Also, software is ultimately written by humans, for humans. A very overlooked thing in many circles. It seems so simple, but folks off in the deeply technical universe of engineering should remember it.
Maybe this is why infrastructure as code feels so obtuse to me, there isn't many opportunities to explain yourself in the code and so most of the communication is done in the docs.
Perhaps there is an opportunity for a more natural language approach to infrastructure as code, but I am not sure how far you can take that since the reality is still a lot of arbitrary delcarations.
“It is very difficult to talk of the operations of the mind with perfect propriety and exactness; because common language has seldom made any very nice distinctions among them, but has generally call’d by the same term all such as nearly resemble each other. And as this is a source almost inevitable of obscurity and confusion in the author; so it may frequently give rise to doubts and objections in the reader, which otherwise he wou’d never have dream’d of.”
––A Treatise of Human Nature, by David Hume, pg. 76
I doubt a "natural language" approach would work; the natural language for the domain has to be very precise. Just like the "natural language" for software has to be similar to either C or LISP. You are giving precise technical instructions which will be run by a machine.
"Hey, just spin up a server with enough capacity, alright?" Yes, sure. Please define capacity.
Those definitions are usually in the docs, which is why the code makes so many references to the docs.
Many infra-as-code systems allow you to add comments to the docs, which (if done well) can really help.
And infra-as-code is much superior to infra-as-let's-click-on-the-gui-until-it-works' Yes, it is hard to read (I know, I do it for a living). But at least you have a document which describes the infra!
I think a lot of the infrastructure-as-code motivation comes from struggling with documentation. At least it provides a unambiguous source of truth which is going to be correct. Understanding it and it’s quirks can be tedious but it’s better than port scanning and guessing hostnames to figure out what infrastructure exists.
I've never seen it done in infrastructure as code files, but I probably would if I were writing it. Or I'd at least include some explanation in the readme.
I do it all the time in YAML config and HCL, especially stuff which is not obvious. That's the main benefit of YAML over JSON, to be able to communicate to human readers through comments.
Some stuff, like which CIDRs are necessary for which connection, are only obvious in the moment you write it, not 1 or 2 years later or for another person who joins your project next week.
Software is written by humans for solving a new problem with new constraints and requirements. Code being understandable to other humans is a desirable side-effect, but in no way primary. Some problems are inherently hard and few humans have the capacity to reason about them, no matter how nicely-formatted, modularised and linted your code is.
All the more reason not to make it any more complicated than necessary (even if it's in the expense of some code elegance). Not just for other people, but also for the future you, coming to maintain this. Going back to your code after three years, and someone else approaching the code for the first time are not that much different.
Also this is why I love automation, like scripts for everything. It's like a memo to your future self, with the side benefit of sparing you menial work. I keep even single one liner commands as their own shell scripts with a descriptive name.
Try to read and understand everything your write. If it doesn't click immediately or needs too much outside context or "working memory", make it simpler. Failing that, elaborate in comments.
Sometimes it has to be complicated. Complication in one area (usually infrastructure code) enables simplicity in other areas (usually application code). An OS memory manager deals with the complexity of page management, so as to present the nice abstraction of an orderly address space to applications. An SQL query optimiser is complex in order to allow dumb and "understandable" queries from applications.
It isn’t just about code. It’s about the software’s user. Humans and the programming is one part. But the software is for people. The problems being solved are for people. It is humans all the way down. In this way I view software as a very humanistic domain. That is why the theme of communication angle resonates with me so much. Agile is misused a lot but for goodness sake, talk to your users. Even if your user is just you.
This is exactly the thinking I see people trapped in and was pointing out. Focused on solving the engineering problems, not the humans that come before and after the problem (both end users and developers). We know there are hard things in engineering and problem solving ain’t easy. I think in an absolute sense you are right, but in a practical sense, it is more than a desirable side effect. Code once written and solving a problem can live on for a very ling time.
