_remove_useless_states() can change types to ndarrays

[manuelli]

It seems that if you enter the if statement here then it replaces self.A withself.A with zeros(1,1) which is of type ndarray. This will cause problems later as the library is taking advantage of the fact that * is matrix multiplication not elementwise multiplcation as it would be for ndarrays.

Here is an example of the problem

sys1 = control.StateSpace(0,0,0,0)
sys2 = control.StateSpace(1,0,0,0)
print "type(sys1.A)", type(sys1.A)
print "type(sys2.A)", type(sys2.A)


type(sys1.A) <type 'numpy.ndarray'>
type(sys2.A) <class 'numpy.matrixlib.defmatrix.matrix'>

[roryyorke]

Fixed in #107 (and it's follow-on, #108); above sample now gives

type(sys1.A) <class 'numpy.matrixlib.defmatrix.matrix'>
type(sys2.A) <class 'numpy.matrixlib.defmatrix.matrix'>

[slivingston]

The link to the source code given in the opening post is floating (i.e., it tracks the tip of master branch). For posterity, a link that is fixed with respect to the current master is

python-control/control/statesp.py

Lines 182 to 186 in 58a3ff2

	if all(useless == range(self.states)):
	# All the states were useless.
	self.A = zeros((1, 1))
	self.B = zeros((1, self.inputs))
	self.C = zeros((self.outputs, 1))

.

[roryyorke]

Bug not present in d8b07cb; most likely fixed by #110, which superseded #107 and #108.

[slivingston]

Yes, I agree that it is fixed. I just ran the example from the opening post using the current tip of master branch (32f13bc), and the output is

type(sys1.A) <class 'numpy.matrixlib.defmatrix.matrix'>
type(sys2.A) <class 'numpy.matrixlib.defmatrix.matrix'>

as reported earlier in #82 (comment).