Do we want to do a callback to primal heuristics when in PHASE_PRICE1?

[spoorendonk]

The user callback to do primal heuristics is called here

Dip/Dip/src/DecompAlgo.cpp

Line 2198 in cf4f848

m_app->APPheuristics(m_xhat, getOrigObjective(), m_xhatIPFeas);

This also done when m_phase == PHASE_PRICE1 and m_xhat therefore not necessarily is complete, i.e., a valid LP relaxation. The user would expect the m_xhat to feasible LP relaxation I guess?

I suggest to only call APPheuristics when m_phase != PHASE_PRICE1.

Perhaps also have a parameter to control the frequency of calls so not to call in every iteration. For now this is for the user to control manually.

[tkralphs]

I guess it does make sense only to do primal heuristics in Phase 2. In branch-and-cut, there are primal heuristics that can take an infeasible point as an input, but I'm not sure it really makes sense here.

And yes, parameter to control the frequency would be good. It can always be set to 1 to get the current behavior.

[spoorendonk]

I will do a PR