Skip to content
Permalink
3.0
Go to file
 
 
Cannot retrieve contributors at this time
1039 lines (902 sloc) 31 KB
<?php
/*
* This file is part of the Symfony package.
*
* (c) Fabien Potencier <fabien@symfony.com>
*
* For the full copyright and license information, please view the LICENSE
* file that was distributed with this source code.
*/
namespace Symfony\Component\DomCrawler;
use Symfony\Component\CssSelector\CssSelectorConverter;
/**
* Crawler eases navigation of a list of \DOMNode objects.
*
* @author Fabien Potencier <fabien@symfony.com>
*/
class Crawler implements \Countable, \IteratorAggregate
{
/**
* @var string The current URI
*/
protected $uri;
/**
* @var string The default namespace prefix to be used with XPath and CSS expressions
*/
private $defaultNamespacePrefix = 'default';
/**
* @var array A map of manually registered namespaces
*/
private $namespaces = array();
/**
* @var string The base href value
*/
private $baseHref;
/**
* @var \DOMDocument|null
*/
private $document;
/**
* @var \DOMElement[]
*/
private $nodes = array();
/**
* Whether the Crawler contains HTML or XML content (used when converting CSS to XPath).
*
* @var bool
*/
private $isHtml = true;
/**
* Constructor.
*
* @param mixed $node A Node to use as the base for the crawling
* @param string $currentUri The current URI
* @param string $baseHref The base href value
*/
public function __construct($node = null, $currentUri = null, $baseHref = null)
{
$this->uri = $currentUri;
$this->baseHref = $baseHref ?: $currentUri;
$this->add($node);
}
/**
* Removes all the nodes.
*/
public function clear()
{
$this->nodes = array();
$this->document = null;
}
/**
* Adds a node to the current list of nodes.
*
* This method uses the appropriate specialized add*() method based
* on the type of the argument.
*
* @param \DOMNodeList|\DOMNode|array|string|null $node A node
*
* @throws \InvalidArgumentException When node is not the expected type.
*/
public function add($node)
{
if ($node instanceof \DOMNodeList) {
$this->addNodeList($node);
} elseif ($node instanceof \DOMNode) {
$this->addNode($node);
} elseif (is_array($node)) {
$this->addNodes($node);
} elseif (is_string($node)) {
$this->addContent($node);
} elseif (null !== $node) {
throw new \InvalidArgumentException(sprintf('Expecting a DOMNodeList or DOMNode instance, an array, a string, or null, but got "%s".', is_object($node) ? get_class($node) : gettype($node)));
}
}
/**
* Adds HTML/XML content.
*
* If the charset is not set via the content type, it is assumed
* to be ISO-8859-1, which is the default charset defined by the
* HTTP 1.1 specification.
*
* @param string $content A string to parse as HTML/XML
* @param null|string $type The content type of the string
*/
public function addContent($content, $type = null)
{
if (empty($type)) {
$type = 0 === strpos($content, '<?xml') ? 'application/xml' : 'text/html';
}
// DOM only for HTML/XML content
if (!preg_match('/(x|ht)ml/i', $type, $xmlMatches)) {
return;
}
$charset = null;
if (false !== $pos = stripos($type, 'charset=')) {
$charset = substr($type, $pos + 8);
if (false !== $pos = strpos($charset, ';')) {
$charset = substr($charset, 0, $pos);
}
}
// http://www.w3.org/TR/encoding/#encodings
// http://www.w3.org/TR/REC-xml/#NT-EncName
if (null === $charset &&
preg_match('/\<meta[^\>]+charset *= *["\']?([a-zA-Z\-0-9_:.]+)/i', $content, $matches)) {
$charset = $matches[1];
}
if (null === $charset) {
$charset = 'ISO-8859-1';
}
if ('x' === $xmlMatches[1]) {
$this->addXmlContent($content, $charset);
} else {
$this->addHtmlContent($content, $charset);
}
}
/**
* Adds an HTML content to the list of nodes.
*
* The libxml errors are disabled when the content is parsed.
*
* If you want to get parsing errors, be sure to enable
* internal errors via libxml_use_internal_errors(true)
* and then, get the errors via libxml_get_errors(). Be
* sure to clear errors with libxml_clear_errors() afterward.
