Emacs-ng: Emacs with Deno runtime and TypeScript
github.com
github.com
I am not an Emacs user but I like Scheme (scm user here) under nvi, and a lot of people could profit from a Modern Guile based Emacs environment.
It works in principle but is slow due to some impedance mismatches that have to be bridged. Especially, Emacs has a special string format internally that is not native utf-8.
Andy Wingo is making a great job with JIT-support for Guile. We have to see whether emacs-guile can become fast enough to replace native-elisp. At this time it doesn't look like it...
Ain't gonna get far once the main developer loses interest in the project. Emacs users use Emacs because they want Lisp and don't care about web rendering and javascript runtimes ... Oh yeah and Rust is thrown in too for good measure.
Good 90% of Emacs code is in the extensions which are in Lisp and nobody is going to rewrite them to Javascript even if the code was running 50x faster.
It's like saying that a browser skin is "built on Rust" when inside it's still Webkit and Chromium.
I'm certain you're telling the truth. And I'm sure that's the case for a lot of old-timers emacs users. But I have the feeling it's less and less true.
People that have used emacs for a long time don't need anything else, so they don't want any change.
People for whom emacs doesn't cover everything want change, but clash against the first group of people (who have been there/contributed for longer), and in the end migrate somewhere else.
But my way of working is : many little projects in various languages, much text reports with Latex/markdown, lots of notes (orgmode), bit of email, bit of IRC. In a traditional business, I'd had to use beefier IDE such as IntelliJ (hat beats emacs without a doubt) and maybe Word. But for the rest, emacs fills all the little holes...
One area where I find emacs lacking is appointment (org mode is too big for little things such as quick reminders for today's stuff); a good calculator (calc is very clumys if you compare it to SpeedCrunch for example); a good calendar (emacs calendar is a nightmare to use, for exampe why on earth doesn't display what happens in a day below the calendar and instead forces me to hit 'd' which opens a new mostly empty buffer...); a good console on Windows (on Linux vterm is mostly perfect). Email support is OK with Wanderlust but a nightmare to set up.
Also, since it's very old, there's this warm comforting feeling that will last forever. And also I'm GPL zealot, which helps too :-)
Given enough time I can make Emacs do pretty much anything. I just don't have that much free time (anymore).
You can have C / Rust / any_lang extensions like RipGrep to boost some intensive workloads as well.
Uh, in the age of the 386 and 4MB of RAM, if you had that setup against a 486 an 8MB, you chose between running X and running Emacs. OFC you could run jmacs perfectly.
I would love a normal extension language, that also isn’t a hack (like Python integration). I’m not a JS guy but I’ll take it over lisp.
I’m not sure what the web render things are, but if it means some kind of browser integration, that sounds good too. I use org mode, often with html export and latex maths, having it render quickly and seamlessly would be a win for me.
Though, it's true that the many elisp-code around usca heavy burden. But how much of that is actually still used? And how fast would a propsering community replace them really? Hacking-friendly tools tend to have a wild growing ecosystem. Emacs itself is not shy of having many clones of similar featursets and regular new implementations of old but still popular features.
In that view I don't thing a new language would a real problem long term for the prosperty of emacs. It might be even a benefit, as it mixes in new blood, more people and a broader ecosystem from outside the emacs-scope.
Sure the Elisp bytecode itself doesn't evaluate faster, but that distinction isn't really important here. The end-to-end perf of certain functions can be improved in a backwards-compatible way.
I don't see the point of polluting Emacs with Javascript.
I'm reminded of when an old C hand tells me that Rust will never replace C because it's too complex, brittle, and frustrating for the kinds of applications C programmers use C for. I always tell them: Rust isn't aimed at you -- it's aimed at your replacement.
The Emacs-ng team doesn't have to convince the Emacs community -- they only have to convince their replacements. Much of computing in the very near future will be built on two languages: Rust and JavaScript. By basing Emacs-ng on those two languages, they just opened Emacs up to extension and hacking by a huge community who have no interest in touching Emacs's ancient, doddering Lisp nor its C underpinnings (C being, inherently, unsafe at any speed to work in). So ugly as it is, from a social standpoint it's absolutely the right approach and may well overtake GNU Emacs in terms of ecosystem size and vibrancy by the late 2020s.
