Google Summer of Code 2013: Emacsy

Emacsy was accepted as a 2013 Google Summer of Code Project, which is great news! Emacsy is an embeddable Emacs-like library for GNU Guile Scheme. It’s an attempt to extract “the Emacs way” of using, augmenting, and extending an application while it is running. Emacsy harbors no ambition to become a text editor because there is already a great Emacsy text editor: Emacs. I’d like to use this post to address some features and implementation details that I’m personally excited about: how to do online help, going beyond Emacs to do job control, and some comments on whether Emacsy’s “shell” ought to be special.

The GSoC proposal has many details about the project, but perhaps I should say who this blog entry is written for, as I see a handful distinct audiences: application users, integrators, and contributors.

  • Application users are the users of some application X that Emacsy has been integrated into it. As Emacsy hasn’t been released, there is no real application users to speak of yet, but they’re important to keep in mind.

  • Integrators are developers who embed Emacsy into their own application. They need to know enough to integrate with it but don’t necessarily care about how Emacsy does its thing.

  • Contributors are developers who are interested in the architecture and inner-workings of Emacsy and can contribute to its development.

These Emacsy blog posts will be of most interest to would-be contributors and integrators, and I will assume some familiarity with Emacs. There’s a tension between features that might be nice to have but would incur a large burden on integrators. It’ll be helpful to keep that tension in mind when deciding what features to implement and how.

Help will always be given in Emacsy to those who ask for it

Emacs has the most comprehensive online help system of any application; there is really nothing else like it. This is the reason for its claim to be self-documenting. Unfortunately because there isn’t anything else like it, it seems a bit alien at first. If you don’t know what the key sequence M-C-\ does, then hit C-h k M-C-\ and Emacs will tell you that it’s bound to the procedure indent-region. (If this notation C-h for control-h is unfamiliar, please see Sacha Chua’s excellent one pager for beginners.) If you don’t know what the procedure indent-region does, then hit C-h f indent-region and Emacs will describe it and even provide a link to the source code. One question I’ve been struggling with is, how can Emacsy provide online help like Emacs does? Several ideas come to mind:

  • Text buffer, duh. One idea is to duplicate the way Emacs does it by showing a read only text buffer. In Emacs all things are text buffers, but Emacsy does not make this presumption. Also if one must display a text buffer that means every integrator must figure out how to display this text. What if the view is larger than the window? Scrollbar, please. What if there are links in the text? Should they be clickable? Of course! Integrator, please implement these text features that may have nothing to do with your application. This seems like too high a burden for integrators, even if it would be cool.

  • Minibuffer to the max. Just use the minibuffer for everything. We can get away with using the minibuffer for a lot of things but that would not allow us to have a comprehensive online help system that was comparable to Emacs’.

  • Use the Web, Luke. This idea struck me recently. What if when we wanted to show a lot of information, say about some particular function C-h f some-fun, we opened a link to some internal web server? Emacsy serves up the information over HTTP and the browser gives us a lot of features like links, styling, rich media, and perhaps some customization possibilities. The implementor only needs to send URL requests to some handler application. Fancier implementors could handle these URLs internally in a web view.

I’m pretty taken with this last idea. It retains the ability to let the application be fully self-documenting without imposing a heavy burden on the implementor.

Going Beyond Emacs: Job Control

One feature I’m particularly excited about is job control. How many times have you accidentally stumbled upon some network operation that causes Emacs to freeze? Granted it’ll come back if you wait, or you can quit C-g it. But wouldn’t it be nice to just suspend that command like you would in the shell? Just hit C-z to suspend it. Then maybe type M-x background to continue running it in the background. That’s a feature that we can implement in Emacsy, but how? Allowing for job control brings up the issue of multitasking.

Multitask like it’s 1999

How shall Emacsy support running multiple jobs some of which are in the background? I’m going to suggest a cooperative multitasking approach rather than a pre-emptive multitasking approach. Heresy, I know, and I know it’s not the ’90s, but hear me out.

I like to think of Emacsy as inheriting from both Emacs and Unix. Unix and Emacs are somewhat odd bedfellows though. Unix is about small tools that do one thing well and compose easily; the shared state is the filesystem with little context, and it’s generally not focused on interactive apps. Emacs on the other hand is about having all tools, big and small, at one’s finger tips (key bindings); the shared state is Emacs’ process memory with lots of context, and it’s focused on interactive use.

For an operating system like Unix pre-emptive multitasking is absolutely preferred. Note though that the memory of each process is isolated from one another, so you don’t have to worry about one process overwriting another’s memory. However, an Emacsy command will not have its memory isolated from other commands, so it seems inadvisable to use a pre-emptive multitasking mechanism like threads since the commands could then overwrite shared memory unless very carefully synchronized. Instead I suggest using coroutines to implement a cooperative multitasking system. Only one piece of Emacsy code runs at any one time; thus a whole class of race conditions evaporate. This doesn’t preclude one from using threads within Emacsy since threads are natively supported by GNU Guile, but they will have to work out the synchronization details themselves.

Why not vimy?

There are two text editors of note: Emacs and vim, so why Emacsy and not vimy? Partly because I couldn’t do a vimy project justice because I’m not a vim user, but I think there is a larger reason than that. I think vim is an excellent text editor. Some of my friends are absolute wizards with it. I’ve had glimers of understanding the power and unity of its terse sub-language. I am envious of the dot . command. However, making the best text editor, sub-language and all, does not necessarily mean it would transfer well to other domains. Is an insert mode necessary in a CAD program? What is the terse sub-language of manipulating chemical models? I think making something worthy of the name vimy would be very challenging in another domain.

That said, I don’t see any reason why someone couldn’t write a vim-like UI using Emacsy as a starting place. Borrowing more from the Unix tradition, consider the shell. The creators of Unix didn’t make the shell special. It is just another program unlike DOS’ shell, but that required a lot of ingenuity to invent the right six system calls to make that happen (Torvalds, “Just for Fun” 54). In my proposal, I emphasized the Key-Lookup-Execute-Command-Loop (KLECL), but in a way that is essentially Emacs’ shell. One could write a different shell, a vim-like shell if they preferred. That is, if Emacsy takes care to not make the shell special.

That’s it for now

Those are some of my current musings for Emacsy. I’m going to try to record further ideas on Emacsy’s github wiki. Feel free to contribute your own ideas there or write a comment below. Otherwise, I’ll be on twitter or hanging out in #guile on freenode.net.

Published: June 05 2013

blog comments powered by Disqus