Add views display always defaults to attachment

Thanks for reporting!

git bisecting this now.

#2244447: Translation of low-level info/annotations leads to circular dependencies is what broke this, not yet sure why or how.

Okay here's the fix, this could use a test.

Hmm, I can't get this to fail locally. But manual testing of it fails.

This is an alternate approach to fix it up a level.

So this technically still needs works?

Coming here from #2299309: Adding a "REST export" Views display adds an "Attachment" display instead. This 90% case functionality is just plain broken, so I agree critical seems appropriate.

What's happening before is $plugins[$id] is a TranslationWrapper object like so:

array(7) {
  ["block"]=>
  object(Drupal\Core\StringTranslation\TranslationWrapper)#514 (4) {
    ["string":protected]=>
    string(5) "Block"
    ["arguments":protected]=>
    array(0) {
    }
    ["options":protected]=>
    array(0) {
    }
    ["stringTranslation":protected]=>
    NULL
  }

Casting it to a string produces just "Block" and makes the link work again.

I guess the first thing I'd ask is why $plugin['title'] is a TranslationWrapper object in the first place; I would imagine most code would (like this code) assume 'title' is, well, a title. ;) Is that something that needs to be fixed even further upstream?

Digging into SimpleTest results, I can confirm that for whatever reason, SimpleTest's browser is *not* hitting this issue:

And a debug($this->getUrl()) shows /block_2 as well. Hrm.

AFAICS, this also "fixes" the bug. Which explains why Simpletest passes.

Huh. Yep, I confirm that fixes it indeed.

There's a minor UI regression which is that now when you click "Add" you get:

As opposed to before you'd get:

That seems like a suitable trade-off for it working, you know, at all. ;) Still though, it might be good to understand how #2244447: Translation of low-level info/annotations leads to circular dependencies broke functionality that's been unchanged since Views 7.x-3.x was first branched. It might allow us to find other instances that are broken too which we haven't found yet.

Meh. I dunno. Granted, I'm definitely not the savviest JS person around, but I dug into this for another hour and can't figure it out.

Tentatively doing this. I'd still like to ultimately get down to the bottom of this (maybe nod_ or someone else JS-inclined could help?) but a normal follow-up of "regression: Fix Add Add Add on Views UI displays" seems preferable to Views displays not working at all for any JS-enabled browser.

The other fix was equally functional (and probably would fix other bugs), it was needs work only for tests...

Oh good point. And since the tests would have caught this had the bug not only manifested in JS...

Committed and pushed to #5 to 8.x (the extra assertion doesn't hurt anything). Thanks!