In addition to our opportunistic write-tests-as-you-go approach, we should also direct our testing efforts on the functions that most *need* to be tested. This includes (possibly) * Big functions * Volatile functions (ones that have changed many times) * Functions that have had lots of bugs in them in the past * Functions nobody understands * Functions with high complexity * Functions with high complexity-to-line ratio * Functions with high code-to-comments ratio We should also look at what's _hardest_ to test, including possibly * Anything that hits the OS directly * Anything that's windows only. * Anything that calls many other functions * Anything that calls many other modules * Anything that indirectly calls most of the rest of Tor.