htmlpurifier update - compatibility issue with language library autoloader

8 files changed, 494 insertions, 235 deletions

diff --git a/library/HTMLPurifier/ChildDef/Chameleon.php b/library/HTMLPurifier/ChildDef/Chameleon.php
index 15c364ee3..7439be26b 100644
--- a/library/HTMLPurifier/ChildDef/Chameleon.php
+++ b/library/HTMLPurifier/ChildDef/Chameleon.php
@@ -14,33 +14,52 @@ class HTMLPurifier_ChildDef_Chameleon extends HTMLPurifier_ChildDef
 
     /**
      * Instance of the definition object to use when inline. Usually stricter.
+     * @type HTMLPurifier_ChildDef_Optional
      */
     public $inline;
 
     /**
      * Instance of the definition object to use when block.
+     * @type HTMLPurifier_ChildDef_Optional
      */
     public $block;
 
+    /**
+     * @type string
+     */
     public $type = 'chameleon';
 
     /**
-     * @param $inline List of elements to allow when inline.
-     * @param $block List of elements to allow when block.
+     * @param array $inline List of elements to allow when inline.
+     * @param array $block List of elements to allow when block.
      */
-    public function __construct($inline, $block) {
+    public function __construct($inline, $block)
+    {
         $this->inline = new HTMLPurifier_ChildDef_Optional($inline);
-        $this->block  = new HTMLPurifier_ChildDef_Optional($block);
+        $this->block = new HTMLPurifier_ChildDef_Optional($block);
         $this->elements = $this->block->elements;
     }
 
-    public function validateChildren($tokens_of_children, $config, $context) {
+    /**
+     * @param HTMLPurifier_Node[] $children
+     * @param HTMLPurifier_Config $config
+     * @param HTMLPurifier_Context $context
+     * @return bool
+     */
+    public function validateChildren($children, $config, $context)
+    {
         if ($context->get('IsInline') === false) {
             return $this->block->validateChildren(
-                $tokens_of_children, $config, $context);
+                $children,
+                $config,
+                $context
+            );
         } else {
             return $this->inline->validateChildren(
-                $tokens_of_children, $config, $context);
+                $children,
+                $config,
+                $context
+            );
         }
     }
 }
diff --git a/library/HTMLPurifier/ChildDef/Custom.php b/library/HTMLPurifier/ChildDef/Custom.php
index b68047b4b..128132e96 100644
--- a/library/HTMLPurifier/ChildDef/Custom.php
+++ b/library/HTMLPurifier/ChildDef/Custom.php
@@ -8,28 +8,42 @@
  */
 class HTMLPurifier_ChildDef_Custom extends HTMLPurifier_ChildDef
 {
+    /**
+     * @type string
+     */
     public $type = 'custom';
+
+    /**
+     * @type bool
+     */
     public $allow_empty = false;
+
     /**
-     * Allowed child pattern as defined by the DTD
+     * Allowed child pattern as defined by the DTD.
+     * @type string
      */
     public $dtd_regex;
+
     /**
-     * PCRE regex derived from $dtd_regex
-     * @private
+     * PCRE regex derived from $dtd_regex.
+     * @type string
      */
     private $_pcre_regex;
+
     /**
      * @param $dtd_regex Allowed child pattern from the DTD
      */
-    public function __construct($dtd_regex) {
+    public function __construct($dtd_regex)
+    {
         $this->dtd_regex = $dtd_regex;
         $this->_compileRegex();
     }
+
     /**
      * Compiles the PCRE regex from a DTD regex ($dtd_regex to $_pcre_regex)
      */
-    protected function _compileRegex() {
+    protected function _compileRegex()
+    {
         $raw = str_replace(' ', '', $this->dtd_regex);
         if ($raw{0} != '(') {
             $raw = "($raw)";
@@ -57,33 +71,31 @@ class HTMLPurifier_ChildDef_Custom extends HTMLPurifier_ChildDef
 
