Running ALAPScheduleAnalysis directly on a circuit fails

[eendebakpt]

Environment

• Qiskit version: 1.2.4
• Python version: 3.12
• Operating system: Windows

What is happening?

Running the ALAPScheduleAnalysis pass directly on a circuit fails. Running it via a passmanager works.

How can we reproduce the issue?

A minimal example

import numpy as np
from qiskit.circuit import QuantumCircuit
from qiskit.transpiler import PassManager, InstructionDurations
from qiskit.transpiler.passes import ALAPScheduleAnalysis

durations = InstructionDurations( [("reset", None, 10),
     ("cz", None, 120), ("cx", None, 120), ("crx", None, 160),
     ("rx", None, 80), ('ry', None, 80), ('rz', None, 0),  ("measure", None, 30_000)]
)

circ = QuantumCircuit(2)
circ.rx(np.pi/2, 0)
circ.rz(-np.pi/2, 0)

pm = PassManager([ALAPScheduleAnalysis(durations)])
circ = pm.run(circ) # works

circ = QuantumCircuit(2)
circ.rx(np.pi/2, 0)
circ.rz(-np.pi/2, 0)

p=ALAPScheduleAnalysis(durations)
p(circ) # fails

What should happen?

Usually a qiskit pass can be applied directly to a quantum circuit, so it seems reasonable to expect the same from schedule passes.

Any suggestions?

No response

[jakelishman]

ALAPScheduleAnalysis (and all the other built-in scheduling passes) make use of a PassManager feature of requiring that certain passes are run before them. In this case, I think it's the pass that just converts all time units into one unified thing (dt preferentially), but I honestly wouldn't be surprised at all to learn that that pass has a side-effect of initialising the analysis state for the other scheduling passes.

When you __call__ a transpiler pass cold, it skips all the pass-manager infrastructure that manages setting up the executor and handling the requirements, etc, which will be why it's not calling that correctly, and getting out-of-sync. I think the best way to take this is to have BasePass.__call__ simply instantiate a PassManager with self and return the result of running that on the input. That'll also reduce a lot of messy duplicated code, which has clearly got a bit out-of-sync with what PassManager itself does now.