[WIP] esm: basic wasm module env

[MylesBorins]

The current tests for Wasm modules is only testing modules that were compiled from Wast to Wasm.

When trying to get some basic functions working that were compiled via emscripten I quickly found that those modules were expecting a gloabal env that is generated for the .js wrapper.

Here's a link to the upstream source--> https://github.com/emscripten-core/emscripten/blob/incoming/src/preamble.js#L1117-L1158

This is a really naive PR tracking my work trying to get off the shelf wasm working as modules, it seems like there are lots of edge cases. I'm not 100% we want to ship this specific, and arguably internal, details of emscripten in core... hope to use this PR for a discussion in figuring out what makes the most sense.

@nodejs/modules

[nodejs-github-bot]

Lite-CI: https://ci.nodejs.org/job/node-test-pull-request-lite-pipeline/3593

[MylesBorins]

/cc @ofrobots@linclark@tschneidereit@lukewagner

[xtuc]

We would be against shipping that, as it's emscripten specific. It's a well known issue, I filed an issue (on the wrong repo) here: WebAssembly/esm-integration#5.

In order to make it work I used to run a transformation the binary to point to an actual ESM file: https://github.com/wasm-tool/emscripten/blob/master/loader.js#L69-L71. It would make sense to me to do something similar for testing.

I'm happy to help if you need.

[devsnek]

agreed with xtuc. it would also cause confusion for wasms built from rust which uses env for unresolved symbols

[lukewagner]

Agreed with xtuc as well. Moreover, if you use a pure upstream clang + wasi-sysroot compiler, iiuc, there is no env import; WASI effectively takes over this role. Hopefully we can get Emscripten to emit pure-WASI builds (or have a flag to do so) in the future.

[chjj]

I think ESM wasm is flawed for this reason. Lots of wasm modules have very complicated setups. IMO, import "./whatever.wasm" should return a compiled Module as the default export rather than an Instance. This would solve the problem of easy instantiation for everything (even emscripten), but Chrome pre-emptively broke this approach by limiting both Modules and Instances to 4kb for synchronous instantiation.

There's so much wrong with wasm module instantiation that it's practically unfixable at this point. I spent a few days trying to write an emscripten wrapper that provided consistent behavior between the browser and node -- it's impossible because of Chrome's arbitrary 4kb limit for sync instantiation. Surely ESM wasm would be the answer? Nope, that's broken too. =/

[ofrobots]

@lukewagner

Moreover, if you use a pure upstream clang + wasi-sysroot compiler, iiuc, there is no env import; WASI effectively takes over this role.

FYI, this doesn't seem to be the case just yet.. wasi-sdk-5.0 still ends up emitting env dependencies for me:

 ... (type (;17;) (func (param i32 i32 i32 i32 i32 i32 i32))) (import "env" "__indirect_function_table" (table (;0;) 3 funcref)) (import "env" "__stack_pointer" (global (;0;) (mut i32))) (import "env" "__memory_base" (global (;1;) i32)) (import "env" "__table_base" (global (;2;) i32)) (import "wasi_unstable" "fd_prestat_get" (func $__wasi_fd_prestat_get (type 0))) (import "wasi_unstable" "fd_prestat_dir_name" (func $__wasi_fd_prestat_dir_name (type 3))) (import "wasi_unstable" "environ_sizes_get" (func $__wasi_environ_sizes_get (type 0))) (import "wasi_unstable" "environ_get" (func $__wasi_environ_get (type 0))) (import "wasi_unstable" "args_sizes_get" (func $__wasi_args_sizes_get (type 0))) (import "wasi_unstable" "args_get" (func $__wasi_args_get (type 0))) (import "env" "main" (func $main (type 0))) (import "wasi_unstable" "proc_exit" (func $__wasi_proc_exit (type 4))) (import "wasi_unstable" "fd_fdstat_get" (func $__wasi_fd_fdstat_get (type 0))) ... 

[lukewagner]

Oops, sorry to give you incorrect information; I had thought that was the goal. Maybe @sunfishcode can shed some light on what the plan is?

[sunfishcode]

@ofrobots Statically-linked WASI executables don't have env dependencies in them. It looks like you may have used -Wl,-shared or similar here. Dynamic linking is not supported in WASI yet.

[ofrobots]

Indeed, there was confusion on my part about to link a wasm library without a main. -shared wasn't the answer.

[MylesBorins]

Feels like there has been good discussion.. closing in lieu of #27850