Building a shell is a great exercise, but honestly having to deal with string parsing is such a bother that it robs like 2/3 of the joy along the way. I once built a very simple one in Go [0] as a learning exercise and I stopped once I started getting frustrated with all the corner cases.
A common problem I noticed is that if you took certain courses in computer science, you may have a pre-conceived notion of how to parse programming languages, and the shell language doesn't quite fit that model
I have seen this misconception many times
In Oils, we have some pretty minor elaborations of the standard model, and it makes things a lot easier
To summarize that, I'd say that doing as much work as possible in the lexer, with regular languages and "lexer modes", drastically reduces the complexity of writing a shell parser
And it's not just one parser -- shell actually has 5 to 15 different parsers, depending on how you count
Author here, and yeah, I agree. I skipped writing a parser altogether and just split on whitespace and | so that I could get to the interesting bits.
For side-projects, I have to ask myself if I'm writing a parser, or if I'm building something else; e.g. for a toy programming language, it's way more fun to start with an AST and play around, and come back to the parser if you really fall in love with it.
Can say the same for control characters in terminals. I even think maybe it's just easier to ditch them all and use QT to build a "terminal" with clickable urls, something similar to what TempleOS does.
Fun read! I built a minimal Linux shell [0] in c and Zig last year which does not depend on libc. It was a great way to learn about execve, the new-ish clone3 syscall and how Linux starts a process. Parsing strings is the least fun part of the building the shell.
There's a very good exercise on Codecrafters (https://app.codecrafters.io/courses/shell/overview) to walk you through writing your own shell. I found it enlightening, as well as a good way to learn a new language.
Unix shells are conceptually simple but hide a surprising amount of complexity under the hood that we take for granted. I recently had build my own PTY controller. There were so many edge-cases to deal with. It took weeks of stress testing and writing many tests to get it right.
Bit of pedantry but I don't think traditional unix shell (like this) follows repl model; the shell is not usually doing printing of the result of evaluation. Instead the printing happens more as a side effect of the commands.
Even if you view the system as a whole the printing is deeply intertwined with the evaluation, which is very different from repl where eval returns a value and print prints it
I remember my first shell programming I ever did was batch in windows back in the 3.11/95 days.
The first line was always to turn off echo, and I've always wondered why that was a decision for batch script. Or I'm misremembering. 30 years of separation makes it hard to remember the details.
Great article. There are many things every developer should do when starting to learn programming or when trying to improve their skills. This is one of them. I once built a shell-like programming language (not an interpreter). If anyone reading this wants to improve their skills, I strongly suggest building your own shell from scratch.
Is there a (real) shell whose code is relatively short and self contained and would be valuable to read? This was always something I wanted to do but never quite spent time to look for a good one to explore.
It depends on what you are looking for. My recommendation for learning "how is X done in a shell" is the OpenBSD ksh: https://github.com/ibara/oksh
It's what they use for /bin/sh, it has everything that a complete shell needs (including a mechanism for providing command completions) and has code that is much easier to read than bash or zsh.
Although not the same... Destroy All Software has videos on building your own shell using Ruby. I watched it to learn and it was a lot of fun to watch him basically building a shell, I'm not really a Ruby guy, but it was easy to grasp. It's not free, you would need a subscription, but its worth the watch otherwise.
I think there's a good one if you search around for "xv6 sh.c". Hard to tell immediately from a google search just now since there are many implementations (people do it in school) and github's currently blocking requests from my phone.
Also helpful may be running strace on your shell, then reviewing the output line by line to make sure you understand each. This is a VERY instructive exercise to do in general.
Very cool. Currently working on the beginning of a small text editor so this part seemed interesting and was curious of any overlap. Thanks for the interesting post!
Interesting. I wanted to do toast | bash to let the AI drive the computer but the bash shell really got in the way. Too much complexity. The things that annoy humans, $ expansion, special characters, etc don't work for AI either. Ended up writing a custom shell for AI (and humans). When a tool gets in the way, sometimes it just time to change the tool.
