I’m way behind on posting here. Here is a quick rundown of what I’ve been up to.

The biggest thing is that I managed to run Kythe! It took me half a week to get it to work with a hello world program, but once I had that I was good to go. This meant I got to play with Bazel again, which was nicely familiar.

A lot of the confusion was that remote Bazel repositories don’t quite work for C++ in the current state. My approach of referring to the Kythe indexer from a Bazel repository with my hello world program just wasn’t going to work, even though I kept trying. In the end, I just copied the binaries into my tree and set up the configuration needed to run them.

I gave a demo of the web UI that comes with Kythe, run against some data structure code that had Alice written. There’s a somewhat hilarious blowup in size: 7 kB of source results in a couple of serving database tables. I also did a hackish port of the SQLite build over to Bazel so I could run Kythe over it. The 1.5 MB of source resulted in a ~80 MB database. The flipside is that the API server responds to most queries in a few milliseconds.

Keeping with Kythe, I’m currently working on an alternative web UI. The provided one is really good to understand how nodes are linked together in the graph structure that Kythe builds up, but it’s too noisy and clunky to use as a day-to-day code exploration tool. As an example, it shows the syntactic relationship of a variable reference being a child of the function it occurs in.

I’m pretty excited about this project because one of the things I wanted to do at RC was to learn JavaScript (well, ECMAScript 2015). The Kythe web UI is in Clojurescript using Om, which I’m not currently interested in learning, so I’ll be starting my own from scratch.

Just to keep things slightly weird, I’m using RxJS, the Reactive Extensions for JavaScript. I’m trying to emulate the architecture used in Elm in place of the Flux architecture that’s more commonly used with React. The Elm architecture is pleasantly functional, consisting of

  • a model type Model;
  • an action type Action that specifies all the kinds of actions that exist;
  • an update function update :: Action -> Model -> Model; and
  • a view function view :: Model -> Html that renders the model.

These are all wired up with functional reactive stuff—Signals in Elm, Observables in RxJS. The user input events are translated into a stream of actions. A stream of models is created by folding the update function over the action stream. Finally, the view is kept up to date by mapping the view function over the model stream.

I have no real idea what I’m doing, so I’m starting off with translating the Elm TodoMVC implementation to this cobbled together set of tools. This should be done fairly soon, then I’ll go back to the Kythe code browser.