Node.js Node-API Team, 10 Jun 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Node.js N-API Team meeting - 10 June 2019

Description

https://github.com/nodejs/bootstrap/issues/114

A

So welcome to the napi tea meeting for the 10th of June 2019. We had a little bit of a slow start as we worked to get the streaming going, but we're up now and we'll go through the regular agenda. Does anybody have any announcements or anything they would like to share.

B

C

Not here, okay,.

A

So I'm going to open up the the repo and the milestone as we go through so okay I have to get to the right one. Of course,.

A

And just before, I do that I also want I will share that, but milestones share.

A

So that I, don't forget, I want to flip back to the update that was sent him.

A

So I'm, just looking at my email and Kevin, sent me this update because he wasn't gonna be able to make it today.

A

A

Nope, it's not showing me the right thing, sorry, I'm still finding that email from Kevin there we go okay, so he just wanted update us that he's created the PR for a sync worker and the optional callback, and they be asked that we talk about the comment that he had recording. If there's no function provided an API, M must be provided.

A

He also responded to Gabriel's comment for using, regarding use cases for threats, a function with no function, callback and no doubt on API this for it's a function, PR code review is code. Reviews were waiting on poll number four for two and it says I've cleared the docs, but need to make some changes into the order prior to creating a PR. On note, on add, add on API. So it sounds like he's making good progress on that. Maybe we should start with this question so that we don't forget about them.

A

The first one is the discussed comment. If no function is provided, an API M must be provided so with anybody for anybody who's not up to speed I guess it turns out that we were. We were making one of the parameters optional, but it was through that parameter that we were getting the Nappy the napi m1 needed. So one of the suggestions was just to make it so you needed to pass the end API M as a parameter instead in those new signatures, and the question is if anybody has any concerns about that or no.

C

It's absolutely needed I. Think because you know in the kaabah like you need to be able to make an API calls. Otherwise it's it's sort of emasculated. You know if you can't make an API calls in the callback yep and.

A

I know I can't think of any downsides to requiring that when you create the object, you need to have an MS yeah.

C

Although let me just think like you, you will get an end from an API though so so the wrapper for the callback should get an end, so you don't necessarily need to store the end right. What would you what would you use the store the end for, but.

A

Here let me pull up the issue.

A

C

Okay, if okay, no no I I, do get it, though, like you can't even create the thread safe function without an end right so as before, because what what the the semantics of a threats a function are? Here's a here's, an a p-value which I know is function, value, you'd, write and I. Give.

A

C

To you- and you give me back another function, which can be called from any thread that sort of the transaction would not be create thread safe function. Now, if I give you nothing, then you know it is still possible to call nappy, create thread safe function on the sea side right, but but on the node on API side, you were never passing an end. You were passing in a function which was which until now, what's really required right, but the function contained and then right.

C

So you were getting you're passing in the end implicitly, whereas now, if the function is optional, you're cutting out both the function and the end, and that's not good right.

A

Oh yeah yeah, we've always within every object in no data on API. We store the M so that so I wouldn't want us to make this like the only one that didn't. Even if there wasn't some other good reason.

C

Well, the reason we don't pass the end and and we stored the end, is because yeah we always associate and know that on API we always associate the end with with with the value, so there's always nappy and nappy value together and then wrapped into some class depending on there and what the net value contains right and but but in order to call the sea level API, you need the end explicitly right. Yeah.

A

C

So if, on the c++ side you pass nothing you you don't pass a nappy colon colon function, then you don't have an end at all right, yeah.

A

C

Optional yeah, so I, don't know if there's any other parameter that that we pass to nappy, create thread sake, function which could have one and implicitly stored in it, because I think the function is the only like nappy value that we pass into the create. But.

A

In this case, we're talking about new async worker or signatures so like you can see the code I'm sharing like it's. It's an example like right here, so it's we. We were that the complaint was like for async worker that you had to pass in a function, even though we never needed it right.

A

C

Oh wait a minute. Are we talking about threats, a function, our async worker, AC.

A

C

My understanding is that we're making the function optional for threats a function, not racing worker well.

