These feature tests were taking too long, we can't run them for every
single model.
I'm taking the approach of using one different model for each test, but
in theory only using a few models covering every possible scenario
would be enough.
This is a mistake I made in commit f2ef27d3. Back then I thought we
needed to keep Globalize.locale and I18n.locale in sync, but the truth
is it automatically happens when setting Globalize.locale to nil.
So now we can use I18n.with_locale (at least in the tests) and forget
about Globalize, which will make it easier to switch to Mobility in the
future.
Now factories define default headings for investments, so there's no
need to create a group and a heading to create an investment.
Likewise, in order to create a heading it isn't necessary to specify a
group anymore; specifying the budget is enough.
It's possible that there are more similar cases we haven't simplified
yet; I'm only addressing the obvious ones.
These settings are enabled by default.
It could be argued explicitely enabling the features makes tests more
consistent, because they'll work if we change the default setting. It
could also be argued that it makes tests more expressive because it
makes the reader realize certain things will only work if a setting is
enabled.
However, we were only doing so in a few tests. The truth is, thousands
of our tests depend on certain features being enabled. So IMHO we should
be consistent and either set them on every test, or not at all. I'm
choosing the latter option for simplicity.
We already configure `I18n.locale` and we reset Globalize's fallbacks
before every test.
On the other hand, RSpec automatically resets anything which is stub
with `allow`, so there's no need to use `and_call_original` in an
`after` block.
Settings are stored in the database, and so any changes to the settings
done during the tests are automatically rolled back between one test and
the next one.
There were also a few places where we weren't using an `after` block but
changing the setting at the end of the test.
When creating a budget investment with an unverified manager (for
example, a manager who isn't part of the local census), there's a
request to `Budgets::InvestmentsController#suggest`. Since the manager
isn't verified, suggestions can't be obtained.
There are serveral ways to fix this problem:
* Add a `suggest` action to Management::Budgets::InvestmentsController,
doing the same thing the main `suggest` action does.
* Give unverified users permission to access investment suggestions
* Give managers permission to access investment suggestions
I've chosen the last one because I thought it was simple and only
changed existing behaviour for managers, but any other solution would be
as valid. I haven't added the `phase: "accepting"` condition to keep it
simple, since a read-only action like this one in the management portal
isn't gonna create security risks.
While the test should work without this assertion, the intention of the
test is to check what happens if a user selects and investment and, once
it's selected, clicks the "next" link.
When we were inserting a row or replacing an existing one (just like we
do when we click the link to select an investment), we were entering a
row containing all columns, and all of them were displayed even if they
had been excluded using the column selector.
This caused the table to move in a strange way, which sometimes made the
investment selection tests fail.
This way we can reduce the number of locales used in the test
environment as well, while still testing every possible scenario for
remote translations.
We accidentally removed the `count` option in commit 55fb14ac, which
made the translation return a hash.
The test is a bit hacky, which makes me think changing the user
interface would probably be a better solution.
In JavaScript, when there isn't a `break` or `return` statement inside a
`switch` case, the next case will be executed as well.
That wasn't a problem here because CoffeeScript automatically inserts a
`return` statement in this specific situation. However, since we don't
want to return the result of the `hide()` operation, it might be easy to
accidentally remove the `return` statement, causing the code to break.
I've added a test for the scenario where neither `break` nor `return`
statements are present, so we don't run into this error.
We were performing an AJAX request and then a "normal" request without
checking the AJAX request had finished. Sometimes it resulted in the
normal request finishing because the AJAX request did, causing the test
to fail.
We're getting a failure on Travis in one of these tests. Debugging shows
the AJAX request rendering the first page (after clicking the "Previous"
link) takes too long and sometimes it exceeds Capybara's timeout.
After running the test thousands of times, the only way I've found to
clearly reduce the number of times the test fails is to reduce the
number of records shown on the first page. Other experiments, like
adding an `includes(:author)` to the query getting the proposals in the
controller, or adding `author: user` to the `create_list` part of the
test (so only one author needs to be fetched when rendering the
proposals) show inconsistent results regarding performance.
Note we still need at least 10 proposals for the test for several users,
to guarantee two users will never get the same records during the test
(or at least the probability they get the same records is one in
millions).
This number was not the important one; the important one is the one
given by the recounts.
Note we're also removing the votes by date, since they're also system
votes.
System count isn't a relevant number because the important one is the
number of votes counted by poll officers. We're still maintaining it for
a month in case poll officers would like to review the results.
When clicking the button "Search", the link "newest" is already present,
so capybara might click the "newest" link before the "Search" request is
finished, leading to unexpected results.
Checking the page to make sure the "Search" request has finished before
clicking the "newest" link solves the problem.
The `type: :feature` is automatically detected by RSpec because these
tests are inside the `spec/features` folder. Using `feature` re-adds a
`type: :feature` to these files, which will result in a conflict when we
upgrade to Rails 5.1's system tests.
Because of this change, we also need to change `background` to `before`
or else these tests will fail.
We're also adding a rubocop rule so we dont' accidentally add these
keywords again.
