Add scip::ObjExprhdlr class to objscip

[kkofler]

The intent is to be able to use SWIG director classes to implement expression handlers through bindings, in particular, to add support for expression handlers to JSCIP. That needs the C++ class to build upon.

src/objscip/objexprhdlr.h, src/objscip/objexprhdlr.cpp: New files. C++ wrapper for expression handlers. The files contain the scip::ObjExprhdlr class and the functions SCIPincludeObjExprhdlr, SCIPfindObjExprhdlr, and SCIPgetObjExprhdlr.

src/CMakeLists.txt (objscipsources): Add objscip/objexprhdlr.cpp.
(objscipheaders): Add objscip/objexprhdlr.h.

[svigerske]

@kkofler Are you ok with the current changes to your changes?

[kkofler]

Yes, looks good now. Thank you for your improvements to my contribution.

[svigerske]

Thank you for the contribution!

[kkofler]

Unfortunately, now that I am finally far enough with my Java bindings to be able to really test this, I have found an annoying issue, but I am not sure about the proper fix: The problem is: The Java ExpressionHandler class wants to call SCIPincludeObjExprhdlr in the constructor (in any constructor, even the copy constructor) so that it can remember the corresponding C SCIP_EXPRHDLR * pointer. As a result, the implementation of the clone operation (called by the ObjExprhdlr implementation of copyhdlr) is also going to indirectly call SCIPincludeObjExprhdlr from the Java code.

However, the C++ code also tries to directly call SCIPincludeObjExprhdlr in copyhdlr:

      /* call include method of expression handler object */
      SCIP_CALL( SCIPincludeObjExprhdlr(scip, newobjexprhdlr, TRUE) );

That was not my idea: At least ObjConshdlr does the exact same thing, I just adapted that code here.

As a result, the expression handler is going to be included twice, with both (duplicate) SCIP_EXPRHDLRs pointing to the same ObjExprhdlr. This eventually results in a double-free, and then all hell may break loose. (In my tests, I ended up running into a mysterious assertion failure that turned out to be caused by memory corruption: the second free was use-after-free-ing a SCIP * pointer that was already overwritten, and complaining that it did not match the passed SCIP *. But the behavior is undefined, so it can fail in many other ways.)

I tried simply commenting out that SCIP_CALL( SCIPincludeObjExprhdlr(scip, newobjexprhdlr, TRUE) ); line in objexprhdlr.cpp (pushing the reponsibility of calling that method down to the clone implementation) and that fixed it for me. However, then we should also change that behavior in objconshdlr.cpp (and any other objscip classes that do the same, if any) for consistency.

Fixing this in the JSCIPOpt Java code is not easy, also because, as I already mentioned, SCIPincludeObjExprhdlr is what gives me the SCIP_EXPRHDLR * pointer.

So I am not sure what the proper fix is here.