mirrors/htmlpurifier
0
0
Fork 0
mirror of https://github.com/ezyang/htmlpurifier.git synced 2024-12-22 08:21:52 +00:00
Code Releases Activity

Rewrite FixNesting implementation to be tree-based.

Browse Source 
This mega-patch rips out the FixNesting implementation and the related
ChildDef components.  The primary algorithmic change is to convert from
use of tokens to tree nodes, which are far more amenable to the style
of processing that FixNesting uses.  Additionally, FixNesting has been
changed to go bottom-up rather than top-down, in order to avoid needing
to implement backtracking.

This patch simplifies a good deal of the relevant logic, since we no
longer need to continually recalculate the nesting structure when
processing things.  However, the conversion to the alternate format
incurs some overhead, so for small inputs these changes are not a win.
One possibility to greatly reduce the constant factors here is to switch
to entirely using libxml's representation, and never serializing tokens;
this would require one to rewrite injectors, however.

The iterative post-order traversal in FixNesting is a bit subtle, but
we have essentially reified the stack and continuations.

We've removed support for %Core.EscapeInvalidChildren.

Signed-off-by: Edward Z. Yang <ezyang@mit.edu>
This commit is contained in:
Edward Z. Yang 2013-10-20 22:18:59 -07:00
parent b3640e1af6
commit 0767bbc12d
22 changed files with 358 additions and 698 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
NEWS
View File

@ -15,10 +15,15 @@ NEWS ( CHANGELOG and HISTORY ) HTMLPurifier
# URI parsing algorithm was made more strict, so only prefixes which
looks like schemes will actually be schemes. Thanks
Michael Gusev <mgusev@sugarcrm.com> for fixing.
# %Core.EscapeInvalidChildren is no longer supported, and no longer does
anything.
! New directive %Core.AllowHostnameUnderscore which allows underscores
in hostnames.
- Eliminate quadratic behavior in DOMLex by using a proper queue.
Thanks Ole Laursen for noticing this.
- Rewritten MakeWellFormed/FixNesting implementation eliminates quadratic
behavior in the rest of the purificaiton pipeline. Thanks Chedburn
Networks for sponsoring this work.
- Made Linkify URL parser a bit less permissive, so that non-breaking
spaces and commas are not included as part of URL. Thanks nAS for fixing.
- Fix some bad interactions with %HTML.Allowed and injectors. Thanks

7
configdoc/usage.xml
View File

@ -406,11 +406,6 @@
<line>53</line>
</file>
</directive>
<directive id="Core.EscapeInvalidChildren">
<file name="HTMLPurifier/ChildDef/Required.php">
<line>86</line>
</file>
</directive>
<directive id="Cache.SerializerPath">
<file name="HTMLPurifier/DefinitionCache/Serializer.php">
<line>171</line>
@ -498,7 +493,7 @@
</directive>
<directive id="Core.DisableExcludes">
<file name="HTMLPurifier/Strategy/FixNesting.php">
<line>67</line>
<line>54</line>
</file>
</directive>
<directive id="Core.EscapeInvalidTags">

8
library/HTMLPurifier/ChildDef.php
View File

@ -1,7 +1,7 @@
<?php
/**
* Defines allowed child nodes and validates tokens against it.
* Defines allowed child nodes and validates nodes against it.
*/
abstract class HTMLPurifier_ChildDef
{
@ -41,12 +41,12 @@ abstract class HTMLPurifier_ChildDef
/**
* Validates nodes according to definition and returns modification.
*
* @param HTMLPurifier_Token[] $tokens_of_children Array of HTMLPurifier_Token
* @param HTMLPurifier_Node[] $children Array of HTMLPurifier_Node
* @param HTMLPurifier_Config $config HTMLPurifier_Config object
* @param HTMLPurifier_Context $context HTMLPurifier_Context object
* @return bool|array true to leave nodes as is, false to remove parent node, array of replacement child tokens
* @return bool|array true to leave nodes as is, false to remove parent node, array of replacement children
*/
abstract public function validateChildren($tokens_of_children, $config, $context);
abstract public function validateChildren($children, $config, $context);
}
// vim: et sw=4 sts=4

