CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

[3.5.6] - 2026-01-05

Fixed

• Potential issues found with PHPStan 2 on level 8.

[3.5.5] - 2025-08-05

Fixed

• Removed the overriding validateAndSanitizeInput() method from the Paginate HTTP step to ensure features like staticUrl() and useInputKeyAsUrl() work correctly.
• The Paginate HTTP step now also supports receiving an array of URLs, initiating pagination separately for each one.

Deprecated

• The Crwlr\Crawler\Steps\Loading\Http\Paginate class. It shall be removed and its behavior implemented in the Http class directly, in the next major version.

[3.5.4] - 2025-07-28

Fixed

• An issue in the SimpleWebsitePaginator when used with stop rules.

[3.5.3] - 2025-06-10

Fixed

• Issues with passing cookies from the cookie jar to the headless browser when using the useBrowser() method on Http steps, in cases where the loader wasn’t globally configured to use the browser for all requests.

[3.5.2] - 2025-05-16

Fixed

• The Result::toArray() method now converts all objects contained in the Result array (at any level of the array) to arrays. Also, if the only element in a result array has some autogenerated key containing "unnamed", but the value also is an associative array with string keys, the method only returns that child array.

[3.5.1] - 2025-04-23

Fixed

• An issue that occurred, when a step uses the PreStepInvocationLogger. As refiners also use the logger, a newer logger (replacing the PreStepInvocationLogger) is now also passed to all registered refiners of a step.
• Enable applying refiners to output properties with array value. E.g. if a step outputs an array of URLs (['https://...', 'https://...']), a UrlRefiner will be applied to all those URLs.

[3.5.0] - 2025-04-10

Added

• Dynamically building request URLs from extracted data: Http steps now have a new staticUrl() method, and you can also use variables within that static URL - as well as in request headers and the body - like https://www.example.com/foo/[crwl:some_extracted_property]. These variables will be replaced with the corresponding properties from input data (also works with kept data).
• New Refiners:
  • DateTimeRefiner::reformat('Y-m-d H:i:s') to reformat a date time string to a different format. Tries to automatically recognize the input format. If this does not work, you can provide an input format to use as the second argument.
  • HtmlRefiner::remove('#foo') to remove nodes matching the given selector from selected HTML.
• Steps that produce multiple outputs per input can now group them per input by calling the new Step::oneOutputPerInput() method.

[3.4.5] - 2025-04-09

Fixed

• When feeding an Http step with a string that is not a valid URL (e.g. https://), the exception when trying to parse it as a URL is caught, and an error logged.

[3.4.4] - 2025-04-04

Fixed

• As sometimes, XML parsing errors occur because of characters that aren't valid within XML documents, the library now catches XML parsing errors, tries to find and replace invalid characters (with transliterates or HTML entities) and retries parsing the document. Works best when you additionally install the voku/portable-ascii composer package.

[3.4.3] - 2025-04-03

Fixed

• When providing an empty base selector to an Html step (Html::each(''), Html::first(''), Html::last('')), it won't fail with an error, but instead log a warning, that it most likely doesn't make sense.
• The Step::keep() methods now also work when applied to child steps within a group step.

[3.4.2] - 2025-03-08

Fixed

• Issue when using Http::get()->useBrowser()->postBrowserNavigateHook(). Previously in this case, when the loader is configured to use the HTTP client, the post browser navigate hook was actually not set because of an issue with the order, things happened internally.

[3.4.1] - 2025-03-08

Fixed

• Since, when using the Chrome browser for loading, we can only execute GET requests:
  • The loader now automatically switches to the HTTP client for POST, PUT, PATCH, and DELETE requests and logs a warning.
  • A warning is logged when attempting to use "Post Browser Navigate Hooks" with POST, PUT, PATCH, or DELETE requests.
  • Consequently, the useBrowser() method, introduced in v3.4.0, is also limited to GET requests.

[3.4.0] - 2025-03-06

Added

• Two new methods to the base class of all Http steps:
  • skipCache() – Allows using the cache while skipping it for a specific loading step.
  • useBrowser() – Switches the loader to use a (headless) Chrome browser for loading calls in a specific step and then reverts the loader to its previous setting.
• Introduced the new BrowserAction::screenshot() post browser navigate hook. It accepts an instance of the new ScreenshotConfig class, allowing you to configure various options (see the methods of ScreenshotConfig). If successful, the screenshot file paths are included in the RespondedRequest output object of the Http step.

[3.3.0] - 2025-03-02

Added

• New BrowserActions to use with the postBrowserNavigateHook() method:
  • BrowserAction::clickInsideShadowDom()
  • BrowserAction::moveMouseToElement()
  • BrowserAction::moveMouseToPosition()
  • BrowserAction::scrollDown()
  • BrowserAction::scrollUp()
  • BrowserAction::typeText()
  • BrowserAction::waitForReload()
• A new method in HeadlessBrowserLoaderHelper to include the HTML content of shadow DOM elements in the returned HTML. Use it like this: $crawler->getLoader()->browser()->includeShadowElementsInHtml().

Changed

• The BrowserAction::clickElement() action, now automatically waits for an element matching the selector to be rendered, before performing the click. This means you don't need to put a BrowserAction::waitUntilDocumentContainsElement() before it. It works the same in the new BrowserAction::clickInsideShadowDom() and BrowserAction::moveMouseToElement() actions.

Deprecated

• BrowserAction::clickElementAndWaitForReload() and BrowserAction::evaluateAndWaitForReload(). As a replacement, please use BrowserAction::clickElement() or BrowserAction::evaluate() and BrowserAction::waitForReload() separately.

[3.2.5] - 2025-02-26

Fixed

• When a child step is nested in the extract() method of an Html or Xml step, and does not use each() as the base, the extracted value is an array with the keys defined in the extract() call, rather than an array of such arrays as it would be with each() as base.

[3.2.4] - 2025-02-25

Fixed

• Trying to load a relative reference URI (no scheme and host/authority, only path) via the HttpLoader now immediately logs (or throws when loadOrFail() is used) an error instead of trying to actually load it.

[3.2.3] - 2025-01-28

Fixed

• Fix deprecation warning triggered in the DomQuery class, when trying to get the value of an HTML/XML attribute that does not exist on the element.

[3.2.2] - 2025-01-17

Fixed

• Warnings about loader hooks being called multiple times, when using a BotUserAgent and therefore loading and respecting the robots.txt file, or when using the Http::stopOnErrorResponse() method.

