Friday, December 08, 2006

OSDC 2006, Melbourne Australia

Day 1



I attended this year's Open Source Developers Conference in Melbourne, Australia, from 6-8 December 2006, after a week in Brisbane at The University of Queensland to research and use the library.

The fabulous exhibition of aerial photos by artist Yann Arthus-Bertrand was in Melbourne and set up by the Yarra River near the centrally located Flinders Street Station. The Internet version is good and I recommend viewing it, but it doesn't compare with interacting with a live crowd. I had the good fortune to see this exhibit a couple of years ago while in Bristol, England.

Randal Swartz gave the first keynote. He presented a tried and true talk regarding his position on Open Source licenses. He is a BSD/Artistic adherent, compared with Richard Stallman's insistence on the GPL. Randal and I discussed this in some detail after dinner and he saw my point regarding the desire to keep free code free - while being pragmatic about the need to license looser than the GPL to encourage use. In return, I freely acknowledge his point that very unrestrictive licenses encourage the wider use of Open Source software. It was a pleasure to discuss this with someone who both knew the licenses and could have a discussion about it without getting religious and emotional.

On the other hand, Randal did get emotional when I asked him about Perl 6 and Parrot ;) He made the fine point that Perl 6 will continue to scare him until a real release date becomes guessable. He needs to prepare training material (and pay for it) in advance, and that makes it a big deal for him.

My talk on RESTful Software Development and Maintenance went quite well and I was pleased to see that people were interested. It was a bit risky discussing pure future-oriented research at a (pragmatic) Open Source conference, but I'm glad I did it. Lots of interesting discussions ensued, fostered by breaking the mould a bit. I changed the talk at the last minute to include an longer introduction on why software maintenance is so hard, primarily to see whether it resonated with people. It did, and I plan to blog a bit of it as I get my head around it better.

Dr. Damian Conway presented his amazingly cool The Da Vinci Codebase (see the O'Reilly page on this here). If you get the chance, it is not to be missed.

Paul Fenwick of Perl Training Australia provided some entertainment on the train ride back to Flinders Station by showing the World of Warcraft version of the song The Internet is for Porn on his laptop at the highest volume. Reactions from other passengers were, umm, mixed. Several (male) commuters watched avidly while a group of Sikhs went stoic. Paul's girlfriend put up with it, while suggesting headphones occasionally.

Day 2



The second day of the conference was not tailored directly to my interests, but I took the opportunity to learn about other areas. The recent advances on Python's ctypes library are very useful and will, upon release, allow python programmers to have wonderfully powerful access to OS and hardware. I also learned a bit about the Mono project, the Boo scripting language for Microsoft CLI/Mono and attended a cool talk on the Australian Synchrotron and its use of Open Source software. It is worth noting that the control systems engineer from the synchrotron had no idea what licenses his free software was under because he was "not interested in I.P." I consider that sort of attitude dangerous to the entire Open Source software movement. Very little software is actually in the public domain due to implicit copyright rules, so licenses are the only protection for Open Source users and developers alike.

The highlights of the day were the incredibly high quality lightening talks. These included an understandable five-minute description of the RSA algorithm, a quick-and-dirty Catalyst application and some useful discussion on community building.

Andrae Muys gave a five-minute lightening talk on Mulgara. He presented (in five minutes!) his theory that RDF data stores exactly match the relational fifth-normal form, where each table contains exactly one data element and some number of foreign keys. He used C.J. Date's classic parts-suppliers database description and extended it with optional data. It was a good explanation of Mulgara's sweet spot.

At night, Andrae and I went to a funky little bar off of Little Lonsdale Street called the Horse Bazaar. The walls were covered with moving projections of city scenes and a live band played rapid Gypsy music. We were supposed to meet Randal and crew, but they never showed.

Day 3



Scott Penrose gave a keynote on his project Zaltana. Zaltana uses HTML tag attributes as hints for styling, including Javascript behavior. I haven't looked at it in detail yet, but it appears to be a similar idea to Behavior. Behavior uses CSS class attributes instead of namespace-id'd special attributes. Zaltana's goal is to wrap existing content with minimal effort with a new or custom style. I was more interested in it until I realized that the existing application has to be directly modified - I was hoping that some form of dynamic filter or preprocessing script could be used to map CSS class attributes to actions, but no.

Miles Byrne spoke on Programs are just words: Designing domain-specific languages in Ruby. I like Ruby, even though I have only coded a couple of toy applications in it so far. Miles gets the award for the best cat pictures in the conference. He gave an overview of the Ruby language, and then pushed on to many, many examples of dynamic languages written in Ruby. He showed examples for HTML parsing, Rake, ActiveRecord and several others.