*
* @param string $content The HTML content
* @param string $charset The charset
*/
public function addHtmlContent($content, $charset = 'UTF-8')
{
$internalErrors = libxml_use_internal_errors(true);
$disableEntities = libxml_disable_entity_loader(true);
$dom = new \DOMDocument('1.0', $charset);
$dom->validateOnParse = true;
set_error_handler(function () {throw new \Exception();});
try {
// Convert charset to HTML-entities to work around bugs in DOMDocument::loadHTML()
$content = mb_convert_encoding($content, 'HTML-ENTITIES', $charset);
} catch (\Exception $e) {
}
restore_error_handler();
if ('' !== trim($content)) {
@$dom->loadHTML($content);
}
libxml_use_internal_errors($internalErrors);
libxml_disable_entity_loader($disableEntities);
$this->addDocument($dom);
$base = $this->filterRelativeXPath('descendant-or-self::base')->extract(array('href'));
$baseHref = current($base);
if (count($base) && !empty($baseHref)) {
if ($this->baseHref) {
$linkNode = $dom->createElement('a');
$linkNode->setAttribute('href', $baseHref);
$link = new Link($linkNode, $this->baseHref);
$this->baseHref = $link->getUri();
} else {
$this->baseHref = $baseHref;
}
}
}
/**
* Adds an XML content to the list of nodes.
*
* The libxml errors are disabled when the content is parsed.
*
* If you want to get parsing errors, be sure to enable
* internal errors via libxml_use_internal_errors(true)
* and then, get the errors via libxml_get_errors(). Be
* sure to clear errors with libxml_clear_errors() afterward.
*
* @param string $content The XML content
* @param string $charset The charset
* @param int $options Bitwise OR of the libxml option constants
* LIBXML_PARSEHUGE is dangerous, see
* http://symfony.com/blog/security-release-symfony-2-0-17-released
*/
public function addXmlContent($content, $charset = 'UTF-8', $options = LIBXML_NONET)
{
// remove the default namespace if it's the only namespace to make XPath expressions simpler
if (!preg_match('/xmlns:/', $content)) {
$content = str_replace('xmlns', 'ns', $content);
}
$internalErrors = libxml_use_internal_errors(true);
$disableEntities = libxml_disable_entity_loader(true);
$dom = new \DOMDocument('1.0', $charset);
$dom->validateOnParse = true;
if ('' !== trim($content)) {
@$dom->loadXML($content, $options);
}
libxml_use_internal_errors($internalErrors);
libxml_disable_entity_loader($disableEntities);
$this->addDocument($dom);
$this->isHtml = false;
}
/**
* Adds a \DOMDocument to the list of nodes.
*
* @param \DOMDocument $dom A \DOMDocument instance
*/
public function addDocument(\DOMDocument $dom)
{
if ($dom->documentElement) {
$this->addNode($dom->documentElement);
}
}
/**
* Adds a \DOMNodeList to the list of nodes.
*
* @param \DOMNodeList $nodes A \DOMNodeList instance
*/
public function addNodeList(\DOMNodeList $nodes)
{
foreach ($nodes as $node) {
if ($node instanceof \DOMNode) {
$this->addNode($node);
}
}
}
/**
* Adds an array of \DOMNode instances to the list of nodes.
*
* @param \DOMNode[] $nodes An array of \DOMNode instances
*/
public function addNodes(array $nodes)
{
foreach ($nodes as $node) {
$this->add($node);
}
}
/**
* Adds a \DOMNode instance to the list of nodes.
*
* @param \DOMNode $node A \DOMNode instance
*/
public function addNode(\DOMNode $node)
{
if ($node instanceof \DOMDocument) {
$node = $node->documentElement;
}
if (null !== $this->document && $this->document !== $node->ownerDocument) {
throw new \InvalidArgumentException('Attaching DOM nodes from multiple documents in the same crawler is forbidden.');
}
if (null === $this->document) {
$this->document = $node->ownerDocument;
}
// Don't add duplicate nodes in the Crawler
if (in_array($node, $this->nodes, true)) {
return;
}
$this->nodes[] = $node;
}
/**
* Returns a node given its position in the node list.
*
* @param int $position The position
*
* @return Crawler A new instance of the Crawler with the selected node, or an empty Crawler if it does not exist
*/
public function eq($position)
{
if (isset($this->nodes[$position])) {
return $this->createSubCrawler($this->nodes[$position]);
}
return $this->createSubCrawler(null);
}
/**
* Calls an anonymous function on each node of the list.
*
* The anonymous function receives the position and the node wrapped
* in a Crawler instance as arguments.
*
* Example:
*
* $crawler->filter('h1')->each(function ($node, $i) {
* return $node->text();
* });
*
* @param \Closure $closure An anonymous function
*
* @return array An array of values returned by the anonymous function
*/
public function each(\Closure $closure)
{
$data = array();
foreach ($this->nodes as $i => $node) {
$data[] = $closure($this->createSubCrawler($node), $i);
}
return $data;
}
/**
* Slices the list of nodes by $offset and $length.
*
* @param int $offset
* @param int $length
*
* @return Crawler A Crawler instance with the sliced nodes
*/
public function slice($offset = 0, $length = null)
{
return $this->createSubCrawler(array_slice($this->nodes, $offset, $length));
}
/**
* Reduces the list of nodes by calling an anonymous function.
*
* To remove a node from the list, the anonymous function must return false.
*
* @param \Closure $closure An anonymous function
*
* @return Crawler A Crawler instance with the selected nodes
*/
public function reduce(\Closure $closure)
{
$nodes = array();
foreach ($this->nodes as $i => $node) {
if (false !== $closure($this->createSubCrawler($node), $i)) {
$nodes[] = $node;
}
}
return $this->createSubCrawler($nodes);
}
/**
* Returns the first node of the current selection.
*
* @return Crawler A Crawler instance with the first selected node
*/
public function first()
{
return $this->eq(0);
}
/**
* Returns the last node of the current selection.
*
* @return Crawler A Crawler instance with the last selected node
*/
public function last()
{
return $this->eq(count($this->nodes) - 1);
}
/**
* Returns the siblings nodes of the current selection.
*
* @return Crawler A Crawler instance with the sibling nodes
*
* @throws \InvalidArgumentException When current node is empty
*/
public function siblings()
{
if (!$this->nodes) {
throw new \InvalidArgumentException('The current node list is empty.');
}
return $this->createSubCrawler($this->sibling($this->getNode(0)->parentNode->firstChild));
}
/**
* Returns the next siblings nodes of the current selection.
*
* @return Crawler A Crawler instance with the next sibling nodes
*
* @throws \InvalidArgumentException When current node is empty
*/
public function nextAll()
{
if (!$this->nodes) {
throw new \InvalidArgumentException('The current node list is empty.');
}
return $this->createSubCrawler($this->sibling($this->getNode(0)));
}
/**
* Returns the previous sibling nodes of the current selection.
*
* @return Crawler A Crawler instance with the previous sibling nodes
*
* @throws \InvalidArgumentException
*/
public function previousAll()
{
if (!$this->nodes) {
throw new \InvalidArgumentException('The current node list is empty.');
}
return $this->createSubCrawler($this->sibling($this->getNode(0), 'previousSibling'));
}
/**
* Returns the parents nodes of the current selection.
*
* @return Crawler A Crawler instance with the parents nodes of the current selection
*
* @throws \InvalidArgumentException When current node is empty
*/
public function parents()
{
if (!$this->nodes) {
throw new \InvalidArgumentException('The current node list is empty.');
}
$node = $this->getNode(0);
$nodes = array();
while ($node = $node->parentNode) {
if (XML_ELEMENT_NODE === $node->nodeType) {
$nodes[] = $node;
}
}
return $this->createSubCrawler($nodes);
}
/**
* Returns the children nodes of the current selection.
*
* @return Crawler A Crawler instance with the children nodes
*
* @throws \InvalidArgumentException When current node is empty
*/
public function children()
{
if (!$this->nodes) {
throw new \InvalidArgumentException('The current node list is empty.');
}
$node = $this->getNode(0)->firstChild;
return $this->createSubCrawler($node ? $this->sibling($node) : array());
}
/**
* Returns the attribute value of the first node of the list.
*
* @param string $attribute The attribute name
*
* @return string|null The attribute value or null if the attribute does not exist
*
* @throws \InvalidArgumentException When current node is empty
*/
public function attr($attribute)
{
if (!$this->nodes) {
throw new \InvalidArgumentException('The current node list is empty.');
}
$node = $this->getNode(0);
return $node->hasAttribute($attribute) ? $node->getAttribute($attribute) : null;
}
/**
* Returns the node name of the first node of the list.
*
* @return string The node name
*
* @throws \InvalidArgumentException When current node is empty
*/
public function nodeName()
{
if (!$this->nodes) {
throw new \InvalidArgumentException('The current node list is empty.');
}
return $this->getNode(0)->nodeName;
}
/**
* Returns the node value of the first node of the list.