[3.2.1] - 2025-01-13

Fixed

• Reuse previously opened page when using the (headless) Chrome browser, instead of opening a new page for each request.

[3.2.0] - 2025-01-12

Added

• RespondedRequest::isServedFromCache() to determine whether a response was served from cache or actually loaded.

[3.1.5] - 2025-01-10

Fixed

• Another improvement for getting XML source when using the browser, in cases where Chrome doesn't identify the response as an XML document (even though a Content-Type header is sent).

[3.1.4] - 2025-01-10

Fixed

• HttpLoader::dontUseCookies() now also works when using the Chrome browser. Cookies are cleared before every request.

[3.1.3] - 2025-01-10

Fixed

• Further improve getting the raw response body from non-HTML documents via Chrome browser.

[3.1.2] - 2025-01-08

Fixed

• When loading a non-HTML document (e.g., XML) via the Chrome browser, the library now retrieves the original source. Previously, it returned the outerHTML of the rendered document, which wrapped the content in an HTML structure.

[3.1.1] - 2025-01-07

Fixed

• When the validateAndSanitize() method of a step throws an InvalidArgumentException, the exception is now caught, logged and the step is not invoked with the invalid input. This improves fault tolerance. Feeding a step with one invalid input shouldn't cause the whole crawler run to fail. Exceptions other than InvalidArgumentException remain uncaught.

[3.1.0] - 2025-01-03

Added

• New method HeadlessBrowserLoaderHelper::setPageInitScript() ($crawler->getLoader()->browser()->setPageInitScript()) to provide javascript code that is executed on every new browser page before navigating anywhere.
• New method HeadlessBrowserLoaderHelper::useNativeUserAgent() ($crawler->getLoader()->browser()->useNativeUserAgent()) to allow using the native User-Agent that your Chrome browser sends by default.

[3.0.4] - 2024-12-18

Fixed

• Minor improvement for the DomQuery (base for Dom::cssSelector() and Dom::xPath()): enable providing an empty string as selector, to simply get the node that the selector is applied to.

[3.0.3] - 2024-12-11

Fixed

• Improved fix for non UTF-8 characters in HTML documents declared as UTF-8.

[3.0.2] - 2024-12-11

Fixed

• When the new PHP 8.4 DOM API is used, and HTML declared as UTF-8 contains non UTF-8 compatible characters, it does not replace them with a � character, but instead removes it. This behaviour is consistent with the data returned by Symfony DomCrawler.

[3.0.1] - 2024-12-10

Undeprecated

• Removed deprecations for all XPath functionality (Dom::xPath(), XPathQuery class and Node::queryXPath()), because it's still available with the net DOM API in PHP 8.4.

[3.0.0] - 2024-12-08

The primary change in version 3.0.0 is that the library now leverages PHP 8.4’s new DOM API when used in an environment with PHP >= 8.4. To maintain compatibility with PHP < 8.4, an abstraction layer has been implemented. This layer dynamically uses either the Symfony DomCrawler component or the new DOM API, depending on the PHP version.

Since no direct interaction with an instance of the Symfony DomCrawler library was required at the step level provided by the library, it is highly likely that you won’t need to make any changes to your code to upgrade to v3. To ensure a smooth transition, please review the points under “Changed.”

Changed

• BREAKING: The DomQuery::innerText() method (a.k.a. Dom::cssSelector('...')->innerText()) has been removed. innerText exists only in the Symfony DomCrawler component, and its usefulness is questionable. If you still require this variant of the DOM element text, please let us know or create a pull request yourself. Thank you!
• BREAKING: The DomQueryInterface was removed. As the DomQuery class offers a lot more functionality than the interface defines, the purpose of the interface was questionable. Please use the abstract DomQuery class instead. This also means that some method signatures, type hinting the interface, have changed. Look for occurrences of DomQueryInterface and replace them.
• BREAKING: The visibility of the DomQuery::filter() method was changed from public to protected. It is still needed in the DomQuery class, but outside of it, it is probably better and easier to directly use the new DOM abstraction (see the src/Steps/Dom directory). If you are extending the DomQuery class (which is not recommended), be aware that the argument now takes a Node (from the new DOM abstraction) instead of a Symfony Crawler.
• BREAKING: The Step::validateAndSanitizeToDomCrawlerInstance() method was removed. Please use the Step::validateAndSanitizeToHtmlDocumentInstance() and Step::validateAndSanitizeToXmlDocumentInstance() methods instead.
• BREAKING: The second argument in Closures passed to the Http::crawl()->customFilter() has changed from an instance of Symfony Crawler class, to an HtmlElement instance from the new DOM abstraction (Crwlr\Crawler\Steps\Dom\HtmlElement).
• BREAKING: The Filter class was split into AbstractFilter (base class for actual filter classes) and Filter only hosting the static function for easy instantiation, because otherwise each filter class also has all the static methods.
• BREAKING: Further, the signatures of some methods that are mainly here for internal usage, have changed due to the new DOM abstraction:
  • The static GetLink::isSpecialNonHttpLink() method now needs an instance of HtmlElement instead of a Symfony Crawler.
  • GetUrlsFromSitemap::fixUrlSetTag() now takes an XmlDocument instead of a Symfony Crawler.
  • The DomQuery::apply() method now takes a Node instead of a Symfony Crawler.

Deprecated

• Dom::xPath() method and
• the XPathQuery class as well as
• the new Node::queryXPath() method.

Added

• New step output filter Filter::arrayHasElement(). When a step produces array output with a property being a numeric array, you can now filter outputs by checking if one element of that array property, matches certain filter criteria. Example: The outputs look like ['foo' => 'bar', 'baz' => ['one', 'two', 'three']]. You can filter all outputs where baz contains two like: Filter::arrayHasElement()->where('baz', Filter::equal('two')).

[2.1.3] - 2024-11-05

Fixed

• Improvements for deprecations in PHP 8.4.

[2.1.2] - 2024-10-22

Fixed

• Issue when converting cookie objects received from the chrome-php library.

