Novel Automation

This post is a follow up to the interlude in the /posts/programming-tutorials post, which part of an ongoing series of posts on programmer training and related issues in technological literacy and education.

In short, creating novel automations is hard. The process would have to look something like:

  1. Realize that you have an unfulfilled software need.
  2. Decide what the proper solution to that need is. Make sure the solution is sufficiently flexible to be able to support all required complexity.
  3. Then sit down, open an empty buffer and begin writing code.

Not easy. [1]

Something I've learned in the past few years is that the above process is relatively uncommon for actual working programmers: most of the time you're adding a few lines here and there, testing various changes or adding small features built upon other existing systems and features.

If this is how programming work is actually done, then the kinds of methods we use to teach programmers how to program should hold some resemblance to the actual work that programmers do. As an attempt at a case study, my own recent experience:

I've been playing with Buildbot for a few weeks now for personal curiosity, and it may be useful to automate some stuff for the Cyborg Institute. Buildbot has its merits and frustrations, but this post isn't really about buildbot. Rather, the experience of doing buildbot work has taught me something about programming and about "building things," including:

  • When you set up buildbot, it generates a python configuration file where all buildbot configuration and "programming" goes.

    As a bit of a sidebar, I've been using a base configuration derived from the buildbot configuration for buildbot itself, and the fact that the default configuration is less clean and a big and I'd assumed that I was configuring a buildbot in the "normal way."

    Turns out I haven't, and this hurts my (larger) argument slightly.

    I like the idea of having a very programmatic interface for systems that must integrate with other components, and I really like the idea of a system that produces a good starting template. I'm not sure what this does for overall maintainability in the long term, but it makes getting started and using the software in a meaningful way, much more possible.

  • Using organizing my buildbot configuration as I have, modeled on the "metabuildbot," has nicely illustrated the idea software is just a collection of modules that interact with each other in a defined way. Nothing more, nothing less.

  • Distributed systems are incredibly difficult to get people to conceptualize properly, for anyone, and I think most of the frustration with buildbot stems from this.

  • Buildbot provides an immediate object lesson on the trade-offs between simplicity and terseness on the one hand and maintainability and complexity on the other.

    This point relates to the previous one. Because distributed systems are hard, it's easy to configure something that's too complex and that isn't what you want at all in your Buildbot before you realize that what you actually need is something else entirely.

    This doesn't mean that there aren't nightmarish Buildbot configs, and there are, but the lesson is quite valuable.

  • There's something interesting and instructive in the way that Buildbot's user experience lies somewhere between "an application," that you install and use, and a program that you write using a toolkit.

    It's clearly not exactly either, and both at the same time.

I suspect some web-programming systems may be similar, but I have relatively little experience with systems like these. And frankly, I have little need for these kinds of systems in any of my current projects.


[1]Indeed this may be why the incidence of people writing code, getting it working and then rewrite it from the ground up: writing things from scratch is an objectively hard thing, where rewriting and iterating is considerably easier. And the end result is often, but not always better.
comments powered by Disqus