Regression: instruction `label` must now be a string

[garrison]

Environment

• Qiskit version: main
• Python version:
• Operating system:

What is happening?

I am witnessing a CI failure in the circuit-knitting-toolbox that very likely has something to do with the merging of #12459. (We run CI against Qiskit main on a cron job every day.)

Here is one such error:

    def _split_barriers(circuit: QuantumCircuit):
        """Mutate an input circuit to split barriers into single qubit barriers."""
        for i, inst in enumerate(circuit):
            num_qubits = len(inst.qubits)
            if num_qubits == 1 or inst.operation.name != "barrier":
                continue
            barrier_uuid = uuid4()
    
            # Replace the N-qubit barrier with a single-qubit barrier
>           circuit.data[i] = CircuitInstruction(
                Barrier(1, label=barrier_uuid), qubits=[inst.qubits[0]]
            )
E           TypeError: 'UUID' object cannot be converted to 'PyString'

How can we reproduce the issue?

from uuid import uuid4
from qiskit.circuit import Barrier, CircuitInstruction

CircuitInstruction(Barrier(1, label=uuid4()), (0,))

What should happen?

It should work like before.

Any suggestions?

It is probably sufficient to call str(...) on any object before passing to Rust, and to add a test.

[jakelishman]

An instruction label should always have been a string, and certainly for Barrier it's been documented that way since 2022.

[jakelishman]

(btw, you're quite likely to see a lot more problems in the near future, including performance and memory regressions on main while we make large-scale changes to the internals. Everything should be fast and memory-efficient by the time of 1.2, but the intermediate states will very likely temporarily regress.)

[garrison]

An instruction label should always have been a string, and certainly for Barrier it's been documented that way since 2022.

Thanks for the reply, Jake. We can work around this pretty easily in ckt, obviously, and this is probably something I should have noticed in code review, except I was unfamiliar with uuid4() so didn't realize it returns an object, not a string. The reason I raised an issue here is because I suspect there is a chance this specific change may break other code in the wild, so perhaps it is worth converting the label to a str, and making a DeprecationWarning if you really don't want users to ever do it. But it's up to you as Qiskit maintainers. I am just trying to do my part to contain the fallout across the ecosystem from these fundamental changes.

[garrison]

Jake, it looks like if qiskit were modified as I suggest to convert the label to str, that would still have broken us, since ckt later checks isinstance(label, UUID). So I think you have it correct right now: the code I provided should just lead to an error. I am going to close this as "won't fix."

[garrison]

(Also, sorry I was in such a rush to open this issue. I wanted to do so before I forgot, but it turns out my mental model of what was happening was wrong still.)

[jakelishman]

Sorry, I dropped responding to this at the time.

I am just trying to do my part to contain the fallout across the ecosystem from these fundamental changes.

I meant to say: thank you very much for this - it is super helpful. Sorry for being curt before.