python-mip SEGFAULT when adding lazy constraint generator to simple model in CBC

output.log

Welcome to the CBC MILP Solver 
Version: devel 
Build Date: Nov 15 2020 

Starting solution of the Linear programming relaxation problem using Primal Simplex

Clp0000I Optimal - objective value 20
Clp0032I Optimal objective 20 - 0 iterations time 0.002, Idiot 0.00

Starting MIP optimization


ERROR while running Cbc. Signal SIGSEGV caught. Getting stack trace.
/home/rs/in/env/lib/python3.7/site-packages/mip/libraries/lin64/libCbcSolver.so(_Z15CbcCrashHandleri+0x14c) [0x7fd62b6fa83c]
/lib/x86_64-linux-gnu/libc.so.6(+0x37840) [0x7fd62c726840]
./libClp.so.0(_ZN8ClpModel12emptyProblemEPiPdb+0x226) [0x7fd62aaab466]
./libClp.so.0(_ZN10ClpSimplex7startupEii+0x231) [0x7fd62ab272e1]
./libClp.so.0(_ZN14ClpSimplexDual12startupSolveEiPdi+0x2d) [0x7fd62ab3d6dd]
./libClp.so.0(_ZN14ClpSimplexDual4dualEii+0x1d8) [0x7fd62ab41378]
./libClp.so.0(_ZN10ClpSimplex4dualEii+0x65) [0x7fd62ab18c05]
./libOsiClp.so.0(_ZN21OsiClpSolverInterface6crunchEv+0x637) [0x7fd62adff097]
./libOsiClp.so.0(_ZN21OsiClpSolverInterface7resolveEv+0x13a3) [0x7fd62ae0e723]
./libCbc.so.0(_ZN8CbcModel7resolveEP18OsiSolverInterface+0xbd) [0x7fd62b3db64d]
./libCbc.so.0(_ZN8CbcModel7resolveEP11CbcNodeInfoiPdS2_S2_+0x14c) [0x7fd62b3dbb3c]
./libCbc.so.0(_ZN8CbcModel14branchAndBoundEi+0x840) [0x7fd62b3f8a50]
/home/rs/in/env/lib/python3.7/site-packages/mip/libraries/lin64/libCbcSolver.so(_Z8CbcMain1iPPKcR8CbcModelPFiPS2_iER19CbcSolverUsefulData+0xbb31) [0x7fd62b710321]
/home/rs/in/env/lib/python3.7/site-packages/mip/libraries/lin64/libCbcSolver.so(Cbc_solve+0x11e3) [0x7fd62b6a6333]
/home/rs/in/env/lib/python3.7/site-packages/cffi.libs/libffi-806b1a9d.so.6.0.4(ffi_call_unix64+0x4c) [0x7fd62bdcadec]
/home/rs/in/env/lib/python3.7/site-packages/cffi.libs/libffi-806b1a9d.so.6.0.4(ffi_call+0x165) [0x7fd62bdc9f55]
/home/rs/in/env/lib/python3.7/site-packages/_cffi_backend.cpython-37m-x86_64-linux-gnu.so(+0x1edf6) [0x7fd62bfecdf6]
python(_PyObject_FastCallKeywords+0x4eb) [0x5ce4ab]
python() [0x53ed81]
python(_PyEval_EvalFrameDefault+0x49ba) [0x54653a]
python(_PyEval_EvalCodeWithName+0x252) [0x53f732]
python(_PyFunction_FastCallKeywords+0x482) [0x5cd982]
python(_PyEval_EvalFrameDefault+0x732) [0x5422b2]
python(_PyEval_EvalCodeWithName+0x252) [0x53f732]
python(_PyFunction_FastCallKeywords+0x482) [0x5cd982]
python(_PyEval_EvalFrameDefault+0x732) [0x5422b2]
python(_PyEval_EvalCodeWithName+0x252) [0x53f732]
python(PyEval_EvalCode+0x23) [0x541b73]
python() [0x62d09e]
python(PyRun_FileExFlags+0x97) [0x62d157]
python(PyRun_SimpleFileExFlags+0x17f) [0x62dfcf]
python() [0x6525ae]
python(_Py_UnixMain+0x2e) [0x65290e]
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xeb) [0x7fd62c71309b]
python(_start+0x2a) [0x5d531a]

script.py

#
# maximize  x + y
# s.t.      (x-4)^2 + (y-2)^2 - 9 <= 0
#           0 <= x, y <= 10
#           y integer
#
# using a lazy constraint callback to generate gradient cuts
#

import mip

def g(x, y):
    return (x - 4)**2 + (y - 2)**2 - 3**2

def gradient_cut(x0, y0, x, y):
    dg = 2 * x0 - 8, 2 * y0 - 4
    return g(x0, y0) +  dg[0] * (x - x0) + dg[1] * (y - y0) <= 0

class GradientCutGenerator(mip.ConstrsGenerator):
    def __init__(self, x, y, TOL=1e-3):
        self.x = x
        self.y = y
        self.TOL = TOL

    def generate_constrs(self, model: mip.Model, depth: int = 0, npass: int = 0):
        x = model.translate(self.x)
        y = model.translate(self.y)

        violation = g(x.x, y.x)
        print(f'### sol = ({x.x:.2f}, {y.x:.2f}), viol = {violation:.2f}')
        if violation < self.TOL:
            return

        # Generate cut
        model.add_lazy_constr(gradient_cut(x.x, y.x, x, y))

model = mip.Model(solver_name="CBC")
model.verbose = 1

x = model.add_var("x", var_type=mip.CONTINUOUS, lb=0, ub=10)
y = model.add_var("y", var_type=mip.INTEGER, lb=0, ub=10)
model.objective = mip.maximize(x + y)

model.lazy_constrs_generator = GradientCutGenerator(x, y)

status = model.optimize()

script2.py

import mip

class CutGen(mip.ConstrsGenerator):
    pass

model = mip.Model(solver_name="CBC")
x = model.add_var("x", var_type=mip.BINARY)
model.lazy_constrs_generator = CutGen()
status = model.optimize()

versions

Python 3.7.3
mip 1.13.0
Linux t440p 4.19.0-14-amd64 #1 SMP Debian 4.19.171-2 (2021-01-30) x86_64 GNU/Linux