Yup, job control is a huge mess. I think Bill Joy was able to modify the shell, the syscall interface, and the terminal driver at the same time to implement the hacky mechanism of job control. But a few years later that kind of crosscutting change would have been harder
One thing we learned from implementing job control in https://oils.pub is that the differing pipeline semantics of bash and zsh makes a difference
In bash, the last part of the pipeline is forked (unless shopt -s lastpipe)
In zsh, it isn't
$ bash -c 'echo hi | read x; echo $x' # no output
$ zsh -c 'echo hi | read x; echo $x'
hi
And then that affects this case:
bash$ sleep 5 | read
^Z
[1]+ Stopped sleep 5 | read
zsh$ sleep 5 | read # job control doesn't apply to this case in zsh
^Zzsh: job can't be suspended
So yeah the semantics of shell are not very well specified (which is one reason for OSH and YSH). I recall a bug running an Alpine Linux shell script where this difference matters -- if the last part is NOT forked, then the script doesn't run
I think there was almost a "double bug" -- the script relied on the read output being "lost", even though that was likely not the intended behavior
FWIW here is another piece of trivia about job control: the API means you can't spawn a child process "safely" in POSIX -- you have to trust that that the executable you're spawning is well-behaved (or use more advanced Linux process isolation)
In this case it was the Zed editor spawning the zsh shell:
39 comments
[0] https://github.com/lourencovales/codecrafters/blob/master/sh...
I have seen this misconception many times
In Oils, we have some pretty minor elaborations of the standard model, and it makes things a lot easier
How to Parse Shell Like a Programming Language - https://www.oilshell.org/blog/2019/02/07.html
Everything I wrote there still holds, although that post could use some minor updates (and OSH is the most bash-compatible shell, and more POSIX-compatible than /bin/sh on Debian - e.g. https://pages.oils.pub/spec-compat/2025-11-02/renamed-tmp/sp... )
---
To summarize that, I'd say that doing as much work as possible in the lexer, with regular languages and "lexer modes", drastically reduces the complexity of writing a shell parser
And it's not just one parser -- shell actually has 5 to 15 different parsers, depending on how you count
I often show this file to make that point: https://oils.pub/release/0.37.0/pub/src-tree.wwz/_gen/_tmp/m...
(linked from https://oils.pub/release/0.37.0/quality.html)
Fine-grained heterogenous algebraic data types also help. Shells in C tend to use a homogeneous command* and word* kind of representation
https://oils.pub/release/0.37.0/pub/src-tree.wwz/frontend/sy... (~700 lines of type definitions)
|so that I could get to the interesting bits.For side-projects, I have to ask myself if I'm writing a parser, or if I'm building something else; e.g. for a toy programming language, it's way more fun to start with an AST and play around, and come back to the parser if you really fall in love with it.
[0] https://gist.github.com/rrampage/5046b60ca2d040bcffb49ee38e8...
The first line was always to turn off echo, and I've always wondered why that was a decision for batch script. Or I'm misremembering. 30 years of separation makes it hard to remember the details.
set -xrather thanecho.> the shell is not usually doing printing of the result of evaluation
I always include $? in the prompt, so I guess I can say it does print the result of the evaluation.
It's what they use for /bin/sh, it has everything that a complete shell needs (including a mechanism for providing command completions) and has code that is much easier to read than bash or zsh.
Something that I also would recommend is the design document for the plan9 rc shell; it is a worthwhile read for anybody interested in shells: https://doc.cat-v.org/plan_9/4th_edition/papers/rc
An implementation is also available if one wants to look at how it could be done: https://github.com/rakitzis/rc
https://www.destroyallsoftware.com/screencasts/catalog/shell...
Marc Rochkind's book Avanced UNIX Programming implemented a basic shell, through iterations. You can see the first at e.g. here https://github.com/gmarler/AUPv2/blob/master/c5/sh0.c
It might be a bit old too. The book is very good but again, quite old. There seem to be free copies of it on the net.
BTW, does anyone know if Marc Rochkind is alive? His site basepath.com seems to be for sale :-(
Also helpful may be running strace on your shell, then reviewing the output line by line to make sure you understand each. This is a VERY instructive exercise to do in general.
IIRC readline uses a
char *internally since the length of a user-edited line is fairly bounded.One thing we learned from implementing job control in https://oils.pub is that the differing pipeline semantics of bash and zsh makes a difference
In bash, the last part of the pipeline is forked (unless shopt -s lastpipe)
In zsh, it isn't
And then that affects this case: So yeah the semantics of shell are not very well specified (which is one reason for OSH and YSH). I recall a bug running an Alpine Linux shell script where this difference matters -- if the last part is NOT forked, then the script doesn't runI think there was almost a "double bug" -- the script relied on the
readoutput being "lost", even though that was likely not the intended behaviorIn this case it was the Zed editor spawning the zsh shell:
How to Lose Control of your Shell - https://registerspill.thorstenball.com/p/how-to-lose-control...
zsh has a bug where it doesn't do the job control cleanup properly in some cases -- when fork-exec() optimizations happen.
This can mess up the calling process. For example, here you could no longer kill Zed by hitting Ctrl-C, even after zsh is done.
My comment: https://lobste.rs/s/hru0ib/how_lose_control_your_shell#c_dfl...