Can you summarize what made the switch worthwhile for you?
I've been using zsh/omz with a customized powerlevel10k for years now and I see fish mentioned all the time but I didn't really find a compelling reason to try it out for myself.
My niche use case is that since fish comes out of the box with a whole bunch of very nice features (syntax highlighting on the shell, really good autocompletion) it's easy for me to stick it into the set of shell scripts I use to turn a new Ubuntu VM into a dev productivity machine. [1]
I could probably get Zsh to do all the same stuff I would want fish to do out of the box, but it would make reading through and reasoning about the shell scripts more difficult for third parties to audit, who may not be familiar with either tool before they try it.
only a few things that are probally also possible in zsh
fish functions are super easy to create on the fly from a command
updating $path is fish_add_path {path here}
setting global variable is easy set -Ux FOO bar adding -g makes it global
most of these require me to update my .zshrc and source ~/.zshrc as far as i am aware
You just install it and it works. I change devices, or wipe out my devices often enough that just having a familiar shell installed in one line feels comfy. I usually don't customize things a lot, I pick something with good default and rarely change anything on it. I don't want to bother downloading plugins and addons, or fonts, or themes, just for a command interpreter.
For me it came down to speed and maintenance. Fish is incredibly fast and I spend almost no time configuring it, so it’s trivial to bring up on a new system and I don’t have to exercise self-control to avoid tweaking it (that’s a personal failing, not other shells).
It's easy to install. I used to have some fisher plugins, but gave up on those since none of them were really essential.
When I set up a new computer I can just install fish and starship prompt and I'm mostly done. My fish configuration file is just aliasing and sourcing configs from other software, it's been a while since I had to touch it.
I start using fish, really like its feature set, but then have to give up and switch back to zsh when I want to get anything done. Unlike zsh, Fish isn't compatible with bash syntax, so a lot of existing scripts and software just doesn't work right. Even today people write scripts as if bash already exists and they seem to always target it. I wish fish had some form of compatibility layer.
Are sourcing bash scripts in a fish shell? If you can either run bash ./xyz or add the shebang #!/bin/bash to run scripts with the intended interpreter
Fish runs bash scripts just fine, as it does any script with a shebang line. IT is increasingly compatible with random CLI commands copy and pasted from the internet.
For those that don't work right away, just type `bash` to enter bash shell, run the command return back to Fish.
If a significant part of your day is spent copying CLI command from the internet, then yes, perhaps Bash is a better choice for you.
zsh, oh-my-zsh, p10k, fzf, and as of recently zsh-autosuggestions plugin for oh-my-zsh is what I ride on. I keep hearing about fish, and I see what it's doing but I'm not sure what it would bring to the table compared to the setup?
So fish comes with nice defaults. It could happen that you have a complicated zsh setup, then migrate to fish and you find the defaults are fine.
The fish scripting language is nicer (more logical) than the zsh one. That's because zsh needs to be Bourne-ish for historical reasons, and fish has decided to deviate from it. Existing sh, bash, zsh scripts will continue to work just fine. You can write new scripts in fish, or keep writing scripts in zsh or whatever.
Yeah I tried zsh a few times and never really thought it gave me that much relative to my existing bash config.
Fish on the other hand has stuck, and I'm still discovering nifty new built in features, like the fact that if you type `kill`, you can tab-complete on process names, so like if I want to kill emacs, I can type `kill ema<tab>` and select from the following:
If there's only one match it completes the PID for you, and of course if you start typing a PID and press tab it shows the process names along with the potential completions.
I use starship for prompt customization, which is nice because my config transfers just fine from bash over to fish.
FWIW, that is also the default behaviour of zsh's standard completion subsystem for kill. You can try it yourself: "zsh -f" to get a shell without reading a fancy config, start compinit, "kill <Tab>".
You can even disable verbose mode via zstyle, should you be the type of person that likes a simple pid list like bash's completion project provides. To me this level of customization is the main feature zsh provides, but not everyone incessantly turns every knob they see.
PSA: simply installing this won't get you many benefits until you read about what it can do and then make an effort to use those things in your workflow.
