Thanks for the contribution. I can see where you're coming from. At the same time, I'm a bit torn. Suppose a user fills a PersistentDict, then (say as a dependency of another package) installs siphash24. Suddenly the content of their PersistentDict became inaccessible. OTOH, they're used as caches most of the time, so who cares? 🤷 I don't know.

Could you look into the CI failures?

CI failures were a mindless typo. I don't have a strong defense for this change, so if you think the value of enforcing consistency even as the package graph changes, I'll most likely just package siphash24 for Void and be done with it.

I'd be willing to give this a shot. But be warned: If it blows up on me in some way, I'll probably just revert it. Also, Ruff had some more complaints. Could you take a look?

That seems reasonable. The latest patch should pass all checks.

https://pypi.org/project/pytools/2024.1.11/

Thanks!

I don't know about this- now users of PersistentDict won't get the benefits of siphash24 unless they install siphash24 manually (but instead get a warning that does not clearly describe how to resolve the issue)...

This affects downstream packages too (see e.g. https://github.com/matthiasdiener/skvlite/actions/runs/10066243771/job/27827181383?pr=13, which tests the size of a PersistentDict-based sqlite file, where the size now depends on whether siphash24 is installed or not).

To be fair, the size of that SQLite file could be considered an implementation detail. 523ff36 rewords the warning so as to be more actionable.