And if you try to solve it using workers you introduce an annoying programming paradigm of async/await and IPC.
Copying data over IPC can be slow, and having to set up IPC to share the data you need to run in a worker is also annoying.
The great thing about the single threaded shared memory model of elisp is that it's a lot easier to code against. And yeah sometimes someone blocks the UI doing synchronous I/O, but that comes at the benefit of making it really easy for people to write addons that do all sorts of things.
A lot of the I/O blocking UI problems could be alleviated by just moving to async I/O primitives without incurring the runtime and developer experience problems of a multi process architecture
I don't want to dismiss the argument; maybe it's just that Emacs packages are bad at avoiding that behavior? But I'm not sure that async I/O primitives are enough for a text editor, even though I would generally agree with you.
On the subject of that last sentence:
> A lot of the I/O blocking UI problems could be alleviated by just moving to async I/O primitives without incurring the runtime and developer experience problems of a multi process architecture
This sounds like an advertisement for Javascript to me? This is one of the reasons I think why NodeJS became so popular when it was first introduced, literally because it was single-threaded with good async I/O primitives (or good enough, this was before Promises were in the language), and that led to huge performance improvements over other server implementations that prioritized workers.
So while I still don't like the kitchen sink approach, I'm still left feeling like a lot of the problems and solutions being proposed here are kind of things that would be addressed by using Deno in Emacs? At least, I'm seeing the appeal. Single-threaded memory pool to help make developer experience better and to improve performance, coupled with decent async I/O primitives -- that's Javascript, that's one of the things the language is sort of good at. You have workers for larger tasks, and that's definitely handy, but most of your programming is happening in a single-threaded event loop using promises.
If I looked at this sentence in isolation, I would think that you were arguing in favor of adding Deno.
That is a pretty strong argument which I agree with. The somewhat dismissive answer I could give around JS in Emacs is that if I wanted to write JS so badly that I was willing to rewrite the ecosystem in order to do it, I'm not sure why I shouldn't just use VS Code.
I'm floating the idea that JS would be a good fit for a lot of these problems, but it is correct to ground that with the idea that having 2 separate coding languages in a text editor just feels weird and cumbersome and I don't know, isn't the ecosystem fragmented enough? I can't actually avoid Elisp in Emacs and I wouldn't be able to just because Javascript was added.
As far as I can tell Emacs already supports the primitives required to implement an async/await style interface. The way it works now, is that the editor is one big event loop. The event loop handles reading keyboard input and eval'ing the elisp functions bound to those keys, and while that's not happening, non-blocking things like subprocesses are being progressed. This happens outside of elisp, inside of the C part of Emacs somewhere.
Right now subprocesses support using custom "sentinels", these are an elisp function that gets called by the event loop whenever the subprocess progresses in some way.
I'm not really a concurrency expert but I think you could work these already existing things into a nice interface. Having built in support for this would be a huge upgrade!
There are libraries like emacs-aio that try to do this, but I don't know very much about them right now, but that should at the least serve as a proof of concept.
async/await APIs are just a fancy abstraction over an event loop and callbacks and whatnot.
Even just having "promises" would be a nice upgrade.
Maybe it's just me but I find the current functions for doing non-blocking stuff very clunky and awkward.
So, yes, you can block the UI thread. There have been ways to not do that basically forever. I welcome better ways to manage this, but there will always be some way to do code on the UI thread, such that it will always be possible to somewhat block things.
The alternative is to make all code preemptive, it seems. And that just begs for race conditions on the shared mutable state that is the editor.
The problem is I had to use Atom for a while seriously before I understood how often it failed for me. If you're just trying out a text editor to see if it's comfortable, you don't see these failure modes as clearly. So I've tried VSCode before, but I don't know if it's better or worse at avoiding hanging the UI, because I don't use it constantly all day long with a ton of packages installed.
If I had to make a list of all the times I get frustrated at Emacs, a long operation that has nothing to do with the UI blocking everything is pretty high on the list, that is a regular complaint that I have. I feel comfortable saying that I wish Emacs packages would get better at this, and maybe there are ways to improve it within Emacs.
But I guess truthfully for all I know VS Code could have the same problem but even worse.
I agree that making code preemptive is probably not a good idea.
Sure, it’s possible to write code that doesn’t block the UI, but it’s very difficult in Emacs and just about every built-in functionality is an offender. This includes network operations, which can block forever. You’d be laughed out of the room if you implemented a text editor today that can lock up completely and indefinitely with M-x package-list-packages.
You are right that it is difficult in many respects. But it is far from impossibly difficult. And most of the tricks you could do to make it easier typically lead to some silly hard edge cases in the code.
Me too.
I routinely run emacs on a Raspberry Pi IV building large rust projects
Blocking has never been an issue
I bury the compilation buffer and carry on as usual
The main problem I have is too many options and possibilities
I think a better thing to do is organise all the third party packages with decent documentation
The documentation in the package ecosystem ranges from barely adequate to woeful with the emphasis on the latter
I genuinely don't think it is - Javascript is just a language. Why not implement Async I/O and WebWorkers APIs in Elisp? Why complicate things with an entirely new language?
And with Elisp native compiled on most platforms, it seems doubtful WebAsm is going to make a noticeable difference in performance.
Yes, it will change its name from Emacs (8 megabits and constantly swapping) to Etacs (8 terabits and constantly swapping). /s
Backward compatibility is the elephant in the room about some of this stuff that most Emacs users absolutely take for granted. I personally have about 20 years worth of Emacs Lisp config files; and others have more decades worth. They largely have continued working for all those years.
I hope that whoever is working on this fancy Deno/Javascript stuff realizes this, and actually talks to some Emacs users if they want any kind of actual adoption.
When I look at the README, I'm seeing one mention of "Warning: This feature is currently unmaintained", and another of "Note: This feature is currently outdated and been disabled. However there is this to bring it back." This...does not inspire confidence in the minds of long-term Emacs users.
Now, maybe I'm being curmudgeonly and unfair. Like the README says, I do want Emacs to be a an editor for the next 40+ years. I just doubt it's going to get there without Emacs Lisp.
Deno and WebRender have good integration with workers, webrender offers the option of providing better GUI integration (although it's not clear exactly how it will be leveraged), jumping to Javascript gets rid of dynamic scoping, and (in my experience, although mileage may very much vary) tends to handle IO-blocking better.
The project's performance justifications it lists are:
> Async I/O from Deno, WebWorkers, and WebAsm, gives you the tools to make Emacs a smoother and faster experience without having to install additional tools to launch as background processes or worry about shared library versions.
I don't disagree necessarily, in some ways adding another language instead of fixing the one that's still going to be the dominant way most integration is handled is a very, for lack of a better word, temporary fix. If Lisp isn't good enough for an editor, then we'd all jump ship to VSCode or whatever. There is something hacky about saying, "we're not leveraging this language well, let's just add another."
But even saying that, I do still (at least at first glance and without looking into it harder) see the appeal.
Even dropping the JS entirely, WASM workers seem like a non-terrible way to get better worker support and to get rid of the IO-blocking-UI problem (which I agree is a problem, and might be one of my biggest criticisms of Emacs). Am I missing something?