Wow. It's so... special.
That said, even though the format is clunky and horrible, it genuinely is self-documenting. I managed to parse it in a surprisingly small amount of code without needing to refer to the documentation at all (which is good because I couldn't find any). Some datasets needed special treatment because the value that came out of the formula specified in the PDS label needed to be offset to produce a physically valid value. But it all worked. They're even mmap friendly. So, ancient and loathesome yes, but also pragmatic and effective yes.
The C++ parsing code, if anyone cares, is at https://github.com/davidgiven/flooded-moon/tree/trunk/terrai... in the pds.h and pdslabel.h files.
Incidentally, if anyone knows of:
(a) a raytracer with volumetric rendering that has a programmatic API (right now I'm using a hacked version of Povray) (b) higher-resolution terrain data of the moon than the 6GB Lunar Reconnaissance Orbiter DEMs available here: http://wms.lroc.asu.edu/lroc/view_rdr/WAC_GLD100
...please let me know?