Potential conshdlr conslock life cycle problem

[Koen-Ligthart]

Describe the bug
Cleaning up the SCIP model results in a ReferenceError in conslock of a constraint handler.

To Reproduce
Construct a single file with the following contents and execute it:

import pyscipopt as ps


class DummyConshdlr(ps.Conshdlr):

    def createCons(self, vars, model):
        cons = self.model.createCons(self, "dummy")
        # cons.data = {"vars": vars, "model": model} # alt 1
        cons.data = {"vars": vars} # alt 2
        return cons

    def conscheck(self, constraints, solution, check_integrality, check_lp_rows, print_reason, completely, **results):
        return {"result": ps.SCIP_RESULT.FEASIBLE}

    def consenfolp(self, constraints, n_useful_conss, sol_infeasible):
        return {"result": ps.SCIP_RESULT.FEASIBLE}

    def conslock(self, constraint, locktype, nlockspos, nlocksneg):
        model = self.model
        for var in constraint.data["vars"]:
            model.addVarLocks(var, nlockspos + nlocksneg, nlockspos + nlocksneg)


def test():

    model = ps.Model()
    vars = [ model.addVar() ]

    conshdlr = DummyConshdlr()
    model.includeConshdlr(conshdlr, "dummy", "dummy")
    model.addPyCons(conshdlr.createCons(vars, model))

    model.optimize()
    

test()

This then first successfully solves the problem, but also results in an unexpected ReferenceError. See the following log:

feasible solution found by trivial heuristic after 0.0 seconds, objective value 0.000000e+00
presolving:
presolving (1 rounds: 1 fast, 0 medium, 0 exhaustive):
 0 deleted vars, 0 deleted constraints, 0 added constraints, 0 tightened bounds, 0 added holes, 0 changed sides, 0 changed coefficients
 0 implications, 0 cliques, 0 implied integral variables (0 bin, 0 int, 0 cont)
transformed 3/3 original solutions to the transformed problem space
presolving solved problem
Presolving Time: 0.00

SCIP Status        : problem is solved [optimal solution found]
Solving Time (sec) : 0.00
Solving Nodes      : 0
Primal Bound       : +0.00000000000000e+00 (3 solutions)
Dual Bound         : +0.00000000000000e+00
Gap                : 0.00 %
Traceback (most recent call last):
  File "XXXXXXXXX", line 21, in conslock
    model.addVarLocks(var, nlockspos + nlocksneg, nlockspos + nlocksneg)
    ^^^^^^^^^^^^^^^^^
ReferenceError: weakly-referenced object no longer exists
Exception ignored in: 'pyscipopt.scip.PyConsLock'
Traceback (most recent call last):
  File "XXXXXXXXX.py", line 21, in conslock
    model.addVarLocks(var, nlockspos + nlocksneg, nlockspos + nlocksneg)
ReferenceError: weakly-referenced object no longer exists
[cons.c:7592] ERROR: Error <0> in function call
[prob.c:1602] ERROR: Error <0> in function call
[prob.c:458] ERROR: Error <0> in function call
[scip_solve.c:2033] ERROR: Error <0> in function call
[scip_solve.c:3511] ERROR: Error <0> in function call
[scip_prob.c:847] ERROR: Error <0> in function call
[scip_general.c:412] ERROR: Error <0> in function call
Traceback (most recent call last):
  File "src/pyscipopt/scip.pxi", line 319, in pyscipopt.scip.PY_SCIP_CALL
Exception: SCIP: unspecified error!
Exception ignored in: 'pyscipopt.scip.Model.__dealloc__'
Traceback (most recent call last):
  File "src/pyscipopt/scip.pxi", line 319, in pyscipopt.scip.PY_SCIP_CALL
Exception: SCIP: unspecified error!

Expected behavior
I expect the code to run without such error and produce a log identical to the situation where alt 1 is uncommented and alt 2 is commented, which yields the same log, except for the part after Gap:, which is not present with alt 1 uncommented. Additionally, I do not expect a different result from the code depending on whether alt 1 or alt 2 is commented.

System

• OS: Windows
• Version 11 Enterprise
• SCIP version 10.0.0
• Installed pyscipopt version 6.0.0 through pip

[Joao-Dionisio]

Hey @Koen-Ligthart , thanks for the issue! Some months ago I was working on something I believe fixes this. See #1194, locally I can run your example.

There are some design questions to decide first, so it will take a little bit to make it into master, and a bit longer to make it to a release. But you can try compiling the version of the PR, and check if your application now works.