Workbench moderation + field encrypt + translations isn't working

As talked with @nerdstein this is probably "just" the problem that field encrypt doesn't support translations at all.

Reporting some progress ...

It turns out we store a different value for the translated value in the DB:

select id, entity_id, entity_revision_id, langcode, encrypted_value__value  from encrypted_field_data;
+----+-----------+--------------------+----------+------------------------------------------------------+
| id | entity_id | entity_revision_id | langcode | encrypted_value__value                               |
+----+-----------+--------------------+----------+------------------------------------------------------+
|  2 |         2 |                  5 | en       | MTIzNDU2Nzg5MDEyMzQ1NjxjPmdyZmcgb2JxbDwvYz4NCg==     |
|  3 |         2 |                  6 | en       | MTIzNDU2Nzg5MDEyMzQ1NjxjPmdyZmcgb2JxbDwvYz4NCg==     |
|  4 |         2 |                  6 | de       | MTIzNDU2Nzg5MDEyMzQ1NjxjPmdyZmcgb2JxbCBRUjwvYz4NCg== |
+----+-----------+--------------------+----------+------------------------------------------------------+

Given that I guess this is a rendering problem.

So @nerdstein and myself discussed that a bit.

There are two possibilities to implement proper translation and revision support:

1. We store one encrypted_value-entity per translation and revision of the source entity. See the first patch. This fixed the issue described well in the issue summary for me.
2. We store one entity per source entity and encrypted_value-entity (entity_type_id, entity_id, field_name) and sync translations and revisions with the source entity. This is basically how the module was designed to work, but it never worked really.

I'm not 100% convinced that going with the way 1) is the only possibility we have, but it certainly simplifies things a lot.

Here is a try to actually add some test coverage just for that case, turns out, either the test or the actual implementation is still wrong.

This time also with the test.

Currently I'm a bit buffled about

    $languages = $entity->getTranslationLanguages();
    foreach ($languages as $language) {

For the testcase the languages are 'en', and 'fr', but the entity internally stores in 'fr' and 'x-default', which is maybe a problem here as well.

Its tricky. I tried to write the tests properly on the storage layer, but I could not yet reproduce the issue on that layer.

While working on the test I realized a couple of issues though:

* We use a hook which is cached, see #2743073: hook_entity_storage_load() is cached
* We potentially re-encrypt '[ENCRYPTED_VALUE]'. I haven't figured out when this happens exactly, but I've seen the behaviour while writing the tests
* The tests require a $storage->reset() call for some reason

I have also reviewed and it appears to be working for me.

Thank you!

This was fixed in #13, so I'm closing the issue now. If you have a similar problem, please open a new issue.