Question

Ruby’s Test::Unit has assert_nothing_raised. Test::Unit has been replaced by MiniTest. Why don’t MiniTest’s assertions / expectations have anything parallel to this? For example you can expect must_raise but not wont_raise.

Answer 1

MiniTest does implement assert_nothing_raised in its Test::Unit compatibility layer, but in its own tests (MiniTest::Unit and MiniTest::Spec) it does not implement any test like this. The reason is, the programmer argues, that testing for nothing raised is not a test of anything; you never expect anything to be raised in a test, except when you are testing for an exception. If an unexpected (uncaught) exception occurs in the code for a test, you’ll get an exception reported in good order by the test and you’ll know you have a problem.

Example:

require 'minitest/autorun'

describe "something" do
  it "does something" do
    Ooops
  end
end

Output:

Run options: --seed 41521

# Running tests:

E

Finished tests in 0.000729s, 1371.7421 tests/s, 0.0000 assertions/s.

  1) Error:
test_0001_does_something(something):
NameError: uninitialized constant Ooops
    untitled:5:in `block (2 levels) in <main>'

1 tests, 0 assertions, 0 failures, 1 errors, 0 skips

Which is exactly what you wanted to know. If you were expecting nothing to be raised, you didn’t get it and you’ve been told so.

So, the argument here is: do not use assert_nothing_raised! It’s just a meaningless crutch. See, for example:

https://github.com/seattlerb/minitest/issues/70

https://github.com/seattlerb/minitest/issues/159

http://blog.zenspider.com/blog/2012/01/assert_nothing_tested.html

On the other hand, clearly assert_nothing_raised corresponds to some intuition among users, since so many people expect a wont_raise to go with must_raise, etc. In particular one would like to focus an assertion on this, not merely a test. Luckily, MiniTest is extremely minimalist and flexible, so if you want to add your own routine, you can. So you can write a method that tests for no exception and returns a known outcome if there is no exception, and now you can assert for that known outcome.

For example (I’m not saying this is perfect, just showing the idea):

class TestMyRequire < MiniTest::Spec
  def testForError # pass me a block and I'll tell you if it raised
    yield
    "ok"
  rescue
    $!
  end
  it "blends" do
    testForError do
      something_or_other
    end.must_equal "ok"
  end
end

The point is not that this is a good or bad idea but that it was never the responsibility of MiniTest to do it for you.