tags var $block_elements = 'div|blockquote|pre|code|h\d|script|ol|ul'; // Elements that should not have

and
tags within them. var $skip_elements = 'pre|ol|ul'; // Tags we want the parser to completely ignore when splitting the string. var $ignore_elements = 'a|b|i|em|strong|span|img|li'; /** * Main Processing Function * */ function convert($str) { if ($str == '') { return ''; } $str = ' '.$str.' '; // Standardize Newlines to make matching easier if (strpos($str, "\r") !== FALSE) { $str = str_replace(array("\r\n", "\r"), "\n", $str); } /* * Reduce line breaks * * If there are more than two consecutive line * breaks we'll compress them down to a maximum * of two since there's no benefit to more. * */ $str = preg_replace("/\n\n+/", "\n\n", $str); /* * Convert quotes within tags to temporary marker * * We don't want quotes converted within * tags so we'll temporarily convert them to * {@DQ} and {@SQ} * */ if (preg_match_all("#\<.+?>#si", $str, $matches)) { for ($i = 0; $i < count($matches['0']); $i++) { $str = str_replace($matches['0'][$i], str_replace(array("'",'"'), array('{@SQ}', '{@DQ}'), $matches['0'][$i]), $str); } } /* * Add closing/opening paragraph tags before/after "block" elements * * Since block elements (like ,

, etc.) do not get
		 * wrapped in paragraph tags we will add a closing 

tag just before * each block element starts and an opening

tag right after the block element * ends. Later on we'll do some further clean up. * */ $str = preg_replace("#(<)(".$this->block_elements.")(.*?>)#", "

\\1\\2\\3", $str); $str = preg_replace("#(block_elements.")(.*?>)#", "\\1\\2\\3

", $str); /* * Convert "ignore" tags to temporary marker * * The parser splits out the string at every tag * it encounters. Certain inline tags, like image * tags, links, span tags, etc. will be adversely * affected if they are split out so we'll convert * the opening < temporarily to: {@TAG} * */ $str = preg_replace("#<(/*)(".$this->ignore_elements.")#i", "{@TAG}\\1\\2", $str); /* * Split the string at every tag * * This creates an array with this prototype: * * [array] * { * [0] = * [1] = Content contained between the tags * [2] = * Etc... * } * */ $chunks = preg_split('/(<(?:[^<>]+(?:"[^"]*"|\'[^\']*\')?)+>)/', $str, -1, PREG_SPLIT_DELIM_CAPTURE|PREG_SPLIT_NO_EMPTY); /* * Build our finalized string * * We'll cycle through the array, skipping tags, * and processing the contained text * */ $str = ''; $process = TRUE; foreach ($chunks as $chunk) { /* * Are we dealing with a tag? * * If so, we'll skip the processing for this cycle. * Well also set the "process" flag which allows us * to skip

 tags and a few other things.
			 *
			 */
			if (preg_match("#<(/*)(".$this->block_elements.").*?\>#", $chunk, $match))
			{
				if (preg_match("#".$this->skip_elements."#", $match['2']))
				{
					$process =  ($match['1'] == '/') ? TRUE : FALSE;		
				}
		
				$str .= $chunk;
				continue;
			}
		
			if ($process == FALSE)
			{
				$str .= $chunk;
				continue;
			}
			
			//  Convert Newlines into 

and
tags $str .= $this->format_newlines($chunk); } // FINAL CLEAN UP // IMPORTANT: DO NOT ALTER THE ORDER OF THE ITEMS BELOW! /* * Clean up paragraph tags before/after "block" elements * * Earlier we added

tags before/after block level elements. * Then, we added paragraph tags around double line breaks. This * potentially created incorrectly formatted paragraphs so we'll * clean it up here. * */ $str = preg_replace("#

({@TAG}.*?)(".$this->block_elements.")(.*?>)#", "\\1\\2\\3", $str); $str = preg_replace("#({@TAG}/.*?)(".$this->block_elements.")(.*?>)

#", "\\1\\2\\3", $str); // Convert Quotes and other characters $str = $this->format_characters($str); // Fix an artifact that happens during the paragraph replacement $str = preg_replace('#(

\n*

)#', '', $str); // If the user submitted their own paragraph tags with class data // in them we will retain them instead of using our tags. $str = preg_replace('#()

#', "\\1", $str); // Final clean up $str = str_replace( array( '

', '

', '

', '

', '{@TAG}', '{@DQ}', '{@SQ}', '

' ), array( '

', '

', '

', '

