Related
Most posts on Rails profiling recommend Ruby-Prof. To use Ruby-Prof I need to write at least one new test for each controller action, then manually compare the results to see what's taking the longest and might be a candidate for optimization.
This is good if I already know exactly what request I'm focusing on. It seems less good if I'm trying to identify the hot spots in the first place. Given that I already have a huge integration test suite covering all the app's functionality that I care about, it seems like what I really want to do is:
Run the entire test suite and capture the time spent in each controller action. (Or model method, or whatever level of granularity I want.)
Print two lists, of worst-case and average-case times in each controller action.
Sort each list and start investigating the longest-running controller actions, now using Ruby-prof or other profiling tools to drill down into the call stack. The worst-case times will identify request params that might be problematic (i.e., trigger slow code on the backend), without my having to think of them all when I write the performance test.
Is there some reason people don't use the integration test suite in this way, rather than basically duplicating it with a second performance test suite? I have not seen it suggested. Before I write code to do something like this (presumably with a before_action in ApplicationController, is there already a tool for this?
I think that the automated tests will not tell you anything about performance. You need real data. For example, your tests probably won't use indexes, but if you create 10,000 records without an index you may find a performance issue.
I need to write at least one new test for each controller action
Why would you performance test each controller action?
In my limited experience, performance testing was done after deploying the app and tested very specific things. I tested a chunk of code that was slow or code that I thought might be slow.
Also, if you use online performance tools it is not necessary to change your code. The online tool runs against an instance of the app that has been deployed.
When writing specs for a simple Rails app, is the following a correct approach for full test coverage?
Write feature specs for all user stories
Write controller specs to ensure that individual action responses are correct and all required variables are set
Write model specs to ensure all methods, validations,e tc. are working as intended
Write mailer specs
Write routing specs
Is this enough, too much (e.g. can I skip some lower-level specs if I've written feature specs), or not enough? Why?
You don't need to write specs for every object in every layer either to get 100% test coverage or to test-drive (require you to implement) all of the important behavior in your application. Instead, as behavior-driven development (BDD) advises, write specs outside in, and write lower-level specs only as necessary.
The most important measure of test completeness is requirement coverage: it's helpful for each user story, and each detail of each story that requires new code, to be represented in at least one test. If you're following typical agile practices (mentioning user stories suggests that you are) your tests are probably the only place where you record your requirements, so you probably can't put a number on this kind of coverage. It's also helpful to have
line coverage (what most people mean when they say test coverage), meaning that every line of code is exercised by at least one test, and
integration coverage, meaning that every method call from one class to another is exercised by at least one test.
For each story,
Write only the feature specs that will test-drive all of the story's distinct happy paths.
Write additional feature specs to ensure integration coverage of architecturally interesting minor variations of happy paths and of sad paths. For example, I often write three feature specs for a story that involves a form: one where the user fills in every possible field and succeeds, another where the user fills in as little information as possible and still succeeds (ensuring that unspecified values and defaults work as intended), and one where the user makes a mistake, fails, corrects the mistake and succeeds.
At this point you've already test-driven every layer (controllers, models, views, helpers, mailers, etc.) into existence, with only feature specs.
Write model and helper specs to drive out detailed requirements which live entirely in those classes. For example, once you've written a single sad-path feature spec that establishes that entering one particular invalid attribute sends the user back to edit their form submission and displays a message, you can handle other invalid attributes entirely by writing more examples in that model's spec that test that model attributes are validated, and let the architecture that you've already test-driven propagate the errors back to the user.
Note that although your feature specs already test the happy paths through model and helper methods, as soon as you start writing examples for a method for minor or error cases, you'll probably want to write the happy-path example or examples for that method too, so you can see the entire description of the method in one place, and so you can test the method fully just by running all its examples and not also have to run any feature specs.
You might not need some kinds of specs at all:
Well-factored controller actions are short and have few or no conditionals, so you often won't need any controller specs at all. Write them only when needed, and stub out model, mailer, etc. behavior to keep them simple and fast.
Similarly, views and mailers should have few or no conditionals (complex code should be refactored into helper and model methods), so you often won't need view or mailer specs at all.
Your feature specs will have test-driven all the routes you need, so you probably won't need routing specs. I've only ever gotten use out of routing specs when I had to do a major refactor of routes, as when upgrading from one major version of Rails to the next.
