[gecode-users] Parameter/Options passing to engines through script

[Luca Di Gaspero]

Hi all,

I am trying to re-engineer some piece of code we wrote for enabling Large Neighborhood Search within Gecode and I am facing the problem (which has also some general sense) of passing specific parameters (or options objects) to an engine that has to be managed through the Script<Space>::run<…>() execution method.

Since I would like to contribute the LNS code to the Gecode community, before undertaking modifications in my Gecode working copy I would like to agree it with Gecode maintainers since it will have a (I would say minor) impact on the framework. 

Other solutions to the problem from more "knowledgeable users" are, of course, welcome. :-)

Coming to the problem, in detail, the specific option object that is passed to the run method is dispatched in the Script::run function and a brand new Search::Option object is sent to the actual engine. Unfortunately this means that the specific engine could not have access to its specific options and I found it a general limitation.
Of course it is possible to overcome this limits with workarounds, but they will somehow break the general framework philosophy.

To be generic enough, my proposal is to add a "Gecode::Options* parameters" attribute to the Search::Options class in order to allow sending this specific information to the engine. This variable should be set in the Script::run(Options, Script) dispatching, e.g., by invoking some "parameters()" function on the "Gecode::Options& option" object passed to run.

Just sketching the "pseudo diff" of the tentative modification it will be something like this:

At search.hh around line 210 add the line
Gecode::Options* parameters;

At script.hpp around line 300 add the line
so.parameters = o.parameters();

And, at options.hpp around line 148 add the lines

inline
Options* parameters() {
  return NULL; // for default no specific option is passed to the engine
}

This way, the engine has access to parameters and it can cast it to the opportune parameter class.

Thanks in advance for comments and/or suggestions.

All the best,

Luca