)', $Text);
|
||||
|
||||
$Text = preg_replace("/\[url\]([$URLSearchString]*)\[\/url\]/ism", '[$1]($1)', $Text);
|
||||
$Text = preg_replace("/\#\[url\=([$URLSearchString]*)\](.*?)\[\/url\]/ism", '[#$2]($1)', $Text);
|
||||
$Text = preg_replace("/\[url\=([$URLSearchString]*)\](.*?)\[\/url\]/ism", '[$2]($1)', $Text);
|
||||
|
||||
|
||||
$Text = preg_replace("/\[img\](.*?)\[\/img\]/", '![' . t('image/photo') . '](' . '$1' . ')', $Text);
|
||||
$Text = preg_replace("/\[img\=(.*?)\](.*?)\[\/img\]/", '![' . t('image/photo') . '](' . '$2' . ')', $Text);
|
||||
|
||||
// Perform MAIL Search
|
||||
$Text = preg_replace("(\[mail\]([$MAILSearchString]*)\[/mail\])", '[$1](mailto:$1)', $Text);
|
||||
$Text = preg_replace("/\[mail\=([$MAILSearchString]*)\](.*?)\[\/mail\]/", '[$2](mailto:$1)', $Text);
|
||||
|
||||
$Text = str_replace('*', '\\*', $Text);
|
||||
$Text = str_replace('_', '\\_', $Text);
|
||||
|
||||
$Text = str_replace('`','\\`', $Text);
|
||||
|
||||
// Check for bold text
|
||||
$Text = preg_replace("(\[b\](.*?)\[\/b\])is",'**$1**',$Text);
|
||||
|
||||
// Check for italics text
|
||||
$Text = preg_replace("(\[i\](.*?)\[\/i\])is",'_$1_',$Text);
|
||||
|
||||
// Check for underline text
|
||||
// Replace with italics since Diaspora doesn't have underline
|
||||
$Text = preg_replace("(\[u\](.*?)\[\/u\])is",'_$1_',$Text);
|
||||
|
||||
// Check for strike-through text
|
||||
$Text = preg_replace("(\[s\](.*?)\[\/s\])is",'**[strike]**$1**[/strike]**',$Text);
|
||||
|
||||
// Check for over-line text
|
||||
// $Text = preg_replace("(\[o\](.*?)\[\/o\])is",'<span class="overline">$1</span>',$Text);
|
||||
|
||||
// Check for colored text
|
||||
// Remove color since Diaspora doesn't support it
|
||||
$Text = preg_replace("(\[color=(.*?)\](.*?)\[\/color\])is","$2",$Text);
|
||||
|
||||
// Check for sized text
|
||||
// Remove it since Diaspora doesn't support sizes very well
|
||||
$Text = preg_replace("(\[size=(.*?)\](.*?)\[\/size\])is","$2",$Text);
|
||||
|
||||
// Check for list text
|
||||
$endlessloop = 0;
|
||||
while ((((strpos($Text, "[/list]") !== false) && (strpos($Text, "[list") !== false)) ||
|
||||
((strpos($Text, "[/ol]") !== false) && (strpos($Text, "[ol]") !== false)) ||
|
||||
((strpos($Text, "[/ul]") !== false) && (strpos($Text, "[ul]") !== false)) ||
|
||||
((strpos($Text, "[/li]") !== false) && (strpos($Text, "[li]") !== false))) && (++$endlessloop < 20)) {
|
||||
$Text = preg_replace_callback("/\[list\](.*?)\[\/list\]/is", 'diaspora_ul', $Text);
|
||||
$Text = preg_replace_callback("/\[list=1\](.*?)\[\/list\]/is", 'diaspora_ol', $Text);
|
||||
$Text = preg_replace_callback("/\[list=i\](.*?)\[\/list\]/s",'diaspora_ol', $Text);
|
||||
$Text = preg_replace_callback("/\[list=I\](.*?)\[\/list\]/s", 'diaspora_ol', $Text);
|
||||
$Text = preg_replace_callback("/\[list=a\](.*?)\[\/list\]/s", 'diaspora_ol', $Text);
|
||||
$Text = preg_replace_callback("/\[list=A\](.*?)\[\/list\]/s", 'diaspora_ol', $Text);
|
||||
$Text = preg_replace_callback("/\[ul\](.*?)\[\/ul\]/is", 'diaspora_ul', $Text);
|
||||
$Text = preg_replace_callback("/\[ol\](.*?)\[\/ol\]/is", 'diaspora_ol', $Text);
|
||||
$Text = preg_replace("/\[li\]( *)(.*?)\[\/li\]/s", '* $2' ,$Text);
|
||||
}
|
||||
|
||||
// Just get rid of table tags since Diaspora doesn't support tables
|
||||
$Text = preg_replace("/\[th\](.*?)\[\/th\]/s", '$1' ,$Text);
|
||||
$Text = preg_replace("/\[td\](.*?)\[\/td\]/s", '$1' ,$Text);
|
||||
$Text = preg_replace("/\[tr\](.*?)\[\/tr\]/s", '$1' ,$Text);
|
||||
$Text = preg_replace("/\[table\](.*?)\[\/table\]/s", '$1' ,$Text);
|
||||
|
||||
$Text = preg_replace("/\[table border=(.*?)\](.*?)\[\/table\]/s", '$2' ,$Text);
|
||||
