Possibility of improvement in the AbstractInsertionCalculator class

[arielmoraes]

I've done some tests and it seems that when using the getClass().getSimpleName() method of a failed constraint the CPU usage is higher. It could be changed to a static string field in each implemented constraint. Testing with 250+ jobs, 5 vehicles, 1000 iterations, fast regret, 4 threads, compiled fast matrix costs and removing the getSimpleName reduces the execution time from ~18 seconds to ~14 seconds using an i5 processor.

[oblonski]

You are correct. This is unnecessary overhead. However, the current approach has been the least intrusive. If we changed it, we d need to force the implementor of a contraint to implement getName() also. Do you agree? Or better, what would you suggest to implement it?

[arielmoraes]

I think it is feasible to let the developers implement the getName(), thus the actual available constraints probably suffice the majority of use cases.

[balage1551]

I'd comment this not really knowing the actual talk about. :-) The project moves to Java 8, so there is a new feature for extending an interface without breaking the existing code: default implementation. So the default implementation for getName() could cache the getClass().getSimpleName() into a static field and return that. If the talk was about something else, I'm sorry to interfere.

[oblonski]

👍 Thanks @balage1551. Thus, I am adding it to our v2 branch.

[oblonski]

Todo:

• [ ] extend HardXXXConstraint interface with default getName() method
• [ ] implement .getName() in existing core constraints
• [ ] change to .getName() method in reason tracker