Where unit testing fails
hmemcpy.com
hmemcpy.com
Listen kids, testing is a tool for HELPING THE DEVELOPER, not for using to engage in a "more pious than thou" dick-swinging my Cucumber is bigger than yours idiocy.
Testing is about giving YOU THE DEVELOPER useful and quick feedback about if you're on the right path, and if you've broken something, and for warning people who come after you if they've broken something. It's not an arcane methodology that somehow has some magical "making your code better" side-effect...
The whole concept of "test driven development" is hocus, and I speak this as someone who writes a lot of tests, and who charges a lot of money for fixing test suites. Instead: developer-driven testing. Give your developers useful tools for solving problems and supporting themselves, rather than disappearing in to some testing hell where you're doing it a certain way because you're supposed to.
I've worked on numerous projects where TDD has (and continues) to prove invaluable.
So regardless how much experience you've got, I've had experience to the contrary, thus the "whole concept" isn't hocus.
Why have you come to believe this?
Have I had experience (and much value) out of sometimes writing tests for certain problem classes before writing any code? Yes. Changes to existing functionality are often a good candidate.
Does TDD as a methodology suggest you should ALWAYS write your tests first? Everything I've read suggests so.
And "as any fule kno" [sic], this is idioicy during a design or hacking or greenfield phase of development. Allowing your tests to dictate your code (rather than influence the design of modular code) and to dictate your design because you wrote over-invasive test.
tl;dr: Writing tests before code works pretty well in some situations. Test Driven Development, as handed down to us mortals by Agile Testing Experts and other assorted shills, is hocus.
For me, TDD, in practice means switching back and forth between writing tests and production code. That is, write some test code, run, fail, fix the production code, write some more test code. It is incremental.
EDIT: Fixed a typo
What? So it works in some situations, but not all. Thus hocus?
I'm no 'Agile Testing Expert', but I have worked for the last 2-3 years doing TDD exclusively (as in ALWAYS writing tests first). It isn't hocus. It is effective. I can understand it doesn't suit all problems, and can potentially be a hinderance sometimes, but that is the same for ANY methodology/practice. I don't quite understand your hostility towards it.
What do you mean by 'developer driven testing'?
The hocus is that you must use it, because some guru said so.
My experience is the same as (what I believe is) the usual criticism of tests-first.
I often need to rethink the API to my functions/methods, sometimes more than once. This goes for both external and internal APIs, so it doesn't help with detailed interface specifications between modules.
Test-first is a problem, since it needs a fixed interface before I'm ready to lock down the interface.
I tend to instead write inside-out; a few pages of code and then write tests for that code. That model isn't really kosher/halal/etc, according to TDD?
TDD suggests 3 steps: 1. Write one test 2. Write the minimal code to make it pass 3. Refactor the code you've got.
Try not to think of tests as immutable. You can throw away tests when they don't provide value.
Finally, make sure you're using an IDE with refactoring support. That helps immensely when you make an API change which will affect your tests.
Interesting TDD argument. I'll buy a book or something and try again when my present hobby project is done and I have time for experimenting.
(IDE refactoring needs a stricter language than I prefer. Also, without an IDE and using my old eyes I have 2 A4 of code + 1 A4 of bash on a 24" monitor...)
The examples are all Java, but the accompanying website (http://www.growing-object-oriented-software.com/) has links to reimplementations in several other languages.
The point of using TDD to write interfaces is that you first write the code you wish you had, and then implement it. The other way around you could end up with code that is optimized for the code inside the module, and not optimized for using it.
Of course now you have to rewrite your tests and your code when changing an API, but at least you know that no other part of your application got broken by the new API.
I personally like the TDD approach to write interfaces, since then I won't take shortcuts just because it is easier to do in the specific module that I'm programming/extending. It will actually produce code that can be used easily from the outside and removes a lot of dependencies in the process.
Then why do it exclusively?
Testing is about helping developers, and recognizing that automated testing is about benefit to developers, rather than cargo-culting a workflow and decreeing that one size fits all.
Writing tests first as a tool to be deployed where it works is "Developer Driven Testing" - focussing on making the developer more productive by choosing the right tool for the job. Generalizing a bunch of testing rules and saying This Is The One True Way Even When It Isn't (I mean seriously, you continued to use it when it was a hindrance?!) - that's not right.
Sounds like you are interpreting TDD in a rather extreme way. If you haven't found it useful, then fine don't use it. But that doesn't mean it is universally bad, and thus doesn't deserve the "It's all hocus and you are a stupid zealot if you do it"
Making tests a central part of the process because they're useful to developers? Awesome. Dictating a workflow to developers that works in some cases as the One True Way: ridiculous.
I tend to agree with you that TDD as a development methodology is deeply flawed. I have never found that the act of writing a test case and making it pass has helped me with the actual creative process of solving a problem via code. In fact, I'll frequently dash out a method I'm writing quickly, and then go back and reconstruct it piece by piece via TDD.
