Modern Linux tools
ikrima.dev
ikrima.dev
Learn the classic tools, learn them well, and your life will be much easier.
Many folks nowadays don't get how lucky they are, not having to do UNIX development on a time-sharing system, although cloud systems kind of replicate the experience.
ed (pronounced as distinct letters, /ˌiːˈdiː/)[1] is a line editor for Unix and Unix-like operating systems. It was one of the first parts of the Unix operating system that was developed, in August 1969.[2] It remains part of the POSIX and Open Group standards for Unix-based operating systems
so it is a bug in those distros.
Only to feel totally handicapped when logging in into a busybox environment.
I'm glad I learned how to use vi, grep, sed..
My only change to an environment is the keyboard layout. I learned Colemak when I was young. Still enjoying it every day.
Agreed, but that doesn't stop you from using/learning alternatives. Just use your preferred option, based on what's available. I realise this could be too much to apply to something like a programming language (despite this, many of us know more than one) or a graphics application, but for something like a pager, it should be trivial to switch back and forth.
Awk and sed.
I like the idea of new tools though. But knowing the building blocks is useful. The “Unix power tools” book was useful to get me up to speed.. there are so many of these useful mini tools.
Miller is one I’ve made use of (it also was available for my distro)
Not everybody is a sysadmin manually logging into lots of independent, heterogeneous servers throughout the day.
Same goes for a bunch of other tools that have "modern" alternatives but the "classic" ones are already installed/available on most default distribution setups.
The infra may be cattle but debugging via anal probe err SSH is still the norm.
apt-get/pacman/dnf/brew install <everything that you need>
You'll need install those and other tools (your favorite browser, you favorite text editor, etc) anyway if you're changing your OS.
> or SSH anywhere
When you connect through SSH you don't have GUI and that's not a reason for avoiding using GUI tools, for example.
> even use a mix of these on my personal computer and the traditional ones elsewhere
I can't see the problem, really. I use some of those tools and they are convenient, but it doesn't matter that I can't work without that. For example, bat: it doesn't replace cat, it only outputs data with syntax highlight, makes my life easier but if I don't have it, ok.
The point is that sometimes you're SSHing to a lightweight headless server or something and you can't (or can't easily) install software.
I personally haves an ansible playbook to ~setup all my commonly used tooling on ~any cli I will use significantly; (almost) all local installs to avoid need for root. It runs in ~minute - and I have all the Niceties. If it's not worth spending that minute to run; then i won't be on the machine long enough for it to matter.
^^ Yep. Totally this. I've become entirely too accustomed to all the little niceties of a well-crafted toolchain that covers all my needs at any given moment. It was worth the time invested to automate installing and configuring all the fancy newfangled stuff I've built up muscle-memory for. :)
One major difference can emerge from the fact that using a tool regularly inevitably builds muscle memory.
You’re accustomed to a replacement command-line tool? Then your muscle memory will punish you hard when you’re logged into an SSH session on another machine because you’re going to try running your replacement tool eventually.
You’re used to a GUI tool? Will likely bite you much less in that scenario.
Yes.
> Then your muscle memory will punish you hard
No.
I'm also used to pt-br keyboards, it's easier to type in my native language, but it's ok if I need to use US keyboards. In terms of muscle memory, keyboards are far harder to adapt.
A non-tech example: if I go to a Japanese restaurant, I'll use chopsticks and I'm ok with them. At home, I use forks and knives because they make my life easier. I won't force myself to use chopsticks everyday only for being prepared for Japanese restaurants.
If only it were so simple. Not every tool comes from a package with the same name, (delta is git-delta, "z" is zoxide, which I'm not sure I'd remember off the top of my head when installing on a new system). On top of that, you might not like the defaults of every tool, so you'll have config files that you need to copy over or recreate (and hopefully sync between the computers where you use these tools).
That said I do think nix provides some good solutions for this. It gives you a nice clean way to list the packages you want in a nixfile and also to set their defaults and/or provide some configuration files. It does still require some maintenance (and I choose to install the config files as editable, which is not very nix-y, but I'd rather edit it and then commit the changes to my configs repo for future deploys than to have to edit and redeploy for every minor or exploratory change), but I've found it's much better than trying maintain some sort of `apt-get install [packages]` script.
But yes, in the eventual case that I don't have Nix I can very much use the classic tools. It is not a binary choice, you can have both.
For example, in $CURRENT_JOB we have a bastion host that gives access to the databases (not going to discuss if this is a good idea or not, this is how my company does). 90% of time I can do whatever I need just with what the bastion host offers (that doesn't have Nix), if I need to do further analysis I can copy some files between the bastion host and my computer to do further analysis.
