Semi-closing a hole

I think the test repeatability issue would be best solved by having the test framework able to select the random hash. One of the biggest things that's made my test code easier to write is having the framework cause the function that returns the current time return a constant instead of the actual time. Then there's something to have the random password salt be a particular value. Along with those, it's quite reasonable to have the hash order be fixed. (For that matter, it would even be nice to tie all these together, so that your application generates obviously-wrong times if your hash isn't random, so you'll notice.)