Hubot, a Physical Manifestation


Hubot, a Physical Manifestation

Recently all Resineers came together in a single place for our annual summit, and there were several jokes about whether Hubot, our virtual PA, would be joining us. I decided to spend a few quick minutes piping what Hubot says on our chat service (flowdock) to an LED grid. At this point a wiser head than mine would point out certain truths about “a few quick minutes” but this, dear reader, is where my learning journey began.

tl,dr; or “the crux of the problem”

What I used

  • Scroll Bot Kit - a quick note here. The project uses the pins as the only mechanical connection holding the RPi in place. I’m not sure if there’s a better way, but it is certainly on my ponder list.
  • A Flowdock instance - other chat services do exist, happen to use this one.
  • Optional, a USB power bank.

The journey begins

  1. Bright eyed and bushy tailed I set out. Loads of my other work integrates using the Flowdock SDK, which is published on npm. So I went looking for Scroll pHat support within nodeJS. A quick search found a WIP library but at this stage I knew I wanted scrolling text, and that library marks scrolling text as a todo.
  2. If Scroll pHat will not come to JS, Flowdock will come to Python. With JS out of the equation for supporting the Scroll pHat I took a quick look around for Python libraries that would help me with Flowdock. Probably too quick as it turns out, but that’s for ‘next steps’. I knew that I wanted to register a handler against their web stream, which eliminated a couple of options, and I also did not care about posting to their endpoints.
  3. I slept on it. I’m often quite surprised by how much pondering my brain does while sleeping, and I’d love to skim-read a scholarly article on this. I went to sleep with the bits that would solve the problem, even though I did not realise it at the time, and woke with an idea.
  4. Just execute both. I realised that I knew how to listen to Flowdock in NodeJS, I knew how to send web requests from NodeJS, I knew how to listen to web requests in Python and I had a library to control the Scroll pHat in Python. This made a complete chain from Flowdock to Scroll pHat.
  5. Wrestle with Docker a bit. Getting the execution environment to manage both npm install and pip install took a little learning, that I’m sure others would have just done in five minutes. In the end I went from our node base image (because apt-get python had less setup than node). For future similar projects I’m likely to just start from the dockerfile.template from this repository. It installs a bunch of stuff from apt-get, a couple of things from npm and a couple of things from pip, before executing a shell script that creates the Python web server as a background task and also the flowdock listener.
    • Use apt-get python-numpy, no seriously. The compile time on python-numpy if it gets bought in as a pip dependency is approximately forever.
  6. Long execution vs web requests. The scrolling of text across the screen required a relatively long execution time, but it was best if the web request resolved relatively rapidly. This lead to the LedManager class in http_to_led, which uses a thread to manage the scrolling and holds a queue of messages to allow the web request to return promptly.
  7. Funky screen-saver. The Scroll pHat library had a couple of cool example projects, and I decided that in-between the marquee text the screen should display a pattern, to confirm things are powered up. A bit of interesting maths with sine curves, pythagoras and a time-based offset and I had a ripple effect.
  8. Summit! So, packed carefully in a lunchbox and full of expectations PhysHubot went to summit. I mostly left PhysHubot alone, I hope that Hubot isn’t the jealous sort, because, well there was this Beast v3 tile that I really wanted to put some code on.

Next steps

  • Bring it all onto one language and one process. PyFlowdock, this is where you come in.
  • Mount the RPi Zero W in a manner that won’t impose mechanical wear on the pins.