On livecoding annotations and visualisations

I’ve just finished reading Charlie Roberts’ interactive web essay on annotations and visualisations for live code. If you haven’t read it, go do that now—and make sure you have some speakers/headphones to listen to the code examples, because it would be a real shame to read it without watching/listening to his ideas in action. All the code samples (with the visual annotations) are live-editable, both running “whole blocks” and re-evaluating individual expressions with ctrl+enter.

The essay lays out some guiding principles:

There are three principles that guide the design of the presented annotations and visualizations.

  1. If a value changes over time (typically related to musical progression or signal processing), display its current state, either by adding a visualization / annotation or by modifying the source code to reflect the current value.
  2. Make annotations and visualizations as proximal as possible to the code fragment responsible for generating the data they are representing.
  3. In addition to displaying when values change, whenever possible also provide some indication of when a value is being read, in particular when there is a direct affect on musical / sonic output.

It then goes on to show some example visualisation in gibberwocky, both in an “all in” demo and then in bite-sized listings which show off the different specific ideas. In general, I really like the ideas, and it’s something Andrew Sorensen and I have written about before in our paper Visual Code Annotations for Cyberphysical Programming (2013) in the 1st International Workshop on Live Programming (LIVE)1 (it’s reference #4 in the essay). I’ll refer to this a bit in this post, so let’s call it “the LIVE paper”.

I also think that the three points listed above are pretty solid, especially in a multimedia livecoding context (maybe even in a broader context). One thing I like about the visual annotations provided is that they’re mostly ASCII (or ASCII-ish). This is not so important when deploying them in the web browser (since you can do so much fancy styling stuff with CSS & js these days) but it’s really important when dealing with… ahem, more venerable editors. I ended up having to use some unholy Emacs hacks with overlays to get the original annotations discussed in the LIVE paper working.

I think that displaying hidden state in comments is a good compromise, and avoids the need for fancy “extra-textual” overlays. Not that overlays aren’t sometimes useful, but there’s a lot you can show with inline text-decoration hacks and adding a few comments to provide ascii text to decorate (when it isn’t explicitly represented in the first place).

Mixing annotations and live edits

But, (you can probably tell that there was going to be a “but” somewhere) many of the cool annotations displayed don’t work while the code is being edited2. In some cases they actually break the code (try editing one of the 'x*ox*xo--x*x*o' patterns while the code’s running—you’ll end up with new stuff in your pattern that you didn’t put there).

The problem isn’t so obvious when all the code listings are fully-formed on page load, and you can just play them as is. But when you try and mess with the code then you’ll see what I mean (again, I really suggest that you try it—it’s super-cool being able to mess around with the live code in the browser).

I feel this particularly keenly because I’m a clean-slate livecoder (as is Charlie), so I’m always moving through an incomplete code state until I get something which will even run. This isn’t just a problem for clean-slate livecoding, though—even tweaks to existing code which introduce “bad” code states (from the visualisation’s perspective) will cause these issues.

We talked about this in the LIVE paper—the fact that there’s a distinction between the “state of the world” vs “state of the code”. This is a fundamental challenge for the sorts of inline code visualisation/annotations shown in the essay, because it’s using the code as the “raw material” for displaying information about the world (beat/timing and other “hidden variables”, audio engine state and output, etc.).

There are a few different ways to tackle this problem:

  1. mark some annotations as “safe” for code being edited, and some “unsafe”
  2. when code is being edited, turn off all annotations (or at least for that expression)
  3. some sort of “grand unified theory” of the delta between the current code state and the current execution state, and reconcile these to provide a maximal set of acceptable visualisations (this approach includes a coq program to formally verify that you’re doing everything right)

I think the pragmatic choice is #2, as long as the “disabled edits” section is kept as small as possible. It’s still going to be a pain, though, because when I’m livecoding I’m tweaking stuff all the time, so it’s likely that a lot of the code will spend a lot of the time with the visualisations disabled.

I’ll leave #3 up to people smarter than me 😁 (although perhaps there are heuristics which could do a decent job).

Finally, I can identify with Charlie when he writes:

But, in the end, the feedback provided by these annotations and visualizations have become a critical part of my live coding practice.

I get this—in my experience (when I’ve had even more limited annotations than the ones he shows in the essay) are just as much for my benefit as the audience. Livecoding is hard, and any extra information you can get about what’s going on with your code is super helpful.

In principle visual code annotations can be even more useful to the livecoder because they allow her to “audition” algorithmic changes to the code without3 actually eval-ing the code and changing the music. Here’s an example from the essay:

Euclid(9,16)/* 1010110101011010 */ 

I might not always know what the 9,6 euclidean rhythm is, but I’m by now fairly used to looking at 10100010 sequences and “hearing” the rhythm in my head. I could poke around with the parameters in a live set, exploring the parameter space (and thinking through the effect it’ll have on the music) and then only evaluating the code when I’m satisfied. That’s super powerful—the equivalent of the DJ cueing the next track with one can on their ear, and one ear in the club—something which I don’t have currently in my livecoding setup (although others might).

Conclusion

You can probably tell that I think there’s a productive research agenda here—and I hope Charlie continues with it. I hope to help out myself, too. I guess my main point is just to shout from the rooftops:

any code visualisation/annotation techniques must be robust for code which is currently being edited

I’m not just talking about technical issues, either; obviously any demo/prototype is going to have those, but they’re fixable. I think there are deeper issues with trying to use live text as both the description of program behaviour and as a “view” on the hidden state of the program.

Anyway, this is just a blog post, so I’m off the hook with regard to rigour, accountability and just general good scholarship, right 😜

Addendum: thoughts on web publishing

I love that this essay/paper is published online—the interactive examples are crucial to getting the point across. I know that some conferences & journals these days allow html submission (nicer for reading on mobile, anyway) and other multimedia artefacts (audio/video recordings) but it’s still hard to get traction for this sort of rich, interactive in-browser work. The fact that at the end Charlie has to say:

If you’re going to cite this website in an academic paper, please consider also citing either reference #1 or reference #7 given above; citations of such papers count more in academia than citations of a website. Plus, there’s further information in them not covered in this essay. Thank you!

Oh well. Mad props to Charlie for putting this out there for comment and discussion, and hopefully there are a new generation of publications4 where this stuff can be front-and-centre, not just a weird “supplemental web materials” section to a traditional pdf.

  1. if you don’t have institutional access to the IEEE then just email me and I can send you a pre-print. 

  2. Charlie, if I’m doing it wrong, please let me know :) 

  3. Yes, I know that the real-time feedback of hearing the sound is crucial, and I’m not for a second saying that we do away with it, but there are some situations where I want to check what the result of an algorithmic/parameter change might be without inflicting it on the audience. 

  4. Distill is great, but it’s pretty DL/AI-focussed. The livecoding community needs something similar (although it does privilege livecoding environments which work in the brower, so that’s not ideal either. Hmm.)