[rspec-users] stub_chain together with should_receive

David Chelimsky dchelimsky at gmail.com
Fri Nov 26 06:26:03 EST 2010

On Nov 26, 2010, at 2:24 AM, medihack wrote:

> David, sorry for double posting (it seems I am working too much and
> forgetting about what I already asked) ... and thanks for your answer.
> How about a bit more convenient way for future releases. Something
> like:
> MyModel.stub_chain(:tag_counts, { :offset =>
> 0 }, :limit, :order).and_return([])
> it could also work for multi parameters:
> MyModel.stub_chain(:tag_counts, { :my_method => [:param1, :param2,
> param3] }, :limit, :order).and_return([])
> What do you think?

I would be opposed to this feature.

There is an old guideline in TDD that suggests that you should listen to your tests because when they hurt there is usually a design problem. Tests are clients of the code under test, and if the test hurts, then so do all of the other clients in the codebase. Shortcuts like this quickly become an excuse for poor designs. I want it to stay painful because it _should hurt_ to do this.

I understand that chains like this are common in Rails apps thanks to good ideas like composable finders (which generally do not violate Demeter), but I don't think the parallel chains should appear in client code or in specs. e.g. if this is a controller spec, the model should expose a single method that wraps this, and if it's the model spec, the spec should just call the method that wraps the chain with different inputs and and specify the expected outcomes.

Even if I were in favor of the concept, the example above is confusing because it is a stub that becomes a message expectation. Constrained stubs like this (e.g. double.stub(:method).with(:arg).and_return(value)) are confusing enough as it is because they don't look like message expectations, but they act sort of like them. I've considered deprecating "with" on stubs and pushing toward an inline fake implementation instead:

account.stub(:withdraw) do |*args|
  if args.first == Money.new(200, :USD)
    # do something
    # do something else

RSpec already supports this, and it makes the intent of the spec much more clear and localized than:

  account.stub(:withdraw).with(Money.new(200, :USD)).and_return(TransactionResult.new)

In this case, if account receives withdraw() with anything other than Money.new(200, :USD), the outcome needs to be defined elsewhere in the spec, and these things quickly become spread out and difficult to understand.

That's my 2¢, but feel free to try to convince me otherwise :)


More information about the rspec-users mailing list