Repository navigation
Unable to use serde macros in additional_derives when importing wasi::http types #812
Description
I have a simple library testing out the wasi:http types. When I try to add an additional_derives: [serde::Serialize, serde::Deserialize] to the wit_bindgen::generate! macro definition I get the below error. Not sure if thise repo or the wasi-http repo would be better for this issue, but trying here first.
Reproduction can be found in this repo. Uncommenting this line will cause the error.
error[E0277]: the trait bound `wasi::io::error::Error: Deserialize<'_>` is not satisfied
--> src/lib.rs:5:1
|
5 | / wit_bindgen::generate!({
6 | | world: "wasi-http-tests",
7 | | exports: {
8 | | "wasi-http-tests:component/client/wasi-http-tests-client": WasiHttpTestsClient,
9 | | },
10 | | additional_derives: [serde::Serialize, serde::Deserialize]
11 | | });
| |__^ the trait `Deserialize<'_>` is not implemented for `wasi::io::error::Error`
|
= help: the following other types implement trait `Deserialize<'de>`:
bool
char
isize
i8
i16
i32
i64
i128
and 137 others
note: required by a bound in `newtype_variant`
--> /Users/lnj/.cargo/registry/src/index.crates.io-6f17d22bba15001f/serde-1.0.193/src/de/mod.rs:2119:12
|
2117 | fn newtype_variant<T>(self) -> Result<T, Self::Error>
| --------------- required by a bound in this associated function
2118 | where
2119 | T: Deserialize<'de>,
| ^^^^^^^^^^^^^^^^ required by this bound in `VariantAccess::newtype_variant`
= note: this error originates in the macro `wit_bindgen::generate` (in Nightly builds, run with -Z macro-backtrace for more info)
error[E0277]: the trait bound `wasi::io::error::Error: Serialize` is not satisfied
--> src/lib.rs:5:1
|
5 | / wit_bindgen::generate!({
6 | | world: "wasi-http-tests",
7 | | exports: {
8 | | "wasi-http-tests:component/client/wasi-http-tests-client": WasiHttpTestsClient,
9 | | },
10 | | additional_derives: [serde::Serialize, serde::Deserialize]
11 | | });
| |__^ the trait `Serialize` is not implemented for `wasi::io::error::Error`
|
= help: the following other types implement trait `Serialize`:
bool
char
isize
i8
i16
i32
i64
i128
and 136 others
note: required by a bound in `serialize_newtype_variant`
--> /Users/lnj/.cargo/registry/src/index.crates.io-6f17d22bba15001f/serde-1.0.193/src/ser/mod.rs:936:12
|
928 | fn serialize_newtype_variant<T: ?Sized>(
| ------------------------- required by a bound in this associated function
...
936 | T: Serialize;
| ^^^^^^^^^ required by this bound in `Serializer::serialize_newtype_variant`
= note: this error originates in the macro `wit_bindgen::generate` (in Nightly builds, run with -Z macro-backtrace for more info)
Activity
pchickey commentedon Jan 10, 2024
wasi::io::error::Erroris a resource type, so it will never be able to impl Serialize or Deserialize. So, the bug here is that the derives are being applied to the type alias created for https://github.com/WebAssembly/wasi-http/blob/main/wit/types.wit#L7landonxjames commentedon Mar 27, 2024
Another related bug I just encountered: When a
variantcontains a case and the type associated with that case is aresourcetheserdemacros cause a build failure. So the following wit fails to buildresource perf-event { constructor(); parse-string: func(input: string) -> result<some-event, string>; } resource constructor-perf-event { constructor(input: string); get-data: func() -> some-event; } variant supported-events { none, my-event(constructor-perf-event), foo(perf-event) } resource variant-perf-event { constructor(input: supported-events); get-child-event: func() -> supported-events; }Due to the
resourcetypes not implementingSerialize.I think this is further evidence to motivate some kind of attribute system in WIT so things like
serdetraits can be properly targeted to the correct types.LUK3ARK commentedon May 26, 2024
This is correct, if I add additional derives, I expect it to only be added to records and not custom resources, or at least a way to specify what resources not to apply to.
Is there any solution currently to using serde for types but ignoring for resources/ fixing for resources?
thomastaylor312 commentedon Aug 29, 2024
I am pretty sure this code is mostly unchanged since I wrote it and based on my quick skim again, it does only target records and anything that gets turned into a rust
enum. But that is only at the top level. So if arecordorvariantcontains a resource within it, it won't work (because the top level resource doesn't get the attributes). FWIW, I knew we'd run into this sooner rather than later, but we figured it would be good to have so you at least didn't have to do the manual remote derive stuff for something simple (plus this was back far enough I probably didn't know how to solve it well enough at that point).The fix here would be to change the generation code
wit-bindgen/crates/rust/src/interface.rs
Line 1292 in 02f0784
wit-bindgen/crates/rust/src/interface.rs
Line 1388 in 02f0784
Resolveso that you could walk the whole type and check if any nested type is a resource before adding the additional derives.