It will be interesting to see if this prediction pans out. There have been similar predictions about C++, Java, TCL, php, Perl, etc. it seems “the one true language” never emerges, but it is just around the corner. Though I really hope the future is more polyglot, because that is just way more fun.
I hope so too, but... I've worked with people who really don't want to touch anything besides JavaScript. Such people could be coaxed to use Rust to write kernel drivers and the like, due to its safety guarantees -- but not C, nor C++. JavaScript is in everything. Both of the major DEs for Linux embed JavaScript interpreters. Microsoft is adding it to their office suite, to coexist with and eventually supplant VBA. Inasmuch as developers are still writing desktop apps, odds are good they're Electron apps. It's as near to a universal language as we've had in decades. Instructions are being added to ARM to better support JavaScript.
The dream of the Lisp machine (which Emacs to some extent embodies) is dead. Computers of the future will be, largely, JavaScript machines.
- On JS machines, I doubt it, because if some better language it's better than JS, that could yield to huge losses to ARM CPU producers. And that will happen, sooner or later.
>I hope so too, but... I've worked with people who really don't want to touch anything besides JavaScript.
These people are professionaly dead if they want to create something more performant than a bullshit, slow JS application.
QT5 it's being used in LOTS of professional software. Not even WASM Google Earth can be close to the performance of Google Earth Pro used in offices for professional tasks.
Eh, no. Not even close.
Back in the day I wrote a personal patch for BTTV in order to support my video card with different tuner and radio settings. It was damn hard for a C newbie like me. For these people it would be a nightmare.
The point of Rust is to enable newbies to write systems software.
Those are using VS Code already, isn't it?
There is definitely ample room for a hacker-centric text editor with a minimal core where every aspect of UI and behavior can be customizable through TS.
This is not really a criticism of VSCode. If you want to prevent the kind of cross-extension conflicts and unstability that has plagued emacs since forever, a restricted extension runtime is absolutely the best way to go. But there are certainly people who would want to *build* an editor uniquely tailored to their preferences out of low level blocks and deal with the complexity associated therein.
Atom was expected to fill that niche, but it's web based UI is too slow for most large projects.
Ah, another Silicon Valley. Guru. Gook luck with that.
Good luck replacing C, Unix which are everwhere, and not a tiny niche compared to these hipster JS trends. Not even close. EVERY teleco background is tied to C and Unix for standards and data exchanging/defining protocols.
And well, even Scheme being a niche, it's having good stuff such as Guix, Artanis and a JIT.
JavaScript runtimes will written in C++, and will run on OS's written in C.
"Built on" is the wrong phrase to use here, because the actual foundations aren't Rust or JavaScript and will never be.
C++ is slowly but surely displacing C on the OS side too. The field is obviously full of, hmm, time-proven code but new project tends to use C++.
Wake me up when it happens.
I don't think this is a fair comparison. Emacs is a specific piece of software where implementation details are part of the offering. I don't care if parts of the Linux kernel get rewritten in Rust, I have never (extensively) looked at its code. I'll like it as much.
Being able to effortlessly examine and change other people's code is a big thing for me in Emacs. Many big Emacs users make money programming in Lisps too, it's not just some vague preference they're not willing to act on.
To implement a real mode, you WILL have to call elisp code all the time. Emacs does almost nothing asynchronously. Every tiny thing in Emacs triggers a lot of elisp that assumes sequential execution. You cannot do all the hard work in an isolated context and call Emacs once you're done. Most of Emacs performance issues have to do with latency, not long running computations. With libgccjit, native-comp Elisp is consistently 4-5 times faster (yes, really), but the actual perceived improvement is much less. Can hardly tell a difference. Emacs still hangs at all the same stuff and there's no easy fix. You'd have to completely change how Emacs works at which point it may as well be something different. I don't see how this kind of a thing can incrementally take over.
If you want to make my life better find me a way to fix long lines or get rid of GC pauses without breaking everything.
I would like to see Emacs/Guile. Emacs need a more modern elisp (or in this case, Scheme) implementation. It's much easier to make it compatible with the mountain of existing (and useful!) elisp code if you are also in a form of Lisp. That's still not trivial.
