Dotfiles (Tom Ryder)

This is my personal repository of configuration files and scripts for $HOME, including most of the settings that migrate well between machines.

This repository began as a simple way to share Vim and tmux configuration, but a lot of scripts and shell configuration have been added over time, making it into a personal suite of custom Unix tools.


$ mkdir -p ~/.local/src
$ git clone https://dev.sanctum.geek.nz/code/dotfiles.git ~/.local/src/dotfiles
$ cd ~/.local/src/dotfiles
$ git submodule init
$ git submodule update
$ make
$ make -n install  # Check output carefully
$ make install

For the default all target, you'll need a POSIX-fearing userland, including make(1) and m4(1).

The installation Makefile overwrites things standing in the way of its installed files without backing them up, so read the output of make -n install before running make install carefully, to make sure you aren't going to lose anything unexpected. If you're still not sure, install it in a temporary directory first, so you can explore:

$ tmpdir=$(mktemp -d)
$ make install HOME="$tmpdir"
$ env -i HOME="$tmpdir" TERM="$TERM" "$SHELL" -l

The default install target will install these targets and all their dependencies:

  • install-bin
  • install-curl
  • install-ex
  • install-git
  • install-gnupg
  • install-less
  • install-login-shell
  • install-man
  • install-readline
  • install-vim

The install-login-shell target looks at your SHELL environment variable, and tries to figure out which shell's configuration files to install, falling back on install-sh.

The remaining files can be installed with the other install-* targets. Try awk -f bin/mftl.awk Makefile in the project's root directory to see a list.


To keep a set of make targets useful for a specific user or host, you can list them in a newline-separated file ~/.config/dotfiles/config, and install using that with the special install-conf target. This can include macro settings for the Makefile, too:

$ cd
$ cat .config/dotfiles/config
$ make -C .local/src/dotfiles install-conf


Configuration is included for:

  • Bourne-style POSIX shells, sharing a .profile, an ENV file, and some helper functions:
  • Abook---curses address book program
  • cURL---Command-line tool for transferring data with URL syntax
  • Dillo---A lightweight web browser
  • finger(1)---User information lookup program
  • Git---Distributed version control system
  • GNU Emacs---Extensible text editor
  • GnuPG---GNU Privacy Guard, for private communication and file encryption
  • i3---Tiling window manager
  • less---Terminal pager
  • mpv---Media player
  • Mutt---Terminal mail user agent
  • mysql(1)---Command-line MySQL client
  • ncmpcpp---ncurses based MPD client inspired by ncmpc
  • Newsboat---Terminal RSS/Atom feed reader
  • psql(1)---Command-line PostgreSQL client
  • Parcellite---clipboard manager for X
  • Perl::Critic---static source code analysis engine for Perl
  • Perl::Tidy---reformats Perl source code
  • Readline---GNU library for user input used by Bash, MySQL, and others
  • Subversion---Apache Subversion, a version control system
  • tidy---HTML/XHTML linter and tidier
  • tmux---Terminal multiplexer similar to GNU Screen
  • Vim---Vi IMproved, a text editor
  • X11---Windowing system with network transparency for Unix

There is also some slightly customized support for multi-version environment management for three major scripting languages:

The configurations for shells, Mutt, tmux, and Vim are the most likely to be of interest. The i3 configuration is limited mainly to changing window switching key bindings to match Vim's.


On GNU/Linux, I use Bash; on *BSD, I use some variant of Korn Shell, preferably ksh93 if it's available.

POSIX core

My ~/.profile and other files in sh are written in POSIX shell script, so they should work in most POSIX-conforming sh(1) implementations. Please email me if you find a case where they don't!

Further shell snippets to run on login are sourced from ~/.profile.d by ~/.profile. Most of these boil down to exporting variables appropriate to the system and the software it has available.

Configuration that should be sourced for all conforming interactive shells is kept in ~/.shrc, with subscripts read from ~/.shrc.d. There's a ~/.shinit shim to act as ENV.

GNU Bash

My Bash scripts are written to work with GNU Bash v3.0 or newer. This is why I use older syntax for certain things such as appending items to arrays:


This doesn't work for arrays with sparse indices; compare this to the much nicer syntax available since 3.1-alpha1, which does:


I do use some features that are only available in versions after v3.0, such as newer shopt options like dirspell, or variables like PROMPT_DIRTRIM. These are set only after testing BASH_VERSINFO appropriately.


