This is a remote pair programming codecast with Wassel Alazhar on an interesting problem called Good Friends.
After implementing a bit of the problem, we discuss about what we have done. We observe a lot of duplication and we debrief.
This is a remote pair programming codecast with Wassel Alazhar on an interesting problem called Good Friends.
We pick a problem not at all known by me and we figure it out with #TDD on first view. Interesting to see how you need to analyze before rushing to write tests
At SoCraTes France I paired with Rémy and Johan to show them how I code using TDD as if you Meant It. We had a few hours to try it, and then Rémy told me he would like me to present this to the Grenoble Software community. So this is how we arranged a remote live coding session on TDD as if you Meant It.
During this session I am using Poker Hands as a problem, and I start with the usual Think-Red-Green-Refactor and Behavior Slicing. Then a lot of refactoring is going on. I invite you to watch the video and come with questions and improvement ideas.
At the end of the remote live coding session we had a remote discussion about what happened. Here is in form of Q&A.
Question: When you want to show the duplication to be sure it is not accidental, you try to duplicate three times the same code. Do we need to do this when we do just copy/paste? It doesn’t really help the tests.
Answer: The fact that you can duplicate the test it shows it is easy enough and we don’t necessarily use the Rule of Three. But there are other cases when it’s more complicated and just duplicating the tests would not show the same. And there is still some duplication that doesn’t prove in the production code. We find that it’s not really the same behavior, and it’s often a surprise. The fact that I can make the duplication the third time is just making the point that we need to remove the duplication. Otherwise I need to search for duplication in another way.
Question: It seems that we create these tests just to follow the Rule of Three, not because we want to.
Answer: I don’t want to create the tests just to follow the constraint of The Rule of Three, I just want to be sure enough that what I am doing is on the good path. I don’t want to rush into refactoring. It is a question of how well you know your domain and how much you trust your instincts. The TDD as if you Meant It comes with the idea you don’t trust your instincts and you need proof.
Question: If you want to create some duplication, we can use parametrized tests.
Answer: Yes, for the tests that are duplicated we can use parametrized tests and compress the code. And probably at the end there will be very little code.
Question: Have you detected different styles of doing TDD
Answer: Yes, but most of the problems I see that people fail to respect the rules and don’t find a way to respect the rules. But of course, there are very different styles. Some go very functional, some like to use more state, I go in very small steps. I like to go in very small steps and focus on triangulation. I like to use it as a machine that is generating duplication in all these tests and then because I have duplication, I extract the methods and the classes. You do need to refactor a lot. And you need very good tools to refactor automatically.
Question: I think this technique is a lot more difficult to use with Legacy Code, is it?
Answer: Yes, it is more difficult. But you don’t need to use it too often with legacy code. I can use it on legacy code, but first of all I surround the code I want to change, put some interfaces around it, and then use the same steps to generate the code to rewrite it with very small steps in a simpler way. There might be situations where I prefer to rewrite some legacy code instead of refactoring. When using TDD as if you Meant It with legacy code you need to take very small steps, and do a lot of refactoring in small steps, understand the flow of dependency injection and understand the patterns. All these I learned by using TDD as if you Meant it a lot. I think it helps you to refactor better. You have this refactoring muscle. With legacy code you have a lot of primitives and so much misplaced code. Here with new code you have the same, and you need to understand how it makes sense to put it together. So I think you can use TDD as if you Meant It especially for the refactoring part.
After in the previous episode I took too bigger steps, during this episode I start all over and take smaller steps. The main differences are:
TDD as if you Meant It: Clean-up before next Triangulation (Episode 13) About This episode is about failure to refactor because of trying to rush to the next step of Triangulation and taking too bigger…
TDD as if you Meant It: Separate Production from Test Code (Episode 12) About When using TDD as if you Meant It we need to have a clear distinction between the test code, and the…
TDD as if you Meant It: Refactor and Clean-Up (Episode 11) About From Wikipedia we learn that Code refactoring is the process of restructuring existing computer code—changing the factoring—without changing its external behavior So clearly not any…
TDD as if you Meant It: Add Missing Tests for untested Branches (Episode 10) About During the first episode of this series I added some initial tests where I didn’t cover all the cases. Now…
TDD as if you Meant It: Experiment Design Alternatives when Stuck (Episode 9) About During this episode you will see how I formulate experiments to decide where it is a good idea to get the…
TDD as if you Meant It: Create Duplication in order to remove it (Episode 8) About After a couple of episodes when I showed how we can get out of TDD as if you Meant…
This is the second episode of an alternative evolutionary design approach. During the (Episode 5) I refactored the code to generate a builder. The alternative presented in Episode 6 and 7 is to stop using TDD as if you Meant It for a while and start using some traditional TDD.
During this second episode for creating the builder with traditional TDD, I am trying to focus on finishing the problem. This means I am trying to get in a flow state, where everything works well at the end. Also that means that I am often taking bigger steps than what you saw in the first 5 episodes. This is because I consider the risk being smaller, and because I know where I want to get.
Because I am focusing on the problem, I am doing more refactoring. And the type of refactoring is faster, in bigger and more direct steps. I want to leave a nice code base behind so I try to have the best code I can for the moment. So then during the next episode I can use this code with ease.
At one extreme I am in the point where I know exactly what I need. In this case any form of Evolutionary Design or TDD is useless. I might want to add automated tests (while adding the code or after) to perform the checks on the code just for regressions.
At another extreme I don’t know at all what I need, but I have the specs. So I start with them and I focus on the principles of refining the code concepts in order to extract the appropriate design entities.
During this codecast I will situate myself very close to the extreme where I know pretty well what I want. But careful to be sure you always introspect and clarify if you have all the information to take a design decision without enough proof. Your solution might be too complicated or just inappropriate.
This is the first episode of an alternative evolutionary design approach. During the last episode (Episode 5) I refactored the code to generate a builder. The alternative presented in Episode 6 and 7 is to stop using TDD as if you Meant It for a while and start using some traditional TDD.
In a traditional TDD approach I am creating some structures up-front. I can create more or less design up-front, but still the minimum I create is the production class or function I want to test. This approach is different from TDD as if you Meant It where I am extracting the code from the test method, to the test class and to a production class.
In a traditional TDD approach we can have different approaches: bottom-up, top-down, middle-top, middle-bottom, etc. But I take this decision depending on the clear need I observe in the current moment.
Whenever I start with traditional TDD, I need to define the clear specifications. Or in other terms, I define the clear needs of the design structure I need to add. In this current case I defined the fact that I need a fluent builder. So my current tests will reflect the need in the tests. In this approach the evolutionary part of the design is marginal. As I am very clear with what I need, I don’t need a lot of proof. This is not always the case.
At one extreme I am in the point where I know exactly what I need. In this case any form of Evolutionary Design or TDD is useless. I might want to add automated tests (while adding the code or after) to perform the checks on the code just for regressions.
At another extreme I don’t know at all what I need, but I have the specs. So I start with them and I focus on the principles of refining the code concepts in order to extract the appropriate design entities.
During this codecast I will situate myself very close to the extreme where I know pretty well what I want. But careful to be sure you always introspect and clarify if you have all the information to take a design decision without enough proof. Your solution might be too complicated or just inappropriate.