It also needs a better front-end renderer, across platforms. I could understand if the "NG" part were to be just about that. But WebRender is an optional feature. Then again, it looks and runs perfectly fine for me on Linux, it's when I'm using on OSX that I see the warts. This is doubtless in no small part due to patches being refused because they are for proprietary platforms.
Async I/O is needed but I am not sure we need to bring Typescript and a full JS environment to get that.
It doesn't make sen to have multiple interpreters jury-rigged up (the horror!) And it also doesn't make sense to throw away all the lisp by having no interpreter for emacs lisp. Guile does support emacs lisp just for this purpose.
> opaque maintenance/contributions, poor performance inherited from core design decisions made decades ago, the burden of backwards compatibility for legacy systems, etc.
The main goal of NeoVim is to ditch the backwards compatibility and simplify the codebase. Sure, it isn't a from-scratch rewrite, but it's a very deep fork.
Not really. I contribute to Neovim and in my experience the architecture is the same as in Vim. The features that were added in Neovim first do have a different architecture and are better designed though. However these new features only make a tiny fraction of all of Vim's features.
(it should also be mentioned the old vimscript support isn't going away, you can still use old configs)
And as a user of emacs, the stability has been quite nice. The speed has mostly only hurt around things I'd expect to hurt. (Long long long lines, for example.)
I can see the allure of language servers. But, to me they are basically a modern cscope. With json being the transport. Not bad, but hardly magical or unforseen. (Granted, getting the pockets of microsoft has been nice.)
What sort of spontaneous breaks are you referencing?
I missed the native-comp note the first read through. Thanks!
Guile-Emacs should.
Even then, there is no other ecosystem that has quite the volume of powerful tools and approaches that HTML/CSS/JS have.
GNU Artanis, forget Hurd, get Guix. Also, on HTML/CSS/JS, they are still 20 years behind the Pascal IDE from Borland or Lazarus, and don't let me talk on NPM and dependencies.
On speed and features, QT5 slaps JS so hard that trying to get something as performant it's a ridiculous claiming.
And I use OpenBSD and CWM, but FFS, JS is a toy compared, to, for example, Lazarus IDE, QTCreator and their compilers/backends running fast as fuck software even on scrap computers.
And Turbo Pascal/Lazarus had RAD developing features since 20 years aso as I said. Something much faster and more native (totally native :p) than any JS toolkit is trying to do ever. Add a control? Drag and drop, edit the bindings. Cross compiling? Lazarus does that even for damn Windows 95 from a Linux machine. And so on.
This could unfold into a project which demand relativ little work, while pushing it to a level where it can compare with VS Code. It could become a good bridge beween modern VS Code and old GNU Emacs that way.
How is that different from what Emacs package managers provide?
In VS Code you can search for ne plugins from inside the application (which emacs also allows), while also showing all information including pictures and animations about the plugin in-app (which emacs does not allow). VS Code also shows which bindings, functions and settings the plugin adds (in emacs only indirect possible, with effort).
And best part: it just works. With emacs there are always a gazzilion problems coming with plugins. Be it some conflict or just additional work neccessary to make it work. Thouhg this is mostly a solvable problem with emacs, it just does not get solved. The Emacs communities solultion for this is the usage of distributions, which come with other problems and have only support for the more popular plugins.
But dynamic scope, and e-lisp, isn't the only thing weird about Emacs. Emacs also calls "files" "buffers", and calls "windows" "frames" and "frames" "windows". It has weird keyboard shortcuts. Every third command also copies to the clipboard as a side-effect, which means you constantly obliterate the contents of the clipboard while getting ready to paste what you thought was there. Oh, but to address this problem, it makes the clipboard into a ring of clipboards, so that when you replace the clipboard with something else, you can still access the other clipboards in a ring by hitting the paste command multiple times in a row. Oh, but if you press another key in between, you get lost in clipboard madness.
Emacs also has an "undo", but no "redo" command. Except that when you "undo", it actually pushes the "undo" itself onto the list of "things to undo", so that you can undo that undo, in order to redo. But people also need to undo multiple undos before getting interrupted with a redo, so Emacs only undoes the undos if you press some other key after undoing a few dos in a row. But hey, this means it doesn't need a "redo" command.