A terminal session with my prompt looks something like this:

~$ ssh remote
remote:~$ cd .local/src/dotfiles
remote:~/.local/src/dotfiles(master+!)$ git status
M  bash/bashrc.d/prompt.bash
A  init
remote:~/.local/src/dotfiles(master+!)$ foobar
foobar: command not found
remote:~/.local/src/dotfiles(master+!)<127>$ sleep 5 &
[1] 28937

The hostname is elided if not connected via SSH. The working directory with tilde abbreviation for $HOME is always shown. The rest of the prompt expands based on context to include these elements, in this order:

  • Whether in a Git repository if applicable,

  • The current version control branch, tag, or commit/revision if applicable, and punctuation to show repository status including reference to upstreams at a glance. Subversion support can also be enabled, in which case a git: or svn: prefix is added appropriately for disambiguation.

  • The number of running background jobs, if non-zero.
  • The exit status of the last command, if non-zero.

You can set PROMPT_COLOR, PROMPT_PREFIX, and PROMPT_SUFFIX too, which all do about what you'd expect.

If you start up GNU Bash, Korn shell, or Z shell, and that doesn't match your login shell, the prompt should display an appropriate prefix.

This is all managed within the prompt function. There's some mildly hacky logic on tput codes included such that it should work correctly for most common terminals using both termcap(5) and terminfo(5), including *BSD systems. It's also designed to degrade gracefully for eight-color and no-color terminals.


If a function can be written in POSIX sh without too much hackery, I put it in sh/shrc.d to be loaded by any POSIX interactive shell. Those include:

  • Four functions for using a "marked" directory, which I find a more manageable concept than the pushd/popd directory stack:
    • md() marks a given (or the current) directory.
    • gd() goes to the marked directory.
    • pmd() prints the marked directory.
    • xd() swaps the current and marked directories.
  • Ten other directory management and navigation functions:
    • bd() changes into a named ancestor of the current directory.
    • gt() changes into a directory or into a file's directory.
    • lgt() runs gt() on the first result from a loc(1df) search.
    • mkcd() creates a directory and changes into it.
    • pd() changes to the argument's parent directory.
    • rd() replaces the first instance of its first argument with its second argument in $PWD, emulating a feature of the Zsh cd builtin that I like.
    • scr() creates a temporary directory and changes into it.
    • sd() changes into a sibling of the current directory.
    • ud() changes into an indexed ancestor of a directory.
    • vr() tries to change to the root directory of a source control repository.
  • Two editor wrapper functions:
    • e() invokes $EDITOR, or ed(1) if not set.
    • v() invokes $VISUAL, or vi(1) if not set.
  • bc() silences startup messages from GNU bc(1).
  • ed() tries to get verbose error messages, a prompt, and a Readline environment for ed(1).
  • gdb() silences startup messages from gdb(1).
  • grep() tries to apply color and other options good for interactive use if available.
  • hgrep() allows searching $HISTFILE.
  • ls() tries to apply color and other options good for interactive use if available.
    • la() runs ls -A if it can, or ls -a otherwise.
    • ll() runs ls -Al if it can, or ls -al otherwise.
  • path() manages the contents of PATH conveniently.
  • scp() tries to detect forgotten hostnames in scp(1) command calls.
  • sudo() forces -H for sudo(8) calls so that $HOME is never preserved; I hate having root-owned files in my home directory.
  • tor() is just a terse shortcut for using Torsocks to anonymize TCP connections from the current shell.
  • tree() colorizes GNU tree(1) output if possible (without having LS_COLORS set).
  • x() is a one-key shortcut for exec startx.

There are a few other little tricks defined for other shells providing non-POSIX features, as compatibility allows:

  • keep() stores ad-hoc shell functions and variables (Bash, Korn Shell 93, Z shell).
  • prompt() sets up my interactive prompt (Bash, Korn Shell, Z shell).
  • pushd() adds a default destination of $HOME to the pushd builtin (Bash).
  • vared() allows interactively editing a variable with Readline, emulating a Z shell function I like by the same name (Bash).
  • ver() prints the current shell's version information (Bash, Korn Shell, Z shell).


