RSpec failures caused by I18n.locale

Recently, I ran into a RSpec issue that an example failed when running with all the tests, but it can pass when it's ran separately.

It's obvious that this test was affected by some other tests. Then I found this interesting RSpec failure caused by I18n.

This blog post is a note about how I found the issue and how I solved it.

Find the order dependent failures

Since there were not many tests in this project, and they were not running randomly, it was easy for me to find the test files that caused the failing example.

But during my research, I also found that RSpec has a more elegant way to find the order dependent failures: rspec --bisect

rspec --bisect

When we run rspec --bisect, RSpec will:

  1. run all the tests to find the failing examples
  2. run failing examples independently to check if it's order-dependent
  3. bisect all the examples to find the minimum example set to reproduce the issue

It's way faster than testing it by hand.

rspec --seed 51012 --bisect
# Bisect started using options: "--seed 51012"
# Running suite to find failures... (18.73 seconds)
# Starting bisect with 1 failing example and 39 non-failing examples.
# Checking that failure(s) are order-dependent... failure appears to be order-dependent

# Round 1: bisecting over non-failing examples 1-39 .. ignoring examples 21-39 (21.18 seconds)
# Round 2: bisecting over non-failing examples 1-20 .. ignoring examples 11-20 (21.46 seconds)
# Round 3: bisecting over non-failing examples 1-10 . ignoring examples 1-5 (9.71 seconds)
# Round 4: bisecting over non-failing examples 6-10 . ignoring examples 6-8 (9.49 seconds)
# Round 5: bisecting over non-failing examples 9-10 .. ignoring example 10 (18.34 seconds)
# Bisect complete! Reduced necessary non-failing examples from 39 to 1 in 1 minute 30.53 seconds.

# The minimal reproduction command is:
#   rspec './spec/controllers/children_controller_spec.rb[1:1:5:1]' './spec/features/section_spec.rb[1:1:1]' --seed 51012


Turns out minitest also provides an extra package that does the same thing: seattlerb/minitest-bisect. Check it out if you are using minitest.

Failing examples caused by i18n.locale

After I found the example that caused my feature spec to fail, I started debugging the issue. Turns out, it's pretty simple:

  1. A controller spec created some account fixtures, and set its language to French
  2. We're setting the i18n.locale value according to user account's language, so i18n.locale = :fr now
  3. My feature spec is using Capybara to find a button with text "Create", but since now the i18n.locale is :fr, the button is not showing "Create" (which is for :en locale). Thus this feature spec failed.

And this seems to be an old issue for i18n: I18n.locale cached in rspec · Issue #256 · svenfuchs/i18n

We have no choice but to reset i18n.locale for every test (by setting it directly or setting account fixture's default language to :en so that i18n.locale is set to :en every time).

One More Thing: git bisect

Since the i18n.locale behavior in our controller was introduced long time ago, I didn't get the chance to use git bisect, which can help us to find the commit that caused the test to fail.

If you are interested in this command, you can also check these documents: