How the examples are run

No output on this site is typed by hand. Each rebuild runs every example in a sandbox and the pages show what came back.

This build

The run took 212 seconds. Examples that could not run in the sandbox (43 of them) say so on their page and show no output:

The sandbox

• Image: Debian 13.7 ("trixie"), the official debian:trixie-slim image pinned by its digest, with every package installed from the Debian snapshot archive of 2026-10-15. Rebuilding the image installs exactly the same versions, so a change in output means a change in the example, not a silent upgrade.
• User and place: each example runs as an ordinary user, dev, in /home/dev/demo, on a host named demo, with LANG=C.UTF-8, TZ=UTC and an 80-column terminal width. A few examples (such as chown) run as root and say so.
• Fresh every time: before each example the home directory is restored from the sample project, /tmp is emptied and leftover processes are stopped, so no example depends on another.
• No internet: the sandbox has no network except loopback. A small HTTP test server answers on localhost:8080 and an SSH server on the host alias server, so curl, wget, ssh, scp and rsync examples show real transfers.
• What you see: standard output and standard error, interleaved as a terminal shows them, and the exit status when it is not 0. Output over 60 lines is cut and the page says how many lines were left out.
• Things that vary: process IDs, timings, dates and key fingerprints come from the run that built the page. File dates in the sample project are set to a few days before each run, so ls -l and find -mtime look as they would on a working machine.

Package versions in this run

On Debian and Ubuntu, a minimal install's awk is mawk; in this sandbox awk is GNU awk (/usr/bin/gawk), and the awk page notes where the two differ.

The sample project

Every example runs in the same small directory: logs, CSV and JSON data, some source code and a few images, all made up for this site (the IP addresses are from the ranges reserved for documentation). Seeing the same files again and again makes outputs easy to compare across commands.

~/demo
├── .env
├── access.log
├── data/
│   ├── config.json
│   ├── fruits.txt
│   ├── hosts.txt
│   ├── inventory.tsv
│   ├── numbers.txt
│   ├── poem.txt
│   ├── sales.csv
│   ├── servers.json
│   ├── settings.conf
│   ├── users.csv
│   └── windows.txt
├── file with spaces.txt
├── link-to-notes -> notes.txt
├── logs/
│   ├── app.log
│   ├── app.log.1
│   └── error.log
├── notes.txt
├── photos/
│   ├── beach.png
│   ├── forest.png
│   └── IMG_2041.png
├── README.md
├── scripts/
│   ├── backup.sh
│   └── greet.sh
├── src/
│   ├── app.py
│   ├── lib/
│   │   └── helpers.py
│   ├── main.c
│   ├── Makefile
│   └── util.sh
└── todo.txt

Checks that stop a build

• Every example declares the exit status it should end with (0 unless the page is showing a failure). If any example ends differently, or times out, the build fails and the previous version of the site stays up.
• An example whose output was not recorded in the latest run cannot be published.
• Every internal link is checked; a broken one fails the build.
• The command-line explainer must pass its whole test corpus.

The explainer's test corpus

121 command lines, each with an expected breakdown into tokens that was checked by hand and is now fixed (command, option, option value, argument, operator, redirection, keyword, assignment, here-document, comment and the nested commands inside $( ), <( ), xargs and find -exec). This build: 121 passed, 0 failed. They cover:

• assignments: 6
• comments: 2
• keywords and compound commands: 12
• long options and values: 10
• nested commands: 11
• pipes and lists: 8
• quoting and expansion: 14
• redirections: 12
• simple options: 40
• unknown options flagged: 6

The explainer describes options only from the notes on this site's command pages. An option that is not in those notes is reported as unknown; the corpus includes such cases, so a change that starts guessing fails the tests.

How the text is written

• All explanations, pitfalls and option descriptions are written for this site. No manual or info page text is reproduced; each page names its manual section (for example grep(1)) so you can read the original.
• The GNU vs BSD and macOS notes describe systems the sandbox does not run. They are written from those systems' documentation and only where the difference is well established; they are not run.
• The systemctl and journalctl pages are the exception to "every example runs": systemd is not running in the sandbox, so their examples are marked as not run and show no output.

Sources and licences

• Text, examples, sample project and illustrations: written and drawn for this site.
• Outputs: produced by Debian 13.7 packages, which are free software under their own licences (GNU coreutils, findutils, grep, sed, gawk, tar and bash under the GPL; curl under the curl licence; OpenSSH under a BSD-style licence; jq under the MIT licence, among others). Showing a program's output does not copy the program.
• Manual pages: cited by name and section only.
• No third-party data, images, fonts or scripts are used.