8
library/HTMLPurifier/ChildDef/Chameleon.php
View File

@ -41,22 +41,22 @@ class HTMLPurifier_ChildDef_Chameleon extends HTMLPurifier_ChildDef
}
/**
* @param array $tokens_of_children
* @param HTMLPurifier_Node[] $children
* @param HTMLPurifier_Config $config
* @param HTMLPurifier_Context $context
* @return bool
*/
public function validateChildren($tokens_of_children, $config, $context)
public function validateChildren($children, $config, $context)
{
if ($context->get('IsInline') === false) {
return $this->block->validateChildren(
$tokens_of_children,
$children,
$config,
$context
);
} else {
return $this->inline->validateChildren(
$tokens_of_children,
$children,
$config,
$context
);

21
library/HTMLPurifier/ChildDef/Custom.php
View File

@ -73,31 +73,20 @@ class HTMLPurifier_ChildDef_Custom extends HTMLPurifier_ChildDef
}
/**
* @param array $tokens_of_children
* @param HTMLPurifier_Node[] $children
* @param HTMLPurifier_Config $config
* @param HTMLPurifier_Context $context
* @return bool
*/
public function validateChildren($tokens_of_children, $config, $context)
public function validateChildren($children, $config, $context)
{
$list_of_children = '';
$nesting = 0; // depth into the nest
foreach ($tokens_of_children as $token) {
if (!empty($token->is_whitespace)) {
foreach ($children as $node) {
if (!empty($node->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 . ',';
}
$list_of_children .= $node->name . ',';
}
// add leading comma to deal with stray comma declarations
$list_of_children = ',' . rtrim($list_of_children, ',');

4
library/HTMLPurifier/ChildDef/Empty.php
View File

@ -24,12 +24,12 @@ class HTMLPurifier_ChildDef_Empty extends HTMLPurifier_ChildDef
}
/**
* @param array $tokens_of_children
* @param HTMLPurifier_Node[] $children
* @param HTMLPurifier_Config $config
* @param HTMLPurifier_Context $context
* @return array
*/
public function validateChildren($tokens_of_children, $config, $context)
public function validateChildren($children, $config, $context)
{
return array();
}

117
library/HTMLPurifier/ChildDef/List.php
View File

@ -2,6 +2,12 @@
/**
* 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
{
@ -17,115 +23,55 @@ class HTMLPurifier_ChildDef_List extends HTMLPurifier_ChildDef
public $elements = array('li' => true, 'ul' => true, 'ol' => true);
/**
* @param array $tokens_of_children
* @param array $children
* @param HTMLPurifier_Config $config
* @param HTMLPurifier_Context $context
* @return array
*/
public function validateChildren($tokens_of_children, $config, $context)
public function validateChildren($children, $config, $context)
{
// Flag for subclasses
$this->whitespace = false;
// if there are no tokens, delete parent node
if (empty($tokens_of_children)) {
if (empty($children)) {
return false;
}
// the new set of children
$result = array();
// current depth into the nest
$nesting = 0;
// a little sanity check to make sure it's not ALL whitespace
$all_whitespace = true;
$seen_li = false;
$need_close_li = false;
$current_li = false;
foreach ($tokens_of_children as $token) {
if (!empty($token->is_whitespace)) {
$result[] = $token;
foreach ($children as $node) {
if (!empty($node->is_whitespace)) {
$result[] = $node;
continue;
}
$all_whitespace = false; // phew, we're not talking about whitespace
if ($nesting == 1 && $need_close_li) {
$result[] = new HTMLPurifier_Token_End('li');
$nesting--;
$need_close_li = false;
}
$is_child = ($nesting == 0);
if ($token instanceof HTMLPurifier_Token_Start) {
$nesting++;
} elseif ($token instanceof HTMLPurifier_Token_End) {
$nesting--;
}
if ($is_child) {
if ($token->name === 'li') {
// good
$seen_li = true;
} elseif ($token->name === 'ul' || $token->name === 'ol') {
// we want to tuck this into the previous li
$need_close_li = true;
$nesting++;
if (!$seen_li) {
// create a new li element
$result[] = new HTMLPurifier_Token_Start('li');
} else {
// backtrack until </li> found
while (true) {
$t = array_pop($result);
if ($t instanceof HTMLPurifier_Token_End) {
// XXX actually, these invariants could very plausibly be violated
// if we are doing silly things with modifying the set of allowed elements.
// FORTUNATELY, it doesn't make a difference, since the allowed
// elements are hard-coded here!
if ($t->name !== 'li') {
trigger_error("Only li present invariant violated in List ChildDef", E_USER_ERROR);
return false;
}
break;
} elseif ($t instanceof HTMLPurifier_Token_Empty) { // bleagh
if ($t->name !== 'li') {
trigger_error("Only li present invariant violated in List ChildDef", E_USER_ERROR);
return false;
}
// XXX this should have a helper for it...
$result[] = new HTMLPurifier_Token_Start('li', $t->attr, $t->line, $t->col, $t->armor);
break;
} else {
if (!$t->is_whitespace) {
trigger_error(
"Only whitespace present invariant violated in List ChildDef",
E_USER_ERROR
);
return false;
}
}
}
}
} else {
// start wrapping (this doesn't precisely mimic
// browser behavior, but what browsers do is kind of
// hard to mimic in a standards compliant way
// XXX Actually, this has no impact in practice,
// because this gets handled earlier. Arguably,
// we should rip out all of that processing
$result[] = new HTMLPurifier_Token_Start('li');
$nesting++;
$seen_li = true;
$need_close_li = true;
if ($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
}
$result[] = $token;
}
if ($need_close_li) {
$result[] = new HTMLPurifier_Token_End('li');
}
if (empty($result)) {
return false;
@ -133,9 +79,6 @@ class HTMLPurifier_ChildDef_List extends HTMLPurifier_ChildDef
if ($all_whitespace) {
return false;
}
if ($tokens_of_children == $result) {
return true;
}
return $result;
}
}

