minion_rs / minion_c++: internal Minion checks throw SIGTERM; we probably want an exception instead?

[niklasdewally]

When testing #230, Felix and I mistakenly used a watchneq constraint with bound variables. This triggered, during search, an internal minion check that SIGTERM-ed the entire executing process:

Cannot use watchneq with BOUND or SPARSEBOUND variables.
Please use neq as a replacement or Please use DISCRETE variables instead.

I think it would be useful for these to throw an exception over FFI instead of terminating the program. An immediate use-case for this would be improperly type-checked constraints inside our cargo tests : currently, the failing test SIGTERMs the entire test suite whereas an exception would be passed upwards and reported as a test failure.

[ChrisJefferson]

I suggest trying to remove (almost) every 'exit' and 'abort' in minion (we can remove almost all of them in both library and main mode).

There is already a 'try/catch' in the 'main' of minion, so I suggest making a new error type ( FatalMinionError ?) which contains a string and return code, replacing all exit and abort with throwing that, then catching them in 'main' and printing out the message (and exiting with the return code).

The only thing to watch out for is places where minion does 'catch(...)', which would catch these objects. Fortunately, I don't think there are any catches in minion which would cause a problem.

After doing that, the library code can just catch the same error objects.

I'm not sure if anyone feels confident enough with their C++-foo to try doing that :) I'm happy to look at this in about a week, if no-one else has tackled it before then.

[niklasdewally]

I'm not sure if anyone feels confident enough with their C++-foo to try doing that :) I'm happy to look at this in about a week, if no-one else has tackled it before then.

I think I can do this; I more or less know where checks happen and where errors need to go.