dir/library/HTMLPurifier/Injector/AutoParagraph.php
Alexander Kampmann 355c42cb30 Merge branch 'master' of https://github.com/friendica/friendica
Conflicts:
	include/config.php
	update.php
2012-04-05 13:39:15 +02:00

346 lines
13 KiB
PHP

<?php
/**
* Injector that auto paragraphs text in the root node based on
* double-spacing.
* @todo Ensure all states are unit tested, including variations as well.
* @todo Make a graph of the flow control for this Injector.
*/
class HTMLPurifier_Injector_AutoParagraph extends HTMLPurifier_Injector
{
public $name = 'AutoParagraph';
public $needed = array('p');
private function _pStart() {
$par = new HTMLPurifier_Token_Start('p');
$par->armor['MakeWellFormed_TagClosedError'] = true;
return $par;
}
public function handleText(&$token) {
$text = $token->data;
// Does the current parent allow <p> tags?
if ($this->allowsElement('p')) {
if (empty($this->currentNesting) || strpos($text, "\n\n") !== false) {
// Note that we have differing behavior when dealing with text
// in the anonymous root node, or a node inside the document.
// If the text as a double-newline, the treatment is the same;
// if it doesn't, see the next if-block if you're in the document.
$i = $nesting = null;
if (!$this->forwardUntilEndToken($i, $current, $nesting) && $token->is_whitespace) {
// State 1.1: ... ^ (whitespace, then document end)
// ----
// This is a degenerate case
} else {
if (!$token->is_whitespace || $this->_isInline($current)) {
// State 1.2: PAR1
// ----
// State 1.3: PAR1\n\nPAR2
// ------------
// State 1.4: <div>PAR1\n\nPAR2 (see State 2)
// ------------
$token = array($this->_pStart());
$this->_splitText($text, $token);
} else {
// State 1.5: \n<hr />
// --
}
}
} else {
// State 2: <div>PAR1... (similar to 1.4)
// ----
// We're in an element that allows paragraph tags, but we're not
// sure if we're going to need them.
if ($this->_pLookAhead()) {
// State 2.1: <div>PAR1<b>PAR1\n\nPAR2
// ----
// Note: This will always be the first child, since any
// previous inline element would have triggered this very
// same routine, and found the double newline. One possible
// exception would be a comment.
$token = array($this->_pStart(), $token);
} else {
// State 2.2.1: <div>PAR1<div>
// ----
// State 2.2.2: <div>PAR1<b>PAR1</b></div>
// ----
}
}
// Is the current parent a <p> tag?
} elseif (
!empty($this->currentNesting) &&
$this->currentNesting[count($this->currentNesting)-1]->name == 'p'
) {
// State 3.1: ...<p>PAR1
// ----
// State 3.2: ...<p>PAR1\n\nPAR2
// ------------
$token = array();
$this->_splitText($text, $token);
// Abort!
} else {
// State 4.1: ...<b>PAR1
// ----
// State 4.2: ...<b>PAR1\n\nPAR2
// ------------
}
}
public function handleElement(&$token) {
// We don't have to check if we're already in a <p> tag for block
// tokens, because the tag would have been autoclosed by MakeWellFormed.
if ($this->allowsElement('p')) {
if (!empty($this->currentNesting)) {
if ($this->_isInline($token)) {
// State 1: <div>...<b>
// ---
// Check if this token is adjacent to the parent token
// (seek backwards until token isn't whitespace)
$i = null;
$this->backward($i, $prev);
if (!$prev instanceof HTMLPurifier_Token_Start) {
// Token wasn't adjacent
if (
$prev instanceof HTMLPurifier_Token_Text &&
substr($prev->data, -2) === "\n\n"
) {
// State 1.1.4: <div><p>PAR1</p>\n\n<b>
// ---
// Quite frankly, this should be handled by splitText
$token = array($this->_pStart(), $token);
} else {
// State 1.1.1: <div><p>PAR1</p><b>
// ---
// State 1.1.2: <div><br /><b>
// ---
// State 1.1.3: <div>PAR<b>
// ---
}
} else {
// State 1.2.1: <div><b>
// ---
// Lookahead to see if <p> is needed.
if ($this->_pLookAhead()) {
// State 1.3.1: <div><b>PAR1\n\nPAR2
// ---
$token = array($this->_pStart(), $token);
} else {
// State 1.3.2: <div><b>PAR1</b></div>
// ---
// State 1.3.3: <div><b>PAR1</b><div></div>\n\n</div>
// ---
}
}
} else {
// State 2.3: ...<div>
// -----
}
} else {
if ($this->_isInline($token)) {
// State 3.1: <b>
// ---
// This is where the {p} tag is inserted, not reflected in
// inputTokens yet, however.
