Stage 2.7 roadmap

[nicolo-ribaudo]

Before Stage 3, we still need to investigate multiple aspects of the proposal. Each topic should be discussed in its own issue, but I'm writing all of them here to keep track of the progress.

---

TODOs

• Interop with WASM modules (How does this impact WASM modules? #22)
  The current WASM-ESM integration proposal makes WASM modules async from JS's perspective. This meant that their execution would never be deferred. Can this be changed?
• More performance analysis
  We need to investigate more the performance benefits of this proposal, and collect more data. Additionally, we need to explain why top-level await suddenly causing a module to not be deferred anymore is not a performance footgun.
• Early errors (related to import * potentially doesn't need to parse/link #9)
  How critical are early errors? Is it ok to defer them until execution, similarly to what happens with dynamic import? This would allow some systems to entirely defer loading modules.
  EDIT: After talking with people working on various runtime implementations, it's clear that those implementations would still be able to avoid loading by pre-computing metadata about files.
• Deferred re-exports (related to Bloomberg Feedback #18) (Add support for "optional" re-exports #30, Add support for "defer" re-exports #31)
  It would be useful to support something like export defer { f } from "./foo", that causes the execution of ./foo only if f is actually imported form this module.
  EDIT: The surface area for re-exports is much larger due to the tree-shaking semantics, I plan to ask the committee to leave it at stage 2 for longer.
• Dynamic import syntax (PR: Add dynamic import.defer() syntax to mirror import.source() #28)
  Given that import source is now stage 3, we should probably add dynamic import.defer() to this proposal similar to import.source().

---

Reviews

• @guybedford (note: Guy was initially a co-author of the proposal with Yulia, but both of them stepped down from active development long ago during Stage 1)
• @dminor
• Editors
  • @bakkot
  • @michaelficarra
  • @syg

[nicolo-ribaudo]

• @bakkot
• @michaelficarra
• @syg

I was planning on proposing this for Stage 2.7 at the April meeting, and I remembered just a few days ago (because I was pinged for a review) that editor reviews are needed for Stage 2.7 and not for Stage 3.

The spec text at https://tc39.es/proposal-defer-import-eval/ is complete and is what I plan to propose. Regarding the open PRs in this repo:

• Throw when doing import defer * of a module with no exports #34 is for a possible minor change to how evaluation happens that I need to run through the committee, mostly because I don't have a preference on whether we should do it or not.
• Add support for "optional" re-exports #30&Add support for "defer" re-exports #31 are something that I will ask to the committee to "leave behind" as a separate stage 2 proposal, because while working on them I realized that they are very different from import defer. They can be thus ignored.

I know that this ping is coming way too late, so I understand if it's impossible for you to take a look before plenary :)

[guybedford]

My review can be considered complete after #38 is resolved, with the PR in #39 and #35 already landed. The implementation approach is really neat, nice work.

[dminor]

This looks good to me, thank you for the good work here.

[nicolo-ribaudo]

@bakkot @michaelficarra @syg I'm planning to re-propose this for 2.7 at the next meeting, I would appreciate if you could find some time to review the spec text.

If you already took a look at this last time, there have been two PRs to handle a bug that @guybedford found with re-entrancy and top-level errors: #39 and #43

[dminor]

I've had another look, I think this is good for 2.7.

[michaelficarra]

@nicolo-ribaudo Why do you use an early error to restrict ImportClause in ImportDeclaration to NameSpaceImport when defer is present? Why not just add an alternative to ImportDeclaration?

[michaelficarra]

@nicolo-ribaudo There's also a misspelling of [[Specifer]] that occurs multiple times.

edit: I'll just turn this into a review feedback comment.

• evaluatePromise doesn't need to be declared in step 8.c. You can have more than one declaration (at the points where you currently assign it) as long as all paths to the first use pass through exactly one declaration.
• I don't like that we're using an out parameter in GatherAsynchronousTransitiveDependencies. It looks like we can easily avoid it with some list-concatenations, right?

[parzhitsky]

@nicolo-ribaudo If you don't mind me asking, is 2.7 an arbitrary number between 2 and 3? Why "Stage 2.7" and not, for example, "Stage 2.5", or "Stage 2.1", or "Stage 2.9"?

[nicolo-ribaudo]

@parzhitsky You can read the discussion at https://github.com/tc39/notes/blob/main/meetings/2023-11/november-30.md#continuation-of-the-new-stage-discussion