Copy and deepcopy of a non-parametrized operation only references `op.matrix`

Issue description

When copying a non-parametrized operation, e.g. CNOT, the matrix is only copied as a reference. This is the case when using both copy.copy and copy.deepcopy. The following:

import pennylane as qml
import copy

op = qml.CNOT(wires=[0,1])
op2 = copy.copy(op)
op3 = copy.deepcopy(op)

print("Before:")
print(op2.matrix)
print(op3.matrix)

op.matrix[0, 0] = 9

print("\nAfter:")
print(op2.matrix)
print(op3.matrix)

outputs:

Before:
[[1 0 0 0]
 [0 1 0 0]
 [0 0 0 1]
 [0 0 1 0]]
[[1 0 0 0]
 [0 1 0 0]
 [0 0 0 1]
 [0 0 1 0]]

After:
[[9 0 0 0]
 [0 1 0 0]
 [0 0 0 1]
 [0 0 1 0]]
[[9 0 0 0]
 [0 1 0 0]
 [0 0 0 1]
 [0 0 1 0]]

• Expected behavior: (What you expect to happen) I would expect at least deepcopy to make a deep copy of all the internal data.

• Actual behavior: (What actually happens) Only references are being made to the internally saved matrix.

• Reproduces how often: (What percentage of the time does it reproduce?) Every time

[albi3ro]

For the CNOT case, the matrix is a class variable, so even independent objects have correlated matrix values:

op1 = qml.CNOT(wires=(0,1))
op2 = qml.CNOT(wires=(2,3))
op1.matrix[0,0] = 9
print(op1.matrix)
print(op2.matrix)

array([[9, 0, 0, 0],
       [0, 1, 0, 0],
       [0, 0, 0, 1],
       [0, 0, 1, 0]])
array([[9, 0, 0, 0],
       [0, 1, 0, 0],
       [0, 0, 0, 1],
       [0, 0, 1, 0]])

[josh146]

Would it be good to add a warning if a user attempts to deepcopy an operation? Especially if other things might not work as well when doing that?

I lean towards not, since we can’t properly distinguish between PL itself performing deepcopies or users performing deepcopies. The core itself might create deepcopies, which would result in a cascade of warnings appearing to unexpected users 😆