crash with specific problem data --- non-deterministic behaviour

[jmd....@free.fr]

Hi,

I had a few cases where cvxopt would randomly fail or succeed on a given problem. Attached is a pickled set of problem data that should cause a 'math domain error' in misc.jnrm2 (called by misc.update_scaling). Just run the following lines :

import cvxopt

import pickle

with open('cvxopt_fails2.pk', 'r') as f :

    prob = pickle.load(f)

solverOutput = cvxopt.solvers.conelp(prob['c'], prob['G'], prob['h'], prob['dims'], prob['A'], prob['b'])

It would help to avoid those sqrt() math domain error and smoothly exit cvxopt with 'optimal_inaccurate' or similar 

Thanks,

JM