M by N Code Explosions

Back to Perlis-Thompson Principle

Some concrete examples:

• Linux distros!
  • I know this from Oil. It is annoying to have 10 different distros! You have M applications and N packaging tasks. There are some tools that try to ameliorate this.
• Language package managers {Python pip, JavaScript npm, Rust cargo, ...} x {Linux, Windows, OS X, ...}. This partly explains the XKCD comic -- Python packaging is messy and hard to test.
• Build systems likewise have a (M languages) x (N operating systems) problem, e.g. Zig https://lobste.rs/s/cykija/zig_build_system#c_eizab2
• Shell Completion
  • M applications to write completions for, and N shells.
  • This is what Shellac (Shell Autocompletion) was supposed to solve! However I think most shells including bash are too weak to support it at the "user level"
• NUL terminated vs lines of text
  • 2 formats x N operations
  • xargs -0 and grep -z, sort -z, head -z
  • compare with QSN
• Google's cluster management: You're using 3.5 distributed operating systems / cluster managers when you use Kubernetes
• Google service turnup (Broccoli Man, etc.)
  • Unix and Microservice Platforms by Brandom Bloom
  • auth/metrics/config/monitoring are M things you have to do for every service.
• Protobuf generated code.
  • {types ... } x {copy, assign, serialize, deserialize, DebugPrint, ...}
• Some parts of Windows? Well COM reduces the code explosion, just like Unix pipes do. It's dynamic / runtime software composition, not static / compile time.
• Docker
  • duplicates push and pull
  • duplicates remote storage -- i.e. with the registry

Non-open source

• The "Send To" functionality in iOS and Android operating systems.

Perlis-Thompson Problems

These are like Mx N code explosions, but it's often 2x2 or 2x N. There are 2 things instead of 1.

• Three Perlis-Thompson Problems in Docker/K8s: https://oilshell.zulipchat.com/#narrow/stream/266575-blog-ideas/topic/Perlis-Thompson.20Problems.20in.20Docker.2FK8s
  1. Separate storage for code and data with Docker registry. Should just be git. Although arguably git is only good for source code!
  2. multistage builds -- should be 2 stages of the same thing!
  3. persistent processes vs. batch processes -- should be one mechanism