I can just drop it into the environment and pull in tools that I need using nix.
Some of them are smart but sometimes I want dumb, for example, ripgrep respects gitignore, and often, I don't want that. Though in this case, there is an option to turn it off (-uuu). That's a common theme with these tools too, they are trying to be smart by default and you need option to make them dumb.
So no, these tools are not "objectively superior", they are generally more advanced, but it is not always what you need. They complement classic tools, but in no way replace them.
IMO, it's worth spending some time to clean up your setup for smooth transition to new machines in the future.
I never asked for such behaviour, and I have no time for pretty "modern" opinions in a base software.
Often, when I read "modern", I read "immature".
I am not ready to replace my stable base utilities for some immature ones having behaviour changes.
The scripts I wrote 5 years ago must work as is.
> ripgrep is a line-oriented search tool that recursively searches the current directory for a regex pattern. By default, ripgrep will respect gitignore rules and automatically skip hidden files/directories and binary files. (To disable all automatic filtering by default, use rg -uuu.)
Because hey guess what: you can still use grep! So I built something different.
Note: if you want to make ripgrep not do .gitignore filtering, set `RIPGREP_CONFIG_PATH` to point to a config file that contains `-uu`.
Sources:
- https://github.com/BurntSushi/ripgrep/blob/master/GUIDE.md#c...
- https://github.com/BurntSushi/ripgrep/blob/master/GUIDE.md#a...
That's on me!
--smart-case --no-messages --hidden --ignore-vcs
and then point to it with
.zshenv 3:export RIPGREP_CONFIG_PATH="$HOME/.rgrc"
Not perfect and sometimes I reach for good old fashioned escaped \grep but most of the time it's fine.
Regarding ripgrep: if it's not bug-for-bug compatible with grep, it’s deemed useless. Yet, if it is identical, then why bother using it at all? What kind of logic is that?
Wow, that is so cool. This looks a lot more approachable than other sandboxing tools.
E.g. I have ls set up aliased to eza as part of my custom set of configuration scripts. eza pretty much works as ls in most scenarios.
If I'm in an environment which I control and is all configured as I like it, then I get a shinier ls with some nice defaults.
If I'm in another environment then ls still works without any extra thought, and the muscle memory is the same, and I haven't lost anything.
If there's a tool which works very differently to the standard suite, then it really has to be pulling its weight before I consider using it.
Then use grep, what’s your point? grep is not going away because ripgrep is better, but ripgrep might become more available?
I also notice you’re saying "if", so you’re not. So again, what’s your point?
Often there are plenty of of paths open to getting a decent environment as you go:
Mostly, I rely on ansible scripts to install and configure the tools I use.
One fallback I haven't seen mentioned, that can get a lot of mileage from it: use sshfs to mount the target system locally. This allows you to use local tool & setup effectively against another machine!
[1] Not like the time one of my friends "wardialed" every number in my local calling area and posted the list to a BBS and I found that some of them could be logged into with "uucp/uucp" and the like. I think Bell security knew he rang everybody's phone in the area but decided to let billing handle the problem because his parents had measured service.
I have a chef cookbook that sets up all the tools I like to have on my VMs. When I bootstrap a VM it includes all the stuff I want like fish shell and other things that aren’t standard. The chef cookbook also manages my SSH keys and settings.
not really contradicted by:
> exa: modern replacement for ls/tree, not maintained
I have a very limited set of additional tools I tend to install on systems, and they are in my default ansible-config, so will end up on systems quickly, but I try to keep this list short and sweet.
95% of the systems I manage are debian or ubuntu, so they will use mostly the same baseline, and I then add stuff like ack, etckeeper, vim, pv, dstat.
As a greybeard linux admin, I agree with you though. This is why when someone tells me they are learning linux the first thing I tell them is to just type "info" into the terminal and read the whole thing, and that will put them ahead of 90% of admins. What I don't say is why: Because knowing what tooling is available as a built-in you can modularly script around that already has good docs is basically the linux philosophy in practice.
Of course, we remember the days where systems only had vi and not even nano was a default, but since these days we do idempotent ci/cd configs, adding a tui-editor of choice should be trivial.
What are you talking about? I'm still living those days in modern day AWS with latest EC2 machines!
The ones that leap out at me are ripgrep (a genuinely excellent recursive grepper), jq (a JSON processor - there is no alternative to this in the standard unix toolkit), and hyperfine (benchmarking).
Is something like that possible? Seems like you could conceivablely dump these small file size tools into a temp folder and and use them and that could be automated.