// $Text = preg_replace("/\[table border=0\](.*?)\[\/table\]/s", '<table border="0" >$1</table>' ,$Text);
|
||||
|
||||
|
||||
// $Text = str_replace("[*]", "<li>", $Text);
|
||||
|
||||
// Check for font change text
|
||||
// $Text = preg_replace("(\[font=(.*?)\](.*?)\[\/font\])","<span style=\"font-family: $1;\">$2</span>",$Text);
|
||||
|
||||
|
||||
$Text = preg_replace_callback("/\[code\](.*?)\[\/code\]/is",'stripdcode_br_cb',$Text);
|
||||
|
||||
// Check for [code] text
|
||||
$Text = preg_replace("/(\[code\])+(.*?)(\[\/code\])+/is","\t$2\n", $Text);
|
||||
|
||||
|
||||
|
||||
|
||||
// Declare the format for [quote] layout
|
||||
// $QuoteLayout = '<blockquote>$1</blockquote>';
|
||||
// Check for [quote] text
|
||||
$Text = preg_replace("/\[quote\](.*?)\[\/quote\]/is",">$1\n\n", $Text);
|
||||
$Text = preg_replace("/\[quote=(.*?)\](.*?)\[\/quote\]/is",">$2\n\n", $Text);
|
||||
|
||||
// Images
|
||||
|
||||
// html5 video and audio
|
||||
|
||||
$Text = preg_replace("/\[video\](.*?)\[\/video\]/", '$1', $Text);
|
||||
|
||||
$Text = preg_replace("/\[audio\](.*?)\[\/audio\]/", '$1', $Text);
|
||||
|
||||
// $Text = preg_replace("/\[iframe\](.*?)\[\/iframe\]/", '<iframe src="$1" width="425" height="350"><a href="$1">$1</a></iframe>', $Text);
|
||||
|
||||
// [img=widthxheight]image source[/img]
|
||||
// $Text = preg_replace("/\[img\=([0-9]*)x([0-9]*)\](.*?)\[\/img\]/", '<img src="$3" style="height:{$2}px; width:{$1}px;" >', $Text);
|
||||
|
||||
$Text = preg_replace("/\[youtube\]https?:\/\/www.youtube.com\/watch\?v\=(.*?)\[\/youtube\]/ism",'http://www.youtube.com/watch?v=$1',$Text);
|
||||
$Text = preg_replace("/\[youtube\]https?:\/\/www.youtube.com\/embed\/(.*?)\[\/youtube\]/ism",'http://www.youtube.com/watch?v=$1',$Text);
|
||||
$Text = preg_replace("/\[youtube\]https?:\/\/youtu.be\/(.*?)\[\/youtube\]/ism",'http://www.youtube.com/watch?v=$1',$Text);
|
||||
$Text = preg_replace("/\[youtube\]([A-Za-z0-9\-_=]+)(.*?)\[\/youtube\]/ism", 'http://www.youtube.com/watch?v=$1', $Text);
|
||||
|
||||
$Text = preg_replace("/\[vimeo\]https?:\/\/player.vimeo.com\/video\/([0-9]+)(.*?)\[\/vimeo\]/ism",'http://vimeo.com/$1',$Text);
|
||||
$Text = preg_replace("/\[vimeo\]https?:\/\/vimeo.com\/([0-9]+)(.*?)\[\/vimeo\]/ism",'http://vimeo.com/$1',$Text);
|
||||
$Text = preg_replace("/\[vimeo\]([0-9]+)(.*?)\[\/vimeo\]/ism", 'http://vimeo.com/$1',$Text);
|
||||
|
||||
|
||||
$Text = str_replace('[nosmile]','',$Text);
|
||||
|
||||
// oembed tag
|
||||
// $Text = oembed_bbcode2html($Text);
|
||||
|
||||
// If we found an event earlier, strip out all the event code and replace with a reformatted version.
|
||||
|
||||
if(x($ev,'start')) {
|
||||
|
||||
$sub = format_event_diaspora($ev);
|
||||
|
||||
$Text = preg_replace("/\[event\-summary\](.*?)\[\/event\-summary\]/is",'',$Text);
|
||||
$Text = preg_replace("/\[event\-description\](.*?)\[\/event\-description\]/is",'',$Text);
|
||||
$Text = preg_replace("/\[event\-start\](.*?)\[\/event\-start\]/is",$sub,$Text);
|
||||
$Text = preg_replace("/\[event\-finish\](.*?)\[\/event\-finish\]/is",'',$Text);
|
||||
$Text = preg_replace("/\[event\-location\](.*?)\[\/event\-location\]/is",'',$Text);
|
||||
$Text = preg_replace("/\[event\-adjust\](.*?)\[\/event\-adjust\]/is",'',$Text);
|
||||
}
|
||||
|
||||
$Text = preg_replace("/\<(.*?)(src|href)=(.*?)\&\;(.*?)\>/ism",'<$1$2=$3&$4>',$Text);
|
||||
|
||||
$Text = preg_replace_callback('/\[(.*?)\]\((.*?)\)/ism','unescape_underscores_in_links',$Text);
|
||||
|
||||
*/
|
||||
|
||||
// Remove any leading or trailing whitespace, as this will mess up
|
||||
// the Diaspora signature verification and cause the item to disappear
|
||||
$Text = trim($Text);
|
||||
|
|
Loading…
Reference in a new issue