mirror of
https://github.com/renbaoshuo/S2OJ.git
synced 2024-11-23 12:38:40 +00:00
225 lines
7.0 KiB
PHP
225 lines
7.0 KiB
PHP
|
<?php
|
||
|
|
||
|
/**
|
||
|
* Definition for tables. The general idea is to extract out all of the
|
||
|
* essential bits, and then reconstruct it later.
|
||
|
*
|
||
|
* This is a bit confusing, because the DTDs and the W3C
|
||
|
* validators seem to disagree on the appropriate definition. The
|
||
|
* DTD claims:
|
||
|
*
|
||
|
* (CAPTION?, (COL*|COLGROUP*), THEAD?, TFOOT?, TBODY+)
|
||
|
*
|
||
|
* But actually, the HTML4 spec then has this to say:
|
||
|
*
|
||
|
* The TBODY start tag is always required except when the table
|
||
|
* contains only one table body and no table head or foot sections.
|
||
|
* The TBODY end tag may always be safely omitted.
|
||
|
*
|
||
|
* So the DTD is kind of wrong. The validator is, unfortunately, kind
|
||
|
* of on crack.
|
||
|
*
|
||
|
* The definition changed again in XHTML1.1; and in my opinion, this
|
||
|
* formulation makes the most sense.
|
||
|
*
|
||
|
* caption?, ( col* | colgroup* ), (( thead?, tfoot?, tbody+ ) | ( tr+ ))
|
||
|
*
|
||
|
* Essentially, we have two modes: thead/tfoot/tbody mode, and tr mode.
|
||
|
* If we encounter a thead, tfoot or tbody, we are placed in the former
|
||
|
* mode, and we *must* wrap any stray tr segments with a tbody. But if
|
||
|
* we don't run into any of them, just have tr tags is OK.
|
||
|
*/
|
||
|
class HTMLPurifier_ChildDef_Table extends HTMLPurifier_ChildDef
|
||
|
{
|
||
|
/**
|
||
|
* @type bool
|
||
|
*/
|
||
|
public $allow_empty = false;
|
||
|
|
||
|
/**
|
||
|
* @type string
|
||
|
*/
|
||
|
public $type = 'table';
|
||
|
|
||
|
/**
|
||
|
* @type array
|
||
|
*/
|
||
|
public $elements = array(
|
||
|
'tr' => true,
|
||
|
'tbody' => true,
|
||
|
'thead' => true,
|
||
|
'tfoot' => true,
|
||
|
'caption' => true,
|
||
|
'colgroup' => true,
|
||
|
'col' => true
|
||
|
);
|
||
|
|
||
|
public function __construct()
|
||
|
{
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* @param array $children
|
||
|
* @param HTMLPurifier_Config $config
|
||
|
* @param HTMLPurifier_Context $context
|
||
|
* @return array
|
||
|
*/
|
||
|
public function validateChildren($children, $config, $context)
|
||
|
{
|
||
|
if (empty($children)) {
|
||
|
return false;
|
||
|
}
|
||
|
|
||
|
// only one of these elements is allowed in a table
|
||
|
$caption = false;
|
||
|
$thead = false;
|
||
|
$tfoot = false;
|
||
|
|
||
|
// whitespace
|
||
|
$initial_ws = array();
|
||
|
$after_caption_ws = array();
|
||
|
$after_thead_ws = array();
|
||
|
$after_tfoot_ws = array();
|
||
|
|
||
|
// as many of these as you want
|
||
|
$cols = array();
|
||
|
$content = array();
|
||
|
|
||
|
$tbody_mode = false; // if true, then we need to wrap any stray
|
||
|
// <tr>s with a <tbody>.
|
||
|
|
||
|
$ws_accum =& $initial_ws;
|
||
|
|
||
|
foreach ($children as $node) {
|
||
|
if ($node instanceof HTMLPurifier_Node_Comment) {
|
||
|
$ws_accum[] = $node;
|
||
|
continue;
|
||
|
}
|
||
|
switch ($node->name) {
|
||
|
case 'tbody':
|
||
|
$tbody_mode = true;
|
||
|
// fall through
|
||
|
case 'tr':
|
||
|
$content[] = $node;
|
||
|
$ws_accum =& $content;
|
||
|
break;
|
||
|
case 'caption':
|
||
|
// there can only be one caption!
|
||
|
if ($caption !== false) break;
|
||
|
$caption = $node;
|
||
|
$ws_accum =& $after_caption_ws;
|
||
|
break;
|
||
|
case 'thead':
|
||
|
$tbody_mode = true;
|
||
|
// XXX This breaks rendering properties with
|
||
|
// Firefox, which never floats a <thead> to
|
||
|
// the top. Ever. (Our scheme will float the
|
||
|
// first <thead> to the top.) So maybe
|
||
|
// <thead>s that are not first should be
|
||
|
// turned into <tbody>? Very tricky, indeed.
|
||
|
if ($thead === false) {
|
||
|
$thead = $node;
|
||
|
$ws_accum =& $after_thead_ws;
|
||
|
} else {
|
||
|
// Oops, there's a second one! What
|
||
|
// should we do? Current behavior is to
|
||
|
// transmutate the first and last entries into
|
||
|
// tbody tags, and then put into content.
|
||
|
// Maybe a better idea is to *attach
|
||
|
// it* to the existing thead or tfoot?
|
||
|
// We don't do this, because Firefox
|
||
|
// doesn't float an extra tfoot to the
|
||
|
// bottom like it does for the first one.
|
||
|
$node->name = 'tbody';
|
||
|
$content[] = $node;
|
||
|
$ws_accum =& $content;
|
||
|
}
|
||
|
break;
|
||
|
case 'tfoot':
|
||
|
// see above for some aveats
|
||
|
$tbody_mode = true;
|
||
|
if ($tfoot === false) {
|
||
|
$tfoot = $node;
|
||
|
$ws_accum =& $after_tfoot_ws;
|
||
|
} else {
|
||
|
$node->name = 'tbody';
|
||
|
$content[] = $node;
|
||
|
$ws_accum =& $content;
|
||
|
}
|
||
|
break;
|
||
|
case 'colgroup':
|
||
|
case 'col':
|
||
|
$cols[] = $node;
|
||
|
$ws_accum =& $cols;
|
||
|
break;
|
||
|
case '#PCDATA':
|
||
|
// How is whitespace handled? We treat is as sticky to
|
||
|
// the *end* of the previous element. So all of the
|
||
|
// nonsense we have worked on is to keep things
|
||
|
// together.
|
||
|
if (!empty($node->is_whitespace)) {
|
||
|
$ws_accum[] = $node;
|
||
|
}
|
||
|
break;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
if (empty($content)) {
|
||
|
return false;
|
||
|
}
|
||
|
|
||
|
$ret = $initial_ws;
|
||
|
if ($caption !== false) {
|
||
|
$ret[] = $caption;
|
||
|
$ret = array_merge($ret, $after_caption_ws);
|
||
|
}
|
||
|
if ($cols !== false) {
|
||
|
$ret = array_merge($ret, $cols);
|
||
|
}
|
||
|
if ($thead !== false) {
|
||
|
$ret[] = $thead;
|
||
|
$ret = array_merge($ret, $after_thead_ws);
|
||
|
}
|
||
|
if ($tfoot !== false) {
|
||
|
$ret[] = $tfoot;
|
||
|
$ret = array_merge($ret, $after_tfoot_ws);
|
||
|
}
|
||
|
|
||
|
if ($tbody_mode) {
|
||
|
// we have to shuffle tr into tbody
|
||
|
$current_tr_tbody = null;
|
||
|
|
||
|
foreach($content as $node) {
|
||
|
switch ($node->name) {
|
||
|
case 'tbody':
|
||
|
$current_tr_tbody = null;
|
||
|
$ret[] = $node;
|
||
|
break;
|
||
|
case 'tr':
|
||
|
if ($current_tr_tbody === null) {
|
||
|
$current_tr_tbody = new HTMLPurifier_Node_Element('tbody');
|
||
|
$ret[] = $current_tr_tbody;
|
||
|
}
|
||
|
$current_tr_tbody->children[] = $node;
|
||
|
break;
|
||
|
case '#PCDATA':
|
||
|
assert($node->is_whitespace);
|
||
|
if ($current_tr_tbody === null) {
|
||
|
$ret[] = $node;
|
||
|
} else {
|
||
|
$current_tr_tbody->children[] = $node;
|
||
|
}
|
||
|
break;
|
||
|
}
|
||
|
}
|
||
|
} else {
|
||
|
$ret = array_merge($ret, $content);
|
||
|
}
|
||
|
|
||
|
return $ret;
|
||
|
|
||
|
}
|
||
|
}
|
||
|
|
||
|
// vim: et sw=4 sts=4
|