Segmentation fault when running LPs repeatedly

[frogeyedpeas]

Hello I am using a MAC OS Big Sur Version 11.3.1

When I run the solver multiple times SOMETIMES it segmentation faults, I dont know if this is because of qsopt_ex or because of the cython library. Just copy the code below into a file called reproducing_seg_fault.py and run the program like 20 or so times consecutively. One of the times will result in a segfault (see here)

Code here:

import qsoptex

import faulthandler
faulthandler.enable()

#import logging
#logging.basicConfig(level=logging.NOTSET)

class Polytope:
    def __init__(self, A, b):
        self.A = A
        self.b = b
        self.status = None
        """
            The status variable carries the status of a most recent solve, if it is set to None then an LP run has not occurred yet 
        """
        self.p = None

    def dimension(self):
        return len(self.A[0])
    
    def inequality_count(self):
        return len(self.A)

    #given an objective array c we attempt to maximize it and return the status, surprising the "c" tells us what the dimension of the problem is 
    def maximize(self, c, flag_integers=True): 
        self.p = qsoptex.ExactProblem()
        for var_number in range(len(c)):
            self.p.add_variable(name='x' + str(var_number), 
                objective=c[var_number], 
                lower=0, upper=1)

        for constraint_number in range(self.inequality_count()):
            self.p.add_linear_constraint(qsoptex.ConstraintSense.LESS, 
                dict(('x' + str(var_number), self.A[constraint_number][var_number]) for var_number in range(len(c))), 
                rhs = self.b[constraint_number], 
                name='c' + str(constraint_number))

        self.p.set_objective_sense(qsoptex.ObjectiveSense.MAXIMIZE)
        self.p.set_param(qsoptex.Parameter.SIMPLEX_DISPLAY, 1)
        print("before seg fault: ")
        self.status = self.p.solve()
        print("after seg fault: ")


    

if __name__ == '__main__':

    #running one at a time never results in a segfault
    #running both of these repeatedly results in a segfault 

 
    A_matrix =  [[1037066, 2074132, 1648264, 796528, 1593056, 686112, 1372224, 244448, 488896, 977792, 1955584, 1411168, 322336, 644672, 1289344, 78688, 157376, 314752, 629504, 1259008], [-1037066, -2074132, -1648264, -796528, -1593056, -686112, -1372224, -244448, -488896, -977792, -1955584, -1411168, -322336, -644672, -1289344, -78688, -157376, -314752, -629504, -1259008]]
    b_matrix =  [2463098, -2463098]
    c_matrix =  [0, -1, 1, -1, 1, -1, 0, 0, 0, 0, -1, 0, -1, -1, 0, 1, 0, 1, 1, -1]

    test_polyhedron = Polytope(A_matrix, b_matrix)
    
    test_polyhedron.maximize( c_matrix)

    A_matrix =  [[1037066, 2074132, 1648264, 796528, 1593056, 686112, 1372224, 244448, 488896, 977792, 1955584, 1411168, 322336, 644672, 1289344, 78688, 157376, 314752, 629504, 1259008], [-1037066, -2074132, -1648264, -796528, -1593056, -686112, -1372224, -244448, -488896, -977792, -1955584, -1411168, -322336, -644672, -1289344, -78688, -157376, -314752, -629504, -1259008]]
    b_matrix =  [2463098, -2463098]
    c_matrix =  [0, -1, 1, -1, 1, -1, 0, 0, 0, 0, -1, 0, -1, -1, 0, 1, 0, 1, 1, -1]

    test_polyhedron = Polytope(A_matrix, b_matrix)

    test_polyhedron.maximize( c_matrix)

[frogeyedpeas]

here is my python version Python 3.8.2 (default, Jun 8 2021, 11:59:35) [Clang 12.0.5 (clang-1205.0.22.11)] on darwin and the shell i'm using is zsh

[frogeyedpeas]

So I have noticed that if I don't have two systems back to back I cannot get the segfault to occur. This suggests to me that some kind of caching routine (maybe reading from a bad cache) is causing the segfaults in the first place.

[frogeyedpeas]

So with some inspection it appears that line 481 in the file qsoptex.pyx is to blame

r = cqsoptex.QSexact_solver(self._c_qsdata, NULL, NULL, NULL, cqsoptex.PRIMAL_SIMPLEX, &status)

I'm out of my depth here but I can follow that line of code the to the file csqoptex.pxd line 96

int QSexact_solver(mpq_QSdata* problem, mpq_t* const x, mpq_t* const y, QSbasis* const basis, int simplexalgo, int* status)

and this basically appears to wrap a direct call to the qsoptex library in exact.h

[frogeyedpeas]

The segfault is caused by line 721 here: https://github.com/jonls/qsopt-ex/blob/master/qsopt_ex/exact.c particulary just accessing the qslp->colnames structure see below:

It appears that the colnames pointer is being accessed like an array on elements which do not cast like strings