As long as you always write a test before you write new code, you'll always have 100% line coverage.
That testing strategy sounds really comprehensive. If you had all of these tests in place you would have great test coverage. However it would take you longer to deliver your project. You would also not be agile as someone who is doing more limited testing. Testing has to suit the project. Don't over test. Over testing can cost time and money. Don't under test. Under testing can cost time and money.
There are right ways to do unit testing. There are right ways to do integration testing. The glove has to fit. If your application is largely front end facing then perhaps it's best to start with integration tests. If your writing a back end application or perhaps an API then unit tests maybe a better place to start. I think approaching with one style of testing and then expanding to different styles is a better start than to try and test every layer of your application.
Why not start with simple unit tests? They are easy to write. Write these tests and then track how many bugs you ship. Are you letting in too many bugs? Are you having a lot of regression issues? Are there bugs that are getting through to production that your suite is not picking up? If the answer is yes then maybe it's time to write some higher level tests. Remember the higher level a test is the more development cost you will have to pay.
If your not shipping bugs then you have no reason to write any more tests. Remember the end goal here. We want to ship bug free code. If we can write one test and one test alone that will ensure we are doing this then there is no reason to test any further.
It seems, that there are two totally different approaches to testing, and I would like to cite both of them.
The thing is, that those opinions were stated 5 years ago (2007), and I am interested, what has changed since then and which way should I go.
Brandon Keepers:
The theory is that tests are supposed to be agnostic of the
implementation. This leads to less brittle tests and actually tests
the outcome (or behavior).
With RSpec, I feel like the common approach of completely mocking your
models to test your controllers ends up forcing you to look too much
into the implementation of your controller.
This by itself is not too bad, but the problem is that it peers too
much into the controller to dictate how the model is used. Why does it
matter if my controller calls Thing.new? What if my controller decides
to take the Thing.create! and rescue route? What if my model has a
special initializer method, like Thing.build_with_foo? My spec for
behavior should not fail if I change the implementation.
This problem gets even worse when you have nested resources and are
creating multiple models per controller. Some of my setup methods end
up being 15 or more lines long and VERY fragile.
RSpec’s intention is to completely isolate your controller logic from
your models, which sounds good in theory, but almost runs against the
grain for an integrated stack like Rails. Especially if you practice
the skinny controller/fat model discipline, the amount of logic in the
controller becomes very small, and the setup becomes huge.
So what’s a BDD-wannabe to do? Taking a step back, the behavior that I
really want to test is not that my controller calls Thing.new, but
that given parameters X, it creates a new thing and redirects to it.
David Chelimsky:
It’s all about trade-offs.
The fact that AR chooses inheritance rather than delegation puts us in
a testing bind – we have to be coupled to the database OR we have to
be more intimate with the implementation. We accept this design choice
because we reap benefits in expressiveness and DRY-ness.
In grappling with the dilemma, I chose faster tests at the cost of
slightly more brittle. You’re choosing less brittle tests at the cost
of them running slightly slower. It’s a trade-off either way.
In practice, I run the tests hundreds, if not thousands, of times a
day (I use autotest and take very granular steps) and I change whether
I use “new” or “create” almost never. Also due to granular steps, new
models that appear are quite volatile at first. The valid_thing_attrs
approach minimizes the pain from this a bit, but it still means that
every new required field means that I have to change
valid_thing_attrs.
But if your approach is working for you in practice, then its good! In
fact, I’d strongly recommend that you publish a plugin with generators
that produce the examples the way you like them. I’m sure that a lot
of people would benefit from that.
Ryan Bates:
Out of curiosity, how often do you use mocks in your tests/specs?
Perhaps I'm doing something wrong, but I'm finding it severely
limiting. Since switching to rSpec over a month ago, I've been doing
what they recommend in the docs where the controller and view layers
do not hit the database at all and the models are completely mocked
out. This gives you a nice speed boost and makes some things easier,
but I'm finding the cons of doing this far outweigh the pros. Since
using mocks, my specs have turned into a maintenance nightmare. Specs
are meant to test the behavior, not the implementation. I don't care
if a method was called I just want to make sure the resulting output
is correct. Because mocking makes specs picky about the
implementation, it makes simple refactorings (that don't change the
behavior) impossible to do without having to constantly go back and
"fix" the specs. I'm very opinionated about what a spec/tests should
cover. A test should only break when the app breaks. This is one
reason why I hardly test the view layer because I find it too rigid.
