Concerns about multiple logical assertions

[AlexArchive]

The consensus among the community appears to be that each test should have but one logical assertion.

If a developer wants to test that a controller returns a model of the correct type explicitly, they must also test that a view with a particular name is rendered - that's two logical assertions.

[Test]
public void Index_ReturnsCorrectModelType()
{
    repository
        .Setup(repo => repo.All())
        .Returns(PostsMother.CreateEmptyPosts());

    controller
        .WithCallTo(c => c.Index(1))
        .ShouldRenderDefaultView()
        .WithModel<PagedList<Post>>();
}

I suspect that this kind of test goes against your philosophy but I am going to bring it up for discussion anyway.

Links:

• http://osherove.com/blog/2005/4/14/try-to-avoid-multiple-asserts-in-a-single-unit-test.html
• http://stackoverflow.com/questions/2430429/are-multiple-asserts-bad-in-a-unit-test-even-if-chaining
• https://github.com/ploeh/Booking/blob/1ccdcc61b699b80f1be0ee5a87c7981d5baace2d/BookingWebModel.UnitTest/HomeControllerTests.cs#L30

[robdmoore]

FYI - I do plan on responding to this and the other one. The week is just getting the better of me so far ;)

[AlexArchive]

It's all good mate - I understand.

[robdmoore]

Hi,

So my stance on multiple asserts mirrors something that @pawelpabich once said on an internal Readify site:

Personally I try to test one logical concept. Assert.XXXX is simply an implementation detail and don't mind using more than one.

So if I look at asserting that a particular controller action returns a view with a model that has a certain structure then I see that as being one logical concept. Splitting out the assertions to have a separate test for the fact it's an action vs the fact the model type is right vs the structure of the model is correct (like your third link) seems really verbose, a waste of developer time and makes it harder to understand what you are actually asserting in the first place in my opinion.

I wouldn't see any point for instance in having a different test for model type vs view - if they didn't return a view there is no point checking the model type since there is no model. I can understand that there might be a situation though where you don't care about what the view name was and want to check the model. If that's a scenario you are interested in I'm happy to talk about adding a new method off of ActionResult that checks for the model, bypassing the view check :).

[AlexArchive]

You makes some good points and ones that I find hard to argue because they are ones that I have found to be true in practice.

I intend on doing some more research and then continuing the discussion. In particular, I want to see if there are any astute arguments for the approach used in the third link.

I can understand that there might be a situation though where you don't care about what the view name was and want to check the model.

Well if you are willing to support the testing of the model in isolation for any reason - then those who want to write a test like the one I described can and it won't matter whether we agree or not - not unless you want to encourage what you perceive to be a good practice that is.

[robdmoore]

As fair warning it's a fairly hotly contested subject with lots of opinions on both sides. I tend to side on the favour of pragmatism :)

If you don't care about the view, but do care about the model then it's valid to want to test that. If someone wants to use the library in a particular way then I don't mind even if I disagree with their approach :p