         $this->_pcre_regex = $reg;
     }
-    public function validateChildren($tokens_of_children, $config, $context) {
+
+    /**
+     * @param HTMLPurifier_Node[] $children
+     * @param HTMLPurifier_Config $config
+     * @param HTMLPurifier_Context $context
+     * @return bool
+     */
+    public function validateChildren($children, $config, $context)
+    {
         $list_of_children = '';
         $nesting = 0; // depth into the nest
-        foreach ($tokens_of_children as $token) {
-            if (!empty($token->is_whitespace)) continue;
-
-            $is_child = ($nesting == 0); // direct
-
-            if ($token instanceof HTMLPurifier_Token_Start) {
-                $nesting++;
-            } elseif ($token instanceof HTMLPurifier_Token_End) {
-                $nesting--;
-            }
-
-            if ($is_child) {
-                $list_of_children .= $token->name . ',';
+        foreach ($children as $node) {
+            if (!empty($node->is_whitespace)) {
+                continue;
             }
+            $list_of_children .= $node->name . ',';
         }
         // add leading comma to deal with stray comma declarations
         $list_of_children = ',' . rtrim($list_of_children, ',');
         $okay =
             preg_match(
-                '/^,?'.$this->_pcre_regex.'$/',
+                '/^,?' . $this->_pcre_regex . '$/',
                 $list_of_children
             );
-
-        return (bool) $okay;
+        return (bool)$okay;
     }
 }
 