I find the bash-completion package a bit too heavy for my tastes, and turn it off using a stub file installed in ~/.config/bash_completion. The majority of the time I just want to complete paths anyway, and this makes for a quicker startup without a lot of junk functions in my Bash namespace.

I do make some exceptions with completions defined in files in ~/.bash_completion.d for things I really do get tired of typing repeatedly:

  • Bash builtins: commands, help topics, shell options, variables, etc.
  • find(1)'s more portable options
  • gpg(1) long options
  • make(1) targets read from a Makefile
  • man(1) page titles
  • pass(1) entries
  • ssh(1) hostnames from ~/.ssh/config

For commands that pretty much always want to operate on text, such as text file or stream editors, I exclude special file types and extensions I know are binary. I don't actually read the file, so this is more of a heuristic thing, and sometimes it will get things wrong.

I also add completions for my own scripts and functions where useful. The completions are dynamically loaded if Bash is version 4.0 or greater. Otherwise, they're all loaded on startup.

Korn shell

These are experimental; they are mostly used to tinker with MirBSD mksh, AT&T ksh93, and OpenBSD pdksh. All shells in this family default to a yellow prompt if detected.

Z shell

These are experimental; I do not like Z shell much at the moment. The files started as a joke (exec bash). zsh shells default to having a prompt colored cyan.


My mail is kept in individual Maildir-format directories under ~/mail, with the system mail spool in e.g. /var/mail/tejr being where most unfiltered new mail is spooled. I use Getmail, maildrop, and msmtp; the configurations for these are not included here. I sign whenever I have some indication that the recipient might be using a PGP implementation, and I encrypt whenever I have a public key available for them. The GnuPG and S/MIME interfacing is done with GPGme, rather than defining commands for each crypto operation. I wrote an article about this setup if it sounds appealing.

You'll need Abook installed if you want to use the query_command I have defined, and msmtp for the sendmail command.


These are just generally vi-friendly settings, and there isn't much out of the ordinary. Note that the configuration presently uses a hard-coded 256-color color scheme, and uses non-login shells, with an attempt to control the environment to stop shells thinking they have access to an X display.

The shell scripts in bin include tm(1df), a shortcut to make attach into the default command if no arguments are given and sessions do already exist. My ~/.inputrc file binds Alt+M to run that, and Tmux in turn binds the same key combination to detach.


The majority of the Vim configuration is just setting options, with a fair few mappings and remappings, both global and buffer-local. It's extensively commented.

XDG Basedirs

The XDG Base Directory Specification's environment variables are checked on startup, and appropriate directories are added to the start and end of 'runtimepath'. I use these separate directories for machine-local configuration, usually in ~/.config/vim, while all the files that this suite installs land in ~/.vim. Backups, swap files, persistent undo data, saved views, and the viminfo file all live under XDG_CACHE_HOME, normally ~/.cache/vim.


I define my own filetype.vim and scripts.vim, so that filetype detection works in a way I like, and loads quickly. They are unlikely to suit you as they are, but if you want to use it, you can extend them with your favorite filetypes in custom ftdetect rules.


If the logic for doing something involves more than a few lines or any structures like functions that can be decoupled from $MYVIMRC, I like to implement it as a plugin in ~/.vim/plugin and/or ~/.vim/autoload, with documentation for each in ~/.vim/doc.

They eventually get either discarded if I stop using them, or spun off into their own repositories and added to this repository as submodules under vim/bundle if I don't. Some of them I upload to vim.org.

Filetype plugins

I apply some replacement or supplementary configuration specific to file types I often edit in ~/.vim and ~/.vim/after, in the ftplugin, indent, and syntax subdirectories. Some of these filetype plugins or extensions may also eventually be removed to be separately distributed, and installed via submodules instead.


I define a few of my own :compiler scripts for ~/.vim/compiler to check and lint appropriate filetypes. I bind checking---"does it run?"---and linting---"is it good?"---with separate local leader maps; for example, for perl filetypes, <LocalLeader>c switches makeprg to perl -c for checking, and <LocalLeader>l to perlcritic for linting.

