Project Layout (v3.0.0, 2022-04-01)

I'm a polyglot programmer. I work in a variety of languages but mostly in C#, Typescript, and Rust. Every few years, I try a new language to see if I can pick up new ideas or if one “fits” my current mental state better. This is also why I've done a lot dozens of other languages; I would say I know over thirty languages but I'm only a “master” in a handful.

I also flit from project to project. I have my writing and games. I have little one-off programs and ones that I hope will become a major thing. But, like everything else in my life, I'm “gloriously unfocused” on my tasks which means I have to minimize the speed that I get into a project.

Tools Selection

One of the earliest approaches I had to try getting a proper environment at the per-project level was asdf. It worked out fairly well for a few years, but then I noticed that my various novels and stories were getting fragile. There were limitations that asdf couldn't handle easily which meant I needed something more reliable. That led me into Nix which is my current setup because entering the directory sets up that project's settings while still giving me the reproducibility I need for my novels.

This means that most of my projects now have a ./flake.nix and a ./flake.lock in the root level.

Building, Releasing, and Actions

Because I've fallen in love with Semantic Releases and Conventional Commits, a lot of my processes are built around those. In earlier projects, that usually meant that almost every project also included Node in some form so I could use semantic-release. That also meant I could use package.json to handle versioning.

Though, recent thoughts have suggested that I need to break that “one tool fits all” approach. Mostly it came while working on Nitride and this website. I found myself trying to have “one build system” to create everything related to the site, including handling Javascript and CSS/SASS. Those are two very complicated projects in C#, so I realize it made sense that instead of creating a Nitride task to call webpack, I really should just call webpack directly. In other words, the Unix philosophy.

This is where being a polyglot and using different tools comes into play. I have a website that does C#, Typescript, and SASS at the same time. Which one is the “root”, which command drives everything? What about a Rust project? Or something else?

That has kind of led me to my current approach. Instead of always packaging Node in my projects, I really should have a standard location to handle the various actions/targets that apply to any project. Right now, that seems to be shell scripts.

With shell scripts, I just have to know that ./scripts/build.sh will do whatever is needed to build the target. Same with ./scripts/test.sh and ./scripts/release.sh. A Rust project may call Cargo, a .NET project will call dotnet, and polyglot will call any and all needed to build it.

This will give me room to experiment. If I decide I want to play with Cake for my .NET projects, then it still works out because it is just a consistent place. If I want to use Lefthook instead of Husky, I can.

I also went with .sh suffixes on the files because while I mostly code in Linux, I also want to support Powershell and Windows. That way, it is also clear that build.sh and build.ps1 probably result in the same end-result, but specific for that language. (I know Powershell runs on Linux too.)

Obviously, some documentation would be required, but that could be a README.md file in that directory. That will look nice in GitLab and give documentation.

Metadata

Categories:

Tags: