My Emacs config lives in a massive org-mode file and I often want to re-tangle from outside Emacs (because, e.g., I broke my Emacs session for some reason). I could just use a simple Makefile, but that’s no fun. Instead, I’ve turned my config into a self-tangling script by adding the following two lines to the top of my init.org file and marking it as executable:

#!/usr/bin/env -S sh -c 'emacs --batch --eval="(setq org-confirm-babel-evaluate nil)" --file "$0" -f org-babel-tangle'
# -*- mode: org; lexical-binding: t -*-

Now I can run ./init.org in my ~/.emacs.d directory and my config will tangle itself into the correct files.

This might sound like a weird gimmick – to be honest, it kind of is – but I’ve been tangling my config this way for years at this point and it “just works” with no dependencies other than Emacs.

A similar trick can be used to execute all org-babel blocks in an org-mode file, making org-mode a viable meta-scripting language for tying together scripts written in different languages – even scripts executing across multiple machines via TRAMP.

#!/usr/bin/env -S sh -c 'emacs --batch --eval="(setq org-confirm-babel-evaluate nil)" --file "$0" -f org-babel-execute-buffer'
# -*- mode: org; lexical-binding: t -*-
#+TITLE: Hello World

#+BEGIN_SRC elisp
(message "Hello World!")
#+END_SRC

While there are many ways to pipe to emacs, they all involve either shuttling text by repeatedly calling emacsclient or writing to a temporary file. However, neither are necessary.

Basically, while emacs can’t (yet) read from a named pipe (FIFO), it can read standard output from a process so, one gratuitous use of cat later…

(defun pager-read-pipe (fname)
  (let ((buf (generate-new-buffer "*pager*"))
        (pname (concat "pager-" fname)))
    (with-current-buffer buf (read-only-mode))
    (switch-to-buffer buf)

    (let ((proc (start-process pname buf "/usr/bin/cat" fname)))
      (set-process-sentinel proc (lambda (proc e) ()))
      (set-process-filter proc (lambda (proc string)
                                 (when (buffer-live-p (process-buffer proc))
                                   (with-current-buffer (process-buffer proc)
                                     (save-excursion
                                       ;; Insert the text, advancing the process marker.
                                       (let ((inhibit-read-only t))
                                         (goto-char (process-mark proc))
                                         (insert string)
                                         (set-auto-mode)
                                         (set-marker (process-mark proc) (point))))))))
      proc)))

…and you can read a from named pipe. As an added bonus, this function will try to autodetect the correct mode.

To actually use this, I recommend the following shell script:

#!/bin/bash
set -e

cleanup() {
    trap - TERM INT EXIT
    if [[ -O "$FIFO" ]]; then
        rm -f "$FIFO" || :
    fi
    if [[ -O "$DIR" ]]; then
        rmdir "$DIR" || :
    fi
}
trap "cleanup" TERM INT EXIT

SOCKET="${XDG_RUNTIME_DIR:-/run/user/$UID}/emacs/server"

# Create a named pipe in /dev/shm
DIR=$(mktemp -d "/dev/shm/epipe-$$.XXXXXXXXXX")
FIFO="$DIR/fifo"
mkfifo -m 0600 "$DIR/fifo"

# Ask emacs to read from the names socket.
emacsclient -s "$SOCKET" -n --eval "(pager-read-pipe \"$FIFO\")" >/dev/null <&-

exec 1>"$FIFO"
cleanup # Cleanup early. Nobody needs the paths now...
cat

You will probably need to set the SOCKET variable to your emacs socket filename.

Usage:

dmesg --follow | epipe