The main benefit writing tests first gives you is that you know your test is actually working (i.e. because it first fails) and that the code you implement actually passes it. It also keeps you honest because it prevents you from writing "extraneous" code, i.e. code that exists but is not specified by a test.
So yeah, I totally agree with you that the cult of TDD kind of sucks. But I haven't stumbled upon a better way yet. I've often thought it would be cool if there was some way for e.g. your version control software to be tied to the code you write that passes certain test(s), such that you could somehow verify that without that code your test fails and with that code your test passes. For the present though, TDD tends to work pretty well for me.
Labouring under the idea that Tests Must Come First (and everything I've seen, and everything I /do/ see now suggests that that is the central idea in TDD - you write a test, then you write the code to pass it) without pivoting to see that testing is a useful practice in so much as it helps developers is the wrong approach.
The problem is that ideologies like TDD try to come up with simple rules for how to develop (i.e. always write tests first). In reality, it's better to use your brain when you're deciding how to approach something (i.e. is this a good situation for writing tests first?)
I often code three or four commits worth at once when on a roll and it's a pain doing multiple staged adds, stashes, tests, and commits to store each sub-change on its own. Having a tool to help with this would be great.
When you write something you run it to make sure it works. Bottle that and call it a test. And then merely do this before you write the code. "What will I be done when I get this code to do?"
Don't plan, and spec, your whole API before you've written a line of code or anything, but do write a one-liner to check whatever it is you're going to do so you know when you're done.
Ideally you'll write a test (still just a one-liner) for every corner-case that you cover in code, because you need a failing test to know if your fix actually worked.
And face it, if you're doing "real" development work you've got business goals, and a spec with features intended to deliver on that, and you need to cover every tiny detail. If you don't have the spec ('fail on values over n') directly linked to the code that implements it, it will get lost, and if you aren't testing it you can't legitimately claim to be delivering it.
A spec, such as you'd get by performing useful TDD, is a requirement for modern software development. You could write it all up-front in the planning phase (waterfall style), or all after in the testing phase (waterfall style again), or you could interleave it with research and development, which if you think about it, sounds kind of agile by comparison...
For example lets take a simple newsletter signup form with only a field to enter an email address.
Now you shouldn't write a test to see what happens when you enter an email, a wrong email, if the confirmation email is sent, etc. Start with the simplest case: "If I fill in my email, and I press klick, then it should show me a confirmation that I singed up." Not more. Now make that pass, and you have working code!
Now you can add tests that verify that a wrong or empty email is rejected, now make that pass. More working code!
With TDD you should never end up with a lot of tests and no code that solves the problem. In that case you are actually writing too many tests and not enough code.
This is useful because you're constantly making and checking hypotheses: "This code should make a test fail in X way... yup. Okay, now this code should make the test pass. Yup. Now this change shouldn't change behavior, and the tests should still pass. Oops! Undo. Oh, that method has to be called in the constructor. All right, this change should work. Good."
(If you're interested, my book goes into more detail, and it's online: http://jamesshore.com/Agile-Book/test_driven_development.htm... . I also have a extensive screencast of real-world TDD here: http://jamesshore.com/Blog/Lets-Play/ )
Despite the extremists, TDD really is a useful tool--and that's all it is. Sure, TDD has been turned into yet another vehicle for programmers to engage in religious wars, but you can't really blame TDD for that, any more than you can blame Richard Stallman for emacs vs. vi wars.
Ravi's article: http://ravimohan.blogspot.com/2007/04/learning-from-sudoku-s...
Peter Norvig's solver: http://norvig.com/sudoku.html
Ron Jeffries' attempts:
http://xprogramming.com/articles/sudokumusings/
http://xprogramming.com/articles/oksudoku/
http://xprogramming.com/articles/sudoku2
http://xprogramming.com/articles/sudoku4
http://xprogramming.com/articles/sudoku5
And, as dessert (Ron is very frank about his failures):
http://xprogramming.com/articles/roroncemore/
"This is surely the most ignominious debacle of a project listed on my site, even though others have also not shipped. (Sudoku did not ship and will not. Shotgun will go forward if the Customer wants to.)"
Achieving this, in my experience, comes down to disciplined used of mocks and stubs. I've seen people lean too heavily on them, or not heavily enough. The biggest problem I've seen, particularly common in the Java world due to jMock's datedness, is over specifying expectations on mocks. Stubs which can provide canned responses (whether through a framework or manually rolled) are absolutely and totally the most underused yet useful tool in your testing arsenal.
Perhaps I'm doing it wrong, but I've found in my experience that bugs I miss in testing increase proportionally with my use of mocks or stubs. Invariably, when I'm forced to stub something and replicate its behavior I miss something subtle that comes back to bite me.