Simon Raik-Allen presented on JasperReports (see the Sourceforge project for downloads). JasperReports is a Java library for generating nicely formatted SQL reports. It is LGPL licensed.

JasperReports has some commercial add-ons from JasperSoft, the initial developer. JasperSoft is busy creating an enhanced infrastructure layer focusing on the semantics. Given that they already have OLAP support and industry-sufficient GUIs and query- and report-building tools, the semantic mappings are about all that is left. I wouldn't want to be working on Crystal Reports this year and wonder how threatening this will be to data integration companies like Experian, Nimaya and Data Infinity.

Distributed Development



Martin Pool and Robert Collins reported on the Bazaar distributed version control system that does not have a centralized repository (GPL license). They have some very radical, but potentially useful, ideas. For example, they prefer to encourage many, many branches - even a branch per bug fix/feature addition. Merges to the mainline are also done early and often and become the equivalent to an update. They thus have some tools, such as reports, to manage and view the status of all those branches, not all of which have been released as OSS yet (although at least one is available as a free service on LaunchPad). Bazaar is for use specifically by distributed development teams and is used by Ubuntu Linux, a sizable project. That makes it particularly interesting to me since I am working on software maintenance for large and distributed teams.

Martin noted that an anti-pattern has developed in Bazaar use; some developers tend to put all of their work in their own branch instead of one branch per feature. Robert also gave a number of good reasons for having a main line, or trunk.

The fear of forking is often thrown up as a reason not to do version control the Bazaar way. Martin and Robert suggest that good distributed version control also makes potential reintegration easier.

Merging (especially often) can have some pitfalls. Adam Kennedy asked particularly about code conflicts. He wanted to know in advance of a merge whether a merge would be clean. Martin said that they are thinking about this, but the solution for the moment is to just ensure that the main line is always clean.

Andrew Bennetts carried on this theme after lunch. He talked about using Bazaar and Patch Queue Manager to ensure that tests always pass on the main line. The only way to ensure the main line is safe, of course, is to also use Patch Queue Manager (or some similar systems) to manage dependencies.

Andrew suggested that code reviews are more important for distributed teams than for teams who can informally and routinely overhear each others' conversations. This is an interesting thought because it begins to sound like a methodological requirement for larger OSS systems (that have historically not operated that way, possibly due to size). VOIP is also making distributed pair programming more affordable, especially for OSS projects.

The lightening talks were awesome, again. I am suddenly thinking about a conference format that starts with a series of lightening talks so that individuals can have a basis for choosing which sessions to attend.

Richard Jones showed the Selenium IDE, a test suite IDE for Firefox. Selenium tests themselves are cross-browser. There are heaps of (poorly implemented) test suites for browsers, but this one seems worth a look. It is also OSS under an Apache 2.0 license. The addition of an OSS IDE to this market makes it very easy to construct tests.

A talk was given on SELinux, when what I really needed was a talk on chcon :)

A wild talk was from a guy who implemented Lisp without (most) parentheses. It was actually readable! Kinda cool...

Jon Oxer proposed a really simple, but complete, language he called OSDcLang (actually yesterday). Today, Paul Fenwick introduced the ACME::OSDC perl module to compile it. Scary. Cool, but scary.

There were too many lightening talks to really capture. One guy proposed that OSS apps have source code inspection/editing access in the GUIs. Another introduced a perl module he just wrote and uploaded to simplify for/else syntax.

Andrae and I had a discussion with Adam Kennedy regarding spam prevention. I won't steal his thunder, but I am going to introduce him to friends at Yahoo! who are constantly looking for better anti-spam techniques. Adam is great at solving general problems and categorizes well between people and automation. I have some hopes for this one. I thought I had a solution last week, but it was shot down upon more detailed review.

I learned a lot about Perl internals after dinner by listening to Randal, Adam, Paul, Andrae and some others discussing Adam's attempt to make a perl parser. It turns out that nothing, not even perl, parses perl. There is no grammer, and there are places in the code where perl actually makes probabilistic guesses about how a token should be interpreted based on following information. Throw in dynamic language extensions and you start to see where this is going. Adam was trying to parse perl entirely in order to provide an API for editors to use. He still has some hope, but I was surprised at how hard it was.

Overall, it was a great conference. I look forward to coming next year, hopefully to Hobart, Tasmania.

No comments:

Post a Comment