Add a test for the default tracker plugin

Yes, a Kernel test sounds most reasonable here. Without a DB connection, you won't get very far (and mocking that is bordering on the absurd), but there's no reason for going full Functional.

OK, here it is, and I actually spotted a problem through this. And not even with the new LIFO option – on the contrary, something that's only a problem with FIFO:

@@ -266,6 +266,12 @@ public function trackItemsUpdated(array $ids = NULL) {
         if ($ids_chunk) {
           $update->condition('item_id', $ids_chunk, 'IN');
         }
+        // Update the status of unindexed items only if the item order is LIFO.
+        // (Otherwise, an item that's regularly being updated might never get
+        // indexed.)
+        if ($this->configuration['indexing_order'] === 'fifo') {
+          $update->condition('status', static::STATUS_INDEXED);
+        }
         $update->execute();
       }
     }

Open question: Should we do the same for trackAllItemsUpdated()? Or is that expected to update all items to the same "changed" timestamp, no matter whether they were indexed before?

A second set of fails I had to fix (since I went a bit berserk and aimed for a 100% coverage) was that, while all the other methods took care to not let an exception escape (since the interface doesn't allow for exceptions), the ones for returning the various counts (and getRemainingItems()) didn't. To be consistent, I added try/catch blocks for those, too.
(The possible exceptions are, in all cases, just database exceptions that should only occur when the database suddenly goes away. Actually, hiding these exceptions and just carrying on seems a bit dangerous, but if the database connection is lost I don't really know what else we could do. Probably, whatever event we wanted to track in such a case would fail anyways.)

Thanks a lot for reviewing!

We can replace this with the following, so we don't have to add the @todo.
It also looks like the returncallback isn't used here. At least, when I replaced that locally all tests still passed.

The callback ensures that a test fails if an exception occurs within a method. So, it's expected that the tests would still pass without this whole block. Maybe we should just add a comment to make that clearer? Patch attached.

I also actually like that we add the @todo, since we don't have one in the code base for that issue yet anyways.

I've never used this "prophecy" thing. Seems like modern nonsense. ;P Is it really better than a normal mock, or just different syntax?

So, is this RTBC with the addition of the comment?

Great to hear, thanks a lot again for reviewing!
Committed.