I try to start with just one or two tests to actually help do things that are tedious or require multiple steps. It takes some time to automate a good test but once you do it immediately starts saving time because you don't have to run the same sequence a thousand times while developing. You can think of it more like a macro that saves you time.

Once you write the main test it's easy then to run it with all combinations of good and bad input. By doing that you'll often wind up hitting a pretty good percentage of your code.

Then as bugs are discovered due to unexpected input you can just keeping adding more input situations.

Smart automated testing just takes all that extra test work you're already doing and saves it as you go along.

No need to try to invent extra things to test. You just test what you would have tested anyway by hand.

I used to work at a company where automated testing was sending you emails about what your commits broke. It does help at improving code quality.

1. Write some code

2. Test it

3. Debug your code

4. Test it

5. Debug your code

6. Test it

...

39. Debug your code

40. Test it; now it finally seems to work

If all your "test it" steps are being done manually, you're being very inefficient. A good unit test can actually make development go by faster with the added bonus of defending your code from changes down the road that might screw things up.

TDD isn't "THE" way, but test coverage helps. It's not fun (at least not for me), but it's less aggravating than breaking something 6 months down the road in some non-obvious way. I'm human, so I assume I'll screw something up eventually. Having test coverage helps keep me from shooting myself in the foot later.

You should often be able to at least create an automated acceptance test for what you're doing (e.g., "as a user I want to click this button and see XYZ"). This is usually extremely decoupled from the implementation so it should survive refactoring. So then do your exploratory code, get the test passing, and then refactor, introducing lower-level tests.

If that doesn't seem doable you might be taking on a task that doesn't have a good set of requirements. Writing code without any concrete use case in mind is fun and all but that kind of code should usually be limited to a prototyping sandbox.

Consider (for instance) a program to translate ABC music format to proper sheet music. It's easy to say the basic requirement: "The program has to accurately translate ABC format to legible and easy to read sheet music." But even a start at automating a test for that would require converting a graphical representation of sheet music back to the basic notes, and that problem is at least an order of magnitude harder than writing the original program, without factoring in the "easy to read" bit at all. (PS This is a real issue for me, if someone knows of a decent open source sheet music to MIDI/ABC convertor I'd love to hear about it.)

A unit test for a piece of code like this might be "Given that the time signature for the music is 3:4, the software puts 3:4 in the appropriate place on each line of output".

You then might write a variety of cases testing that it deals correctly with (say) a changing time signature at some point in the piece.

The upside of this is when you try and fix another bug which has a knock on effect on this bit of code, lots of your tests are going to fail- immediately identifying where the problem is (or at least letting you know there is one!)

But okay, is it really so much easier to write a test which just tests changing the time signature? It's still going to require doing OCR. And if I'm really testing it, I've got to make sure the time signature change comes at the correct point, which requires OCR on the notes around it. Also the bar lines (to makes sure the time signature change is reflected in the notes per bar and not just by writing a new time signature out and otherwise ignoring it).

Now, it's perfectly reasonable to have unit tests that the code correctly recognizes time signature changes in ABC format. (And it looks like I forgot to write them: I see inline key changes in the tests but not inline time signature changes.) But that's only testing half of the problem; and it's the easier half by far.

Don't you still have the same problem colomon described here, though? Testing your stated condition "the software puts 3:4 in the appropriate place on each line of output" still implies some form of image recognition on the graphical output.

Prototype all you want without tests, but once you've settled on a design & are ready to make it production-ready you should spend time at least writing unit tests for your work or (ideally) take what you've learned & re-apply to a clean design written in a test-first manner.

You might think this is a waste of time, but putting code into production without tests is going to give you more trouble in the long run.

It doesn't slow me down, and I can be sure that my feature is there and works even when somebody refactors our software.

  1. Write a lot of code
  2. Test it
  3. It works!
  4. Test it
  5. It works!
  6. Test it
  7. It works!
  ...
  15. Test it
  16. It works!
  17. Are you done?

TDD works fucking great. If you know what you're doing.

