Lessons from a skeleton

I've been working on a vanity project recently; as part of the exercise, I decided to try to built out a walking skeleton. The basic idea behind a walking skeleton is to front load a bunch of plumbing concerns. The Ur skeleton is probably Hello World - in effect a warm up exercise to ensure that we can compile a program, link with our I/O library, run the result. In 2020, we normally add into this a CI or CD pipeline, source control, and so on. In my case, I'm working toward a REST API with serverless reference implementation, so the walking skeleton includes configuring a number of AWS offerings. First lesson: I targeted an transient deliverable. I had considered obeying the testing goat and writing a django implementation, then doing a lift and shift. The problem is that we are now front loading the wrong pain -- I don't care, in the long term, about the leaky abstractions I'll encounter in Django because that is just a temporary structure. The lessons of Django are unlikely to transfer to the intended staging, so I might as well skip that pain and move more directly toward my destination. Second lesson: in so far as is possible, work inside out.

Everything should be built top-down, except the first time.

Walking skeleton is very much a "the first time" situation; there's a lot of comfort to be had in introducing a single new element to something that has been proven to be sound. Working outward tends to concentrate the uncertainty cloud in the new elements, because the lower level traps have already been discovered. Third lesson: manual first. We're not done until we've achieved our goal of a one click deploy. BUT working through the skeleton manually first helps to clarify the automation requirements. This is a form of separation of concerns; we start out with two problems -- one is to align our own custom design into the general purpose model of our supplier, the other is to automate the steps of the general purpose model. Keeping these two ideas separate allows us to bring our concentration to bear on a single concern.