Normative: Add Source Phase Imports

[lucacasonato]

https://github.com/tc39/proposal-source-phase-imports

[lucacasonato]

The exact layout of this record depend on #3057. There is cleanup to be done around how and where exactly phases are passed around (and if they are part of this record), but this will be less merge conflict prone once Nicolo's PR is merged.

The PR is technically correct as is, so I'll defer this work until Import Attributes is landed.

[michaelficarra]

You'll also have to add [[Phase]] to LoadedModuleRequest Records.

[ljharb]

i'm not sure we have the concept of an "abstract" class constructor; can't it be a normal constructor that throws both when [[Call]]ed or when it's its own new.target?

[bakkot]

@ljharb Do you mean you want it to be possible to do like

let AbstractModuleSource = /* some coe which gets AbstractModuleSource somehow */;
class A extends AbstractModuleSource {}

new A()

and have that not throw?

If so, that can't be done while keeping the property mentioned above, that these things are supposed to all have a [[ModuleSourceRecord]] internal slot holding a ModuleSource Record - there's no way to provide one when constructing from user code.

---

Or do you just mean you don't like the wording "abstract class constructor" here, and are OK with the behavior but want different wording?

[ljharb]

The semantics are fine, i indeed just mean the word "abstract" doesn't appear in the spec in this context afaik, and I don't think we should introduce it short of a proposal to add some kind of language feature for that.

[michaelficarra]

can't it be a normal constructor that throws both when [[Call]]ed or when it's its own new.target?

Yes, we should do this, and we should get rid of

The module source constructors do not perform a super call to it

[syg]

What is the resolution of this? The current version I'm reading has an %AbstractModuleSource% constructor that unconditionally throws for both [[Call]] and [[Construct]].

[lucacasonato]

The intention here is the following:

• User code can not [[Call]] and [[Construct]] the %AbstractModuleSource% constructor.
• Host objects can inherit from %AbstractModuleSource%.
• Host objects (like WebAssembly.Module) that inherit from %AbstractModuleSource% should be able to be constructed by user code (if the relevant host object constructor allows this).

[ljharb]

User code should be able to class extends AbstractModuleSource, though, otherwise you can't self-host a JS implementation and get the right internal slot.

[guybedford]

Since only hosts are able to implement custom module sources, and not user code, this is explicitly designed to ensure class extends AbstractModuleSource doesn't work.

[bakkot]

Below this is spelled "Module Source Object", rather than "ModuleSource Object". It should be consistent (and ideally <dfn>'d).

[guybedford]

Added a space in tc39/proposal-source-phase-imports#52. I don't want to call this a definition since these are a class of objects as opposed to a specific object here.

[bakkot]

@ljharb Do you mean you want it to be possible to do like

let AbstractModuleSource = /* some coe which gets AbstractModuleSource somehow */;
class A extends AbstractModuleSource {}

new A()

and have that not throw?

If so, that can't be done while keeping the property mentioned above, that these things are supposed to all have a [[ModuleSourceRecord]] internal slot holding a ModuleSource Record - there's no way to provide one when constructing from user code.

---

Or do you just mean you don't like the wording "abstract class constructor" here, and are OK with the behavior but want different wording?

[jmdyck]

Rather than making a bunch of suggestions here, I've created a PR against this PR's branch: lucacasonato#1

[jmdyck]

In ExportEntries(), there's an invocation of ModuleRequests that needs to accommodate this change in the return-type.

[michaelficarra]

Suggested change

	<p>It returns either a normal completion completion for the ModuleSource Object corresponding to this source Module Record's source phase (<emu-xref href="#sec-module-source-objects"></emu-xref>), or a throw completion.</p>
	<p>It returns either a normal completion containing the Module Source Object corresponding to this Module Record, or a throw completion.</p>

Assuming we <dfn> "Module Source Object" below.

[michaelficarra]

What does "record" mean here?

Suggested change

	<p>Hosts may define their own Module Source subclasses for custom module record types.</p>
	<p>Hosts may define their own Module Source subclasses for custom module types.</p>

Also, wouldn't they be %AbstractModuleSource% subclasses?

Suggested change

	<p>Hosts may define their own Module Source subclasses for custom module record types.</p>
	<p>Hosts may define their own %AbstractModuleSource% subclasses for custom module types.</p>

[lucacasonato]

I guess we're trying to say "Hosts may define their own %AbstractModuleSource% subclasses for custom Module Record subtypes/subclasses/subrecords."

Not sure about that last word.

[syg]

lgtm % some questions.

P.S. Be sure to run npm run format.

[syg]

Speaking as a delegate, not editor, I've gotten feedback that there is a somewhat strong intuition that source refers to unparsed code. Are champions still open to bikeshedding the phase name?

[Jack-Works]

the unparsed source will be import.asset in the future. I don't know if there is a better name for this

[lucacasonato]

Yup. Wasm uses module, but within the module harmony group we are of the opinion that module refers to a module instance (source + context).

This is reflected in the designs for Module expressions, Module declarations, and Module loaders.

[syg]

Internally in V8 we bikshedded handle and instantiable.

[syg]

What is the resolution of this? The current version I'm reading has an %AbstractModuleSource% constructor that unconditionally throws for both [[Call]] and [[Construct]].

[jmdyck]

Reminder that I created lucacasonato#1 a month ago.

[jmdyck]

Suggested change

	String
	a String

[guybedford]

This is now superseded by #3492.