[2.1.1] - 2024-10-21

Fixed

• Also add cookies, set during headless browser usage, to the cookie jar. When switching back to the (guzzle) HTTP client the cookies should also be sent.
• Don't call Loader::afterLoad() when Loader::beforeLoad() was not called before. This can potentially happen, when an exception is thrown before the call to the beforeLoad hook, but it is caught and the afterLoader hook method is called anyway. As this most likely won't make sense to users, the afterLoad hook callback functions will just not be called in this case.
• The Throttler class now has protected methods _internalTrackStartFor(), _requestToUrlWasStarted() and _internalTrackEndFor(). When extending the Throttler class (be careful, actually that's not really recommended) they can be used to check if a request to a URL was actually started before.

[2.1.0] - 2024-10-19

Added

• The new postBrowserNavigateHook() method in the Http step classes, which allows to define callback functions that are triggered after the headless browser navigated to the specified URL. They are called with the chrome-php Page object as argument, so you can interact with the page. Also, there is a new class BrowserAction providing some simple actions (like wait for element, click element,...) as Closures via static methods. You can use it like Http::get()->postBrowserNavigateHook(BrowserAction::clickElement('#element')).

[2.0.1] - 2024-10-15

Fixed

• Issue with the afterLoad hook of the HttpLoader, introduced in v2. Calling the hook was commented out, which slipped through because the test case was faulty.

[2.0.0] - 2024-10-15

Changed

• BREAKING: Removed methods BaseStep::addToResult(), BaseStep::addLaterToResult(), BaseStep::addsToOrCreatesResult(), BaseStep::createsResult(), and BaseStep::keepInputData(). These methods were deprecated in v1.8.0 and should be replaced with Step::keep(), Step::keepAs(), Step::keepFromInput(), and Step::keepInputAs().
• BREAKING: Added the following keep methods to the StepInterface: StepInterface::keep(), StepInterface::keepAs(), StepInterface::keepFromInput(), StepInterface::keepInputAs(), as well as StepInterface::keepsAnything(), StepInterface::keepsAnythingFromInputData() and StepInterface::keepsAnythingFromOutputData(). If you have a class that implements this interface without extending Step (or BaseStep), you will need to implement these methods yourself. However, it is strongly recommended to extend Step instead.
• BREAKING: With the removal of the addToResult() method, the library no longer uses toArrayForAddToResult() methods on output objects. Instead, please use toArrayForResult(). Consequently, RespondedRequest::toArrayForAddToResult() has been renamed to RespondedRequest::toArrayForResult().
• BREAKING: Removed the result and addLaterToResult properties from Io objects (Input and Output). These properties were part of the addToResult feature and are now removed. Instead, use the keep property where kept data is added.
• BREAKING: The signature of the Crawler::addStep() method has changed. You can no longer provide a result key as the first parameter. Previously, this key was passed to the Step::addToResult() method internally. Now, please handle this call yourself.
• BREAKING: The return type of the Crawler::loader() method no longer allows array. This means it's no longer possible to provide multiple loaders from the crawler. Instead, use the new functionality to directly provide a custom loader to a step described below. As part of this change, the UnknownLoaderKeyException was also removed as it is now obsolete. If you have any references to this class, please make sure to remove them.
• BREAKING: Refactored the abstract LoadingStep class to a trait and removed the LoadingStepInterface. Loading steps should now extend the Step class and use the trait. As multiple loaders are no longer supported, the addLoader method was renamed to setLoader. Similarly, the methods useLoader() and usesLoader() for selecting loaders by key are removed. Now, you can directly provide a different loader to a single step using the trait's new withLoader() method (e.g., Http::get()->withLoader($loader)). The trait now also uses phpdoc template tags, for a generic loader type. You can define the loader type by putting /** @use LoadingStep<MyLoader> */ above use LoadingStep; in your step class. Then your IDE and static analysis (if supported) will know what type of loader, the trait methods return and accept.
• BREAKING: Removed the PaginatorInterface to allow for better extensibility. The old Crwlr\Crawler\Steps\Loading\Http\Paginators\AbstractPaginator class has also been removed. Please use the newer, improved version Crwlr\Crawler\Steps\Loading\Http\AbstractPaginator. This newer version has also changed: the first argument UriInterface $url is removed from the processLoaded() method, as the URL also is part of the request (Psr\Http\Message\RequestInterface) which is now the first argument. Additionally, the default implementation of the getNextRequest() method is removed. Child implementations must define this method themselves. If your custom paginator still has a getNextUrl() method, note that it is no longer needed by the library and will not be called. The getNextRequest() method now fulfills its original purpose.
• BREAKING: Removed methods from HttpLoader:
  • $loader->setHeadlessBrowserOptions() => use $loader->browser()->setOptions() instead
  • $loader->addHeadlessBrowserOptions() => use $loader->browser()->addOptions() instead
  • $loader->setChromeExecutable() => use $loader->browser()->setExecutable() instead
  • $loader->browserHelper() => use $loader->browser() instead
• BREAKING: Removed method RespondedRequest::cacheKeyFromRequest(). Use RequestKey::from() instead.
• BREAKING: The HttpLoader::retryCachedErrorResponses() method now returns an instance of the new Crwlr\Crawler\Loader\Http\Cache\RetryManager class. This class provides the methods only() and except() to restrict retries to specific HTTP response status codes. Previously, this method returned the HttpLoader itself ($this), so if you're using it in a chain and calling other loader methods after it, you will need to refactor your code.
• BREAKING: Removed the Microseconds class from this package. It has been moved to the crwlr/utils package, which you can use instead.

Added

• New methods FileCache::prolong() and FileCache::prolongAll() to allow prolonging the time to live for cached responses.

Fixed

• The maxOutputs() method is now also available and working on Group steps.
• Improved warning messages for step validations that are happening before running a crawler.
• A PreRunValidationException when the crawler finds a problem with the setup, before actually running, is not only logged as an error via the logger, but also rethrown to the user. This way the user won't get the impression, that the crawler ran successfully without looking at the log messages.

[1.10.0] - 2024-08-05

Added

• URL refiners: UrlRefiner::withScheme(), UrlRefiner::withHost(), UrlRefiner::withPort(), UrlRefiner::withoutPort(), UrlRefiner::withPath(), UrlRefiner::withQuery(), UrlRefiner::withoutQuery(), UrlRefiner::withFragment() and UrlRefiner::withoutFragment().
• New paginator stop rules PaginatorStopRules::contains() and PaginatorStopRules::notContains().
• Static method UserAgent::mozilla5CompatibleBrowser() to get a UserAgent instance with the user agent string Mozilla/5.0 (compatible) and also the new method withMozilla5CompatibleUserAgent in the AnonymousHttpCrawlerBuilder that you can use like this: HttpCrawler::make()->withMozilla5CompatibleUserAgent().