12
library/HTMLPurifier/ChildDef/Optional.php
View File

@ -20,20 +20,20 @@ class HTMLPurifier_ChildDef_Optional extends HTMLPurifier_ChildDef_Required
public $type = 'optional';
/**
* @param array $tokens_of_children
* @param array $children
* @param HTMLPurifier_Config $config
* @param HTMLPurifier_Context $context
* @return array
*/
public function validateChildren($tokens_of_children, $config, $context)
public function validateChildren($children, $config, $context)
{
$result = parent::validateChildren($tokens_of_children, $config, $context);
// we assume that $tokens_of_children is not modified
$result = parent::validateChildren($children, $config, $context);
// we assume that $children is not modified
if ($result === false) {
if (empty($tokens_of_children)) {
if (empty($children)) {
return true;
} elseif ($this->whitespace) {
return $tokens_of_children;
return $children;
} else {
return array();
}

71
library/HTMLPurifier/ChildDef/Required.php
View File

@ -50,30 +50,24 @@ class HTMLPurifier_ChildDef_Required extends HTMLPurifier_ChildDef
public $type = 'required';
/**
* @param array $tokens_of_children
* @param array $children
* @param HTMLPurifier_Config $config
* @param HTMLPurifier_Context $context
* @return array
*/
public function validateChildren($tokens_of_children, $config, $context)
public function validateChildren($children, $config, $context)
{
// Flag for subclasses
$this->whitespace = false;
// if there are no tokens, delete parent node
if (empty($tokens_of_children)) {
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
@ -82,51 +76,33 @@ 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;
@ -135,9 +111,6 @@ class HTMLPurifier_ChildDef_Required extends HTMLPurifier_ChildDef
$this->whitespace = true;
return false;
}
if ($tokens_of_children == $result) {
return true;
}
return $result;
}
}

56
library/HTMLPurifier/ChildDef/StrictBlockquote.php
View File

@ -43,68 +43,50 @@ class HTMLPurifier_ChildDef_StrictBlockquote extends HTMLPurifier_ChildDef_Requi
}
/**
* @param array $tokens_of_children
* @param array $children
* @param HTMLPurifier_Config $config
* @param HTMLPurifier_Context $context
* @return array
*/
public function validateChildren($tokens_of_children, $config, $context)
public function validateChildren($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;
$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) {
$depth++;
if ($block_wrap) {
$block_wrap->children[] = $node;
} else {
$ret[] = $node;
}
if ($token instanceof HTMLPurifier_Token_End) {
$depth--;
}
}
if ($is_inline) {
$ret[] = $block_wrap_end;
}
return $ret;
}

