Calling `synth(registerless=False)` on a classical function synth labels the qubits wrongly

[1ucian0]

Environment

• Qiskit Terra version: 0.23.0.dev0+dcd7bea
• Python version: 3.9

What is happening?

In https://github.com/Qiskit/qiskit-terra/pull/9337#discussion_r1067055542 @Cryoris noticed that classical_function synthesis labels the qubits in the reversed order.

How can we reproduce the issue?

:

from qiskit.circuit.classicalfunction import classical_function
from qiskit.circuit.classicalfunction.types import Int1

@classical_function
def grover_oracle(a: Int1, b: Int1, c: Int1) -> Int1:
    return (a and b and not c)

quantum_circuit = grover_oracle.synth(registerless=False)
print(quantum_circuit.draw())

     c: ──■──
          │  
     b: ──■──
          │  
     a: ──o──
        ┌─┴─┐
return: ┤ X ├
        └───┘

What should happen?

in that example, the open control is on c, not on a.

Any suggestions?

No response

[1ucian0]

I'm not sure here what's expected. It could be either:

     a: ──■──
          │  
     b: ──■──
          │  
     c: ──o──
        ┌─┴─┐
return: ┤ X ├
        └───┘

or

     c: ──o──
          │  
     b: ──■──
          │  
     a: ──■──
        ┌─┴─┐
return: ┤ X ├
        └───┘

Tweedledum seems to prefer the first option.

[Cryoris]

What do we expect the function signature to look like: def oracle(q_0, q_1, q_2) or def oracle(q_2, q_1, q_0)? It seems the "Qiskit way" would be q_2, q_1, q_0 so the top labelling in your example would seem correct to me 🙂

[jakelishman]

You could argue either way - def oracle(q_0, q_1, q_2) satisfies *args where args[0] == q_0 and so on, so the indexing matches the actual usage, whereas def oracle(q_2, q_1, q_0) just looks kind of like how bitstrings are printed. The former is probably better, because it matches the semantics when passing qargs in a list to things like QuantumCircuit.append.