SVM in sequence segmenation fault

When you try to train several SVMs in sequence we have segmentation fault.

Follow a code that reproduces the issue (2 SVMs in sequence):

import bob.learn.libsvm
import numpy
numpy.random.seed(10)

for i in range(2):
    pos = numpy.random.normal(0., 1, size=(100, 2))
    neg = numpy.random.normal(1., 1, size=(100, 2))
    data = [pos, neg]

    trainer = bob.learn.libsvm.Trainer()
    trainer.kernel_type = 'LINEAR'
    trainer.cost = 1
    trainer.train(data)

However, if you free the data variables in the end of loop for, everything runs fine

import bob.learn.libsvm
import numpy
numpy.random.seed(10)

for i in range(2):
    pos = numpy.random.normal(0., 1, size=(100, 2))
    neg = numpy.random.normal(1., 1, size=(100, 2))
    data = [pos, neg]

    trainer = bob.learn.libsvm.Trainer()
    trainer.kernel_type = 'LINEAR'
    trainer.cost = 1
    trainer.train(data)

    del data
    del pos
    del neg   

I'm still debugging, but it seems that the issue is somewhere here in the bindings (https://gitlab.idiap.ch/bob/bob.learn.libsvm/blob/master/bob/learn/libsvm/trainer.cpp#L616)