284
library/HTMLPurifier/ChildDef/Table.php
View File

@ -59,230 +59,164 @@ class HTMLPurifier_ChildDef_Table extends HTMLPurifier_ChildDef
}
/**
* @param array $tokens_of_children
* @param array $children
* @param HTMLPurifier_Config $config
* @param HTMLPurifier_Context $context
* @return array
*/
public function validateChildren($tokens_of_children, $config, $context)
public function validateChildren($children, $config, $context)
{
if (empty($tokens_of_children)) {
if (empty($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;
// 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();
$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
// INVARIANT: if $is_collecting, then !empty($collection)
// The converse does NOT hold, see [WHITESPACE]
$tag_index = 0; // the first node might be whitespace,
// so this tells us where the start tag is
$tbody_mode = false; // if true, then we need to wrap any stray
// <tr>s with a <tbody>.
// <tr>s with a <tbody>.
foreach ($tokens_of_children as $token) {
$is_child = ($nesting == 0);
$ws_accum =& $initial_ws;
if ($token === false) {
// terminating sequence started
} elseif ($token instanceof HTMLPurifier_Token_Start) {
$nesting++;
} elseif ($token instanceof HTMLPurifier_Token_End) {
$nesting--;
foreach ($children as $node) {
if ($node instanceof HTMLPurifier_Node_Comment) {
$ws_accum[] = $node;
continue;
}
// handle node collection
if ($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 'tbody':
$tbody_mode = true;
// fall through
case 'tr':
$content[] = $collection;
break;
case 'caption':
if ($caption !== false) {
break;
}
$caption = $collection;
break;
case 'thead':
case 'tfoot':
$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.
// access the appropriate variable, $thead or $tfoot
$var = $collection[$tag_index]->name;
if ($$var === false) {
$$var = $collection;
} 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.
$collection[$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;
} else {
// add the node to the collection
$collection[] = $token;
}
}
// terminate
if ($token === false) {
switch ($node->name) {
case 'tbody':
$tbody_mode = true;
// fall through
case 'tr':
$content[] = $node;
$ws_accum =& $content;
break;
}
if ($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();
$is_collecting = false;
$tag_index = 0;
continue;
case 'caption':
// there can only be one caption!
if ($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 =& $content;
}
switch ($token->name) {
case 'caption':
case 'colgroup':
case 'thead':
case 'tfoot':
case 'tbody':
case 'tr':
$is_collecting = true;
$collection[] = $token;
continue;
default:
// [WHITESPACE] Whitespace is added to the
// collection without triggering collection
// mode. This is a hack to make whitespace
// 'sticky' (that is to say, we ought /not/ to
// drop whitespace.)
if (!empty($token->is_whitespace)) {
$collection[] = $token;
$tag_index++;
}
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;
}
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;
}
// INVARIANT: all members of content are non-empty. This can
// be shown by observing when things are pushed onto content:
// they are only ever pushed when is_collecting is true, and
// collection is the only thing ever pushed; but it is known
// that collections are non-empty when is_collecting is true.
$ret = array();
$ret = $initial_ws;
if ($caption !== false) {
$ret = array_merge($ret, $caption);
$ret[] = $caption;
$ret = array_merge($ret, $after_caption_ws);
}
if ($cols !== false) {
foreach ($cols as $token_array) {
$ret = array_merge($ret, $token_array);
}
$ret = array_merge($ret, $cols);
}
if ($thead !== false) {
$ret = array_merge($ret, $thead);
$ret[] = $thead;
$ret = array_merge($ret, $after_thead_ws);
}
if ($tfoot !== false) {
$ret = array_merge($ret, $tfoot);
$ret[] = $tfoot;
$ret = array_merge($ret, $after_tfoot_ws);
}
if ($tbody_mode) {
// a little tricky, since the start of the collection may be
// whitespace
$inside_tbody = false;
foreach ($content as $token_array) {
// find the starting token
// INVARIANT: token_array is not empty
$t = NULL;
foreach ($token_array as $t) {
if ($t->name === 'tr' || $t->name === 'tbody') {
break;
// 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[] = $current_tr_tbody;
}
} // iterator variable carries over
if ($t->name === 'tr') {
if ($inside_tbody) {
$ret = array_merge($ret, $token_array);
$current_tr_tbody->children[] = $node;
break;
case '#PCDATA':
assert($node->is_whitespace);
if ($current_tr_tbody === null) {
$ret[] = $node;
} else {
$ret[] = new HTMLPurifier_Token_Start('tbody');
$ret = array_merge($ret, $token_array);
$inside_tbody = true;
$current_tr_tbody->children[] = $node;
}
} elseif ($t->name === 'tbody') {
if ($inside_tbody) {
$ret[] = new HTMLPurifier_Token_End('tbody');
$inside_tbody = false;
$ret = array_merge($ret, $token_array);
} else {
$ret = array_merge($ret, $token_array);
}
} else {
trigger_error("tr/tbody in content invariant failed in Table ChildDef", E_USER_ERROR);
break;
}
}
if ($inside_tbody) {
$ret[] = new HTMLPurifier_Token_End('tbody');
}
} else {
foreach ($content as $token_array) {
// invariant: everything in here is <tr>s
$ret = array_merge($ret, $token_array);
}
$ret = array_merge($ret, $content);
}
if (!empty($collection) && $is_collecting == false) {
// grab the trailing space
$ret = array_merge($ret, $collection);
}
array_pop($tokens_of_children); // remove phantom token
return ($ret === $tokens_of_children) ? true : $ret;
return $ret;
}
}