diff --git a/library/HTMLPurifier/ChildDef/Empty.php b/library/HTMLPurifier/ChildDef/Empty.php
index 13171f665..a8a6cbdd2 100644
--- a/library/HTMLPurifier/ChildDef/Empty.php
+++ b/library/HTMLPurifier/ChildDef/Empty.php
@@ -9,10 +9,28 @@
  */
 class HTMLPurifier_ChildDef_Empty extends HTMLPurifier_ChildDef
 {
+    /**
+     * @type bool
+     */
     public $allow_empty = true;
+
+    /**
+     * @type string
+     */
     public $type = 'empty';
-    public function __construct() {}
-    public function validateChildren($tokens_of_children, $config, $context) {
+
+    public function __construct()
+    {
+    }
+
+    /**
+     * @param HTMLPurifier_Node[] $children
+     * @param HTMLPurifier_Config $config
+     * @param HTMLPurifier_Context $context
+     * @return array
+     */
+    public function validateChildren($children, $config, $context)
+    {
         return array();
     }
 }
diff --git a/library/HTMLPurifier/ChildDef/List.php b/library/HTMLPurifier/ChildDef/List.php
new file mode 100644
index 000000000..891b9f6f5
--- /dev/null
+++ b/library/HTMLPurifier/ChildDef/List.php
@@ -0,0 +1,86 @@
+<?php
+
+/**
+ * Definition for list containers ul and ol.
+ *
+ * What does this do?  The big thing is to handle ol/ul at the top
+ * level of list nodes, which should be handled specially by /folding/
+ * them into the previous list node.  We generally shouldn't ever
+ * see other disallowed elements, because the autoclose behavior
+ * in MakeWellFormed handles it.
+ */
+class HTMLPurifier_ChildDef_List extends HTMLPurifier_ChildDef
+{
+    /**
+     * @type string
+     */
+    public $type = 'list';
+    /**
+     * @type array
+     */
+    // lying a little bit, so that we can handle ul and ol ourselves
+    // XXX: This whole business with 'wrap' is all a bit unsatisfactory
+    public $elements = array('li' => true, 'ul' => true, 'ol' => true);
+
+    /**
+     * @param array $children
+     * @param HTMLPurifier_Config $config
+     * @param HTMLPurifier_Context $context
+     * @return array
+     */
+    public function validateChildren($children, $config, $context)
+    {
+        // Flag for subclasses
+        $this->whitespace = false;
+
+        // if there are no tokens, delete parent node
+        if (empty($children)) {
+            return false;
+        }
+
+        // the new set of children
+        $result = array();
+
+        // a little sanity check to make sure it's not ALL whitespace
+        $all_whitespace = true;
+
+        $current_li = false;
+
+        foreach ($children as $node) {
+            if (!empty($node->is_whitespace)) {
+                $result[] = $node;
+                continue;
+            }
+            $all_whitespace = false; // phew, we're not talking about whitespace
+
+            if ($node->name === 'li') {
+                // good
+                $current_li = $node;
+                $result[] = $node;
+            } else {
+                // we want to tuck this into the previous li
+                // Invariant: we expect the node to be ol/ul
+                // ToDo: Make this more robust in the case of not ol/ul
+                // by distinguishing between existing li and li created
+                // to handle non-list elements; non-list elements should
+                // not be appended to an existing li; only li created
+                // for non-list. This distinction is not currently made.
+                if ($current_li === false) {
+                    $current_li = new HTMLPurifier_Node_Element('li');
+                    $result[] = $current_li;
+                }
+                $current_li->children[] = $node;
+                $current_li->empty = false; // XXX fascinating! Check for this error elsewhere ToDo
+            }
+        }
+        if (empty($result)) {
+            return false;
+        }
+        if ($all_whitespace) {
+            return false;
+        }
+        return $result;
+    }
+}
+
+// vim: et sw=4 sts=4
diff --git a/library/HTMLPurifier/ChildDef/Optional.php b/library/HTMLPurifier/ChildDef/Optional.php
index 32bcb9898..b9468063b 100644
--- a/library/HTMLPurifier/ChildDef/Optional.php
+++ b/library/HTMLPurifier/ChildDef/Optional.php
@@ -9,15 +9,34 @@
  */
 class HTMLPurifier_ChildDef_Optional extends HTMLPurifier_ChildDef_Required
 {
+    /**
+     * @type bool
+     */
     public $allow_empty = true;
+
+    /**
+     * @type string
+     */
     public $type = 'optional';
-    public function validateChildren($tokens_of_children, $config, $context) {
-        $result = parent::validateChildren($tokens_of_children, $config, $context);
-        // we assume that $tokens_of_children is not modified
+
+    /**
+     * @param array $children
+     * @param HTMLPurifier_Config $config
+     * @param HTMLPurifier_Context $context
+     * @return array
+     */
+    public function validateChildren($children, $config, $context)
+    {
+        $result = parent::validateChildren($children, $config, $context);
+        // we assume that $children is not modified
         if ($result === false) {
-            if (empty($tokens_of_children)) return true;
-            elseif ($this->whitespace) return $tokens_of_children;
-            else return array();
+            if (empty($children)) {
+                return true;
+            } elseif ($this->whitespace) {
+                return $children;
+            } else {
+                return array();
+            }
         }
         return $result;
     }
diff --git a/library/HTMLPurifier/ChildDef/Required.php b/library/HTMLPurifier/ChildDef/Required.php
index 4889f249b..0d1c8f5f3 100644
--- a/library/HTMLPurifier/ChildDef/Required.php
+++ b/library/HTMLPurifier/ChildDef/Required.php
@@ -7,17 +7,21 @@ class HTMLPurifier_ChildDef_Required extends HTMLPurifier_ChildDef
 {
     /**
      * Lookup table of allowed elements.
-     * @public
+     * @type array
      */
     public $elements = array();
+
     /**
      * Whether or not the last passed node was all whitespace.
+     * @type bool
      */
     protected $whitespace = false;
+
     /**
-     * @param $elements List of allowed element names (lowercase).
+     * @param array|string $elements List of allowed element names (lowercase).
      */
-    public function __construct($elements) {
+    public function __construct($elements)
+    {
         if (is_string($elements)) {
             $elements = str_replace(' ', '', $elements);
             $elements = explode('|', $elements);
@@ -27,29 +31,43 @@ class HTMLPurifier_ChildDef_Required extends HTMLPurifier_ChildDef
             $elements = array_flip($elements);
             foreach ($elements as $i => $x) {
                 $elements[$i] = true;
-                if (empty($i)) unset($elements[$i]); // remove blank
+                if (empty($i)) {
+                    unset($elements[$i]);
+                } // remove blank
             }
         }
         $this->elements = $elements;
     }
+
+    /**
+     * @type bool
+     */
     public $allow_empty = false;
+
+    /**
+     * @type string
+     */
     public $type = 'required';
-    public function validateChildren($tokens_of_children, $config, $context) {
+
+    /**
+     * @param array $children
+     * @param HTMLPurifier_Config $config
+     * @param HTMLPurifier_Context $context
+     * @return array
+     */
+    public function validateChildren($children, $config, $context)
+    {
         // Flag for subclasses
         $this->whitespace = false;
 
