Home GitHub Patreon
RSS Twitter

Annoying prefix argument

Problem

A prefix argument changes the behaviour of a command in a way which forms a useful stand-alone command. We don't want to supply the command with the prefix argument all the time.

Solution

Write an interactive wrapper and pre-fill the arguments

Motivation

It is awkward and annoying to constantly use prefix-arguments for commonly used operations. Emacs has a rich history of commands using prefix arguments to subtly (or not so subtly) modify the behaviour of the base command. If you often use a specialized variant you can save yourself a lot of pain by simply building a specialized solution. It is not uncommon to see two prefix arguments (C-u C-u), but I've seen four and even five once or twice.

Example

Emacs comes with this amazing functionality called marks. They act a bit like "automatic" bookmarks within the current buffer (there is also a global version but I won't go into that right now). Marks live in the mark ring1. This allows you to remember multiple marks at the same time and recall them in a lifo order.

When you move around the buffer, Emacs pushes the position where you started from to the mark ring and lets you jump back by popping the mark.

Most commands which move the point, like isearch, beginning-of-buffer, imenu, next-defun set the mark in somewhat intelligent way. For example, if you call next-defun multiple times it pushes the mark only when you call it for the first time. That way, when you recall the mark you jump to the "original" defun five functions up without having to recall five times.

This is all good and well, but dealing with marks turned out rather frustrating for me. The most common action, namely recalling the mark, is bound to arather awkward C-u C-SPC. It doesn't help in the least that the command is called set-mark-command, in fact, the same command you use to set the mark, just with a prefix argument.

Here we can identify a rather common Emacs annoyance: prefix argument changes the behaviour of a command into something completely different; but we want to use both versions in a simple way!

How can we solve this? We write our own interactive wrapper!

Resolution

First we need to figure out which argument and with what value we will need to pre-fill. Most of the time reading the documentation with C-h f will be enough to give us a clue, but sometimes inspecting the source is the only way to be sure.

In our example, we know that we call the function with C-u which corresponds to Elisp value (4) (that is a list with 4 in it). Each extra C-u multiplies the number in the list by 4. There is only one argument to the function set-mark-command so that's the position we will need to fill. The result is then the following function.

(defun my-jump-to-mark ()
  "Jump to the local mark, respecting the `mark-ring' order.
This is the same as using \\[set-mark-command] with the prefix argument."
  (interactive)
  (set-mark-command '(4)))

We can bind this command to a simple key binding, for example

(global-set-key (kbd "M-`") 'my-jump-to-mark)

Sometimes, the prefix argument isn't passed as an argument but is read from a special variable current-prefix-arg. When you call a command using a prefix argument, Emacs automatically sets this variable to the corresponding Elisp value.

Before we continue, a little excursion into how Emacs maintains value bindings. When you use let to set the value of a special variable2, this value becomes current until the execution of the let block is finished. This means that this value is available globally for all the subroutines called from the let block, as well as for all the subroutines called from these and so on.

Armed with this knowledge, we can try it on an example. The comand align-regexp changes the meaning of its 3rd argument based on the value of the special variable current-prefix-arg. We will write an interactive wrapper which we will call without any prefix argument. This means Emacs will set the value of current-prefix-arg to nil. We will use dynamic binding to set the value to (4) and then call the original function.

Because we want to respect the interactive behaviour of the original function we will use call-interactively which calls the function providing arguments according to its interactive calling specification.

The result is then quite simple again.

(defun my-align-regexp ()
  "Call `align-regexp' with `current-prefix-arg' set to (4)."
  (interactive)
  (let ((current-prefix-arg '(4)))
    (call-interactively 'align-regexp)))

Notice that we could have used this technique in the first example as well, because the interactive specifications "p" and "P" simply check the value of current-prefix-arg and pass it forward. The code would look like so:

(defun my-jump-to-mark-2 ()
  "Jump to the local mark, respecting the `mark-ring' order.
This is the same as using \\[set-mark-command] with the prefix argument."
  (interactive)
  (let ((current-prefix-arg '(4)))
    (call-interactively 'set-mark-command)))

We can't, however, use the first method on align-regexp because none of its arguments directly corresponds to the prefix argument—it is only inside the body that this is checked.

The moral of the story is that we should always prefer passing the argument directly as it is cleaner and less magical. Only if the prefix argument is not direct but checked inside the body of the function we shall use dynamic binding.

Footnotes:

1
A ring in Emacs is basically a stack where popped elements go to the bottom
2
A variable is special if defined using defvar.

Last updated at: 2017-06-04 15:39
Found a typo? Edit on GitHub!