Introduction⌗

Currently, our toolkit is written in Bash 5. The problem is that it’s hard to maintain it. Sure, Bash can be combined with YAML/JSON parsers to do some cool stuff, but it just complicates things. It’s harder to debug such scripts because literally several languages are mixed in one place and each of them has its own unique syntax.

We proved ourselves that it’s possible to do exciting stuff in Bash. But the end goal of this project not to prove ourselves what we can, but make codebase maintainable. As a result we’re going to rewrite it entirely in Go.

Problems with Bash⌗

We identified these issues with our preferred shell (just the most bothering are mentioned):

  • No readable functions to manipulate strings, just variable expansions. While it saves typing, it makes interpreting scripts harder for newbies.
  • No strong dynamic typing. By strong term I mean something like in Python. Bash allows to interpret the same variable value as a string or number in different contexts. While it may seem convenient at the first glance, it’s error-prone. Also, undefined variables are treated as zeros in arithmetic expressions which may hide typos.
  • No way to actually return some structured data, even just one-dimensional arrays. While structuring is possible: you can output YAML/JSON and then interpret it via yq/jq, the way it’s done is not efficient and inconvenient.
  • LSP server for Bash is not ideal, now it’s impossible to perform variable or function renaming via RMB in VS Code. It means that the only way to rename something is to change all occurrences of it and check whether just right identifiers are affected by this change.
  • Bats plugin for Visual Studio Code doesn’t provide intellisence for .bats files, just snippets and syntax highlighting. While it’s not a big issue for small unit test files, it still makes test writing more error-prone as no one tells feedback about what you wrote instantly and no one suggests what can be used in a specific scope.

What does it mean?⌗

We don’t reject Bash completely, we acknowledge that it’s a good shell. But for the tasks we mostly perform it’s not the right tool. We’re going to still use it for CI/CD pipelines, but constrain shell scripts to short ones and require rewrite them if they are too long.

Note that we don’t deprecate Bash tooling right now, we are going to close some opened issues in its repo (especially, fix bugs), but that’s all. Today we are closing some issues marking them as invalid to implement this stuff in Go later.