So, I agree that the elisp runtime is a weird thing about Emacs, and something that is rational to change. But the thing is... if you're going to change that... why not everything else about Emacs?
I mean, at some point Emacs looks a lot like Bitcoin Core, or Wikipedia -- a beautiful historical community with weird consensus rules that flowered into a very exotic, strange, and usually-functional artifact. Although sometimes it doesn't work as well as the newer technologies.
Using "buffer" as a name for text being edited is hardly worthy of the title weird, vim does it too. Lots of other text editors as well. "Buffer" != "file", the first exists in the memory of the text editor process, the second exists on the file system. Their contents may differ, and one may exist without the other.
Likewise, calling the subdivisions of the screen "windows" is something which vim does too, and likely other editors also. For a text mode editor, it makes perfect sense. When you then port such an editor to a GUI, you end up with "windows" which are subdivisions of the GUI "window" – vim has that too, not just emacs. (Unlike Emacs, vim doesn't have the "frame" concept, because vim doesn't appear to support multiple GUI windows.)
It doesn't matter. When people talk of editing documents in Word, they speak of opening a file, not a buffer. That's the metaphor people are used to. It may not make sense, but saying "well, ackshually, files refer to the copies on disk while buffers refer to the copies in memory" just confuses people more. Have we learned nothing since Macintosh 1984? Actually, we haven't; we've gone backwards. The spatial Finder was based on very deep psychological notions of how humans relate to the world: through spatial orientation and object manipulation. Meaning that there is only one place on the screen where a given file appears, and it appears at the same place every time unless the user moves it. In short, it works like an actual desktop with real physical pieces of paper. And Apple brushed it aside in favor of NeXT's Unixoid "file manager".
> Likewise, calling the subdivisions of the screen "windows" is something which vim does too, and likely other editors also. For a text mode editor, it makes perfect sense. When you then port such an editor to a GUI, you end up with "windows" which are subdivisions of the GUI "window" – vim has that too, not just emacs.
Again, it doesn't matter. The ordinary user's concept of a "window" maps directly to a GUI window, so if you expect your program to be used in a graphical environment without confusing the shit out of people, you need to adopt that environment's terms and concepts.
Meet. Your. Users. Where. They. Are.
I think you're missing the point. Emacs is an editor that is tied to programmers, where most of its users are programmers. It's not a general purpose text editor.
And while MacOS as an operating system for general purpose usage has done incredibly well due to its intuitive design first approach when it comes to developers many have pointed "your users" have pointed out that its interface and philosophy is restricting.
> It doesn't matter. When people talk of editing documents in Word, they speak of opening a file, not a buffer.
But nothing in emacs says you can't open a file. Just that the contents once it's opened are in a buffer. But if you open files outside of emacs they'll be opened as independent windows. By the same token we shouldn't be able to have a window split in frames, because Word opens each file in a window? The naming is a necessity for disambiguation.
And in fact, the idea of what you have open is a file does introduce problems. How many people you know *ucked up because they didn't save changes? They afterwards learned the hard way what is a buffer that every program uses (even in photoshop, or other software) but didn't learn it's name, just that you should save it every N minutes...
But who are your users?
I started using emacs back in the 1990s, when I was in high school; I actually started using it under OS/2 (as part of EMX), later switched to running it under Linux. I never had trouble with the concept or name "buffers". I just read the manual and the manual explains everything very clearly. (I really loved the GNU Emacs manual, it was a pleasure to read.) In more recent years, I've abandoned emacs completely for vim, but vim calls them buffers too. I am totally at home with a text editor with "buffers", and changing the name isn't going to make anything less confusing for me – if anything, the change is going to confuse me.
Tools like emacs and vim are not targeted at "ordinary" users. If a tool isn't aimed at "ordinary" users, it shouldn't be judged for not meeting their needs, since it was never trying to.
I even wrote a text editor myself (half-finished personal project, I've never released it publicly, don't know if I ever will, it was really just to prove to myself I could do it). And it has "buffers" too, because that is the concept and terminology I am comfortable with.
We're talking about the power tools of our profession. Nobody designs a band-saw to meet their users where they are either. These are apps we spend 8hrs+ per day living in, and frankly the buffer wording is superior because a buffer is not a file, but a file is in a buffer.
