Better ESM

[james-pre]

According to the documentation for emcc -o, specifying an output with .mjs outputs an ES module:

.mjs : ES6 JavaScript module (+ separate .wasm file if emitting WebAssembly).

However, the output is not an ES module, it is an Immediately Invoked Function Expression, and quite an ugly one at that:

var Module = (() => {
	var _scriptName = import.meta.url;

	return async function (moduleArg = {}) {
		var moduleRtn;

		var Module = moduleArg;

		var readyPromiseResolve, readyPromiseReject;
		var readyPromise = new Promise((resolve, reject) => {
			readyPromiseResolve = resolve;
			readyPromiseReject = reject;
		});

		// ...

		var HEAP8;

		// ...

		Module['HEAP8'] = HEAP8 = new Int8Array(b);

		// ...

		return moduleRtn;
	};
})();
(() => {
	// Create a small, never-async wrapper around Module which
	// checks for callers incorrectly using it with `new`.
	var real_Module = Module;
	Module = function (arg) {
		if (new.target) throw new Error('Module() should not be called with `new Module()`');
		return real_Module(arg);
	};
})();
export default Module;

I think it would make a lot more sense to actually use ES modules under the hood in this case. This would support JS JIT optimizations better, in addition to much cleaner syntax. Moreover, I find that when compiling GNU coreutils, I get significant code duplication (over 25,000 lines).

I am very interested in the possibility of de-duplicating Emscripten code— perhaps with a proper package or module(s). This wouldn't be easy, and I can't guarantee that I would be able to work on a PR.

[sbc100]

Thanks for the feedback. There are efforts underway to try to improve the ES6 modules that we generate. There was an RFC that went out a while back on the mailing list: https://groups.google.com/d/msgid/emscripten-discuss/dc8c4405-60ae-4c02-9bf4-92c89408f48an%40googlegroups.com?utm_medium=email&utm_source=footer

There is a related google doc: https://docs.google.com/document/d/1ccH3NsRQgEc1HXWI_IFAnMbTM0Vn1lP_EhJmdLRzYvs. As well as github "project" for this topic: https://github.com/orgs/emscripten-core/projects/2

One piece of this work has already landed in the form of the new/experimental -sMODULARIZE=instance setting: #22867. It still needs more work but the idea here is that each module could be and instance, instead of an instance factory.

Regarding removing duplication between for programs that include multiple emscirpten-generated modules, that is complicated by the fact that all of the emscripten-generated JS functions are closures over a lot of global state. Each function assumes that it has access to a lot of module-specific globals. e.g. wasmImports, wasmMemory, HEAPU8, etc, etc.

[james-pre]

Hi @sbc100, thank you for the detailed response! It looks like the link to the project gives a 404, but the other information has been extremely helpful.

This will help my use case a lot. I'm looking to use my library, @zenfs/core, in place of Emscripten's FS. Using ESM would make this easier, as at the moment it is very hard to navigate the generated code.

In any case, I appreciate you taking the time to link to all the relevant information and for writing out a response. Please let me know if you'd like some assistance on the JS/TS, I have a lot of experience and would be happy to help.

[sbc100]

Oops, looks like the project was marked as "Private" for some reason. I changed it to "Public".