Fix a bug of missing pulse library entry in PulseQobj parsing

[nkanazawa1989]

Summary

When a user tries to get calibration data from Target, it may fail because of partly missing calibration data, e.g.

backend.target["some_gate"][(0, 1)].calibration

This may occur when the backend doesn't support waveform payload, while Qiskit doesn't have definition of the corresponding parametric pulse that the backend uses.

Details and comments

This PR adds exception handling for this case. Namely, None value is returned instead of raising an error.

[qiskit-bot]

One or more of the the following people are requested to review this:

• @Qiskit/terra-core
• @nkanazawa1989

[coveralls]

Pull Request Test Coverage Report for Build 7532191522

Warning: This coverage report may be inaccurate.

We've detected an issue with your CI configuration that might affect the accuracy of this pull request's coverage report.
To ensure accuracy in future PRs, please see these guidelines.
A quick fix for this PR: rebase it; your next report should be accurate.

• 0 of 0 changed or added relevant lines in 0 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+1.8%) to 89.354%

---

Totals
Change from base Build 7463210826:	1.8%
Covered Lines:	59498
Relevant Lines:	66587

---

💛 - Coveralls

[wshanks]

The code changes look okay here, but I would a little more explanation of the issue. I don't know all the details of the qobj code, but I am guessing only a few people at most know it better than me, so if I am having trouble understanding what the issue is here probably most other people will as well.

From the release note, I understand that the problem occurs when the backend does not support a waveform payload while Qiskit does not have a definition for the parametric pulse shape reported. So is the situation that the defaults json includes a parametric_pulse entry with a name that is not in the ParametricPulseShapes Enum? So when the converter tries to convert a parametric pulse, it hits an AttributeError here:

qiskit/qiskit/qobj/converters/pulse_instruction.py

Line 920 in b8d21ae

pulse = ParametricPulseShapes.to_type(instruction.pulse_shape)(**params, name=pulse_name)

because to_type uses getattr:

qiskit/qiskit/qobj/converters/pulse_instruction.py

Lines 80 to 89 in b8d21ae

	def to_type(cls, name: str) -> library.SymbolicPulse:
	"""Get symbolic pulse class from the name.
	Args:
	name: Qobj name of the pulse.
	Returns:
	Corresponding class.
	"""
	return getattr(library, cls[name].value)

and that causes the converter to fall back to _convert_generic here:

qiskit/qiskit/qobj/converters/pulse_instruction.py

Lines 662 to 665 in b8d21ae

	try:
	method = getattr(self, f"_convert_{instruction.name}")
	except AttributeError:
	method = self._convert_generic

If that is the case, I don't think the part about supporting a waveform payload matters? You wouldn't expect a pulse to be in the waveform pulse library if it was reported as a parametric pulse.

[wshanks]

ch does not have to be set as an attribute of PulseQobjInstruction either. Should we just hope that either qubits or ch is set? Every instruction should have a ch set except acquire which should have qubits set so I think for a valid qobj that is reasonable (it is weird that this block assumed qubits would always be defined here rather than ch).

[nkanazawa1989]

Yeah both ch and qubits are optional as you say. For now having either ch and qubits is valid assumption.

[wshanks]

What is the situation where the error is meaningless? This changes the code so that the case where a calibration schedule was unloadable behaves like no calibration was reported by the backend rather than raising an exception. If the user wanted to use the calibration, the user's code will still have a problem. Maybe there is some code that eagerly handles calibrations and can work okay when they are not available?

[nkanazawa1989]

Qobj is parsed when they are called for the first time (i.e. when a user call .calibration of target instruction) and the meaningless error raised when the user called calibration in their program (non-pulse user will never see this). It tries to raise a QiskitError but the error message required .qubits that the play instruction payload cannot provide, and this resulted in AttributeError instead. It was very hard to understand what was happening.

I expect the user implements the code that checks if calibration is available on their side.

[wshanks]

Could we emit a warning (UserWarning?) here with the message from the QiskitError? Here we are treating a calibration loading error as equivalent to no calibration being provided and the user might want to know when there is a loading error.

[nkanazawa1989]

Fair enough. Added an error and test in bf723eb

[wshanks]

Looks good!

[ElePT]

Approving to unblock the merge, thanks a lot!