It often leads to tests breaking without the app breaking when
changing little things in the view. I'm finding the same problem with
mocks. On top of all this, I just realized today that mocking/stubbing
a class method (sometimes) sticks around between specs. Specs should
be self contained and not influenced by other specs. This breaks that
rule and leads to tricky bugs. What have I learned from all this? Be
careful where you use mocking. Stubbing is not as bad, but still has
some of the same issues.
I took the past few hours and removed nearly all mocks from my specs.
I also merged the controller and view specs into one using
"integrate_views" in the controller spec. I am also loading all
fixtures for each controller spec so there's some test data to fill
the views. The end result? My specs are shorter, simpler, more
consistent, less rigid, and they test the entire stack together
(model, view, controller) so no bugs can slip through the cracks. I'm
not saying this is the "right" way for everyone. If your project
requires a very strict spec case then it may not be for you, but in my
case this is worlds better than what I had before using mocks. I still
think stubbing is a good solution in a few spots so I'm still doing
that.
I think all three opinions are still completely valid. Ryan and I were struggling with the maintainability of mocking, while David felt the maintenance tradeoff was worth it for the increase in speed.
But these tradeoffs are symptoms of a deeper problem, which David alluded to in 2007: ActiveRecord. The design of ActiveRecord encourages you to create god objects that do too much, know too much about the rest of the system, and have too much surface area. This leads to tests that have too much to test, know too much about the rest of the system, and are either too slow or brittle.
So what's the solution? Separate as much of your application from the framework as possible. Write lots of small classes that model your domain and don't inherit from anything. Each object should have limited surface area (no more than a few methods) and explicit dependencies passed in through the constructor.
With this approach, I've only been writing two types of tests: isolated unit tests, and full-stack system tests. In the isolation tests, I mock or stub everything that is not the object under test. These tests are insanely fast and often don't even require loading the whole Rails environment. The full stack tests exercise the whole system. They are painfully slow and give useless feedback when they fail. I write as few as necessary, but enough to give me confidence that all my well-tested objects integrate well.
Unfortunately, I can't point you to an example project that does this well (yet). I talk a little about it in my presentation on Why Our Code Smells, watch Corey Haines' presentation on Fast Rails Tests, and I highly recommend reading Growing Object Oriented Software Guided by Tests.
Thanks for compiling the quotes from 2007. It is fun to look back.
My current testing approach is covered in this RailsCasts episode which I have been quite happy with. In summary I have two levels of tests.
High level: I use request specs in RSpec, Capybara, and VCR. Tests can be flagged to execute JavaScript as necessary. Mocking is avoided here because the goal is to test the entire stack. Each controller action is tested at least once, maybe a few times.
Low level: This is where all complex logic is tested - primarily models and helpers. I avoid mocking here as well. The tests hit the database or surrounding objects when necessary.
Notice there are no controller or view specs. I feel these are adequately covered in request specs.
Since there is little mocking, how do I keep the tests fast? Here are some tips.
Avoid excessive branching logic in the high level tests. Any complex logic should be moved to the lower level.
When generating records (such as with Factory Girl), use build first and only switch to create when necessary.
Use Guard with Spork to skip the Rails startup time. The relevant tests are often done within a few seconds after saving the file. Use a :focus tag in RSpec to limit which tests run when working on a specific area. If it's a large test suite, set all_after_pass: false, all_on_start: false in the Guardfile to only run them all when needed.
I use multiple assertions per test. Executing the same setup code for each assertion will greatly increase the test time. RSpec will print out the line that failed so it is easy to locate it.
I find mocking adds brittleness to the tests which is why I avoid it. True, it can be great as an aid for OO design, but in the structure of a Rails app this doesn't feel as effective. Instead I rely heavily on refactoring and let the code itself tell me how the design should go.
This approach works best on small-medium size Rails applications without extensive, complex domain logic.
Great questions and great discussion. #ryanb and #bkeepers mention that they only write two types of tests. I take a similar approach, but have a third type of test:
Unit tests: isolated tests, typically, but not always, against plain ruby objects. My unit tests don't involve the DB, 3rd party API calls, or any other external stuff.
Integration tests: these are still focused on testing one class; the differences is that they integrate that class with the external stuff I avoid in my unit tests. My models will often have both unit tests and integration tests, where the unit tests focus in the pure logic that can be tested w/o involving the DB, and the integration tests will involve the DB. In addition, I tend to test 3rd party API wrappers with integration tests, using VCR to keep the tests fast and deterministic, but letting my CI builds make the HTTP requests for real (to catch any API changes).
Acceptance tests: end-to-end tests, for an entire feature. This isn't just about UI testing via capybara; I do the same in my gems, which may not have an HTML UI at all. In those cases, this exercises whatever the gem does end-to-end. I also tend to use VCR in these tests (if they make external HTTP requests), and like in my integration tests, my CI build is setup to make the HTTP requests for real.
As far as mocking goes, I don't have a "one size fits all" approach. I've definitely overmocked in the past, but I still find it to be a very useful technique, especially when using something like rspec-fire. In general, I mock collaborators playing roles freely (particularly if I own them, and they are service objects) and try to avoid it in most other cases.
Probably the biggest change to my testing over the last year or so has been inspired by DAS: whereas I used to have a spec_helper.rb that loads the entire environment, now I explicitly load just the class-under test (and any dependencies). Besides the improved test speed (which does make a huge difference!) it helps me identify when my class-under-test is pulling in too many dependencies.
I've been writing tests for a while now and I'm starting to get the hang of things. But I've got some questions concerning how much test coverage is really necessary. The consensus seems pretty clear: more coverage is always better. But, from a beginner's perspective at least, I wonder if this is really true.
Take this totally vanilla controller action for example:
def create
#event = Event.new(params[:event])
if #event.save
flash[:notice] = "Event successfully created."
redirect_to events_path
else
render :action => 'new'
end
end
Just the generated scaffolding. We're not doing anything unusual here. Why is it important to write controller tests for this action? After all, we didn't even write the code - the generator did the work for us. Unless there's a bug in rails, this code should be fine. It seems like testing this action is not all too different from testing, say, collection_select - and we wouldn't do that. Furthermore, assuming we're using cucumber, we should already have the basics covered (e.g. where it redirects).
The same could even be said for simple model methods. For example:
def full_name
"#{first_name} #{last_name}"
end
Do we really need to write tests for such simple methods? If there's a syntax error, you'll catch it on page refresh. Likewise, cucumber would catch this so long as your features hit any page that called the full_name method. Obviously, we shouldn't be relying on cucumber for anything too complex. But does full_name really need a unit test?
You might say that because the code is simple the test will also be simple. So you might as well write a test since it's only going to take a minute. But it seems that writing essentially worthless tests can do more harm than good. For example, they clutter up your specs making it more difficult to focus on the complex tests that actually matter. Also, they take time to run (although probably not much).
But, like I said, I'm hardly an expert tester. I'm not necessarily advocating less test coverage. Rather, I'm looking for some expert advice. Is there actually a good reason to be writing such simple tests?
My experience in this is that you shouldn't waste your time writing tests for code that is trivial, unless you have a lot of complex stuff riding on the correctness of that triviality. I, for one, think that testing stuff like getters and setters is a total waste of time, but I'm sure that there'll be more than one coverage junkie out there who'll be willing to oppose me on this.
For me tests facilitate three things:
They garantuee unbroken old functionality If I can check that
nothing new that I put in has broken
my old things by running tests, it's
a good thing.
They make me feel secure when I rewrite old stuff The code I
refactor is very rarely the trivial
one. If, however, I want to refactor
untrivial code, having tests to
ensure that my refactorings have not
broken any behavior is a must.
They are the documentation of my work Untrivial code needs to be
documented. If, however, you agree
with me that comments in code is the
work of the devil, having clear and
concise unit tests that make you
understand what the correct behavior
of something is, is (again) a must.
Anything I'm sure I won't break, or that I feel is unnessecary to document, I simply don't waste time testing. Your generated controllers and model methods, then, I would say are all fine even without unit tests.
The only absolute rule is that testing should be cost-efficient.
Any set of practical guidelines to achieve that will be controversial, but here are some advices to avoid tests that will be generally wasteful, or do more harm than good.
Unit
Don't test private methods directly, only assess their effects indirectly through the public methods that call them.
Don't test internal states
Only test non-trivial methods, where different contexts may get different results (calculations, concatenation, regexes, branches...)
Don't assess things you don't care about, e.g. full copy on some message or useless parts of complex data structures returned by an API...
