Since wiping my Github over a year ago I’ve taken a much more, let’s say, impatient approach to my work.
I have a lot of things I want to do, and a lot of things I want to see exist. Most of the things I want to see exist are niche enough that they won’t exist if I don’t make them. So, over time, I’ve refined a few anti-patterns and anti-techniques that let me build the worst possible version of a thing I myself will just barely put up with using. Let’s dive into some of them.
Using Git as a database
A few months after moving to Finland almost illegally and reading a really good book I promptly threw everything I had learned about elegant and composable design in the garbage and started making myself a flashcard deck to learn Finnish. The result was surprisingly popular, popular enough that I even made a sequel which I still use mostly unchanged to this day.
There’s just one problem: I lost the SQLite databases I used to build both of these databases. Good on me for not overengineering even more, I guess, and using a whole Postgres server for my dinky little web-scraping experiments, but I really do not have the time anymore to go back and run the code I used to create these databases from Tatoeba.
Not so anymore. Now when I want to scrape web data, I scrape directly to a GitHub repo until it becomes painfully obvious this is a bad idea. (Hat tip to Simon Willison’s Weblog for pioneering the technique! You are an inspiration to me to do better, even if I don’t.)
Using shell scripts and pandoc instead of a big boy HTML parser
It gets worse. I not only use Git as a database, I make intermediate Git databases, out of the raakaa-aineita, by wholesale git submodule-ing the original database time and time again until I get what I actually want. Thank god ``git submodule –recursive` is not the default, we would be here all day.
Now HTML is a pretty nice format, actually. I could use BeautifulSoup and Python to very easily extract whatever I wanted from the webpage. It would probably be less work overall, and certainly less fiddling to get what I want. But I don’t even do that! I run pandoc --from=html --to=commonmark and then run everything through 15 lines of sed and then dump the results into a new file! Which I then automatically commit to Git!
Are there benefits to doing any of this? For a fool like me, absolutely. It’s easy to rg and fd my way to whatever intermediate file is choking my silly little news archive today.
Doing everything on a Raspberry Pi running cronjobs in my closet
“These aren’t that bad for projects of this small a scale.” Just you wait.
This Christmas I got a Raspberry Pi. I turned it on, tossed the same Ubuntu Server image I use for everything these days on it, and immediately dedicated 2 of its extremely expensive 4 gigs of RAM to a tmpfs. I have somewhere on the order of a dozen cronjobs on this thing which commit Git sins on a recuurring basis to this /ram, and the only reason I put it all in /ram in the first place was to avoid wearing down my SD card too quickly. Then another dozen or so which clone Git repos back down to /ram on startup.
It’s fast, it’s easy to work with, and it’s brittle as fuck. I paid money for this which could have gone into VTSAX. But I chose to be a silly man.
“Higher order” dependency management
Back in the day, when I knew little of the world, I did pip/apt-get install whatever and moved on with my life.
When a hasty emigration from the States forced me to adapt my CS hobby into an actual career or die trying, I spent some time trying to look more professional. I wrote Poetry, I committed every facet of Hypermodern Python to memory, I submitted a heavily overpackaged coding assignment to Wolt. And then I realized I could just as easily stick things in a Docker container and call it a day.
Now, knowing slightly more about the world, I do pip/apt install whatever.
If the day ever comes where I need to go further, I trust GPT-4 to eat my READMEs and poop out a nice tidy Dockerfile. To date, I have never needed to do this.
“Just use Python”-ing all the things
The day will come that I can hand off a Haskell / Scheme / Go / C# / Rust / Fish / Idris / Zig / JavaScript / TypeScript / Elm / OCaml / ReasonML / Coq / Agda / Lean / Scala / Java / Julia / Perl / PHP / SQL / Prolog / Smalltalk program to someone else I have never met before, explain nothing, and expect them to be able to figure it out. On that day I will praise the sun and get my small army of goblins in a damp cave to begin work on the New Testament of my work. Cleaner, faster, less prone to en-Pi-ification.
Until then, lingua franca goes brrr.
Using CC Zero whenever possible
I believe the public domain license reduces FUD and dead-weight loss, encourages copying (LOCKSS), gives back (however little) to Free Software/Free Content, and costs me nothing.
-Gwern, emphasis mine
Amen. Especially to that last part.
The culture that is the CLIzation of everything
Above is the first draft of an actual program I am actually writing, to automatically generate flashcards from the JSON partly-refined aineita of one of those intermediate Git databases. Currently not one of those 3 commands actually does any flashcard generating.
And yet they are exactly what I want. I see this image and I know what the next 5 actions are going to be. Eventually this main.py will bloat to a 500 line monstrosity, just like all of my other projects, and I will think about maybe factoring it out into at minimum a separate cli.py and main.py, and then I will forget about it and move on with my life.
I understand many people will consider putting Minimum Viable CLIs around things to be a good pattern, not an anti-pattern. Not so! I do this because I know I don’t have the patience any more to write “real” web apps or “real” GUIs, which I myself never end up using, even if they make the project more accessible to 99% of the rest of the world.
I’m sorry. I will not change. Please forgive me my past and future trespasses. The spirit is willing, but the body is weak and needs to get back to his (handmade) flashcards now.