Which is a major selling point. We do not want anything unique.
Ripgrep accept most basic flags that grep uses and is way faster (when searching for code under VC), and that makes using it very convenient, even before considering the features on top.
I would prefer that you also write your HN comments in Rust, if that's possible, for better performance and safety. In the future, there will only be Rust. You have to accept Rust into your life.
Fish has been great for years.The Rust rewrite will help the devs catch issues earlier in development. Hardly a main a feature.
The awesome level tab-completion and auto-completion are the firs things to notice about Fish. Also, overall cleaner syntax and a several nice built-in commands that don't ship with Bash.
Not sure if you're talking about starship or fish, but starship is by far my favorite prompt customization tool, regardless of what language it's written in. It works across bash, zsh, fish, and others, so you can pretty easily get the same prompt in any shell on your system. It's easy to tweak as needed, but is pretty great out of the box.
I'm a developer of 10 years, and use the command line for Git, Brew and a few other things, though I'm normally an IDE developer. I believe the CL is as powerful as everyone says, but what are some useful things I can use it for? I use Zsh on a Mac some of the time.
Some folks grew up on systems that could fail at any time and leave them with nothing but the tools in /sbin to fix it with. No /usr, no bash/zsh, just sh and vi if you were lucky. You learn to love the cool new things, but you still have a mental bag packed with fsck and ed in case you need to bug out.
It's silly, really because no systems are like that any more, but the things you learned at 2am when you were 22 tend to stick with you.
I strive to speak multiple languages, but I think it still worthwhile to be really good at my primary language of English.
Same with computers and servers. I interact with a lot of them, but my primary computer that I use >80% of the time is worthwhile to optimize with some tooling that's not available on all the others.
You bet wrong, because naturally approaching 50y, I have my tools for quite some time, for the stuff I care about, not random examples on the Internet.
A UNIX shell is a very bad approximation of a real programming language REPL.
It is no accident that sh scripts turn into Perl, Python, Tcl, Lisp,... when one wants to keep sanity.
As we're in a zsh story, I'll point out it comes with a clever file renaming interface out of the box¹. It allows full access to all the advanced glob operators zsh provides too².
On the second one - if that’s in code than a good IDE will not only do it with a single keypress, but will also (optionally) do it intelligently eg change only in the code, not the comments, show you a browseable preview, and offer undo should you get it wrong.
I like the shell but I also think IDEs are underrated.
Colorful prompt with context information, powerline glyphs, autocomplete, command highlighting, easy navigation through strings, plugins and a plugin ecosystem that extends your shell like crazy
Rename hundreds of files at a time. Find stupid errors in GB sized csv/json/.. data files. Automate git bisect. Automate workflows by writing a short loop or 3 commands in a row you can easily repeat. Down/upload big datasets. Convert text files any way you can think of.
For #2 once you’ve got a command you like, maybe one that generates a metric, you can now re-run it with ease from your history, saved text file, or an alias. You could also share it with teammates.
Please correct me if there are other ways to achieve the above, shell is the only way I know
- CLI encourages automation (shell commands are easier to repeat/parametrize then GUI steps)
- shell instructions are more git friendly than screenshots (even if a process can't/shouldn't be automated. Commands are easier to document in a reusable/updatable way)
- CLI can be used to run tests in a way similar how CI does it (more reproducible)
Command line tools are easier to extend/combine with existing pipelines (rg,jq,xargs): read input from stdin/write output to stdout, report on stderr, return non-zero code on error. It enables you to create adhoc tools that you wouldn't bother otherwise (unrelated: LLMs also have this property by making your skill set much broader (though very shallow with current LLMs)).
Shell also has Forth-like property (compose with: retry, timeout, setuid, exec, xargs, env, ssh, etc) https://www.oilshell.org/blog/2017/01/13.html
ZSH + Zimfw + Powerlevel10k is probably the fastest "bling"-y shell I managed to put together. Would usually smoke Fish, but I haven't compared recently
I've never noticed slowness in OMZ, or really any shell. Or at least, when something feels slow I've always assumed it's the application I'm using, or network congestion, not the shell itself. In what situations do you notice a shell being slow?
Initializing a new session is not instantaneous. Even on my M3 Max it probably takes a couple hundred milliseconds, which is fine.
When I first tried it over a decade back, IIRC on an 11" Air, each new session was probably closer to a couple seconds. I recall there were a couple plugins in particular that were major bottlenecks. Replaced those and got the init speed to about half that but it annoyed me enough to do a clean zsh setup.
It turned slow for a whole after I installed anaconda. If you use conda and your shell is slow, try commenting out the code that loads conda, because it's probably that.
nvm likes to slow things down too. Not as bad as conda, but I did add some code to delay load it.
Do you know how fish would prevent getting slowed down by external scripts in its configuration file? Or does it just not support those scripts perhaps?
It's wild to me how slow conda is. Just in ordinary day to day use, with nothing weird, you sometimes need to wait insane amounts of time, even on incredibly fast systems. Whatever they're doing, it's wrong and bad.
Started using fish, moved back in 3 minutes. Remembered I have this awesome shortcut in Zsh where I type what I need to do, hit cmd+G, and some ChatGPT replaces my input with an example command.
I backtracked all the way to Bash. Because I seem to write some amounts of bash scripts, it seems living inside bash is helpful. And it's really not much I feel like I am missing in my use from zsh (or fish). Bash even supports CTRL+R for history with search nowadays.
Yes, I've had to get clear where they differ, but the Fish experience is that much better that it seems worth it.
I've considered replacing more my bash scripts with Fish scripts, but it seems Fish has a slower start start-up time for scripting (not that it matters most of the time) and is more designed for interactive use.
My only issue with fish is that it has had a standing issue[1] for over 12 years where functions and blocks cannot be backgrounded. This makes backgrounding a series of commands in a script nearly impossible.
The defaults it ships out of the box makes the shell actually usable. Unsure I could ever go back to a regular bash/zsh prompt.
A lot of people will tell you this is slow and you've got to use X,Y,Z instead. If you're new, I'd strongly recommend just sticking with this, it's much easier to configure.
zsh autosuggestions has been pretty slow for me compared to fish out of the box, switched a year or so back and have been pretty happy. maybe something has changed now :thinking-emoji:
oh-my-fish is pretty much similar in capabilities and has a good assortment of themes.
All I used that came with zsh were git shortcuts like gc -m "comment", gst for git status which i just recreated in my fish config.
Do yourself a favor and try to configure your shell without ohmyzsh first. You’d be surprised at how few plugins actually require it, and you save yourself from horrible performance.
Yeah I have a ton of customization including a really sweet looking prompt (that includes branch name & even git remote project name). Shortcut aliases and custom functions for everything that I use often. Atuin for Ctrl+R history searching.
All without using any “plugins” or ohmyzsh. All just within my .zshrc, and a few extra files that get “sourced” depending which computer I’m on. Only occasional copy-pasting a few things I liked from StackOverflow.
Basically instantaneous speed. Don’t even notice any performance penalties from all of the customization and styling.
I did have a performance issue early on, but I profiled my zsh startup time and saw 97.5% of my shell startup was consumed by NVM (node version manager), and another 2.4% was RVM (ruby). Switched to FNM and Chruby instead, and it was zippy.
And it’s all mine, so there are no “updates” to break anything. Throw it all in source control and you can keep improving it without risk of losing your previous working config.
I do myself a favor by not doing this, install ohmyzsh and start working. Seeing what other people have done with shell, vim and all sorts of things, I know this is a rabbit hole and I could spend endless time on it. That's why I use ohmyzsh, VSCode and other tools and only tweak settings when necessary. And I can be actually productive even when I get a new machine.
Performance? If you are talking about the startup performance, no it doesn't bother me. Again, nothing is slowly enough to cause noticeable delays in basic typing/editing and the bottleneck is almost never on them. (I regularly work on codebase of hundreds to thousands of files or more.) So I don't spend time worrying about saving a few seconds in total per day when I can use my brain elsewhere.
