Managed to reproduce the errors with the tests from CI (seems to fail in both ubuntu 24.04 and macos-15 arm64): https://github.com/colmap/colmap/actions/runs/20923717516/job/60115792676.

Strange enough, I cannot reproduce this issue locally.

I am confused. So why did this not trigger before as we have 3.10 running in CI? What's the concrete issue? If we don't test 3.10 in CI anymore, I expect we'll frequently break master because changes slip through that will be incompatible with 3.10?

I am confused. So why did this not trigger before as we have 3.10 running in CI? What's the concrete issue? If we don't test 3.10 in CI anymore, I expect we'll frequently break master because changes slip through that will be incompatible with 3.10?

We did not hit the type annotation bug before. Alternatively, what about enabling both 3.10 and 3.11 in the PR, but skip mypy for 3.10? This will make the CI run a bit longer.

Let's see how it goes. In the first breakage in master, let's reenable 3.10 in PR checks.