$token = array($this->_pStart(), $token);
} else {
// State 3.2: <div>
// -----
}
$i = null;
if ($this->backward($i, $prev)) {
if (
!$prev instanceof HTMLPurifier_Token_Text
) {
// State 3.1.1: ...</p>{p}<b>
// ---
// State 3.2.1: ...</p><div>
// -----
if (!is_array($token)) $token = array($token);
array_unshift($token, new HTMLPurifier_Token_Text("\n\n"));
} else {
// State 3.1.2: ...</p>\n\n{p}<b>
// ---
// State 3.2.2: ...</p>\n\n<div>
// -----
// Note: PAR<ELEM> cannot occur because PAR would have been
// wrapped in <p> tags.
}
}
}
} else {
// State 2.2: <ul><li>
// ----
// State 2.4: <p><b>
// ---
}
}
/**
* Splits up a text in paragraph tokens and appends them
* to the result stream that will replace the original
* @param $data String text data that will be processed
* into paragraphs
* @param $result Reference to array of tokens that the
* tags will be appended onto
* @param $config Instance of HTMLPurifier_Config
* @param $context Instance of HTMLPurifier_Context
*/
private function _splitText($data, &$result) {
$raw_paragraphs = explode("\n\n", $data);
$paragraphs = array(); // without empty paragraphs
$needs_start = false;
$needs_end = false;
$c = count($raw_paragraphs);
if ($c == 1) {
// There were no double-newlines, abort quickly. In theory this
// should never happen.
$result[] = new HTMLPurifier_Token_Text($data);
return;
}
for ($i = 0; $i < $c; $i++) {
$par = $raw_paragraphs[$i];
if (trim($par) !== '') {
$paragraphs[] = $par;
} else {
if ($i == 0) {
// Double newline at the front
if (empty($result)) {
// The empty result indicates that the AutoParagraph
// injector did not add any start paragraph tokens.
// This means that we have been in a paragraph for
// a while, and the newline means we should start a new one.
$result[] = new HTMLPurifier_Token_End('p');
$result[] = new HTMLPurifier_Token_Text("\n\n");
// However, the start token should only be added if
// there is more processing to be done (i.e. there are
// real paragraphs in here). If there are none, the
// next start paragraph tag will be handled by the
// next call to the injector
$needs_start = true;
} else {
// We just started a new paragraph!
// Reinstate a double-newline for presentation's sake, since
// it was in the source code.
array_unshift($result, new HTMLPurifier_Token_Text("\n\n"));
}
} elseif ($i + 1 == $c) {
// Double newline at the end
// There should be a trailing </p> when we're finally done.
$needs_end = true;
}
}
}
// Check if this was just a giant blob of whitespace. Move this earlier,
// perhaps?
if (empty($paragraphs)) {
return;
}
// Add the start tag indicated by \n\n at the beginning of $data
if ($needs_start) {
$result[] = $this->_pStart();
}
// Append the paragraphs onto the result
foreach ($paragraphs as $par) {
$result[] = new HTMLPurifier_Token_Text($par);
$result[] = new HTMLPurifier_Token_End('p');
$result[] = new HTMLPurifier_Token_Text("\n\n");
$result[] = $this->_pStart();
}
// Remove trailing start token; Injector will handle this later if
// it was indeed needed. This prevents from needing to do a lookahead,
// at the cost of a lookbehind later.
array_pop($result);
// If there is no need for an end tag, remove all of it and let
// MakeWellFormed close it later.
if (!$needs_end) {
array_pop($result); // removes \n\n
array_pop($result); // removes </p>
}
}
/**
* Returns true if passed token is inline (and, ergo, allowed in
* paragraph tags)
*/
private function _isInline($token) {
return isset($this->htmlDefinition->info['p']->child->elements[$token->name]);
}
/**
* Looks ahead in the token list and determines whether or not we need
* to insert a <p> tag.
*/
private function _pLookAhead() {
$this->current($i, $current);
if ($current instanceof HTMLPurifier_Token_Start) $nesting = 1;
else $nesting = 0;
$ok = false;
while ($this->forwardUntilEndToken($i, $current, $nesting)) {
$result = $this->_checkNeedsP($current);
if ($result !== null) {
$ok = $result;
break;
}
}
return $ok;
}
/**
* Determines if a particular token requires an earlier inline token
* to get a paragraph. This should be used with _forwardUntilEndToken
*/
private function _checkNeedsP($current) {
if ($current instanceof HTMLPurifier_Token_Start){
if (!$this->_isInline($current)) {
// <div>PAR1<div>
// ----
// Terminate early, since we hit a block element
return false;
}
} elseif ($current instanceof HTMLPurifier_Token_Text) {
if (strpos($current->data, "\n\n") !== false) {
// <div>PAR1<b>PAR1\n\nPAR2
// ----
return true;
} else {
// <div>PAR1<b>PAR1...
// ----
}
}
return null;
}
}
// vim: et sw=4 sts=4