Top down development has you starting at the highest point in the application that you can. From here you code down until there is nothing else left to develop. Once you reach this point you should be code complete. Along the way you may need to stub out areas that have not yet been created, or designed.
Bottom up development has you starting at the lowest point in the application. The idea being that this part of the application has the most complexity or will be the most important. You will build the system up from a series of smaller components.
Top down development and bottom up development was introduced to myself in my early days of university. At the time the distinction didn't really mean much - I was very much a developer who would work from the bottom up.
Over time I have completely switched my stance on this. I believe agile practices and TDD are the reason for this change. I feel so strongly about this that I would go as far as to claim that within an agile team - bottom up development is an anti pattern.
Consider the following tasks to be completed on a team of four developers.
- Create controller - main entry point, request mapping.
- Create service - service layer, simple business logic.
- Database query - thin wrapper around complex DB query.
With a bottom up approach a pair of developers could work on the complex database query. After some time they would have this working. The other two developers could start with the controller or service.
The problem with this approach comes from the painful integration process. The developers working on the service might be coding against the interface the team discussed during a planning session, while the developers on the query may have had to change their approach.
This example is trivial, but imagine a story with thirty tasks, more developers and more complexity and this bottom up approach is difficult. Over the past few years my top down approach has evolved.
My first step would be to stub out the workflow with the above implementation. There is no real logic here - only the objects collaboration is implemented. At this stage there are no tests, TDD would not be used. After all there is no logic here. The code is so simple it can be reasoned about with peer review, planning sessions and so on.
At this stage all of the tasks are open for any developer to pick up. If a breaking change was required, there would be no way for one pair to commit these changes without the other pair knowing. Another benefit of this approach is that an end to end acceptance test could be wrapped around the functionality from the get go.
As part of these tasks each developer would use TDD. Remember no tests exist at this point. Building up the tests in stages would ensure the logic of how the objects collaborate is preserved, and ensures that the actual domain logic that is implemented is correct. Does this mean we aren't doing TDD? No, of course not. The tests will drive the implementation. If we need to introduce new objects that is fine - these simply become implementation details that the other devs need not worry about as long as the workflow is not broken.
This approach to top down development isn't new, though many don't appreciate its benefits. I plan on expanding on this style of pragmatic TDD in the coming months.