Component model: `bindgen!` doesn't generate some structures

[tareksander]

I'm working on a new WASI proposal and I'm implementing an example implementation with wasmtime. I have a correct WIT definition AFAIK and I'm generating the bindings with the bindgen macro. For some reason, the key-data record and key variant didn't generate a corresponding struct/enum. There's an error inside the generated bindgen code "cannot find type "KeyData" in this scope".

[alexcrichton]

Thanks for the report! I believe that issue should be fixed in #8065.

One other issue is that the move variant is defined twice: one here and one here. That's probably something which should be a WIT-level error, however, rather than late in bindgen.

Also, for the future in case you find it useful, WASMTIME_DEBUG_BINDGEN=1 can help provide some more comprehensible error messages as the source will be available. That wouldn't have helped you here as it was a bindgen bug, but if further issues arise it may help!

[tareksander]

One other issue is that the move variant is defined twice: one here and one here. That's probably something which should be a WIT-level error, however, rather than late in bindgen.

Yes, that I have also notices and fixed locally, but the type error persisted anyways.

If I understand #8065 correctly, right now bindgen thinks the types are unused, and therefore it doesn't generate them? I have restructured the WIT definitions anyways, I'll see if the bug also happens then.

[alexcrichton]

Yes the event type and key-data aren't actually used by any imports or exports at this time (function-wise), so Wasmtime thinks they're dead types and doesn't generate bindings for one but does generate bindings for the other. If a function uses the types then bindings should get generated correctly. (note that regardless these are bugs in Wasmtime to fix!)