Closed t-imamichi closed 3 years ago
I thought qiskit-aqua was supposed to stay frozen to the current qiiskit-terra, qiskit-ignis in pypi forever. If we extend the requirements to the latest version in master, then aqua will break eventually and we will have to fix frozen code. I will soon create a PR for qiskit-tutorials eliminating all aqua usage there. I would be ok with this change here until qiskit-tutorials doesn’t use aqua anymore but once it changes, I believe terra should remove aqua from its CI and we should go back to the previous requirements.
I'm OK to fix terra. Anyway we need to fix these lines, otherwise terra CI does not pass. https://github.com/Qiskit/qiskit-terra/blob/e1d49102dd18bf6ffdf28e5657e1cde4091dc609/azure-pipelines.yml#L709-L710
Yes, that is why I am ok with this PR being merged but then reverted once aqua is not a dependency in qiskit-tutorials and then they should remove aqua from their CI and we should revert. Currently, I agree I don't see another way because even if they try to use aqua from pypi they will get the same error. The only way I see to fix terra CI while aqua is there would be to not use anything from master in their CI for qiskit-tutorials. I am not sure they would agree to that change.
This seems to be an emergency in Terra as all their PRs are failing because of this right? I will approve this fix for now.
Yes, https://github.com/Qiskit/qiskit-terra/pull/6160 fails even if it fixes only a typo. This is the CI log https://dev.azure.com/qiskit-ci/qiskit-terra/_build/results?buildId=26430&view=logs&jobId=47acf835-66bb-5b0f-07e4-6f53ffe80e20&j=47acf835-66bb-5b0f-07e4-6f53ffe80e20&t=d6cc4d9a-83df-59b0-cfe4-173c81c4d987
Could you double check whether this PR is enough to fix the terra CI issue?
It should fix the current version conflict with ignis and terra there. The aqua code passes their tutorias because the aqua CI runs their tutorials too as double check. The difference is that our CI reverts to terra, ignis pypi if master cannot satisfy our requirements.. That is why our CI still works.
As far as other problems down the line is harder to say if there are other sub-dependencies that may cause conflicts without running their CI.
Thank you for your information. I finally understand why aqua CI works fine even if terra CI fails.
Yes, in this PR I might have just extended the upper bounds and leave the lower bounds but if it will be reverted later, it is not so important.
I see. I adjust this PR as you suggest. What do you think?
Yes, just change the upper bounds as to leave it for now installable with pypi terra/ignis as well as master. The current change doesn't let you use terra/ignis from pypi even though aqua works fine with them.
Thanks. I reverted the lower bounds.
Summary
Fix the CI error of terra.
The following causes an error:
See https://github.com/Qiskit/qiskit-terra/pull/6160 CI log: https://dev.azure.com/qiskit-ci/qiskit-terra/_build/results?buildId=26430&view=logs&jobId=47acf835-66bb-5b0f-07e4-6f53ffe80e20&j=47acf835-66bb-5b0f-07e4-6f53ffe80e20&t=d6cc4d9a-83df-59b0-cfe4-173c81c4d987
Details and comments