Yay, a response from the Subversion dev mailing list! They pointed me here, which is a pretty good description of the format. I’m not sure I understand why they chose to model it on RFC-822, since it’s not a very efficient format, and also because they don’t adhere to it particularly strictly. The key thing that I wasn’t sure about (and which is clarified in this document) is that the Content-length and Prop-content-length properties are lengths in bytes, not characters. Makes sense, otherwise they’d have to mess about with the variable-length representations of Unicode.
In the meanwhile, my script is happily dumping revisions, and I’ve confirmed that they can be loaded up OK. It’s not half as inefficient as I thought it was going to be: I dumped 255 revisions (8.54GB) in about an hour and a half. The other half of the performance equation is loading the dumpfiles into Subversion, which takes roughly the same amount of time.
I imagine the script might be useful to other people, so I’ll probably write some documentation and post it here once I’m done with my testing.
You just don’t update your blog often enough… and after shaking off the bad habit with email, I have acquired its sister bad habit: checking blogs every hour on the hour…i know, life would be simpler if i subscribed to RSS feeds, but I don’t really get them… anyway, i wouldn’t want to read your post in my email browser, i’d want to see the webpage.
the point is, you should update your blog more often. i’m dreadfully bored and currently pissed off about three different things at the same time… was fine and patient and quiet about the first two, and then the third one just came along, and now I’m mad, to the point that I can’t settle down to work and i’m writing this comment on your subversive dumping post. this is subversive dumping. hah.
a.