Today I learnt that these two tests are, not only non-functionally different, but functionally different too:

Test A:

it "should run without raising an exception" do
  some_code_to_test
end

Test B:

it "should run without raising an exception" do
  expect {
    some_code_to_test
  }.not_to raise_error
end

Like JUnit, rspec reacts to tests which raise exceptions by failing the test,  right? So these two tests should (functionally) behave identically?

Wrong!

Well, sometimes wrong.

The difference (or rather, at least one of the differences — perhaps there are more) lies in SystemExit. If some_code_to_test raises this, typically by calling Kernel#exit,  then the test runner stops, treats this test as successful, doesn’t  show the output of this test, and silently skips all subsequent tests:

it "runs test 1" do
end

it "should run without raising an exception (2)" do
  some_code_to_test
end

it "runs test 3" do
end

// and then run it:

rachel@shinypig test$ bundle exec rspec --format doc spec/my_spec.rb

Kernel
  runs test 1

Finished in 0.00089 seconds (files took 0.08703 seconds to load)
2 examples, 0 failures

which reports that it ran 2 examples, but only shows one of them, and doesn’t even mention the one that it skipped completely.

On the other hand, if we wrap the code being tested using “expect … not_to raise_error”:

rachel@shinypig test$ bundle exec rspec --format doc spec/my_spec.rb

Kernel
  runs test 1
  should run without raising an exception (2) (FAILED - 1)
  runs test 3

Failures:

1) Kernel should run without raising an exception (2)
     Failure/Error: expect { some_code_to_test }.not_to raise_error

expected no Exception, got #<SystemExit: exit> with backtrace:
         # ./spec/my_spec.rb:4:in `exit'
         # ./spec/my_spec.rb:4:in `some_code_to_test'
         # ./spec/my_spec.rb:11:in `block (3 levels) in <top (required)>'
         # ./spec/my_spec.rb:11:in `block (2 levels) in <top (required)>'
     # ./spec/my_spec.rb:11:in `block (2 levels) in <top (required)>'

Finished in 0.01099 seconds (files took 0.07536 seconds to load)
3 examples, 1 failure

Failed examples:

rspec ./spec/my_spec.rb:10 # Kernel should run without raising an exception (2)

So now it runs all three tests, showing all three results, including one failure.

To  be honest I’m surprised that the rspec test runner doesn’t deal with  this natively: if a test raises an error, it should always be a failure (to my mind anyway), including SystemExit. Maybe there’s a bug / pull request for rspec somewhere discussing this, where the idea was rejected. I might go digging…


This content originally appeared on my Medium account