*
* @return string The node value
*
* @throws \InvalidArgumentException When current node is empty
*/
public function text()
{
if (!$this->nodes) {
throw new \InvalidArgumentException('The current node list is empty.');
}
return $this->getNode(0)->nodeValue;
}
/**
* Returns the first node of the list as HTML.
*
* @return string The node html
*
* @throws \InvalidArgumentException When current node is empty
*/
public function html()
{
if (!$this->nodes) {
throw new \InvalidArgumentException('The current node list is empty.');
}
$html = '';
foreach ($this->getNode(0)->childNodes as $child) {
$html .= $child->ownerDocument->saveHTML($child);
}
return $html;
}
/**
* Extracts information from the list of nodes.
*
* You can extract attributes or/and the node value (_text).
*
* Example:
*
* $crawler->filter('h1 a')->extract(array('_text', 'href'));
*
* @param array $attributes An array of attributes
*
* @return array An array of extracted values
*/
public function extract($attributes)
{
$attributes = (array) $attributes;
$count = count($attributes);
$data = array();
foreach ($this->nodes as $node) {
$elements = array();
foreach ($attributes as $attribute) {
if ('_text' === $attribute) {
$elements[] = $node->nodeValue;
} else {
$elements[] = $node->getAttribute($attribute);
}
}
$data[] = $count > 1 ? $elements : $elements[0];
}
return $data;
}
/**
* Filters the list of nodes with an XPath expression.
*
* The XPath expression is evaluated in the context of the crawler, which
* is considered as a fake parent of the elements inside it.
* This means that a child selector "div" or "./div" will match only
* the div elements of the current crawler, not their children.
*
* @param string $xpath An XPath expression
*
* @return Crawler A new instance of Crawler with the filtered list of nodes
*/
public function filterXPath($xpath)
{
$xpath = $this->relativize($xpath);
// If we dropped all expressions in the XPath while preparing it, there would be no match
if ('' === $xpath) {
return $this->createSubCrawler(null);
}
return $this->filterRelativeXPath($xpath);
}
/**
* Filters the list of nodes with a CSS selector.
*
* This method only works if you have installed the CssSelector Symfony Component.
*
* @param string $selector A CSS selector
*
* @return Crawler A new instance of Crawler with the filtered list of nodes
*
* @throws \RuntimeException if the CssSelector Component is not available
*/
public function filter($selector)
{
if (!class_exists('Symfony\\Component\\CssSelector\\CssSelectorConverter')) {
throw new \RuntimeException('Unable to filter with a CSS selector as the Symfony CssSelector 2.8+ is not installed (you can use filterXPath instead).');
}
$converter = new CssSelectorConverter($this->isHtml);
// The CssSelector already prefixes the selector with descendant-or-self::
return $this->filterRelativeXPath($converter->toXPath($selector));
}
/**
* Selects links by name or alt value for clickable images.
*
* @param string $value The link text
*
* @return Crawler A new instance of Crawler with the filtered list of nodes
*/
public function selectLink($value)
{
$xpath = sprintf('descendant-or-self::a[contains(concat(\' \', normalize-space(string(.)), \' \'), %s) ', static::xpathLiteral(' '.$value.' ')).
sprintf('or ./img[contains(concat(\' \', normalize-space(string(@alt)), \' \'), %s)]]', static::xpathLiteral(' '.$value.' '));
return $this->filterRelativeXPath($xpath);
}
/**
* Selects a button by name or alt value for images.
*
* @param string $value The button text
*
* @return Crawler A new instance of Crawler with the filtered list of nodes
*/
public function selectButton($value)
{
$translate = 'translate(@type, "ABCDEFGHIJKLMNOPQRSTUVWXYZ", "abcdefghijklmnopqrstuvwxyz")';
$xpath = sprintf('descendant-or-self::input[((contains(%s, "submit") or contains(%s, "button")) and contains(concat(\' \', normalize-space(string(@value)), \' \'), %s)) ', $translate, $translate, static::xpathLiteral(' '.$value.' ')).
sprintf('or (contains(%s, "image") and contains(concat(\' \', normalize-space(string(@alt)), \' \'), %s)) or @id=%s or @name=%s] ', $translate, static::xpathLiteral(' '.$value.' '), static::xpathLiteral($value), static::xpathLiteral($value)).
sprintf('| descendant-or-self::button[contains(concat(\' \', normalize-space(string(.)), \' \'), %s) or @id=%s or @name=%s]', static::xpathLiteral(' '.$value.' '), static::xpathLiteral($value), static::xpathLiteral($value));
return $this->filterRelativeXPath($xpath);
}
/**
* Returns a Link object for the first node in the list.
*
* @param string $method The method for the link (get by default)
*
* @return Link A Link instance
*
* @throws \InvalidArgumentException If the current node list is empty or the selected node is not instance of DOMElement
*/
public function link($method = 'get')
{
if (!$this->nodes) {
throw new \InvalidArgumentException('The current node list is empty.');
}
$node = $this->getNode(0);
if (!$node instanceof \DOMElement) {
throw new \InvalidArgumentException(sprintf('The selected node should be instance of DOMElement, got "%s".', get_class($node)));
}
return new Link($node, $this->baseHref, $method);
}
/**
* Returns an array of Link objects for the nodes in the list.
*
* @return Link[] An array of Link instances
*
* @throws \InvalidArgumentException If the current node list contains non-DOMElement instances
*/
public function links()
{
$links = array();
foreach ($this->nodes as $node) {
if (!$node instanceof \DOMElement) {
throw new \InvalidArgumentException(sprintf('The current node list should contain only DOMElement instances, "%s" found.', get_class($node)));
}
$links[] = new Link($node, $this->baseHref, 'get');
}
return $links;
}
/**
* Returns a Form object for the first node in the list.
*
* @param array $values An array of values for the form fields
* @param string $method The method for the form
*
* @return Form A Form instance
*
* @throws \InvalidArgumentException If the current node list is empty or the selected node is not instance of DOMElement
*/
public function form(array $values = null, $method = null)
{
if (!$this->nodes) {
throw new \InvalidArgumentException('The current node list is empty.');
}
$node = $this->getNode(0);
if (!$node instanceof \DOMElement) {
throw new \InvalidArgumentException(sprintf('The selected node should be instance of DOMElement, got "%s".', get_class($node)));
}
$form = new Form($node, $this->uri, $method, $this->baseHref);
if (null !== $values) {
$form->setValues($values);
}
return $form;
}
/**
* Overloads a default namespace prefix to be used with XPath and CSS expressions.
*
* @param string $prefix
*/
public function setDefaultNamespacePrefix($prefix)
{
$this->defaultNamespacePrefix = $prefix;
}
/**
* @param string $prefix
* @param string $namespace
*/
public function registerNamespace($prefix, $namespace)
{
$this->namespaces[$prefix] = $namespace;
}
/**
* Converts string for XPath expressions.
*
* Escaped characters are: quotes (") and apostrophe (').
*
* Examples:
* <code>
* echo Crawler::xpathLiteral('foo " bar');
* //prints 'foo " bar'
*
* echo Crawler::xpathLiteral("foo ' bar");
* //prints "foo ' bar"
*
* echo Crawler::xpathLiteral('a\'b"c');
* //prints concat('a', "'", 'b"c')
* </code>
*
* @param string $s String to be escaped
*
* @return string Converted string
*/
public static function xpathLiteral($s)
{
if (false === strpos($s, "'")) {
return sprintf("'%s'", $s);
}
if (false === strpos($s, '"')) {
return sprintf('"%s"', $s);
}
$string = $s;
$parts = array();
while (true) {
if (false !== $pos = strpos($string, "'")) {
$parts[] = sprintf("'%s'", substr($string, 0, $pos));
$parts[] = "\"'\"";
$string = substr($string, $pos + 1);
} else {
$parts[] = "'$string'";
break;
}
}
return sprintf('concat(%s)', implode($parts, ', '));
}
/**
* Filters the list of nodes with an XPath expression.
*
* The XPath expression should already be processed to apply it in the context of each node.
*
* @param string $xpath
*
* @return Crawler
*/
private function filterRelativeXPath($xpath)
{
$prefixes = $this->findNamespacePrefixes($xpath);
$crawler = $this->createSubCrawler(null);
foreach ($this->nodes as $node) {
$domxpath = $this->createDOMXPath($node->ownerDocument, $prefixes);
$crawler->add($domxpath->query($xpath, $node));
}
return $crawler;
}
/**
* Make the XPath relative to the current context.
*
* The returned XPath will match elements matching the XPath inside the current crawler
* when running in the context of a node of the crawler.
