feat(cli): replace the classic coreutils tools with modern equivalents

Adds Rust/Go replacements for the day-to-day utilities and shadows four of
them with aliases. Only read-only commands are shadowed (cat, du, df, ps), so a
wrong flag costs a retype rather than data; rm, grep, find and sed keep their
originals and the replacements are reached by their own names.

The aliases land in .zshrc, so they apply to interactive zsh only -- scripts,
`sudo <cmd>` and anything exec'd by another program still get the real binary.

New on every host: dust, dysk, procs, trash-cli, doggo, xh, ouch, jnv, hexyl,
fq and tealdeer. dysk is used rather than duf, which is unmaintained upstream.

git gains difftastic behind a `git dft` alias. diff.external is deliberately
left unset so delta remains the renderer for git diff/show and for anything
parsing them.

The work box gains kubecolor, aliased over kubectl; it wraps the real kubectl
and drops colour when stdout is not a terminal, so pipes are unchanged.

home/README.md documents the alias map, the flag incompatibilities (including
the two that fail silently: dust -s is --apparent-size, and procs reads a bare
`aux` as a search keyword) and the rationale for what was left alone.
This commit is contained in:
Emma Thorpe
2026-08-19 17:03:07 +01:00
parent d654eac1e2
commit d9464009f0
5 changed files with 219 additions and 25 deletions
+3
View File
@@ -180,6 +180,9 @@ automatically.
- Interactive shell features (zsh, tmux, git, ssh, CLI tools, auto-tmux):
[`home/README.md`](./home/README.md).
- Which classic utilities are shadowed by modern replacements, and the flag
differences that will bite:
[`home/README.md` → "Replacing the classics"](./home/README.md#replacing-the-classics).
- All Sway / tmux / foot / zsh keyboard shortcuts:
[`home/KEYBINDINGS.md`](./home/KEYBINDINGS.md).
+146 -25
View File
@@ -16,7 +16,7 @@ Keyboard shortcuts have their own reference: [`KEYBINDINGS.md`](./KEYBINDINGS.md
Shared by every host via [`default.nix`](./default.nix); the work box also layers
[`work.nix`](../users/emmathorpe/work.nix) on top (its own ssh config, extra
packages, and the C#/Helm language servers). The committer identity (name, email,
packages, kubecolor, and the C#/Helm language servers). The committer identity (name, email,
signing key) comes from the user registry
([`../users/registry.nix`](../users/registry.nix)), not this module.
@@ -35,26 +35,34 @@ signing key) comes from the user registry
| History substring search | type a fragment, then ↑/↓ cycles matching past commands — works in foot, iTerm2 and the Linux TTY (both CSI and SS3 arrow encodings bound) |
| Prompt | hostname is prefixed when over SSH |
**Aliases:** `ls`/`ll`/`la`/`lt``eza` (icons + git), `cls``clear`. git aliases live in git.nix (below).
**Aliases:** `ls`/`ll`/`la`/`lt``eza` (icons + git), `cls``clear`,
`cat`/`du`/`df`/`ps` → their modern equivalents (see "Replacing the classics").
git aliases live in git.nix (below).
