Node.js Node-API Team, 12 Aug 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Node.js Node-API Team meeting 2022/08/12

Description

No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).

A

So welcome to the noodle api meeting on august 12th and let's begin with our standard routine and I'm going to share my screen.

A

Okay, can you see my screen.

B

A

Great so the first one is about the full coverage, but I didn't see jack is here, so I'm afraid we don't have any.

A

I don't have any news to report so.

A

If there is any signature before we continue to the next topic,.

B

Now, let's continue.

A

Yeah there's still issues.

A

This is about hungary. I didn't participate in this problem very well.

A

I believe we have um a conversation on this program in these previous meetings. So do we have any updates on this.

A

Oh there, it is.

B

Are there new commits.

A

um No, so the api is still undrafted and there are no new updates and we have no use to share.

A

So maybe we can move to the next topic.

A

Let's see another.

A

It's a palestinian function.

A

B

Yeah, we need kevin to comment tonight.

A

Yeah I didn't participate in these conversations so and I need time to dig into the problem to see what's going on here,.

A

I'll leave it to here if there is no need to report, and if we got some time at the end of the meeting we can maybe we can come back to this. Let's see another one.

B

Yeah, I'm still working yeah, so I think we have basic example on using this proxy, but uh now I'm trying to match with with non fps.

B

I'm going to move this example to examples folder, so I just uh it's still in draft here, but I'm going to close it in this repo and the whole thing will be just as simple. Instead of our examples.

A

Yeah right, so, if I go correctly, the pro the example here is using the proxy.js api right. We don't have any native proxy api file code correctly.

B

Yes, it's just an example which how to use javascript proxy api proxy standard object. No additional node api was introduced.

A

I'm I'm wondering if it is possible to expose the proxy network api, I'm not sure about the xda and other engines if, if they are exposing proxy nato api, but it's a possible solution to it, I mean.

B

B

Enchromis that don't have uh they do have first concept of host object, which which is like proxy like, but they don't have anything which looks like object, template inside of v8.

A

I mean proxy, you see, we have a javascript proxy, but we didn't expose a node api to create proxy directly.

B

A

B

But I don't see much benefit to expand the api for that. Yet.

A

Yeah, I'm I'm just thinking that could be a that could be an alternative, so I will read it to you to see what's the test approach here, if we can do it in the do it with the js proxy api, then grid.

B

Yeah, it definitely works so even even this code, if you look on files changed, uh it's only a unit test is added here and having what we do just using normal node api to create, create proxy and and do the rest from there. So I think it's. uh This approach definitely works.

A

Yeah, that will be.

B

But this code is too complex whatever. Currently it works, it's a little bit simplified. I mean it's more streamlined.

B

Yeah in this example, we do have we see like if you stop here. Like you see, we have a get proxy handler, so at this point, we're just simply defining uh four different uh proxy traps like uh get set on keys and get on property descriptor.

B

So, for example, in our code, where we use json on top of, say, chakra, we in order to simulate this host object. Behavior, we pretty much just do this proxy, just with four methods.

B

They're, usually enough to to simulate object, behavior, but proxy has more proxy, has ability to being called as a function as a constructor, a lot of other things.

A

Is this a javascript proxy? uh Yes, this is a javascript. uh Do we have.

B

Is there an example to show how this looks in javascript like how this is exported to javascript? Is there a js.

A

File: okay, yeah: this is a jstor okay,.

B

Internally, this c plus plus code just simply wraps up um javascript proxy, which is technically a javascript standard. So this approach must work cross engine pretty much any javascript which supports proxy. This approach would work.

A

B

Okay, so here what we are doing is we are creating a javascript proxy and trying to import that into c plus plus. Is it yes? Okay,.

A

So the next step for this pro for this issue is that we you you are going to move this example to the another add-on example repository. Is that right? Yes, yes, that sounds great um and properly, we can add another wrap class in the no diagonal guy. If you see it's beneficial to have.