Is there a security issue with that? Do any of these tools need more permission than the remote session would have?
Maybe the main issue is portability of these apps?
This is certainly a common sentiment (I've felt it myself) so is it at all possible?
> cat clone with syntax highlighting and git integration
doesn't make any sense because cat is not really meant for viewing files. You should be comparing your tool with the more/less/most family of tools, some of which can already do syntax highlighting or even more complex transforms.
Actual LOL. Indeed. I was working for a large corporation at one point and a development team was explaining their product. I asked what its differentiators were versus our competitors. The team replied that ours was written in Go. #faceplam
Sure. You can drive really slowly in a sports car. But if you're looking for travel options for a long distance journey are you going to pick the sports car or the bicycle.
Also I have actually yet to find slow unmaintainable brittle garbage written in Go or Rust. I'm sure it's possible but it's vastly less likely.
Citation needed. /s
This is called "the fallacy of grey". Very common.
https://www.lesswrong.com/posts/dLJv2CoRCgeC2mPgj/the-fallac...
An example from my personal experience: I used to think that oxipng was just a faster optipng. I took a closer look recently and saw that it is more than that.
See: https://op111.net/posts/2025/09/png-compression-oxipng-optip...
That's the problem. How good they are ? Who can tell ? The basic UNIX tools didn't come in "one day" like most of these "rust tools".
I just enjoy seeing others incrementally improve on our collective tool chest. Even if the new tool isn’t of use to me, I appreciate the work that went into it. They’re wonderful tools in their own right. Often adding a few modern touches to make a great tool just a little bit better.
Thank you to those who have put in so much effort. You’re making the community objectively better.
I genuinely don't know what is going on here.
> I genuinely don't know what is going on here.
I basically live in my music library. However, every single pop artist offers songs that I don't like, are not in my library, and mysteriously have many millions of albums sold.
I genuinely don't know what is going on here.
Joking aside, have you ever tried to use some of these tools ? I use to not understand why people where using vim until I really tried.
No.
> I use to not understand why people where using vim until I really tried.
There's your problem. I respectfully suggest installing Emacs.
I use the command line a lot too and this is one of my most common commands, and I don't know of an elegant way to do it with the builtin Unix tools.
(And I have basically the same question for finding files matching a regex or glob [ignoring the stuff I obviously don't want], e.g., `fd '.foo.*'`.)
Hits in hidden files is not really a pain point for me
find . -type f -name "*.foo" | grep -v '/\.' | xargs grep bar
If traversing those hidden files/directories were expensive, I'd tell `find` itself to exclude them. This also lets me switch `xargs` for `find`'s own `-exec` functionality:
find . -path '*/\.*' -prune -o -type f -name "*.foo" -exec grep bar {} +
For the record, I think `git grep` is probably the best builtin solution to the problem I gave, but personally I don't know off-hand how to only search for files matching a glob and to use the current directory rather than the repository root with `git grep` (both of which are must haves for me). I'd also need to learn those same commands for different source control systems besides git (I use one other VCS regularly).
Makes sense. If I had to do this frequently, I'd add a function/alias encapsulating that `find` incantation to my .bashrc, which I keep in version control along with other configuration files in my home directory. That way, when moving to a new environment, I can just clone that repo into a fresh home directory and most of my customizations work out-of-the-box.
1. I don't recommend using shell functions or aliases for this (e.g., `bashrc`) because then these scripts can't be called from other contexts, e.g., like Vim and Emacs builtin support for shell commands. This can easily be solved by creating scripts that can be called from anywhere (my personal collection of these scripts is here https://github.com/robenkleene/Dotfiles/tree/master/scripts). Personally, I only use Bash functions for things that have to do with Bash's runtime state (e.g., augmenting PATH is a common one).
2. The more important part though, is I don't always want to search in `*.foo`, I want a flexible, well-designed, API that allows me to on-the-fly decide what to search.
#2 is particularly important and drifts in philosophy of tooling, a mistake I used to make is building my workflow today into customizations like scripts. This is a bad idea because then the scripts aren't useful as your tasks change, and hopefully your tasks are growing in complexity over time. I.e., don't choose your tools based on your workflow today, otherwise you're building in limitations. Use powerful tools that will support you no matter what task you're performing, that scale practically infinitely. "The measure of a bookshelf is not what has been read, but what remains to be read."
find . -type f -name '*.foo' -not -path '*/.*' -print0 | xargs -0 grep bar find . -name '.*?' -prune -o -name '*.foo' -exec grep bar /dev/null {} +
alias grep='grep --exclude-dir .\* --exclude .\*'
grep -r --include \*.foo barIs that a zshism?