A

This is, this is different. This is 49, which is around actually adding new signatures, which don't require me.

B

A

Right and it's in that round,.

C

A

The question you know the discussion was: hey: okay, we're removing the function parameter. So you don't need to you know. You now have new signatures where, if you don't want to use this function, you don't have to pass.

C

A

People were just passing down.

C

A

C

Yeah, this is the exact same situation as with red state function. That's why I was confused yeah yeah yeah. We, if you want to avoid passing the end of explicitly you need to be passing something into the constructor that that is, is a subclass of nappy, : : value, because nothing else.

A

C

The EM died it and and I mean I guess. The point is that you don't want to pass anything that is derived from that be : : value, because you may just want the the callback to just do something purely native, and so so then. Yes, you have to pass the end, because, because the the the constructor for the async worker or for the thread-safe function has to in turn called the N API for for creating the lower-level object, and that requires an end. So there's no yeah.

B

C

No way except to pass the end, but we need not pass it as a nappy underscore and we can pass it as an at b, : : n, because it's pretty much the same thing. I think.

A

They end up being the same thing right, yeah,.

C

Yeah but I mean for for yeah absolutely, but for consistency with with with I mean for I, don't know for I guess for our static reasons. We should pass an Appy call and call an end, because we don't want to allude to the fact that that you know you need the lower level data types in order to use the higher level bindings. Yeah.

A

A

Am trying to see if there's somewhere else, that we're doing the same thing where we're just passing the end.

C

A

I think, actually that's the way it's documented.

A

I think the so there's the doc. So the dog.

C

Well, you know all the Constructors for all the different values there would have to be an end of the past at some point like you know, can like, if you call objects, calling calling you right.

A

Right but I'm, looking in in the code calling em.

A

Yeah I see okay, so here's like proper disrupter. It is a nappy corn, : mmm yeah.

C

Actually, I added that, because originally property descriptors didn't have didn't, have the nth parameter, but for good reasons you need to pass it. So that's how that ended up. Okay,.

A

So that needs to be a comment on this is that it should be yeah.

C

A nappy, :, : M, it doesn't make a big difference, but it may in the future. It.

A

Definitely looks better in terms of for.

C

A

Just going to comment, these should be.

C

Maybe you may want to comment the line above because this is actually the the initialization of the constructor parameter rather than okay.

A

C

A

Okay, but otherwise we agree that that's yeah, adding that em just makes sense. Yep.

C

A

Sounds good otherwise, I guess I think that's what he wanted to ask about specifically because he had that question, but then we should also go in and review if we can yep. Okay. So moving on to the next question, the next one respond to Gables request for comments regarding use cases for thread-safe functions with no function call back. So do you have any thing on that? You want to share or talk about yeah.

C

Basically, my impression is from from this first issue in the milestone. Is that we're pretty much everything that's currently in experimental in napi has very very good reasons to be stabilized or to be declared stable because everybody I pinged, there are three items: I forget the third one, but but we have date and time we have the threats, a function of being being optional, and then we have a third one. I forget which one that is is.

A

C

Yes begin yes, so these reasonings somebody.

A

Comment about big and not working on some platforms, but I assume we have testing which work must work across all of our supported platforms.

C

Yeah yeah, I, I, guess I'm, not sure version eight of node I believe has begin right so like on the on the engine side. So so, if they all have begins, then we can go ahead. If not, then we need to think we need to think carefully, but.

A

Yeah, that's like I'm, not sure that it does, but we can check that well,.

C

Yeah, quick: we can do a quick check, yeah.

A

That was going to be my next question. Is there are these all back portable if wanted yeah exactly.

C

Yeah yeah yeah. Let me just quickly download it because I've had a million versions of note, but then I I installed everything note version is 8.16, so no, then then const x equals 5 and semicolon syntax error invalid or unexpected token. This is a problem right.

A

C

Me if I'm wrong, if, if the, if a sequence of numerical digits ends in an n, then that means big right. Okay, so yeah, it doesn't have begin right. It's a bit of a problem so.

A

That would be I mean it doesn't mean we can't put them all into 5 and 9 experiment. We'll just means that 5 can never be backported. That's.