You don’t have to follow a rabbit hole, these tweaks take no time at all and you can stop whenever. I used Oh My Zsh until I got fed up with the startup slowness. I open a bunch of terminal windows during the day and every one of them had a second or two of delay before I could do anything. I decided to do myself a favour and no longer put up with it. It took me a handful of minutes to figure out which plugins I actually cared about (three) and get rid of Oh My Zsh entirely. Everything about this change is better for me: it’s easier and faster to redo the setup when I do a clean install, and it’s instant in everyday use which means it’s no longer frustrating.
If you add more than a handful of plugins in any of these IDE layers for Neovim, like LunarVim, LazyVim, AstroNvim, &c., it can slow things down to an unacceptable level quickly.
No need to rely on your terminal to provide performance. LunarVim is very slow even though I use Alacritty, it was the main reason I stopped using it and switched to LazyVim.
If I'm being quite honest, I think LazyVim is the more polished project. I may end up switching to that; I've been impressed by what I've seen so far (as well as using the lazy.nvim package manager).
While I love p10k and use it myself, note that p10k is just a theme, while ohmyzsh is a theme manager (that comes with default themes) + plugin manager + a collection of aliases + other QoL stuff.
p10k is not just a theme, it implements a whole prompt to realize features like instant and transient prompt. For example showing your current Kubernetes namespace while you are typing a kubectl command.
Yep, I use zsh with 2 plugins. One for syntax highlighting commands and another for showing auto-suggestions. It's really fast. The rest is nearly a default zsh set up in terms of zsh configuration. Everything is documented in my dotfiles https://github.com/nickjj/dotfiles.
My prompt is a 1 liner that shows your git branch as well as coloring up $ to be red or not based on if the last command failed. Coincidentally I just released a blog post today on coloring up your prompt based on if the last command failed at https://nickjanetakis.com/blog/color-your-shell-prompt-red-i... , there's solutions for both zsh and bash.
* Exporting shell variables such as HISTFILE isn't necessary, and can be unsafe. For example, if you decide to pop open unconfigured bash from within a zsh session it would intermingle its history on write.
* There is a smart keyboard subsystem that neatly manages terminal differences¹, making it much nicer compared to inserting {system,term}-specific escapes in to your config.
Hoping this comes across in the helpful spirit it was intended, not as nitpicky unwanted code review.
* For shell history I don't think I've encountered that issue yet. For example if I have zsh loaded and run `bash` and then within that bash session run `whoami` and exit bash, my `history` within zsh doesn't include the whoami command, only `bash`. Is there another workflow that would produce intermingled history?
* Ah, this is likely in reference to the home / end / insert / etc. key binds I added? I think I came up with those from following a 2007 blog post[0] which I found here[1]. Do you know what the nicer versions would be for those specific keys? The docs are coming up empty with specific examples and most Google results return the escaped references.
For the first; Unless your bash config overrides the environment variable you've set it will use $HISTFILE from the calling shell, which is why "HISTFILE=<location>" without the export fixes it. You can test it with:
It isn't just intermingling either, bash may truncate the file depending on how it is configured too. You can see this by performing the same procedure as above but starting with 1000 lines in the trash file, bash will truncate the file to 500 lines by default on exit. It shouldn't happen in your case as you're also exporting HISTSIZE, but it can if you have a lot of multiline history events as bash treats them differently to zsh.
The point wasn't just about HISTFILE really though, very few of the shells own configuration variables are useful in child processes. (LC_* and PATH being the obvious exceptions that spring to mind.)
---
For the second; The doc I linked to above shows how to use zkbd to handle terminal differences in a clean way. Run the wizard, and then you can use the $key array as in $key[Home]. The neat thing to do with zkbd is have the wizard run on startup when it can't find its definitions file, that way you'll get correct behaviour whenever you play with a new terminal type.
Another option is to use the terminfo database¹ directly, if you trust it to be correct for your terminals. "zmodload terminfo", and then the database is available through the $terminfo array as in $terminfo[khome] or using the echoti function².
What felt like an initial thirty second throwaway comment yesterday has turned in to a pile of replies that feel like a poorly conceived Ted talk now, forgive me for that.