6
library/HTMLPurifier/ConfigSchema/schema/Core.EscapeInvalidChildren.txt
View File

@ -2,9 +2,11 @@ Core.EscapeInvalidChildren
TYPE: bool
DEFAULT: false
--DESCRIPTION--
When true, a child is found that is not allowed in the context of the
<p><strong>Warning:</strong> this configuration option is no longer does anything as of 4.6.0.</p>
<p>When true, a child is found that is not allowed in the context of the
parent element will be transformed into text as if it were ASCII. When
false, that element and all internal tags will be dropped, though text will
be preserved. There is no option for dropping the element but preserving
child nodes.
child nodes.</p>
--# vim: et sw=4 sts=4

9
library/HTMLPurifier/Node.php
View File

@ -29,6 +29,15 @@ abstract class HTMLPurifier_Node
*/
public $armor = array();
/**
* When true, this node should be ignored as non-existent.
*
* Who is responsible for ignoring dead nodes? FixNesting is
* responsible for removing them before passing on to child
* validators.
*/
public $dead = false;
/**
* Returns a pair of start and end tokens, where the end token
* is null if it is not necessary. Does not include children.

7
library/HTMLPurifier/Node/Text.php
View File

@ -12,6 +12,13 @@
class HTMLPurifier_Node_Text extends HTMLPurifier_Node
{
/**
* PCDATA tag name compatible with DTD, see
* HTMLPurifier_ChildDef_Custom for details.
* @type string
*/
public $name = '#PCDATA';
/**
* @type string
*/

366
library/HTMLPurifier/Strategy/FixNesting.php
View File

@ -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
@ -25,23 +25,8 @@
* @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.
*
* @warning This algorithm (though it may be hard to see) proceeds from
* a top-down fashion. Thus, parents are processed before
* children. This is easy to implement and has a nice effiency
* benefit, in that if a node is removed, we never waste any
* time processing it, but it also means that if a child
* changes in a non-encapsulated way (e.g. it is removed), we
* need to go back and reprocess the parent to see if those
* changes resulted in problems for the parent. See
* [BACKTRACK] for an example of this. In the current
* implementation, this backtracking can only be triggered when
* a node is removed and if that node was the sole node, the
* parent would need to be removed. As such, it is easy to see
* that backtracking only incurs constant overhead. If more
* sophisticated backtracking is implemented, care must be
* taken to avoid nontermination or exponential blowup.
* @todo Enable nodes to be bubbled out of the structure. This is
* easier with our new algorithm.
*/
class HTMLPurifier_Strategy_FixNesting extends HTMLPurifier_Strategy
@ -55,23 +40,19 @@ class HTMLPurifier_Strategy_FixNesting extends HTMLPurifier_Strategy
*/
public function execute($tokens, $config, $context)
{
//####################################################################//
// Pre-processing
//$node = HTMLPurifier_Arborize::arborize($tokens, $config, $context);
//$new_tokens = HTMLPurifier_Arborize::flatten($node, $config, $context);
// 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();
$excludes_enabled = !$config->get('Core.DisableExcludes');
// 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);
// setup the context variable 'IsInline', for chameleon processing
// is 'false' when we are not inline, 'true' when it must always
// be inline, and an integer when it is inline for a certain
@ -85,278 +66,115 @@ 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 < $size;) {
// We need to implement a post-order traversal iteratively, to
// avoid running into stack space limits. This is pretty tricky
// to reason about, so we just manually stack-ify the recursive
// variant:
//
// function f($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.
//################################################################//
// Gather information on children
$parent_def = $definition->info_parent_def;
$stack = array(
array($top_node,
$parent_def->descendants_are_inline,
$parent_def->excludes, // exclusions
0)
);
// 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;
}
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;
}
$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 = $definition->info_parent_def;
};
if ($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 {
// 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;
}
$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 {
$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) && $excludes_enabled) {
foreach ($exclude_stack as $lookup) {
if (isset($lookup[$tokens[$i]->name])) {
$excluded = true;
// no need to continue processing
break;
$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) {
$e->send(E_WARNING, 'Strategy_FixNesting: Node reorganized');
}
}
}
}
//################################################################//
// Perform child validation
if ($excluded) {
// there is an exclusion, remove the entire node
$result = false;
$excludes = array(); // not used, but good to initialize anyway
} else {
// DEFINITION CALL
if ($i === 0) {
// special processing for the first node
$def = $definition->info_parent_def;
} else {
$def = $definition->info[$tokens[$i]->name];
}
if (!empty($def->child)) {
// have DTD child def validate children
$result = $def->child->validateChildren(
$child_tokens,
$config,
$context
);
} 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 [BACKTRACK]
$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 {
$e->send(E_WARNING, 'Strategy_FixNesting: Node reorganized');
}
}
// perform replacement
array_splice($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);
}
}

5
tests/HTMLPurifier/ChildDef/CustomTest.php
View File

@ -3,6 +3,11 @@
class HTMLPurifier_ChildDef_CustomTest extends HTMLPurifier_ChildDefHarness
{
public function setUp()
{
parent::setUp();
}
public function test()
{
$this->obj = new HTMLPurifier_ChildDef_Custom('(a,b?,c*,d+,(a,b)*)');

6
tests/HTMLPurifier/ChildDef/ListTest.php
View File

@ -24,12 +24,6 @@ class HTMLPurifier_ChildDef_ListTest extends HTMLPurifier_ChildDefHarness
$this->assertResult('<li>asdf</li><li />');
}
public function testIllegal()
{
// XXX actually this never gets triggered in practice
$this->assertResult('<li /><b />', '<li /><li><b /></li>');
}
public function testOlAtBeginning()
{
$this->assertResult('<ol />', '<li><ol /></li>');

9
tests/HTMLPurifier/ChildDef/RequiredTest.php
View File

@ -68,15 +68,10 @@ class HTMLPurifier_ChildDef_RequiredTest extends HTMLPurifier_ChildDefHarness
$this->obj = new HTMLPurifier_ChildDef_Required('#PCDATA | b');
$this->assertResult('Out <b>Bold text</b><img />', 'Out <b>Bold text</b>');
}
public function testPCDATAAllowedWithEscaping()
public function testPCDATAAllowedJump()
{
$this->obj = new HTMLPurifier_ChildDef_Required('#PCDATA | b');
$this->config->set('Core.EscapeInvalidChildren', true);
$this->assertResult(
'Out <b>Bold text</b><img />',
'Out <b>Bold text</b>&lt;img /&gt;'
);
$this->assertResult('A <i>foo</i>', 'A foo');
}
}

