Handle solution counts differently

[ChrisJefferson]

• 0 solutions does not throw an exception
• 'conjure_solutions' is always defined and always a list (this list is sorted)

Implements #23

I wasn't 100% happy with how I handle 0 solutions. I was going to return "None", but jupyter uses that to denote printing nothing at all, which is confusing. Therefore I'm currently printing [] (empty list). I could change this to "No solution found".

I also sort the solutions -- this is partly to make testing easier (as it means my tests don't keep changing), but I image users might also get annoyed if their notebooks keep changing. Of course, if you only ask for 1 solution to a problem with many solutions, you might keep getting a different solution each time with some solvers.

[ozgurakgun]

I often use minion's static ordering to get the same order in the output. We could stick to that in testing?

I don't mind the sorting either, just a thought.

[ChrisJefferson]

How do I change the solver (can I do it without interactively using the GUI?)

This would also get around the problem I think I'm having where the version of chuffed / savilerow we currently distribute seems to crash if the problem gets optimised away to nothing (which is why the test is currently failing)

[ChrisJefferson]

looking, might need to add another kind of magic which lets us control config from strings instead of GUI?

[ozgurakgun]

The %%conjure line can take arguments. --solver minion on that line should do it?

[ozgurakgun]

I am not a fan of the conjure_settings UI btw, it was a good idea to explore but I don't find myself reaching for that very often at all...

[ChrisJefferson]

I didn't try --solver minion, but I assumed it wouldn't work as it would clash with the --solver which comes from the UI, which is set to chuffed. Will try now.

[ChrisJefferson]

So, I added --solver minion which fixed the crash. It actually didn't solve the solution ordering problem, because my problem is so simple the generated minion instance is completely empty (I guess that is what was upsetting chuffed), and then savilerow doesn't generate solutions in lexicographical order when it creates them itself.

[ozgurakgun]

We could pass --savilerow-options -O0 as well so minion solves it?

[ozgurakgun]

happy to merge this now?

[ChrisJefferson]

Yes, all done I think!

The problem with solution ordering wasn't what I thought at all (I should always check my assumptions). The problem is that conjure-notebook was reading the solution files in a random order!