Remove string translation services from TestBase container

Do you perhaps need to put the translation services into WebTestBase if you take them out of TestBase? It looks like the failing tests extend WebTestBase.

I'm spotting these causes:

1. The Installer (web) tests are trying to generate form button label translations, but do not have a container available. Unlike WebTestBase, Drupal is not installed yet. We need to change those tests to actually resemble the minimal bootstrap container that is built in install_begin_request(), ideally even with the identical logic for discovering string translations. Before attempting that, it would be good to do #2175459: Convert Drupal\system\Tests\InstallerTest into a new Drupal\simpletest\InstallerTestBase first.
2. The Drupal\system\Tests\Plugin\* classes appear to be unit tests. I suspect they're using t() in assertion messages.
3. Ditto for Drupal\system\Tests\Common\SizeUnitTest.

I would actually recommend to remove these entire lines, too:

  $this->container = new ContainerBuilder();
  \Drupal::setContainer($this->container);

Instead, the patch in #2171315: Ensure a NULL container in UnitTestCase::setUP will allow us to simply nuke the global state container:

  \Drupal::setContainer(NULL);

I wonder whether these are appropriate unit tests in the first place?

If a test needs a service container, then it ought to be DUTB?

→ Let's simply change all of those into DUTB?

Closely related side-note: My ultimate goal for D8 is to remove UnitTestBase prior to 8.0. Unit tests should exclusively be PHPUnit tests only. That's going to leave us with DrupalUnitTestBase + WebTestBase for the non-unit scenarios.

(At that point, we could consider to rename DrupalUnitTestBase into UnitTestBase, although I'm not sure whether that's a good idea. A name like ExtensionTestBase or KernelTestBase would probably be more accurate.)

Last patch was not the patch you intended to upload; that's why I canceled it ;)

Attached patch changes those tests to DUTB.

Also added a @todo for #2171315: Ensure a NULL container in UnitTestCase::setUP, even though I'd actually prefer to get that in first, so we can nuke $this->container entirely here.

Did not touch the installer tests either, waiting for #2175459: Convert Drupal\system\Tests\InstallerTest into a new Drupal\simpletest\InstallerTestBase

I wonder whether these are appropriate unit tests in the first place?

If a test needs a service container, then it ought to be DUTB?

→ Let's simply change all of those into DUTB?

Well, we have unit-testable code in the background for format_plural() but not for format_size(), see #2157945: Deprecate format_size() and use Drupal\Core\StringTranslation\ByteSizeMarkup instead
PluginTestBase certainly can be tested using pure unit tests.

Let's do that all in follow ups.

Re-rolled. Let's see what remains.

1. Fixed HtmlIdentifierUnitTest.
2. Fixed TableSortExtenderUnitTest.
3. Fixed DiscoveryTestBase.
4. Fixed InstallerTestBase requires a minimal mock container in order to issue requests.

Let's quickly move forward here to prevent a new test to introduce a new dependency on the minimal mock container of TestBase! :-)

19: test.translation.19.patch queued for re-testing.

Merged HEAD.

23: test.translation.23.patch queued for re-testing.

23: test.translation.23.patch queued for re-testing.

23: test.translation.23.patch queued for re-testing.

Merged 8.x.

Great! This is almost identical to #2275965: Remove all services from the TestBase container.

1. +++ b/core/modules/simpletest/lib/Drupal/simpletest/TestBase.php
   @@ -1229,7 +1209,7 @@ private function restoreEnvironment() {
   -    $this->container->set('current_user', $this->originalUser);
   +    \Drupal::getContainer()->set('current_user', $this->originalUser);
   

   Can we remove that line entirely? This operates on the original container. I guess there is no way that this service is tampered with from within tests?

2. +++ b/core/modules/system/lib/Drupal/system/Tests/Common/HtmlIdentifierUnitTest.php
   @@ -7,13 +7,12 @@
   +class HtmlIdentifierUnitTest extends DrupalUnitTestBase {
   
   +++ b/core/modules/system/lib/Drupal/system/Tests/Common/SizeUnitTest.php
   @@ -8,12 +8,12 @@
   +class SizeUnitTest extends DrupalUnitTestBase {
   
   +++ b/core/modules/system/lib/Drupal/system/Tests/Common/TableSortExtenderUnitTest.php
   @@ -7,13 +7,13 @@
   +use Drupal\simpletest\DrupalUnitTestBase;
   ...
   +class TableSortExtenderUnitTest extends DrupalUnitTestBase {
   
   +++ b/core/modules/system/lib/Drupal/system/Tests/Plugin/Discovery/DiscoveryTestBase.php
   @@ -7,12 +7,12 @@
   +abstract class DiscoveryTestBase extends DrupalUnitTestBase {
   
   +++ b/core/modules/system/lib/Drupal/system/Tests/Plugin/PluginTestBase.php
   @@ -17,7 +17,15 @@
   +abstract class PluginTestBase extends DrupalUnitTestBase {
   

   Can we use KernelTestBase instead?

1. Removed current_user futzing in TestBase::restoreEnvironment().
2. Converted DUTB to KTB for affected tests.

Great. Running a couple of unit-tests and web-tests through the UI did not reveal any problems, this is ready for inclusion.

+++ b/core/modules/simpletest/src/TestBase.php
@@ -1071,29 +1071,9 @@ private function prepareEnvironment() {
+    // Ensure there is no service container.
+    $this->container = NULL;
+    \Drupal::setContainer(NULL);

As \Drupal::getContainer() is documented to return a ContainerInterface, I think this is slightly weird. Is there any value in NULL over new ContainerBuilder()?

Merged 8.x.

@tstoeckler: Passing a NULL value is explicitly supported and desired in this case, because we do not want to build + set a new container, but instead, we explicitly want to reset the container in \Drupal to NULL.

Prevent var_export() in default assertion message from getting invoked in DiscoveryTestBase.

It looks like the plugin discovery tests did not run against the actual plugin_test.module plugins in HEAD. Otherwise, the var_export() PHP warning would appear in HEAD already.

With that, we should be back to RTBC.

Merged 8.x. Back to RTBC.

Merged 8.x.

Committed/pushed to 8.x, thanks!