So I hacked PieSpy to feed it with data from the Lunarstorm community instread of IRC. I wrote a brief description in Swedish about it all. The main story is that I made a big-ass PieSpy social network graph out of guestbook data from Lunarstorm, and the current result can be viewed here, and a friend used Zoomify to throw together a Flash zoomification of the same image, which can be viewed here. Pretty cool. In hacking termes, most of my work went into changing PieSpy to draw a labeled grid onto the image and append an index to the image, containing an alphabetical list of all the nodes with pointers to each node's coordinates on the image. A smaller example can be viewed here. A package with source and required libraries is available here (it's the PieSpy distribution changed to use my own main class instead of the standard IRC bot thing).

I hope this year will prove more satisfying than the last.

Not much has been happening lately, as I yesterday came back home after spending five days in Sälen, skiing, drinking glühwein and in general having a darn good time. I learned the beauty of off-pist skiing - getting away from the public pistes and ski lifts and to the untouched, deep snow was really cool.

For my Swedish audience: yesterday, I reluctantly decided to publish all my diary postings from a certain web community here. Reluctantly, because some of these are really personal and may contain information that I might not want to publish publicly on the web - the diaries on the community are only available to registered members, and you can always see who has read your diary, which gives a comforting sense of control when writing about personal things. But I wanted to have somewhere to post the things that aren't of interest to non-swedes and the techie audience of Advogato, and didn't want to end up with three diares/blogthingies. To export all my diaries from the community, I made a small command-line version of my program Nular that uses basic screen-scraping techniques to retreive the data, and hacked a function that allows a user to save all diary entries to a directory. Then, I created a cron job that does this once a day so I can continue posting on the community as usual and all posts will show up on my web page as well (and complemented with a script that I can easily trigger manually when I want to force an update). The program and cron job runs on my coLinux Gentoo installation on my Win2K box at home, and uses rsync to transfer the data my real web server, which in turn contains a simple JSP page that reads the file structure to generate indexes, display the data and so on.

A few days ago, I was experimenting around with compiling static binaries with gcj to see if I could easily distribute programs such as md5i in native form to people who don't have or don't want Java. It turned out that the executable size was somewhere around 4 MB - for a Java program which has a bytecode of a few kilobytes. It would be nice to be able to tell GCJ to only link those classes that are referenced in the code - and perhaps be able to supplement that list if you use dynamic class references in the code. For example, I use the Sun MD5 implementation by invoking MessageDigest.getInstance("MD5"), so the compiler/linker would need a hint as to which classes I actually need as it is not explicitly reference syntactically. But one should be able to easily produce a list of needed classes by running the program and passing -verboce:class to the JVM. For all I know, gcj may already do this, and my small program really depends on over four megabytes of runtime class code. Anyway, another thing I'd like is to be able to suppress the WARNING: could not properly read security provider files when the program is run in machines without a GCJ installation, as in most cases, the standard GNU security provider will work fine.