*
* @param string $xpath
*
* @return string
*/
private function relativize($xpath)
{
$expressions = array();
$unionPattern = '/\|(?![^\[]*\])/';
// An expression which will never match to replace expressions which cannot match in the crawler
// We cannot simply drop
$nonMatchingExpression = 'a[name() = "b"]';
// Split any unions into individual expressions.
foreach (preg_split($unionPattern, $xpath) as $expression) {
$expression = trim($expression);
$parenthesis = '';
// If the union is inside some braces, we need to preserve the opening braces and apply
// the change only inside it.
if (preg_match('/^[\(\s*]+/', $expression, $matches)) {
$parenthesis = $matches[0];
$expression = substr($expression, strlen($parenthesis));
}
if (0 === strpos($expression, 'self::*/')) {
$expression = './'.substr($expression, 8);
}
// add prefix before absolute element selector
if (empty($expression)) {
$expression = $nonMatchingExpression;
} elseif (0 === strpos($expression, '//')) {
$expression = 'descendant-or-self::'.substr($expression, 2);
} elseif (0 === strpos($expression, './/')) {
$expression = 'descendant-or-self::'.substr($expression, 3);
} elseif (0 === strpos($expression, './')) {
$expression = 'self::'.substr($expression, 2);
} elseif (0 === strpos($expression, 'child::')) {
$expression = 'self::'.substr($expression, 7);
} elseif ('/' === $expression[0] || '.' === $expression[0] || 0 === strpos($expression, 'self::')) {
$expression = $nonMatchingExpression;
} elseif (0 === strpos($expression, 'descendant::')) {
$expression = 'descendant-or-self::'.substr($expression, strlen('descendant::'));
} elseif (preg_match('/^(ancestor|ancestor-or-self|attribute|following|following-sibling|namespace|parent|preceding|preceding-sibling)::/', $expression)) {
// the fake root has no parent, preceding or following nodes and also no attributes (even no namespace attributes)
$expression = $nonMatchingExpression;
} elseif (0 !== strpos($expression, 'descendant-or-self::')) {
$expression = 'self::'.$expression;
}
$expressions[] = $parenthesis.$expression;
}
return implode(' | ', $expressions);
}
/**
* @param int $position
*
* @return \DOMElement|null
*/
public function getNode($position)
{
if (isset($this->nodes[$position])) {
return $this->nodes[$position];
}
}
/**
* @return int
*/
public function count()
{
return count($this->nodes);
}
/**
* @return \ArrayIterator
*/
public function getIterator()
{
return new \ArrayIterator($this->nodes);
}
/**
* @param \DOMElement $node
* @param string $siblingDir
*
* @return array
*/
protected function sibling($node, $siblingDir = 'nextSibling')
{
$nodes = array();
do {
if ($node !== $this->getNode(0) && $node->nodeType === 1) {
$nodes[] = $node;
}
} while ($node = $node->$siblingDir);
return $nodes;
}
/**
* @param \DOMDocument $document
* @param array $prefixes
*
* @return \DOMXPath
*
* @throws \InvalidArgumentException
*/
private function createDOMXPath(\DOMDocument $document, array $prefixes = array())
{
$domxpath = new \DOMXPath($document);
foreach ($prefixes as $prefix) {
$namespace = $this->discoverNamespace($domxpath, $prefix);
if (null !== $namespace) {
$domxpath->registerNamespace($prefix, $namespace);
}
}
return $domxpath;
}
/**
* @param \DOMXPath $domxpath
* @param string $prefix
*
* @return string
*
* @throws \InvalidArgumentException
*/
private function discoverNamespace(\DOMXPath $domxpath, $prefix)
{
if (isset($this->namespaces[$prefix])) {
return $this->namespaces[$prefix];
}
// ask for one namespace, otherwise we'd get a collection with an item for each node
$namespaces = $domxpath->query(sprintf('(//namespace::*[name()="%s"])[last()]', $this->defaultNamespacePrefix === $prefix ? '' : $prefix));
if ($node = $namespaces->item(0)) {
return $node->nodeValue;
}
}
/**
* @param string $xpath
*
* @return array
*/
private function findNamespacePrefixes($xpath)
{
if (preg_match_all('/(?P<prefix>[a-z_][a-z_0-9\-\.]*+):[^"\/:]/i', $xpath, $matches)) {
return array_unique($matches['prefix']);
}
return array();
}
/**
* Creates a crawler for some subnodes.
*
* @param \DOMElement|\DOMElement[]|\DOMNodeList|null $nodes
*
* @return static
*/
private function createSubCrawler($nodes)
{
$crawler = new static($nodes, $this->uri, $this->baseHref);
$crawler->isHtml = $this->isHtml;
$crawler->document = $this->document;
$crawler->namespaces = $this->namespaces;
return $crawler;
}
}
You can’t perform that action at this time.