A few days ago, my article was posted to lobste.rs and it got a comment from Celeritas linking me to this argument in favor of vim-style editing rather than kakoune-style editing. I was typing up a response in the thread, but it ended up getting pretty long, so I decided to turn it into a sort of follow-up post.
That article is written by noctuid, a pretty prolific emacs community member (i remember reading their evil guide when I first started using doom something like 6 years ago now). I’ve never used kakoune, but even though I ultimately disagree with their take I can relate to many parts of it. Meow fixes many of the issues they mention with kakoune. I think that with editor discussions in general there’s an exceptional amount of subjective bias because of the cost it takes to learn new things. I’ll be biased because I sunk the cost to learn kakoune grammar, and that’s what I currently use. I don’t think there are many people who’ve switched the other way.
In this post I’d like to comment on some of the critiques noctuid makes about kakoune, especially with regards to navigation and selection and how meow seems to address many of their points. In some sense, this is a bit more of a sobered take compared to my post 2 years ago. It comes with a bit more experience and knowledge of how meow works internally.
2 Critique points
2.1 Reverse grammar hurts key reuse
The idea is that operator-pending mode is a separate context in which
keys can be re-used, like how
w means “word” after
d but “next
word” in general. Noctuid takes issue with having to use a modifier to
select text objects.
I think it’s really just a layout issue – indeed
in meow we use
. on the recommended layout. When you’re
selecting a text object, that is a separate context in which keys
are interpreted differently. You can reuse every key to select a text
object. I might be misunderstanding what was meant by this part, but
as I see it you’re losing nothing.
2.2 Navigation vs Selection
When I first started using meow, I might have agreed that automatic selections were a bit distracting, but once you start ignoring them it really becomes a non-issue. In meow, any movement key or command that does not act on a selection cancels the active selection, meaning that if you make a motion without wanting to act on it the selection will probably disappear on your very next command. However, it probably is a very valid observation that most people will very rarely make a motion and THEN realize they want to act on it. I just think that there is a real benefit to automatic selections, and it’s worth it for the cases where you select intending to act.
2.3 Visual selections don’t prevent mistakes; overlays do
Noctuid correctly points out that visual selections don’t really prevent you from making mistakes, they only make it obvious and easy to fix when you do.
Luckily meow has visual selections and overlays. Meow implements nearly exactly what noctuid describes, but with great integration (compared to avy) and really smooth ergonomics for the vast majority of everyday editing use cases. In short, most movement commands in meow trigger overlays over the target locations of all possible repetitions of that command, letting you quickly and visually repeat a command some number of times.
I don’t agree that visual selections don’t prevent mistakes – I
couldn’t tell you how many times I’ve fat fingered 8 instead of 7 or
simply changed my mind after making a selection. Additionally, having
both gives you two options: either press the right number by reading
the hints, or spam. Some commands in meow reuse keys for their expand
versions. For instance, if you select a line with
pressing the same key again selects one more line. If you mark a whole
expands by one word instead of selecting the next word only.
A subtle point which even many long time users of meow might not know the specifics of is this expandable selections. Every time you make a selection in meow, it is marked with some information that informs meow what to do on a repeat: whether it is an “expand” selection or a “select” selection, and what object the selection is targeting. The commands that make an “expand” selections are
- up/down/right/left expansions
- word/symbol based expansions
- line expansions
- block expansions (paren/bracket pairs,
up the tree).
The commands that make “select” selections and therefore do not reuse keys are
- find/till character
- general text-objects (perhaps might change in the future, there’s a lot of discussion about this on github)
This whole setup makes expanding selections much easier, and basically gives you same user experience as highlighting text with a mouse: choosing your selection one object at a time. In my experience, I tend to spam keys to select <5 text objects, it’s just faster and takes less thought. In this “spam” selection mode, visual selection indeed does prevent selection errors, it’s slower and less efficient but exact.
2.4 Visual confirmation is unnecessary
From the article:
Furthermore, the effect of many motions and text objects is obvious; visual confirmation is often completely unnecessary. Deleting the current word, sentence, or parentheses block does not involve uncertainty. t and f are basically the “dumbest” motions in that they don’t move by clear units like paragraphs or words but instead jump to arbitrary characters. The majority of my operator usage has no possibility for mistakes apart from mistyping.It doesn’t make sense to me to optimize a modal model around rarer cases like the dtf example, even if I thought a visual selection was the ideal way to handle these cases.
I mostly agree with this point. A lot of my editing tasks are just
“delete this word” or “change inside parens” which are so fast I
barely see the selection before it’s gone. In this case though, visual
selections cost nothing. So, I don’t agree with is
the tradeoff part of their argument. It does make sense to optimize
modal models around error prone cases, especially when there is
minimal (and to me, mostly aesthetic) cost for doing so. I don’t think
these cases are that rare; jumping to the
n th occurrence of a
character or selecting
n lines is quite common!
2.5 Evil integration is better
In the “why not other emacs packages?” section, noctuid mentions that emacs packages use different keybindings for different actions, and that modal editing packages face a fundamental convention difference for motion keys (jk vs C-n, C-p). This is a good point to mention meow’s “motion” state, meant to address exactly this.
Essentially, it lets you override your motion keys in every mode that steals them
from you automatically. Evil expects you to remap your keys in every
one of your modes. Meow decides which modes to use motion in using a
funny little heuristic that checks whether keys self-insert into the
buffer and recursively against a predefined list of parent modes. Any
command that’s hidden is rebound under the
Hyper modifier, So, if
you need to use the command that used to be bound to your motion keys, you
H-<key> under your leader prefix. It’s a dead simple idea,
and it’ll always work.
While motion works for interaction-style packages like elfeed and magit, unfortunately, this is ultimately a valid criticism. Meow is forced to adapt to packages that nontrivially change your editing experience itself, like company, polymode, sly and cider to name a few. For these, meow comes with some “shims” that mostly just turn on the motion mode when its needed. Inevitably, this leads to bad user experience with some missed packages – but luckily, most compatibility issues are trivially solved by just turning on motion mode when it’s necessary, and indeed most shims that meow has do just that.
3 Ending thoughts
Noctuid’s post is 6 years old, and my response is obviously unfair since they’re writing about kakoune, and I’m talking about a modern package that has had the chance to learn from the many modal editing packages. Still, many of the critiques are about kakoune’s model itself, which meow uses. I still think it’s a good idea, or at least surely not a “a broken solution to a non-issue.” The instantaneous feedback and visual selections cost is mostly aesthetic, with a benefit of making selection errors trivial to notice and fix before acting on them.
At the end of the day, people edit differently – editors should be molded to the desires of the users, not the other way around.