B

Yes, I think, inside of our c plus plus wrapper. We can add this convenience class, like you said.

A

Yeah and the people are moving from land to knowledge api. It will be helpful if we can, if we have a equivalent class for the nand version of the object, template or something like that.

B

Unfortunately, we can't do like it's not possible to define just some macro and people who using none just automatically works. The reason is actually inside of none they're using directly some uh v8 api, so none is not at least from what I see. None is not independent api. It somehow like extends v8 api.

B

So to my understanding it's not. I don't really understand the purpose of that.

A

um 9 is great for compatibility between different node versions, so people can write one single version of their add-on and they their code can work across different versions. Different variants of v8, since viet api is constantly changing, so they don't have to update their own add-on code since nan is going to and nine is going to give a provider a macros and help functions to alleviate the to to remove the changes between different va versions. So that's the purpose of them. If I, uh if I understand it correctly, so that's the that's, that's it.

B

I understand but uh look in this api. I think it's a little bit deficient in terms of, for example, then we access property like get and setter for property, uh not only only using strings. So if you properly happen to be symbol as a name using symbol as a name, none api will not able to help you.

B

A

Yeah, I suppose that's because of that's that's about the backward compatibility issue since the old version of it in not supporting symbols, but I'm not sure about the exact reason that it's my assumption.

B

You see, okay,.

A

So, let's quick and we'll come back to this. If you have any progress on it,.

B

Okay, thank you.

A

So this mark has done so I'm going to move to the next topic.

A

Let's see if there's anything new.

A

Dino is implementing the laker, but I'm not sure if there is anything we can do here.

A

So, let's wait for nicola.

A

And michael is not here too so.

A

We'll wait for my code to vote.

A

Not sure I don't see anything new in this issue.

A

We'll come back to this later, so this one is.

A

Nicola is not here so we'll wait for nicola for this issue.

A

And let's end this.

B

So about this known scene, language bindings, one of my interests is actually c sharp binding and I had a look at this implementation recently.

B

I don't know if anyone had experience of it, but it seems like project is still in it's not completed. Yet it's very, very early bids.

A

Yeah, it said it's on the in the early stage. I have no knowledge on windows development, so I don't have any internet experience. So I'm not sure I can share any design here.

B

Because I want to have it for say in react, native react, native windows: we want to have modules to be written not only in c plus plus, but also in c sharp, and I was considering uh to have a wrapper for node, api and c, and it seems like this one is not complete. At least it will not able to fit my needs. Our needs react with this project.

B

I may actually at some point I want to kind of have a different wrapper or something for c sharp uh for now that guy.

A

Yeah- and that sounds very awesome if we can have one complete wrapper in the c-sharp.

A

So is there any next step for this before we can move to the next topic.

B

A

It would be great to have findings in this various languages so that we can expand the node api coverage in much broader community support. So, let's move on so and let's check the issue list of papers with node api.

A

I didn't I didn't. I didn't find anything new in the last week and.

A

Also in the let's see not I don't like that.

A

I didn't see anything new in the last week so is now. Is there anything anyone would like to bring up before? We continue to the issue list.

B

Can I ask you a little bit about your pull requests for finalizers uh yeah, so my understanding you, you do implement this finalizer uh queue right.

A

B

A

B

So think, every time we have a finalizer, they put first in the queue and at some point we're just simply draining this here.

A

Yeah so let's see.

A

Here the big callback.

A

So every time a reference has been, its persistence is being collected and there will go back is invoked. The reference is being incurred to the environment's finalizer queue and the enviro, the environment will.

A

Schedule on second pass go back with, with the setup immediate and in the seller immediate go back, it will be the queue.

A

So that's the basic design of the panel cancelable panorama, uh cancellable finalize the second particle back, since we can remove analyzers from the queue and it's defined in the media. We can remove it from the panelist queue if it's deleted early directly by user, deleted or yeah. It's mainly because you deleted early before the before the second pass go back, so it will be removed from the queue and there won't need any wrapping around the second pascal back parameters like we have before, which is quite cumbersome to understand it's hard to understand.