The only thing I use JQ for at work is parsing the copilot API response so I remember what the model names are - that's it! TBH, I could just skip it and read the json
Also if I had to guess, the type of person who criticizes tools like jq, fzf, and ripgrep and struggles to understand some people's need for benchmarking tools like hyperfine would likely disapprove of fish as well.
The OP list is significantly cheapened by the various ls alternatives.
ripgrep is something I have installed, but use only via text editor integrations. fzf is nice for building ad-hoc TUIs. fd may make sense (I’m told it’s faster than find), but I already know enough find.
The “next gen ls” family of tools in the article is baffling.
If any fledgling designers are here, then take note and add this to your list of examples to avoid.
Those "foundational GNU tools" just suck, sure, people are familiar with them and they are everywhere, but they just plain suck.
For many common operations you'd want to do by default with grep/find and so on, you have to type mountains of random gibberish to get it done. And that random gibberish isn't something that rolls of your tongue either, thus at minimum you'd define truckload of aliases.
OR you can use a tool(s) that has marginally "sane defaults" and marginally sane UX out of the box.
It really isn't that complicated. This has nothing to do with "rust".
perl -ne 'map{$h{lc$_}++}/(\w+)/g;END{map{print"$h{$_} $_\n"}sort{$h{$b}<=>$h{$a}}keys%h}'exa modern replacement for ls/tree, not maintained
"not maintained" doesn't smell "modern" to me...
Yeeeah, nope.
Damned if you do and damned if you don't.
the future life expectancy is proportional to its current age
https://difftastic.wilfred.me.uk/
It's a huge improvement over purely character-based diffs.
That’s because they’re not GNU coreutils, they’re BSD coreutils, which are spartan by design. (FWIW, this is one of my theories for why Linux/GNU dominated BSD: the default user experience of the former is just so much richer, even though the system architecture of the latter is arguably superior.)
I know I have hyperfine, fd, and eza on my Windows 11, and maybe some more I cannot remember right now.
They are super easy to install too, using winget.
`fd`: first I find that the argument semantic is way better than `find`, but that is more a bonus than a real killer feature. Now, it being much, much faster than `find` on most setup, I would consider a valuable feature. But the killer feature for me is the `-x` argument. It allows calling another command on the individual search result, which `find` can also do with `xargs` and co. But `fd` provide a very nice placeholder syntax[0], which remove the need to mess with `basename` and co. to parse the filename and make a new one, and it executes in parallel. For example, it makes converting a batch of image a fast and readable one line : `fd -e jpg -x cjxl {} {.}.jxl`
`rg` a.k.a `ripgrep` : Honestly it is just about the speed. It is so much faster than `grep` when searching through a directory, it opens up a lot of possibilities. Like, searching for `isLoading` on my frontend (~3444 files) is instant with rg (less than 0.10s) but takes a few minutes with grep.
But there is one other thing that I really like with `ripgrep` and that I think should be a feature of any "modern" CLI tool : It can format its output in JSON. Not that I am a big fan of JSON, but at least it is a well-defined exchange format. "Classic" CLI tool just output in a "human-readable" format which might just happen to be "machine-readable" if you mess with `awk` and `sed` enough. But it makes piping and scripting just that much more annoying and error & bug prone. Being able to output json, `jq` it and feed it to the next tool is so much better and feel like the missing chain of the terminal.
The big advantage of the CLI is that it is composable and scriptable by default. But it is missing a common exchange format to pass data, and this is what you have to wrangle with a lot of time when scripting. Having json, never mind all the gripes I have with this format, really join everything together.
Also, honorable mention for `zellij` which I find to be a much saner UX-wise alternative to `tmux`, and the `helix` text editor, which for me is neo-vim but with, again, a better UX (especially for beginner) and a lot more battery included feature while remaining faster (IMEX) than nvim with matching plugin for feature-parity.
EDIT: I would also add difftastic ( https://github.com/Wilfred/difftastic ) which is a syntax aware diff tool. I don't use it much, but it does makes some diff so so much easier to read.
[0] https://github.com/sharkdp/fd?tab=readme-ov-file#placeholder...
Then I tried them and it was such a night and day performance difference that they're now immediate installs on any new system I use.
That was inherited from find, it has "-exec". Even uses the same placeholder, {}, though I'm not sure about {.}
grep will try to search inside .git. If your project is Javascript, it might be searching inside node_modules, or .venv if Python. ripgrep ignores hidden files, .gitignore and .ignore. You could try using `git grep` instead. ripgrep will still be faster, but the difference won't be as dramatic.