C

A

So that's where I think we should figure out like. Is it because the optional the I mean the optional argument can clearly be backported right, yeah.

C

Yeah yeah yeah, that's yeah that doesn't require buying from anything except notes and.

A

Then the others, what we're sorry, what were the 3 again? Can you remind so.

C

Date time, that's, that's! That's a no-brainer because I mean that's, that's just basic JavaScript. So so that's not a problem. The threats, a function optional is also not a problem. Yeah, the only one. That's that's not back portable is begin right.

A

So the question would be: does it make sense to wait on that one and do a 5 with just the other two or.

C

A

You get it well.

C

You know if we, if we include, begins in five, then we cannot claim that 5 is available on all LTS versions of no no.

A

C

That's always gonna be.

A

The case like I, don't think we can always we're not always gonna be able to back for it. I mean I'm, not that's, not impossible, I mean I, don't think we can always ensure that you know a new version of an API is going to be back portable to all LTS versions right. It may be still something to try and do if we can write well.

C

I well yeah, exactly yeah I mean as a policy. It would be good to to sort of support all LTS versions, bright.

A

C

At all possible, with a new release, so I mean you know: I cannot predict what what new proposed does we will have for api's, but I mean if we, if we split this into two releases and then had six, contain solely big int and and and available on ten and twelve, that that would be of an option, and then we would say that five is available on all LTS six is available on ten and twelve. We could go that way or we could experimental, though, though it does sound like there is serious usage of it.

C

So it you know in terms of our criterion whereby you know real world usage has to be evident. It I believe it meets that criteria right so and.

A

C

In it begins as in ten, is it I believe so.

C

A

Mean that's just the end of the year right, yeah.

C

Yeah, let's see so node context, yeah, that's fine, tens! Fine! It.

A

Just would mean by until the end of the year like if we did a sex, that would basically mean then we can't do. We can't back port anything else. That is a that actually requires a version pump right, yeah.

C

Yeah true yeah yeah, so so so we wouldn't, we would be bucketing everything everything that shows up potentially in an api, but until the by the end of the year into experimental. Well,.

A

We could still make a seven and an eight and like in master and tender, saw it just couldn't be something that we go all the way back right.

C

Yeah so yeah yeah, I, guess I, guess you're right. If we do release, if we do stabilize begins as six or even as five it doesn't matter. At that point, we we are sort of forcing ourselves to wait until eight goes out of scope and is no longer supported before we can ever again claim that something is support.

A

Is across yeah yeah.

C

So yeah mmm but I mean okay. Let's not forget, we do have four, which is supported across all LTS and yes,.

A

C

Fairly comprehensive yeah- and you know there is X I mean you know the I guess. The the folks who are using begins in the real world do not expect to use the Gynt on eight because they know that the engine doesn't support it so yep. So it's not going to come as a surprise that we don't back port it. There.

A

Is the other issue where we were talking about there's actually, two definitions of nappy version, the one in node version and the one in just something: API yeah.

C

Jess native API, we.

A

Can leave the default as for even though later versions support five or six.

C

Okay, that's that's a bit confusing to me like do we actually define nappy version twice, yeah.

A

If you look at the issue that, unfortunately for naming version naming reasons it turns out to be yes, one one is ends up being used internally as the version that you're the version of node you're running with supports.

A

So, for example, we put a five out if we don't change it in the other one, it's the default that you get unless you override it.

A

C

A

Epi version two experimental, the experimental versions. Similarly, if say, the default was four. You could actually set it to five to get what's in five, but it means that you have to sort of explicitly choose to move up as opposed to you know,.

C

A

That's interesting is that lets you say well we're gonna, leave it at four, because that's gonna work across all the LTS versions, and if you want to use it really enough to actually opt into it by selecting five, but then.

C

Right, oh yeah, you've opted and/or.

A

You you know it's, it's not necessarily the preferred one.

C

Right well, but currently, currently we have only two levels right. We have. We have experimental and least right. So if you want to use something that is not a released version, you would define your unity version as experimental or you would define that be experimental, and then you get everything outside the release. Well,.

