HTMLRestrictions' diff operation bug: diff(<tag attr="A B">, <tag attr>) should return an empty result

For test coverage: I could add a test to HTMLRestrictionsTest like:

  public function test3278394(): void {
    $a = HTMLRestrictions::fromString('<tag attr="A B">');
    $b = HTMLRestrictions::fromString('<tag attr>');
    $this->assertSame([], $a->diff($b)->getAllowedElements());
    $this->assertSame(['tag' => ['attr' => TRUE]], $b->diff($a)->getAllowedElements());
  }

… but that would be silly because we already have the massive \Drupal\Tests\ckeditor5\Unit\HTMLRestrictionsTest::testOperations() with an enormous data provider.

Turns out there's a wrong expectation in the expectations, and that's how we missed this bug! Fixed test expectation in ed4a73cd3fea1b2b0598803d0be4ae895239b22a.

Now ready for review!

Uh oh — I fixed \Drupal\Tests\ckeditor5\Unit\HTMLRestrictionsTest::testOperations() but broke two other tests 😅

Now the only remaining failure is this:

Testing Drupal\Tests\ckeditor5\Unit\SmartDefaultSettingsTest
..........F.                                                      12 / 12 (100%)

Time: 00:00.039, Memory: 4.00 MB

There was 1 failure:

1) Drupal\Tests\ckeditor5\Unit\SmartDefaultSettingsTest::testCandidates with data set "Attribute needed, multiple matches" 
Failed asserting that two arrays are identical.
--- Expected
+++ Actual
@@ @@
             )
             'a' => Array &4 (
                 1 => Array &5 (
-                    'test_attr_values' => 1002
-                    'test_tags_and_attr_values' => 2002102
+                    'test_attr_values' => 0
+                    'test_tags_and_attr_values' => 2001100
                 )
             )
             'b' => Array &6 (
                 1 => Array &7 (
-                    'test_attr_values' => 1002
-                    'test_tags_and_attr_values' => 2002102
+                    'test_attr_values' => 0
+                    'test_tags_and_attr_values' => 2001100
                 )
             )
         )
@@ @@
         '-attributes-none-' => Array &8 (
             'test_all_attrs' => 100000
             'test_attrs' => 1100
-            'test_attr_values' => 1002
+            'test_attr_values' => 0
             'test_tags_and_attrs' => 2001100
-            'test_tags_and_attr_values' => 2002102
+            'test_tags_and_attr_values' => 2001100
         )
     )
 )

Mysterious, right? 🤓🙈

Let's unravel it. 🕵️

1. The meaning of those integers: a "surplus score" — see #3231328-43: SmartDefaultSettings should select the CKE5 plugin that minimizes creation of HTML restriction supersets for a handy overview.
2. We can see that all of the failures are because the expected surplus scores are higher than the ones actually being computed. In other words: the surplus scores became lower in a few cases. Why?
3. … because diff() now correctly handles an extra case. In the failing test, the desire is to support <foo bar> (meaning ANY value on the bar attribute on the foo tag), but test_attr_values only supports <foo bar="a b">. In other words: there is no surplus at all here. With the fix in this issue, that changes the score to 0 instead of 1002. The score used to be 1002 (meaning "two surplus attribute values" and "one surplus attribute") because <foo bar="a b"> DIFF <foo bar> === <foo bar="a b"> in HEAD, but we're fixing that to be <foo bar="a b"> DIFF <foo bar> = ∅, which completely explains that difference in score.

In other words: this is another example of something that was wrong in the test expectations!

Fixed in 3af97b37934a63f4f29bf0fd8acd4bb2c115995e.

6aa08d5fc as patch, to test against all 3 4 branches.

#8 OMG 🤣 🙈

Fixed! 😁 (This is a pre-existing bug, but totally reasonable to fix here IMHO. Which is why the patch size grew a bit: there were three places within the same file which contained this comment bug.)

Random unrelated Layout Builder test fail. Re-testing.

Not yet backported 😅