[1.9.5] - 2024-07-25

Fixed

• Prevent PHP warnings when an HTTP response includes a Content-Type: application/x-gzip header, but the content is not actually compressed. This issue also occurred with cached responses, because compressed content is decoded during caching. Upon retrieval from the cache, the header indicated compression, but the content was already decoded.

[1.9.4] - 2024-07-24

Fixed

• When using HttpLoader::cacheOnlyWhereUrl() to restrict caching, the filter rule is not only applied when adding newly loaded responses to the cache, but also for using cached responses. Example: a response for https://www.example.com/foo is already available in the cache, but $loader->cacheOnlyWhereUrl(Filter::urlPathStartsWith('/bar/')) was called, the cached response is not used.

[1.9.3] - 2024-07-05

Fixed

• Add HttpLoader::browser() as a replacement for HttpLoader::browserHelper() and deprecate the browserHelper() method. It's an alias and just because it will read a little better: $loader->browser()->xyz() vs. $loader->browserHelper()->xyz(). HttpLoader::browserHelper() will be removed in v2.0.
• Also deprecate HttpLoader::setHeadlessBrowserOptions(), HttpLoader::addHeadlessBrowserOptions() and HttpLoader::setChromeExecutable(). Use $loader->browser()->setOptions(), $loader->browser()->addOptions() and $loader->browser()->setExecutable() instead.

[1.9.2] - 2024-06-18

Fixed

• Issue with setting the headless chrome executable, introduced in 1.9.0.

[1.9.1] - 2024-06-17

Added

• Also add HeadlessBrowserLoaderHelper::getTimeout() to get the currently configured timeout value.

[1.9.0] - 2024-06-17

Added

• New methods HeadlessBrowserLoaderHelper::setTimeout() and HeadlessBrowserLoaderHelper::waitForNavigationEvent() to allow defining the timeout for the headless chrome in milliseconds (default 30000 = 30 seconds) and the navigation event (load (default), DOMContentLoaded, firstMeaningfulPaint, networkIdle, etc.) to wait for when loading a URL.

[1.8.0] - 2024-06-05

Added

• New methods Step::keep() and Step::keepAs(), as well as Step::keepFromInput() and Step::keepInputAs(), as alternatives to Step::addToResult() (or Step::addLaterToResult()). The keep() method can be called without any argument, to keep all from the output data. It can be called with a string, to keep a certain key or with an array to keep a list of keys. If the step yields scalar value outputs (not an associative array or object with keys) you need to use the keepAs() method with the key you want the output value to have in the kept data. The methods keepFromInput() and keepInputAs() work the same, but uses the input (not the output) that the step receives. Most likely only needed with a first step, to keep data from initial inputs (or in a sub crawler, see below). Kept properties can also be accessed with the Step::useInputKey() method, so you can easily reuse properties from multiple steps ago as input.
• New method Step::outputType() with default implementation returning StepOutputType::Mixed. Please consider implementing this method yourself in all your custom steps, because it is going to be required in v2 of the library. It allows detecting (potential) problems in crawling procedures immediately when starting a run instead of failing after already running a while.
• New method Step::subCrawlerFor(), allowing to fill output properties from an actual full child crawling procedure. As the first argument, you give it a key from the step's output, that the child crawler uses as input(s). As the second argument you need to provide a Closure that receives a clone of the current Crawler without steps and with initial inputs, set from the current output. In the Closure you then define the crawling procedure by adding steps as you're used to do it, and return it. This allows to achieve nested output data, scraped from different (sub-)pages, more flexible and less complicated as with the usual linear crawling procedure and Step::addToResult().

Deprecated

• The Step::addToResult(), Step::addLaterToResult() and Step::keepInputData() methods. Instead, please use the new keep methods. This can cause some migration work for v2, because especially the add to result methods are a pretty central functionality, but the new "keep" methodology (plus the new sub crawler feature) will make a lot of things easier, less complex and the library will most likely work more efficiently in v2.

Fixed

• When a cache file was generated with compression, and you're trying to read it with a FileCache instance without compression enabled, it also works. When unserializing the file content fails it tries decoding the string first before unserializing it.

[1.7.2] - 2024-03-19

Fixed

• When the useInputKey() method is used on a step and the defined key does not exist in input, it logs a warning and does not invoke the step instead of throwing an Exception.

[1.7.1] - 2024-03-11

Fixed

• A PHP error that happened when the loader returns null for the initial request in the Http::crawl() step.

[1.7.0] - 2024-03-04

Added

• Allow getting the whole decoded JSON as array with the new Json::all() and also allow to get the whole decoded JSON, when using Json::get(), inside a mapping using either empty string or * as target. Example: Json::get(['all' => '*']). * only works, when there is no key * in the decoded data.

Fixed

• Make it work with responses loaded by a headless browser. If decoding the input string fails, it now checks if it could be HTML. If that's the case, it extracts the text content of the <body> and tries to decode this instead.

[1.6.2] - 2024-02-26

Fixed

• When using HttpLoader::cacheOnlyWhereUrl() and a request was redirected (maybe even multiple times), previously all URLs in the chain had to match the filter rule. As this isn't really practicable, now only one of the URLs has to match the rule.

[1.6.1] - 2024-02-16

Changed

• Make method HttpLoader::addToCache() public, so steps can update a cached response with an extended version.

[1.6.0] - 2024-02-13

Added

• Enable dot notation in Step::addToResult(), so you can get data from nested output, like: $step->addToResult(['url' => 'response.url', 'status' => 'response.status', 'foo' => 'bar']).
• When a step adds output properties to the result, and the output contains objects, it tries to serialize those objects to arrays, by calling __serialize(). If you want an object to be serialized differently for that purpose, you can define a toArrayForAddToResult() method in that class. When that method exists, it's preferred to the __serialize() method.
• Implemented above-mentioned toArrayForAddToResult() method in the RespondedRequest class, so on every step that somehow yields a RespondedRequest object, you can use the keys url, uri, status, headers and body with the addToResult() method. Previously this only worked for Http steps, because it defines output key aliases (HttpBase::outputKeyAliases()). Now, in combination with the ability to use dot notation when adding data to the result, if your custom step returns nested output like ['response' => RespondedRequest, 'foo' => 'bar'], you can add response data to the result like this $step->addToResult(['url' => 'response.url', 'body' => 'response.body']).

