Deriving uniffi::Record together with ZeroizeOnDrop does not compile. Solvable?

[Sajjon]

I've recently tried to add Zeroize on my UniFFI records / enums, like so:

#[derive(
    Zeroize,
    ZeroizeOnDrop,
    Clone,
    uniffi::Record,
)]
pub struct Mnemonic {
    pub words: Vec<BIP39Word>, // BIP39Word: Zeroize, Clone, uniffi::Record
}

Results in compilation error:

cannot move out of type `Mnemonic`, which implements the `Drop` trait
move occurs because `obj.words` has type `Vec<BIP39Word>`, which does not implement the `Copy` trait

Hmm.... is this even solvable?

[jplatte]

UniFFI could certainly generate code that works without destructuring, but for zeroize in particular I think you'd be defeating the purpose of that derive since UniFFI would be creating copies of the bytes that don't get zeroized on drop.

[Sajjon]

but for zeroize in particular I think you'd be defeating the purpose of that derive since UniFFI would be creating copies of the bytes that don't get zeroized on drop.

Right that is what I feared, but maybe some optimization could be done in "that realm", e.g. if UniFFI could check if the Source or Target implements Zeroize and could use it on any copies, so that only the the resulting Target / Source contains those bytes, but any "ephemeral copies" are zeroed out during the conversion?

[jplatte]

It's impossible for a macro to check whether a trait is implemented, trait solving happens after macro expansion (trait solving depends on the output of all macros in a given source tree incl. dependencies).

It would be possible to introduce #[uniffi(zeroize)] as a helper attribute though. My hunch is that there's probably no appetite for it by the maintainers, but maybe you're lucky 😅

[Sajjon]

It's impossible for a macro to check whether a trait is implemented, trait solving happens after macro expansion (trait solving depends on the output of all macros in a given source tree incl. dependencies).

Right that makes sense, thanks for explaining @jplatte ! I'm still fairly new to Rust (started in December last year) and have only beeng doing macros since February.

It would be possible to introduce #[uniffi(zeroize)] as a helper attribute though

Yes that would really be amazing! I might try to give a PR a go later on!

[mhammond]

I'd suggest you start with a justification first - I can only speak for myself, but @jplatte is correct regarding appetite - I doubt we want to support every cool-sounding crate out there. In this specific case, the hurdle to overcome is already noted above - uniffi makes many copies of records with the expectation that many of them live in the foreign code - in these scenarios, zeroize seems, frankly, pointless.