All right, that was finally easy, I don’t know how the agent is implemented, but it handles very well this very simple custom test:
LOCKED
12
85 328
625 297
625 427
800 325
800 250
800 400
800 475
313 197
625 100
313 400
800 100
800 150
11
2 6
0 7
9 2
3 1
2 5
4 1
9 1
0 9
7 8
8 10
8 11
0
6
3
4
5
6
10
11
Facts:
- This test is compliant with the rules
- This test is “solvable”, it has a least one solution
- This test is even easy ! Takes 10 seconds to solve on “paper”, it’s a very simple and obvious graph
- And nows comes the fun part: out of the 24 Java published solutions, only 10 are able to solve it !!!
Yes, I did spent 10 minutes copy/pasting the Java solutions and evaluating them against my custom test.
It’s a bit disturbing to see 60% of the published Java solutions not able to solve this easy case, although they can solve quite complex ones such as IDE test 4/5/6. While I said earlier that I’m not satisfied with my own solution, at least it solves this test.
I don’t know about the other languages (I only have access to Java solutions since I only solved it using Java), but my guess is that we would have the same kind of ratios.
And I’m quite sure I could easily raise this ratio even further with a little more complex test.
So, open questions:
a) am I wrong somewhere on my facts ?
b) should this puzzle be “redesigned” ?
c) is your solution solving this test ? 