Alas, that's a big IF. Most of the stuff I do, I'm just figuring shit out.

Mostly, like when designing a new library, I work outside-in. I imagine how I'd want to do something, writing the client pseudocodeish stuff first, and then trying to make the implementation support my idealized programming mental model.

I end up throwing away A LOT of code. Getting something short and sweet takes a lot of experimentation, most of which are duds.

Though my personal approach of outside-in is trivially like TDD, it's not nearly as rigorous. Were I to be as thorough as TDD, I'd be spending all my time writing tests. Which seems pointless, for code I'm like just going to throw away.

Anyway. Much respect for the guy who wrote that first TDD book. It's one of the few methodological strategies that works as advertised.

Then, once I've gotten an idea of the problem, I can start writing out some pending tests that help me figure out structure, and then I'll start into the strict TDD loop of write a bit of test, watch it fail, make it pass, write more test, etc.

http://www.amazon.com/Test-Driven-Development-Kent-Beck/dp/0...

That is actually the primary goal of tdd, to free you from the more mundane aspects of the code/run/debug loop. The secondary goal is to give you a good base for changing the code later and finding out what broke, again without a ton of manual actions. But as useful as that is (and it is extremely useful), it doesn't hold a candle to the first benefit.

I do this all the time. Two reasons:

1. I can keep in "the flow" for an extended period of time. This is more important if the code is especially complicated. If I have to stop every few minutes to fix trivial errors, it's easy to forget important details of how everything is supposed to work.

2. Not having any feedback forces you reason about the code before writing it. It's very easy to fall into the trap of writing code, then waiting until it is tested to find the errors. Thinking before writing is the fundamental skill that TDD encourages, but you don't need TDD in order to do it.

> only the people out of college write code for multiple hours straight, then debug everything afterwards.

Knuth wrote TeX in a notebook and did not test it for a good six months afterwards, though I am not aware if he was out of college at the time.

Reading that again, I'm sorry if it came off as sort of attacky, but I really meant that as a "from my personal experience with the people I have worked with over my career" type qualification :)

I can buy #1, but only when it is something you've done a bajillion times before. When you are getting feedback every few minutes, you know exactly what introduced the problem, and don't waste time tracking things down. If you do miss something fundamental and have several hours work behind you, you tend to be more inclined to hack out something to make it work, where if you catch it a few minutes in, you can adjust your design to take it into account. I also find I can keep in the flow pretty easily with constant feedback, and I use simple todo lists to make sure I don't lose track of things.

As for 2, at least for me, I don't think there is any comparison between thinking about how things should work, and knowing if things do work before writing. TDD is definitely not a replacement for deep thought and planning, but I think that is a different beast then working out the details as you are writing them, which is where it comes into play

> only the people out of college write code for multiple hours straight, then debug everything afterwards.

I sort of did it again there, I should have qualified it more :) In my experience, the better programmers I have worked with, paired with, and watched code in videos will get feedback as quickly and often as makes sense, be it with tests or without them. I know if I wrote TeX in a notebook, it would be a guaranteed unmitigated disaster :)

Not all of us have 10 years to work on everything. ;)

Tests can easily do more harm than good if you let them give you a false sense of security.

This is what I find the major advantage of a comprehensive test-suite to be, I don't have to worry as much about breaking any part of the system as a whole - if my suite passes, then I know everything I've worked on so far works, not just the bit I think I changed.

But we must keep in mind that maintaining test code has a cost. Automated testing is not a holy grail and it isn't useful 100% of the time. You should carefully decide what code is worth testing via automation.

Unit tests in general don't catch any regressions, they only help you develop.

Functional tests might be useful, but only if you are testing something that is not likely to change much. E.g.: It is not worth to automate testing the UI if you are going to completely redesign it next month.

Manual testing can actually be cheaper than maintaining test code.

I think this is only in the Rails community, where all new things is quick to be pronounced "the new right way to do things", not just in testing.

