clean_class Twig filter does not work with Views rewriting

Patch applies cleanly and works as expected. Before and after screenshots attached
Note - I did uncheck the "Link label to the referenced entity" label in addition to following the steps above to test

before

after

Html::getClass would need its parameter documentation updated if going with this solution.

You've got a point. I can't argue with that. Views code feels the same way obviously.

+++ b/core/tests/Drupal/Tests/Component/Utility/HtmlTest.php
@@ -88,6 +91,11 @@ public function providerTestCleanCssIdentifier() {
+    // Test Html::getClass() handles Drupal\Component\Render\MarkupInterface input.

This comment is over 80 characters. Could be fixed on commit.

#14 is interesting. I do think the documentation should indeed be updated, and those others fixed in followup issues, but I'll get a third opinion.

@xjm asked me to look at this. I agree with @xjm this should be documented on the @param of the method so setting back to needs work.

Looking at the patch as a whole, I think the cast to string is okay. We have plenty of places where objects that can be cast to strings work like this - as pointed out by #14 - and fortunately this has nothing to do with safeness. I think it is totally okay if a cleaned css class loses any concept of safeness. I contemplated asking for a stricter type check before casting - ie something like MarkupInterface but I think that that is just unnecessary complexity.

I considered also swapping out the implementation in TwigExtension so only the twig filter does the cast to string but also I think that that is additional complexity for no gain.

Also since this is an API widening - ie. strings are still accepted and now stringy objects are too I think this is a patch safe bug fix.

+++ b/core/tests/Drupal/Tests/Component/Utility/HtmlTest.php
@@ -88,6 +91,11 @@ public function providerTestCleanCssIdentifier() {
+    $markup = HtmlTestMarkup::create($this->randomMachineName());
+    $string = Html::cleanCssIdentifier(Unicode::strtolower((string) $markup));
+    $this->assertSame(Html::getClass($markup), $string, 'Markup object is converted to CSS class.');

Rather than using random string and asserting against something programmatic let's just hard code a string and assert against that. Also in PHPUnit the expected value comes first in the assert. So the test should be something like:

$markup = HtmlTestMarkup::create('BLAH_CLASS');
$this->assertSame('blah-class', Html::getClass($markup));

Also adding "Needs followup" to investigate the other places an object-to-string cast is not documented in allowed values, as listed in #14. Thanks!

@benjifisher good point re stringy params. Tricky. The problem is when we only support PHP7 and use primitive typecasting this is going to become more issue. I wonder what happens in that case - does the String typehint force a cast or throw an error?

declare(strict_types=1);

function a(string $blah) {
  var_dump($blah);
}

class b {
  public function __toString() {
    return 'Hello world';
  }
}

$b = new b();
a($b);

This throws the following error:
PHP Fatal error: Uncaught TypeError: Argument 1 passed to a() must be of the type string, object given
So yeah I think mixed is the way to go because it is honest and more future proof.

Just moving it into the right queue.

It would work with {{ type|render|clean_class }} (see: https://www.drupal.org/node/2728915#comment-11207931).

I am moving this to RTBC based on #26

The Follow Up issue exists here https://www.drupal.org/node/2851911

As #17 states, I think casting to string is fine. It is acceptable that a cleaned up CSS class loses the state of safeness because special characters are being replaced anyway. Also, I have no other concerns since the string is going from safe to unsafe, meaning it will be escaped again.

I removed an unused use statement in the test file on the commit.

Committed d2a5fe7 and pushed to 8.4.x. Thanks!

This could be also potentially backported into 8.3.x.

Based on #17 and a private conversation with @xjm this is safe to be backported to 8.3.x.

Cherry-picked 4b369f3 and pushed to 8.3.x.