I use fish shell only now. Used to only write bash, but I’ve started writing some fish scripts. I wouldn’t try to plan too much WRT shell scripting up front. Just fix your pain points as you go.

Use

set -x

For debugging

That’s the way I do it:

#!/usr/bin/env nix
#! nix shell nixpkgs#nushell <optionally more dependencies>  --command nu

<script content>

But those scripts are only used by me

Do you use bash? Yes because it is everywhere and available by default.

Yes, using bash on all boxen.

Scripts start with #!/bin/sh ,because, that gives quicker execution times.

Any simple aliases, I put in .bash_aliases

Tried tcsh and zsh around 30yrs ago, all bash since then.

• Fish. Much, much saner defaults.
• I am writing #!/usr/bin/env sh for dead simple scripts, so they will be a tiny bit more portable and run a tiny bit faster. The lack of arrays causes too much pain in longer scripts. I would love to use Fish, but it lacks a strict mode.
• No, why would I?
• I used to share all my dotfiles, scripts included, but I was too afraid that I would publish some secrets someday, so I stopped doing that. For synchronizing commands, aliases and other stuff between computers I use Chezmoi.
• To use Fish instead of fighting with start up time of Zsh with hundreds of plugins
• Always use the so-called “strict mode” in Bash, that is, the set -euo pipefail line. It will make Bash error on non-zero exit code, undefined variables and non-zero exit codes in commands in pipe. Also, always use shellcheck. It’s extremely easy to make a mistake in Bash. If you want to check the single command exit code manually, just wrap it in set +e and set -e.
• Consider writing your scripts in Python. Like Bash, it also has some warts, but is multiplatform and easy to read. I have a snippet which contains some boilerplate like a main function definition with ArgumentParser instantiated. Then at the end of the script the main function is called wrapped in try … except KeyboardInterrupt: exit(130) which should be a default behavior.
• Absolutely not a bad practice. If you need to use them on a remote server and can’t remember what they stand for, you can always execute type some_command. Oh, and read about abbreviations in Fish. It always expands the abbreviation, so you see what you execute.

I use sh to attempt to keep it compatible with POSIX systems.

I use pain bash. Never really tried zsh and fish, since most of my Linux work is on servers and I don’t really care for extra features.

I try and write idempotent scripts when possible.

I wouldn’t create those aliases on a fleet because writing them to the configuration file of your shell in an idempotent fashion is hacky and my VMs are like cattle.

#!/usr/bin/env bash

A folder dotfiles as git repository and a dotfiles/install that soft links all configurations into their places.

Two files, ~/.zshrc (without secrets, could be shared) and another for secrets (sourced by .zshrc if exist secrets).

Am I missing something - doesn’t bash have tab completion or of the box?

I use bash as my interactive shell. When ~20 years ago or so I encountered “smart” tab completion for the first time, I immediately disabled that and went back to dumb completion, because it caused multi-second freezes when it needed to load stuff from disk. I also saw it refuse to complete filenames because they had the wrong suffix. Maybe I should try to enable that again, see if it works any better now. It probably does go faster now with the SSDs.

I tried OpenBSD at some point, and it came with some version of ksh. Seems about equivalent to bash, but I had to modify some of my .bashrc so it would work on ksh. I would just stick to the default shell, whatever it is, it’s fine.

I try to stick to POSIX shell for scripts. I find that I don’t need bashisms very often, and I’ve used systems without bash on them. Most bash-only syntax has an equivalent that will work on POSIX sh. I do use bash if I really need some bash feature (I recently wanted to set -o pipefail, which dash cannot do apparently, and the workaround is really annoying).

Do not use #!/bin/sh if you’re not writing bash-only scripts. This will break on Debian, Ubuntu, BSD, busybox etc. because /bin/sh is not bash on those systems.

• Do you use bash? If not, which one do you use? zsh, fish? Why do you do it?
• Do you write #!/bin/bash or #!/bin/sh? Do you write fish exclusive scripts?

I use bash, and I use #!/bin/bash for my scripts. Some are POSIX compliant, some have bashisms. But I really don’t care about bashisms, since I explicitly set the bash as interpreter. So no, no fish exclusive scripts, but some “bash exclusive” scripts. Since fish is aimed towards being used as interactive shell I don’t see a real reason to use it as interpreter for scripts anyways.

• Do you have two folders, one for proven commands and one for experimental?
• Do you publish/ share those commands?
• Do you sync the folder between your server and your workstation?

I have my scripts in $HOME/.scripts and softlink them from a directory in $PATH. Some of the scripts are versioned using Git, but the repository is private and I do not plan sharing them because the repoand the scripts scripts contain some not-tho-share information and mostly are simply not useful outside my carefully crafted and specific environment. If I want to share a script, I do it individually or make a proper public Git repository for it.

Since my server(s) and my workstations have different use cases I do not share any configuration between them. I share some configuration between different workstations, though. My dotfiles repository is mainly there for me to keep track of changes in my dotfiles.

is it bad practice to create a handful of commands

It becomes bad practice if it is against your personal or corporate guidelines regarding best practices. While it is not particularly bad or insecure, etc. to create bash scripts containing a single command, maybe use an alias instead. The $1 is automatically the first parameter after typing the alias in the shell.

alias podup="podman compose up -d"
alias poddown="podman compose down"
alias podlog="podman logs -f --tail 20"

Not quite sure about the podman syntax, if podman exec /bin/sh -it "$1" also works, you can use alias podenter="podman exec /bin/sh -it, Otherwise a simple function would do the trick.

Btw, if you ever wondered why Debian uses dash as /bin/sh (the switch was a bit annoying at the time), I think the reasoning was something like this:

• dash is a bit faster, which might have saved a second or two on boot times (this was before systemd). Same applies to compilation times, configure scripts run faster with dash.
• A bunch of #!/bin/sh scripts in Debian did not actually work if you replaced /bin/sh with another shell, which I guess some people wanted to do. Making dash the default /bin/sh forced everyone to fix their scripts.

Also some history on the abomination that is m4sh, famously used by GNU autoconf configure.ac scripts. Apparently when autoconf was released in 1991, there were still some Unix systems that shipped some 70s shells as the default /bin/sh. These shells do not support shell functions, which makes creating any sort of shell programming library pretty much impossible (I guess you could make a folder full of scripts instead of functions). They decided to use m4 preprocessor macros instead, as a sort of poor man’s replacement for functions.

In hindsight, it wish they had told commercial Unix sysadmins to install a proper /bin/sh or gtfo. But the GNU people thought it was important to make it as easy as possible to install free software even on commercial Unices.

Bash script for simple things (although Fish is my regular shell) and Node or Python scripts for complex things. Using #!/usr/bin/env node works just like it would for Bash so you know.

Several things

• write bash and nothing else (except posix sh)
• find a good way to take notes. It shouldn’t be in your bashrc
• only write fish for fish config
• use $!/usr/bin/env bash