[Please, read me carefully. If at any point in time you think I'm preaching to stop writing tests then STOP reading further and FORGET what read so far. Thanks!]
One might think that tests only bring value, that is muchos dolares. If this is the case then why many teams stop maintaining their tests at some point? Or they @Ignore a test forever when it is too difficult to fix? Isn’t it like throwing away ‘value’. Why on earth you are throwing away $$$!?
Thing is that tests bear maintenance cost: keeping them green, refactoring along with the code, etc. Let’s talk about net value of a test: Net = Value – Cost. If you write crappy tests then the net value might be negative. Since I visited quite a few archeological sites digging in legacy code I saw many tests with negative net value.
When half of your team wants to run ‘svn delete‘ on the ‘FitnesseRoot‘ folder it might mean your tests’ net value is plummeting down. Obviously, fitnesse is just an example, any tool can be a blessing or a curse depending on your skill in misusing it. (Dear god of Agile, I swear I love fitnesse… but only with GivWenZen).
Let’s think about it. I’ve got 3 tests with a respective value I conjured using cosmic energy.
Test 1. Adds an article to the content db. Value: 50 $, Cost: 20 $
Test 2. Adds an article & edits the article. Value: 50 $, Cost: 20 $
Test 3. Adds an article & edits it & deletes it. Value: 50 $, Cost: 20 $
The cost is relatively high when compared with value because those tests are deep functional tests that require the whole system to be set up; they use funky frameworks to drive the web browser; they are kind of painful to debug & maintain.
Total cost is easy, just sum up the cost and be scared by 60$.
Total value is tricky. To your great surprise (hopefully ;)) it is not 150$… but 55$ !!! Why? Read again test #3 – it covers the same cases as test #1 & #2 so the value of those tests does not really sum up! I threw 5$ extra because by having the functionality in separate tests we have better defect localization when tests fail. Hence 55$.
This means the net value of tests is: -5$ (minus five dollars).
Funny enough, the tests mentioned are a real case from a gig I was few years ago. I argued with the QA lead that keeping those tests separate does not make sense. To my despair, I failed. 6 months later nobody was maintaining functional suite of web-clicking tests in this project.
My message distilled:
1. Understand the net value of tests.
2. Look after your tests as if they are your newborn daughter!
Now go back to the wilderness and keep writing sane tests!
this is a great post Szczepan. I just google searched for it specifically so that I could send the link to my new team.
You are discounting the value to the marginal value but not the costs to the marginal costs.
Even with duplication, the costs of initially writing the tests would go down massively. With proper factoring of the tests also the maintenance costs would.
I would expect something like 55 – 25.
Great point Dierk, I missed that. Yet, I think that in my particular situation the discount for cost is a bit smaller, e.g: 55 – 35. Therefore I still prefer 50 – 20 (30 net value) than 55 – 35 (20 net value).
I will update my post to accommodate cost discount.