> I deal with custom domains, authentication, and 3rd party API calls in almost every app we have. IMO, this adds 20% or more to the (already high) testing overhead just to get these things configured right in the test suite

We do the same (tests for interaction with EC2, Github, and a few other providers). It is more expensive, but we find it more worthwhile too. Normally, 3rd party APIs are insufficiently specified, especially for error conditions. So when we have a failure in production, we can easily add tests to make sure we handle that edge case in future.

Just use Test::Unit and move on with your life. Write some tests. That's what counts.

Imagine the current HTML5 Pandora having bugs with one particular song in Chrome only. How do you test for that using Test::Unit?

I forgot to mention we also use Socket.io for real time push updates to user Backbone models. A good number of bugs don't crop up until another user modifies data and those changes get pushed out to another person.

It's pretty painful to think "oh, this really needs a test, but I haven't got a test suite set up and besides, I don't know to write a test of this kind".

Writing tests for edge cases we see in production is the most valuable thing we do. We use Airbrake to find the bugs, and then we add a test for it, if possible (it's not always possible).

That gives us good confidence that other changes aren't fucking things up. It's also a pretty sane strategy for growing a test suite when you inevitably have some portion of your code which has no tests at all.

This is why regression tests are my favorite type of test. The need for the test has been confirmed by real world usage and once you create the regression test to fail, fix the bug, and pass the test, you won't have to ever worry about users seeing that bug again :)

Good for you. Extremists on all sides are usually wrong.

Shoot for "test MOST OF the things" or "test the MOST IMPORTANT things" or even "test just enough things so that you know if change Y totally breaks MOST IMPORTANT feature Z".

If you have thousands of full stack integration tests that takes an hour to run, you're not going to run them as often as you should be, if at all, ad might as well delete them.

(What's worse is when changes come in faster than the CI system can run the tests. Then you don't know which change broke your build because many changes were tested at once.)

Anything longer than 10 seconds is too long, in my opinion. As soon as you can get up to get coffee while your tests are running, you've lost a lot of productivity. (I recently finished a project where the tests took about ten minutes to run. That meant I could only change code 50 times per day. If the tests had taken 10 seconds, I would have been able to be make 300 changes per day. That's a 6x productivity increase right there.

Fast running test suites are absolutely essential.

It's also nice to be reasonably confident that your commit won't break the build, for which you should probably run a good chunk of the tests before committing.

There are some workflows (Gerrit springs to mind) where you can let the CI server work on your code without breaking anything else, but even then there's a cost to the context switch when a test failure means you have to return to a piece of code you thought you'd finished.

1. Protecting me from stupid mistakes like using wrong variable in parameters, etc. (yes, it is embarrassing to have something like this, but I better be embarrassed by test and fix it before anybody seen it than be embarrassed by somebody else hitting it when using my code).

2. Ensuring refactoring and adding new things didn't break anything.

3. After "hard" bug has been found, ensuring it never reoccurs again.

As for dealing with authentication, etc. - that's what unit tests are for, testing stuff that is under these layers directly. And I don't see it matters what you are using for tests - almost any framework would do fine, it's having tests that matters, not how you run them.

I think you can unit-test javascript too, though I never had to deal with it myself so I don't know how.

This tool is very widely used in Haskell, but it's been ported to a whole bunch of other languages and could make your testing more thorough. In Haskell it's also easy to use and more fun than normal tests, but I don't know what it would be like in a different language.

I feel that way often too but I write test more as a specification for how I want the code to work then as a catch all bugs thing.

"- I deal with custom domains, authentication, and 3rd party API calls in almost every app we have. IMO, this adds 20% or more to the (already high) testing overhead just to get these things configured right in the test suite - More code is moving to front-end Javascript stuff ... so, now I have to write Rails tests AND JS tests? Sounds delightful"

I feel your pain, I code stuff that use WebGL currently and I find it hard to test that stuff.

About bugs in production, after you find a bug write a test that exercises that bug. Then make the test pass. That way, you're unlikely to ever have a regression on that bug.

For browser-side UI tests, selenium is very useful.