First, people rely too heavily on mocks/stubs/fakes/(whatever you want to call them). This has gotten better in the past 4 or so years (in my mind, largely because of Rails and it being "acceptable" to hit a DB in a test - though it might actually predate Rails). I think this problem is pretty straightforward to understand (again, especially when you look at the Rails way to test a DB interaction versus a more "traditional" way).
However, there are instances where mocks/fakes/stubs are important. Some outside dependencies might not be accessible during testing, might not be predictable/deterministic, might be too slow or might simply involve too much setup. Also, solely relying on "real" implementations can also make your tests too brittle. Why should the logic for can_legally_drink? break if the DB column is changed from DOB to dateofbirth (ok, that's an extreme and poor example, but you get the idea, hopefully).
Anyways, assuming you agree that sometimes a fake is simply better, you get into mocks vs stubs. To me a stub is dumb and a mock is strict. Stubs also automatically reply with canned answers, mocks don't. A mock is used to assert that a certain expected call was made. A stub is used just to get your test to move over a line of code.
The problem is that people use mocks over and over again. Re-specifying the same expectations in every test..making X tests break when you change the behavior of the interaction, versus just 1 test. As semi-pseudocode (and yes, it might be better to just hit the DB in this case):
def self.login_user(name, password)
user = Store.get_user_by_name(name)
return user.nil? || user.password != password ? nil : user
end
it "gets the user from the store" do
Store.should_receive(:get_user_by_name).with('leto')
User.login_user ('leto', 'ghanima')
end
it "returns nil if the passwords don't match" do
Store.stub!(:get_user_by_name).and_return(User.new)
User.login_user('leto', 'ghanima').should be_nil
end
Some framework even return smart canned answers. So instead of returning nil, they'll default to a new instance of the type, or an empty array, or some default scalar value (like an empty string). This is particularly useful when you just want to get passed a null reference (which is exceedingly common).
Does that make sense?
I can't seem to find a blog post discussing the merits, but IME strict mocking leads to a lot of tests breaking just because you added one more piece of functionality to a method. Now, this causes you to edit all sorts of existing tests to cause them to ignore a call to collaborator. While you can extract common expectations and such to eliminate duplication, you still end up with many tests failing due to one small change in behavior.
Normally your tests tend to assert one or two behaviors per test. With strict mocking, your tests become awkwardly cluttered by setting up expectations/interactions for things that are unrelated to what they are asserting.
With nice mocking, it's much easier to get away with only the expectations/interactions that are directly relevant to what a single test is asserting. This also means that you get more fine-grained failures, which speeds up diagnostics.
Don't get me wrong, I'm sure that all these guys and chicks that live and die by TDD are smart (probably smarter than me), but what they are building is starting to get more and more out of touch with real-world requirements (I wanted to say somth about how this resembles a schizophrenic world, but for the moment I'll stop short of that).
My father was a civil engineer, at a time when they didn't have AutoCAD available and all that fancy computer stuff. As he elegantly put it: "if I mess things up in my work people will die, either in 2 weeks or in 20 years, when the next big earthquake will hit us". Well, I never saw him "testing" building apartment-blocks, or industrial buildings, or roads, he made sure that what was on the sheet of papers where the construction plans were drawn would get built as accurate as possible in real life. And keep in mind that he was managing construction workers, many of them close to illiterate, even former convicts, and not CS-graduates. That's the job of all civil engineers from all over the world. So, if they can do it, why in the name of God do we programmers write tests for mundane stuff like sudoku-solvers? Something is not right.
At the point where you're writing tests to impress other programmers (see article) about how much of a 'craftsman' you are, you lose. When you're writing tests that help you as a developer, and help get the software out more quickly and with higher quality, then you're doing it right.
But to me that's not the whole purpose of unit tests - the most important part for me is that it tells you very quickly when you have bugs in your implementation. Somehow the author neglects to address this very central point.
Don't just write all the test for every little function in the beginning thats stupid, How do you even know what helper function you need to implment.
It is not a coincidence that one of the reasons TDD'ers love rspec so much is that the "code reads like documentation" - it's even called r spec.
Perhaps that is ok for mature projects. But I often hear TDD and agile used together as if they weren't in conflict and I get confused. Then again, I've only ever worked on new projects.
Maybe I'm missing something?
That's an interesting question - would a unit test prevent this error? To answer this, let's consider what happened: the database didn't respond in an appropriate way, and probably returned an error. I'm not exactly sure what exact error was displayed, but WordPress has caught this situation, and handled it the best it could - displaying a proper error. I would write a unit test for this exact scenario - using a fake (mock) object, simulate the database timeout, and make sure that the error was caught and handled appropriately. Other then that there's no real other way to gracefully recover.
When writing code it is imperative to to get great feedback, as often as possible. In a number of cases, TDD helps you do that. In cases where it don't, well, it doesn't, so don't use it.
Testing doesn't help you come up with good algorithms. It can help you with not messing up an implementation, though.