Meeting your users where they are is how we got everybody moving from Textmate to Atom and now to VSCode as the new hotness that would finally kill Emacs and vi. And yet that's the mentality that prevented those newer editors from having the legs to really last. Meanwhile other power tools like Eclipse which absolutely meet nobody where they are are going strong too.
I'd like to think that too, but the sad fact is that most developers these days use VSCode. It is the new hotness that killed Emacs and vi. Emacs and vi are both mere footnotes compared to VSCode in terms of market and mind share, and even Eclipse has been largely supplanted within its niche by the far easier IntelliJ suite.
I suppose vi and Emacs will be around as long as they have diehards still using them, but they're not capturing new users' hearts and minds like they used to. Most professional developers I've met hear Emacs and think it's something that last ran on dusty old DEC iron -- and that includes people who worked at DEC in the 80s! The fact that Emacs is still maintained and runs on modern systems came as a genuine surprise to them.
Emacs and vi held on for as long as they did because there was a significant contingent of developers and technical people still familiar with a traditional Unix environment up until the 2000s or so. As some of these people aged out of the work force, they were replaced by an even larger contingent of younger developers, who only ever grew up with Windows and Mac, and have a reasonable expectation that programs work the Windows and Mac way. By and large they don't even have much interest in Linux on the desktop. Linux is for servers/the cloud. Their dev workstation is a Mac. And Emacs and vi are relics from the before times, whose 1970s UIs just aren't worth grappling with because there are better options available. Those are the people Emacs has to reach if it is to stay relevant. Which means that one way or another it will be "modernized" whether we like it or not.
According to StackOverflow's 2019 developer survey [0], 50.7% of editors used VSCode, 25.4% used Vim. So that doesn't really support your claim that vi is a "mere footnote". 25% of market share isn't 50%, but it is still decent and more than a mere "footnote".
[0] https://insights.stackoverflow.com/survey/2019#development-e...
People overestimate the value of massive community and apply a winner takes all mentality too often. You don't need a giant community, and almost by definition giant communities tend to have a lower average user quality leading to lower quality of libraries and support (looking at you JS ecosystem).
They think the world revolutes around them with JS and hip editors while in the end they are the utter minority in serious IT/dev environments.
Bullshit, a lot of Linux/Unix developers used Emacs in the 90's and early OO's.
I was there.
I mean - if you're going to ignore everything that makes Emacs great on your way to cavil about how it's different, why not go all out, right?
Try `C-w`, `C-k` and `C-y` inside a Bash shell sometime, you might be surprised.
`M-w` is an exception, though.
Blender is, I think, another example of this... It's quite good, but relative to the other 3D editing tools available, it has an absolute space-alien UI. Quite internally consistent; a bear to get into if you learned 3DS Max or Maya first.
When did you last start Blender? This used to be true. Nowadays it's at least on par with commercial tools if not better.
* Moving an object in Blender: Select, press 'G', now the mouse floats the object without the need to hold a mouse button down, optionally hit 'x', 'y', or 'z' to constrain to those axes. You can also take advantage of the 3D cursor.
* Moving an object in 3DS Max: Select to bring up manipulation widgets, click and drag the image of the widgets to move the object.
Starting to learn Blender if you learned 3D scene creation in Max is like starting to use emacs if you learned word processing in Microsoft Word. The metaphors aren't categorically better or worse; they're very different.
> Emacs also calls "files" "buffers", and calls "windows" "frames" and "frames" "windows"
Well, as someone pointed out, buffers are buffers, files are files, why would your editor try to fool you regarding the fact that what you have in front of you is in a buffer (that might be persisted into a file) and not exactly a file? Every buffer tells you if it's pointing to some file...
Frames aren't windows. Like tabs in a browser aren't browsers or windows either?
> It has weird keyboard shortcuts.
Yeah, the worse part of it all is you can't even change them, right?
> Oh, but if you press another key in between, you get lost in clipboard madness.
Yeap sometimes that happens, on the other hand the times it doesn't and you can cycle through things in ways that are not easy to do in other editors is great.
> Emacs also has an "undo", but no "redo" command
This ties back to the ring design? You can switch the direction you're going with two key-strokes. Granted, if you change the buffer (outside of just navigating the ring or copying into buffer, save, etc) then it does get weird, but it's usually pretty consistent and useful. It's like having a step by step git to your inputs into the buffer.