A

Today, today, because those two numbers are set, the same we've back ported for where and the default is four so.

C

A

Don't do anything you just get four, which is the default, or you said mr. to get the you know the.

C

A

We you know if we made a five and a six but left the default is 4. It would mean that by default you would get four and only be able to use these functions in four right.

B

A

Then you know you could opt in by defining your nappy version as five or six in.

C

Addition to being able to define that be experimental, yeah.

A

I'm, like you yeah, you could do experimental, which will get you everything yeah.

B

C

A

Other option would be five or six to get specific things.

C

A

So I'd be tempted to say like if, if we weren't going to be able to backport it, we might want to leave the default for.

C

A

C

Yeah I'm just I'm I, don't know it's just a vague sensation, but but I'm thinking, like you know, if we you know at what point are we actually causing people to have to make code changes so so as to be a bi-stable, because at that point it just defeats the purpose right like if people say if they, if they build their code with number defined, nappy version five then then yeah it should still be a bi-stable, because as no J's moves forward, eventually yeah.

A

They can go to five, but if they use any of the new functions in five.

C

Well and which they will cuz, otherwise, they wouldn't have said, go to five right right.

A

In which case they'd be able to go back to the previous, you know they're stable, but not for version. Eight. That's.

C

Right, yeah and and eventually eventually they're there they're there add-on without and he started, recompiling or anything we'll all of a sudden be supported on all LTS versions, because eight will just drop off the radar right.

A

C

Okay, yeah, that's okay, I! Guess! That's reasonable! I! Guess.

A

I mean like so the in our mind. Is it like it's probably worth doing a five and a six so that we can back port them separate? We can do some back porting if we want yeah.

C

I'm not sure if we you.

A

Rush to back port those things but having the possibility, probably isn't a bad idea. Yeah.

C

I mean, as a rule as a rule, I think. If and when we release on UN API version wishes. We should strive very strongly to to to having it available on all. We.

A

Might get some pushback on eight but anyway, I think if we do that, if we do it in master, so that we have the option, that's the best way to start, and then we can well.

C

Well, one thing's for sure right like if we decide to to handle begin separately because of this back party is true. Then then it's gonna take a while for us to back port everything else anyway right, so so so by the time we have woven version five through all the LTS versions. You.

A

C

You know we may get some feedback on that, and then we can we can. We can handle that in in in at the time at which we choose to release begin. So I think it might be worth just getting the process started for for date and threats, a function right.

A

C

A

Do five and then once that's unless there's some begin rush that makes sense to me exactly.

C

Yes, yes, right: okay,.

A

C

Cool, so okay, so then yeah I guess I can get the back. Porting sort of steamroll are going okay.

A

So that's yes, I think: we've discussed what Kevin wanted us to cover other than there is a pull request for forty two, which he probably would also like. Some yeah.

C

I'm actually I'm actually in the process of reviewing that one: okay, I'm, the the thread safe function, I think that's.

A

Quite a bit one and I think you're, probably the best review that so yeah.

C

Iii think it's it's! It's pretty good, like there's, there's not much left to do on there. So I'm, just gonna give it a final once-over but I think it's gonna be good to go. Yeah.

A

Maybe if you can, let me know once you're happy with it. I'll take a look but I. Okay.

C

A

Probably gonna have the most feedback, so, okay.

C

Yeah I mean I, it's it's gone through a few rounds of review, and, and now it's so far at least it's mostly just like you know like. Maybe we should rename this function kind of thing right.

A

C

But there might be something towards the bottom: there I haven't.

D

Gotten to the bottom.

C

Of yet so that's where that is okay,.

A

Okay, so I think we've covered what Kevin asked me to update on. Let's look at the things they're tagged, so we I think we already talked about what should already got what should go into n API version, 5 yep.

C

A

Submissions no Plus, Jess, interactive, so Jim I know you posted a the birds of the feather abstract right.

B

So when I, the way the wave, node plus Jas interactive works the way their submissions work is you you give a primary speaker and a secondary speaker, the primary speaker? If the session is accepted, they they get a full conference. Pass a secondary speaker gets a partial reimbursement for that. So what I was hoping we could discuss as a group is who I should put as primary and who I should put his secondary for.

