separate compiling from parsing

This commit is contained in:
Emanuil 2013-09-24 01:19:17 +03:00
parent d306ee3db5
commit f8119fa3cb
1 changed files with 274 additions and 202 deletions

View File

@ -89,10 +89,12 @@ class Parsedown
# ~
$text = trim($text, "\n");
$text = preg_replace('/\n\s*\n/', "\n\n", $text);
$text = trim($text, "\n");
$text = $this->parse_lines($text);
$lines = explode("\n", $text);
$text = $this->parse_block_elements($lines);
# Decodes escape sequences (leaves out backslashes).
@ -110,270 +112,340 @@ class Parsedown
# Private Methods
#
private function parse_lines($text, $context = null)
private function parse_block_elements(array $lines, $context = '')
{
$lines = explode("\n", $text);
$lines []= null;
$elements = array();
$line_count = count($lines);
$element = array(
'type' => '',
);
$markup = '';
foreach ($lines as $index => $line)
foreach ($lines as $line)
{
# ~
# Empty
if (isset($line) and $line !== '' and $line[0] >= 'A')
if ($line === '')
{
$simple_line = $line;
$element['interrupted'] = true;
unset($line);
}
# Setext Heading (-)
if (isset($line) and $line !== '' and isset($paragraph) and preg_match('/^[-]+[ ]*$/', $line))
{
$setext_heading_text = $this->parse_inline_elements($paragraph);
$markup .= '<h2>'.$setext_heading_text.'</h2>'."\n";
unset($paragraph, $line);
$element['type'] === 'code' and $element['text'] .= "\n";
continue;
}
# Rule
# Lazy Blockquote
if (isset($line) and $line !== '' and preg_match('/^[ ]{0,3}([-*_])([ ]{0,2}\1){2,}[ ]*$/', $line))
if ($element['type'] === 'blockquote' and ! isset($element['interrupted']))
{
$rule = true;
$line = preg_replace('/^[ ]*>[ ]?/', '', $line);
unset($line);
}
elseif (isset($rule))
{
$markup .= '<hr />'."\n";
$element['lines'] []= $line;
unset($rule);
continue;
}
# List
# Lazy List Item
# Unlike other types, consequent lines of type "list items" may not
# belong to the same block.
if (isset($line) and $line !== '' and preg_match('/^([ ]{0,3})(\d+[.]|[*+-])[ ](.*)/', $line, $matches)) # list item
if ($element['type'] === 'li')
{
$list_item_indentation = $matches[1];
$list_item_type = ($matches[2] === '-' or $matches[2] === '+' or $matches[2] === '*')
? 'ul'
: 'ol';
if (isset($list_items)) # subsequent
if (preg_match('/^([ ]{0,3})(\d+[.]|[*+-])[ ](.*)/', $line, $matches))
{
if ($list_item_indentation === $list_indentation and $list_item_type === $list_type)
if ($element['indentation'] !== $matches[1])
{
$list_items []= $list_item;
$list_item = $matches[3];
$element['lines'] []= $line;
}
else
{
$list_item .= "\n".$line;
unset($element['last']);
$elements []= $element;
$element = array(
'type' => 'li',
'indentation' => $matches[1],
'last' => true,
'lines' => array(
preg_replace('/^[ ]{0,4}/', '', $matches[3]),
),
);
}
}
else # first
{
$list_indentation = $list_item_indentation;
$list_type = $list_item_type;
$list_item = $matches[3];
$list_items = array();
continue;
}
unset($line);
}
elseif (isset($list_items)) # incomplete list item
{
if (isset($line) and ($line === '' or $line[0] === ' '))
if (isset($element['interrupted']))
{
$line and $line = preg_replace('/^[ ]{0,4}/', '', $line);;
$list_item .= "\n".$line;
unset($line);
}
else # line is consumed or does not belong to the list item
{
$list_item = rtrim($list_item, "\n");
$list_items []= $list_item;
$markup .= '<'.$list_type.'>'."\n";
foreach ($list_items as $list_item)
if ($line[0] === ' ')
{
$list_item_text = strpos($list_item, "\n") !== false
? $this->parse_lines($list_item, 'li')
: $this->parse_inline_elements($list_item);
$element['lines'] []= '';
$markup .= '<li>'.$list_item_text.'</li>'."\n";
$line = preg_replace('/^[ ]{0,4}/', '', $line);;
$element['lines'] []= $line;
continue;
}
$markup .= '</'.$list_type.'>'."\n";
unset($list_items);
}
}
# Code Block
if (isset($line) and $line !== '' and preg_match('/^[ ]{4}(.*)/', $line, $matches))
{
if (isset($code_block))
{
$code_block .= "\n".$matches[1];
}
else
{
$code_block = $matches[1];
}
unset($line);
}
elseif (isset($code_block))
{
if (isset($line) and $line === '')
{
$code_block .= "\n";
$line = preg_replace('/^[ ]{0,4}/', '', $line);;
$element['lines'] []= $line;
# » continue;
}
else
{
$code_block = rtrim($code_block);
$code_block_text = htmlentities($code_block, ENT_NOQUOTES);
