My type A
, which can contain anything that implements trait Trait
, is serialisable, although the type implementing the trait Trait
might not be. In my case, it cannot be - it's a private asymmetric key:
extern crate serde;
#[macro_use]
extern crate serde_derive;
use serde::de::DeserializeOwned;
use serde::Serialize;
trait Trait {
type SerialisableType: Clone + Serialize + DeserializeOwned;
fn inner(&self) -> &Self::SerialisableType;
}
#[derive(Serialize, Deserialize)]
enum A<T: Trait> {
Variant0(B<T>), // *** NOTE: Compiles if this is commented ***
Variant1(T::SerialisableType),
}
#[derive(Serialize, Deserialize)]
struct B<T: Trait> {
inner: T::SerialisableType,
}
// ==============================================
struct NonSerialisable {
serialisable: Serialisable,
}
impl Trait for NonSerialisable {
type SerialisableType = Serialisable;
fn inner(&self) -> &Self::SerialisableType {
&self.serialisable
}
}
#[derive(Clone, Serialize, Deserialize)]
struct Serialisable(Vec<u8>);
#[derive(Serialize, Deserialize)]
enum E {
Variant0(A<NonSerialisable>),
Variant1(B<NonSerialisable>),
}
fn main() {}
playground
This errors out with:
error[E0277]: the trait bound `NonSerialisable: serde::Serialize` is not satisfied
--> src/main.rs:43:10
|
43 | #[derive(Serialize, Deserialize)]
| ^^^^^^^^^ the trait `serde::Serialize` is not implemented for `NonSerialisable`
|
= note: required because of the requirements on the impl of `serde::Serialize` for `A<NonSerialisable>`
= note: required by `serde::Serializer::serialize_newtype_variant`
error[E0277]: the trait bound `NonSerialisable: serde::Deserialize<'_>` is not satisfied
--> src/main.rs:43:21
|
43 | #[derive(Serialize, Deserialize)]
| ^^^^^^^^^^^ the trait `serde::Deserialize<'_>` is not implemented for `NonSerialisable`
|
= note: required because of the requirements on the impl of `serde::Deserialize<'_>` for `A<NonSerialisable>`
= note: required by `serde::de::VariantAccess::newtype_variant`
If I comment out A::Variant0
, as mentioned in the inline comment in the code, then it compiles fine. This makes me think that the compiler is unable to deduce that B<T>
is serialisable, but then it actually is able to deduce that because it can figure out E
is serialisable which would require B
to be serialisable as well.
Where is the problem?
During macro expansion the compiler has not yet determined which B
is being referred to inside Variant0
or how that B
may use its type parameters. As such, macro expansion infers trait bounds that would work for the most common cases of what B
might be, like if B
were Box
or Vec
. In those cases serializing B<T>
would require T: Serialize
and deserializing B<T>
would require T: Deserialize<'de>
.
You can provide handwritten generic type bounds to replace the inferred bounds.
#[derive(Serialize, Deserialize)]
#[serde(bound = "")]
enum A<T: Trait> {
Variant0(B<T>),
Variant1(T::SerialisableType),
}
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With