The dividing line between variables that should be prefixed by export are ones that are generally useful to child processes and ones that are not.
For example, $PATH is useful and should be exported. You probably want scripts you start to have access to all the things you have installed, not just the locations of the default search path:
PATH= =zsh -fc 'print -l $path'
$HISTFILE/$PROMPT/etc control current shell behaviour and are unlikely to be of any use to child processes executed by the shell, and do not need to be exported. Remembering, of course, that they'll be set by any new interactive shells anyway as they too will read their startup files when launched.
Oftentimes it doesn't make much difference. However, some such as $CDPATH can cause non-obvious bugs when they're exported, and some such as $HISTFILE can cause data corruption or even loss when the planets align against you.
Of course, their config is the best according to the benchmark (and ohmyzsh is the slowest option), but DIY configs are also covered, particularly possible performance optimizations.
Like practically everything zsh, it can be configured. The select-word-style zle widget¹ ships with zsh, and has a bash compatible setting(among others). The interface is super powerful, and given that you can change it in individual zle widgets you can make it context aware if you wish too.
Anything after (besides the path and the GPG setup) that is an add-on that I can remove and still function well. I've learned to live without Git info. But this is very personal and I have not been into active development for quite a while.
Loading time is not an issue for me. I run tmux so my sessions stay alive and performance is very quick. M2 MacBook Air with 16GB RAM, using Warp terminal. Using OMZ with Dracula Theme and mosh for remote servers.
I don't either. -hell, since I found out that "set -o emacs" gives me history and autocompletion I go even further and usually only use /bin/sh on NetBSD.
I'm not sure what the use case for zsh and fish are, but I'm guessing it's different than mine.
(edit/update: apparently, judging by a quicky netbsd 10 vm install, you don't even need to explicitly "set -o emacs". /bin/sh does it by default)
I use and like omyzsh, but find it super annoying how it deals with updating. It either nags you constantly to update, or if you set it to "auto," then it's constantly spamming you about how it's updating, and when you create a new shell session (say, in tmux), it causes significant lag before you can use the new shell and pollutes the console with tons of output. There has to be a better way, like some optional mode that just runs an update service in 3am silently using systemd.
I used zsh for a few years before oh-my-zsh became a thing.
I have to admit that I don't find it more convenient.
It takes an ecosystem where things have a way of working, and establishes new conventions on top of those using a new set of environment variables. So if you know zsh, you don't know oh-my-zsh.
A vanilla zsh will feel very bare.
A vanilla oh-my-zsh will feel very feature rich.
If you end up configuring zsh anyways, I don't see the point at all.
153 comments
[ 4.1 ms ] story [ 219 ms ] thread[0] https://github.com/romkatv/powerlevel10k [1] https://fishshell.com/
I've been using zsh/omz with a customized powerlevel10k for years now and I see fish mentioned all the time but I didn't really find a compelling reason to try it out for myself.
I could probably get Zsh to do all the same stuff I would want fish to do out of the box, but it would make reading through and reasoning about the shell scripts more difficult for third parties to audit, who may not be familiar with either tool before they try it.
[1]: https://github.com/hiAndrewQuinn/shell-bling-ubuntu
fish functions are super easy to create on the fly from a command updating $path is fish_add_path {path here} setting global variable is easy set -Ux FOO bar adding -g makes it global
most of these require me to update my .zshrc and source ~/.zshrc as far as i am aware
When I set up a new computer I can just install fish and starship prompt and I'm mostly done. My fish configuration file is just aliasing and sourcing configs from other software, it's been a while since I had to touch it.
For all the rest, perhaps give https://github.com/edc/bass a try.
Don't even change your default shell with chsh ,rather just point your terminal emulator to launch fish
For those that don't work right away, just type `bash` to enter bash shell, run the command return back to Fish.
If a significant part of your day is spent copying CLI command from the internet, then yes, perhaps Bash is a better choice for you.
Nearly everything I want, fancy autocompletion, syntax highlighting, git-aware prompt just worked out of the box.
The fish scripting language is nicer (more logical) than the zsh one. That's because zsh needs to be Bourne-ish for historical reasons, and fish has decided to deviate from it. Existing sh, bash, zsh scripts will continue to work just fine. You can write new scripts in fish, or keep writing scripts in zsh or whatever.
Fish on the other hand has stuck, and I'm still discovering nifty new built in features, like the fact that if you type `kill`, you can tab-complete on process names, so like if I want to kill emacs, I can type `kill ema<tab>` and select from the following:
If there's only one match it completes the PID for you, and of course if you start typing a PID and press tab it shows the process names along with the potential completions.I use starship for prompt customization, which is nice because my config transfers just fine from bash over to fish.
You can even disable verbose mode via zstyle, should you be the type of person that likes a simple pid list like bash's completion project provides. To me this level of customization is the main feature zsh provides, but not everyone incessantly turns every knob they see.
And the website is amazing. “Finally, a command line shell for the 90s”
https://github.com/IlanCosman/tide
1. https://github.com/ohmyzsh/ohmyzsh/wiki/Cheatsheet
Still wanting to try out fish and hopefully soon!
Ripgrep accept most basic flags that grep uses and is way faster (when searching for code under VC), and that makes using it very convenient, even before considering the features on top.
I would prefer that you also write your HN comments in Rust, if that's possible, for better performance and safety. In the future, there will only be Rust. You have to accept Rust into your life.
The awesome level tab-completion and auto-completion are the firs things to notice about Fish. Also, overall cleaner syntax and a several nice built-in commands that don't ship with Bash.
Soon a Rust userland will be table stakes.
I mean, surely you already know?
Anything where feeding input between different tasks, or where you want to use regex, or similar things the cl will be superior.
The idea is to be able to use any computer/system/devices easily instead of one perfectly.
Some folks grew up on systems that could fail at any time and leave them with nothing but the tools in /sbin to fix it with. No /usr, no bash/zsh, just sh and vi if you were lucky. You learn to love the cool new things, but you still have a mental bag packed with fsck and ed in case you need to bug out.
It's silly, really because no systems are like that any more, but the things you learned at 2am when you were 22 tend to stick with you.
Same with computers and servers. I interact with a lot of them, but my primary computer that I use >80% of the time is worthwhile to optimize with some tooling that's not available on all the others.
And a shell is a REPL.
A UNIX shell is a very bad approximation of a real programming language REPL.
It is no accident that sh scripts turn into Perl, Python, Tcl, Lisp,... when one wants to keep sanity.
¹ https://zsh.sourceforge.io/Doc/Release/User-Contributions.ht...
² https://zsh.sourceforge.io/Doc/Release/Expansion.html#Filena...
I like the shell but I also think IDEs are underrated.
Pick any X and Y that maximises your ability to feel superior, I suppose.
For #1 this blog post has some good examples of quickly chaining together commands (though personally I rarely use xargs) https://drewdevault.com/2020/12/12/Shell-literacy.html
For #2 once you’ve got a command you like, maybe one that generates a metric, you can now re-run it with ease from your history, saved text file, or an alias. You could also share it with teammates.
Please correct me if there are other ways to achieve the above, shell is the only way I know
- CLI encourages automation (shell commands are easier to repeat/parametrize then GUI steps) - shell instructions are more git friendly than screenshots (even if a process can't/shouldn't be automated. Commands are easier to document in a reusable/updatable way) - CLI can be used to run tests in a way similar how CI does it (more reproducible)
Command line tools are easier to extend/combine with existing pipelines (rg,jq,xargs): read input from stdin/write output to stdout, report on stderr, return non-zero code on error. It enables you to create adhoc tools that you wouldn't bother otherwise (unrelated: LLMs also have this property by making your skill set much broader (though very shallow with current LLMs)). Shell also has Forth-like property (compose with: retry, timeout, setuid, exec, xargs, env, ssh, etc) https://www.oilshell.org/blog/2017/01/13.html
Paired with zoxide and fzf for the zsh keyboard shortcuts are all I need to be very productive.
Starship.rs is very useful as well as a theme but pure and pl10k are also great.