A

So I believe this implementation can, let's see we remove very remote many lines here and just a few lines new. So it can. I believe it can be much simpler. Okay,.

B

Sounds good! uh It's! Okay! If I leave some comments, I have uh I looked into this implementation a little bit and I do have some questions and ideas like, for example, one question: I have like uh your choice of uh data structure for finalizer queue. uh You chose another to set and I wonder why. Why not queue something.

A

um I choose the under unordered cell because we need to remove the analyzer from the queue efficiently, so we don't have to look into the queue to find the item. We need to remove um and I'm very looking forward to your comments to your questions. You I'm very welcome.

A

You can comment in the pr so that we can discontinue the discussion on the pro request on the github. So please feel free to drop a comment in the pr okay.

B

You can see.

A

Here we are removed, it's user delete, the analyzer uh delete a reference before we before we run the panoramizer in the second pass callback, so we have to remove it from the queue.

A

So if we have using a q and the structure here, I suppose we need to iterate over the q to find the atom we need to remove. If I recall correctly, so that's why we should. I choose the unordered set here so that we can remove it efficiently.

B

It might distinguish from some for previous meetings, node api, uh that and so we we did call finalizers according to some order and now it's an ordered like I just wonder: if do you think it's may affect something or not.

A

Actually, with my test with v8 gate is not determined to invoke the finalize the weak callback of the object in an order in determined order, and maybe maybe I can have another con- have another concrete test to verify it since I just just play around with it, prove it and it with a very naive.

A

So I can verify it, but with my preliminary preliminary um test around the order, so that's that's why I added a comment here that the invocation order of the analysis is not determined and we do have a test for the order here. If I recalculate, I just got a problem when I implement this pr.

B

Oh, I think I understand what you've done. You pretty much said like. uh There is no order if we already inside of uh this uh finalizer queue, but we still keep order uh on environment shutdown right.

A

um At the environment chardonnay, we have two kinds of finalizers and we will yeah. This is the. This is the two list of the finalizers. The first list is that the rep analyze the reference that don't have any vandalizers registered, so they can be.

A

So you have to.

A

Run the final to delete the panel to delete the rough list item that have a finalizer first, since they can calling into they can delete other references and they can call into javascript I'm not. No. They cannot according to javascript and shut down, so they they still can delete other references that don't have a finalizer.

A

So we will run the will download the analyze, the ones that we they have a panel registered and then we'll delete the the references or any items that don't have a callback register. So this is the order we are guaranteed to when environment is children and we don't have any other kind of analyzers or another invocation order.

B

Listen: okay makes sense.

A

B

Thanks, let's, let's move on yeah I'll leave my comments. I really like this change and at some point I believe one of my iterations for finalizer changes. I also try to implement with q, so think seems like ideas, kind of circulating around and um definitely it's a.

A

B

Really good direction. I really like it.

A

Yeah that would be awesome. I'm very looking forward to your reviews on it. Please feel free to ask any questions on the github.

A

Is there any other topics on this issue on this pr.

B

No, I don't have questions.

A

Okay, so let's move on to.

A

A

Just wait a minute.

B

Just uh to tell you that one place, I kind of think this finalizer queue could be useful, is uh remember.

B

I had this pr about this reference counting one thing I need out of this reference, counting new reference, counting that it must be able to work from any thread, and if we have a penalizer queue I would be able to even then the ref count goes to zero and when different thread, I will have ability not delete, object immediately reference immediately, but rather put it in a queue, and then we can write thread only at this point. We delete it. So it's some kind of design. I have in mind to utilize uq in the future.

A

Yeah, that sounds a good point, but I suppose that the current implementation of node environment, that immediately is not thread safe, actually, and we do have a threshold version of that immediate in the node core and it you facilitate the meal taxes. So it's still.

A

And yeah everybody is possible and if it's beneficial to your reference content, I do it. It's words to try.

B

Okay, let's continue.

A

um By this one, I didn't have any time to look into this in the last week, but I believe it was worth to explore if we can, um if it's, how uh how adam authors can be benefit benefit from this api and I'm planning to look into this in the next weekend, and since I got some free time, and so I have nothing new to report here, do you have any updates.

B

No, I didn't look at this, but it will be great if you see.

A

Well then, we are moving on to the next topic.

A

um This one- and I have noticed several updates several new comments on this pr, but I didn't have much time to take a look on it and, as you can see, that ana has requested changes on the pr. As as it's about the earth world.

A

If it's worthy to introduce, they introduced embedders into node api, since the node api currently is mainly focused on the add-on assets.

A

I believe I don't have any updates on this issue, since I didn't have much time to look into it.

B

Yeah, I didn't look at this stuff too. You know I I I'll check just see I'm curious to learn about it.

A

Yes, it would be beneficial if you can take a look at the api design, see if it can fit in the hermes engines and another engines. You are familiar.

B

No, I think like uh in terms of uh embedding node.js. This api will be effectively.

B

It will be kind of very node.js specific. I think it's uh that's why I kind of previously I proposed this. This api must be separated from node api. I mean no depend. I mean, like whatever been used by module developers, pretty much api we discussed before. I think it's. This is different set of api.

B

It should be, should be totally different if you like, we want to make it abi safe in terms of kind of abi compatibility and that works between different compilers and stuff, because c, plus plus, is not abi safe between compilers, like, for example, I can't use uh compiled dll on windows. If it's compiled, say in clank, I can't use it really from msvc if it's c, plus plus, because I have using different name c, plus plus name bandwidth scheme, so the only universal one should be c based, but.

A

Yeah, that's difficult! Sorry, yeah go ahead.

A

I mean that's reasonable to me since that's the main topic in this in this thread. So I believe this api is a expected expected to the embedders rather than add-on. Authors.

A

So, since there is no updates on this- and maybe we can move on to another topic,.

A

Jazz inheritance from citrus plus, I didn't see any updates. I mean any new comments on this pr.

A

So do we have any new to on this?

A

B

I don't see any changes here, but um I believe one one kind of more practical way to achieve this inheritance is maybe to use proxies. I don't know something, but uh I'm curious to try. You know we'll see.

A

um Sorry, but I'm not following the point why our inheritance related to proxies.

B

No, I mean like if we want to support some kind of inheritance.

B

It may be one of the possible ways which I believe I know I didn't, try it yet, but uh one one path I want to explore if we can use proxies for that reason, because this this pr, if, if you remember uh it, was the kind of the developer exposing some internal guts of object, template.

B

Yeah, like a function, template yeah, function, template, sorry and we said function. Template is not something that we see in other languages, but the problem probably still exists, like people asking how we can inherit, how can implement inheritance and- and I believe one of the possibilities will be using proxies.

A

But I remember correctly: function. Template cannot inherit from directly from a js value, since the function template can be is created as a context, agnostic data structure and it can be created before any context is instantiated.

A

So I I may need time to look into this. Why it's beneficial for inheriting js from c plus class.

B

A

Definitely don't want.

B

To expose function templates it's what uh gist of this pr. I guess next question. Okay, if we can't use function, template what will be the next best thing to achieve this goal? Can we do something else.

A

I see it still cannot.

A

And inherit a javascript value, it's just the function, template inheriting one another function, template and you that's an that's, not that's not sufficient for the problem. We have, let's say inviting a event target or event emitter, since they are javascript values.

B

Right and the only thing like uh which I I believe might happen: if we're going to utilize javascript proxies, so it's it should be potentially potentially doable. I don't know it's. I really need to try it. First, let's see yeah.

A

B

Finish my example: first before I go to this step,.

A

Yeah sounds great: let's wait for your examples before we can come back to this.

A

And I will just drop a comment here about the fact that we cannot enhance the javascript class from with this api.

A

It may not be met the problem about inheriting javascript classes, I'll do it later, since I I need to sum up some reason on it.

B

And if you see like our last comment was exactly considering all the different scenarios and uh developers say like uh uh some of them is not achievable. Oh.

A

Yes and there kevin has covered in this.

B

We discussed this july 2nd.

A

Yeah sounds great.

A

So these two are your works. I believe you have continued going on this work.

B

Yeah, like uh this finalizers uh will probably proceeds uh so think, first one with about finalizers, so I to build on top of your pr with addition of this feature like we discussed last time, so I have found this feature implementation from from pr remember. We have this issue with get all properties. I had implemented the prototype there, so I pulled it on top put it on top of your code, and so we can actually run finalizers either immediately.

B

If they have, they have no javascript inside or they can actually run as a part of a set immediate.

B

So it will be changed on top of your pr, so it will be not the queue is the queue will be run, not necessarily as a part of set immediate, but uh I think that some of these finalizers will be run immediately.

A

Now we can rename the uh enqueue to something more agnostic about the enqueue like say.

A

um Sorry, I didn't come up with any name immediately, but I believe we can rename the message. If you, then it's not describing the condition very well right.

B

And second, one about this, references is mostly I I do want to have. I want to kind of complete it. The one thing which I'm missing right now is actually weak, references which I want to have.

A

Sorry, weak references.

B

A

um About the week, references- and I um I just want to mention that um there is a not api implementation on the edge devices, they don't support, weak references and as now since they their engines are depends on the reference content and garbage collection. So they are asserting.

A

A

If the reference count goes down to zero, they cannot and.

A

Basically, they are going to release the value when the reference cannot go to zero and.

A

And they will no longer get the value once that's once, and so there are no weak reference in their implementation.

B

Okay, let me ask a question so think: the weak reference, I'm I have in mind it's something which is like part of a javascript language standard. uh It's not it's, not the s6, but I think it's upcoming standard they're introducing something called a weak reference. I believe.

A

Yeah but we don't have any native api for those equivalent to javascript. We create similar things.

A

I'm not sure these engines are exposing it, since, um if I recall correctly there, you know any weak reference api.

A

B

Yeah they do have a big craft support.

A

Yeah they do have a big reference, but I need to take a look and but um I mean yeah, there is no api to create a reference in the native api. You have to create in the javascript.

A

A

B

I'll get back to it and uh see thanks for your comment, uh but I totally agree with you. I don't really like our current reference at all. The fact that uh the name of the account kind of reference count goes to zero, we're becoming some kind of magically some weak reference.

B

I think it's. uh I don't really like this design, because I don't understand it and it's very difficult to explain what I really think we should be doing. Instead, if actually separate this problem into two different things, one of them say from c plus plus they want to reference a javascript object.

B

So simple reference counter works, just fine right in terms of if reference count goes to zero, we just kill our reference and automatically in earnings file, but we have another problem which I believe, what caused this reference to be more complex is how to reference from javascript how to reference uh c, plus plus objects, and uh in this this scenario we have uh two things to consider: one of them. We need to identify c, plus plus objects, and this javascript object goes out the scope, so we can actually delete this resources and second one.

B

We must be able, from c plus plus, say we don't want to have this reference anymore right. We we kind of want to destroy it on our own, and I wonder if we can somehow separate this concept of reference counting for one end and referencing from another end.

A

I believe the current behavior is just mimicking. The v8 persistent behavior since feared persistent behavior can be and we can set weak and set the position to a weak, persistent and we, when the fundraiser quebec is not present, then we just depend on the garbage collection to see if we can still reach to the javascript value.

A

So that's what I can share on the behavior of the of the api resonances.

B

Right, but if we look at the even v8 behavior v8 does not have ref counting there right. We have either persistent object or it becomes this weak object.

