When I removed this and tested it locally, the feature spec still passed, but it is failing on CI, so it looks like we still need this `execute_script` use until we find a better Capybara strategy.
In addition to getting rid of an "execute_script" instance, this approach is slightly higher fidelity, as it renders the hovercard at the correct place on the page and doesn't reach into jQuery to trigger the hovercard.
This call to `execute_script` would directly call jQuery's `sortable` function on an element to make it sortable. However, this isn't necessary; the test can still drag the "Cat People" aspect to the desired position without it.
The apparition driver does not support "scroll_by" without going through "execute_script", but we can better express the intention of this test and make the scrolling action a bit more general by replacing "scroll a bit" with "scroll to [some element]" — in this case, we need to scroll until the "Change" button is visible so that we can press it.
This change_settings.feature spec is the only place that used "I scroll a bit".
This set of specs — in particular, all the calls to `this.view.$el.find(...).trigger('click') — are changing the environment in ways that cause I18n missing key errors elsewhere.
There are failures in `app_spec.js` if this spec file is run before it, and I haven't found a way to reset the environment to avoid the issue.
This is from Flaburgan: https://github.com/diaspora/diaspora/pull/8305/files
I can't find a better way around it. The specs that call `onChangedUnreadStatus` change the environment in ways we can't figure out how to reset cleanly.
This was failing if `app.aspect` got set by another spec than ran before it (such as one of the other specs in the same file, when run in a different order).
These tests were failing because of an error thrown from `this.header.render()`, unless `head_view_spec.js` had already been run to set app.notificationsCollection.
These cases of modifying app.stream can cause other specs to fail, depending on test order. Here we either don't modify them if the tests still pass without manually stubbing `addNow`, or we cache and restore the properties we need to test.