But `this`, this, this: > I agree that using a JS runtime makes more sense that Emacs's weird e-lisp runtime, with all of its dynamic-scope weirdness.
If the JS runtime, with the million of hours and $ poured into it to make it performant at the runtime engine, libraries and general language improvement, when compared to others created decades ago (and mostly voluntarily kept) didn't have an edge in some things, it would probably be the biggest facepalm in programming history ever.
Having said all this, I'm happy people keep experimenting and building stuff for others to use, even if in JS, or Rust, or Zig, or brainfuck.
They don't have default bindings (unfortunate), but try installing a recent build of Emacs 28 and binding these commands to the keys of your choice:
- undo-only
- undo-redo
Okay.
$ cat fib.scm
(define (fibonacci n)
(if (<= n 1)
n
(+ (fibonacci (- n 1)) (fibonacci (- n 2)))))
(fibonacci 40)
$ time guile fib.scm
guile fib.scm 8.52s user 0.01s system 99% cpu 8.559 total
$ time node fib.js
node fib.js 1.29s user 0.01s system 100% cpu 1.296 total
$ time deno run fib.js
deno run fib.js 1.41s user 0.02s system 99% cpu 1.423 total
edit: for reference, the equivalent C code runs in 0.27s on my laptop.
edit edit: okay, okay, compiling the Guile code before running it speeds it up to an exciting 8.16s execution time, still about 6x slower than deno.
guile --version?
Guile always compiles code before running it, but adding a JIT it's only from v > 2.91.x, and by default at v3.
$ time guile3 fib.scm
guile3 fib.scm 2.01s user 0.02s system 99% cpu 2.047 total
Guile3 will improve that, too.
Chicken compiles to C
$ csc fib.scm
$ time ./fib
real 0m8.461s
user 0m8.383s
sys 0m0.061s
$ cat fib.rkt
#lang racket
(define (fibonacci n)
(if (<= n 1)
n
(+ (fibonacci (- n 1)) (fibonacci (- n 2)))))
(fibonacci 40)
$ time racket fib.rkt
102334155
real 0m1.396s
user 0m1.142s
sys 0m0.127s
$ sysctl -a | grep .brand_string
machdep.cpu.brand_string: Intel(R) Core(TM) i9-8950HK CPU @ 2.90GHzFor lulz, I clocked a dumb loop version.
(defun fib (n)
(loop repeat n
for a = 1 then b
for b = 1 then c
for c = (+ a b)
finally (return a)))
Edit: I confess the speed of the loop got me curious and.... javascript has numeric overflow. That would be a hell of a surprise if you weren't expecting it.
Edit2: (I realize overflow isn't the right term, but it is amusing that I can calc the 10000th fib in elisp... not so much in javascript. And yes, even elisp falls over sooner than sbcl does.)
Granted, this is also true of fixednum in common lisp. At least "biting you" with that should only be performance. (I can't think of many other ways for that one to get you.)
Edit: I should further ack that "getting around" it is as simple as adding "n" to one of the constants. So, it is nicely done.
I wonder if there’s some room for improvement since using GMP is pretty standard for most languages.
Edit: forgot to link to a BigInt comparison in major language implementations http://www.wilfred.me.uk/blog/2014/10/20/the-fastest-bigint-...
OFC I would rewrite that as a iterative form (and any Schemer here could :D), but here we are testing performance :D.
I suppose there is a bit of something in there about using the right algorithm. But I can understand picking one that will take time for comparing speed. I'm not sure either is really representative of work you expect an editor to do, though. Curse of benchmarks.
Intuition: if you were to expand and flatten the recursive calls, you'd end up with something like
(+ 1 0 1 1 0 1 1 0 ... 1 0)
with exactly (fib n) 1s (base case (fib 1)), and no more than (fib n) 0s (base case (fib 0)).
Simple example:
(fib 3)
= (+ (fib 2) (fib 1))
= (+ (+ (fib 1) (fib 0)) 1)
= (+ (+ 1 0) 1)
= (+ 1 0 1)That is, knowing all of these facts, it is still impressive how fast you can get (fib 1000000) using iteration and bignums.
