[gecode-users] View selection approach in Gecode

[Lars Otten]

Hi everyone,

I've started looking into random view selection for branching and hence
examined the way the next view to be branched on is selected in Gecode.
After comprehending the source code (kernel/branching.icc and int/branch*) I
can't help wondering why it's done the way it currently is.

To be specific: Why did you choose this 'incremental' approach, handing one
view after another to the view selection class instance, instead of just
passing the ViewArray right from the start? The latter would be comparable
to the value selection process, where the whole view is passed and not one
value after another...

As I see it, I can just modify this scheme (or rather add my own next to it)
since this 'incremental' approach won't work for my purposes; nevertheless I
think I'm missing something here and I'd like to understand your design
before I start implementing my random view selection so I can avoid
potential pitfalls...

Thanks in advance (once more),
/Lars