A

The birds of a feather or I guess.

B

It's a feather and for the workshop okay.

A

So the workshops the same thing, so you get one and a half. What was it, what was the second person gets uh the.

B

First person gets a full ride and then the second secretary person gets a partial okay, so.

A

We should make those different.

B

We could and I.

A

Hope to be covered some other way. So I was me right.

B

Right so I'm thinking either Nicola or Gabriel ought to be the first speaker and if.

C

Ya, probably a bad person to be the first speaker, because I am very unlikely to go I'm still hoping, but but it depends on a myriad logistical things that have to happen here. So let's not waste the potential free ride for somebody. So I don't know. Maybe maybe you two should be you and Nicola. That is, should be primary and secondary.

C

You know, and vice-versa and the other one so.

B

The fact that you were accepted as a speaker would would not be influential in your ability to come.

C

Yeah yeah I mean the the logistics are our personal here. So.

B

C

Type logistics, so okay.

B

C

Hoping to get everything ironed out and then and then put in a travel request which, hopefully until will approve but I'm I'm, not a hundred percent sure, which is why I'd rather not I'd, rather not take up like a valuable. You know: freeride type spot, okay,.

B

A

I think if you two or the niccola.

B

Would you prefer to be the primary speaker for the workshop or for the birds of a feather.

C

D

It's your choice: okay! It's it's! uh Okay! For me, it's the same. Okay,.

B

Okay, oh I'll figure it out and I'll get both of those submissions in today's, so that we're not waiting to the very last minute excellent.

A

B

I'm, assuming they'll give us one or the other. That is what I'm assuming yeah.

A

I, don't know I mean a birds of a feather is a pretty light thing right. So right.

B

A

What kind of submissions they get and all that? Okay, all.

B

Right; okay, that.

A

C

Yeah I have one small thing regarding workshops: I'm sort of kind of trying to test the waters here in Santa Clara to see if anybody is interested in in doing like, like a native add-on workshop, so I'm trying to see how how I can do things.

C

I didn't tell like to reserve rooms and and maybe float the idea on Twitter see if anybody wants to come down for a workshop, so I'm very, very, very wet behind the years that organizing such things and so just wanted to- let you know and I'll, probably at some point when I have all the Intel logistics lined up I'll, just tweet about it and say you know if anybody's interested and I'll have a time in a place by then hopefully and- and you know, maybe try to get like- maybe 10 10 15 people.

C

You know that that kind of thing, so, just just so you know, and I'll probably be using your material Jim. If you don't mind, I would probably come down Gabriel. Oh, you would yeah.

B

C

B

C

I'll try to give you as much as much heads-up as possible. Okay, okay,.

A

It would definitely be good to have a few workshops like that I've been trying to think of where else we can do them. So that's good.

C

Yeah yeah, yeah I, don't know I, don't know what I don't know if it'll fly, but if it doesn't so, will you Jim right right, yes or Drive.

B

A

Okay, so I think that's this one we've covered off. This works right.

B

All right, yep and.

A

I guess yeah, so, just as a reminder, I know that there were a couple of other ones in terms of people accuse another submission and I. Think Gabriel. You had one too, but so just a reminder. Today's the the end of Frye I think Friday is the last day to submit this week right.

A

Okay, um so there's do we need some additional async worker variants. I think we talked about that I'm, not sure, there's anything new to talk about that.

C

Yeah yeah I think yeah I think it's. Basically we concluded that it's it's it's a good idea and yeah some of them. Some of them might actually be based on thread-safe function rather than rather than async worker, because I'm not sure how we can do like acing progress worker, for example, right.

A

C

I, don't know if we have in napi I, don't know if we have support for progress. So if.

A

We enter, and it would be good to do it with existing API, as opposed to adding API is right, because the.

C

More absolutely yeah yeah.

A

We be adding them is to support like migration, as opposed to anything else. I think yeah.

C

Yeah so now that I think about it, some more so if we did not be threat see if we implemented I'm using nappy threats, a function would actually be getting into like it's like thread implementation territory, because because.