Fixed

• Improvement regarding the timing when a store (Store class instance) is called by the crawler with a final crawling result. When a crawling step initiates a crawling result (so, addToResult() was called on the step instance), the crawler has to wait for all child outputs (resulting from one step-input) until it calls the store, because the child outputs can all add data to the same final result object. But previously this was not only the case for all child outputs starting from a step where addToResult() was called, but all children of one initial crawler input. So with this change, in a lot of cases, the store will earlier be called with finished Result objects and memory usage will be lowered.

[1.5.3] - 2024-02-07

Fixed

• Merge HttpBaseLoader back to HttpLoader. It's probably not a good idea to have multiple loaders. At least not multiple loaders just for HTTP. It should be enough to publicly expose the HeadlessBrowserLoaderHelper via HttpLoader::browserHelper() for the extension steps. But keep the HttpBase step, to share the general HTTP functionality implemented there.

[1.5.2] - 2024-02-07

Fixed

• Issue in GetUrlsFromSitemap (Sitemap::getUrlsFromSitemap()) step when XML content has no line breaks.

[1.5.1] - 2024-02-06

Fixed

• For being more flexible to build a separate headless browser loader (in an extension package) extract the most basic HTTP loader functionality to a new HttpBaseLoader and important functionality for the headless browser loader to a new HeadlessBrowserLoaderHelper. Further, also share functionality from the Http steps via a new abstract HttpBase step. It's considered a fix, because there's no new functionality, just refactoring existing code for better extendability.

[1.5.0] - 2024-01-29

Added

• The DomQuery class (parent of CssSelector (Dom::cssSelector) and XPathQuery (Dom::xPath)) has a new method formattedText() that uses the new crwlr/html-2-text package to convert the HTML to formatted plain text. You can also provide a customized instance of the Html2Text class to the formattedText() method.

Fixed

• The Http::crawl() step won't yield a page again if a newly found URL responds with a redirect to a previously loaded URL.

[1.4.0] - 2024-01-14

Added

• The QueryParamsPaginator can now also increase and decrease non first level query param values like foo[bar][baz]=5 using dot notation: QueryParamsPaginator::paramsInUrl()->increaseUsingDotNotation('foo.bar.baz', 5).

[1.3.5] - 2023-12-20

Fixed

• The FileCache can now also read uncompressed cache files when compression is activated.

[1.3.4] - 2023-12-19

Fixed

• Reset paginator state after finishing paginating for one base input, to enable paginating multiple listings of the same structure.

[1.3.3] - 2023-12-01

Fixed

• Add forgotten getter method to get the DOM query that is attached to an InvalidDomQueryException instance.

[1.3.2] - 2023-12-01

Fixed

• When creating a CssSelector or XPathQuery instance with invalid selector/query syntax, an InvalidDomQueryException is now immediately thrown. This change is considered to be not only non-breaking, but actually a fix, because the CssSelector would otherwise throw an exception later when the apply() method is called. The XPathQuery would silently return no result without notifying you of the invalid query and generate a PHP warning.

[1.3.1] - 2023-11-30

Fixed

• Support usage with the new Symfony major version v7.

[1.3.0] - 2023-10-28

Added

• New methods HttpLoader::useProxy() and HttpLoader::useRotatingProxies([...]) to define proxies that the loader shall use. They can be used with a guzzle HTTP client instance (default) and when the loader uses the headless Chrome browser. Using them when providing some other PSR-18 implementation will throw an exception.
• New QueryParamsPaginator to paginate by increasing and/or decreasing one or multiple query params, either in the URL or in the body of requests. Can be created via static method Crwlr\Crawler\Steps\Loading\Http\Paginator::queryParams().
• New method stopWhen in the new Crwlr\Crawler\Steps\Loading\Http\AbstractPaginator class (for more info see the deprecation below). You can pass implementations of the new StopRule interface or custom closures to that method and then, every time the Paginator receives a loaded response to process, those stop rules are called with the response. If any of the conditions of the stop rules is met, the Paginator stops paginating. Of course also added a few stop rules to use with that new method: IsEmptyInHtml, IsEmptyInJson, IsEmptyInXml and IsEmptyResponse, also available via static methods: PaginatorStopRules::isEmptyInHtml(), PaginatorStopRules::isEmptyInJson(), PaginatorStopRules::isEmptyInXml() and PaginatorStopRules::isEmptyResponse().

Deprecated

• Deprecated the Crwlr\Crawler\Steps\Loading\Http\PaginatorInterface and the Crwlr\Crawler\Steps\Loading\Http\Paginators\AbstractPaginator. Instead, added a new version of the AbstractPaginator as Crwlr\Crawler\Steps\Loading\Http\AbstractPaginator that can be used. Usually there shouldn't be a problem switching from the old to the new version. If you want to make your custom paginator implementation ready for v2 of the library, extend the new AbstractPaginator class, implement your own getNextRequest method (new requirement, with a default implementation in the abstract class, which will be removed in v2) and check if properties and methods of your existing class don't collide with the new properties and methods in the abstract class.

Fixed

• The HttpLoader::load() implementation won't throw any exception, because it shouldn't kill a crawler run. When you want any loading error to end the whole crawler execution HttpLoader::loadOrFail() should be used. Also adapted the phpdoc in the LoaderInterface.

[1.2.2] - 2023-09-19

Fixed

• Fix in HttpCrawl (Http::crawl()) step: when a page contains a broken link, that can't be resolved and throws an Exception from the URL library, ignore the link and log a warning message.
• Minor fix for merging HTTP headers when an Http step gets both, statically defined headers and headers to use from array input.

[1.2.1] - 2023-08-21

Fixed

• When a URL redirects, the trackRequestEndFor() method of the HttpLoader's Throttler instance is called only once at the end and with the original request URL.

[1.2.0] - 2023-08-18

Added

• New onCacheHit hook in the Loader class (in addition to beforeLoad, onSuccess, onError and afterLoad) that is called in the HttpLoader class when a response for a request was found in the cache.

Deprecated

• Moved the Microseconds value object class to the crwlr/utils package, as it is a very useful and universal tool. The class in this package still exists, but just extends the class from the utils package and will be removed in v2. So, if you're using this class, please change to use the version from the utils package.

[1.1.6] - 2023-07-20

Fixed

• Throttling now also works when using the headless browser.