Stub all the things in unit tests, they're called unit tests because you're only testing one class, not its collaborators. With stubs/spies, you test the messages you send them without testing their internal logic.
Consider private nested classes as private methods
Integration
Don't try to test all the combinations in integration tests. That's what unit tests are for. Just test happy-paths or most common cases.
Don't use Cucumber unless you really BDD
Integration tests don't always need to run in the browser. To test more cases with less of a performance hit you can have some integration tests interact directly with model classes.
Don't test what you don't own. Integration tests should expect third-party dependencies to do their job, but not substitute to their own test suite.
Controller
In controller tests, only test controller logic: Redirections, authentication, permissions, HTTP status. Stub the business logic. Consider filters, etc. like private methods in unit tests, tested through public controller actions only.
Others
Don't write route tests, except if you're writing an API, for the endpoints not already covered by integration tests.
Don't write view tests. You should be able to change copy or HTML classes without breaking your tests. Just assess critical view elements as part of your in-browser integration tests.
Do test your client JS, especially if it holds some application logic. All those rules also apply to JS tests.
Ignore any of those rules for business-critical stuff, or when something actually breaks (no-one wants to explain their boss/users why the same bug happened twice, that's why you should probably write at least regression tests when fixing a bug).
See more details on that post.
More coverage is better for code quality- but it costs more. There's a sliding scale here, if you're coding an artificial heart, you need more tests. The less you pay upfront, the more likely it is you'll pay later, maybe painfully.
In the example, full_name, why have you placed a space between, and ordered by first_name then last_name- does that matter? If you are later asked to sort by last name, is it ok to swap the order and add a comma? What if the last name is two words- will that additional space affect things? Maybe you also have an xml feed someone else is parsing? If you're not sure what to test, for a simple undocumented function, maybe think about the functionality implied by the method name.
I would think your company's culture is important to consider too. If you're doing more than others, then you're really wasting time. Doesn't help to have a well tested footer, if the main content is buggy. Causing the main build or other developer's builds to break, would be worse though. Finding the balance is hard- unless one is the decider, spend some time reading the test code written by other team members.
Some people take the approach of testing the edge cases, and assume the main features will get worked out through usage. Considering getter/setters, I'd want a model class somewhere, that has a few tests on those methods, maybe test the database column type ranges. This at least tells me the network is ok, a database connection can be made, I have access to write to a table that exists, etc. Pages come and go, so don't consider a page load to be a substitute for an actual unit test. (A testing efficiency side note- if having automated testing based on the file update timestamp (autotest), that test wouldn't run, and you want to know asap)
I'd prefer to have better quality tests, rather than full coverage. But I'd also want an automated tool pointing out what isn't tested. If it's not tested, I assume it's broken. As you find failure, add tests, even if it's simple code.
If you are automating your testing, it doesn't matter how long it takes to run. You benefit every time that test code is run- at that point, you know a minimum of your code's functionality is working, and you get a sense of how reliable the tested functionality has been over time.
100% coverage shouldn't be your goal- good testing should be. It would be misleading to think a single test of a regular expression was accomplishing anything. I'd rather have no tests than one, because my automated coverage report reminds me the RE is unreliable.
The primary benefit you would get from writing a unit test or two for this method would be regression testing. If, sometime in the future, something was changed that impacted this method negatively, you would be able to catch it.
Whether or not that's worth the effort is ultimately up to you.
The secondary benefit I can see by looking at it would be testing edge cases, like, what it should do if last_name is "" or nil. That can reveal unexpected behavior.
(i.e. if last_name is nil, and first_name is "John", you get full_name => "John ")
Again, the cost-vs-benefit is ultimately up to you.
For generated code, no, there's no need to have test coverage there because, as you said, you didn't write it. If there's a problem, it's beyond the scope of the tests, which should be focused on your project. Likewise, you probably wouldn't need to explicitly test any libraries that you use.
For your particular method, it looks like that's the equivalent of a setter (it's been a bit since I've done Ruby on Rails) - testing that method would be testing the language features. If you were changing values or generating output, then you should have a test. But if you are just setting values or returning something with no computation or logic, I don't see the benefit to having tests cover those methods as if they are wrong, you should be able to detect the problem in a visual inspection or the problem is a language defect.