A

B

But when we reference it from c plus plus we for persistent object, we implemented ref counting right, but for we call it, we have no ref counting.

A

Yeah, we don't have reference content in the javascript weekly reference.

A

Honestly, there is no weaker. There is no reference containing the javascript in the javascript language, since that's implementation defined behavior.

A

But um so the next step for the pr is that you are still looking into the um I'm, not sure if you, what which, which which approach you are choosing to do as the next step.

B

Yeah like uh let me let me work on the next uh week or two and uh I'll definitely.

A

Yeah, it sounds great, so we'll come back to this next, I don't.

B

Have solution yet it's just still thinking.

A

Yeah, that's, okay. I don't have a solution. I just I just comment about my thoughts on this.

B

A

um I don't have any new on desktop on this document.

A

And at the park do you have any link to this document if you would like to work on c plus class? If so, I can share this document with you and you can pick up any part of your interest. Oh yes, sure, oh yeah, please share it with me.

A

um Can you put in the chat.

B

A

And yeah, okay, I post it on the chat. Okay. Thank you.

A

Thank you very much. So it's all about the optimization point we can do in the node api implementation in the core and we are currently focusing on the finalizer part you can possibly you can find a.

A

If you would like to work- and do let's say the refactor on this function, to adopt analyzer interface can be a good path to pick up, since there can be, there is no functionality change and it's all about the code refactoring. Is that any open issue? For that?

A

I know we just have a issue check the whole consolidation, okay,.

A

Yeah we just have a checking issue for the whole project. Okay, we don't have and separate issue for each atom.

A

B

Me take a look at it. Thank you.

A

Yeah, if you are interested in the implementation, please comment in the issue so that we know you are working on. It shall be quit.

B

A

A

um I don't have anything else to discuss. Do you have anything else to discuss before we close.

B

No, I just want to share that I'm kind of um looking changes in hermes engine api and they recently added a big implementation.

B

So it feels like we should be able to have um more complete node api implementation for hermes soon. So.

A

That sounds awesome. It sounds like.

B

Quite exciting.

A

Yeah, the a complete support of another api for armis is definitely helpful, so I don't want others to adopt, not apa.

B

And also I'm going to have, I finally been approved for the repo uh which implements gsi on top of node api gsi is actually in c plus plus based interface used by uh implemented by meta inside of react native, and it's been used inside of react native community.

B

uh So I do have this gsi on top of node api implemented already, uh but it's currently in 2d in three different reapers. So I don't have one single ripper just for that for that project.

A

Yeah sounds great: is there any um let's say links that can be shared publicly.

B

A

B

So this repo is, it's not public, yet it's going to be public soon as soon as I put it there. But uh if, if you talk about code itself, code can be found today inside of our react native windows, public reaper.

B

Can probably should, should I post it in a in a chat just just if somebody listen to us, I want to people kind of be aware.

A

Or you can also about an issue in the api, stable node, so that we can check it exit in this repo so that we can regularly check if there is any problem or there is anything new. We need to discuss on your implementation in the jsi. If I call it correctly, so you can regularly check the implementation and to see if there is anything new to discuss or is there anything behavior? We should to revisit.

B

Okay sounds good yeah, because some of my additions like, for example, serif, counting nodes and stuff many of them is coming from requirement to implement gsi on top of node api, because gsi had some requirements to api and so on. That's why I needed to extend node api and some of these extensions, which I have. um I want them to be part of the core. No no of node api.

A

Yeah, I believe, that's very valuable for us to discuss the comparability of node between different engines.

B

And so thank you yeah the call implementation just two files like a header file, cvp file. I posted them inside of this inside of chat.

A

A

Yeah great, I maybe I can take it with it- take a briefing to look into it like look after the meeting. That's.

B

Great will be nice appreciate it.

A

So before we close is there anything we need to discuss.

B

No, that's it. That's it from me.

A

Great have a good weekend, bye-bye.