This is done by rescuing `StandardError` instead of `Exception`, which should
still cover most if not all errors raised by the Curb and Feedjira libraries. We
no longer try to rescue fatal errors, like `NoMemoryError`, which is what
RuboCop was complaining about.

An even better fix for this could be to rescue an even narrower set of errors,
but I'll leave that until the next time.

Fix RuboCop lint offences

Only one autofocus element per page is allowed.

> ### delete([other_str]+) -> new_str
>
> Returns a copy of *str* with all characters in the intersection of its
> arguments deleted. Uses the same rules for building the set of characters as
> `String#count`.

See <http://ruby-doc.org/core-2.0.0/String.html#method-i-delete>.

Fix RuboCop performance offences

Autofocus

...through some light refactorings.

...through a small refactoring of
`Sinatra::AuthenticationHelpers#needs_authentication?`.

Matt and I agreed that 120 characters is a sensible maximum line length for
modern editors, see
<#395 (comment)>.

Also set the `MaxLineLength` option for two other cops, since they're related.

Joins lines in the Fever API files to fix warnings from
`Style/IfUnlessModifier`.

Despite my best efforts, I find it hard to keep all methods below or at the
default maximum length of 10 lines. 15 lines is a somewhat arbitrary compromise
that's subject to future tweaking.

See also: <https://robots.thoughtbot.com/sandi-metz-rules-for-developers>

Fixes Metrics/ClassLength by extracting `StoryRepository#expand_absolute_urls`
and `StoryRepository#normalize_url` to a separate helper module.

Fix RuboCop metrics offences

Adds group_id to Feeds#edit

… new rubocop warnings (#541)

These gems aren't dependencies anywhere. Not sure why they are still
lurking in our Gemfile.lock, but every time I run bundle it removes
them, so might as well commit that.

This makes it so that we can pass input and output to ChangePassword in
order to test against them, and prevent them from outputting in the
tests.

ActiveRecord no longer wants us to pass SQL fragments to `order`. This
instead makes use of the underlying framework Arel to generate the
query.

I'm a little surprised this was possible, but #533 introduced a version
of ActiveRecord that did not match the constraints of
`delayed_job_active_record`. I thought bundler was more strict, but it
somehow allowed this one to slip through. This fixes the issue.

fix a rubocop alert from the newer version faker
work around a capybara change

And auto generate the configuration

This is super janky, grabbing the instance variable out of the instance,
but there doesn't seem to be a great way to test for the underlying
issue, which is that in some cases jobs don't work when passing the
entire record. I thought also about inspecting the job more closely, but
wasn't able to think of a great solution there, either. I'd like to
refactor this class some down the road to not try to juggle both
`@feeds_ids` and `@feeds` variables.

I enabled settings for test mode, which doesn't seem to noticeably
impact performance, but gives us a little bit more test coverage.

* Add tests for `.fetch_by_ids`, `.delete`, `.list`, and `.in_group`
* Rearrange test file to mirror ordering of class
* Fix deprecation warning for `lower(name)`
* Refactor `in_group` query to use AR syntax

This adds `rubocop-rails`, `rubocop-rake`, and `rubocop-rspec`. Even
though this project doesn't use Rails, a lot of the same linters will
apply.

Also regenerates `.rubocop_todo.yml` based on the new rules.

The test wasn't actually executing the method under test, so it wasn't
verifying anything.

Also made it so that auth check applies in test mode as well. Needed to
make a couple of other changes to tests to get them working.

This brings test coverage up to 100%, so I'm also enabling the
`minimum_coverage` threshold to make sure it stays there.

There doesn't appear to be anything in [the major release][release] that
should impact us.

[release]: https://github.com/defunkt/unicorn/releases/tag/v6.0.0

This blocks upgrading to ActiveRecord 6.x, as the query for
`true`/`false` values is different for some reason depending on the
database, and it fails tests with SQLite. The simplest solution in this
case is probably to remove SQLite. The production setup uses Postgres,
so better to have a consistent database setup, anyway. This also frees
us up to take advantage of Postgres specific features.

I needed to update the format of the json being returned in the tests.
I'm not sure if it actually ends up changing the result when the dates
are converted to JSON strings, but it doesn't look like anything client
or server relies particularly on the format, so we should be fine.

Also updates the formatting in `schema.rb` by running `rake db:migrate`.
All of the changes look to be cosmetic.

This updates minor versions of gems.

The [coveralls gem][gem] hasn't been updated in more than a year. The
[reborn][reborn] is a more up to date version that doesn't lock us down
on an older version of SimpleCov. We may want to consider removing
Coveralls entirely and just using the CI coverage, perhaps also keeping
the artifacts for easy reference.

[gem]: https://github.com/lemurheavy/coveralls-ruby
[reborn]: https://github.com/tagliala/coveralls-ruby-reborn