# Decodes encoded escape sequences if present.
strpos($code_block_text, "\x1A\\") !== FALSE and $code_block_text = strtr($code_block_text, $this->escape_sequence_map);
$markup .= '<pre><code>'.$code_block_text.'</code></pre>'."\n";
unset($code_block);
continue;
}
}
# Blockquote
# Quick Paragraph
if (isset($line) and $line !== '' and preg_match('/^[ ]*>[ ]?(.*)/', $line, $matches))
if ($line[0] >= 'A')
{
if (isset($blockquote))
{
$blockquote .= "\n".$matches[1];
}
else
{
$blockquote = $matches[1];
}
unset($line);
}
elseif (isset($blockquote))
{
if (isset($line) and $line === '')
{
$blockquote .= "\n";
}
else
{
$blockquote = $this->parse_lines($blockquote);
$markup .= '<blockquote>'."\n".$blockquote.'</blockquote>'."\n";
unset($blockquote);
}
goto paragraph; # trust me
}
# Atx Heading
# Setext Header (---)
if (isset($line) and $line !== '' and $line[0] === '#' and preg_match('/^(#{1,6})[ ]*(.+?)[ ]*#*$/', $line, $matches))
if ($element['type'] === 'p' and preg_match('/^[-]+[ ]*$/', $line))
{
$atx_heading_level = strlen($matches[1]);
$atx_heading = $this->parse_inline_elements($matches[2]);
unset($line);
}
elseif (isset($atx_heading))
{
$markup .= '<h'.$atx_heading_level.'>'.$atx_heading.'</h'.$atx_heading_level.'>'."\n";
unset($atx_heading);
}
# Setext Heading (=)
if (isset($line) and $line !== '' and isset($paragraph) and preg_match('/^[=]+[ ]*$/', $line))
{
$setext_heading_text = $this->parse_inline_elements($paragraph);
$markup .= '<h1>'.$setext_heading_text.'</h1>'."\n";
unset($paragraph, $line);
$element['type'] = 'h.';
$element['level'] = 2;
continue;
}
# Paragraph
# Horizontal Rule
if (isset($simple_line))
if (preg_match('/^[ ]{0,3}([-*_])([ ]{0,2}\1){2,}[ ]*$/', $line))
{
$line = $simple_line;
$elements []= $element;
unset($simple_line);
$element = array(
'type' => 'hr',
);
continue;
}
if (isset($line) and $line !== '')
# List Item
if (preg_match('/^([ ]{0,3})(\d+[.]|[*+-])[ ](.*)/', $line, $matches))
{
substr($line, -2) === ' ' and $line = substr_replace($line, '<br />', -2);
if (isset($paragraph))
{
$paragraph .= "\n".$line;
}
else
{
$paragraph = $line;
}
}
elseif (isset($paragraph))
{
$paragraph_text = $this->parse_inline_elements($paragraph);
$elements []= $element;
if ($context === 'li')
$element = array(
'type' => 'li',
'ordered' => isset($matches[2][1]),
'indentation' => $matches[1],
'last' => true,
'lines' => array(
preg_replace('/^[ ]{0,4}/', '', $matches[3]),
),
);
continue;
}
# Code
if (preg_match('/^[ ]{4}(.*)/', $line, $matches))
{
if ($element['type'] === 'code')
{
if ( ! $markup and $index + 1 === $line_count)
$element['text'] .= "\n".$matches[1];
}
else
{
$elements []= $element;
$element = array(
'type' => 'code',
'text' => $matches[1],
);
}
continue;
}
