Explicitly defining bounds with multiple traits leads to spurious bounds

[Ekleog]

The following code:

#[derive(Educe)]
#[educe(Clone, Debug(bound(T: Debug)))]
pub struct Obj<T: Object, LocalDb: ClientSideDb> {
    ptr: DbPtr<T>,
    data: Arc<T>,

    #[educe(Debug(ignore))]
    db: Arc<ClientDb<LocalDb>>,
}

Leads to the Clone impl having a T: Debug bound, while I'd expect only the Debug implementation to have this bound.

Unfortunately I guess fixing this will require bumping educe to 0.6, as it's technically a breaking change, though it's also a bugfix.

PS: sorry for the non-minimized example, I'm in a hurry and just copy-pasted my current code; hopefully it'd also reproduce with an empty struct as expected! If not please let me know and I'll try to make a minimized example.

[ijackson]

I can repro this.

[ijackson]

The output seems to depend on the ordering!

[ijackson]

I think the root cause is that TraitHandler::trait_meta_handler takes &mut DeriveInput and each impl assembles its where clauses in the .generics field there.