Second tutorial fails with development version of Qiskit

[garrison]

The second tutorial fails in CI with the development version of Qiskit.

It must be a recent change causing this, because the first cron-job CI failure was today. I am guessing Qiskit/qiskit#13482 is the culprit. EDIT: I have since verified this.

Example failure at https://github.com/Qiskit/qiskit-addon-cutting/actions/runs/12060889091/job/33635443761#step:6:407

[mtreinish]

I'm pretty sure that error can only be raised by the basis translator or equivalence library. Tracing through the code it will come up if a parameter expression being assigned a new value (either a float or new param) in an equivalence rule isn't found in the circuit from the rule. Looking through Qiskit/qiskit#13482 my guess is that it's probably a Python side shared ref issue. The PR substitutes the param in place on cached instances of standard gates on parameter assignment but if that python object was added to more than one circuit the param from the pycache is going to mismatch with the parameter table of the other circuit. This could be coming up somehow in the basis translator/equivance library after running it multiple times somehow. But this is just a guess. My gut feeling in Qiskit/qiskit#13482 was to invalidate the py cache instead of trying to update it, but the update seemed to work and all the tests passed so I figured it was fine. It might be worth trying to just clear the py cache on parameter assignment instead to see if that fixes the issue. @Cryoris can you give that a try?

[Cryoris]

FYI this is tracked in Qiskit/qiskit#13504. We found a potential fix, which is not included in 1.3.0 yet though. Until then, this can be resolved by turning off disabling parallel processing in Qiskit, i.e.

export QISKIT_PARALLEL=FALSE

or by disabling caching of Python gates in Rust, by

export QISKIT_NO_CACHE_GATES=1
python setup.py build_rust --inplace --release