Thursday, March 06, 2008

@SD West 08: Highlights of Day Four

HTTP for Web Developers (Jason Hunter)

After his yesterday's talk about caching, Jason detailed the core of the HTTP magic. Indeed modern web development tools very often abstract HTTP out of the development paradigm. You end up with developers who talk about controls on forms as if they were developing Access applications (do not laugh, it happened to me).

HTTP is a wild world: with browsers and servers interpreting the standard in their own manner (often leading to bastardized additions to the standard itself), developers need to know what is happening under the hood.

I will not detail all what Jason talked about but I am always amazed by the amount of extra knowledge you can get when an expert revisits the basics! A highly recommended exercise for anyone who does things with the webernet.

Clean Code: Functions in Java (Robert C. Martin)

Woohoo! A new class from Uncle Bob! It is of course impossible to properly summarize an hour and half of such a magistral session. The main concept that this talk detailed is the following:
"Making code more readable allows us to write code faster, because we end-up reading code twenty times more than writing code when we are in the process of writing code."

So how to achieve this? The general idea is that a function should be an executive summary of what is happening below it: it must be short and should not mix concepts from all layers of the application, in order to remain understandable. This must lead to a recursive functional decomposition and the creation of functions with meaningful and descriptive names all the way down through the abstraction layers. And by the way: if you find hard to find a good name for a method, it is probably because it does too much things.

Here is short summary of the talk:
  • Write small function, and if possible, even write smaller ones.
  • In a method: do one thing to preserve cohesion, i.e. one level of abstraction.
  • Three is the absolute maximum number of arguments: think about how two arguments are already confusing (right order?).
  • No side effects: a method should have no strange temporal coupling, it should always have the same semantics whenever it is called.
  • Throw exceptions instead of returning error codes because they get mixed with the natural outcome of the method and force ugly calling code.
Someone in the audience asked a question that is dear to me: "there is nothing new here, why keeping repeating it?" Uncle Bob replied: "because we do not believe it". This sounds like the classic "Shema Yisreal", with prophets and priests repeating the same things again and again because the people was unfaithful. We need to hear again and again to internalize things so we practice them. Writing code as if reading it matters is not natural but we can reach this goal thanks to method extraction refactoring.

Ouch. I will spend the rest of my life refactoring my own code.

Parallel or Perish!! - Are you Ready? (James Reinders)

After assuring us that multi-core are not a temporary trick to gain performance that will be deprecated later by faster single core processors, James made very clear the urgency and necessity of a mindset shift towards parallel programming, especially for client side developers.

He then gave us several tips that you can read in an article he wrote for DDJ a while ago.

I have really appreciated his remark on how to consider leveraging multicore parallelism in light of an increased work load, and not only by looking after accelerating existing code.

I was surprised by James remark about the lack of parallelism abstraction in Java: since version 1.5, the JDK offers a wealth a concurrency oriented high-level constructs (like collections, conditions, mutexes, futures and whatnot...). He might refer to the thread management themselves, but again I find that executors are offering an interesting way to splitting work around threads. But he confessed his main focus is on C/C++ though!

I am in fact more surprised by the number of Java developers who do not have clear (or at least basic) guidelines about how they write code that runs concurrently without coughing. As James said: we all have to learn and think about parallel development. Coming from Intel, this shows how much learning we can expect in the coming years!

In the meantime, thanks to either smart scheduling or pure randomness, this keynote was followed by two sessions on Java concurrency and parallelism: so I had an immediate opportunity to keep climbing the learning curve!

Thousands of Threads and Blocking I/O (Paul Tyma)

I was really looking forward this session and was not disappointed. The breadth and depth of the material Paul shared with the audience was worth attending. He did a great job debunking some myths about synchronous vs. asynchronous server models, all backed with hard facts. Here are a few of them:
  • Java asynchronous NIO has higher throughput than Java IO (false)
  • Thread context switching is expensive (false)
  • Synchronization is expensive (false, usually)
  • Thread per connection servers cannot scale (false)
Visit Paul's blog for more details and even the slides of the presentation. All in all, this talk confirmed my quasi-gutsy feeling about concurrent development: let the threads flow and be smart about their touch points!

Anti-Patterns in Software Projects: Human Factor (Rob Daigneau)

Writing code as a very mental process, hence is subjected to our human nature, with all its up and down sides. Rob gave a great presentation about how human factor affects software development and how can leaders of all sorts act to make the workplace a better place. This of course spawned a lot of lively discussions, as everybody has so much to say about what happens in their own life in software development.

It is hard to summarize all what Rob said but I really liked his emphasis on passion and how essential it is to let it burn in developers (without let them burning it, which is destructive). I also appreciate one of his final word: except if we write life-critical software, well, it does not really matter that much so better having fun while working. Sounds like a good advice to me.

Developer Bowl

Last year Developer Bowl was all about Googlers and their incredible supremacy in term of computer science knowledge. With Google being in no shortage of big brains, I walked in the theater with the expectation to get a fair deal of deja-vu...

This year was as entertaining as last year's edition. The questions, which were partially submitted by DDJ's readers, were much less oriented on the fundamentals of computer science and more on history and anecdotes. So this year Google did not pass the first round and IBM won versus Intel!

Oh well, it is all rigged anyway ;-)