Segfault when using reoptimization and concurrent solving

[waweber]

Describe the bug
A segfault occurs during reoptimization when using concurrent solving.

To Reproduce
Run this program:

from pyscipopt import Model

model = Model()
x = model.addVar(lb=0.0, ub=10.0)
y = model.addVar(lb=0.0, ub=10.0)
z = x + y
model.addCons(y >= 2*x)
model.setObjective(z, "maximize")

model.enableReoptimization()
model.solveConcurrent()

print(f"x = {model.getVal(x)}")
print(f"y = {model.getVal(y)}")
print(f"z = {model.getVal(z)}")

model.freeReoptSolve()

model.addCons(x <= 3)
model.solveConcurrent()

print(f"x = {model.getVal(x)}")
print(f"y = {model.getVal(y)}")
print(f"z = {model.getVal(z)}")

The issue occurs with any combination of solving methods that uses solveConcurrent()

Expected behavior
The program runs the same way as when normal model.optimize() is called.

System

• Fedora Linux 40
• SCIP from master branch
• PySCIPOpt compiled from master branch

Additional context
Not 100% sure if this is from PySCIPOpt or a bug in SCIP itself

[Joao-Dionisio]

Hey @waweber! Apologies for the delay, August-time is always difficult.

Are there any updates on this? You can use solveConcurrent normally if you don't reoptimize, right?

[waweber]

Correct, I can use solveConcurrent without reoptimization, or reoptimize without solveConcurrent, but not both.

Notably, if I use solveConcurrent on the initial optimization, then use the single threaded optimize when reoptimizing, it also segfaults.

[Joao-Dionisio]

Hey @waweber ! Sorry for the nearly 2-year long delay!!!!!!!!!

I digged a bit into the SCIP code, and it turns out that there is a segmentation fault: reoptimization and concurrent solving are really not compatible. There is some cleanup shenanigans going on that subsequent solvers invalidly try to access. It would be pretty cool to be able to do both, sure, but it's seems to be shaping up to be a quite considerable rework. As such, we're thinking of just retunring an error, for now.