No Neovim support

The configuration doesn't explicitly support Neovim, although most of it will probably work; you would probably just comment out the settings for a few of the removed options.


Where practical, I make short scripts into POSIX (but not Bourne) sh(1), awk(1), or sed(1) scripts in ~/.local/bin. I try to use shell functions only when I actually need to, which tends to be when I need to change the state of the user's current shell, or to limit a change in behavior only to interactive shells.

These scripts are installed by the install-bin target:

  • Three SSH-related scripts:
    • sls(1df) prints hostnames read from ssh_config(5) files. It uses slsf(1df) to read each one.
    • sra(1df) runs a command on multiple hosts read from sls(1df) and prints output.
    • sta(1df) runs a command on multiple hosts read from sls(1df) and prints the hostname if the command returns zero.
  • Five URL-related shortcut scripts:
    • hurl(1df) extracts values of href attributes of <a> tags, sorts them uniquely, and writes them to stdout; it requires pup.
    • murl(1df) converts Markdown documents to HTML with pandoc(1) and runs the output through hurl(1df).
    • urlc(1df) accepts a list of URLs on stdin and writes error messages to stderr if any of the URLs are broken, redirecting, or are insecure and have working secure versions; requires curl(1).
    • urlh(1df) prints the values for a given HTTP header from a HEAD response.
    • urlmt(1df) prints the MIME type from the Content-Type header as retrieved by urlh(1df).
  • Three RFC-related shortcut scripts:
    • rfcf(1df) fetches ASCII RFCs from the IETF website.
    • rfct(1df) formats ASCII RFCs.
    • rfcr(1df) does both, displaying in a pager if appropriate, like a man(1) reader for RFCs.
  • Five toy random-number scripts (not for sensitive/dead-serious use):
    • rndi(1df) gets a random integer within two bounds.
    • rnds(1df) attempts to get an optional random seed for rndi(1df).
    • rnda(1df) uses rndi(1df) to choose a random argument.
    • rndf(1df) uses rnda(1df) to choose a random file from a directory.
    • rndl(1df) uses rndi(1df) to choose a random line from files.
  • Four file formatting scripts:
    • d2u(1df) converts DOS line endings in files to UNIX ones.
    • u2d(1df) converts UNIX line endings in files to DOS ones.
    • stbl(1df) strips a trailing blank line from the files in its arguments.
    • stws(1df) strips trailing spaces from the ends of lines of the files in its arguments.
  • Seven stream formatting scripts:
    • sd2u(1df) converts DOS line endings in streams to UNIX ones.
    • su2d(1df) converts UNIX line endings in streams to DOS ones.
    • slow(1df) converts uppercase to lowercase.
    • supp(1df) converts lowercase to uppercase.
    • tl(1df) tags input lines with a prefix or suffix, basically a sed(1) shortcut.
    • tlcs(1df) executes a command and uses tl(1df) to tag standard output and standard error lines, and color them if you want.
    • unf(1df) joins lines with leading spaces to the previous line. Intended for unfolding HTTP headers, but it should work for most RFC 822 formats.
  • Six simple aggregate scripts for numbers:
    • max(1df) prints the maximum.
    • mean(1df) prints the mean.
    • med(1df) prints the median.
    • min(1df) prints the minimum.
    • mode(1df) prints the first encountered mode.
    • tot(1df) totals the set.
  • Three quick-and-dirty HTML tools:
    • htenc(1df) encodes.
    • htdec(1df) decodes.
    • htrec(1df) wraps a tags around URLs.
  • Two internet message quoting tools:
    • quo(1df) indents with quoting right angle-brackets.
    • wro(1df) adds a quote attribution header to its input.
  • Six Git-related tools:
    • fgscr(1df) finds Git repositories in a directory root and scrubs them with gscr(1df).
    • grc(1df) quietly tests whether the given directory appears to be a Git repository with pending changes.
    • gscr(1df) scrubs Git repositories.
    • isgr(1df) quietly tests whether the given directory appears to be a Git repository.
    • jfc(1df) adds and commits lazily to a Git repository.
    • jfcd(1df) watches a directory for changes and runs jfc(1df) if it sees any.
  • Two time duration functions:
    • hms(1df) converts seconds to hh:mm:ss or mm:ss timestamps.
    • sec(1df) converts hh:mm:ss or mm:ss timestamps to seconds.
  • Three pipe interaction tools:
    • pst(1df) runs an interactive program on data before passing it along a pipeline.
    • ped(1df) runs pst(1df) with $EDITOR or ed(1).
    • pvi(1df) runs pvi(1df) with $VISUAL or vi(1).
  • Two editor wrapper tools:
    • mked(1df) creates paths to all its arguments before invoking $EDITOR.
    • mkvi(1df) creates paths to all its arguments before invoking $VISUAL.
  • ap(1df) reads arguments for a given command from the standard input, prompting if appropriate.
  • apf(1df) inserts arguments to a command with ones read from a file, intended as a framework for shell wrappers or functions.
  • ax(1df) evaluates an AWK expression given on the command line; this is intended as a quick way to test how AWK would interpret a given expression.
  • bcq(1df) runs bc(1), quieting it down if need be.
  • bel(1df) prints a terminal bell character.
  • bl(1df) generates a given number of blank lines.
  • bp(1df) runs br(1df) after prompting for an URL.
  • br(1df) launches $BROWSER.
  • ca(1df) prints a count of its given arguments.
  • cf(1df) prints a count of entries in a given directory.
  • cfr(1df) does the same as cf(1df), but recurses into subdirectories as well.
  • chc(1df) caches the output of a command.
  • chn(1df) runs a filter over its input a given number of times.
  • clog(1df) is a tiny timestamped log system.
  • clrd(1df) sets up a per-line file read, clearing the screen first.
  • clwr(1df) sets up a per-line file write, clearing the screen before each line.
  • csmw(1df) prints an English list of monospace-quoted words read from the input.
  • dam(1df) buffers all its input before emitting it as output.
  • ddup(1df) removes duplicate lines from unsorted input.
  • defang(1df) prevents dangerous URLs from being made into clickable links.
  • dub(1df) lists the biggest entries in a directory.
  • edda(1df) provides a means to run ed(1) over a set of files preserving any options, mostly useful for scripts.
  • eds(1df) edits executable script files in EDSPATH, defaulting to ~/.local/bin, for personal scripting snippets.
  • exm(1df) works around a screen-clearing quirk of Vim's ex mode.
  • finc(1df) counts the number of results returned from a set of given find(1) conditions.
  • fnl(1df) runs a command and saves its output and error into temporary files, printing their paths and line counts.
  • fnp(1df) prints the given files to standard output, each with a plain text heading with the filename in it.
  • gms(1df) runs a set of getmailrc files; does much the same thing as the script getmails in the getmail suite, but runs the requests in parallel and does up to three silent retries using try(1df).
  • grec(1df) is a more logically-named grep -c.
  • gred(1df) is a more logically-named grep -v.
  • gwp(1df) searches for alphanumeric words in a similar way to grep(1).
  • han(1df) provides a keywordprg for Vim's Bash script file type that will look for help topics. You could use it from the shell too.
  • igex(1df) wraps around a command to allow you to ignore error conditions that don't actually worry you, exiting with 0 anyway.
  • ix(1df) posts its input to the ix.io pastebin.
  • jfp(1df) prints its input, excluding any shebang on the first line only.
  • loc(1df) is a quick-search wrapped around find(1).
  • maybe(1df) is like true(1) or false(1); given a probability of success, it exits with success or failure. Good for quick tests.
  • mex(1df) makes given filenames in $PATH executable.
  • mi5(1df) is a crude preprocessor for m4.
  • mim(1df) starts an interactive Mutt message with its input.
  • mftl(1df) finds usable-looking targets in Makefiles.
  • mkcp(1df) creates a directory and copies preceding arguments into it.
  • mkmv(1df) creates a directory and moves preceding arguments into it.
  • motd(1df) shows the system MOTD.
  • msc(1df) crudely counts messages in an mbox.
  • mw(1df) prints alphabetic space-delimited words from the input one per line.
  • oii(1df) runs a command on input only if there is any.
  • onl(1df) crunches input down to one printable line.
  • osc(1df) implements a netcat(1)-like wrapper for openssl(1)'s s_client sub-command.
  • p(1df) prints concatenated standard input; cat(1) as it should always have been.
  • pa(1df) prints its arguments, one per line.
  • phpcsff(1df) wraps around PHP-CS-Fixer to make it a source code filter suitable for use as an equalprg in Vim.
  • pp(1df) prints the full path of each argument using $PWD.
  • pph(1df) runs pp(1df) and includes a leading $HOSTNAME:.
  • paz(1df) print its arguments terminated by NULL chars.
  • pit(1df) runs its input through a pager if its standard output looks like a terminal.
  • pwg(1df) generates just one decent password with pwgen(1).
  • qat(1df) disables stty echo for the duration of a paste.
  • rep(1df) repeats a command a given number of times.
  • rgl(1df) is a very crude interactive grep(1) loop.
  • shb(1df) attempts to build shebang lines for scripts from the system paths.
  • sqs(1df) chops off query strings from filenames, usually downloads.
  • sshi(1df) prints human-readable SSH connection details.
  • stex(1df) strips extensions from filenames.
  • sue(8df) execs sudoedit(8) as the owner of all the file arguments given, perhaps in cases where you may not necessarily have root sudo(8) privileges.
  • swr(1df) allows you to run commands locally specifying remote files in scp(1)'s HOST:PATH format.
  • td(1df) manages a to-do file for you with $EDITOR and git(1); I used to use Taskwarrior, but found it too complex and buggy.
  • tm(1df) runs tmux(1) with attach-session -d if a session exists, and new-session if it doesn't.
  • trs(1df) replaces strings (not regular expression) in its input.
  • try(1df) repeats a command up to a given number of times until it succeeds, only printing error output if all three attempts failed. Good for tolerating blips or temporary failures in cron(8) scripts.
  • umake(1df) iterates upwards through the directory tree from $PWD until it finds a Makefile for which to run make(1) with the given arguments.
  • uts(1df) gets the current UNIX timestamp in an unorthodox way that should work on all POSIX-compliant operating systems.
  • vest(1df) runs test(1) but fails with explicit output via vex(1df).
  • vex(1df) runs a command and prints true or false explicitly to stdout based on the exit value.
  • vic(1df) tries to run a POSIX-compliant vi(1).
  • xrbg(1df) applies the same randomly-selected background to each X screen.
  • xrq(1df) gets the values of specific resources out of xrdb -query output.

There's some silly stuff in install-games:

  • aaf(6df) gets a random ASCII Art Farts comic.
  • acq(6df) allows you to interrogate AC, the interplanetary computer.
  • aesth(6df) converts English letters to their full width CJK analogues, for aesthetic purposes.
  • squ(6df) makes a reduced Latin square out of each line of input.
  • kvlt(6df) translates input to emulate a style of typing unique to black metal communities on the internet.
  • philsay(6df) shows a picture to accompany pks(6df) output.
  • pks(6df) laughs at a randomly selected word.
  • rndn(6df) implements an esoteric random number generation algorithm.
  • strik(6df) outputs s̶t̶r̶i̶k̶e̶d̶ ̶o̶u̶t̶ struck out text.
  • rot13(6df) rotates the Latin letters in its input.
  • uuu(6df) uuuuu uuuu uu uuuuuu uuuuuuu u uuu uuuuu.
  • xyzzy(6df) teleports to a marked location on the filesystem.
  • zs(6df) prefixes "z" case-appropriately to every occurrence of "s" in the text on its standard input.


The install-bin and install-games targets install manuals for each script. If you want to read the manuals, you may need to add ~/.local/share/man to your ~/.manpath or /etc/manpath configuration, depending on your system.


You can check that both sets of shell scripts are syntactically correct with make check-bash or make check-sh, or make check for everything including the scripts in bin and games. There's no proper test suite for the actual functionality (yet).

There are also optional lint targets, if you have the appropriate tools available to run them:

  • ShellCheck:
    • lint-bash
    • lint-bin
    • lint-games
    • lint-ksh
    • lint-sh
    • lint-x
  • Vint:
    • lint-vim

Future development

See IDEAS.md.

Known issues

See ISSUES.md.


Public domain; see the included UNLICENSE file. It's just configuration and simple scripts, so do whatever you like with it if any of it's useful to you. If you're feeling generous, please join and/or donate to a free software advocacy group, and let me know you did it because of this project: