Unflag Top-Level Await?

[MylesBorins]

We have top-level await flagged on v14.x / master behind --experimental-top-level-await.

I think we should consider unflagging, but keeping a warning when it is used the first time.

Thoughts?

[richardlau]

For clarification, this is unflagging on 14.x or the next semver major?

[devsnek]

I don't think we should unflag until it is stable in V8. I think as a policy we should never enable harmony flags by default.

[jasnell]

Have to agree with @devsnek ... given the dependency on V8, I think we should hold off.

[MylesBorins]

@jasnell I can confirm with the V8 team but my understanding is that the implementation in V8 is stable. Currently what is blocking the stability upstream is the HTML integration

[MylesBorins]

@richardlau I was suggesting to unflag for 14.x. As TLA was specified it's enablement should have 0 observable changes if it is not used.

[devsnek]

@MylesBorins to be clear, I realize the tla implementation itself is stable. the problem is that it is flagged as an unstable feature, and I don't want precedence of messing around with that.

[MylesBorins]

@devsnek while I can understand that, we can be explicit that this is a one-off, not a precedent. This feature is a huge boon to the ecosystem and if what we are shipping is highly unlikely to change I don't think we are benefiting from keeping it flagged.

[guybedford]

I'd be strongly in support of this unflagging as a specific one-off exception for v8 unflagging. Just yesterday I was using this flag to great benefit myself - it's such a common scenario and help to Node.js users I think it would be a massive win for our users that is long overdue already.

[syg]

The V8-side implementation of TLA has been staged since May, so it's been getting clusterfuzz coverage for 2-3 months. From that perspective we can be more confident that the V8 part of this is more stable (in terms of bugs) than not. Note that it is still not as stable as on-by-default.

The current hold up to ship is HTML integration, namely, working out timing issues around firing events like onload and onerror for module scripts, as TLA moves work to the microtask queue.

IOW, the thing preventing us from shipping and being iterated on isn't in V8.

[bnb]

I'd be strongly in support of this unflagging as a specific one-off exception for v8 unflagging. Just yesterday I was using this flag to great benefit myself - it's such a common scenario and help to Node.js users I think it would be a massive win for our users that is long overdue already.

Expanding on this, when first starting to use async/await TLA was one of the big gotchas I ran into. It is a genuinely good piece of DX and as the ecosystem starts moving forward it would be particularly positive, especially as we're moving into a point where a large amount of tooling and content is about to be generated about using modern features of the project, I think it would likely better for the ecosystem's advancement if we can help enable people to use this sooner rather than later - especially if it is relatively stable (despite being unmarked as such) in V8.

Given the unflagged experimental support for ESM and the scoping of TLA to the module target (AFAIK?), I don't think it's a particular stretch to enable it.

[robpalme]

I agree with @bnb and @guybedford that now is the right time to unflag - specifically whilst we have a small user group on ESM.

It ensures that any Node-specific integration risk for TLA affects only brave users. And it also acts as a carrot for would-be ESM users to try ESM out before Node declares ESM non-experimental and widespread adoptions kicks in.

[mmarchini]

I'm also +1, the benefits for our users outweigh the risk of this becoming a precedent (and as @MylesBorins said, we can explicitly say this is not a precedent but instead a one-off).

[jasnell]

I certainly won't block given that it appears to be fairly stable in V8, but I definitely don't want to set any kind of precedence with regards to shipping with harmony flags enabled. This would definitely need to be an exception.

[devsnek]

it might be good to establish 2-3 people (I can be one) who can quickly upstream patches to V8, given we're sort of taking on any potentially broken behaviour here as our own.

[mmarchini]

I can open PRs upstream if necessary (I don't have commit access there though)