June 10, 2012

Impressions of Pittsburgh TechFest

Saturday saw Pittsburgh's first TechFest, a day-long set of sessions with seven (!) tracks covering a number of different technologies and disciplines. I wound up getting there about 15 minutes after lunch ended[1] and ran into a former co-worker, and instead of going to the current session we chatted. Despite the environment (school halls aren't the best place for a hallway track) that's probably the best reason to attend an event like this. Sure, you learn things here and there -- and some of the sessions were long enough to get meaty -- but having a bunch of motivated people around each other is the real payoff.

The two sessions I did see appeared to be pretty different but wound up being quite similar in scope. The first was on Scala and Actors from Josh Suereth. Josh is both super nice and super smart, but that's not always a great combination for a presentation. The last time I saw him speak was in October 2010 for an overview of Scala, which (referring back to my notes) was jam packed with information but unfocused, which is a problem with a lot of technical presentations, including my own.

This was much different. Josh started by demoing in a shell the amount of Scala you needed to know for the presentation, and then moved on. It's so easy to get wrapped up in language intro because it's easy and I'm sure he could riff on Scala nuts and bolts for hours. But while it would be easier for Josh -- no preparation needed! -- that's not going to be as useful for the audience.

The main point of the talk was to discuss actors and some ways that they shape your solutions, to get you thinking in actors and appreciate some core differences as to how they'd shape design. A key point he mentioned multiple times is that you need to think topologically, not lexically (when thinking of error handling) or even in objects. To show what that meant he took a real-world problem (document indexing and searching with scatter-gather) and broke the system down into the important components, pointing out trade-offs and the reasons for doing things like creating temporary actors to hold the state (because the aggregating actor can only process a message at a time, you don't want it to wait).

He went pretty quickly through the material, but he covered a useful patch of ground in enough depth to provide lots of grist for your mill. And he also left enough time at the end to answer some substantive questions, even referncing deep detail (interaction with the Java memory model) about the Akka implementation of actors in response to a question about whether they could be ported to the CLR. It was exactly what I was hoping for.

The second talk I went to was on Hypermedia APIs from Steve Klabnik. I probably shouldn't have gone because (a) I've talked on this myself, and (b) pretty much knew what to expect from Steve. I also figured I'd agree with everything he said. (Which I did.)

But it was still very much a worthwhile talk. Steve knows this material backwards and forwards, and needed no notes to talk cogently and in depth about subjects as varied as his "favorite" RFCs, history of REST and its back-and-forth in the internet arena, and post-structuralism as it applies to decentralized networks of data and people. And while he referenced that it's hard to talk about REST without using its terms, he was able to explain things to newcomers clearly and concisely. The talk was an hour and a half, but it certainly didn't feel that way.

A couple thoughts I had during the talk:

  • It would be interesting to see a concrete example of a media type that could be interpreted different ways depending on its definition. (For example a 'screens' attribute in a media type describing a film's distribution vs 'screens' in a TSA report.) But, as he said, this is also as much art as it is science and to be done well would probably be another many minute talk. Which I look forward to hearing from Steve.
  • He referenced that we keep repeating the mistakes of distributed objects. Another great talk (which might go along with this presentation from Steve Vinoski) might go over why we keep doing so. Why do we keep repeating this? Is it because of the never-ending tension between management who wants interchangeable programmers and repeatable processes (e.g., creating clients through code generation) rather than craftspeople? (Simplistic, but still.)
  • Hashing over media types winds up being a lot of the work done by domain experts so they can talk to each other, or create systems that talk to each other. Think HL7 or any of the X12 domains. It's incredibly hard work for non-trivial domains to get people to talk and understand one another, much harder to get computers to understand one another as well.

So a short, but great day. Looking forward to the next one.


[1] In addition to hanging out with Ella, the entry to Laroche is marked quite sparsely and I drove past it once. Seriously.

Next: Never have to worry about
Previous: Goodbye Kim