*TL;DR; In case you were not aware during the weekend we had experienced Pydantic 2.10.0/2.10.1 issue that impacted both main for development and potentially also released airflow versions (where constraints are not used or Pydantic upgraded by other means). *
This is going to be fixed in 2.10.2 of Pydantic (which is likely to come out in a matter of hours or days). But in the meantime there are workarounds we can tell our users to apply. The result of that release - was that our `main` needed a workaround ( https://github.com/apache/airflow/pull/44317). But more importantly, those releases could (and did) cause issues for previous airflow versions If users did not use constraints when installing airflow from scratch, or upgrade Pydantic independently, airflow would start failing. With our feedback and help in testing it, Pydantic team is working on 2.10.2 and a permanent fix is already merged ( https://github.com/pydantic/pydantic/pull/10962), but seems that they have more issues to handle as Pydantic 2.10 had a number of internal changes that had some unintended side-effects despite being theoretically backwards-compatible (which is a clear manifestation of the Hyrum's law in a full swing https://www.hyrumslaw.com/. More details on those "Hyrum's law effects for Pydantic 2.10 release are kept together by the Pydantic team in https://github.com/pydantic/pydantic/issues/10910 Once 2.10.2 of Pydantic is released we will get rid of the workaround and exclude 2.10.0/2.10.1 of Pydantic from main (and backporting to 2.10.4 eventually - and advocate for yanking the bad Pydantic versions) In the meantime - if anyone reports an issue with "Unable to evaluate type annotation" - a simple workaround to suggest is to install the `eval-type-backport` package alongside the new Pydantic version and airflow - it's a crude, but working workaround until Pydantic 2.10.2 gets released. I will keep you posted - I am following this issue closely. J.