         // if there are no tokens, delete parent node
-        if (empty($tokens_of_children)) return false;
+        if (empty($children)) {
+            return false;
+        }
 
         // the new set of children
         $result = array();
 
-        // current depth into the nest
-        $nesting = 0;
-
-        // whether or not we're deleting a node
-        $is_deleting = false;
-
         // whether or not parsed character data is allowed
         // this controls whether or not we silently drop a tag
         // or generate escaped HTML from it
@@ -58,58 +76,41 @@ class HTMLPurifier_ChildDef_Required extends HTMLPurifier_ChildDef
         // a little sanity check to make sure it's not ALL whitespace
         $all_whitespace = true;
 
-        // some configuration
-        $escape_invalid_children = $config->get('Core.EscapeInvalidChildren');
-
-        // generator
-        $gen = new HTMLPurifier_Generator($config, $context);
-
-        foreach ($tokens_of_children as $token) {
-            if (!empty($token->is_whitespace)) {
-                $result[] = $token;
+        $stack = array_reverse($children);
+        while (!empty($stack)) {
+            $node = array_pop($stack);
+            if (!empty($node->is_whitespace)) {
+                $result[] = $node;
                 continue;
             }
             $all_whitespace = false; // phew, we're not talking about whitespace
 
-            $is_child = ($nesting == 0);
-
-            if ($token instanceof HTMLPurifier_Token_Start) {
-                $nesting++;
-            } elseif ($token instanceof HTMLPurifier_Token_End) {
-                $nesting--;
-            }
-
-            if ($is_child) {
-                $is_deleting = false;
-                if (!isset($this->elements[$token->name])) {
-                    $is_deleting = true;
-                    if ($pcdata_allowed && $token instanceof HTMLPurifier_Token_Text) {
-                        $result[] = $token;
-                    } elseif ($pcdata_allowed && $escape_invalid_children) {
-                        $result[] = new HTMLPurifier_Token_Text(
-                            $gen->generateFromToken($token)
-                        );
+            if (!isset($this->elements[$node->name])) {
+                // special case text
+                // XXX One of these ought to be redundant or something
+                if ($pcdata_allowed && $node instanceof HTMLPurifier_Node_Text) {
+                    $result[] = $node;
+                    continue;
+                }
+                // spill the child contents in
+                // ToDo: Make configurable
+                if ($node instanceof HTMLPurifier_Node_Element) {
+                    for ($i = count($node->children) - 1; $i >= 0; $i--) {
+                        $stack[] = $node->children[$i];
                     }
                     continue;
                 }
+                continue;
             }
-            if (!$is_deleting || ($pcdata_allowed && $token instanceof HTMLPurifier_Token_Text)) {
-                $result[] = $token;
-            } elseif ($pcdata_allowed && $escape_invalid_children) {
-                $result[] =
-                    new HTMLPurifier_Token_Text(
-                        $gen->generateFromToken($token)
-                    );
-            } else {
-                // drop silently
-            }
+            $result[] = $node;
+        }
+        if (empty($result)) {
+            return false;
         }
-        if (empty($result)) return false;
         if ($all_whitespace) {
             $this->whitespace = true;
             return false;
         }
-        if ($tokens_of_children == $result) return true;
         return $result;
     }
 }
diff --git a/library/HTMLPurifier/ChildDef/StrictBlockquote.php b/library/HTMLPurifier/ChildDef/StrictBlockquote.php
index dfae8a6e5..3270a46e1 100644
--- a/library/HTMLPurifier/ChildDef/StrictBlockquote.php
+++ b/library/HTMLPurifier/ChildDef/StrictBlockquote.php
@@ -5,75 +5,97 @@
  */
 class HTMLPurifier_ChildDef_StrictBlockquote extends HTMLPurifier_ChildDef_Required
 {
+    /**
+     * @type array
+     */
     protected $real_elements;
+
+    /**
+     * @type array
+     */
     protected $fake_elements;
+
+    /**
+     * @type bool
+     */
     public $allow_empty = true;
+
+    /**
+     * @type string
+     */
     public $type = 'strictblockquote';
+
+    /**
+     * @type bool
+     */
     protected $init = false;
 
