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.


$ git clone https://sanctum.geek.nz/code/dotfiles.git ~/.dotfiles
$ cd ~/.dotfiles
$ git submodule init
$ git submodule update
$ make
$ make -n install
$ make install

For the default all target, you'll need bash(1), make(1), and m4(1).

The installation Makefile will overwrite 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 to make sure you aren't going to lose anything unexpected. If you're still not sure, install it in a temporary directory so you can explore:

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

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

  • install-bash
  • install-bin
  • install-bin-man
  • install-curl
  • install-ex
  • install-git
  • install-gnupg
  • install-less
  • install-readline
  • install-sh
  • install-vim

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


Configuration is included for:

  • Bourne-style POSIX shells, sharing an ENV file and functions:
  • Abook -- curses address book program including a ~/.profile configured to work with most Bourne-compatible shells
  • cURL -- Command-line tool for transferring data with URL syntax
  • Dunst -- A lightweight X11 notification daemon that works with libnotify
  • finger(1) -- User information lookup program
  • Git -- Distributed version control system
  • GnuPG -- GNU Privacy Guard, for private communication and file encryption
  • GTK+ -- GIMP Toolkit, for graphical user interface elements
  • i3 -- Tiling window manager
  • less -- Terminal pager
  • mail(1) -- classic mail program
  • Mutt -- Terminal mail user agent
  • mysql(1) -- Command-line MySQL client
  • Ncmpcpp -- ncurses music player client
  • Newsbeuter -- Terminal RSS/Atom feed reader
  • psql(1) -- Command-line PostgreSQL client
  • Perl::Critic -- static source code analysis engine for Perl
  • Perl::Tidy -- Perl indenter and reformatter
  • Readline -- GNU library for user input used by Bash, MySQL, and others
  • rxvt-unicode -- Fork of the rxvt terminal emulator with Unicode support
  • Subversion -- Apache Subversion, a version control system
  • tmux -- Terminal multiplexer similar to GNU Screen
  • Vim -- Vi IMproved, a text editor
  • X11 -- Windowing system with network transparency for Unix

The configurations for shells, GnuPG, Mutt, tmux, and Vim are the most expansive, and most likely to be of interest. The i3 configuration is mostly changed to make window switching behave like Vim windows and tmux panes do, and there's a fair few resources defined for rxvt-unicode.


My .profile and other files in sh are written in POSIX shell script, so they should work in most sh(1) implementations. Individual scripts called by .profile are saved in .profile.d and iterated on login for ease of management. 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 POSIX-fearing interactive shells is kept in ~/.shrc, with subscripts read from ~/.shrc.d. There's a shim in ~/.shinit to act as ENV. I make an effort to target POSIX for my functions and scripts where I can so that the same files can be loaded for all shells.

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

As I occasionally have work on very old internal systems, my Bash is written to work with any version 2.05a or newer. This is why I use older syntax for certain things such as appending items to arrays:


Compare this to the much nicer syntax available since 3.1-alpha1, which actually works for arrays with sparse indices, unlike the above syntax:


Where I do use features that are only available in versions of Bash newer than 2.05a, such as newer shopt options or PROMPT_DIRTRIM, they are only run after testing BASH_VERSINFO appropriately.


A terminal session with my prompt looks something like this:

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

The username and hostname are elided if not connected via SSH. The working directory 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, and punctuation to show repository status including reference to upstreams at a glance. Subversion support can also be enabled (I need it at work), in which case a git: or svn: prefix is added appropriately.
  • 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 Bash, Ksh, or Zsh and it detects that it's not normally your $SHELL, the prompt will 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:
    • ad() is a cd shortcut accepting targets like /u/l/b for /usr/local/bin, as long as they are unique.
    • 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.
  • bc() silences startup messages from GNU bc(1).
  • ed() tries to get verbose error messages, a prompt, and a Readline environment for ed(1).
  • env() sorts the output of env(1) if it was invoked with no arguments, just for convenience when running it interactively.
  • gdb() silences startup messages from gdb(1).
  • gpg() quietens gpg(1) down for most commands.
  • grep() tries to apply color and other options good for interactive use if available.
  • hgrep() allows searching $HISTFILE.
  • keychain() keeps $GPG_TTY up to date if a GnuPG agent is available.
  • 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.
  • mysql() allows shortcuts to MySQL configuration files stored in ~/.mysql.
  • 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.
  • 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 Zsh function I like by the same name (Bash).
  • ver() prints the current shell's version information (Bash, Korn Shell, Yash, 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 .bash_completion.d files, 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
  • ftp(1) hostnames from ~/.netrc
  • git(1) subcommands, remotes, branches, tags, and addable files
  • gpg(1) long options
  • make(1) targets read from a Makefile
  • man(1) page titles
  • mysql(1) databases from ~/.mysql/*.cnf
  • 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.