First is parallelism. If you do `grep -r`, most greps (like GNU grep and BSD grep and the one found on macOS) will not use parallelism. ripgrep will. This could easily account for a large perceived difference in search time.
Second is the grep implementation. While GNU grep is generally considered to be quite fast on its own (especially in the POSIX locale), many other grep implementations are not. And indeed, they may be extraordinarily slow. For example, consider this comparison between FreeBSD grep (on macOS) and ripgrep (I clipped the `time` output for the `wc -l` command, which I just used to make the output smaller):
$ time rg -j1 -uuu -w -g '*.c' '[A-Z]{2}_RESUME' | wc -l
22
real 0.769
user 0.105
sys 0.662
maxmem 5104 MB
faults 0
$ time LC_ALL=C grep -E -w -r --include '*.c' '[A-Z]{2}_RESUME' | wc -l
22
real 15.830
user 14.576
sys 1.253
maxmem 2224 MB
faults 1
$ time LC_ALL=C ggrep -E -w -r --include '*.c' '[A-Z]{2}_RESUME' | wc -l
22
real 1.148
user 0.206
sys 0.767
maxmem 2784 MB
faults 1
Any one of these explanations, on its own, can account for huge differences in perceived performance. It's not really possible to know which one (perhaps even all 3) is relevant from pithy descriptions of ripgrep being so much faster than other tools.
Version info for above commands:
$ rg --version
ripgrep 14.1.1 (rev bb88a1ac45)
features:+pcre2
simd(compile):+NEON
simd(runtime):+NEON
PCRE2 10.45 is available (JIT is available)
$ grep -V
grep (BSD grep, GNU compatible) 2.6.0-FreeBSD
$ ggrep -V
ggrep (GNU grep) 3.12
Packaged by Homebrew
Copyright (C) 2025 Free Software Foundation, Inc.
[.. snip ..]Got featured here on HN few weeks ago.
btop has been pretty good for watching a machine to get an overview of everything going on, the latest version has cleaned up how the lazy CPU process listing works.
zoxide is good for cding around the system to the same places. It remembers directories so you avoid typing full paths.
It looks quite fancy but I actually like it more for it's functionality, particularly it's tree view for navigating the processes list. I'm not a big fan of full multicolor in these kinds of tools and so appreciate how easy it is to flip to grey scale mode from the built in colour schemes (even from the TUI settings menu).
I was going to top-post that the Unix/Linux command line tools were designed back in the day when data was pretty much line-oriented, e.g. one database record per line. Since then XML, and more recently JSON, have been invented, and tools like grep and sed just don't work for those formats. But you ninja'd me, sort of.
- bat it's a useless cat. Cat concatenates files. ANSI colour breaks that.
- alias ls='ls -Fh' , problem solved. Now you have * for executables, / for directories and so on.
- ncdu it's fine, perfect for what it does
- iomenu it's much faster than fzf and it almost works the same
- jq it's fine, it's a good example on a new Unix tool
- micro it's far slower than even vim
- instead of nnn, sff https://github.com/sylphenix/sff with soap(1) (xdg-open replacement) from https://2f30.org create a mega fast environment. Add MuPDF and sxiv, and nnn and friends will look really slow compared to these.
It's useless as a cat replacement, I agree. The article really shouldn't call it that, although the program's GitHub page does self-describe it as "a cat clone". It's more of a syntax highlighter combined with a git diff viewer (I do have an issue with that; it should be two separate programs, not one).
1) Piping the contents of some file into a process.
2) Showing the contents of some short file.
Now (1) is better done with redirection (< or >). The only time I use cat is when I'm testing some pipeline where I only want a few lines of input, so I use 'head' or something similar. Once I have the pipeline working right, I edit the command line to replace 'head' with 'cat'. Easier than re-arranging whole words.
And it's rare that (2) is the right solution--too often I find that the file was longer than I thought, and I have to use 'more' (actually 'less').
So a replacement for 'cat' that does color coding sounds pretty much useless to me.
I can't see bat as a "useless cat" or a replacement for cat except for reading source code in the terminal. It's more a like a less with syntax highlight or a read-only vim.
That said, I've never really cared much about missing syntax highlighting for cases where I'm viewing file contents with cat. So the tool doesn't really serve a purpose for me and instead I'll continue to load up vim/neovim if I want to view a file with syntax highlighting.
From the README:
>Whenever bat detects a non-interactive terminal (i.e. when you pipe into another process or into a file), bat will act as a drop-in replacement for cat and fall back to printing the plain file contents
bat works as normal cat for normal uses of cat and a better cat for all those "useless cat" situations we find ourselves in.
YMMV.