Now it’s even rewritten in Rust!
Then you get trapped by the utility and struggle when you land on some remote server's bash shell.
https://github.com/jorgebucaran/fisher
https://github.com/jorgebucaran/awsm.fish#readme
When I first tried it over a decade back, IIRC on an 11" Air, each new session was probably closer to a couple seconds. I recall there were a couple plugins in particular that were major bottlenecks. Replaced those and got the init speed to about half that but it annoyed me enough to do a clean zsh setup.
It turned slow for a whole after I installed anaconda. If you use conda and your shell is slow, try commenting out the code that loads conda, because it's probably that.
nvm likes to slow things down too. Not as bad as conda, but I did add some code to delay load it.
Do you know how fish would prevent getting slowed down by external scripts in its configuration file? Or does it just not support those scripts perhaps?
https://github.com/Schniz/fnm
Yes, I've had to get clear where they differ, but the Fish experience is that much better that it seems worth it.
I've considered replacing more my bash scripts with Fish scripts, but it seems Fish has a slower start start-up time for scripting (not that it matters most of the time) and is more designed for interactive use.
[1]: https://github.com/fish-shell/fish-shell/issues/238
A lot of people will tell you this is slow and you've got to use X,Y,Z instead. If you're new, I'd strongly recommend just sticking with this, it's much easier to configure.
oh-my-fish is pretty much similar in capabilities and has a good assortment of themes.
All I used that came with zsh were git shortcuts like gc -m "comment", gst for git status which i just recreated in my fish config.
Does fish support something similar? If so then I could be encouraged to give it a whirl.
Removing omz, adding antidote and starship. feels (and measureably is) faster and snappier.
All without using any “plugins” or ohmyzsh. All just within my .zshrc, and a few extra files that get “sourced” depending which computer I’m on. Only occasional copy-pasting a few things I liked from StackOverflow.
Basically instantaneous speed. Don’t even notice any performance penalties from all of the customization and styling.
I did have a performance issue early on, but I profiled my zsh startup time and saw 97.5% of my shell startup was consumed by NVM (node version manager), and another 2.4% was RVM (ruby). Switched to FNM and Chruby instead, and it was zippy.
And it’s all mine, so there are no “updates” to break anything. Throw it all in source control and you can keep improving it without risk of losing your previous working config.
But lately I've become really happy with just installing spaceship.
Sure, it doesn't look exactly like I want it to. But it shows git status.
Various features just "pop up" if I install the right things on the system.
https://starship.rs/
Would recommend using a PS1 generator, e.g. [1].
[1] https://bash-prompt-generator.org/
Performance? If you are talking about the startup performance, no it doesn't bother me. Again, nothing is slowly enough to cause noticeable delays in basic typing/editing and the bottleneck is almost never on them. (I regularly work on codebase of hundreds to thousands of files or more.) So I don't spend time worrying about saving a few seconds in total per day when I can use my brain elsewhere.
I had slow starts when I added nvm completion to .zshrc and .. maybe conda environment but that was unrelated to oh-my-zsh
This was years ago, no idea how it is today. But I also have no reason to go back.
> I had slow starts when I added nvm completion to .zshrc
That too, but that’s unrelated. To me that was yet another reason to stop using node altogether, I wasn’t enjoying it anyway.
'Wansmer/symbol-usage.nvim' is one of them. I found that I didn't really need it as much so disabled it and LunarVim became much snappier.
https://github.com/romkatv/powerlevel10k
My prompt is a 1 liner that shows your git branch as well as coloring up $ to be red or not based on if the last command failed. Coincidentally I just released a blog post today on coloring up your prompt based on if the last command failed at https://nickjanetakis.com/blog/color-your-shell-prompt-red-i... , there's solutions for both zsh and bash.
* Exporting shell variables such as HISTFILE isn't necessary, and can be unsafe. For example, if you decide to pop open unconfigured bash from within a zsh session it would intermingle its history on write.
* There is a smart keyboard subsystem that neatly manages terminal differences¹, making it much nicer compared to inserting {system,term}-specific escapes in to your config.
