1.8.0/1.8.1 releases break `dmore/chrome-mink-driver` about mink

[stof]

Can you provide the reproducing case (your usage of Mink that was working before and does not work anymore) ?

[mpdude]

<?php

use Behat\Behat\Hook\Scope\BeforeScenarioScope;
use Behat\MinkExtension\Context\RawMinkContext;

class MyAppContext extends RawMinkContext ...
{
    /**
     * @BeforeScenario
     */
    public function setUpBeforeScenario(BeforeScenarioScope $scope)
    {
        // Workaround for https://github.com/minkphp/Mink/issues/798:
        $this->getSession()->start();

        // $this->getSession()->setCookie() fails as of minkphp/Mink ^1.8
        if (\function_exists('xdebug_is_debugger_active') && xdebug_is_debugger_active()) {
            $this->getSession()->setCookie('XDEBUG_SESSION', 'behat');
        }
    }

    ....

[lpeabody]

For additional context, see https://travis-ci.com/github/lpeabody/Metis/builds/153721427#L1377 for an example of where this breaks. $this->page evaluates to null here https://gitlab.com/DMore/chrome-mink-driver/-/blob/2.7.0/src/ChromeDriver.php#L1058.

It's broken out more verbosely here https://gitlab.com/DMore/chrome-mink-driver/-/issues/94.

[aik099]

@mpdude , actually I see no issue here.

In Selenium (which emulates user actions) you can't set a cookie on a page prior to that page being visited. Same goes for dmore/chrome-mink-driver driver, which does the same as Selenium, but without Selenium in Google Chrome.

Setting headers (e.g. basic authentication, cookies, etc.) prior to page visiting is only possible in headless drivers, like MinkBrowserKitDriver and MinkZombieDriver.

The solution you came up with is correct way to do this. I'm not sure why it even worked, because you've set cookie right after browser window was opened, but prior to page being visited. In that case there is no way to determine at which URL cookie should apply.

[mpdude]

But I should be able to set a request header before requesting a page, right?

[aik099]

But I should be able to set a request header before requesting a page, right?

@mpdude , I guess no. If you can't do this with Selenium you can't do it in dmore/chrome-mink-driver driver, that controls actual web browser.

[mpdude]

I am not sure if we're talking about the same thing: The driver needs the start() method to be called so it connects to the browser. But once that's done, you can add the cookies and/or headers, even before visiting the page.

Am I missing something?

[aik099]

If that is how it works for that driver then it's good thing. It doesn't work that way with other drivers.

Maybe driver code needs to be updated to that it auto-starts session, when it's missing and somebody is attempting to set cookies.

[mpdude]

@aik099 I cross-posted this as an issue here and over at the driver's repo at https://gitlab.com/DMore/chrome-mink-driver/-/issues/94 because I know too little about the contracts between Mink and the drivers and I was unsure if this needs to be fixed in the driver or here.

Mink/src/Driver/DriverInterface.php

Lines 32 to 51 in 07c6a9f

	/**
	* Starts driver.
	*
	* Once started, the driver should be ready to visit a page.
	*
	* Calling any action before visiting a page is an undefined behavior.
	* The only supported method calls on a fresh driver are
	* - visit()
	* - setRequestHeader()
	* - setBasicAuth()
	* - reset()
	* - stop()
	*
	* Calling start on a started driver is an undefined behavior. Driver
	* implementations are free to handle it silently or to fail with an
	* exception.
	*
	* @throws DriverException When the driver cannot be started
	*/
	public function start();

Calling any action before visiting a page is an undefined behavior.
The only supported method calls on a fresh driver are [...]

• setRequestHeader()

If "fresh" means "not started", then probably the driver should not rely on start() being called until the very last moment before visitPage()?

[adorin]

@mpdude , I guess no. If you can't do this with Selenium you can't do it in dmore/chrome-mink-driver driver, that controls actual web browser.

Selenium needs the browser to navigate to the cookie domain before being able to set a cookie. The chrome driver can technically set the cookie to base_url before visiting it, as long as ->start() is called before setCookie.

But since calling setCookie() before visit() is undefined behaviour, that doesn't have much practical value.

Maybe driver code needs to be updated to that it auto-starts session, when it's missing and somebody is attempting to set cookies.

It's possible, but is it a good idea? What if mink called stop() specifically and then we just overrule that with autostart. We'd essentially be taking control over start/stop/reset.

[aik099]

Closing as @DorianCMore agreed to fix it on driver side in https://gitlab.com/DMore/chrome-mink-driver/-/issues/94.

[lpeabody]

I'm failing to see where he has explicitly agreed to fix this in chrome-mink-driver?

[aik099]

@lpeabody , here is a screenshot with relevant text selected from GitLab:

[xurizaemon]

There's an MR for this at https://gitlab.com/DMore/chrome-mink-driver/-/merge_requests/93 now.

Input / feedback welcome - @aik099 I'd especially welcome your thoughts on the solution, I'm pretty fresh to this!

[mpdude]

https://gitlab.com/DMore/chrome-mink-driver/-/merge_requests/93 does not fix the issue:

It fixes that you can set HTTP Basic Auth and/or request headers before calling visit(), but it does not fix calls to setCookie().

You still can call the start() method on the driver yourself and afterwards call setCookie(), even before calling visit().

I find it hard to understand why setCookie() should not be possible when setRequestHeader() is. But that might be because setCookie() is expected to have additional semantics (setting the cookie in the context of a page limits it to... the URL? Domain?) over just a "plain" request header.

[aik099]

@xurizaemon , I've commented on that PR on GitLab website.