Discuss blocking on Zones proposal

[bmeck]

Moving here, as per: nodejs/CTC#166 (comment)

https://github.com/domenic/zones is a proposal that is currently being blocked by Node.

The proposal is something similar to domains and async_hooks in creating a way to pass data through asynchronous tasks.

The blocker has been the handling of errors with the proposal.

I would like to setup a meeting to discuss the blocking behavior now that async_hooks is landed.

[Trott]

@nodejs/async_hooks I guess?

[AndreasMadsen]

We can setup a meeting in the diagnostics WG.

Does this have any relation to angular/zone.js#795? /cc @JiaLiPassion

[JiaLiPassion]

@AndreasMadsen , thanks for cc me, the nodejs/CTC#166 (comment) seems about error handling of zone proposal, it is not directly related to angular/zone.js#795, but I would like to provide any help I can.

[bmeck]

@JiaLiPassion Is the following no longer a goal?

5. Error
   use process.unCaughtException to redirect to ZoneDelegate.handleError.

[JiaLiPassion]

@bmeck ,

1. use process.unCaughtException to redirect to ZoneDelegate.handleError is still a goal, to use async_hooks to implements zone, this is the only way I found to handle error.

2. currently the [WIP] feat(async): fix #740, use async hooks to wrap nodejs async api angular/zone.js#795 is stopped because I can't find a way to implement zone aware promise with current async_hooks API.

[mcollina]

I am 👎 on changing process.on('uncaughtException') default behavior: crashing the process. It should be possible to enable zones if someone wants to, but a lot of modules and applications out there are developed thinking that an error will crash the process.

Generically, I would prefer that zones would create a new context that does not influences anything that happens outside of it.

[JiaLiPassion]

@mcollina , thank you for your advice, currently, my thought is just use async_hooks to simulate original zone.js monkey-patch behavior.

• with zone.js monkey-patch version. zone.js intercept all async callback, so it will be able to catch all errors.

• with async_hooks which is just a notification not an interception, so it is not possible to catch any errors.

So currently I just thought maybe with process.on('unaughtException'), it is a way to do what currently zone.js does.

[AndreasMadsen]

Generically, I would prefer that zones would create a new context that does not influences anything that happens outside of it.

Unfortunately, we don't have a mechanism for that. Domains works by catching the error beforeunaughtException, likely zones will work that way too.

@JiaLiPassion if you want to experiment with it, you can overwrite process._fatalException or process.domain._errorHandler. it is not a public API so there are never any guarantees but process._fatalException especially is not something that is likely to change.

[JiaLiPassion]

@AndreasMadsen , thank you ! I will try it.

[MylesBorins]

Hey all, is there an update on this? should we consider a strategic initiative around it?

[AndreasMadsen]

@JiaLiPassion We now have a public API for catching error the way domain does it. And we have migrated the domain module to use the API.

• docs: https://github.com/nodejs/node/blob/master/doc/api/process.md#processsetuncaughtexceptioncapturecallbackfn
• domain: https://github.com/nodejs/node/blob/master/lib/domain.js#L103

Additionally, there is a general tracking issue for implementing domain using only public APIs nodejs/node#17143

[JiaLiPassion]

@AndreasMadsen , Thank you for the update, I will look into it. And I will also check the never resolved promise issue.

[AndreasMadsen]

@JiaLiPassion we now have the promiseResolve event. so that issue should be possible to solve now too :)

[JiaLiPassion]

@AndreasMadsen , got it, I will check it, thank you!!!

is the new promiseResolve can handle this case?

 const p1 = new Promise(() => {}); // This will never be resolved
 const p2 = new Promise((resolve) => setTimeout(resolve, 1000, p1)); // resolve p2 with p1 after 1s

I will test it.

[AndreasMadsen]

@JiaLiPassion I'm not sure it can. I think it just emits the first resolve value. It would be smart if we could know the "resolved" value.