## CLI tools
| Tool | What it gives you |
| ----------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `fzf` | `Ctrl-R` fuzzy history, `Ctrl-T` file picker, `Alt-C` fuzzy cd (Catppuccin-themed) |
| `zoxide` | `z <fragment>` jumps to frecent directories |
| `direnv` + `nix-direnv` | per-project environments auto-loaded on `cd` (cached Nix dev shells) |
| `eza` | modern `ls` (drives the ls aliases) |
| `bat` | syntax-highlighting pager (Catppuccin Mocha theme); behaves like `cat` when piped; also the `MANPAGER` |
| `ripgrep` / `fd` | fast search (`rg`) and find (`fd`); also back `fzf` |
| `jq` | JSON processor |
| `gh` / `tea` | GitHub and Gitea (`code.emmathe.dev`) CLIs; `gh` uses SSH |
| `nix-index` | `command-not-found`: an unknown command tells you which Nix package provides it (prebuilt DB, no manual indexing) |
| `comma` (`,`) | run an uninstalled program once: `, cowsay hi` |
| `nh` | nicer `nixos-rebuild`/`home-manager` with diffs; `$NH_FLAKE` set to the repo. No scheduled GC (it could reap paths a running generation still references) — collect garbage manually with `nh clean all` / `nix-collect-garbage -d` |
| `btop` | resource monitor, themed Catppuccin Mocha (vendored theme) |
| `lazygit` | git TUI for staging/rebasing, themed to match (`git.nix`) |
| `hyperfine` / `sd` | command-line benchmarking; saner find-and-replace than sed |
| Tool | What it gives you |
| ------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `fzf` | `Ctrl-R` fuzzy history, `Ctrl-T` file picker, `Alt-C` fuzzy cd (Catppuccin-themed) |
| `zoxide` | `z <fragment>` jumps to frecent directories |
| `direnv` + `nix-direnv` | per-project environments auto-loaded on `cd` (cached Nix dev shells) |
| `eza` | modern `ls` (drives the ls aliases) |
| `bat` | syntax-highlighting pager (Catppuccin Mocha theme); behaves like `cat` when piped; also the `MANPAGER` |
| `ripgrep` / `fd` | fast search (`rg`) and find (`fd`); also back `fzf` |
| `jq` | JSON processor |
| `gh` / `tea` | GitHub and Gitea (`code.emmathe.dev`) CLIs; `gh` uses SSH |
| `nix-index` | `command-not-found`: an unknown command tells you which Nix package provides it (prebuilt DB, no manual indexing) |
| `comma` (`,`) | run an uninstalled program once: `, cowsay hi` |
| `nh` | nicer `nixos-rebuild`/`home-manager` with diffs; `$NH_FLAKE` set to the repo. No scheduled GC (it could reap paths a running generation still references) — collect garbage manually with `nh clean all` / `nix-collect-garbage -d` |
| `btop` | resource monitor, themed Catppuccin Mocha (vendored theme) |
| `lazygit` | git TUI for staging/rebasing, themed to match (`git.nix`) |
| `hyperfine` / `sd` | command-line benchmarking; saner find-and-replace than sed |
| `tldr` (tealdeer) | worked examples for a command, alongside `man`; the page cache is refreshed by a `tldr-update` user timer |
| `jnv` / `fq` | interactive jq-filter builder for JSON; jq syntax over binary formats (ELF, PNG, gzip, mp4…) |
| `hexyl` | hex viewer, coloured by byte class |
| `ouch` | one command for every archive format (`ouch d`/`c`/`l`) |
| `dust` `dysk` `procs` | `du` / `df` / `ps` replacements — aliased over the originals, see below |
| `trash-cli` `doggo` `xh` | `rm` (to the XDG trash) / `dig` / `curl` replacements — **not** aliased, see below |
**Theming:** `fzf`, `bat`, `btop`, `lazygit` and `git`'s `delta` pager are all
Catppuccin Mocha, driven from the shared `../lib/catppuccin-mocha.nix` palette / the
@@ -64,6 +72,110 @@ catppuccin upstream themes.
(the editor owns `$EDITOR`/`$VISUAL`); `xdg.mimeApps` maps web→Firefox,
directories→nemo (`desktop.nix`).
## Replacing the classics
Muscle memory is the expensive part of this, not the packages. Four commands are
**shadowed** — the old name now runs a new tool. Everything else keeps a new
name, so the original is never displaced.
### Shadowed by an alias
| You type | You now run | The original is still `command <name>` / `\<name>` |
| -------- | -------------------- | -------------------------------------------------- |
| `cat` | `bat --paging=never` | `command cat` |
| `du` | `dust` | `command du` |
| `df` | `dysk` | `command df` |
| `ps` | `procs` | `command ps` |
Only read-only commands are shadowed, so the worst case of a wrong flag is a
retype rather than lost data. `rm`, `grep`, `curl` and `find` are deliberately
left alone — see "Left alone on purpose" below.
**Where the aliases apply.** They are written into `~/.config/zsh/.zshrc`, so
they exist only in an **interactive zsh**:
- shell scripts, `Makefile` recipes and anything another program `exec`s get the
real coreutils binary — nothing that parses output can break;
- `sudo du -sh /var` runs the real `du`: zsh does not expand an alias after
`sudo`;
- `KUBECONFIG=… kubectl …` **does** expand — zsh expands aliases after a
variable-assignment prefix. That is what makes the kubecolor alias on the work
box (below) useful rather than a special case you have to remember.