Which is still not fast -- on my laptop, Python takes about 9s for fib(1000000). But compared to an exponentially-growing function, that's a blink of an eye.
With scm, using (trace proc):
scm -f fib.scm
call fib 9
call fib 8
call fib 7
call fib 6
call fib 5
retn fib 5
call fib 4
retn fib 3
retn fib 8
call fib 5
call fib 4
retn fib 3
call fib 3
retn fib 2
retn fib 5
retn fib 13
call fib 6
call fib 5
call fib 4
retn fib 3
call fib 3
retn fib 2
retn fib 5
call fib 4
call fib 3
retn fib 2
call fib 2
retn fib 1
retn fib 3
retn fib 8
retn fib 21
call fib 7
call fib 6
call fib 5
call fib 4
retn fib 3
call fib 3
retn fib 2
retn fib 5
call fib 4
call fib 3
retn fib 2
call fib 2
retn fib 1
retn fib 3
retn fib 8
call fib 5
call fib 4
call fib 3
retn fib 2
call fib 2
retn fib 1
retn fib 3
call fib 3
call fib 2
retn fib 1
call fib 1
retn fib 1
retn fib 2
retn fib 5
retn fib 13
retn fib 34
call fib 8
call fib 7
call fib 6
call fib 5
call fib 4
retn fib 3
call fib 3
retn fib 2
retn fib 5
call fib 4
call fib 3
retn fib 2
call fib 2
retn fib 1
retn fib 3
retn fib 8
call fib 5
call fib 4
call fib 3
retn fib 2
call fib 2
retn fib 1
retn fib 3
call fib 3
call fib 2
retn fib 1
call fib 1
retn fib 1
retn fib 2
retn fib 5
retn fib 13
call fib 6
call fib 5
call fib 4
call fib 3
retn fib 2
call fib 2
retn fib 1
retn fib 3
call fib 3
call fib 2
retn fib 1
call fib 1
retn fib 1
retn fib 2
retn fib 5
call fib 4
call fib 3
call fib 2
retn fib 1
call fib 1
retn fib 1
retn fib 2
call fib 2
call fib 1
retn fib 1
call fib 0
retn fib 0
retn fib 1
retn fib 3
retn fib 8
retn fib 21
55
scm -f fib-iter.scm
call fib-iter 1 0 10
call fib-iter 1 1 9
call fib-iter 2 1 8
call fib-iter 3 2 7
call fib-iter 5 3 6
retn fib-iter 55
retn fib-iter 55
retn fib-iter 55
retn fib-iter 55
retn fib-iter 55
55 $ csc -optimize-level 5 fib.scm
$ time ./fib
real 0m5.506s
user 0m5.489s
sys 0m0.010s
$ csc -optimize-level 5 -stack-size 4m fib.scm
$ time ./fib
real 0m5.295s
user 0m5.277s
sys 0m0.011s
I stopped writing Chicken scheme code a few years ago; it was fun due to the ease of interoperating with C, but I found Chibi a better fit for that niche eventually.
> (time (fibonacci 40)) (time (fibonacci 40)) no collections 1.364130664s elapsed cpu time 1.364456065s elapsed real time 0 bytes allocated 102334155
Which Emacs version is this?
Any chance you want to try the native-comp branch and post its numbers here as well?
I can install native-comp via AUR. It'll take some time.
time emacs --script fib-0b03fe9a-6b386e71.eln
emacs --script fib-0b03fe9a-6b386e71.eln 22.47s user 0.09s system 99% cpu 22.664 total
(I copied the .eln from the cache to confirm that's what I'm running against; I got a similar speedup from running against fib.elc.)
That should be considerably relevant in this benchmark.
PS you can also run a wider set of benchmarks to compare against stock Emacs using:
https://elpa.gnu.org/packages/elisp-benchmarks.html
Here some not very updated results:
IIUC, you're saying the fib benchmark gets optimized out at speed 3 (and thus run faster than all the implementations discussed here).
But what about speed 2, which is the current default? If we're 15 times slower on this benchmark, does that mean that Node has much cheaper function calls, or something like that?
Because we have to keep Emacs Lisp functions advice-able (which I agree is a good thing)?
Specifically a single fibonacci example is not a good performance indicator by any means.