[1.1.5] - 2023-07-14

Fixed

• The Http::crawl() step, as well as the Html::getLink() and Html::getLinks() steps now ignore links, when the href attribute starts with mailto:, tel: or javascript:. For the crawl step it obviously makes no sense, but it's also considered a bugfix for the getLink(s) steps, because they are meant to deliver absolute HTTP URLs. If you want to get the values of such links, use the HTML data extraction step.

[1.1.4] - 2023-07-14

Fixed

• The Http::crawl() step now also work with sitemaps as input URL, where the <urlset> tag contains attributes that would cause the symfony DomCrawler to not find any elements.

[1.1.3] - 2023-06-29

Fixed

• Improved Json step: if the target of the "each" (like Json::each('target', [...])) does not exist in the input JSON data, the step yields nothing and logs a warning.

[1.1.2] - 2023-05-28

Fixed

• Using the only() method of the MetaData (Html::metaData()) step class, the title property was always contained in the output, even if not listed in the only properties. This is fixed now.

[1.1.1] - 2023-05-28

Fixed

• There was an issue when adding multiple associative arrays with the same key to a Result object: let's say you're having a step producing array output like: ['bar' => 'something', 'baz' => 'something else'] and it (the whole array) shall be added to the result property foo. When the step produced multiple such array outputs, that led to a result like ['bar' => '...', 'baz' => '...', ['bar' => '...', 'baz' => '...'], ['bar' => '...', 'baz' => '...']. Now it's fixed to result in [['bar' => '...', 'baz' => '...'], ['bar' => '...', 'baz' => '...'], ['bar' => '...', 'baz' => '...'].

[1.1.0] - 2023-05-21

Added

• Http steps can now receive body and headers from input data (instead of statically defining them via argument like Http::method(headers: ...)) using the new methods useInputKeyAsBody(<key>) and useInputKeyAsHeader(<key>, <asHeader>) or useInputKeyAsHeaders(<key>). Further, when invoked with associative array input data, the step will by default use the value from url or uri for the request URL. If the input array contains the URL in a key with a different name, you can use the new useInputKeyAsUrl(<key>) method. That was basically already possible with the existing useInputKey(<key>) method, because the URL is the main input argument for the step. But if you want to use it in combination with the other new useInputKeyAsXyz() methods, you have to use useInputKeyAsUrl(), because using useInputKey(<key>) would invoke the whole step with that key only.
• Crawler::runAndDump() as a simple way to just run a crawler and dump all results, each as an array.
• addToResult() now also works with serializable objects.
• If you know certain keys that the output of a step will contain, you can now also define aliases for those keys, to be used with addToResult(). The output of an Http step (RespondedRequest) contains the keys requestUri and effectiveUri. The aliases url and uri refer to effectiveUri, so addToResult(['url']) will add the effectiveUri as url to the result object.
• The GetLink (Html::getLink()) and GetLinks (Html::getLinks()) steps, as well as the abstract DomQuery (parent of CssSelector (/Dom::cssSelector) and XPathQuery (/Dom::xPath)) now have a method withoutFragment() to get links respectively URLs without their fragment part.
• The HttpCrawl step (Http::crawl()) has a new method useCanonicalLinks(). If you call it, the step will not yield responses if its canonical link URL was already yielded. And if it discovers a link, and some document pointing to that URL via canonical link was already loaded, it treats it as if it was already loaded. Further this feature also sets the canonical link URL as the effectiveUri of the response.
• All filters can now be negated by calling the negate() method, so the evaluate() method will return the opposite bool value when called. The negate() method returns an instance of NegatedFilter that wraps the original filter.
• New method cacheOnlyWhereUrl() in the HttpLoader class, that takes an instance of the FilterInterface as argument. If you define one or multiple filters using this method, the loader will cache only responses for URLs that match all the filters.

Fixed

• The HttpCrawl step (Http::crawl()) by default now removes the fragment part of URLs to not load the same page multiple times, because in almost any case, servers won't respond with different content based on the fragment. That's why this change is considered non-breaking. For the rare cases when servers respond with different content based on the fragment, you can call the new keepUrlFragment() method of the step.
• Although the HttpCrawl step (Http::crawl()) already respected the limit of outputs defined via the maxOutputs() method, it actually didn't stop loading pages. The limit had no effect on loading, only on passing on outputs (responses) to the next step. This is fixed in this version.
• A so-called byte order mark at the beginning of a file (/string) can cause issues. So just remove it, when a step's input string starts with a UTF-8 BOM.
• There seems to be an issue in guzzle when it gets a PSR-7 request object with a header with multiple string values (as array, like: ['accept-encoding' => ['gzip', 'deflate', 'br']]). When testing it happened that it only sent the last part (in this case br). Therefore, the HttpLoader now prepares headers before sending (in this case to: ['accept-encoding' => ['gzip, deflate, br']]).
• You can now also use the output key aliases when filtering step outputs. You can even use keys that are only present in the serialized version of an output object.

[1.0.2] - 2023-03-20

Fixed

• JSON step: another fix for JSON strings having keys without quotes with empty string value.

[1.0.1] - 2023-03-17

Fixed

• JSON step: improve attempt to fix JSON string having keys without quotes.

[1.0.0] - 2023-02-08

Added

• New method Step::refineOutput() to manually refine step output values. It takes either a Closure or an instance of the new RefinerInterface as argument. If the step produces array output, you can provide a key from the array output, to refine, as first argument and the refiner as second argument. You can call the method multiple times and all the refiners will be applied to the outputs in the order you add them. If you want to refine multiple output array keys with a Closure, you can skip providing a key and the Closure will receive the full output array for refinement. As mentioned you can provide an instance of the RefinerInterface. There are already a few implementations: StringRefiner::afterFirst(), StringRefiner::afterLast(), StringRefiner::beforeFirst(), StringRefiner::beforeLast(), StringRefiner::betweenFirst(), StringRefiner::betweenLast() and StringRefiner::replace().
• New method Step::excludeFromGroupOutput() to exclude a normal steps output from the combined output of a group that it's part of.
• New method HttpLoader::setMaxRedirects() to customize the limit of redirects to follow. Works only when using the HTTP client.
• New filters to filter by string length, with the same options as the comparison filters (equal, not equal, greater than,...).
• New Filter::custom() that you can use with a Closure, so you're not limited to the available filters only.
• New method DomQuery::link() as a shortcut for DomQuery::attribute('href')->toAbsoluteUrl().
• New static method HttpCrawler::make() returning an instance of the new class AnonymousHttpCrawlerBuilder. This makes it possible to create your own Crawler instance with a one-liner like: HttpCrawler::make()->withBotUserAgent('MyCrawler'). There's also a withUserAgent() method to create an instance with a normal (non bot) user agent.