### Flag gotchas
These replacements are not drop-in. The two marked **silent** are the dangerous
ones — they succeed and answer a different question than the one you asked.
Everything else fails loudly.
| Old habit | What happens now | Do this instead |
| ------------------- | --------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------- |
| `du -sh dir` | dust prints its usage and exits non-zero — `-h` is not a dust flag | `dust dir` (units are human by default; the total is the last row) |
| `du -s dir` | **silent**: dust's `-s` is `--apparent-size`, not `--summarize` | `dust -d 0 dir` for a single total line |
| `du --max-depth=2` | not recognised | `dust -d 2` |
| `df -h` | dysk rejects `-h` | `dysk` (SI units by default; `-u binary` for 1024-based) |
| `df -i` | not recognised | `dysk -c +inodes` |
| `df -a` | works, same meaning (all mount points) | — |
| `df /some/path` | works, same meaning (the device holding that path) | — |
| `ps aux` | **silent**: `aux` is read as a search keyword, so you get only processes whose command line contains the string "aux" | `procs` lists everything; `procs <pattern>` filters |
| `ps -ef` | `error: unexpected argument '-e'` | `procs` |
| `ps -p 1234` | not recognised | `procs 1234` |
| `procs -a` | **silent**: `-a` is `--and` (combine search keywords), not "all" | drop it — `procs` already shows everything |
| `cat -v` / `cat -e` | `error: unexpected argument` | `cat -A` does work (bat implements show-all); else `command cat -v` |
| `cat -n` | works, but bat's number column, not coreutils' layout | fine to read; `command cat -n` when the exact layout matters |
| `cat <binary>` | prints `<BINARY>` to a terminal instead of dumping the bytes | `hexyl <file>`, or `command cat` to dump |
Useful new capabilities in the same tools: `procs --tree`, `procs --watch`,
`dust -r` (largest at the top), `dysk -s size`, `dysk -f 'type=ext4'`.
**Piping is safe for `cat`.** bat drops all decoration and colour when stdout is
not a terminal, so `cat f | sha256sum` is byte-for-byte what coreutils `cat`
would have given. The others are TUI-shaped tables with no stable format — if
something needs to parse them, use `dysk --json`/`--csv`, `procs --json`, or the
original binary.
### Renamed, not shadowed
| Instead of | Use | Notes |
| --------------------------- | ------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `rm` | `trash` | Moves to the XDG trash. `trash-list`, `trash-restore` (interactive picker), `trash-empty [days]`. It never deletes in place: if it cannot create a trash directory on that filesystem it errors out. |
| `dig` / `nslookup` | `doggo` | `doggo example.com MX @1.1.1.1`; `--json` for scripting. Not aliased — `dig` (from the `bind` closure that other modules pull in) stays where scripts expect it. |
| `curl` (interactive poking) | `xh` | HTTPie syntax: `xh POST api.example/x name=lyra`. `xhs` is `xh --https`. **curl stays installed and unaliased** — it is what scripts and CI use. |
| `tar` / `unzip` / `7z` | `ouch` | `ouch d file.<anything>`, `ouch c out.tar.zst src/`, `ouch l archive`. Format is inferred from the extension. The oh-my-zsh `extract` function still works too. |
| `jq` (exploring a payload) | `jnv` | Interactive filter builder over a JSON file; it prints the jq expression you built. `jq` remains the scripting tool. |
| `hexdump -C` / `xxd` | `hexyl` | `hexyl -n 256 -s 0x40 file` for a window into a large file. |
| `strings` on a known format | `fq` | jq syntax over binary formats: `fq -d elf '.sections[].name' ./bin`. |
| skimming a man page | `tldr` | Worked examples. `man` is untouched (and still rendered through bat). |
### Left alone on purpose
- **`grep`** is not aliased to `rg`. ripgrep is recursive by default, skips
gitignored and hidden files, and uses a different regex dialect (no
backreferences, no POSIX classes in the same form). A `grep` habit silently
producing fewer matches is a worse failure than typing three characters. Type
`rg`.
