distcc is My Buddy

After a conversation Richard (richlowe) and I had last night, I decided to actually try out distcc. It's neat, it works, and it appears to speed compiles up. I was a bit worried about having different versions of GCC, but so far it seems to be handling it just fine (the versions are 3.3, 3.3.1, and 3.3.2). The only quirk I've seen so far is it appears to get stuck at the end of a compile every once in a while. This is mentioned on the distcc Problems page, but I'm running 2.4 on both the machines, not 2.5. Also I wonder if I'm losing my Athlon optimizations, and if so, how that will affect performance. I'm going to try some tests with kylie in the mix as well, to see if it's still worth it with Internet latency thrown in the mix. I'll post them here when I'm finished.

The other interesting thing Jon-Anne (lilyj), Richard (richlowe), and I talked about is was a learning information search tool (roughly comparable to the "agent" craze a few years back). I know there's a GNOME application that does something similar — it "knows" what your doing in GAIM and Nautilus, and constantly updates with information that might be relevant &mdash but we were thinking of something that was more standardized, abstracted, and modular. Some of the stuff we came up with: standard protocol for applications to pass metadata to the daemon (I was thinking something that looked as much like HTTP POST as possible), a pluggable daemon backend, so you could add additional search modules (that either queried different backends, or categorized, analyzed, and ranked the data in different ways), and a standardized output format for the aggregating the various search results (Richlowe suggested RDF, but I was thinking OPML might be more appropriate). For the modules, if you had various modules that queried things like Google, Syndic8, newsgroups, and some sort of filesystem database (Jon-Anne suggested that this should be some sort of associative mesh, rather than a simple keyword database), etc. The other thing is that results would be associated with related searches. It should be smart enough to realize that if you're looking at a cooking-related web site, and you start talking about cows in your IM program, you'd probably be interested in hamburger- and beef-related content.

Interesting stuff. And technically do-able, I think. Does anyone remember the name of the existing GNOME application? I'd like to take another look at it and see what they've got so far.