Another requirement would be to always return the same mapping.

or in the opposite direction, pass the module record to it.

Updated with a parameter module records and the result list must contain the same mapping for the given module record.

I'm not sure I understand this change. I thought the goal of this PR was to make is so that all modules have the same top-level mapping, while this PR is now defining the top-level mapping as potentially different for each module.

This allows hosts to add mapping entries based on the origin of the module records. The goal of the PR is allowing hosts initialize their entries, instead of deriving the mappings based on the originator while an imported module is only evaluated once.

So this allows hosts to implement both fresh-context-per-module and module-based-on-importer?

Not sure what you mean by module-based-on-importer, but my understanding is that if two different importers import the same module, you'd still only evaluate once, and any importer dependency would resolve based on one or the other (presumably the earlier one).

Right, the original motivation of this PR was to give an answer to this question: what is the context of the module in this case?

variable.get(); // undefined

variable.run(1, () => import("x"));

At the top-level of x, does variable.get() return undefined or 1? This PR as-is does not give an answer, and instead laves it host-defined.

To give an answer, this PR could either:

• not do anything, and import("x") would see 1
• have a non-per-module top-level snapshot (stored in the realm record?), and import("x") would see undefined

With this PR both behaviors are possible:

• if the host implements HostGetTopLevelAsyncContextMapping(module) as "Return AsyncContextSnapshot()", then it's 1
• if the host implements HostGetTopLevelAsyncContextMapping as "Return the original top-level snapshot", then it's undefined

FWIW, HostHasSourceTextAvailable has a similar requirement worded as "It must be deterministic with respect to its parameters. Each time it is called with a specific func as its argument, it must return the same result". It might be worth rephrasing to be consistent with that language.

I still think this requirement should be deleted, since it's completely redundant with the signature. Looking through the other host-defined ops, I see a few with non-unused returns (e.g. HostHasSourceTextAvailable and HostResizeArrayBuffer) and none of them have this sort of restatement of the signature as an enumerated requirement.

This host hook is only called once per module record, so this requirement is not necessary.