- **`rm`** is not aliased to `trash-put`. Retraining `rm` to mean "recoverable"
is a habit that follows you onto every machine where it is not — remote hosts,
root shells, containers, CI. Type `trash`.
- **`find`** is not aliased to `fd`; the `-exec`/`-print0` vocabulary has no
equivalent and scripts lean on it. Type `fd`.
- **`sed`** is not aliased to `sd`; `sd` takes real regex and literal
replacements, not sed's expression language. Type `sd`.
- **coreutils itself** is not swapped for `uutils-coreutils`. It is packaged and
tempting, but every Nix builder and shell script on these hosts is written
against GNU behaviour, including its forty-year-old edge cases.
### Work box only: kubectl → kubecolor
On EDaaS (`work.nix`) `kubectl` is aliased to **kubecolor**, which runs the real
kubectl underneath and colourises what comes back. Nothing to relearn: every
flag, subcommand and plugin passes straight through, unrecognised output is
printed verbatim, and colour is dropped automatically when stdout is not a
terminal — so `kubectl get -o json … | jq` is unchanged. The alias also applies
to `KUBECONFIG=prodconfig kubectl …`, per the alias-expansion note above.
Completions are kubectl's own (`compdef kubecolor=kubectl`). Escape hatch as
ever: `command kubectl`.
## tmux
**Auto-start:** opening any interactive terminal — foot, iTerm2, the WSL shell, the
@@ -141,13 +253,21 @@ Pager is **delta**. **commitizen** is installed on every host; `cz` defaults to
Conventional Commits. **lazygit** (themed) is the TUI. The commit-graph is kept
current (`gc`/`fetch.writeCommitGraph`) so `lg` stays fast.
| Aliases | |
| ------------------------ | ------------------------------------------------------------------ |
| `st` `co` `sw` `br` `ci` | status / checkout / switch / branch / commit |
| `last` `unstage` | last commit / unstage |
| `amend` `fixup` `undo` | amend-no-edit / `commit --fixup` / soft-reset HEAD~1 (keep staged) |
| `lg` | graph log, all branches |
| `cz` `cc` | `git cz <sub>` (e.g. `git cz c`) and `git cc` → commitizen prompt |
| Aliases | |
| ------------------------ | ------------------------------------------------------------------------- |
| `st` `co` `sw` `br` `ci` | status / checkout / switch / branch / commit |
| `last` `unstage` | last commit / unstage |
| `amend` `fixup` `undo` | amend-no-edit / `commit --fixup` / soft-reset HEAD~1 (keep staged) |
| `lg` | graph log, all branches |
| `cz` `cc` | `git cz <sub>` (e.g. `git cz c`) and `git cc` → commitizen prompt |
| `dft` | structural (syntax-aware) diff via difftastic; takes `git diff` arguments |
**`git dft` vs `git diff`.** delta stays the default renderer for everything;
`diff.external` is deliberately **not** set, so `git diff`, `git show` and
anything parsing their output are unchanged. Reach for `dft` when a refactor
moved code around and a line-based diff is noise. One wrinkle: `dft` is a
`!`-shell alias, and git runs those from the repository root — pass pathspecs
relative to the root, not to your current directory.
| Behaviour | |
| -------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
@@ -217,6 +337,7 @@ Claude to route new memories there.
| | Personal Linux (sway) | macOS | Work WSL (EDaaS) |
| --------------------------- | --------------------- | ----------------- | --------------------------- |
| Auto-tmux | yes (foot/TTY) | yes (iTerm2) | yes (WSL shell) |
| `kubectl` → kubecolor | no (no kubectl) | no | yes (work module) |
| git email | `iam@emmathe.dev` | `iam@emmathe.dev` | `…@citrix.com` (work) |
| ssh config managed | yes | yes | no (keeps corporate config) |
| ssh-agent | yes | launchd | yes (work module) |
+13
View File
@@ -74,6 +74,11 @@ in
# `cz commit`, `git cz bump`, etc. `git cc` is a shortcut for the prompt.
cz = "!cz";
cc = "!cz commit";
# Structural (syntax-aware) diff, on demand. Set per-invocation via the
# environment rather than `diff.external`, which would also change what
# `git show` and `git log -p --ext-diff` emit for every caller.
# Takes the same arguments as `git diff`: `git dft HEAD~3 -- file`.
dft = "!GIT_EXTERNAL_DIFF=difft git diff";
};
# SSH signing, key from the registry. mkDefault so a host lacking the key
@@ -99,6 +104,14 @@ in
enableGitIntegration = true;
};
# difftastic backs the `dft` alias above. git.enable stays off on purpose:
# the module's git integration sets `diff.external`, which would displace
# delta as the diff renderer everywhere instead of only where asked.
programs.difftastic = {
enable = true;
git.enable = false;
};
# lazygit: TUI for staging/rebasing, themed to Catppuccin Mocha to match.
programs.lazygit = {
enable = true;
+44
View File
@@ -26,8 +26,32 @@ in
pkgs.tea
pkgs.hyperfine # command-line benchmarking
pkgs.sd # saner find-and-replace than sed
# Replacements for the classic coreutils/BSD tools. Only the read-only ones
# are aliased over the original name (see shellAliases below); the rest keep
# their own name so nothing changes shape under a script's feet. The alias
# map and the flag-compatibility differences are documented in
# ./README.md, "Replacing the classics".
pkgs.dust # du: tree-shaped, size-sorted disk usage
pkgs.dysk # df: mounted filesystems (duf is unmaintained upstream)
pkgs.procs # ps: process list with tree, ports and container columns
pkgs.trash-cli # rm: XDG trash; `trash` / `trash-list` / `trash-restore`
pkgs.doggo # dig: DNS lookups
pkgs.xh # curl, for interactive HTTP poking (curl stays for scripts)
pkgs.ouch # tar/unzip/7z/zstd: one command for every archive format
pkgs.jnv # interactive jq filter builder (jq itself stays for scripts)
pkgs.hexyl # hex viewer
pkgs.fq # jq for binary formats
];
# tldr pages: worked examples for a command, next to (not instead of) man.
# enableAutoUpdates defaults on and installs a tldr-update user timer, which
# keeps the page cache fresh -- without it `tldr` fails until first `--update`.
programs.tealdeer = {
enable = true;
settings.display.compact = true;
};
# Resource monitor, themed Catppuccin Mocha to match the rest of the desktop.
# btop does not bundle the theme, so vendor it from catppuccin/btop (pinned).
programs.btop = {
@@ -137,6 +161,26 @@ in
la = "eza --icons --git -la";
lt = "eza --icons --git --tree";
cls = "clear";
# Shadow the classics with their modern equivalents. Only read-only
# commands are shadowed: a wrong flag costs a retype, never data. The
# flag vocabularies are NOT compatible (`du -sh`, `df -h`, `ps aux` all
# fail here) -- see ./README.md, "Replacing the classics".
#
# Blast radius is bounded by where these live: shellAliases lands in
# .zshrc, so only interactive zsh sees them. Scripts, `sudo <cmd>` and
# anything exec'd by another program still get the real binary. To reach
# the original in an interactive shell: `command du` or `\du`.
cat = "bat --paging=never"; # bat is already the PAGER/MANPAGER
du = "dust";
df = "dysk";
ps = "procs";
# `rm` is deliberately NOT aliased to trash-put. Retraining `rm` to mean
# "recoverable" is a habit that follows you onto machines where it does
# not (every remote host, every root shell, every container), and trash
# semantics break down anyway on a different filesystem or on
# root-owned paths. Type `trash` when you want a trash can.
};
};
+13
View File
@@ -50,6 +50,19 @@
];
services.ssh-agent.enable = true;
# Colourised kubectl. enableAlias points `kubectl` at kubecolor, which parses
# the output of the real kubectl underneath and passes anything it does not
# recognise straight through, so every flag and subcommand still works. It
# drops colour automatically when stdout is not a terminal, leaving pipes into
# grep/jq/yq byte-identical. zsh integration reuses kubectl's own completions.
# Note the alias does apply to `KUBECONFIG=... kubectl ...`: zsh expands
# aliases after a variable-assignment prefix.
programs.kubecolor = {
enable = true;
enableAlias = true;
enableZshIntegration = true;
};
# gcx (above) keeps its OAuth tokens in the system keychain and has no
# plaintext fallback, so this WSL box needs something owning
# org.freedesktop.secrets. See home/secret-service.nix for why