Sorry I'm not sure why this would be happening.

This is an example:

use wasm_bindgen::prelude::*;

#[wasm_bindgen]
pub struct Test1 {
    #[wasm_bindgen(skip)]
    pub s: Option<String>,
}

#[wasm_bindgen]
pub struct Test2 {
    #[cfg_attr(feature = "wasm-bindgen", wasm_bindgen(skip))]
    pub s: Option<String>,
}

pub struct Test3 {
    #[wasm_bindgen(skip)]
    pub s: Option<String>,
}

It produces this error:

error: expected non-macro attribute, found attribute macro `wasm_bindgen`
  --> src/lib.rs:11:42
   |
11 |     #[cfg_attr(feature = "wasm-bindgen", wasm_bindgen(skip))]
   |                                          ^^^^^^^^^^^^ not a non-macro attribute

error: expected non-macro attribute, found attribute macro `wasm_bindgen`
  --> src/lib.rs:16:7
   |
16 |     #[wasm_bindgen(skip)]
   |       ^^^^^^^^^^^^ not a non-macro attribute

error[E0277]: the trait bound `String: std::marker::Copy` is not satisfied
  --> src/lib.rs:12:12
   |
12 |     pub s: Option<String>,
   |            ^^^^^^^^^^^^^^ the trait `std::marker::Copy` is not implemented for `String`
   |
   = note: required because of the requirements on the impl of `std::marker::Copy` for `Option<String>`

Test2 or Test3 alone also produces an error.

The error on Test3 is expected because there's no attribute on the struct itself, the error on Test2 is probably because the cfg_attr would be processed after the outer procedural macro. I think you may be able to fix this by placing #[cfg_attr] on the outer wasm_bindgen attribute as well, but that's just a guess.

use wasm_bindgen::prelude::*;

#[cfg_attr(feature = "wasm-bindgen", wasm_bindgen)]
pub struct Test2 {
    #[cfg_attr(feature = "wasm-bindgen", wasm_bindgen(skip))]
    pub s: Option<String>,
}

fails the same way.
I think the problems is in crates/macro-support/src/parser.rs in the function find on line 172.
It looks only for wasm_bindgen and my guess is that cfg_attr is not evaluated yet.

I found a solution.

#![feature(cfg_eval)]
use wasm_bindgen::prelude::*;

#[cfg_eval]
#[cfg_attr(feature = "wasm-bindgen", wasm_bindgen)]
pub struct Test2 {
    #[cfg_attr(feature = "wasm-bindgen", wasm_bindgen(skip))]
    pub s: Option<String>,
}

It works. But in some way it should be the responsibility of the wasm_bindgen macro to do this.

I also have this issue but have not found a fix for it yet. When I try the supposed solution from above I just get syntax errors about expected symbols in this #![feature(cfg_eval)] line.

#![feature(cfg_eval)] // syntax error here
use wasm_bindgen::prelude::*;

#[cfg_eval]
#[cfg_attr(feature = "wasm-bindgen", wasm_bindgen)]
pub struct Test2 {
    #[cfg_attr(feature = "wasm-bindgen", wasm_bindgen(skip))]
    pub s: Option<String>,
}

#[cfg_attr(target_arch = "wasm32", wasm_bindgen::prelude::wasm_bindgen)]
pub struct MyStruct {
    #[cfg_attr(target_arch = "wasm32", wasm_bindgen::prelude::wasm_bindgen(skip))]
    pub my_field: String,

I do not believe that there is something wasm-bindgen can do here. While it is possible for wasm-bindgen to parse #[cfg_attr(...)] attributes, we can't evaluate them, this is just a limitation of how proc-macros work in Rust.

I'm unfamiliar with cfg_eval, so I can't assist there.

it's fine. i guess I'll just wait and see what the future has to offer. In the meantime, this is my workaround:

#[cfg(target_arch = "wasm32")]
#[wasm_bindgen::prelude::wasm_bindgen]
pub struct MyStruct {
    #[wasm_bindgen::prelude::wasm_bindgen(skip)]
    pub my_field: String,
}

#[cfg(not(target_arch = "wasm32"))]
pub struct MyStruct {
    pub my_field: String,
}

a bit cumbersome, but it works.

I believe I found a fix which doesn't require cfg_eval, thus can be used on stable (which motivated me to look into it). It uses similar two-step expansion pattern used to allow cfg_attr in impl blocks: #1208. It is also inspired by the discussion on a similar problem in bon crate: elastio/bon#124.

Overview:

• During #[wasm_bindgen] attribute macro expansion on struct, instead of generating ast for Program, it puts a #[derive(BindgenedStruct)] on it and preserves all the wasm_bindgen atributes as-is.
• When #[derive(BindgenedStruct)] is expanded, cfg_attr are already evaluated, so macro can just processes the leftover attributes and generate the glue code required, as it's done now. Attributes are then automatically removed by the compiler, since macro is defined as #[proc_macro_derive(BindgenedStruct, attributes(wasm_bindgen))].

BindgenedStruct is a placeholder, internal-use name.

I think I have a POC working with tests passing and would be happy to work on getting it to a PR state.

That sounds workable indeed. I'm hoping the implementation can be actually lightweight.

Happy to review a PR! Would need a good testsuite though!

Opened #4351 with proposed fix