Yeah -- so apparently this is a major change, so I guess we can't just add it like this. And I think it's also probably not worth doing a 2.0 only for this?

Yeah I'm afraid this alone doesn't motivate a 2.0 for me :(

For now we could keep the Zeroize impls, and downstream users can write Zeroizing<PrivateKeyDer<'_>>. That is a bit unsatisfying though.

For now we could keep the Zeroize impls, and downstream users can write Zeroizing<PrivateKeyDer<'_>>. That is a bit unsatisfying though.

So one goal in this PR was to avoid having zeroize types/traits in the public API. Do you think we should allow those in order to enable the use of stuff like Zeroizing? Alternatively, I suppose we could provide our own Zeroizing type wrapper that handles this, while still hiding the dependency?

On the assumption we don't want zeroize in the public API, I think the latter is the best way to go -- specifically provide some zeroizing newtypes (maybe generic) that hold a PrivateKey* value. That means downstream users need to opt-in to the drop behaviour change, but that is the situation we are in.

• Published rustls-pki-types v1.12.0 at registry crates-io
• [new tag] v/1.12.0 -> v/1.12.0
• Release notes