It seems like the CI run for the PR was successful (or it wouldn’t have been merged). Are you able to confirm that it was successful? (Edit: I see that the details on the GitHub PR page say that all checks passed.)
The history of that file with the unicode constraints doesn’t include any recent commits.
Has anything changed on the CI machines recently, or could it be that some machines are passing that test, while others aren’t?
MariaDB is only tested on PRs with python 3.13. A fuller matrix is run on main, and it seems 3.10 and 3.13 pass, and 3.11 and 3.12 fail (3.12 most consistently). So that would explain why this wasn’t found during review.
I haven’t looked at the issue in details but the fact it passes on 3.10 and 3.13 but fail on 3.11 and 3.12 makes me believe this has more to do with CI configuration than the patch itself.
@felixxm would you have any pointer on the above? I think we did the right change in the ansible configs… Is there a specific config for MariaDB that was missed? 
@benc Would you have available a MariaDB 10.5 with Python 3.11 or 3.12 to run the failing tests locally?
There was an issue with one node (focal5), I fixed it.
My local MariaDB is 10.11.6. With @felixxm’s comment about fixing the focal5 issue, do we still need a test on MariaDB 10.5? If so, I could look at installing that version to try.
@benc Thank you for your willingness to help! The CI is now green: django-mariadb [Jenkins], so I believe this was solved by the fix applied by @felixxm (thank you!).
I think we can consider this issue solved 