Can not run a WebTestBase or KernelTest base tests from inside a simpletest

So the error in the web test is "file_get_contents(/Volumes/devdisk/dev/sites/drupal8alt.dev/sites/simpletest/254593/.htkey): failed to open stream: No such file or directory" which seems like we fail to properly set up a webtest once we're inside a webtest.

The patch attached shows just how messed up things currently are.

The patch attached fixes running webtestbase tests from the UI in a webtestbase - aka testception :)

We need #2239969: Session of (UI) test runner leaks into web tests in order to fix KernelTestBase tests - but that patch broke WebTestBase testing in the UI so I suggest we commit this one before that.

Well the SimpletestTest::stubTest never tests making a call to anything that makes a request and this is exactly what was broken in #2239969: Session of (UI) test runner leaks into web tests so our test coverage is not enough.

In fact the SimpleTestTest is a mess - the test methods testInternalBrowser() and testUserAgentValidation() should not be in there since they impact the output of stubTest() that is run from testWebTestRunner() when we're in test inception. Having these additional do nothing test methods results in the child test running three test methods two of which do nothing accept print additional Enabled modules: non_existent_module messages.

As you can see from the patches attached in #0 we have a problem using the internal browser once DRUPAL_TEST_IN_CHILD_SITE is set to TRUE (constants which change values - I always disliked that in the simpletest multisite patch). The moment we do a request to the internal browser it will fail - this is because we have a new database prefix but the code in drupal_generate_test_ua() makes it utterly impossible to generate a .htkey file for the child site.

    if (DRUPAL_TEST_IN_CHILD_SITE && $parent_prefix = drupal_valid_test_ua()) {
      if ($parent_prefix != $prefix) {
        throw new \RuntimeException("Malformed User-Agent: Expected '$parent_prefix' but got '$prefix'.");
      }

It seems impossible for this exception to be thrown since drupal_valid_test_ua() is called with the child test prefix in TestBase::prepareEnvironment() - therefore the condition $parent_prefix != $prefix will never be met.

I would argue that this method of test the test runner is completely false and overly complex to understand and very liable to produce false positives - for example at the moment we're not even testing the ability to enable modules successfully when running tests through the UI since the only thing in setUp that we do (when in a childsite) is enable a module that does not exist. It would be far more realistic to just run one of the thousands of tests we have.

The only advantage of SimpleTestTest::testWebTestRunner() and SimpleTestTest::stubTest() is that it makes our assumptions clear. The 2272879.8.patch attached:

• Moves the browser tests to SimpleTestBrowserTest to keep testception down to only 1 test
• Adds a comment to SimpleTestTest so that people are warned not to add additional test methods
• Generates a .htkey in stubTest and logs in a user in stubTest to prove the internal browser works from the UI and that sessions do not get in the way.

BUT: I've attached 2272879-better.8.patch that does not generate the .htkey in from stubTest but includes the same fix as in #2 because I think this is the correct fix...

The drupal_generate_test_ua()/drupal_valid_test_ua() logic was and is invoked in a very controlled way

... since "this logic" Is directly responsible for this bug.

The better patch also includes a new test that just runs other tests - this would just fail if it was included in 2272879.8.patch.

Rerolled patch and we can now test both WebTestBase tests and KernelTestBase tests. Went for the option that is not change bootstrap.inc so that should make this patch acceptable to @sun - since test coverage of UI is more important.

Okay we need to add PHPUnit test coverage and there is no need to cover WebTestBase tests since this is already covered in SimpleTestTest through it's testception.

Test only patch available on #2306649: Running phpunit tests via the UI results in a fatal

re #14

1. sure
2. That's why reviewing having the full context is good... the full context is:
   

       // These cause the eleventh to fourteenth of the fifteen passes asserted in
       // confirmStubResults().
       $this->assertTrue(file_exists(conf_path() . '/settings.testing.php'));
       // Check the settings.testing.php file got included.
       $this->assertTrue(function_exists('simpletest_test_stub_settings_function'));
       // Check that the test-specific service file got loaded.
       $this->assertTrue($this->container->has('site.service.yml'));
       $this->assertIdentical(get_class($this->container->get('cache.backend.database')), 'Drupal\Core\Cache\MemoryBackendFactory');
   

3. I don't think so - it is only needed by stubTest because this is a test running in test.
4. Because stubTest is writing the file. #2246725: Make sure TestSitePath exist before creating .htkey is not really related - I guess that might be due to using the drush test runner. I run tests after a drush install all the time. But I do it using run-tests.sh

In my mind all the changes in the interdiff attached to #14 are out of scope since the fail and pass properties were not added or changed by this patch but it improves test readability so I'm ok with it.

rerolled - context conflict with #2322889: Various setUp() and tearDown() methods are not protected