Changed

• BREAKING: The FileCache now also respects the ttl (time to live) argument and by default it is one hour (3600 seconds). If you're using the cache and expect the items to live (basically) forever, please provide a high enough value for default the time to live. When you try to get a cache item that is already expired, it (the file) is immediately deleted.
• BREAKING: The TooManyRequestsHandler (and with that also the constructor argument in the HttpLoader) was renamed to RetryErrorResponseHandler. It now reacts the same to 503 (Service Unavailable) responses as to the 429 (Too Many Requests) responses. If you're actively passing your own instance to the HttpLoader, you need to update it.
• You can now have multiple different loaders in a Crawler. To use this, return an array containing your loaders from the protected Crawler::loader() method with keys to name them. You can then selectively use them by calling the Step::useLoader() method on a loading step with the key of the loader it should use.

Removed

• BREAKING: The loop feature. The only real world use case should be paginating listings and this should be solved with the Paginator feature.
• BREAKING: Step::dontCascade() and Step::cascades() because with the change in v0.7, that groups can only produce combined output, there should be no use case for this anymore. If you want to exclude one steps output from the combined group output, you can use the new Step::excludeFromGroupOutput() method.

[0.7.0] - 2023-01-13

Added

• New functionality to paginate: There is the new Paginate child class of the Http step class (easy access via Http::get()->paginate()). It takes an instance of the PaginatorInterface and uses it to iterate through pagination links. There is one implementation of that interface, the SimpleWebsitePaginator. The Http::get()->paginate() method uses it by default, when called just with a CSS selector to get pagination links. Paginators receive all loaded pages and implement the logic to find pagination links. The paginator class is also called before sending a request, with the request object that is about to be sent as an argument (prepareRequest()). This way, it should even be doable to implement more complex pagination functionality. For example when pagination is built using POST request with query strings in the request body.
• New methods stopOnErrorResponse() and yieldErrorResponses() that can be used with Http steps. By calling stopOnErrorResponse() the step will throw a LoadingException when a response has a 4xx or 5xx status code. By calling the yieldErrorResponse() even error responses will be yielded and passed on to the next steps (this was default behaviour until this version. See the breaking change below).
• The body of HTTP responses with a Content-Type header containing application/x-gzip are automatically decoded when Http::getBodyString() is used. Therefore, added ext-zlib to suggested in composer.json.
• New methods addToResult() and addLaterToResult(). addToResult() is a single replacement for setResultKey() and addKeysToResult() (they are removed, see Changed below) that can be used for array and non array output. addLaterToResult() is a new method that does not create a Result object immediately, but instead adds the output of the current step to all the Results that will later be created originating from the current output.
• New methods outputKey() and keepInputData() that can be used with any step. Using the outputKey() method, the step will convert non array output to an array and use the key provided as an argument to this method as array key for the output value. The keepInputData() method allows you to forward data from the step's input to the output. If the input is non array you can define a key using the method's argument. This is useful e.g. if you're having data in the initial inputs that you also want to add to the final crawling results.
• New method createsResult() that can be used with any step, so you can differentiate if a step creates a Result object, or just keeps data to add to results later (new addLaterToResult() method). But primarily relevant for library internal use.
• The FileCache class can compress the cache data now to save disk space. Use the useCompression() method to do so.
• New method retryCachedErrorResponses() in HttpLoader. When called, the loader will only use successful responses (status code < 400) from the cache and therefore retry already cached error responses.
• New method writeOnlyCache() in HttpLoader to only write to, but don't read from the response cache. Can be used to renew cached responses.
• Filter::urlPathMatches() to filter URL paths using a regex.
• Option to provide a chrome executable name to the chrome-php/chrome library via HttpLoader::setChromeExecutable().

Changed

• BREAKING: Group steps can now only produce combined outputs, as previously done when combineToSingleOutput() method was called. The method is removed.
• BREAKING: setResultKey() and addKeysToResult() are removed. Calls to those methods can both be replaced with calls to the new addToResult() method.
• BREAKING: getResultKey() is also removed with setResultKey(). It's removed without replacement, as it doesn't really make sense any longer.
• BREAKING: Error responses (4xx as well as 5xx), by default, won't produce any step outputs any longer. If you want to receive error responses, use the new yieldErrorResponses() method.
• BREAKING: Removed the httpClient() method in the HttpCrawler class. If you want to provide your own HTTP client, implement a custom loader method passing your client to the HttpLoader instead.
• Deprecated the loop feature (class Loop and Crawler::loop() method). Probably the only use case is iterating over paginated list pages, which can be done using the new Paginator functionality. It will be removed in v1.0.
• In case of a 429 (Too Many Requests) response, the HttpLoader now automatically waits and retries. By default, it retries twice and waits 10 seconds for the first retry and a minute for the second one. In case the response also contains a Retry-After header with a value in seconds, it complies to that. Exception: by default it waits at max 60 seconds (you can set your own limit if you want), if the Retry-After value is higher, it will stop crawling. If all the retries also receive a 429 it also throws an Exception.
• Removed logger from Throttler as it doesn't log anything.
• Fail silently when robots.txt can't be parsed.
• Default timeout configuration for the default guzzle HTTP client: connect_timeout is 10 seconds and timeout is 60 seconds.
• The validateAndSanitize...() methods in the abstract Step class, when called with an array with one single element, automatically try to use that array element as input value.
• With the Html and Xml data extraction steps you can now add layers to the data that is being extracted, by just adding further Html/Xml data extraction steps as values in the mapping array that you pass as argument to the extract() method.
• The base Http step can now also be called with an array of URLs as a single input. Crawl and Paginate steps still require a single URL input.

Fixed

• The CookieJar now also works with localhost or other hosts without a registered domain name.
• Improve the Sitemap::getUrlsFromSitemap() step to also work when the <urlset> tag contains attributes that would cause the symfony DomCrawler to not find any elements.
• Fixed possibility of infinite redirects in HttpLoader by adding a redirects limit of 10.

