module: refactor to use iterable-weak-map

[bcoe]

Using an iterable WeakMap (a data-structure that uses WeakRef and
WeakMap), we are able to: stop relying on Module._cache to
serialize source maps; stop requiring an error object when calling
findSourceMap().

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• documentation is changed or added
• commit message follows commit guidelines

[nodejs-github-bot]

Review requested:

• @nodejs/modules

[bcoe]

@joyeecheung @addaleax when I try to define FinalizationRegistry and WeakRef in primordials, I get a compilation error because the WeakRef and FinalizationRegistry are undefined at the time when primordials is initially loaded.

Any thoughts?

[devsnek]

WeakRef and FinalizationRegistry cannot be used with snapshots because they can be disabled using a flag at runtime.

[bcoe]

I think this means we can't use them for a utility like this?

We don't want source maps to stop working because the flag isn't set.

[aduh95]

It's a V8 --harmony flag, that's not supported and undocumented in Node.js. I'd say it's fine to use them.

[benjamingr]

I don't understand how this works - this appears to create a new WeakRef each time the key is set even if the key "value" already exists in the map? Is this intentional?

[benjamingr]

var m = new Map();
var o = {};
var r1 = new WeakRef(o);
var r2 = new WeakRef(o);
m.set(r1, "foo");
console.log(m.get(r1)); // foo 
console.log(m.get(r2)); // undefined

[bcoe]

this appears to create a new WeakRef each time the key is set even if the key "value" already exists in the map? Is this intentional

I'll add unit tests for the IterableWeakMap implementation, and will make sure this case is covered.

[benjamingr]

I'm saying - I don't understand how it works and to my understanding it's not supposed to. A test is fine, but a "this works because X" is also fine :]

[bcoe]

@benjamingr I'm pretty sure you've identified a bug in the IterableWeakMap shared on the TC39 spec page -- we should only create a new WeakRef if there's not already an entry, otherwise we should perform an update.

Once I've fixed this, I'll share the updated code, and can provide a walk through of the logic if it's still confusing.

[benjamingr]

Sure, feel free to ping me when updated and I'll take a look :]

[bcoe]

CC: @mcollina, I noticed a conversation you had on Twitter contemplating using IterableWeakMaps for a similar purpose.

[ljharb]

to call it out explicitly: i think these primordials todos should be considered blockers

[bcoe]

@ljharb @aduh95, how about the approach of exposing the makeSafe from primordials, and freezing WeakRef and FinalizationRegistry when iterable-weak-map.js is loaded:

// TODO(aduh95): Add WeakRef to primordials
const SafeWeakRef = makeSafe(
  globalThis.WeakRef,
  class SafeWeakRef extends globalThis.WeakRef {}
);

// This class is modified from the example code in the WeakRefs specification:
// https://github.com/tc39/proposal-weakrefs
// Licensed under ECMA's MIT-style license, see:
// https://github.com/tc39/ecma262/blob/master/LICENSE.md
class IterableWeakMap {
// ...
}}

source_map_cache.js pretty early on in the application lifecycle, _I think, before any userland code.

[ljharb]

That seems fine, but I’m not sure what’s wrong with eagerly freezing it, like all the other primordials?

[targos]

When the primordials are prepared, we only have access to a subset of builtins. V8 does not provide us the ones that can be disabled with flags by the user.

[ljharb]

ahhhh thanks, that makes sense.

[benjamingr]

Generally look fine, lmk when it's ready and I'll check the branch out and play with it a bit :]

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/34111/

[bcoe]

@benjamingr happy with how this is looking?

[mcollina]

lgtm

[bcoe]

Landed in 8fa9035