When is it appropriate to do RDD instead of TDD?

Thanks for that. It seems you have to take on many roles. I imagine you’re also a mentor to each student employee. It must be rewarding when things go according to plan.

I produce use-case tests and they can write the unit-tests and associated functions.

I have some troubles understanding this. Which of the following is true, if any? You produce use-case tests and they are free to solve the problem

  • however they like
  • as long as they provide unit tests along with solution
  • by writing unit-tests first

I’m probably too harsh but if example based unit tests are expected and accepted, I can see you’re walking dangerously close to a slippery slope. As soon as you start to believe that your program works because all unit tests pass you’re missing the fact that those unit tests test just a specific examples. If program works (more or less) correctly in general then that’s only down to your team doing good enough job. Desired outcome and proud moment for you and “first timers” but tests have hardly anything to do with it.

My approach to tests depends a lot on the nature of the modification I am applying to the system.

When I don’t have a clear vision of how I’m going to implement a new module I tend to prefer to sketch a test on outermost layer of the application I’m working on (given that It still has a quick feedback cycle.)
e.g If it is a web API I’m working on it would be an API test

I start writing the happy path scenario test on the outermost layer of the application.

From the moment I have the tests written and failing I start writing the code on a top-down way using RDD when writing pure functions. I like to use (comment …) forms to test each function I’m writing.

I write the necessary code until my tests are passing.

At this point, I have a much more clear vision of how I want to design that module, so I make the design changes using TDD.

When I am happy with the design I start to write the corner case scenarios as unit tests on the innermost layers.

In another scenario, when I know exactly how I’m going to design a certain module, for example, when there is a clear pattern to follow on the project, I prefer to use TDD and build the code bottom-up.


REPL and Tests are not replacement of each other. as I think most answers in this thread would agree with this.

Tests are invaluable, I need it because:

  1. it’s the safety guard of regression bugs
  2. integration/e2e tests (that covers the major usage patterns of the product) are what gives me confidence when doing major code refactoring
  3. test cases that accompanies new features code are (to a large degree) self-describing docs and functional design specs
  4. writing test first in most cases could make(or “force”) one have much clear expectation/design of the to-be-written code.

In a word, no serious software project could live without tests.

The REPL is also invaluable. I enjoy it because:

  1. Play/explore with things to get a better understanding of everything. This is also why the jupyter notebook is so popular in data science world.
  2. The instant feedback loop that easily brings one into the flow state.
  3. With tools like hashp , debux, postmortem It’s like have a lightweight debugger constantly at your discretion.

Ofc, these are all about REPL and Tests, not RDD and TDD. But the difference is small enough that I don’t think it’s worth to tell them apart.


Two quick points on this from me.

Firstly I completely buy the usual gripes with TDD. It’s a strangely expensive way to build anything, not repeated in any other industry. If you designed the feature badly TDD won’t help - instead the bad design is baked in twice. Also it tests a few highly specific paths only unlike generative tests. But in pair programming for normal Joes like me - just getting one flow built at all with a clear goal set out by the test (rather than relying on each other’s verbal skills) is fairly indispensable, as is the safety net of a barrage of tests to stop you when you mess things up.

Second I love RDD, it provides far faster and more directed feedback and changes how you think about code. But also recently I’ve been using a remote repl to build some tests that churn through user journeys in integrated test environments (with legacy systems that are difficult to reproduce locally.) Developing directly on a externally deployed instance with the same fast feedback as locally is a joy, and it’s a mode of development that TDD could never give you.


+1 for “run tests on save”! I do the same for all unit and integration tests, using kaocha’s --watch feature. It

  1. provides almost the same fast feedback loop as the REPL
  2. reduces the mental load to remember the current state of the REPL variable
  3. ensures a better coverage for all use cases.

Of course this is interleaved with using the REPL, so a typical workflow is like:

  1. Launch the kaocha watcher
  2. Add test
  3. write code, play with the repl, write more code, until the test pass
  4. refine the tests / add more tests
  5. repeat step 3

And I make heavy use kaocha’s foucs/skip feature to only run the tests for the current module I’m working on.

For e2e tests it varies from case to case, e.g. I won’t setup auto run for browser-driven e2e tests since I’m working on frontend / backend code at the same time, so a saving on the cljs file may need some change on the backend to make the tests pass.

This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.