This kata was a lot of fun! I am proud of my solution, as my recursive algorithm turned out to be many magnitudes more efficient than I hoped. I track both the time and the number of calls into my recursive worker function. Here is one of the worst results I saw:

CLUES: [0, 3, 0, 5, 3, 4, 0, 0, 0, 0, 0, 1, 0, 3, 0, 3, 2, 3, 3, 2, 0, 3, 1, 0]
Solved with 373 calls in 31.658 ms.
5 2 6 1 4 3
6 4 3 2 5 1
3 1 5 4 6 2
2 6 1 5 3 4
4 3 2 6 1 5
1 5 4 3 2 6

Now, I need to rework all references to constant sizes to make this solution work for all the other kata like this one ^_^

I have been trying to solve this problem with Java and ran into this same problem. I currently have 220 passing tests and 10 failures. All 10 failures are related to this flaw. I am not able to see the tests' code, so I cannot see what is actually being checked. I tried giving my exception classes with messages that matched the expected unclean termination exception messages, but they are apparently not checked. I thought it might be expecting it in the OutputStream objects, except I discovered that every test of this category passes a null OutputStream.

I am inclined to conclude that this problem's Java test suite was broken at some point in the past and never fixed. The fact that I am replying to a two-year-old comment that was never addressed is incredibly discouraging.