Hey Brooks,

Adjusted diff to be a proper diff with respect to head.

@ngie I think I had a misunderstanding with what these diff's should represent. I had been updating the page with diff's against the previous commits I had posted, for review purposes. Is the standard protocol for this page to host a diff against freebsd head such that it could be directly applied?

Added a python driver script that outputs statistical comparisons using ministat to files. Decided on python rather than shell because I felt limited by the built ins of bourne shell and the bash script I wrote was needlessly complex in order to perform the simple commands.

Added the canary for the bigstruct test. Unsure if DEADBEEF is the right value to use but it's adjustable. The only problem I have left is that of printing out the entire array on failure. I'm unsure of the proper way to do that. I could print both the input array and the improperly sorted array to stderr, but I worry about if that would be very painful to look over and find the problem. Just the input array would probably be enough to reproduce the bug, but should it be printed directly to STDOUT or STDERR or do I have to do it through some ATF_TC medium?

Update the exits and change to use arc4rand, as well as add Makefile.

I have nearly completed a version of this script that uses function pointers rather than enums, but I worry about it. I feel like it makes the code unnecessarily messy, and when I tried to clean it up with typedefs and macros, it makes it much less readable. What do you recommend?

Updated bench to handle usage better and only use wiki if provided

Updated tests to reflect tests of the mergesort implementation and not WikiSort

In terms of manipulating the algorithm, I've had minimal structural changes, except to modify everything for generality and replace some helper functions implementation in order to support the main changes. I would say there was not a major overhaul from a code layout standpoint.

Updated to reflect argument based design.

This benchmark is my initial design without full understanding of statistical validity. Is it at all still usable or it better to just pull out the relevant bits that run the sorting code into separate files and then attempt to bench them with ministat and multiple trials. Is shell a better medium for this?

Made many changes according to recommendations.