In Rails 6.1, the classic autoloader is deprecated.
We were getting an error because we were using `autoload` in the
ActiveStorage plugin for CKEditor:
expected file app/lib/ckeditor/backend/active_storage.rb to define
constant Ckeditor::Backend::ActiveStorage
So we're removing the line causing the error.
Finally, we can now restore all the tests that that failed sometimes
with the classic autoloader and that we modified in commits 2af1fc72f
and 8ba37b295.
When running these tests, under certain conditions, we get a warning
followed by an error:
```
activesupport-6.1.7.7/lib/active_support/dependencies.rb:502:
warning: already initialized constant ActiveStorage::Representations
activesupport-6.1.7.7/lib/active_support/dependencies.rb:502:
warning: previous definition of Representations was here
Failure/Error: raise LoadError, "Unable to autoload constant
'#{qualified_name}', expected #{file_path} to define it"
LoadError: Unable to autoload constant
ActiveStorage::Representations::RedirectController, expected
activestorage-6.1.7.7/app/controllers/active_storage/representations/redirect_controller.rb
to define it
```
The error seems to take place when we request a page in a test that
loads two (or more) ActiveStorage images if ActiveStorage hasn't loaded
yet, although it's a flaky error and so the test doesn't always behave
like this.
We've tested that switching to zeitwerk solves the issue but, since we
aren't switching to zeitwerk in version 2.1.1 and we'd like this version
to run all tests correctly, for now we're changing the tests so they
don't load two records with images.
On of these tests ("Polls Index Polls can be listed") fails on my
machine when run individually. I haven't been able to consistently
reproduce the other ones.
They were marked as pending.
Note Capybara doesn't support finding a button by its `aria-labelledby`
attribute, so we're using the ugly `click_button "Yes"`, like we did in
commit fabe97e50.
Having to wait for a whole page refresh after updating each setting was
painful when modifying several settings.
Even though the navigation is updated immediately to reflect which
sections have been enabled/disabled, there's one gotcha. Changing the
"SDG" setting will not update the user menu (which contains a link to
SDG content) nor the "SDG Configuration" tab; refreshing the page will
be necessary to check these changes. The same happens with the map and
remote census tabs. So in these cases we're making an exception and
sending the form. We might find a better solution in the future.
For this reason, we aren't using the `switch` ARIA role. Some users
might not expect a switch control to refresh the page, just like they
usually don't expect checkboxes to refresh the page. Furthermore, screen
reader support for the `switch` role seems to be inconsistent. For
instance, NVDA with Chrome announces the control as a checkbox instead
of a switch.
Note AJAX is only used for feature settings. Other settings are still
updated with regular HTTP requests.
Since we're now using AJAX requests, we have to make sure to add an
expectation in the homepage tests in order to make sure the request has
finished before starting a new one.
JavaScript is used by about 98% of web users, so by testing without it
enabled, we're only testing that the application works for a very
reduced number of users.
We proceeded this way in the past because CONSUL started using Rails 4.2
and truncating the database between JavaScript tests with database
cleaner, which made these tests terribly slow.
When we upgraded to Rails 5.1 and introduced system tests, we started
using database transactions in JavaScript tests, making these tests much
faster. So now we can use JavaScript tests everywhere without critically
slowing down our test suite.
Content like lowercase letters with `text-transform: uppercase` or
spaces after elements with `display: block` or "You're on page:" are not
seen that way by users with a browser supporting CSS.
So we're testing what most users actually experience.
They were all following the same format.
Note we need to group the `see_all` translation keys together (the same
way it's done with the `most_active` keys) so we don't have an unused
translation warning.
We're also moving the "see all" link in processes outside the feed
content; the same way it's done in debates and proposals and removing
unnecessary classes in the processes feed: the column class is causing
the processes not to be aligned with the debates above them, and the
margin bottom is not needed because the margin of the footer is already
enough.
We were repeating the same code over and over (with a few variants) to
setup tests which require an administrator. We can use a tag and
simplify the code.
This rule was added in Rubocop 0.89.0. However, there are some false
positives when we don't use interpolation but simply concatenate in
order to avoid long lines. Even if there weren't false positives, there
are places where we concatenate to emphasize the point that we're adding
a certain character to a text.
We might reconsider this rule in the future, since we generally prefer
interpolation over concatenation.
In this case the confirmation dialog isn't really necessary since the
action to enable/disable the setting can easily be undone.
Furthermore, these tests were failing with Chrome 83, probably because
we use `confirm_dialog` and then we use `visit` without checking the
page in between.
In theory we shouldn't need to check the page in between because the
request generated by `confirm_dialog` is a synchronous one and so
`visit` isn't executed after the previous request has finished, but
apparently this behavior has changed in Chrome 83.
We could add an expectation before executing the `visit` method, but
that wouldn't improve the usability of the application.