', '<', '"', "'", '' ), $str ); return $str; } // -------------------------------------------------------------------- /** * Format Characters * * This function mainly converts double and single quotes * to entities, but since these are directional, it does * it based on some rules. It also converts em-dashes * and a couple other things. */ function format_characters($str) { $table = array( ' "' => " “", '" ' => "” ", " '" => " ‘", "' " => "’ ", '>"' => ">“", '"<' => "”<", ">'" => ">‘", "'<" => "’<", "\"." => "”.", "\"," => "”,", "\";" => "”;", "\":" => "”:", "\"!" => "”!", "\"?" => "”?", ". " => ".  ", "? " => "?  ", "! " => "!  ", ": " => ":  ", ); // These deal with quotes within quotes, like: "'hi here'" $start = 0; $space = array("\n", "\t", " "); while(TRUE) { $current = strpos(substr($str, $start), "\"'"); if ($current === FALSE) break; $one_before = substr($str, $start+$current-1, 1); $one_after = substr($str, $start+$current+2, 1); if ( ! in_array($one_after, $space, TRUE) && $one_after != "<") { $str = str_replace( $one_before."\"'".$one_after, $one_before."“‘".$one_after, $str); } elseif ( ! in_array($one_before, $space, TRUE) && (in_array($one_after, $space, TRUE) OR $one_after == '<')) { $str = str_replace( $one_before."\"'".$one_after, $one_before."”’".$one_after, $str); } $start = $start+$current+2; } $start = 0; while(TRUE) { $current = strpos(substr($str, $start), "'\""); if ($current === FALSE) break; $one_before = substr($str, $start+$current-1, 1); $one_after = substr($str, $start+$current+2, 1); if ( in_array($one_before, $space, TRUE) && ! in_array($one_after, $space, TRUE) && $one_after != "<") { $str = str_replace( $one_before."'\"".$one_after, $one_before."‘“".$one_after, $str); } elseif ( ! in_array($one_before, $space, TRUE) && $one_before != ">") { $str = str_replace( $one_before."'\"".$one_after, $one_before."’”".$one_after, $str); } $start = $start+$current+2; } // Are there quotes within a word, as in: ("something") if (preg_match_all("/(.)\"(\S+?)\"(.)/", $str, $matches)) { for ($i=0, $s=sizeof($matches['0']); $i < $s; ++$i) { if ( ! in_array($matches['1'][$i], $space, TRUE) && ! in_array($matches['3'][$i], $space, TRUE)) { $str = str_replace( $matches['0'][$i], $matches['1'][$i]."“".$matches['2'][$i]."”".$matches['3'][$i], $str); } } } if (preg_match_all("/(.)\'(\S+?)\'(.)/", $str, $matches)) { for ($i=0, $s=sizeof($matches['0']); $i < $s; ++$i) { if ( ! in_array($matches['1'][$i], $space, TRUE) && ! in_array($matches['3'][$i], $space, TRUE)) { $str = str_replace( $matches['0'][$i], $matches['1'][$i]."‘".$matches['2'][$i]."’".$matches['3'][$i], $str); } } } // How about one apostrophe, as in Rick's $start = 0; while(TRUE) { $current = strpos(substr($str, $start), "'"); if ($current === FALSE) break; $one_before = substr($str, $start+$current-1, 1); $one_after = substr($str, $start+$current+1, 1); if ( ! in_array($one_before, $space, TRUE) && ! in_array($one_after, $space, TRUE)) { $str = str_replace( $one_before."'".$one_after, $one_before."’".$one_after, $str); } $start = $start+$current+2; } // Em-dashes $start = 0; while(TRUE) { $current = strpos(substr($str, $start), "--"); if ($current === FALSE) break; $one_before = substr($str, $start+$current-1, 1); $one_after = substr($str, $start+$current+2, 1); $two_before = substr($str, $start+$current-2, 1); $two_after = substr($str, $start+$current+3, 1); if (( ! in_array($one_before, $space, TRUE) && ! in_array($one_after, $space, TRUE)) OR ( ! in_array($two_before, $space, TRUE) && ! in_array($two_after, $space, TRUE) && $one_before == ' ' && $one_after == ' ') ) { $str = str_replace( $two_before.$one_before."--".$one_after.$two_after, $two_before.trim($one_before)."—".trim($one_after).$two_after, $str); } $start = $start+$current+2; } // Ellipsis $str = preg_replace("#(\w)\.\.\.(\s|
|

)#", "\\1…\\2", $str); $str = preg_replace("#(\s|
|

)\.\.\.(\w)#", "\\1…\\2", $str); // Run the translation array we defined above $str = str_replace(array_keys($table), array_values($table), $str); // If there are any stray double quotes we'll catch them here $start = 0; while(TRUE) { $current = strpos(substr($str, $start), '"'); if ($current === FALSE) break; $one_before = substr($str, $start+$current-1, 1); $one_after = substr($str, $start+$current+1, 1); if ( ! in_array($one_after, $space, TRUE)) { $str = str_replace( $one_before.'"'.$one_after, $one_before."“".$one_after, $str); } elseif( ! in_array($one_before, $space, TRUE)) { $str = str_replace( $one_before."'".$one_after, $one_before."”".$one_after, $str); } $start = $start+$current+2; } $start = 0; while(TRUE) { $current = strpos(substr($str, $start), "'"); if ($current === FALSE) break; $one_before = substr($str, $start+$current-1, 1); $one_after = substr($str, $start+$current+1, 1); if ( ! in_array($one_after, $space, TRUE)) { $str = str_replace( $one_before."'".$one_after, $one_before."‘".$one_after, $str); } elseif( ! in_array($one_before, $space, TRUE)) { $str = str_replace( $one_before."'".$one_after, $one_before."’".$one_after, $str); } $start = $start+$current+2; } return $str; } // -------------------------------------------------------------------- /** * Format Newlines * * Converts newline characters into either

tags or
* */ function format_newlines($str) { if ($str == '') { return $str; } if (strpos($str, "\n") === FALSE) { return '

'.$str.'

'; } $str = str_replace("\n\n", "

\n\n

", $str); $str = preg_replace("/([^\n])(\n)([^\n])/", "\\1
\\2\\3", $str); return '

'.$str.'

'; } } // END Typography Class /* End of file Typography.php */ /* Location: ./system/libraries/Typography.php */