Forced serde_derive dependency

[jean-airoldie]

Using this crate alongside a serde = { version = "1.0", features = ['derive'] } dependency causes behavior like in #2. This is due to this line that assumes that serde_derive is present. However it cannot be replaced by serde::Serialize since this causes serde-rs/serde#1467 caused by rust-lang/rust#55779.

A sketchy work around is to define a

mod serde_derive {
    pub use serde::{Serialize, Deserialize};
}

in the same module where Serialize_tuple is used.

I think that once rust-lang/rust#55779 is fixed, this crate should depend on serde = { version = "1.0", features = ['derive'] } since, from what I understand, feature flags forces the dependency down the chain, while simply depending on serde_derive (like this crate does currently) does not.

[kardeiz]

Thanks for the issue, and I think that is a good solution. I'll keep this open until the SO issue is resolved and I update this crate.

Another workaround (definitely not ideal) for your setup would be to copy the serde_derive dependency from Cargo.lock to Cargo.toml. If you make sure to use the same version that serde depends on, nothing else will need to be installed; it just makes serde_derive available to user code.

[jean-airoldie]

A workaround would be to add a crate helper attribute to the proc macro so that the user can control the name of the dependency crate.

#[derive(Serialize_tuple)]
#[serde_tuple(crate = "serde")]
struct SomeStruct;

[kardeiz]

@jean-airoldie I think that is a good idea, but I think the name crate could be confusing: I think it would need to be something like crate_that_provides_derive_serialize_and_deserialize, or serde_derive_crate (though I think the latter might also be confusing).

Actually, it looks like your initial suggestion (depend on serde with the derive feature), already works. Could you see if the change I just pushed to master works for you?

Looking back, I'm not sure why this crate was depending on serde and serde_derive, since it doesn't use them directly, but depending on serde with the derive feature does have the desirable feature of not requiring serde_derive in the user's dependencies.

[jean-airoldie]

The master branch works for me currently. Should we close this?

[kardeiz]

@jean-airoldie Great! I just went ahead and updated the syn/quote/proc-macro2 dependencies to their 1.x versions and released 0.4.0.

I'll close this now, but let me know if anything doesn't work,