Seems that 'moveto' is no longer supported - how do you test hover actions?

[Saeven]

In the process of upgrading our Behat rig, we landed on this configuration and the latest ChromeDriver and selenium/standalone-chrome image.

    "behat/behat": "v3.12.0",
    "behat/mink": "v1.10.0",
    "behat/mink-extension": "2.3.1",
    "behat/mink-selenium2-driver": "1.6.0",
    "php-webdriver/webdriver": "1.13.1",

We noticed that this feature context no longer works:

 /**
     * @When /^I hover over the element "([^"]*)"$/
     */
    public function iHoverOverTheElement($locator)
    {
        $session = $this->getSession();
        $element = $session->getPage()->find('css', $locator);

        if (null === $element) {
            throw new \InvalidArgumentException(sprintf('No element matching "%s" could be found', $locator));
        }

        $element->mouseOver();
    }

The mouseOver action in Mink triggers $this->getDriver()->mouseOver($this->getXpath()); which in turn results in logs such as:

22:31:45.759 WARN [SeleniumSpanExporter$1.lambda$export$1] - Unable to execute request for an existing session: POST /session/e7c2bdfafc03e7afa763268c81d9c4a1/moveto

Seems moveto is gone? How do we go about testing hovers in this new spec?

Thanks for your guidance!

[andrewnicols]

HI @Saeven,

What version of Behat + Mink were you upgrading from?

We're not suing the official behat/mink-selenium2-driver, we use a driver we worked on with @oleg-andreyev: https://github.com/oleg-andreyev/MinkPhpWebDriver/

This is working fine for us with Behat 3.11.

Oleg's underlying mouseOver function uses the webDriver Action system and does the following:

        $element = $this->findElement($xpath);
        $this->webDriver->action()->moveToElement($element)->perform();

https://github.com/oleg-andreyev/MinkPhpWebDriver/blob/master/src/WebDriver.php#L899-L903

We'e been using this driver with Moodle for the past 2-3 years without any incident - we did spend a lot of time working together to get everything working with minimal backwards compatability issues.

The official Behat Selenium2 driver is very much abandoned. No changes in eight years, and the official maintainer seems to have largely gone AWOL.

In any case, the solution here is to use the webDriver action factory above.

[Saeven]

Thanks for the response @andrewnicols. We're upgrading from this configuration:

selenium/standalone-chrome-debug:3.141.59-oxygen using this configuration:

    "behat/behat": "3.6.1",
    "behat/mink": "1.8.1",
    "behat/mink-extension": "2.3.1",
    "behat/mink-selenium2-driver": "1.2.0",
    "instaclick/php-webdriver": "1.4.7",

This config works, but uses a much older version of Chrome. Upgrading the selenium image forced our hand on the library upgrades, and that's when hover complete broke. I've found a lot of gripes online about moveto being removed somewhere along the chain within Selenium; but no solutions. I'll give the driver you've linked a try!

[Saeven]

Yeah unfortunately even with the two libs swapped in and the extension added to behat.yml, the problem persists:

    "oleg-andreyev/mink-phpwebdriver": "v1.2.1",
    "oleg-andreyev/mink-phpwebdriver-extension": "v1.0.0",

Seems hover got nuked from selenium at some point. I'm going to try rolling selenium back until I find the version where they killed it. Not being able to test hover is a bit of a show-stopper.

Unable to execute request for an existing session: POST \u002fsession\u002fdc6e223e5ebc84a10d8b157250c5ca20\u002fmoveto

[Saeven]

As added information, pertaining to version of selenium/standalone-chrome

Selenium Version	ChromeDriver Version	Status
4.2	103.0.5060.53	works
4.3	103.0.5060.134	breaks

4.4 and on break as well. Looks like something was deprecated, that this entire ecosystem of drivers depends on somewhere in between these versions...

[andrewnicols]

Yes 4.3 deprecated legacy webdriver recently - a few months ago.

But the issue is in your driver in this case.

It's strange because the Oleg driver does not call moveto at all, which suggests that your config is wrong.

Note: You will have to update your behat.yml to modify the Driver extension details. Here's an excerpt of mine:

chrome:
  extensions:
    Behat\MinkExtension:
      webdriver:
        browser: chrome
        wd_host: 'http://localhost:4444/wd/hub'
        capabilities:
          extra_capabilities:
            chromeOptions:
              args:
                - disable-web-security
                - unlimited-storage

[Saeven]

@andrewnicols would you be open to sharing your entire behat.yml so that I can see how the pieces fit? 🙏🏼. The one I'm migrating from looks like so:

# behat.yml
default:
  autoload: [ '%paths.base%/../contexts' ]
  suites:
    core_features:
      paths: [ '%paths.base%/../features' ]
      contexts: [ FeatureContext, CirclicalBehatFixtures\Behat\DatabaseContext ]
  extensions:
    Lemonade\Service\Behat\TimeLoggerExtension:
      enabled_always: true
    Behat\MinkExtension:
      base_url: http://web:80
      files_path: '%paths.base%/../../../public/assets/'
      javascript_session: selenium2
      browser_name: chrome
      selenium2:
        browser: chrome
        capabilities:
          browserName: "chrome"
          browser: "chrome"
          marionette: false

To clarify, here are all my behat-y things in composer.json

    "behat/behat": "v3.12.0",
    "behat/mink": "v1.10.0",
    "behat/mink-extension": "2.3.1",
    "oleg-andreyev/mink-phpwebdriver": "v1.2.1",
    "oleg-andreyev/mink-phpwebdriver-extension": "v1.0.0",

Greatly appreciated.

