Wednesday, October 01, 2008

JAOO 2008, Wednesday

Everything is dead and crap -- but here comes DCI
James O. Coplien held a talk called Not your Grandfather's Architecture: Taking Architecture into the Agile World - take 2. The talk was highly unstructured, but entertaining. James declared a lot of concepts to be crap and/or dead, such as runtime-polymorphism, class diagrams, Java, and aspect orientation. E.g., class diagrams was declared waste of time, because the end user doesn't care about classes -- however, object diagrams were somehow very good (do users care about objects?). Much of the criticism was highly appropriate, but unclearly motivated. Next, James described a paradigm which was described as brand new, and as the way to "save" object orientation: DCI, short for Data, Context, Interaction. The paradigm was illustrated through C++ code examples, and seemed to rely on a special way of combining C++ class templates with multiple inheritance -- mixins, effectively. The goal of the new paradigm is allegedly to increase code readability. However confusingly the concepts were illustrated, I left with a feeling DCI may actually be something we will see more of.

Next, I went to a session about Lego's new WeDo product, presented in a clear and inspiring way by Eik Thyrsted BrandsgÄrd. WeDo is a robotics Lego product, like Lego's Mindstorms NXT. It will be available next year. WeDo has connections to the OLPC project; consequently, Lego has put an effort into making the product inexpensive. WeDo is also targeted at an audience younger than that of NXT, i.e. age 7+. Surprisingly, the presentation contained a lot of idealistic statements - and it was quite convincing: Through world-wide Lego robotics-competitions, kids may have a chance to become fans of technically creative peers -- instead of the teen-band of the year (my interpretation). And due to being affordable, WeDo may reach to a large number of schools and end-users, triggering creativity. Several components of WeDo are open source. Apart from that, Eik described Lego's development model where they take advantage of having partners world wide: A new idea is spawned in Denmark in the morning and made ready for consumption by software developers in the US. When the Danes go home, the Americans spend the day implementing it. Finally, the implementation is sent to the QA people in India. And the cycle starts over. In other words: Lego isn't just outsourcing for financial reasons. The talk was highly refreshing: Instead of paradigm/architecture name dropping (which some of JAOO's other presentations were full of), this was a talk which actually gave a strong urge to write some code! (My son will actually turn seven, soon...)

The Lively Kernel
Before Dan Ingalls' talk, The Lively Kernel, I thought that Google's web-based office applications marked the frontier of online web applications. I was wrong. Ingalls performed an hour of outmost sorcery, dragging, dropping, copying and morphing highly live objects around. The Lively Kernel is a collection of javascript, using SVG as "canvas" instead of HTML. Think of the Lively Kernel as Flash, only built on open standards and without the plugin requirement (as long as you use a modern browser). Lively Kernel even includes a browser based development environment (profiler included), and it uses WebDAV for versioning. I wasn't sure if/how Lively Kernel applications can be integrated with existing web pages. If they can't be well integrated, they may suffer from the un-webbiness problem which is discussed in an episode of the Stackoverflow podcast series. Looking forward, Ingalls mentioned that online collaboration features are somewhere around the corner. SVG has gained momentum now that most modern browsers support it and Wikipedia has started using it extensively; this could be another move forward. I wonder when Microsoft wakes up and adds built-in SVG support in Internet Explorer. By the way: FLOSS Weekly has an interview with Ingalls about the Lively Kernel.

Pattern of boredom
I'm a big fan of domain specific languages (first and foremost: SQL). So I chose to attend Patterns of Internal DSLs by Martin Fowler. External DSLs are like yacc grammars, while internal DSLs are DSLs embedded into the host language. The talk seemed to be aimed at the (many) poor people forced to work with a language lacking expressiveness. Although Fowler did mention how some things can be done rather elegantly in Ruby, most of the talk described various ways of twisting and abusing a Java-like language in order to express data in a way where chatty language boiler plate syntax doesn't hide the essence of the data. (This reminds me of yet another episode of the Stackoverflow podcast series where Spolsky distinguishes between languages in which you can easily express tree literals, and those where you can't.) I would recommend that Java programmers simply accept some lack of expressiveness and don't spend any time on the kind of DSL hacks which Fowler presented.

Parallel extensions to .Net
It's commonly heard that CPUs basically don't grow faster any more, so we need to employ several CPU cores if we want more performance. Natually, this entails that the multiple CPU cores can actually be fed some work -- which can be a serious challenge. In some parts of the IT world, the "parallelization problem" has already been solved: Server-side web applications can easily spread work to several cores, because the work of a web server is inherently concurrent. And SQL databases use a language which is for the most part very declarative, so the better DBMSes are actually able to chunk up work without special effort from the systems developer. But there are other cases where programmers need to implement parallel execution in more or less explicit ways, e.g. in traditional desktop applications, games, and in high-performance computing (HPC). The last JAOO talk I went to was on this subject.

Concurrent Programming with Parallel Extensions to .NET was a suitable name for Joe Duffy's talk. Microsoft is working on making it easy for .Net developers to choose parallel versions of operations and data structures. Joe's talk gave good and comprehensive insight into that. The extensions are not finished yet, but a preview is available from Microsoft's Parallel Computing Development Center. However, even after the hard work from Microsoft's developers, parallel .Net will still be very much in the "sharp knives" category, and Joe strongly urged people to get acquainted with Haskell which he described as the One True North in this space (Duffy has a Haskell logo tattooed on his arm). It's always encouraging to hear a Microsoft employee recommend something in the open source world. Even more encouraging was that Joe told me that he'd heard that parallel extensions for .Net has already been committed for the Mono project.

Conference over
That was it for JAOO 2008 (although some people were going to stay for a few more days, for Scrum tutorials). After the last talks, there were meet the speakers sessions, but I (and many others) didn't have time for that. Maybe JAOO should arrange meet the speakers sessions every day next year?

No comments: