Patterns and Anti Patterns

UNDER CONSTRUCTION

This page is about how to use shell.

Terms

Shell Functions / procs
- Bernstein Chaining / Chain Loading
- $0 Dispatch Pattern (sudo, xargs -P, find -exec)
- "Task file" pattern -- with runproc
  - no data dependencies?
  - TODO: Shell Scripts Are Executable Documentation
  - Same pattern as https://github.blog/2015-06-30-scripts-to-rule-them-all/ , except they use a DIRECTORY, not a single file
- Proc-like: argv -> status, and stdin/stdout/stderr
  - command annotations can tame this generality, but we want proc-like
- Proc as function: stdout is the "return value" (like Elvish). Return value is for errors.
PPPT: Parameterized Parallel Process Table
- you can add a column to a table in parallel
data dependencies: shell vs. make/Ninja
Software Evolution
- Versionless Interchange Format
  - lines of text, JSON
  - Oil: QSN, QTT
- Feature Detection, not Version Detection
Avoid Mini-Languages in Other Tools
- xargs -I {} can be replaced with a shell function and $1
- '\n' in cut, xargs, etc. can be replaced with $'\n'
- xargs -L special tokenization rules should just be -n
- find -exec foo + (for batching args) can also be replaced with find -print0 | xargs -0 (avoiding weird ; vs. +)

The YAML problem. The sh-in-{yaml,docker,systemd,chef,vagrant,python} problem.
- actually 2 variants of this: sh-in-YAML, and YAML-based languages like Github Actions (and Gitlab)
Whitespace Joining Antipattern
- SSH quoting
- eval builtin
- echo builtin
I'm Too Lazy to Write a Lexer. Instead I'll just use the argv array. Used by find, test, and expr.
Deserialize a huge blob -> trivial in-memory operation -> Serialize huge blob. This is bad for performance.
- instead use lazy deserialization

Distributed OS Design: The Inner Platform Anti-Pattern
- Kubernetes suffers from this