2
tests/HTMLPurifier/ChildDef/TableTest.php
View File

@ -76,7 +76,7 @@ class HTMLPurifier_ChildDef_TableTest extends HTMLPurifier_ChildDefHarness
$this->config->set('Output.Newline', "\n");
$this->assertResult(
"\n\t<tbody />\n\t\t<tfoot />\n\t\t\t",
"\n\t\t<tfoot />\n\t<tbody />\n\t\t\t"
"\n\t<tfoot />\n\t\t\t<tbody />\n\t\t"
);
}

2
tests/HTMLPurifier/ChildDefHarness.php
View File

@ -8,8 +8,8 @@ class HTMLPurifier_ChildDefHarness extends HTMLPurifier_ComplexHarness
parent::setUp();
$this->obj = null;
$this->func = 'validateChildren';
$this->to_tokens = true;
$this->to_html = true;
$this->to_node_list = true;
}
}

33
tests/HTMLPurifier/ComplexHarness.php
View File

@ -28,6 +28,14 @@ class HTMLPurifier_ComplexHarness extends HTMLPurifier_Harness
*/
protected $to_tokens = false;
/**
* Whether or not the method deals in a node list.
* If set to true, assertResult() will transparently convert HTML
* to and back from node.
* @type bool
*/
protected $to_node_list = false;
/**
* Whether or not to convert tokens back into HTML before performing
* equality check, has no effect on bools.
@ -54,9 +62,12 @@ class HTMLPurifier_ComplexHarness extends HTMLPurifier_Harness
*/
protected function assertResult($input, $expect = true)
{
if ($this->to_tokens && is_string($input)) {
// $func may cause $input to change, so "clone" another copy
// to sacrifice
// $func may cause $input to change, so "clone" another copy
// to sacrifice
if ($this->to_node_list && is_string($input)) {
$input = HTMLPurifier_Arborize::arborize($this->tokenize($temp = $input), $this->config, $this->context)->children;
$input_c = HTMLPurifier_Arborize::arborize($this->tokenize($temp), $this->config, $this->context)->children;
} elseif ($this->to_tokens && is_string($input)) {
$input = $this->tokenize($temp = $input);
$input_c = $this->tokenize($temp);
} else {
@ -76,6 +87,12 @@ class HTMLPurifier_ComplexHarness extends HTMLPurifier_Harness
}
if ($this->to_html) {
if ($this->to_node_list) {
$result = $this->generateTokens($result);
if (is_array($expect) && !empty($expect) && $expect[0] instanceof HTMLPurifier_Node) {
$expect = $this->generateTokens($expect);
}
}
$result = $this->generate($result);
if (is_array($expect)) {
$expect = $this->generate($expect);
@ -106,6 +123,16 @@ class HTMLPurifier_ComplexHarness extends HTMLPurifier_Harness
return $generator->generateFromTokens($tokens);
}
/**
* Generate tokens from node list
*/
protected function generateTokens($children)
{
$dummy = new HTMLPurifier_Node_Element("dummy");
$dummy->children = $children;
return HTMLPurifier_Arborize::flatten($dummy, $this->context, $this->config);
}
}
// vim: et sw=4 sts=4

18
tests/HTMLPurifier/Strategy/FixNestingTest.php
View File

@ -27,15 +27,6 @@ class HTMLPurifier_Strategy_FixNestingTest extends HTMLPurifier_StrategyHarness
);
}
public function testEscapeBlockInInline()
{
$this->config->set('Core.EscapeInvalidChildren', true);
$this->assertResult(
'<b><div>Illegal div.</div></b>',
'<b>&lt;div&gt;Illegal div.&lt;/div&gt;</b>'
);
}
public function testRemoveNodeWithMissingRequiredElements()
{
$this->assertResult('<ul></ul>', '');
@ -98,15 +89,6 @@ class HTMLPurifier_Strategy_FixNestingTest extends HTMLPurifier_StrategyHarness
);
}
public function testChameleonEscapeInvalidBlockInInline()
{
$this->config->set('Core.EscapeInvalidChildren', true);
$this->assertResult( // alt config
'<span><ins><div>Not allowed!</div></ins></span>',
'<span><ins>&lt;div&gt;Not allowed!&lt;/div&gt;</ins></span>'
);
}
public function testExclusionsIntegration()
{
// test exclusions