The joy of horribly biased benchmarks

I was a bum and did almost nothing today. Well, I did some moderating stuff, but who cares about that? But I'm updating my journal anyways, basically just to bump it so that I can tell at a glance if any of the other developer journals have been updated when I next visit.

Actually, I did one vaguely productive thing: I finally got around to testing that the sortable_string class performed better than std::string when used as a key in a binary tree. On one horribly biased benchmark for std::set::find that does not accurately represent any reasonable input set, sortable_string performed about 3100 times better than the std::string implementation that ships with MSVC 7.1. Of course, that doesn't prove anything except that I know how to skew a test. More reasonable benchmarks slim that number down a bit.