OSCON 2005 was held in a convention center this year, instead of a hotel, because it just got too big (2000+ people). Too big, in fact, for pudge and myself to cover more than a fraction of the talks and the ideas flitting around the hallways. But here's some of what I found cool last week. And if you attended or presented at OSCON and want to tell us about all the neat stuff we missed, please, share your thoughts in the comments, or submit a fact-rich writeup and we'll maybe do a followup story later.
Aaron Boodman's talk on his extension Greasemonkey was a walkthrough of writing a simple GM user script, a discussion of what's coming up, and some Q&A. Greasemonkey 0.5 ("Now With Security!") is in beta: there are multiple security changes that suggest someone really has sat down and thought the whole model through. GM works with Firefox, Seamonkey, Opera, and Windows MSIE (but not, oh please somebody correct this oversight, Safari).
Ruby on Rails is hot; if you want to develop a web app quickly you can't ignore it. It stresses "convention over configuration" with reasonable defaults. The tutorial went from installation to the "hello world" of the web, a blog (!), in a few hours. Anyone have a real-world example of Rails scaling to a large project and lots of traffic?
DarwinBuild is an open-source project from Apple that aids in building the open-source components of Darwin/Mac OS X. Given a build number of Mac OS X, it will fetch and build the software for that version, allowing you to modify the source as needed, making it easy for any developer to modify everything from the kernel to various utilities (just remember to reapply the modifications after running Software Update, if necessary). You can read more about it from, in addition to the web site, the presentation slides.
Google and O'Reilly gave out the 2005 open source awards, with $5000 attached to each. Congratulations to the winners.
Tony Baxter's Shtoom is a cross-platform VoIP client and software framework, written in Python, for writing your own phone applications.
Novell is still moving its employees from Windows to Linux, which we first heard at last year's OSCON. The migration from Microsoft Office to OpenOffice is complete, and the big step, from Windows to Linux, is 50% complete, projected to be 80% by November. Miguel de Icaza gave flashy demos of some Linux desktop applications that didn't impress this cynical observer very much.
PlaceSite is an open-source project looking to bring physical proximity awareness to Internet access at coffeeshops and other meetingplaces: think "local-only Friendster" and you're not far off. They got feedback from a monthlong trial earlier this year and are working on a new version that will be easy to deploy. Could be neat.
In a great 2-hour session on Wednesday, we got to hear from representatives of four leading open source databases about what they've been working on lately. Here are the summaries...
Ingres r3 has an impressive list of big features. Ingres was just open-sourced by Computer Associates this summer, and it's gotten a lot of attention for being a full-featured enterprise database. Ingres supports table partitioning that can be either range-based or hash-based, which can greatly improve performance in many cases. Its optimizer can now come up with parallel execution plans, which can be useful even on single-CPU machines and non-partitioned tables. There's also federated data storage (one can access data stored in another RDBMS through Ingres) and replication. And they're working on a concurrent access cluster, to allow data to be manipulated not just by multiple threads on one machine, but multiple machines.
A side note: Computer Associates was invited by O'Reilly to talk about its recently open-sourcing Ingres. Its representative, while confessing that introducing a new license was "probably the wrong thing to do," said that other licenses wouldn't have worked for them (the GPL "was seen as viral"). The one question that the audience had time to ask was "is Ingres a dump" -- is CA making it open-source to transfer the responsibility of support from the company to the community? The three-part "no" answer was that there are more CA developers working on Ingres now, that Ingres is at the core of their new releases, and that they've sponsored a "million-dollar challenge" to foster community interest. Time will tell I guess.
Firebird 2.0 has been in alpha since January and a beta is expected soon. Since 2000 much of their development has been aimed at making the product easy to install, and making the code easy for a distributed group of developers to work on. This year they're building features on that groundwork. Their design includes 2-phase commits (since the beginning), cooperative garbage collection (as a transaction encounters unneeded data, it removes it) and self-balancing indexes. Backup has been improved. When 2.0 gets to beta, I'm going to check this out, it sounds like very interesting technology (and apparently it will install with four clicks!).
MySQL 5.0 is in beta, and has been feature-frozen since April. Back in 4.1, its abstracted table-type has been put to advantage with odd engines like Archive (only insert, no update); Blackhole for fast replication; and an improvement to MyISAM for logging (allowing concurrent selects with inserts-at-table-end). Their Connector/MXJ lets you run a native MySQL server embedded inside a Java application. In 5.0 we're seeing stored procedures per the SQL:2003 standard, triggers, updatable views, XA (distribution transaction), SAP R/3 compatible server side cursors, fast precision math, a federated storage engine, a greedy optimizer for better handling of many-table joins, and an optional "strict mode" to turn some of MySQL's friendly nonstandard warnings into compliant errors. And they're working on partitioning, ODBC, and letting MySQL Cluster's non-indexed columns to be stored on disk.
PostgreSQL 8.1 is expected to be released in November or December, after a feature-freeze in July -- and it's an impressive list of new features. Their optimizer will make use of multiple indexes when appropriate, which is pretty darn exciting. The recommendation will be that in most cases it will be most efficient to have only single-column indexes and let the optimizer figure out which combination to use. They're implementing a 2-phase commit, they're bringing the automatic vacuum into the core code, and they removed a global shared buffer lock so they're now getting "almost linear" SMP performance scaling. I've never felt the need for Postgres, but I'm definitely going to look at 8.1.