     /**
+     * @param HTMLPurifier_Config $config
+     * @return array
      * @note We don't want MakeWellFormed to auto-close inline elements since
      *       they might be allowed.
      */
-    public function getAllowedElements($config) {
+    public function getAllowedElements($config)
+    {
         $this->init($config);
         return $this->fake_elements;
     }
 
-    public function validateChildren($tokens_of_children, $config, $context) {
-
+    /**
+     * @param array $children
+     * @param HTMLPurifier_Config $config
+     * @param HTMLPurifier_Context $context
+     * @return array
+     */
+    public function validateChildren($children, $config, $context)
+    {
         $this->init($config);
 
         // trick the parent class into thinking it allows more
         $this->elements = $this->fake_elements;
-        $result = parent::validateChildren($tokens_of_children, $config, $context);
+        $result = parent::validateChildren($children, $config, $context);
         $this->elements = $this->real_elements;
 
-        if ($result === false) return array();
-        if ($result === true) $result = $tokens_of_children;
+        if ($result === false) {
+            return array();
+        }
+        if ($result === true) {
+            $result = $children;
+        }
 
         $def = $config->getHTMLDefinition();
-        $block_wrap_start = new HTMLPurifier_Token_Start($def->info_block_wrapper);
-        $block_wrap_end   = new HTMLPurifier_Token_End(  $def->info_block_wrapper);
-        $is_inline = false;
-        $depth = 0;
+        $block_wrap_name = $def->info_block_wrapper;
+        $block_wrap = false;
         $ret = array();
 
-        // assuming that there are no comment tokens
-        foreach ($result as $i => $token) {
-            $token = $result[$i];
-            // ifs are nested for readability
-            if (!$is_inline) {
-                if (!$depth) {
-                     if (
-                        ($token instanceof HTMLPurifier_Token_Text && !$token->is_whitespace) ||
-                        (!$token instanceof HTMLPurifier_Token_Text && !isset($this->elements[$token->name]))
-                     ) {
-                        $is_inline = true;
-                        $ret[] = $block_wrap_start;
-                     }
+        foreach ($result as $node) {
+            if ($block_wrap === false) {
+                if (($node instanceof HTMLPurifier_Node_Text && !$node->is_whitespace) ||
+                    ($node instanceof HTMLPurifier_Node_Element && !isset($this->elements[$node->name]))) {
+                        $block_wrap = new HTMLPurifier_Node_Element($def->info_block_wrapper);
+                        $ret[] = $block_wrap;
                 }
             } else {
-                if (!$depth) {
-                    // starting tokens have been inline text / empty
-                    if ($token instanceof HTMLPurifier_Token_Start || $token instanceof HTMLPurifier_Token_Empty) {
-                        if (isset($this->elements[$token->name])) {
-                            // ended
-                            $ret[] = $block_wrap_end;
-                            $is_inline = false;
-                        }
-                    }
+                if ($node instanceof HTMLPurifier_Node_Element && isset($this->elements[$node->name])) {
+                    $block_wrap = false;
+
                 }
             }
-            $ret[] = $token;
-            if ($token instanceof HTMLPurifier_Token_Start) $depth++;
-            if ($token instanceof HTMLPurifier_Token_End)   $depth--;
+            if ($block_wrap) {
+                $block_wrap->children[] = $node;
+            } else {
+                $ret[] = $node;
+            }
         }
-        if ($is_inline) $ret[] = $block_wrap_end;
         return $ret;
     }
 
-    private function init($config) {
+    /**
+     * @param HTMLPurifier_Config $config
+     */
+    private function init($config)
+    {
         if (!$this->init) {
             $def = $config->getHTMLDefinition();
             // allow all inline elements
diff --git a/library/HTMLPurifier/ChildDef/Table.php b/library/HTMLPurifier/ChildDef/Table.php
index 34f0227dd..3e4a0f218 100644
--- a/library/HTMLPurifier/ChildDef/Table.php
+++ b/library/HTMLPurifier/ChildDef/Table.php
@@ -1,140 +1,222 @@
 <?php
 
