diff options
Diffstat (limited to 'library/HTMLPurifier/Strategy/FixNesting.php')
| -rw-r--r-- | library/HTMLPurifier/Strategy/FixNesting.php | 349 |
1 files changed, 101 insertions, 248 deletions
diff --git a/library/HTMLPurifier/Strategy/FixNesting.php b/library/HTMLPurifier/Strategy/FixNesting.php index f81802391..6fa673db9 100644 --- a/library/HTMLPurifier/Strategy/FixNesting.php +++ b/library/HTMLPurifier/Strategy/FixNesting.php @@ -10,12 +10,12 @@ * document type definitions, such as the chameleon nature of ins/del * tags and global child exclusions. * - * The first major objective of this strategy is to iterate through all the - * nodes (not tokens) of the list of tokens and determine whether or not - * their children conform to the element's definition. If they do not, the - * child definition may optionally supply an amended list of elements that - * is valid or require that the entire node be deleted (and the previous - * node rescanned). + * The first major objective of this strategy is to iterate through all + * the nodes and determine whether or not their children conform to the + * element's definition. If they do not, the child definition may + * optionally supply an amended list of elements that is valid or + * require that the entire node be deleted (and the previous node + * rescanned). * * The second objective is to ensure that explicitly excluded elements of * an element do not appear in its children. Code that accomplishes this @@ -25,24 +25,33 @@ * @note Whether or not unrecognized children are silently dropped or * translated into text depends on the child definitions. * - * @todo Enable nodes to be bubbled out of the structure. + * @todo Enable nodes to be bubbled out of the structure. This is + * easier with our new algorithm. */ class HTMLPurifier_Strategy_FixNesting extends HTMLPurifier_Strategy { - public function execute($tokens, $config, $context) { + /** + * @param HTMLPurifier_Token[] $tokens + * @param HTMLPurifier_Config $config + * @param HTMLPurifier_Context $context + * @return array|HTMLPurifier_Token[] + */ + public function execute($tokens, $config, $context) + { + //####################################################################// // Pre-processing + // O(n) pass to convert to a tree, so that we can efficiently + // refer to substrings + $top_node = HTMLPurifier_Arborize::arborize($tokens, $config, $context); + // get a copy of the HTML definition $definition = $config->getHTMLDefinition(); - // insert implicit "parent" node, will be removed at end. - // DEFINITION CALL - $parent_name = $definition->info_parent; - array_unshift($tokens, new HTMLPurifier_Token_Start($parent_name)); - $tokens[] = new HTMLPurifier_Token_End($parent_name); + $excludes_enabled = !$config->get('Core.DisableExcludes'); // setup the context variable 'IsInline', for chameleon processing // is 'false' when we are not inline, 'true' when it must always @@ -57,272 +66,116 @@ class HTMLPurifier_Strategy_FixNesting extends HTMLPurifier_Strategy //####################################################################// // Loop initialization - // stack that contains the indexes of all parents, - // $stack[count($stack)-1] being the current parent - $stack = array(); - // stack that contains all elements that are excluded // it is organized by parent elements, similar to $stack, // but it is only populated when an element with exclusions is // processed, i.e. there won't be empty exclusions. - $exclude_stack = array(); + $exclude_stack = array($definition->info_parent_def->excludes); // variable that contains the start token while we are processing // nodes. This enables error reporting to do its job - $start_token = false; - $context->register('CurrentToken', $start_token); + $node = $top_node; + // dummy token + list($token, $d) = $node->toTokenPair(); + $context->register('CurrentNode', $node); + $context->register('CurrentToken', $token); //####################################################################// // Loop - // iterate through all start nodes. Determining the start node - // is complicated so it has been omitted from the loop construct - for ($i = 0, $size = count($tokens) ; $i < $size; ) { - - //################################################################// - // Gather information on children - - // child token accumulator - $child_tokens = array(); - - // scroll to the end of this node, report number, and collect - // all children - for ($j = $i, $depth = 0; ; $j++) { - if ($tokens[$j] instanceof HTMLPurifier_Token_Start) { - $depth++; - // skip token assignment on first iteration, this is the - // token we currently are on - if ($depth == 1) continue; - } elseif ($tokens[$j] instanceof HTMLPurifier_Token_End) { - $depth--; - // skip token assignment on last iteration, this is the - // end token of the token we're currently on - if ($depth == 0) break; - } - $child_tokens[] = $tokens[$j]; - } - - // $i is index of start token - // $j is index of end token - - $start_token = $tokens[$i]; // to make token available via CurrentToken - - //################################################################// - // Gather information on parent - - // calculate parent information - if ($count = count($stack)) { - $parent_index = $stack[$count-1]; - $parent_name = $tokens[$parent_index]->name; - if ($parent_index == 0) { - $parent_def = $definition->info_parent_def; - } else { - $parent_def = $definition->info[$parent_name]; - } - } else { - // processing as if the parent were the "root" node - // unknown info, it won't be used anyway, in the future, - // we may want to enforce one element only (this is - // necessary for HTML Purifier to clean entire documents - $parent_index = $parent_name = $parent_def = null; - } - - // calculate context - if ($is_inline === false) { - // check if conditions make it inline - if (!empty($parent_def) && $parent_def->descendants_are_inline) { - $is_inline = $count - 1; - } - } else { - // check if we're out of inline - if ($count === $is_inline) { - $is_inline = false; - } - } - - //################################################################// - // Determine whether element is explicitly excluded SGML-style - - // determine whether or not element is excluded by checking all - // parent exclusions. The array should not be very large, two - // elements at most. - $excluded = false; - if (!empty($exclude_stack)) { - foreach ($exclude_stack as $lookup) { - if (isset($lookup[$tokens[$i]->name])) { - $excluded = true; - // no need to continue processing - break; - } + // We need to implement a post-order traversal iteratively, to + // avoid running into stack space limits. This is pretty tricky + // to reason about, so we just manually stack-ify the recursive + // variant: + // + // function f($node) { + // foreach ($node->children as $child) { + // f($child); + // } + // validate($node); + // } + // + // Thus, we will represent a stack frame as array($node, + // $is_inline, stack of children) + // e.g. array_reverse($node->children) - already processed + // children. + + $parent_def = $definition->info_parent_def; + $stack = array( + array($top_node, + $parent_def->descendants_are_inline, + $parent_def->excludes, // exclusions + 0) + ); + + while (!empty($stack)) { + list($node, $is_inline, $excludes, $ix) = array_pop($stack); + // recursive call + $go = false; + $def = empty($stack) ? $definition->info_parent_def : $definition->info[$node->name]; + while (isset($node->children[$ix])) { + $child = $node->children[$ix++]; + if ($child instanceof HTMLPurifier_Node_Element) { + $go = true; + $stack[] = array($node, $is_inline, $excludes, $ix); + $stack[] = array($child, + // ToDo: I don't think it matters if it's def or + // child_def, but double check this... + $is_inline || $def->descendants_are_inline, + empty($def->excludes) ? $excludes + : array_merge($excludes, $def->excludes), + 0); + break; } - } - - //################################################################// - // Perform child validation - - if ($excluded) { - // there is an exclusion, remove the entire node - $result = false; - $excludes = array(); // not used, but good to initialize anyway + }; + if ($go) continue; + list($token, $d) = $node->toTokenPair(); + // base case + if ($excludes_enabled && isset($excludes[$node->name])) { + $node->dead = true; + if ($e) $e->send(E_ERROR, 'Strategy_FixNesting: Node excluded'); } else { - // DEFINITION CALL - if ($i === 0) { - // special processing for the first node - $def = $definition->info_parent_def; - } else { - $def = $definition->info[$tokens[$i]->name]; - + // XXX I suppose it would be slightly more efficient to + // avoid the allocation here and have children + // strategies handle it + $children = array(); + foreach ($node->children as $child) { + if (!