Hoping this comes across in the helpful spirit it was intended, not as nitpicky unwanted code review.
¹ https://zsh.sourceforge.io/Doc/Release/User-Contributions.ht...
* For shell history I don't think I've encountered that issue yet. For example if I have zsh loaded and run `bash` and then within that bash session run `whoami` and exit bash, my `history` within zsh doesn't include the whoami command, only `bash`. Is there another workflow that would produce intermingled history?
* Ah, this is likely in reference to the home / end / insert / etc. key binds I added? I think I came up with those from following a 2007 blog post[0] which I found here[1]. Do you know what the nicer versions would be for those specific keys? The docs are coming up empty with specific examples and most Google results return the escaped references.
[0]: https://blog.andrewbeacock.com/2007/08/how-to-get-home-end-k...
[1]: https://stackoverflow.com/questions/8638012/fix-key-settings...
The point wasn't just about HISTFILE really though, very few of the shells own configuration variables are useful in child processes. (LC_* and PATH being the obvious exceptions that spring to mind.)
---
For the second; The doc I linked to above shows how to use zkbd to handle terminal differences in a clean way. Run the wizard, and then you can use the $key array as in $key[Home]. The neat thing to do with zkbd is have the wizard run on startup when it can't find its definitions file, that way you'll get correct behaviour whenever you play with a new terminal type.
Another option is to use the terminfo database¹ directly, if you trust it to be correct for your terminals. "zmodload terminfo", and then the database is available through the $terminfo array as in $terminfo[khome] or using the echoti function².
¹ https://zsh.sourceforge.io/Doc/Release/Zsh-Modules.html#The-...
² terminfo(5) contains the mappings, but they're often far less readable than the $key array. $terminfo[kpp] vs $key[PageUp], for example.
The dividing line between variables that should be prefixed by export are ones that are generally useful to child processes and ones that are not.
For example, $PATH is useful and should be exported. You probably want scripts you start to have access to all the things you have installed, not just the locations of the default search path:
$HISTFILE/$PROMPT/etc control current shell behaviour and are unlikely to be of any use to child processes executed by the shell, and do not need to be exported. Remembering, of course, that they'll be set by any new interactive shells anyway as they too will read their startup files when launched.Oftentimes it doesn't make much difference. However, some such as $CDPATH can cause non-obvious bugs when they're exported, and some such as $HISTFILE can cause data corruption or even loss when the planets align against you.
There was just enough time after your talk for 1 more question from the audience:
For my zshrc that you poked around in, is this issue related specifically to HISTFILE or all of the history related env vars?
Of course, their config is the best according to the benchmark (and ohmyzsh is the slowest option), but DIY configs are also covered, particularly possible performance optimizations.
¹ https://zsh.sourceforge.io/Doc/Release/User-Contributions.ht...
[0] https://github.com/liquidprompt/liquidprompt
I'm not sure what the use case for zsh and fish are, but I'm guessing it's different than mine.
(edit/update: apparently, judging by a quicky netbsd 10 vm install, you don't even need to explicitly "set -o emacs". /bin/sh does it by default)
Main culprits were language switchers (nvm, jabba, pyenv). Which I moved to lazy loading.
If I drop the ohmyzsh plugins startup time is 11ms. But, I want some quality of life.
There's a nice benchmarking command in the article if you want to test yours:
for i in $(seq 1 10); do /usr/bin/time $SHELL -i -c exit; done
Zsh already has a module system (autoload), a prompt framework (promptinit), and Git integration (vcs_info) builtin.
I have a pretty extensive Zsh setup in vanilla Zsh: https://github.com/cbarrick/dotfiles
It consists of: - collection of plugins/themes - auto update - way to customize them
Everything could be done directly in .zshrc, it's just a bit more convinient
I have to admit that I don't find it more convenient.
It takes an ecosystem where things have a way of working, and establishes new conventions on top of those using a new set of environment variables. So if you know zsh, you don't know oh-my-zsh.
A vanilla zsh will feel very bare.
A vanilla oh-my-zsh will feel very feature rich.
If you end up configuring zsh anyways, I don't see the point at all.