[Saeven]

@andrewnicols as added detail, the reason I ask, is that if I change my behat.yml as was recommended (combined with the test at this URL) to create something like this:

default:
  autoload: [ '%paths.base%/../contexts' ]
  suites:
    core_features:
      paths: [ '%paths.base%/../features' ]
      contexts: [ FeatureContext, CirclicalBehatFixtures\Behat\DatabaseContext ]
  extensions:
    OAndreyev\MinkPhpWebdriverExtension: ~
    Behat\MinkExtension:
      default_session: webdriver
      webdriver:
        wd_host: "http://web:80"
        browser: "chrome"
        capabilities:
          extra_capabilities:
            chromeOptions:
              args:
                - disable-web-security
                - unlimited-storage

Then I am left with a multitude of these errors:

    BeforeScenario # FeatureContext::beforeScenario()
      Could not open connection: Curl error thrown for http POST to /session with params: {"capabilities":{"firstMatch":[{"browserName":"chrome","goog:chromeOptions":{"args":["disable-web-security","unlimited-storage"]}}]},"desiredCapabilities":{"browserName":"chrome","platform":"ANY","tags":["SaevenPro.local","PHP 7.4.33"],"chromeOptions":{"args":["disable-web-security","unlimited-storage"]},"browser":"firefox","marionette":null,"ignoreZoomSetting":false,"name":"Behat feature suite"}}
      
      Could not resolve host: web (Behat\Mink\Exception\DriverException)

It conversely 'works' with the original configuration, despite moveto being a problem.

[andrewnicols]

Here's my behat.yml: https://gist.github.com/101d2b3bc62d47414bc8267d367e850e

[andrewnicols]

Have you got a project link I can try on, or is it not OSS?

[Saeven]

@andrewnicols here is a reproduction:
https://github.com/Saeven/behat-moveto-test

[andrewnicols]

Hi,

I'm on leave today, but I'll try and find some time this evening or tomorrow to take a look.

[andrewnicols]

Here you go: andrewnicols/behat-moveto-test@2a59d52

The MinkSelenium2 dependency is basically abandoned. When Oleg and I tried to update it we found that the maintainer was a combination of non-responsive, and not keen on the proposed changes. As a result I believe the driver has stalled.

You'll find that much of the Mink and Behat codebase is very stale, that's why FriendsOfBehat exists and attempts to keep things working - their MinkExtension has fixes for PHP 8.0 and up for example which were still missing in the standard Mink extension last I looked.

Oleg's WebDriver has a couple of key difference which may break some of your tests. From memory the main one is that it drops use of syn.js in favour of using WebDriver actions. That doesn't create a change in behaviour as such, but the behaviour of the syn library was really poor - it was synthesising browser events incorrectly. Moving to use the WebDriver actions means that you get the actual Mouse events that the real browser gives rather than some synthesised rubbish that someone once observed with one browser ten years ago.

Other than that I seem to recall that the other big changes were in the element setValue function - the 'official' driver, and the instaclick variant, use an API which calls a clear function, which performs additional change events which a normal user would not. That means that some user interfaces behave incorrectly with the Mink/Instaclick drivers. The other fix in Oleg's driver over the official is in working with Date and colour inputs. Those have broken behaviours in some driver implementations (chromedriver specifically). See https://bugs.chromium.org/p/chromedriver/issues/detail?id=3274&q=date&can=2 for information.
The official driver doesn't work around that issue, which means that if you're not in the US and want to use a DateTime input, it will behave differently between Firefox and Chrome. Oleg's driver uses JS to set the value properly.

Hope this helps

[andrewnicols]

Any success @Saeven

[Saeven]

Hey @andrewnicols!

First, thanks so much for the time you've taken to write that message above. I haven't had a chance to respond; I have been addled with SOC2 prep and a myriad of other small fires. You know the drill!

The composer configuration you initially proposed in a fork of the sample project I had rigged was a breakthrough. It enabled me to run things on M1 macs and update the docker-compose configuration to use newer images.

My first stop after this was to adjust the CI chain (which runs on AWS Ubuntu AMIs). We have some 7000 tests running on our principal app. The change precipitated a few minor failures here and there; this was expected. What wasn't expected, however, was that about halfway through each test run, tests would begin to fail with "failures to POST" random commands. It'd grind in molasses and die.

It seemed time-bound...

First test was to scale vertically; I moved the agents from m3.medium to t2.large and finally *.2xlarge to see if it was a resource problem. Even after throwing more horsepower at it - the same problem... I tried swapping versions of the selenium images; same issue. So, I fired up blackfire to see if it was a memory leak in a lib (again, hunting the time-bound problem), and it was quite stable.

The solution in the end, was a composer change. Swapping out the friendsofphp driver to the mink driver made all of these issues vanish. I've not had the time to investigate why - have to solve the broken tests first.

Finally - I'm left with a few "differences" to solve in how the tests are working; some are prompting significant test rewrites. All in all, a great success. Of those 7000+ tests, I'm currently at 4 failing scenarios left. I'll take it!

Thanks again!

Final composer component that works:

"behat/behat": "v3.12.0",
    "behat/mink": "v1.10.0",
    "behat/mink-extension": "2.3.1",
    "oleg-andreyev/mink-phpwebdriver-extension": "v1.0.0",
    "oleg-andreyev/mink-phpwebdriver": "v1.2.1",
    "php-webdriver/webdriver": "1.13.1",
    ```