My XSLT namespace splitter originally didn’t support IRIs. It was just a standard stylesheet encoded in UTF-8. When I added IRI support, I changed the encoding to UTF-16 which reduced the file size significantly due to the large tables of far-eastern characters in the file.

It was all working fine, I was ready to upload it to the web, I checked it into CVS first. Oops. Do you see the mistake?...

I’d forgot to change the filetype to Text/Unicode. CVS completely mangled both the version in the repository and my working copy. It didn’t just screw up the encoding - it deleted random chunks of the file too.

Luckily Vista’s “Previous Versions”, based on Volume Shadow Copy Service, takes a snapshot of your harddrive on a regular basis, and I was able to just right-click the file, and choose an earlier version of the file to recover. Interestingly, these snapshots appear to have their own browsable UNC paths, eg: \\localhost\C$\@GMT-2007.02.24-08.58.34\projects\newrdfsyntax. Cool.