G028-wit-bindgen-cpp-wrong-enum-discriminant-in-option-variant-payload

[xiaozzzZZzzz240]

wit-bindgen-cpp encodes wrong enum discriminant for enum payload inside variant inside option, producing incorrect values under jco

Bug report: C++ consumers generated by wit-bindgen-cpp lift/lower the enum case inside an option-wrapped variant with a wrong discriminant (e3 becomes e1), yielding wrong output values on jco regardless of provider language.

When a C++ consumer lowers option<list> whose variant case carries an enum payload, the generated binding writes canonical-ABI bytes in place over the source vector's own memory. It stores the variant discriminant (1 for c1) at byte 0 first, clobbering the enum payload stored there by std::variant's layout, then reads the now-corrupted payload — so e3 becomes e1. Any provider language (C, C++, Rust) receives the wrong value.

Environment

component	version
OS	macOS 15.7 (Darwin 24.6.0), x86_64
jco	1.16.1 (source checkout, tag jco-v1.16.1)
Node.js	v22.22.2
wac	wac-cli 0.8.1
wasi-sdk	27.0 (wasm32-wasip2-clang/clang++)

Steps to reproduce

The full chain is: compile the guest sources into a provider component and a consumer component, compose them with wac, then run the composed component. Prebuilt copies of every artifact are included (components/provider.wasm, components/consumer.wasm, composed.wasm), so you can also skip straight to step 5.

Github repo:https://github.com/xiaozzzZZzzz240/G028-wit-bindgen-cpp-wrong-enum-discriminant-in-option-variant-payload

1. Clone this repository:

   git clone <THIS_REPO_URL>
   cd <THIS_REPO_DIR>

2. Compile the provider component from the C sources in src/provider-C/ (toolchain versions in the table above; the WIT package is in wit/):

   wasm32-wasip2-clang -o components/provider.wasm -mexec-model=reactor src/provider-C/world9-wi7n9u47_generated.c src/provider-C/world9_wi7n9u47.c src/provider-C/world9_wi7n9u47_component_type.o

3. Compile the consumer component from the C++ sources in src/consumer-Cpp/:

   wasm32-wasip2-clang++ -std=c++23 -fno-exceptions -o components/consumer.wasm -mexec-model=reactor src/consumer-Cpp/world9b-mul38v8w_generated.cpp src/consumer-Cpp/world9b_mul38v8w.cpp src/consumer-Cpp/world9b_mul38v8w_component_type.o

4. Compose the two components into one runnable component with wac 0.8.1:

   wac plug components/consumer.wasm --plug components/provider.wasm -o composed.wasm

5. Run the composed component with jco:

   npm install -g @bytecodealliance/jco@1.16.1
   jco run composed.wasm

Expected result

The composed component should print exactly the following, then exit 0.

"resinst0"
"resinst1"
"resinst2"
"resinst0"
resource("temres36")
"c2"
true
"c0"
resource("temres39")
(resource("temres40"), "bar")
"c2"
true
"temres39"
"temres44"
some([c1(enum(e3))])
resource("temres6")
"c0"
true
7
true

Actual result

Under jco the same composed.wasm exits with rc=0. stdout:

"resinst0"
"resinst1"
"resinst2"
"resinst0"
resource("temres36")
"c2"
true
"c0"
resource("temres39")
(resource("temres40"), "bar")
"c2"
true
"temres39"
"temres44"
some([c1(enum(e1))])
resource("temres6")
"c0"
true
7
true

stderr:

(empty)

wasmtime on the exact same file: rc=0

"resinst0"
"resinst1"
"resinst2"
"resinst0"
resource("temres36")
"c2"
true
"c0"
resource("temres39")
(resource("temres40"), "bar")
"c2"
true
"temres39"
"temres44"
some([c1(enum(e1))])
resource("temres6")
"c0"
true
7
true

[tschneidereit]

@xiaozzzZZzzz240 thank you for the report. Please note though our AI Tool Use Policy: this report seems pretty clearly fully tool-generated, including the title. Please make sure that you can take on full responsibility for everything you post, and that everything is structured and worded in a way that's suited for consumption by humans.

In this particular case, please at least adapt the title, and make sure you properly reduce the report. If I understand correctly, this has nothing to do with JCO, but given that the report makes heavy mention of it, I'm not entirely sure.