 /**
- * Definition for tables
+ * 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';
-    public $elements = array('tr' => true, 'tbody' => true, 'thead' => true,
-        'tfoot' => true, 'caption' => true, 'colgroup' => true, 'col' => true);
-    public function __construct() {}
-    public function validateChildren($tokens_of_children, $config, $context) {
-        if (empty($tokens_of_children)) return false;
 
-        // this ensures that the loop gets run one last time before closing
-        // up. It's a little bit of a hack, but it works! Just make sure you
-        // get rid of the token later.
-        $tokens_of_children[] = false;
+    /**
+     * @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;
+        $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();
+        $cols = array();
         $content = array();
 
-        $nesting = 0; // current depth so we can determine nodes
-        $is_collecting = false; // are we globbing together tokens to package
-                                // into one of the collectors?
-        $collection = array(); // collected nodes
-        $tag_index = 0; // the first node might be whitespace,
-                            // so this tells us where the start tag is
-
-        foreach ($tokens_of_children as $token) {
-            $is_child = ($nesting == 0);
-
-            if ($token === false) {
-                // terminating sequence started
-            } elseif ($token instanceof HTMLPurifier_Token_Start) {
-                $nesting++;
-            } elseif ($token instanceof HTMLPurifier_Token_End) {
-                $nesting--;
-            }
+        $tbody_mode = false; // if true, then we need to wrap any stray
+                             // <tr>s with a <tbody>.
 
-            // handle node collection
-            if ($is_collecting) {
-                if ($is_child) {
-                    // okay, let's stash the tokens away
-                    // first token tells us the type of the collection
-                    switch ($collection[$tag_index]->name) {
-                        case 'tr':
-                        case 'tbody':
-                            $content[] = $collection;
-                            break;
-                        case 'caption':
-                            if ($caption !== false) break;
-                            $caption = $collection;
-                            break;
-                        case 'thead':
-                        case 'tfoot':
-                            // access the appropriate variable, $thead or $tfoot
-                            $var = $collection[$tag_index]->name;
-                            if ($$var === false) {
-                                $$var = $collection;
-                            } else {
-                                // transmutate the first and less entries into
-                                // tbody tags, and then put into content
-                                $collection[$tag_index]->name = 'tbody';
-                                $collection[count($collection)-1]->name = 'tbody';
-                                $content[] = $collection;
-                            }
-                            break;
-                         case 'colgroup':
-                            $cols[] = $collection;
-                            break;
-                    }
-                    $collection = array();
-                    $is_collecting = false;
-                    $tag_index = 0;
+        $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 {
-                    // add the node to the collection
-                    $collection[] = $token;
+                    // 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 =& $content;
                 }
-            }
-
-            // terminate
-            if ($token === false) break;
-
-            if ($is_child) {
-                // determine what we're dealing with
-                if ($token->name == 'col') {
-                    // the only empty tag in the possie, we can handle it
-                    // immediately
-                    $cols[] = array_merge($collection, array($token));
-                    $collection = array();
-                    $tag_index = 0;
-                    continue;
+                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 =& $content;
                 }
-                switch($token->name) {
-                    case 'caption':
-                    case 'colgroup':
-                    case 'thead':
-                    case 'tfoot':
-                    case 'tbody':
-                    case 'tr':
-                        $is_collecting = true;
-                        $collection[] = $token;
-                        continue;
-                    default:
-                        if (!empty($token->is_whitespace)) {
-                            $collection[] = $token;
-                            $tag_index++;
-                        }
-                        continue;
+                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;
-
-        $ret = array();
-        if ($caption !== false) $ret = array_merge($ret, $caption);
-        if ($cols !== false)    foreach ($cols as $token_array) $ret = array_merge($ret, $token_array);
-        if ($thead !== false)   $ret = array_merge($ret, $thead);
-        if ($tfoot !== false)   $ret = array_merge($ret, $tfoot);
-        foreach ($content as $token_array) $ret = array_merge($ret, $token_array);
-        if (!empty($collection) && $is_collecting == false){
-            // grab the trailing space
-            $ret = array_merge($ret, $collection);
+        if (empty($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);
         }
 
-        array_pop($tokens_of_children); // remove phantom token
+        if ($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 === $tokens_of_children) ? true : $ret;
+        return $ret;
 
     }
 }