As far as the other methods, if you write them, you should probably have a test for them. In Test-Driven Development, this is essential as the tests for a particular method exist before the method does and you write methods to make the test pass. If you aren't writing your tests first, then you still get some benefit to have at least a simple test in place should this method ever change.
I'm currently looking at a hefty Rails test suite. It's nothing I can get into specifics about, but the run time for the entire suite (unit/functional/some integration) can run upward of 5 minutes.
We're completely reliant on fixtures and are we're not mocking and stubbing as much as we should be.
Our next few sprints are going to be completely focused on the test suite, both improving coverage, writing better tests and most importantly writing more efficient tests.
So aside from more mocking and stubbing within our tests, we're considering replacing our fixtures with most likely Factory Girl. I see a lot of happy folks doing similar situations but haven't been able to find a good resource on any minuses of moving to a factory. I have seen some slower benchmarks when using benchmarks from various resources but cannot find a definitive this why factories are good and this is why you might not want to use them.
Can anyone educate me on why or why I shouldn't be using factories?
Thanks!
Oleg's answer is great, but let me offer the perspective of someone who is using both.
Fixtures have sort of been the whipping boy of the Rails community for a while. Everyone understands the drawbacks of fixtures, but no one is really championing their strengths. In my experience, factories by themselves can easily become just as difficult to maintain as fixtures (it really depends on the schema, but I digress). The real strength of factories is in selective replacement of fixture-based pain. Let's talk about a couple specifics.
The first issue is performance. If you can test most of your app without hitting the database then you will see a significant speed up, but for most applications I don't think it's wise to test without hitting the database entirely. At some point you want to test the whole stack. Every time you mock or stub you are making an assumption about an interface that may contain subtle bugs. So, assuming that you need to hit the database on some significant percentage of tests, transactional fixtures (you are using transactional fixtures right?) could well be much much faster than instantiating a whole environment for every test.
I'd say, with the size of your test suite that you really need to look towards Continuous Integration to scale your development to the next level. No matter how much you speed them up, it's still a long time for developers to wait. Maybe look at autotest as well to help at the individual level. But ultimately CI is going to allow you to maintain testing discipline without sacrificing developer agility.
The place where fixtures really shine is in functional/integration testing. The way I look at it is that the fixtures should set up a healthy base state for the app to be tested. Most unit tests don't really need this. You can get very good unit coverage using factories. However when it comes to functional testing, any given page may be hitting dozens of models. I don't want to set up all that stuff in each test. As I construct ever more complex scenarios, I'm getting closer and closer to recreating a global data state which is exactly what fixtures were designed to do in the first place.
One controversial belief I hold is that all else being equal, I prefer one functional test to 20 unit tests (using Rails parlance). Why? Because the functional test proves that the end result that is sent to the user is correct. The unit tests are great for getting at nuances of functionality, but at the end of the day, you could still have a bug along an interface that breaks your entire site. Functional tests are what give me the confidence hitting deploy without actually loading up the page in my browser. I know that I could stub everything out and test both interfaces and get the same coverage, but if I can test the whole stack in one simple test at the expense of a little CPU, I'd much rather do that.
So what are my best practices for fixtures?
Set up a handful for every model to cover the broadest categories of data
When adding a major new feature that cuts across many models and controllers, add some new fixtures to represent the major states
Avoid editing old fixtures except for adding/removing fields
Use factories for more smaller/more localized variations
Use factories for testing pagination or other mass creation that is only needed for a few tests
Also, let me recommend Jay Fields' blog for really good pragmatic testing advice. The thing I like most about Jay's blog is that he always acknowledges that testing is very project-specific, and what works for one project does not necessarily work for another. He's short on dogma and long on pragmatism.
There could be some issues with setting up all dependencies between entities for good test suite. Anyway, it's still much easier than maintaing a lot of fixtures.
Fixtures:
hard to maintain relationships (especially many-to-many);
test suite runtime is usually slower due more DB hits;
tests are very sensitive to changes in schema.
Factories:
you stub everything you don't test at current unit test;
you prepare entities that you are testing with factories. This is where factories show their real advantage — it's easy to set up new test cases, as you don't need to maintain a ton of YAML-files for that;
you concentrate on testing. If tests require changing scenario, you don't shift your mindset. As long as stubs are reasonable and factories are easily customized, you should be fine.
So, factories seem a good way to go. The only possible drawbacks I see, are:
time you spent migrating from fixtures;
keeping a sane set of scenarios can require some effort.