Very slow Trixie emulation?

[stumpylog]

The request to upgrade to QEMU 10 stemmed from slowdowns seen when building a Docker image for arm64 using enumlation. See also docker-library/python#1073

Following #265, I hoped using QEMU 10 would improve the situation, but it doesn't appear to have helped: https://github.com/stumpylog/slow-trixie-qemu-repo/actions/runs/17651118650. In short, emulation of arm64 with a Trixie base image is still slower than using Bookworm, even in a pretty contrived little example. Using a more real world example, the time goes from around 20 minutes to over an hour: https://github.com/paperless-ngx/paperless-ngx/actions/runs/17650033145/job/50159145085

At this point, I'm not sure what to investigate or try. Does anyone have any ideas? What in Trixie changed to make this happen? Does Docker/Buildx/something need to update as well?

[stumpylog]

@crazy-max I'm sorry to directly ping you, I can tell you're busy across a lot of projects. But at this point, I'm not sure what else to try. Thanks to your work, the building is now using a newer QEMU, but it remains quite slow in real world practice. We'd like to provide this update to our users, but over an hour of building isn't feasible.