# Atx Header (#)
if ($line[0] === '#' and preg_match('/^(#{1,6})[ ]*(.+?)[ ]*#*$/', $line, $matches))
{
$elements []= $element;
$level = strlen($matches[1]);
$element = array(
'type' => 'h.',
'text' => $matches[2],
'level' => $level,
);
continue;
}
# Blockquote
if (preg_match('/^[ ]*>[ ]?(.*)/', $line, $matches))
{
if ($element['type'] === 'blockquote')
{
if (isset($element['interrupted']))
{
$text_is_simple = true;
}
else
{
$markup or $markup .= "\n";
$element['lines'] []= '';
unset($element['interrupted']);
}
$markup .= isset($text_is_simple)
? $paragraph_text
: '<p>'.$paragraph_text.'</p>'."\n";
$element['lines'] []= $matches[1];
}
else
else
{
$markup .= '<p>'.$paragraph_text.'</p>'."\n";
$elements []= $element;
$element = array(
'type' => 'blockquote',
'lines' => array(
$matches[1],
),
);
}
unset($paragraph);
continue;
}
# Setext Header (===)
if ($element['type'] === 'p' and preg_match('/^[=]+[ ]*$/', $line))
{
$element['type'] = 'h.';
$element['level'] = 1;
continue;
}
# ~
paragraph:
if ($element['type'] === 'p')
{
if (isset($element['interrupted']))
{
$elements []= $element;
$element['text'] = $line;
unset($element['interrupted']);
}
else
{
$element['text'] .= "\n".$line;
}
}
else
{
$elements []= $element;
$element = array(
'type' => 'p',
'text' => $line,
);
}
}
$elements []= $element;
array_shift($elements);
#
# ~
#
$markup = '';
foreach ($elements as $index => $element)
{
switch ($element['type'])
{
case 'li':
if (isset($element['ordered'])) # first
{
$list_type = $element['ordered'] ? 'ol' : 'ul';
$markup .= '<'.$list_type.'>'."\n";
}
if (isset($element['interrupted']) and ! isset($element['last']))
{
$element['lines'] []= '';
}
$text = $this->parse_block_elements($element['lines'], 'li');
$markup .= '<li>'.$text.'</li>'."\n";
isset($element['last']) and $markup .= '</'.$list_type.'>'."\n";
break;
case 'p':
$text = $this->parse_inline_elements($element['text']);
$text = preg_replace('/[ ]{2}\n/', '<br />'."\n", $text);
if ($context === 'li' and $index === 0)
{
if (isset($element['interrupted']))
{
$markup .= "\n".'<p>'.$text.'</p>'."\n";
}
else
{
$markup .= $text;
}
}
else
{
$markup .= '<p>'.$text.'</p>'."\n";
}
break;
case 'code':
$text = rtrim($element['text'], "\n");
$text = htmlentities($text, ENT_NOQUOTES);
strpos($text, "\x1A\\") !== FALSE and $text = strtr($text, $this->escape_sequence_map);
$markup .= '<pre><code>'.$text.'</code></pre>'."\n";
break;
case 'blockquote':
$text = $this->parse_block_elements($element['lines']);
$markup .= '<blockquote>'."\n".$text.'</blockquote>'."\n";
break;
case 'h.':
$text = $this->parse_inline_elements($element['text']);
$markup .= '<h'.$element['level'].'>'.$text.'</h'.$element['level'].'>'."\n";
break;
case 'hr':
$markup .= '<hr />'."\n";
break;
}
}