For instance let's assume V8 is faster in average at running this kind of nanobenchmarks (probably at least for now it is), but how much does it cost to convert non trivial data structures from Lisp to JS and back in a real case?
I expect there will be a lot of back and forward if the system if JS and Lisp get mixed and have to cooperate.
How much does it cost to go through foreign functions calls?
This are just initial thoughts that are not accounted at all here.
And even benchmarking nano-benchmarks can be surprisingly tricky ;) :) https://github.com/emacs-ng/emacs-ng/issues/187#issuecomment...
My questions here are about whether these results indicate further optimization potential for native-comp (in the default configuration, hopefully).
emacs-ng is an interesting experiment, but I think I will only be able to seriously consider it if it comes to reimplementing the Elisp VM on top of V8. Then it will be a single runtime, and little to no data structure conversion and FFI will be required.
In the meantime, I'll keep following the native-comp progress ;), as well as dreaming of Web Workers in stock Elisp API.
That would cover one major point of complaint with emacs performance, the other would be getting some form of real threading and parallelism into emacs so that long-running elisp doesn't block the UI thread, but that is a pretty huge task. With that said, I am sure that the community realizes this and is slowly working their way towards this.
Emacs performs better than many other applications at important objective measures, eg the time between pressing a key and the character showing up on screen. You might expect vim to be good at this but many terminal emulators are optimised for throughput rather than latency and are slower than gui emacs.
Certainly legacy is a problem but I worry that replacing eg font-lock with something with better performance properties (does this actually do that?) means throwing out the baby with the bath water as so many modes depend on these things.
Typically the reason that emacs becomes slow is due to an overload of features or bad asymptotics. Examples:
- global-auto-revert-mode plus lots of buffers plus some in a slow file system like nfs or maybe lots of open fired buffers
- some mode using a configured alist that works fine for small lists but really sucks for big ones (ie perf is linear in size of config, but often this comes hand in hand with trying to process big inputs so it can be worse.) I think this happened with spacemacs and which-key-mode for example
- flyspell + flycheck + build running in background + some autocompletion server. Even a powerful computer can be slowed down by this. Some of these are likely to slowly improve over time as they become more synchronous. Others are problems you’re likely to have with any editor (a nice emacs thing was that it was easy for me to add some advice to pin my build to not use all my cpu. It can be really hard to poke around in other plugin systems to do that sort of thing without a config option for it)
- just piling on the nodes with something like spacemacs and not paying attention to the performance of the whole. I think doom-emacs shows a lot of promise here.
I do think there are advantages that could be made by modernising emacs’ core and improving data structures (whatever happened to remacs?)
I feel like the fundamental design of emacs is good however and I like:
1. keymaps and the command loop
2. Buffer-local variables, advice, dynamic scoping
3. Documentation
4. Fundamentally text-based interface
Obviously you should disregard all of this though as I am very biased due to my emotional reaction to getting rid of emacs lisp which, frankly, you can pry from my cold dead hands.
————————————
Some other thoughts I had on emacs lisp as a good language for a text editor:
1 https://news.ycombinator.com/item?id=19343908
2 https://news.ycombinator.com/item?id=22881597
3 https://news.ycombinator.com/item?id=18605001
On the “fast” measurement that actually matters: https://news.ycombinator.com/item?id=23432292
In my experience with modern hardware - a gaming PC - Emacs is noticeably sluggish compared to most other editors in a few specific ways: particularly syntax highlighting. When is the last time you used a different editor?
> whatever happened to remacs?
It's still there: https://github.com/remacs/remacs/wiki/Progress
As far as I can tell, its goal isn't to change anything about the core design though; just to re-implement the core in Rust.
It’s not enough to stop me using emacs, but it is enough to make me go out of my way to avoid it (and to grumble about).
Systemd-ng, anyone?
My beefy desktop computer is often already struggling just to run Chrome.
Emacs used to stand for "Eight Megabytes and constly swapping", maybe they could name it now Egacs (Eight Gigabytes ...)
Ditto with Groff+Mom occuping a few MB (totally doable in 1997) vs a GB LaTex install lasting several minutes to render a PostScript, and then running X (if even) just to display the resulting file and nothing more.
Nowadays even QT5 software looks lightweight enough compared to some Electron monsters...