Need a test that ensures barriers are re-combined correctly in transforms.py

[garrison]

See #625 (comment). Such a test is currently lacking, and having 100% test coverage is giving us a false sense of security here.

[lodenrogue]

Any more details you can provide here? An example of what is meant by a barrier or re-combined for those of us new to this repo?

[garrison]

If you take the first tutorial notebook and place a barrier across all qubits at some point in the circuit, there is some special logic to split that barrier into single-qubit barriers so that it will not prevent the circuit from separating into multiple pieces. Then, after separation, those barriers are recombined. #625 fixed this code to work with Qiskit 1.2. It is the stuff in transforms.py that has to do with uuids.

[garrison]

The test should ideally verify the behavior of the seprate_circuit utility function, which is called inside of the partition_problem function, which itself appears in the first tutorial. And the test should live in teat_transforms.py, alongside other tests of the separate_circuit function.

[garrison]

The crux of the problem is that the test suite should fail if the following patch is made to disable (comment out) an important line, but it does not.

diff --git a/circuit_knitting/utils/transforms.py b/circuit_knitting/utils/transforms.py
index 2849d09..715321a 100644
--- a/circuit_knitting/utils/transforms.py
+++ b/circuit_knitting/utils/transforms.py
@@ -135,7 +135,7 @@ def separate_circuit(
             [new_qc.qubits[j] for j in qubits_by_subsystem[label]],
             new_qc.cregs,
         )
-        _combine_barriers(tmp_circ)
+        # _combine_barriers(tmp_circ)
         subcircuits[label] = tmp_circ
 
     return SeparatedCircuits(subcircuits, qubit_map)