Tuesday 6 April 2021

TDD is not a testing strategy

TDD is not a testing strategy

TDD is not a testing strategy

TDD is not a testing strategy

TDD is not a testing strategy

TDD is not a testing strategy

TDD. Is. Not. A. Testing. Strategy.

Just a passing thought. Apropros to absolutely nothing. 'Onest guv.(*)

Dunno if it occurred to you, but that TDD thing? It's not a testing strategy. It's a design strategy.

Let's look at the name. In the name test-driven is a compound adjective: it just modifies the subject. The subject is development. It's about development. It's not about testing.

It's probably better described by BDD, although to me that's a documentation strategy, rather than a development (or testing) one. BDD is about arriving at the test cases (with the client), TDD is about implementing those cases.

The purpose of TDD is to define a process that leads you - as a developer - to pay more attention to the design of your code. It achieves this by forcing you to address the requirement as a set of needs (or cases), eg "it needs to return the product of the two operands". Then you demonstrate your understanding of the case by demonstrating what it is for the case to "work" (a test that when you pass 2 and 3 to the function it returns 6), and then you implement the code to address that case. Then you refine the case, refactor the implementation so it's "nicer", or move on to the next case, and cycle through that one. Rinse and repeat.

But all along the object of the exercise is to think about what needs to be done, break it into small pieces, and code just what's needed to implement the customer need. It also provides a firm foundation to be able to safely refactor the code once it's working. You know: the bit that you do to make your code actually good; rather than just settling for "doesn't break", which is a very low bar to set yourself.

That you end up with repeatable tests along the way is a by-product of TDD. Not the reason you're doing it. Although obviously it's key to offering that stability and confidence for the refactor phase.

Too many people I interact with when they're explaining why it's OK they don't do TDD [because reasons], fall back to the validity / longevity of the tests. It's… not about the tests. It's about how you design your solutions.

Lines of code are not a measure of productivity

Tangential to this, we all know that LOC are not a measure of productivity. There's not a uniform relationship between one line of code and another adjacent line of code. Or ten lines of code in one logic block that represent the implementation of a method are likely to represent less productivity burden than a single line of code nested 14-levels deep in some flow-control-logic monstrousity. We all know this. Not all lines of code are created equal. More is definitely not better. But fewer is also not intrinsically better. It's just an invalid metric.

So why is it that so many people are prepared to count the lines of code a test adds to the codebase as a rationalisation (note: not a justification, because it's invalid) as to why they don't have time to write that test? Or that the test represents an undue burden in the codebase. Why are they back to measuring productivity with LOC? Why won't they let it occur to them that - especially when employing TDD - the investment in the LOC for the test code reduces the investment in the LOC for the production code? And note I am not meaning this as a benefit that one only realises over time having amortised it over a long code lifespan. I mean on the first iteration of code<->test<->release, because the bouncing back and forth between each step there will be reduced. Even for code which this might (although probably won't) be the only iteration the production code sees.

It's just "measure twice, cut once" for code. Of course one doesn't have the limitation in code that one can only cut once; the realisation here needs to be that "measuring" takes really a lot less time than "cutting" when it comes to code.

In closing, if you are rationalising to me (or, in reality: to yourself) why you don't do TDD, and that rationalisation involves lines of code or how often code will be revisited, then you are not talking about TDD. You are pretty much just setting up a strawman to tilt at to make yourself feel better. Hopefully that tactic will seem a little more transparent to you now.

Design your code. Measure your code.



(*) that's a lie. It's obviously a retaliation to a coupla comments I've been getting on recent articles I've written about TDD.