Prepare for Qiskit 2.0 by addressing removed classes and methods by wshanks · Pull Request #1526 · qiskit-community/qiskit-experiments

wshanks · 2025-03-05T05:18:26Z

This change prepares for Qiskit 2.0 by guarding or replacing references to objects removed in Qiskit 2.0.

Detailed list of changes:

Compatibility changes:

Guard references to BackendV1. BackendV1 was removed in Qiskit 2.0. There were only a few remaining references to BackendV1 after the previous removal of pulse support, so it was not hard to keep the remaining support for now but it should be removed when Qiskit 1 is no longer supported.
Update usage of qiskit.result.Result.header to expect a dict (Qiskit 2) as well as a class (Qiskit 1). ExperimentData gets the circuit metadata from the header (by convention the metadata is copied into the result and ExperimentData relies on this).
Add classes MeasLevel and MeasReturnType to enumerate the expected measurement level (kerneled or classified) and return type (averaged or not). Previously, private classes of Qiskit were inadvertently used for these purposes but they were removed in Qiskit 2.0.
Remove null default inst_map transpile options from experiments. These should have been removed along with the rest of pulse support.
Update tests to use GenericBackendV2 or a custom BackendV2 in place of removed FakeOpenPulse2Q and Fake5QV1 backends from Qiskit.
Remove suppression of Qiskit deprecation warnings from the tests.
Remove stray references to ProviderV1 in tests.
Update the saved QuantumVolume results for Qiskit 2.0's change in the quantum volume algorithm.
Vendor deprecation decorators to prepare for their removal from Qiskit's public API (see here). At first it appeared that these methods would be changed late in the Qiskit 2.0 cycle but the change was pushed back in time. Still the plan is to make the functions private, so we need to stop using them.

Other changes

Provide a more specific warning in LocalReadouMitigator if it is used with a backend without a properties() method or with a backend that does not the required readout error properties.
Fix a performance issue in QuantumVolume. When Qiskit Aer is installed, QuantumVolume uses AerSimulator to simulate the outcomes of the quantum volume circuits to determine the heavy output probabilities. In its default instantiation, AerSimulator regenerates its Target every time it is accessed and this generation takes an appreciable amount of time. Because the target was being accessed separately for every circuit, this overhead could accumulate to over a minute in the standard 100 circuit experiment. This overhead was reduced by processing all the circuits in one pass.
Switch FineFrequency tests to use T2HahnBackend in place MockIQBackend since it provides a more realistic backend emulation.
Fix the stacklevel on the deprecation warning for RestlessNode.

eliarbel · 2025-03-05T12:35:21Z

Thanks @wshanks for including my BackendV1 Removal commit here. This is still in draft and I'm sure you'll cover everything eventually, but dropping a couple things just as a reminder:

qiskit.qobj has been removed. So MeasLevel and MeasReturnType should now be imported from qiskit.result. No QobjExperimentHeader and PulseQobjInstruction anymore.
qobj_id is not required anymore when initializing Result (though it can be passed via *kwargs).

wshanks · 2025-03-05T14:09:55Z

Thanks, @eliarbel. I have been keeping an eye on Qiskit/qiskit#13793, but I have only been testing against Qiskit main so there are things being removed there that I haven't caught yet. For MeasLevel, I think I will vendor it in as suggested in Qiskit/qiskit#13793 (comment). In my most recent test with this branch and Qiskit main there were 6 failures, so there is still a little bit of other work to do, too.

eliarbel · 2025-03-05T15:57:27Z

FYI @wshanks, I've opened a PR in qiskit-neko to skip the failing test there: Qiskit/qiskit-neko#55. This will allow us to make progress with the SDK removals while you are still working on your PR.

…55) BackendV1 and qiskit.qobj removals in Qiskit/qiskit#13793 cause the test in qiskit_neko/tests/experiments/test_tomography.py to fail. This is part of the tests that are blocking the Qiskit/qiskit#13793 and Qiskit/qiskit#13872 PR in Qiskit from being merged. There is a PR in Qiskit Experiments that address these removals: qiskit-community/qiskit-experiments#1526. Until it is merged, we need to skip this failing test. Tracked in #54

wshanks · 2025-03-05T16:21:52Z

Sounds good @eliarbel. It is a bit hard to complete this PR right now because testing it requires Qiskit/qiskit#13793 and qiskit-ibm-runtime is not ready for that change plus other changes in Qiskit main so it is hard to test.

* Vendor MeasLevel/MeasReturnType from qiskit.qobj.utils. These classes were never public and the usage here does not rely class identity, just class member values. * Address `Result` header now being a dict rather than a class. * Address removal of `calibrate_instructions` from `GenericBackendV2`. * Fix performance issue with `QuantumVolume` and Qiskit Aer. When Aer was used to calculate the ideal probabilities for quantum volume circuits it was spending most of its time generating a `Target` for transpiling the circuit before simulating it. This was addressed by transpiling all the circuits in one `transpile` call. * Update `QuantumVolume` stored results in tests for Qiskit 2.0 (circuit generation algorithm changed).

wshanks · 2025-03-06T05:24:08Z

Note -- this PR passes the tests for me with Qiskit main which I think at this point has merged all the relevant qobj/backend/pulse removals. I had to test with my own branch of qiskit-ibm-runtime because its main branch still has some pulse references, so we can't say for sure this branch will be good with Qiskit 2.0 and the next qiskit-ibm-runtime release but it is looking good.

nkanazawa1989

LGTM. Just one comment

nkanazawa1989 · 2025-03-06T12:54:49Z

-            if "u_channel_lo" not in self._backend._conf_dict:
-                self._backend._conf_dict["u_channel_lo"] = []  # to avoid qiskit bug
-            self._pulse_conf = PulseBackendConfiguration.from_dict(self._backend._conf_dict)
+                self._v1 = isinstance(backend, BackendV1)


We should at least add deprecation warning in here. I prefer completely removing BackendV1 usage because we cannot guarantee the rest of our software properly works with this backend.

The release note had a line about support for BackendV1 being removed in a future release, but I added an explicit deprecations entry about it as well. Also, I added a deprecation warning here. It is hard to do because this code gets called at varying levels of the stack and Python by default only shows deprecation warnings if the stack level is right. Python 3.12 added a cool skip_file_prefixes option so I used that and it seemed to work well on 3.12+ -- it showed my user level code as responsible for the warning whether I created BackendData directly or created a T1 experiment that creates BackendData four levels lower down in the stack. Users on <=3.11 might not see the warning depending on how they are using BackendV1.

My thought was that we are not planning much change to the repo in the near term and it was easy enough to hide the import here. Qiskit 1 will go out of support in six months and we don't need to support unsupported Qiskit versions, I think. I specified this in the deprecation note.

These functions are planned to be removed from the public Qiskit API in a future release (see Qiskit/qiskit#11296 (review)).

nkanazawa1989

Thanks, let's merge this for #1529

eliarbel mentioned this pull request Mar 5, 2025

Remove references to BackendV1 following its removal in Qiskit 2.0 #1525

Closed

ElePT mentioned this pull request Mar 5, 2025

Unskip tests once downstream packages update to qiskit 2.0 Qiskit/qiskit-neko#54

Open

5 tasks

eliarbel mentioned this pull request Mar 5, 2025

Skip StateTomography test until Qiskit Experiments is ready for 2.0 Qiskit/qiskit-neko#55

Merged

eliarbel and others added 8 commits March 5, 2025 23:32

Remove references to BackendV1 following its removal in Qiskit 2.0

4e2fda0

Guard imports of BackendV1

c2ac105

wip: remove dependence on qobj/backendv1/backend.properties

755f42b

black

ee47dda

Remove stray inst_map references

342e973

black

7a16be0

lint

9987bb9

wshanks force-pushed the prepare-qiskit2-qobj-backendv1-rebase branch from 4a6ee21 to 375fea0 Compare March 6, 2025 04:56

wshanks marked this pull request as ready for review March 6, 2025 04:58

Release note

5274970

wshanks force-pushed the prepare-qiskit2-qobj-backendv1-rebase branch from 375fea0 to 5274970 Compare March 6, 2025 05:19

nkanazawa1989 reviewed Mar 6, 2025

View reviewed changes

wshanks force-pushed the prepare-qiskit2-qobj-backendv1-rebase branch from 95690c5 to 352264a Compare March 6, 2025 14:43

Add deprecation warning for BackendV1

1a11e89

wshanks force-pushed the prepare-qiskit2-qobj-backendv1-rebase branch from 352264a to 1a11e89 Compare March 6, 2025 14:47

Vendor deprecate_func and deprecate_arg from Qiskit

acbf2a6

These functions are planned to be removed from the public Qiskit API in a future release (see Qiskit/qiskit#11296 (review)).

wshanks force-pushed the prepare-qiskit2-qobj-backendv1-rebase branch from 8d5398b to acbf2a6 Compare March 7, 2025 05:07

wshanks mentioned this pull request Mar 7, 2025

Promote curve analysis deprecations from pending to current #1529

Merged

nkanazawa1989 approved these changes Mar 7, 2025

View reviewed changes

wshanks added this pull request to the merge queue Mar 7, 2025

Merged via the queue into qiskit-community:main with commit 974ad78 Mar 7, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Prepare for Qiskit 2.0 by addressing removed classes and methods#1526

Prepare for Qiskit 2.0 by addressing removed classes and methods#1526
wshanks merged 11 commits intoqiskit-community:mainfrom
wshanks:prepare-qiskit2-qobj-backendv1-rebase

wshanks commented Mar 5, 2025 •

edited

Loading

Uh oh!

eliarbel commented Mar 5, 2025

Uh oh!

wshanks commented Mar 5, 2025

Uh oh!

eliarbel commented Mar 5, 2025

Uh oh!

wshanks commented Mar 5, 2025

Uh oh!

wshanks commented Mar 6, 2025

Uh oh!

nkanazawa1989 left a comment

Uh oh!

nkanazawa1989 Mar 6, 2025

Uh oh!

wshanks Mar 6, 2025

Uh oh!

nkanazawa1989 left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

wshanks commented Mar 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Detailed list of changes:

Compatibility changes:

Other changes

Uh oh!

eliarbel commented Mar 5, 2025

Uh oh!

wshanks commented Mar 5, 2025

Uh oh!

eliarbel commented Mar 5, 2025

Uh oh!

wshanks commented Mar 5, 2025

Uh oh!

wshanks commented Mar 6, 2025

Uh oh!

nkanazawa1989 left a comment

Choose a reason for hiding this comment

Uh oh!

nkanazawa1989 Mar 6, 2025

Choose a reason for hiding this comment

Uh oh!

wshanks Mar 6, 2025

Choose a reason for hiding this comment

Uh oh!

nkanazawa1989 left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

wshanks commented Mar 5, 2025 •

edited

Loading