Adding #[wasm_bindgen] means you can't run cargo bench anymore

[fitzgen]

You get link errors and things like that.

Ideally, we should make all the wasm-bindgen generated code guarded by #[cfg(target_arch = wasm32)] or something like that.

The

[lib]
crate-type = ["cdylib"]

stuff is a little harder. It can be commented out just when running cargo bench, but that is clunky. Ideally that would not be necessary either.

[alexcrichton]

One idea would be to use binaries as the output target rather than cdylib, and wasm-bindgen could presumably perform wasm-gc and keep the binaries small (and strip the main function or something like that)

That would still require you to separate code, however, into a binary and a library as only the library would be available for writing benchmarks and tests.

Overall I think the situation isn't great and this is somewhat of a Cargo issue. Ideally you'd be able to customize the crate-type more aggressively than only by what's written in the Cargo.toml.

---

I wonder, though, perhaps the project template for wasm could have two crates? One crate, the primary one, is an rlib which is benchmark-able and testable (and using wasm-bindgen). The other, a cdylib, is purely a reexport of the public contents of the rlib (aka pub use other_crate::*). I believe that should still produce a working wasm target and would be relatively low overhead? (maybe a pain to remember to build a different target for wasm, but scripts may be able to automate this)

[alexcrichton]

cc rust-lang/cargo#5367