[0.6.0] - 2022-10-03

Added

• New step Http::crawl() (class HttpCrawl extending the normal Http step class) for conventional crawling. It loads all pages of a website (same host or domain) by following links. There's also a lot of options like depth, filtering by paths, and so on.
• New steps Sitemap::getSitemapsFromRobotsTxt() (GetSitemapsFromRobotsTxt) and Sitemap::getUrlsFromSitemap() (GetUrlsFromSitemap) to get sitemap (URLs) from a robots.txt file and to get all the URLs from those sitemaps.
• New step Html::metaData() to get data from meta tags (and title tag) in HTML documents.
• New step Html::schemaOrg() (SchemaOrg) to get schema.org structured data in JSON-LD format from HTML documents.
• The abstract DomQuery class (parent of the CssSelector and XPathQuery classes) now has some methods to narrow the selected matches further: first(), last(), nth(n), even(), odd().

Changed

• BREAKING: Removed PoliteHttpLoader and traits WaitPolitely and CheckRobotsTxt. Converted the traits to classes Throttler and RobotsTxtHandler which are dependencies of the HttpLoader. The HttpLoader internally gets default instances of those classes. The RobotsTxtHandler will respect robots.txt rules by default if you use a BotUserAgent and it won't if you use a normal UserAgent. You can access the loader's RobotsTxtHandler via HttpLoader::robotsTxt(). You can pass your own instance of the Throttler to the loader and also access it via HttpLoader::throttle() to change settings.

Fixed

• Getting absolute links via the GetLink and GetLinks steps and the toAbsoluteUrl() method of the CssSelector and XPathQuery classes, now also look for <base> tags in HTML when resolving the URLs.
• The SimpleCsvFileStore can now also save results with nested data (but only second level). It just concatenates the values separated with a |.

[0.5.0] - 2022-09-03

Added

• You can now call the new useHeadlessBrowser method on the HttpLoader class to use a headless Chrome browser to load pages. This is enough to get HTML after executing javascript in the browser. For more sophisticated tasks a separate Loader and/or Steps should better be created.
• With the maxOutputs() method of the abstract Step class you can now limit how many outputs a certain step should yield at max. That's for example helpful during development, when you want to run the crawler only with a small subset of the data/requests it will actually have to process when you eventually remove the limits. When a step has reached its limit, it won't even call the invoke() method any longer until the step is reset after a run.
• With the new outputHook() method of the abstract Crawler class you can set a closure that'll receive all the outputs from all the steps. Should be only for debugging reasons.
• The extract() method of the Html and Xml (children of Dom) steps now also works with a single selector instead of an array with a mapping. Sometimes you'll want to just get a simple string output e.g. for a next step, instead of an array with mapped extracted data.
• In addition to uniqueOutputs() there is now also uniqueInputs(). It works exactly the same as uniqueOutputs(), filtering duplicate input values instead. Optionally also by a key when expected input is an array or an object.
• In order to be able to also get absolute links when using the extract() method of Dom steps, the abstract DomQuery class now has a method toAbsoluteUrl(). The Dom step will automatically provide the DomQuery instance with the base url, presumed that the input was an instance of the RespondedRequest class and resolve the selected value against that base url.

Changed

• Remove some not so important log messages.
• Improve behavior of group step's combineToSingleOutput(). When steps yield multiple outputs, don't combine all yielded outputs to one. Instead, combine the first output from the first step with the first output from the second step, and so on.
• When results are not explicitly composed, but the outputs of the last step are arrays with string keys, it sets those keys on the Result object instead of setting a key unnamed with the whole array as value.

Fixed

• The static methods Html::getLink() and Html::getLinks() now also work without argument, like the GetLink and GetLinks classes.
• When a DomQuery (CSS selector or XPath query) doesn't match anything, its apply() method now returns null (instead of an empty string). When the Html(/Xml)::extract() method is used with a single, not matching selector/query, nothing is yielded. When it's used with an array with a mapping, it yields an array with null values. If the selector for one of the methods Html(/Xml)::each(), Html(/Xml)::first() or Html(/Xml)::last() doesn't match anything, that's not causing an error any longer, it just won't yield anything.
• Removed the (unnecessary) second argument from the Loop::withInput() method because when keepLoopingWithoutOutput() is called and withInput() is called after that call, it resets the behavior.
• Issue when date format for expires date in cookie doesn't have dashes in d-M-Y (so d M Y).

[0.4.1] - 2022-05-10

Fixed

• The Json step now also works with Http responses as input.

[0.4.0] - 2022-05-06

Added

• The BaseStep class now has where() and orWhere() methods to filter step outputs. You can set multiple filters that will be applied to all outputs. When setting a filter using orWhere it's linked to the previously added Filter with "OR". Outputs not matching one of the filters, are not yielded. The available filters can be accessed through static methods on the new Filter class. Currently available filters are comparison filters (equal, greater/less than,...), a few string filters (contains, starts/ends with) and url filters (scheme, domain, host,...).
• The GetLink and GetLinks steps now have methods onSameDomain(), notOnSameDomain(), onDomain(), onSameHost(), notOnSameHost(), onHost() to restrict the which links to find.
• Automatically add the crawler's logger to the Store so you can also log messages from there. This can be breaking as the StoreInterface now also requires the addLogger method. The new abstract Store class already implements it, so you can just extend it.

Changed

• The Csv step can now also be used without defining a column mapping. In that case it will use the values from the first line (so this makes sense when there are column headlines) as output array keys.

[0.3.0] - 2022-04-27

Added

• By calling monitorMemoryUsage() you can tell the Crawler to add log messages with the current memory usage after every step invocation. You can also set a limit in bytes when to start monitoring and below the limit it won't log memory usage.

Fixed

• Previously the use of Generators actually didn't make a lot of sense, because the outputs of one step were only iterated and passed on to the next step, after the current step was invoked with all its inputs. That makes steps with a lot of inputs bottlenecks and causes bigger memory consumption. So, changed the crawler to immediately pass on outputs of one step to the next step if there is one.

[0.2.0] - 2022-04-25

Added

• uniqueOutputs() method to Steps to get only unique output values. If outputs are array or object, you can provide a key that will be used as identifier to check for uniqueness. Otherwise, the arrays or objects will be serialized for comparison which will probably be slower.
• runAndTraverse() method to Crawler, so you don't need to manually traverse the Generator, if you don't need the results where you're calling the crawler.
• Implement the behaviour for when a Group step should add something to the Result using setResultKey() or addKeysToResult(), which was still missing. For groups this will only work when using combineToSingleOutput.