sourceModel loses parameters and is usually slower than original model

[Joao-Dionisio]

Reported in #918.

new_scip = Model(sourceModel=orig_scip) loses 226 parameters including dual presolvers, some heuristics. Solving new_scip is usually slow than solving orig_scip with same random seed. The new copy function can full copy orig_scip problem except solution storage.

Also need to find out why it loses some of the original model's parameters.

[DominikKamp]

Actually, SCIPcopy() does not produce a SCIP clone but is only guaranteed to hold an equivalent model if valid, since some plugins just do not implement the copy callback (I believe some of them by accident since the callback is non-fundamental), so the cause lies in SCIP.

[Joao-Dionisio]

In your opinion as the Issue master, should I create an issue on the gitlab, Dominik? :)

[DominikKamp]

This would be nice. Then we can try to add the copy callbacks and see what happens (as the issue indicates, this might affect performance of subscips). For example the benders parameters get lost as the documentation of SCIPcopyPlugins() notes that SCIPcopyBenders() must be called separately if desired.

[Joao-Dionisio]

As this is being actively worked on in SCIP, I'm closing the PySCIPOPt issue.

[DominikKamp]

Okay, maybe it should be mentioned here that the slowdown is probably due to propagator symmetry not having a copy callback implemented yet.