Why does emscripten.Instantiate and InstantiateForModule differ drastically?

[DejeroDaniel]

One imports all the host functions needed, and the other barely imports anything. The documentation point you towards Instantiate, which won't work for any C++ module requiring exception support, for example.

Why don't they both set the same emscription host functions?

[ncruces]

Maybe @jerbob92 (the creator of wazero-emscripten-embind and our resident emscripten expert) can help?

[jerbob92]

That's because Emscripten dynamically registers host functions as needed, which are called invoke_xxx, where xxx is the function signature. Wazero needs to know which dynamically registered host functions are imported by the module so that it can generate the function as well (https://github.com/tetratelabs/wazero/blob/main/internal/emscripten/emscripten.go#L43). If Wazero wouldn't do this, it would need to register any combination possible, which is not feasible.

I suspect Instantiate is left in to not break external code, but it's probably not very usable in most cases, so I would always go withInstantiateForModule.

Please note that real C++ exceptions is not supported by Wazero yet, I have been building that here:
https://github.com/jerbob92/wazero/blob/feature/implement-emscripten-cpp-exceptions/internal/emscripten/exceptions.go

And it does work, but I don't like the implementation so I'm still waiting for a response on #1654 to make a PR.

[Danlock]

OK. Should I make a PR to change the documentation to use InstantiateForModule (since it sounds like Instantiate should be deprecated)?

[Danlock]

Also as an aside, I'm interested in using wazero to run Tesseract, compiled to WASM via Emscripten.
The Tesseract WASM emscripten imports are minimal enough that I'd figure I could try to implement them myself. (env._emscripten_fs_load_embedded_files and env.system)

Do you think this project is feasible with the current support of emscripten in wazero?

[jerbob92]

OK. Should I make a PR to change the documentation to use InstantiateForModule (since it sounds like Instantiate should be deprecated)?

Yeah, sounds good!

Also as an aside, I'm interested in using wazero to run Tesseract, compiled to WASM via Emscripten. The Tesseract WASM emscripten imports are minimal enough that I'd figure I could try to implement them myself. (env._emscripten_fs_load_embedded_files and env.system)

Do you think this project is feasible with the current support of emscripten in wazero?

I don't know what these imports are, but generally you would need to recompile for a standalone WASM file, then you will have way less emscripten specific imports. I use this branch to compile my stuff so that it uses WASI for things like filesystem operations: emscripten-core/emscripten#18285

So it is feasible, but you might need to do some special steps to get there.