$child->dead) $children[] = $child; } - - if (!empty($def->child)) { - // have DTD child def validate children - $result = $def->child->validateChildren( - $child_tokens, $config, $context); + $result = $def->child->validateChildren($children, $config, $context); + if ($result === true) { + // nop + $node->children = $children; + } elseif ($result === false) { + $node->dead = true; + if ($e) $e->send(E_ERROR, 'Strategy_FixNesting: Node removed'); } else { - // weird, no child definition, get rid of everything - $result = false; - } - - // determine whether or not this element has any exclusions - $excludes = $def->excludes; - } - - // $result is now a bool or array - - //################################################################// - // Process result by interpreting $result - - if ($result === true || $child_tokens === $result) { - // leave the node as is - - // register start token as a parental node start - $stack[] = $i; - - // register exclusions if there are any - if (!empty($excludes)) $exclude_stack[] = $excludes; - - // move cursor to next possible start node - $i++; - - } elseif($result === false) { - // remove entire node - - if ($e) { - if ($excluded) { - $e->send(E_ERROR, 'Strategy_FixNesting: Node excluded'); - } else { - $e->send(E_ERROR, 'Strategy_FixNesting: Node removed'); - } - } - - // calculate length of inner tokens and current tokens - $length = $j - $i + 1; - - // perform removal - array_splice($tokens, $i, $length); - - // update size - $size -= $length; - - // there is no start token to register, - // current node is now the next possible start node - // unless it turns out that we need to do a double-check - - // this is a rought heuristic that covers 100% of HTML's - // cases and 99% of all other cases. A child definition - // that would be tricked by this would be something like: - // ( | a b c) where it's all or nothing. Fortunately, - // our current implementation claims that that case would - // not allow empty, even if it did - if (!$parent_def->child->allow_empty) { - // we need to do a double-check - $i = $parent_index; - array_pop($stack); - } - - // PROJECTED OPTIMIZATION: Process all children elements before - // reprocessing parent node. - - } else { - // replace node with $result - - // calculate length of inner tokens - $length = $j - $i - 1; - - if ($e) { - if (empty($result) && $length) { - $e->send(E_ERROR, 'Strategy_FixNesting: Node contents removed'); - } else { - $e->send(E_WARNING, 'Strategy_FixNesting: Node reorganized'); + $node->children = $result; + if ($e) { + // XXX This will miss mutations of internal nodes. Perhaps defer to the child validators + if (empty($result) && !empty($children)) { + $e->send(E_ERROR, 'Strategy_FixNesting: Node contents removed'); + } else if ($result != $children) { + $e->send(E_WARNING, 'Strategy_FixNesting: Node reorganized'); + } } } - - // perform replacement - array_splice($tokens, $i + 1, $length, $result); - - // update size - $size -= $length; - $size += count($result); - - // register start token as a parental node start - $stack[] = $i; - - // register exclusions if there are any - if (!empty($excludes)) $exclude_stack[] = $excludes; - - // move cursor to next possible start node - $i++; - } - - //################################################################// - // Scroll to next start node - - // We assume, at this point, that $i is the index of the token - // that is the first possible new start point for a node. - - // Test if the token indeed is a start tag, if not, move forward - // and test again. - $size = count($tokens); - while ($i < $size and !$tokens[$i] instanceof HTMLPurifier_Token_Start) { - if ($tokens[$i] instanceof HTMLPurifier_Token_End) { - // pop a token index off the stack if we ended a node - array_pop($stack); - // pop an exclusion lookup off exclusion stack if - // we ended node and that node had exclusions - if ($i == 0 || $i == $size - 1) { - // use specialized var if it's the super-parent - $s_excludes = $definition->info_parent_def->excludes; - } else { - $s_excludes = $definition->info[$tokens[$i]->name]->excludes; - } - if ($s_excludes) { - array_pop($exclude_stack); - } - } - $i++; - } - } //####################################################################// // Post-processing - // remove implicit parent tokens at the beginning and end - array_shift($tokens); - array_pop($tokens); - // remove context variables $context->destroy('IsInline'); + $context->destroy('CurrentNode'); $context->destroy('CurrentToken'); //####################################################################// // Return - return $tokens; - + return HTMLPurifier_Arborize::flatten($node, $config, $context); } - } // vim: et sw=4 sts=4 |