Seemingly incorrect semantics given when using `mem_swap` with `state_result` in Stim simulation

[CalMacCQ]

Unless I'm missing something, mem_swap should have the same semantics as "A swap performed with three cnots". I've found an example for which this does not seem to be the case.

from guppylang import guppy
from guppylang.std.debug import state_result
from guppylang.std.quantum import discard_array, qubit, cx, h
from guppylang.std.builtins import mem_swap, array

from selene_sim import Stim, build

seed = 123


@guppy
def swap_with_cx(q0: qubit, q1: qubit) -> None:
    cx(q0, q1)
    cx(q1, q0)
    cx(q0, q1)

@guppy
def main() -> None:
    qs = array(qubit() for _ in range(2))
    h(qs[1])

    #swap_with_cx(qs[0], qs[1])
    mem_swap(qs[0], qs[1])

    state_result("RESULT", qs)

    discard_array(qs)


instance = build(main.compile())
seeded_stim_instance = Stim(random_seed=seed)
output = instance.run(simulator=seeded_stim_instance, n_qubits=2)
states_dict = seeded_stim_instance.extract_states_dict(output)
stab_list = states_dict["RESULT"].get_reduced_stabilizers()

print(stab_list) # print out the tableau

If I do the swap with mem_swap, I get...

+_Z
+X_

However if I comment out mem_swap and instead use swap_with_cx, I get

+X_
+_Z

[CalMacCQ]

I no longer think this is a bug. Seems I wasn't getting the full picture by just looking at the printout of StabilizerList.

If I make sure use SeleneStimState.get_reduced_stabilizers() rather than SeleneStimState.unreduced_stabilizer_list the appropriate qubit ordering seems to be tracked. See the usage of SeleneStimState.specified_qubits below.

instance = build(main.compile())
seeded_stim_instance = Stim(random_seed=seed)
output = instance.run(simulator=seeded_stim_instance, n_qubits=2)
states_dict = seeded_stim_instance.extract_states_dict(output)
state = states_dict["RESULT"]

stab_list = state.get_reduced_stabilizers()

print("qubit ordering:", state.specified_qubits)
print(stab_list) # print out the tableau

If we run this with the mem_swap we get

qubit ordering: [1, 0]
+_Z
+X_

Now if we do the swap with cx gates instead we get

qubit ordering: [0, 1]
+X_
+_Z

The two tableaux are equivalent if we take the qubit ordering into account.