Dusk: [Feature Request] Headless Chrome

[ejunker]

Chrome 59 (still beta/canary) supports a headless mode.

Before this will work with Dusk, Chromedriver will have to add support for headless chrome. The latest version of Chromedriver is currently 2.29 which Supports Chrome v56-58 which does not include headless Chrome.

I tried to have Dusk run Chrome in headless mode with this code in my DuskTestCase.php

    protected function driver()
    {
        $desiredCapabilities = DesiredCapabilities::chrome();

        $options = new ChromeOptions();
        $options->setBinary('/Applications/Google Chrome Canary.app/Contents/MacOS/Google Chrome Canary');
        $options->addArguments([
            '--headless',
            // avoid an error from a missing Mesa library
            '--disable-gpu',
        ]);

        $desiredCapabilities->setCapability(ChromeOptions::CAPABILITY, $options);

        return RemoteWebDriver::create(
            'http://localhost:9515', $desiredCapabilities
        );
    }

I get this error:

1) Tests\Browser\ExampleTest::testBasicExample
Facebook\WebDriver\Exception\UnknownServerException: unknown error: unrecognized Chrome version: HeadlessChrome/60.0.3073.0
  (Driver info: chromedriver=2.28.455517 (2c6d2707d8ea850c862f04ac066724273981e88f),platform=Mac OS X 10.12.3 x86_64)

I realize that we will have to wait for Chromedriver to add support for headless Chrome but I thought it would be useful to create this issue to have a place to discuss progress.

UPDATE I updated to Chromedriver 2.29 and I no longer get the "unrecognized Chrome version" error but now I run php artisan dusk and it just seems to hang for a bit and then eventually errors out. The hanging appears to be when facebook/webdriver sends a curl request to Chromedriver and it times out waiting for a response. For example, it tries to curl something like http://localhost:9515/session/26935cbeaf02f3d76abdbd43ca28432d/url and times out.

[deleugpn]

If you're going to use it like that, I guess you're right and Chrome Driver needs to support it.

I might have understood it wrong, but I think with headless mode we would be able to talk to the browser directly instead of going through the Chrome Driver, which would be quite a rework to stop relying on Facebook Webdriver and figure out a way for Dusk to communicate with the browser. The upside should be an undeniably boost in performance with the loss of _work delegation_ to Facebook WebDriver. Dusk would have to define a layer for browsers and implement the contract on a HeadlessChrome class, as well as PhantomJS and so on.

[shehi]

@deleugpn , dusk already has phantomjs() capability, doesn't it?!

[deleugpn]

@shehi yes, but if I'm not mistaken, it still relies on Facebook Webdriver for that. My interpretation for the new --headless from Chrome is that we would bypass Facebook WebDriver and talk straight to Chrome. And for that, we would have to rework the way dusk's communication with browsers work.

[OndraM]

@deleugpn Just let me set the record straight - the headless Chrome will still communicate through WebDriver protocol, there is no difference from full-featured Chrome.

The Chromedriver is the way of communicating and remotely controlling Chrome. There is AFAIK nothing like "talk to the browser directly" - the chromedriver is the way of communicating with the browser directly...

The "indirect" way is though Selenium server - but the Selenium server also implements the WebDriver protocol (and passes tha calls to the respective browser driver - like chromedriver), so it is still the same way of communication.

[jonnybarnes]

Chrome 59 just got released on the stable channel. Is there any progress on using it with Dusk?

[deleugpn]

@jonnybarnes I'm waiting on the Chromedriver update to work on it. The support for Chrome 59 was added a little over a month ago into the Chromedriver, but no release to actually use it so far.

[ejunker]

@deleugpn I think Chromedriver may support it. I asked the question on the mailing list: https://groups.google.com/d/topic/chromedriver-users/T5G20PugKVU/discussion

[deleugpn]

@ejunker I'm lousy at C++, but https://github.com/bayandin/chromedriver/commit/7cca62f731168102358b057fedb7fbb28ac21356 came after https://github.com/bayandin/chromedriver/commit/d2b414fd71281a64a1c02a92b65344c2fdf57b63, so I don't really know how that would be possible.

[kolev]

Chrome driver 2.30 is out but still opens the browser on mac.

protected static function buildChromeProcess()
{
    return (new ProcessBuilder())
            ->setPrefix(realpath(__DIR__.'/../bin/chromedriver-'.static::driverSuffix()))
            ->setArguments(['--headless', '--disable-gpu'])
            ->getProcess()
            ->setEnv(static::chromeEnvironment());
}

[jonnybarnes]

This appears to be being worked on: https://twitter.com/laravelphp/status/874256617105129473

Dusk 2.0, which will release with Laravel 5.5, utilizes Chrome 59's new "headless" mode! Easily disable if you want to see the browser. 🔥

[hackel]

This is working great for me with chromedriver 2.30 and Chromium 59. Note that arguments need to be specified without the double hyphens.

@kolev, you're trying to pass the arguments to chromedriver, not Chromium itself.

First, I installed the lbaey/chromedriver package to automatically download the latest version of Chromedriver to vendor/bin.
Here's what I'm using in my TestCase (actually a Behat FeatureContext) to set it all up:

    protected static function buildChromeProcess()
    {
        return (new ProcessBuilder())
            ->setPrefix(base_path('vendor/bin/chromedriver'))
            ->getProcess()
            ->setEnv(static::chromeEnvironment());
    }

    protected function driver()
    {
        $capabilities  = DesiredCapabilities::chrome();
        $chromeOptions = (new ChromeOptions)->addArguments(['headless', 'disable-gpu']);
        $capabilities->setCapability(ChromeOptions::CAPABILITY, $chromeOptions);

        return RemoteWebDriver::create('http://localhost:9515', $capabilities);
    }

[oppiansteve]

It seems I had to replace the 'setPrefix(...)' with setPrefix(implode(DIRECTORY_SEPARATOR, [__DIR__, '..', 'vendor', 'bin', 'chromedriver'])) as the app container doesn't have basePath().

I am hacking this in a clean dev version of Laravel 5.5 though, just for a standalone play.

[sandervanhooft]

Any updates on this? Does it work by default in Laravel 5.5 (stable)?

[driesvints]

Think this can be closed now since it's implemented in the latest Dusk release.