A

C

A function doesn't doesn't yet start any threads on its own right, so yeah we will have to we'll have to look at it in more detail, like I'd also have to look at the way. Nan. Does the progress sync.

A

C

Because if it does have support from node itself for it, then we.

A

C

Want to consider, we may want to consider adding more async work or api's in the in the in the C layer agreed.

A

C

Expose that functionality, the.

A

Other thing that came to mind is like: if we're building something with thread safe function, we may actually want to make this like a module like an an migration module or something I. Don't know yes,.

C

Yes, yeah yeah yeah yeah yeah, because it would have like actual code in it like yeah.

A

Exactly so, I think I think that would be worth thinking about to when we get that get along the way does this you know. Yes, it makes sense to help people migrate, but maybe not actually part of the core module yeah.

C

Yeah yeah cuz that that's that's a that's a fairly big commitment. But if.

A

We tended up putting adding stuff to Cora API, then it would make more sense to potentially be.

C

A

Yes, okay, so nothing more to talk. We just need two cycles to work on it focus on async worker, so this is actually Kevin was working on that. So we talked a bit about I, think the stuff that was going on there and in terms of where do we have.

A

Actually, this is the one specifically I guess, but yeah so Kevin's been working, pushing forward on some of those and that's for one where we were adding some new async worker signatures. Anything else to talk about on that one just.

C

That the one with the destroy is pretty much ready to go. I'm just waiting for another review or or for the seven days to go up and then I'll start seeing it and which.

A

Number is that that's oh.

C

Boy, let me yeah.

A

There's only 11 of them I should be able to find it. Yeah.

C

It's 488, okay,.

A

I was wondering if I'd reviewed, er, yeah yeah.

C

Yeah, so so, for this, the only thing as I said in the comments is that I don't think we can do the ref counting, because because then we have to expose the ability to on ref in to like.

A

C

An API which which in which, as far as I know, is not something that nan has so we already have a. We already have an additional functionality beside nan, which is, which is to suppress this drug right.

C

A

Really straightforward so yeah.

C

Yeah yeah it's, but you do realize that if we were to, if we were to assert in the destructor that if this were implemented using a reference count instead of the suppress destruct flag, then then in the in the in the destructor, we would sometimes run into a situation where the reference count were greater than zero, or in this case the suppress destruct flag were set to true, because if you, if you say, suppress destruct you're, still gonna end up calling the destructor it's just it's just it's going to be flagged at that point very.

A

C

So there there's no point I, don't think, there's any point in asserting that it's zero, unless we also provide like unsurprised destruct, which is a special case of unrest, basic right, yeah, yeah.

A

I, don't think it's your right, because you can call it destroy whenever right exactly.

C

That's right, yeah yeah, so so the only reason we need this is is so that so that the mechanism of destruction is available for overriding. That's.

B

C

And so it discovers that, and and this should bring us up to par with, with with that, okay.

A

Sounds good okay,.

A

A

Think that so an SQLite port, you.

B

Know so, nickel and Gabriel, you may recall earlier helping me diagnose the problem with running L LD, be under vs code. My my breakpoints we're not triggering and.

C

B

Michael and I met last week and the takeaways from that are. We were able to determine the issues in the L LD be plugin for vs code, and it has an adapter model and when I changed the adapter from classic to native it works perfectly. So that was one takeaway. The other takeaway is we were Michael was able to confirm with the MOCA maintainer that it fires off only one process, even though it has a bunch of stuff going on. Unless you take extraordinary measures, if omocha runs it in a single process,.

C

B

That's what we learned last week and so now I'm I'm resuming work on that to get get that get that PRM. Oh okay,.

C

Okay, great great yeah, okay, I, understand why you would want to know about the one process thing yeah.

A

C

That, of course, your breakpoints are occurring right. Yes, in a different process, oh yeah.

B

Yeah, it's working beautifully now excellent, so.

A

I think that's pretty much the end of the list and we're out of time. So unless there's something that somebody wants to bring up I'd say we should call that until next week, all.

B

Right sounds good, okay, talk.

A

To everybody later.