I spent far too much time at the hackathon fighting with gcj. It has been for a good cause though =) Thomas Dudziak and I, after mucking for a while, are leaning towards working with the lucene4c rather than the straight gcj. Lucene4C uses the gcj'd one as well, but puts a more idiommatic C api in front.
In addition to that stuff I played taskmaster with James to flesh out the ActiveMQ bindings for Ruby I started a while back. It looks like we'll stick with the code generation approach for message marshalling/unmarshalling -- it is handy as *all* of the openwire protocol handlers are code generated from the protocol spec. Pretty cool. James's estimate of the openwire protocol performance is about 10% behind the most highly optimized native -- but it has some deeply interesting encodings for performance. I'll take the 10% performance hit for ease of implementation. Once it's up and running I can worry about speed (and maybe do a C library if we really care). We *hope* to have something roughly usable possibly as early as monday. Don't hold me to that, it presumes that I find a nice way to do UTF conversions.
The Cocooners are sick and twisted people (in a good way). Over various beers we got to talking about obscure but... just possibly useful hacks. Tunneling over pings and tcp sequence id's, for instance. On a potentially more interesting note we got to talking about the Groovy classloader which groks maven repositories and between Torsten and Upayavira they had the most brilliant idea I have yet seen. Forget using the maven repository, use koders.com, google, and viewcvs with a compiling classloader. Given import org.apache.Wombat statement it can go get CVS head, compile it, and load it... ;-)
In a Geronimo talk right now. Every other sentence it seems is "Aaron Mulder just added ..." =) Aaron got credit for one of Erin's things (pretty startup) -- though I am not sure who actually coded it. Go Aaron!