Analysis of Emacs keys pressed

Here’s the deal: Emacs keybindings make my fingers hurt. I don’t think I ever experienced RSI before I started using Emacs. I guess I’ve been using Emacs for about 6 years. I’m very efficient with it. I can edit almost as fast as I can think, my fingers never need to take a break. But that efficiency comes at a steep price, I feel.

I hypothesize that chords are to blame, and that I would be happier and less achey if I used a modal set of keybindings, like in Vim, in which every key binding is a single character. Not all the keybindings (e.g. $) are a single key press, but most are.

I’ve tried evil-mode, and it’s pretty poor. It doesn’t provide a proper mapping to Emacs; hitting $ doesn’t actually execute move-end-of-line, it executes evil-end-of-line, which does not integrate with existing modes well at all. It’s catering to Vimers, but it’s not good for Emacs power users.

I suspect that I would like to have a global modal switcher that will make C- and M- implicit somehow, so that a SPC e w is equivalent to typing C-a C-SPC C-e C-w. Before sitting down to develop such a system, tackle the problem of how to start and exit the mode, and how to deal with the meta key, I thought I would collect some statistics. (And actually there are systems like sticky keys or chords for Emacs for tackling stuff like this, so it’s not a scary, new area.)

What I wanted to prove (or collect evidence for) was:

I already had a trivial script to print key presses for screencasts, so I modified that to also store the time and mode in the buffer, and I opened a keys.log file to which I would save the key presses for a day.

I then whipped up a script to read in those statistics and print out a summary, to (hopefully) provide evidence for the above claims.

The output is the following:

Recording start/end: 2013-08-07 09:52:23 UTC/2013-08-08 07:54:23 UTC

Time spent: 22 hours (not 100% activity)

Total key presses: 29687

Commands (including character presses): 22657

Single-key commands: 16457

C- or M- commands: 6200 (27.36%)

Runs of (consecutively) unique C-/M- clusters: min/max/avg/stddev: 1/45/2.25/2.52

Runs of non-unique C-/M- clusters: min/max/avg/stddev: 1/189/3.35/7.04

Key presses used on C-/M- commands: 13230 (44.56%)

Runs of C-f/C-n/C-p/C-b: min/max/avg/stddev: 1/39/2.96/4.63

Key presses used on C-f/C-b/C-n/C-p: 4572 (15.40% of all key presses, 34.56% of C-/M- command key presses)

Top commands used:

 1 |   750 | C-n
 2 |   716 | C-p
 3 |   355 | C-f
 4 |   341 | C-/
 5 |   335 | C-b
 6 |   259 | M-DEL
 7 |   248 | C-z
 8 |   245 | M-b
 9 |   231 | C-e
10 |   221 | M-p
11 |   193 | C-d
12 |   189 | M-f
13 |   157 | C-s
14 |   145 | C-M-u
15 |   142 | C-g
16 |   136 | C-a
17 |   117 | C-y
18 |   109 | C-x C-s
19 |   107 | M-
20 |    92 | C-SPC

Terms:

For unique clusters, I’m doing 2.26 commands per cluster. So if I used sticky keys, or a modal switcher, it would not be a gain. E.g. C f f C vs C-f C-f is no gain, it’s actually more presses due to having to hit C again.

But in terms of non-unique clusters, there’s a gain at 3.44 commands per cluster. That means C f f f C vs C-f C-f C-f, which is one key less pressed. If I’m pressing 9218 keys for C-/M- commands, there might be a 20% decrease in key presses.

I’d love to see a similar analysis done of Vim. How often do Vim users switch from insert mode to normal or presentation mode? I will continue recording my keys for the next couple of days.

Very interesting is how much I use navigation functions. In reaction to this, I’m disabling those keybindings and switching to arrow keys. And I’ve found M-e, a more convenient binding for C-M-u. I will also stop using C-d and use DEL.

Follow-up page →