Just enough configuration to coax it into reading ~/.profile and ~/.shrc.


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


The configuration for GnuPG is intended to follow RiseUp's OpenPGP best practices. The configuration file is rebuilt using m4(1) and make(1) because it requires hard-coding a path to the SKS keyserver certificate authority, and neither tilde nor $HOME expansion works for this.


My mail is kept in individual Maildirs under ~/Mail, with inbox being where most unfiltered mail is sent. 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.


I've butchered the URxvt Perl extensions selection-to-clipboard and selection into a single select extension in ~/.urxvt/ext, which is the only extension I define in ~/.Xresources.

The included .Xresources file assumes that urxvt can use 256 colors and Perl extensions. If you're missing functionality, try changing perl-ext-common to default.

My choice of font is Ubuntu Mono, but the file should allow falling back to the more common Deja Vu Sans Mono. I've found Terminus works well too, but bitmap fonts are not really my cup of tea. The Lohit Kannada font bit is purely to make ಠ_ಠ work correctly. ( ͡° ͜ʖ ͡°) seems to work out of the box.


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

The configuration file is created with m4(1) to allow specifying a color theme. This is just because I use a different color for my work session. The default is a dark grey.

The configuration for Bash includes a tmux function designed to make attach into the default command if no arguments are given and sessions do already exist. The default command is normally new-session.

My ~/.inputrc file binds Alt+M to attach to or create a tmux session, and Tmux in turn binds the same key combination to detach.


The majority of the .vimrc file is just setting options, with a few mappings. I try not to deviate too much from the Vim defaults behaviour in terms of interactive behavior and keybindings.

The configuration is extensively commented, mostly because I was reading through it one day and realised I'd forgotten what half of it did. Plugins are loaded using @tpope's pathogen.vim.


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 tinker with the namespace of the user's current shell.

Installed by the install-bin target:

  • Three SSH-related scripts:
    • sls(1df) prints hostnames read from a ssh_config(5) file. 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 stdout and stderr 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 aggregators 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.
  • ap(1df) reads arguments for a given command from the standard input, prompting if appropriate.
  • apf(1df) prepends 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, or a more suitable application for an URL if it knows of one.
  • 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.
  • ddup(1df) removes duplicate lines from unsorted input.
  • dmp(1df) copies a pass(1) entry selected by dmenu(1) to the X CLIPBOARD.
  • 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.
  • 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).
  • gwp(1df) searches for alphanumeric words in a similar way to grep(1).
  • han(1df) provides a keywordprg for Vim's Bash script filetype 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.
  • 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.
  • onl(1df) crunches input down to one printable line.
  • osc(1df) implements a netcat(1)-like wrapper for openssl(1)'s s_client subcommand.
  • pa(1df) prints its arguments, one per line.
  • 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.
  • plmu(1df) retrieves a list of installed modules from plenv, filters out any modules in ~/.plenv/non-cpan-modules, and updates them all.
  • pwg(1df) generates just one decent password with pwgen(1).
  • 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.
  • 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.
  • 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 fullwidth CJK analogues, for AESTHETIC PURPOSES.
  • kvlt(6df) translates input to emulate a style of typing unique to black metal communities on the internet.
  • 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.
  • xyzzy(6df) teleports to a marked location on the filesystem.
  • zs(6df) prepends "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 they install. There's also an install-dotfiles-man target that uses pandoc(1) to reformat this document as a manual page for section 7 (dotfiles(7df)) if you want that. I haven't made that install by default, because pandoc(1) is a bit heavy.

If you want to use 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, 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).

If you have ShellCheck and/or Perl::Critic, there's a lint target for the shell script files and Perl files respectively. The files don't need to pass that check to be installed.

Known issues

See ISSUES.markdown.

Note for previous visitors

Most of this repository's five-year history was rewritten shortly after I moved it from GitHub to cgit, taking advantage of the upheaval to reduce its size and remove useless binary blobs and third-party stuff that I never should have versioned anyway. If you've checked this out before, you'll probably need to do it again, and per-commit links are likely to be broken. Sorry about that.


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: