Home
Contribute Contact Us Browse all meetings
Home Contribute Contact Us Browse all meetings
Agoric / SES / 10 Apr 2023
Agoric / SES / 10 Apr 2023
Previous Meeting Next Meeting
⏯
Add meeting Rate page Subscribe
youtube image
►
From YouTube: SES Meeting: Promise.ownProperties and ModuleInstnace.link

Description

Discussion of stage 0 proposal for Promise.ownProperties (or another name) to make Promise.all functionality more ergonomic. Suggestion for adding ModuleInstance.link to Layer 0, and subsequent discussion of the difference between reflection (provided in Layer 0) and virtualization (not provided in Layer 0).

A

Mathieu hofman: welcome to the test meeting.. Today we have ashley.

A

Mathieu hofman:, with releasing promise own properties, and then, after that,, we have a guy backward with a module instance. Functions. actually,. You.

B

Ashley claymore: thanks., so ashley claymore: share my screen.

B

Ashley claymore:, so this is a.

B

Ashley claymore: state 0 proposal. and that I've taken over from alex.. So he kindly let.

B

Ashley claymore: me move this across to my.

B

Ashley claymore: might get help. so there's.

B

Ashley claymore:, it's currently. get up a couple, more proposal, weight dictionary and I will to the slides.. It's been more interesting.

B

Ashley claymore:, so the the problem.

B

Ashley claymore: is this.. We go for a little journey of a code base, it could be starts.

B

Ashley claymore: very innocently., there's some sort of async things happening,, so it uses a wait to go grab it.

B

Ashley claymore:, then time passes.. A second thing comes along. and.

B

Ashley claymore:, we await that. ashley claymore: and we've now created a little multiple, where these 2 things.

B

Ashley claymore: could have happened in parallel., but ashley claymore: they,. You know it's just easier to write it. This way. someone comments on the pr. and says, oh,. You've got a waterfall.. Let's make this more efficient.

B

Ashley claymore:, so it switches to promised all everyone's happy.

B

Ashley claymore: we're now doing the 2 things in parallel, and just doing one to 8 for the 2 of them.

B

Ashley claymore: all good. everyone's happy.

B

Ashley claymore: then, you know. time goes by.. Another thing comes along because it's already using promised at all.. So.

B

Ashley claymore: people just had another thing to the promised at all. and then another thing comes along. and then another thing comes along. and then sometimes, and I've seen this happen, for real. Is that.

B

Ashley claymore:, I promised it all- starts to ashley claymore: get a little bit hard to read.

B

Ashley claymore:, as people keep adding things to it, similar to.

B

Ashley claymore: any kind of function that starts off with one parameter, I think, has 2, and ends up with 6 parameters. and by the time you get to there you're kind of regretting the.

B

Ashley claymore: your language, just as old and all ashley, claymore: parameters, and ashley claymore: you're now worried about things getting mixed up. and when you are reading this code as a human.

B

Ashley claymore: I'm. ashley claymore:, getting less and less confident that might be structuring here.

B

Ashley claymore: lines up with this. or even if it does line up.. I want to just like control. click. feature flags in my id and jump to where that's coming from., but I can't.

B

Ashley claymore:, so I have to manually kind of ashley claymore: count, and I'm now ready. not doing much.

B

Ashley claymore: interesting work.. I might counting the number of symbols in a code to try and follow it.

B

Ashley claymore: so.

B

Ashley claymore: have a library.. You have kind of.

B

Ashley claymore: also felt that maybe there's better ways of doing this.

B

Ashley claymore: so that blue bird has., this stop props.

B

Ashley claymore: on. it ashley claymore:, looks similar to this, where,. Instead of pausing and iterable,, you can pass an object. and.

B

Ashley claymore:, it will await each of the properties oil.

B

Ashley claymore: for each kind of ashley, claymore: property,. It will get the value ashley claymore: and await that. ashley claymore: and then give you back an object of the same shape.

B

Ashley claymore:, so this way,, even if you mix up the order of these and mix up the order of those you know the.

B

Ashley claymore:, it's done by name., not ashley, claymore: so.. I still can't collect, control click, jump.

B

ashley claymore:, but at least I can just read the name and read the name.. I don't have to count things or worry about things getting.

B

Ashley claymore: mixed up., so that's the kind of problem.

B

Ashley claymore: and like the rough solution on how that problem could be improved.

B

Ashley claymore:, it doesn't have to look exactly like this., you know.. Maybe it's from entries and object or entries.

B

Ashley claymore:, it's one of the things. That's.

B

Ashley claymore: difficult with this is.

B

Ashley claymore: I'm like a met, your ball. That has a very clear.

B

Ashley claymore: api, like behavior, like semantics. of how do you get all the things it's initiable?, you have to wait for it with objects., it's much less obvious how you get all the things.

B

Ashley claymore:, like the blue bird from my stop props., you know, there being.

B

Ashley claymore: a little bit hand away from what that means,, whereas maybe we need to be very explicit about what it means.. So people don't get confused. About.

B

Ashley claymore:: are we walking the prototype?, how we getting all the innumerable things things like that.

B

Ashley claymore:, when there is very good precedent in the language of.

B

Ashley claymore: kind of what we do, walk like. if you spread an object, or if you do object or entries.

B

Ashley claymore: yeah,, maybe something like using the existing entries kind of button.

B

Ashley claymore: would be a better way to go,, even though it's more of a both,, so that there's, I think, there's definitely things to explore.

B

Markm:, even though this maybe is.

C

Markm: yeah,, when you say that there's pressing in the language,, the problem is that that there's several dimensions on which properties, differs, and and there's precedent for a whole bunch of different choices.

C

Markm: so there's there's you know.

C

Markm: the narrowest precedent.

C

Markm: is things like object, that keys, which is.

C

Markm: only own,, not inherited.,.

C

Markm: only innumerable,, not not innumerable.,.

C

Markm: and only string named not symbol,, name.

C

Markm: and so.

C

Markm:, so I could imagine that you want to., you know another naming.

C

Markm: that you might want. here is a promise, dot., all keys.

C

Markm:, referring to the keys from object that keys, not necessarily saying that's what you should do,, but just the there, you know,, there's.

C

Markm:, I think the triple dot markm: is also just in is certainly just an own and.

C

Markm: and just innumerable., but includes simple name properties as well as streaming properties.

B

Mathieu hofman: is it? yeah.

A

Mathieu hofman: promise from it?. This is a little bit more explicit for that,, because you relied on on.

A

Mathieu hofman: on the object that entries to give you that mathieu hofman:. The question then comes from from entry is also a on the key.. If the key is a promise, or it should only be.

A

Mathieu hofman:, that's that's true.!.

B

Ashley claymore: yeah,. Certainly it opens up that dimension that now we're not,, but definitely not limited to just strings and symbols. Necessarily.

B

Ashley claymore: and presume it like this.. I imagine this is going to. no.. Maybe this isn't going to be a performance bottleneck in the code, because we're waiting, and we're you know we're doing I/o, anyway., but.

B

Ashley claymore:, I imagine this isn't going to be as friendly.

B

Ashley claymore: to optimize, for, compared to,, you know, with the the double indirection, as opposed to just immediately kind of the one layer.

C

Markm:, do not assume that you're doing I up markm: right? I mean this. could just it, could. it could, and often is still just local asynchrony,. If you're trying to.

B

Markm: yeah, I agree., but but nevertheless I I I I'm also.

C

Markm: don't find the the performance difference.

C

Markm: something to work., I don't., I don't think it's something to worry about., so the the from entries and and and entries that would so that would be.

C

Markm: all own,, whether innumerable or not,, and whether string names or symbol name.

B

Ashley claymore:, so the another thing is that there's also this alpha proposal the way to.. I I think this it's.

B

Ashley claymore: we've got to use the way to focus on this tc. 39.

B

Ashley claymore:, I I think that that proposal it pretty relates to this, you know,. If that goes ahead,, maybe we would do this., but I don't think that proposal.

B

Ashley claymore: replaces this, or vice versa.. I think you know they are separate in.

B

Ashley claymore: in the problems they're trying to solve.

B

ashley claymore:. So it's it's related.. I don't think they're actually., but I think they influence each other in too much regard.

B

Ashley claymore:, so the the of the question,. I think that would come up with this is what about all the other promising things?.

B

Ashley claymore:, I think the the use cases when I don't have concrete use cases for the other one.. So I you really have that use case of the promised at all. Problem.

B

Ashley claymore:, so I think this would be another thing of ashley claymore:, you know.. Is it more complete, ashley claymore: to have the fuller api rather than special casing all or does adding all these extra things actually add more burden.? So I think that's another thing I'd. If this proposal and forwards I'd want to kind of explore. then., my gut feeling is it's harder to justify.

B

Ashley claymore: going all in with creating lots of these multiple variants. alex.. I can see your hand.

D

Alex vincent: only because I believe this is the last slide, and I wanted to save my comments for the end.

B

Ashley claymore: you'll, mute, alex.

A

Alex vincent: okay., so I I have a couple of concerns and keep in mind. I'm., the guy who originally pitched this 2 years ago. To this group.-.

D

Alex vincent: the reason that jordan and I abandoned this 2 years ago,.

D

Alex vincent: was a lack of a clear use case for it.

D

Alex vincent:, when we originally pitched it, we saw that we we could achieve roughly the same results with 2 lines of javascript code. Using existing apis and 2 lines is not significantly better than one.

D

Alex vincent:- and I don't think you've sufficiently answered that concern here with this new use. Case., don't get me wrong., it's interesting.

D

Alex vincent:, but I don't think you're quite there yet.

D

Alex vincent:, thank you for pulling this up.

D

Alex vincent: so.

D

Alex vincent:, I'm looking at it from that perspective.. Is it significant enough of an improvement to add to the language, and I'm on the fence about that? I'm not against it., I'm not for it., I'm just on the fence, and that leans towards.

D

Alex vincent:, if there's, if it's not a compelling enough case.

D

alex vincent:, do we need to bring it in,, especially since this kind of thing can be very easily implemented by libraries.

D

Alex vincent:, that is my first concern.. The second concern is about,, possibly.

D

Alex vincent:, because we we could be talking up to 4 different.

D

Alex vincent: 4 different methods that we'd be adding for object analog to the arrays. versions of these is maybe we want to put it in the namespace under promise.. I suggest it's a promise, dictionary, dot, all, or.

D

Alex vincent: promised that dictionary that all settled, etc.,, but that's a separate issue., don't bring it up on the screen.. I just wanted to throw that out. There. I'm really worried about the use case.. Is it a strong enough use case?.

B

Ashley claymore: yeah., so yeah, with the ashley claymore:. The fact that you can achieve the same end result in a 2 liner,, I think, is.

B

Ashley claymore: that the the problem I'm stating is that the problem is on this line, as opposed to being able to achieve this.. So like the problem of these things, here, needing to be in the same order as these things here,, which is trivial here,, but becomes less trivial. When these things.

B

Ashley claymore:, like more complex, and when the number of this increases.

B

Markm:, so so specifically,, who makes make sure, I'm understanding specifically when those become more complex,, as in line expressions.

C

Markm:, if you just,, you know., if, if you just take your complex expressions, and you just have separate.

C

Markm:, you know const shape, p equals cons, color,, p, equals,, etc. then,. By the time you get to that line, all you've got are variable. Names. now you've doubled the number of variables you have shape and shape. p.

C

Markm: and if if the total number is long,, then you still have a then even with just names,, you have accounting problem.

C

Markm: but I, but I think that.

C

Markm: that,, you know that is, would still be the the preferred solution to comparison rather than in line expression.

A

Mathieu hofman:, so one.

A

Mathieu hofman: so., one thing I remember discussing is.

A

Mathieu hofman:, this could be a helper, and libraries act on this. So obviously this is the helper that a lot of.

A

Mathieu hofman: libraries and people, I've written., I sure I've written mine.

A

Mathieu hofman:, which mathieu hofman:, does give credence that this is maybe something which is considering.. If I into the language. however,, I remember discussions on what shape it should have, and it wasn't clear like that.

A

Mathieu hofman: being able to create objects also was the only.

A

Mathieu hofman: I. I remember a discussion about maps that I don't know.. Maybe it was an adjusting proposal, and and not this one.

A

Mathieu hofman:, I don't know., I I know also, there is variations of this,, which is more of a deeply fulfilled.. So if you have more of a tree.

A

Mathieu hofman: like net nested promises and be able to resolve those.

A

Mathieu hofman: we're like each., there's, probably more use cases around.

A

Mathieu hofman: around here that, like when you think,, if we want to solve them or not,, we're just sick with the very simple.

A

Mathieu hofman: just the properties of a record that.

B

Ashley claymore: one thing that I was thinking of the other day.

B

Ashley claymore: another way you could ashley claymore: in kind of what you're saying about how this could be more. The the entries way of this is more general right,, because then you could use it with constructing.

B

Ashley claymore: maps as well.

B

Ashley claymore:, so ashley claymore:, the curls promised to always, and the rox.. It is singular. You have to actually pass the first thing. I mean,. It means we do also have another extension point of the second argument, like here. We could say.

B

Ashley claymore: depth, ashley claymore: like 2, ashley claymore:, which actually means. ashley claymore:. I expect this to be an array of arrays similar to how you have,, like a ray flats kind of depth.

B

Ashley claymore: here. why?, so if this was actually.

B

Ashley claymore: yeah, no,, I like ashley claymore: key.. How are you?.

B

Ashley claymore: you'd get almost like a promise from entries in a.

B

Ashley claymore: in a different way. ashley, claymore: it's,, I think.. This then starts to become ashley, claymore: more complex and like ashley claymore:, I see people less fewer people, ashley claymore:, likely to say, oh,. My promised at all has too many things., I'm gonna.

B

Ashley claymore: rewrite it to use named version. that is.

B

Ashley claymore: there's another potential way of ashley claymore: going for more complex thing., but it's more,, general doesn't., add new, actual apis.

A

Mathieu hofman: no, and I sent like a.

A

Mathieu hofman: hey? it all entries some of the all entries that it's just last time. We discuss this, which basically lets you it it's not as they're going to make,, but it can be.

A

Mathieu hofman: and- and I remember, like when- writing this,, it was like,, not obvious what the right shape was.

A

Mathieu hofman:, so maybe yeah,, maybe it is.

A

Mathieu hofman: bye.

A

Mathieu hofman: right? is that, like there should be a built in helper.

D

Alex vincent:, as I said, I'm on the fence about this.

D

Alex vincent: the reason I not I lean against it is just.. It's easy to implement a it.. It's very easy to implement. a shame for this!.

D

Alex vincent:, that's that's my big concern. is. Is it.

D

Alex vincent: something strong enough where the engine has to do it for us?.

B

Ashley claymore: my, my,, my.

B

Ashley claymore:, my feeling on that is that it's easy to implement a shim for.

B

Ashley claymore: people that are very comfortable with javascript and promises which,. Admittedly, a lot of javascript developers, are.

B

Ashley claymore:, but it compared to some other ashley claymore: shim about api's.. I feel like this is one that.

B

Ashley claymore:, I wouldn't just immediately knock out.. I'd, probably have to stop, pause, and think so.

B

Ashley claymore:, I would certainly ashley claymore:. Imagine it's more. This type of thing you'd want to import from a library rather than quickly. write up yourself.

B

Ashley claymore:, especially as I do find.

B

Ashley claymore: coming to javascript that they're okay. we're facing quick and promises., but it's definitely most concepts.. It takes a little bit longer to really fill.

B

Ashley claymore: really really comfortable with working with them.. My visit.

B

Ashley claymore: I remember, like people.. It was this trick of getting.

B

Ashley claymore: oh,. That's my coming right! Now.! There was always this trick of like combining promised race.

B

I'm promised at all ashley claymore: flipping the resolve to get yeah, it's.. It was things like that. that.. It was clever,, but.

B

Ashley claymore:, it was never ashley, claymore:, always immediately office. what it was doing unless you felt super super comfortable with the promise, help is, and the way promises work.

C

Markm: flipping the resolve.

B

Ashley claymore: I'm trying to remember what it was.. I think it was maybe before we had.

B

Ashley claymore:, it was something where you do promise not race,, but then you switch all for every promise.

B

Ashley claymore:, you change the rejected markm: to a settled and all the settled to a rejected, to kind of get the reverse, trying to remember what it was like.

E

Daniel ehrenberg:, I share alex's skepticism,, but from a slightly different perspective., I think it's.. I think we should include things directly in the language when they have ergonomic value.

E

Daniel ehrenberg: and daniel ehrenberg:, this is just a normal thing to do in any kind of standard library., not everything has to like. give a new, huge capability.. Some things are just.

E

Daniel ehrenberg: a little bit nicer, more convenient.. Anyone can implement iterator dot range,, but I'm a big fan of that proposal.

E

Daniel ehrenberg:, but what convinces me more is like the code sample that that mark had, where you just to find the promises above the promise dot all as variables.

E

Daniel ehrenberg: and then you do the the promise that all kind of simply just with the variable names., it's very easy to verify that they're in order.. I think we have a serious.

E

Daniel ehrenberg: developer understanding problem where people don't.

E

Daniel ehrenberg: really get daniel ehrenberg: that.. You can launch the promises like this, that they'll happen in parallel., and maybe you think of promise that all is more like a fixed thing, where you have to put the promise creation right in there.

E

Daniel ehrenberg: and I'm not sure how to address that that issue.

E

Daniel ehrenberg:, but that's the only real downside I could. I could think of.

E

Daniel ehrenberg: and daniel ehrenberg:, in this discussion, and in in ashley's presentation, we've already covered. why,, there's there's not kind of an obvious single design. Here.

E

Daniel ehrenberg: so.

E

Daniel ehrenberg: yeah,, that's that's why I'm.

E

Daniel ehrenberg:, not extremely opposed., but I'm kind of leaning in the direction of.

E

Daniel ehrenberg:, not including this, this right, now.

E

Daniel ehrenberg:, you know,, if we,. If we want to think about how to improve the ergonomics of using.

E

Daniel ehrenberg: promises and daniel ehrenberg: all all over the place. there's also been that await all proposal.

E

Daniel ehrenberg:, I don't know.. I think we might want to do like a broader study about where people get confused. I'm not, really.. Could I'm really not convinced that this is like the main.

E

Daniel ehrenberg:, the main thing with promised ergonomics at this point.

E

Daniel ehrenberg: may may maybe it is, and I'm missing something. but yeah.

C

Markm: yeah,, I I I. markm: for with regard to the criteria for stage one,, I would.. I would certainly.

C

Markm: agree to let this go to stage one,, but I'm I'm.. Similarly, skeptical.

C

Markm: is false.. I I.

C

Markm:, I hope you've read my markm:, my essay,, the tragedy of the common list., how large languages explode.

C

Markm: this, to me is very much in the category of what I you know of the.

C

Markm: the improvement which is,, wouldn't it be nice if we could write it this way, instead.

C

Markm: and a lot of you know,, and sometimes those are worth it.

C

Markm:, I'm actually supporting the the the pipeline, I think,, because it's it's such a nice improvement in the how you write it.

C

Markm: but you know,, supporting that one good., it's very, very high leverage.. It has broad reach. it's one mechanism that once you learn it, you find yourself using it for a lot of different things.

C

Markm: this one is well,, it's not more syntax., that's good. it's., it's just more api surface,, which is much less costly.

C

Markm:, but it's still just wouldn't it be nice if we could write it this way, instead of that way, where there's not that much improvement.

C

Markm: and the way you have to write it now is not that painful compared to you know,, compared to what you're offering.

C

Markm: and you know other things like., you know the.

C

Markm:, the prior,, the markm: things that that really markm: change, what you can do with the language.

C

Markm:, which reasonable. you know what? what can be achieved with reasonable effort, or, to me much just much higher priority than something like this, and the the incremental learning burden.

C

Markm: of markm:, of adding new methods to the api surface.

C

Markm: that are then become part of every manual part of it.. What you need to learn you need to not just remember.

C

Markm: the benefit when you're using it, of using it., you have to also remember the cost when you're learning the api as a whole of what's this for?. What's that for? and just more things to think about.

C

Markm: and it's those those incremental learning costs., so it's expanding the api surface,, whether it's the things that just very easy not to take into account.

E

Daniel ehrenberg:, I want to register my disagreement with that analysis.. I think we should be making ergonomic improvements. It's just a question of.

E

Daniel ehrenberg: with there is cost., but I don't wanna. I I.

E

Daniel ehrenberg: disagree that we should focus.. I mean,, I think you you you give the example that pipeline is is important and good,, even though it's an economic improvement.

E

Daniel ehrenberg:, I think, we'll probably find other examples in the javascript standard library that might not have extremely broad reach,, but might still be, are going to make improvements that are worth it..

E

Daniel ehrenberg:, I think we just need to analyze these things case case by case.

E

Markm: and it's the daniel ehrenberg:, the more decisive thing is about like.. How much of an environment is it?. Do we have these alternatives that that make sense,? And in this case we we maybe do.

C

Markm: so., I I. so I agree with with all of that.

C

Markm:, but I I think the thing that that that I need to that, I feel like I always need to remind people, is that it's, the overall learning cost of still just having more stuff that we also need to take into account in this trade off,. And that's the thing that is, that is, you know.

C

Markm:, I I think we we suffer the most from it from systematically ignoring.

E

Daniel ehrenberg: don't. think we systematically ignore this in tc. 39. I think we constantly talk about it, and that's good,, because the cost does exist.

E

Daniel ehrenberg:- but I think sometimes this is over emphasized.

B

Ashley claymore:, I think I I ashley claymore:. I agree that ashley claymore:. I think if this did go through to stage one. a a very large emphasis of this would be working out how to research and measure.

B

Ashley claymore:, if this is actually going to help. and how trying to also measure.

B

Ashley claymore: the cost of learning this. in the first place,. You know. add that ashley claymore: now,, but I think, like the technical semantics, and the naming, you know, is like that step, I think. yeah,, the largest part, is actually trying to work out.

B

Ashley claymore:, how much of a problem this is, and how well this solves that, and then get that to their costs. cognitive costs, rather than the technical complexity costs.

B

Ashley claymore: timebox wise.. I'm aware I've used an awful lot of this. This session, now.

E

Daniel ehrenberg: yeah, and we we should go on to the next agenda item., but thank you.. Unless you wanted to respond to my comment,, mark, or something., I don't want to to cut you off from that.

C

Markm: no, just just.. I I think everything that just mentioned is a great thing to investigate in stage one.. So I certainly do support this for a stage. One investigation.

A

Mathieu hofman: all right., so on to the next topic. thanks.. Everyone got it.

F

Guy bedford: thanks. yeah.. I just wanted to pick up a discussion on on the modules virtualization, and I brought it up yesterday in the in the regular mod modules,, discussions, and kari had some.

F

Guy bedford: some good feedback on it, already,, so it'll be. I'll, try and do my best to kind of.

F

Guy bedford: recap where we got to yesterday on that, as well.

F

Guy bedford:, but the the motivating use case for this is.

F

Guy bedford: well,, the the the initial reason was.

F

Guy bedford:, that. and then, according to.

F

Guy bedford: think about,. You know what was left on my wish list here and.

F

Guy bedford: in in the last country meeting shoe had some reservations about the import hook.

F

Guy bedford: and and the the the fact that it.

F

Guy bedford:, during that that they've gone through a lot of effort in in v. 8.

F

Guy bedford: to ensure that you have a synchronous, optimized resol, there.

F

Guy bedford: and they don't want to be running js in the resolver,, and this is something that they've been quite careful about to the point where we're even now,. In a situation where there is divergence between their js and browsers,, at least temporarily.

F

Guy bedford:, because import dot meadow resolve is sync in the browser and async in their js, and and that's been a the phone from there just for a little while, and then still is.

F

Guy bedford: in trying to rectify that., but this this sort of strong stunts on on that that resolve it.

F

Guy bedford: I'm bringing up that argument and then putting some port to that, and then also relating it to other stuff. and.

F

Guy bedford: I. guy bedford:, when when I first presented some of the ideas around modules for import, reflection.

F

Guy bedford:, it was based on a kind of a more of a static linkage, model, and then the import hook. Discussions were kind of around dynamic import,. I think, originally as a sort of the general hook.

F

Guy bedford: and I I was wondering if there's, if there's not still scope to actually consider both of these things, coexisting.

F

And so the idea was,, if you have a.

F

Guy bedford: on on the module instance itself, guy bedford: introduce a new instance method. I'll just share a very rough outline. here.

F

Guy bedford: introduce a new link function on the module instance.

F

Guy bedford:, which takes 2 arguments. the first argument: is, the string specifies to static a link., and the second argument is the module instance to statically link that specifier. again.

F

Guy bedford:, since we have access to modern instances before they've been evaluated.. This opens up a the ability to to link these things before they've been evaluated.. So that's a linkage that is.

F

Guy bedford: perfectly compatible with cycles., and when you then pass that module object to your dynamic imports it', it can already know that static linkage.

F

Guy bedford:, and if it already knows that static linkage., there is no need to call the import hook and go back into js., and so that it,, you know that that might litigate the it's concern. There.

F

Guy bedford:, at the same time, the import hook would still exist.

F

Guy bedford:, so that, if you didn't call link.

F

guy bedford:, the import hook would be used to fulfill guy bedford:, all the everything asynchronously,. And if there is no import hook, it would obviously default to the the default resolver.

F

Guy bedford: and what that gets us is potentially a mitigation for v. eight's concern.

F

Guy bedford: also, not solving virtualization, but giving us an easy first win on virtualization, or at least an easy way to handle some.

F

Guy bedford:, very simple, interceptions.

F

Guy bedford:- and you can also imagine this potentially even extending to workers. so it it might also be another way to to skin the work.. This problem in that.

F

Guy bedford: import hook relates to the kind of dynamic host resolution. We were., you going. you know, out to this parent context., whereas the static linkage is something that could potentially, even.

F

Guy bedford:, you know, existing work, a transfer if you.. If there was some mechanism implemented by which.

F

Guy bedford:, you know mod modules, could be uniquely identified,, regardless of their.

F

Guy bedford: a realm or host guy bedford: environment context., so.

F

Guy bedford:, with something like that, you you could imagine that you could, you know, link something.

F

Guy bedford:, posit or worker, and you would get sort of their equivalents.

F

Guy bedford: versions on on the other side, somehow. there's some most of these to that- would be put to that.. But I was just wondering if that's maybe a way to think about that as well, and then it's only the input. that gets stripped during transfer.

F

Guy bedford: and then there there are a bunch of questions that are still open, like.

F

Guy bedford:, should you be able to link things that haven't that that only are dynamically imported, and on actually static monitoring, ports. and then also.. If you want to have some shorthand like.

F

Guy bedford:, like an object, shorthand, or even a shorthand for linking against like.

F

Guy bedford: namespaces and things like that., but just.

F

Guy bedford: yeah,, but putting some thought to this as something that I just wanted to get feedback from the folks here, and.

F

Guy bedford:, I I guy bedford:, we'll let you pick up with with his his feedback from,, I guess.

G

Caridy: yeah., so I I I see no problem with the api. just a lot more difficult., it's not only about links,. You know how to do reflection, you know,, to give them. what are the things that they're importing.? What are the things that they're exporting on?.

G

Caridy: those are complicated.. You have different api's, if any, of what kind of people the statement that you have, and then there is a problem of what's happening by the in for a dynamic module that I already linked statically by calling link.

G

Caridy: is it going to be escaping the call to the dynamic in in port book and so on.? So the a lot of.

G

Caridy: little details there. technically. yeah,, it will work., but my main focus is not really about the api.. My main focus is to try to understand really what the problem is.

G

Caridy: with respect to implementers and.

G

Caridy: the specific performance issues that they have there, and so on.

G

Caridy:, because the way and this we we talked about that yesterday.

G

Caridy: but my point of view caridy:, is that that concern that they have.

G

Caridy: about virtualization is not really.

G

Caridy: it's.: it's it's not really going to be relevant for the case of layer, 0.

G

Caridy: and the reason why,, I don't think is relevant is because you are in use of that. you you, in in order for you to do.

G

Caridy: module. the link that means that you're linking to certain things., you you're, you have a modeling that you're linking to the dependencies of the module infants., so you're doing all that in use of land. The same way that you will be doing all of that in.

G

Caridy: in the input hooks case. you're, not in.

G

Caridy: the browser intertwine resolution process between things that are created by the user, using a new module creation process versus things that are created by.. They use the agent.

G

Caridy: that are also part of the same module tree.. I don't see that happening,, because in the or in our proposal, we we were very explicit about. once you enter into the wrong of.

G

Caridy: module in terms of creating a user land.. You are in that land until you decide to go and.

G

Caridy: delegate back to the user agent, a a sub 3 of the module that you are trying to construct, and there is no possibility to have.

G

Caridy:, so a single dependencies there.

G

caridy:, because for you to have a signal dependency., you must do it all in use a lot.

G

Caridy: and and that restriction itself., what it tells me is that, when the browser hits these modules that are creating user land.

G

Caridy: that concern that they have is no longer a concern.

G

Caridy: because they use a lot. all these things that happen in the user line. At that point,.

G

Caridy: there's not.: they would not a slow down the things that are being load in.

G

Caridy: in. no, user. Not so I I I'll. I'll push back on.

G

Caridy: the actual concern here and try to understand better what the concern is.

G

Caridy: because I don't think caridy:, it's real for this particular case.

F

Guy bedford: yeah,, I I I don't know how to get for the feedback on that.

F

Guy bedford:, I forget which. guy bedford:, in which spec it was brought up in reference to at the last meeting. We should probably check the notes.

F

Caridy: so I, from what I understand from you.

G

Caridy:, the issue was mostly that they thought, go and construct them on the graph today,, and then they have.

G

Caridy: part of the process is to go and do some.

G

Caridy: some of the steps in a synchronous, because they already have all the stopping them.

G

Caridy: because she goes into sip through it,, and they do it in one goal, and they don't want to go to use a line.

G

Caridy: bye.

G

caridy: that will not change. caridy: because from something that is being created by the user agent and linked by the user. Agent.

G

Caridy:, you will never be able to depend on a thing that was created in user, that is impossible.

G

Caridy: is the other way, around. caridy:, because you don't have a way to to to say use the agent. go on the all this stuff, and then on this particular dependency. I want to take over,, and I want to give you the intense for that that doesn't work.

G

Caridy:, because this api doesn't give you that it. this, I give you the ability to create a room module.

G

Caridy: and then link to water things that are either created by a user land or created by the user agent., but it's never the other way. Around., it's not possible in the api that we are offering.

G

Caridy: so caridy: for the cases where you go into user land,, you're a user not to do the things in a synchron or whatever., no problem.. It will not affect or slow down what they already have.. It's just a new process of it's.. I guess.

F

Guy bedford:, that's kind of from that concern point of view. you're, saying you're gonna have 2 code pods in your module loader, one for custom loaders.

F

Guy bedford: one for native loading.

F

guy bedford: and when you're on the custom, loader path.. It's not such an issue.

F

Guy bedford: and that's like an interesting question to consider with this virtualization. because, like are we.

F

Guy bedford:, are we virtualizing it to the point where you can.

F

You know we would. guy bedford:. Are we leaving native linking as like something? That's? That's.

F

Guy bedford:, special, different. and then you've got this. This new kind of virtualized linking,, which is, as you say,, it's our own user line, thing.

F

Guy bedford: or we, you know, completely unifying those pods., that's a good question, and it will be interesting to hear feedback on that. but from you, know.

F

Guy bedford:, if, if that separation makes sense, or if they still would carry their argument through to the.

F

Guy bedford: the user line loading as well, and say what we wouldn't even want that to to be jumping around.

G

Caridy: my, my, my,, my main concern., so one thing that maybe tactically we can do is find time.. We talk about the us. as well, the phenomenon between visualization and reflection.

G

Caridy:, this is a reflection. api.. You you're able to create module, instances. you're able to link those module instances you're able to.

G

Caridy: require or import those module instances and get the volume that they export or whatever.. So it's not really about utilization in, in the sense that you're not really virtualizing.

G

Caridy: and anything order d10, the import here I will be the one resolving it.. But beyond that there's number to it. you don't control anything., it's just a resolution,. I would say, but not realization.. So if we disconnect the layer 0 from the concept of virtualization,, then we should be fine.

G

Caridy: it's most of the concerns that google has about.

G

Caridy: visualization. as a use case., so that's probably why they push back here.

G

Caridy: so tactically., we should probably disconnect that those 2 concepts.. Then this is not really a mutualization.. This is,, I don't know one name. We can put,, but I I would say my module, intensation,, resolution, or or just simply saying, there's a reflection from module.

G

Caridy:, this is a reflection. api.. We can do what the browsers are. Doing.

G

Caridy: and we can do it in user line. now, now that we're virtualizing the module graph entirely. you're saying,. I want to create an intense,, and I want to make sure that this intense link to our instances that I have a little bit of control, on.

G

Caridy:, so I I'll say that caridy: this is important to clarify, saying,. This is not about virtualization.

G

Caridy:, this is about controlling subtree of the monte cristo and being able to create any user line with the reflection api that we offer.

F

Guy bedford: yeah, it.. It would be great to be able to narrow down that feedback, because,, you know,. If, if there are any concerns of progressing with import hook,, it would be good to know. and and.

F

Guy bedford:, how about have a path forward,- and this is one.

F

Guy bedford:, the the the other things that this kind of does.

F

Guy bedford: bring is guy bedford:, potentially some like slight ergonomic improvements in the sense that, like.

F

Guy bedford:, I guess it it.. There is also like this kind of like.. Maybe it's like you for the software called different., so question,, which is like.. You know we coming at this kind of like whole load of angle, where it's like.

F

Guy bedford:, you have to stop from the route.

F

guy bedford: and you're, implementing the whole graph, and you have this, the single hook, and you need to know everything about everything.

F

Guy bedford: and you need to be. you know, this, this kind of power, virtualizing, user, or do we want to like just yeah.

F

Guy bedford: come, come at it from a direction which we can say, hey,. If you want to.

F

Guy bedford: resolve this to this, you can just inject it here.. Otherwise everything else is going to behave. The same. so and and.

F

Guy bedford: there are. guy bedford:, you know, a bunch of lots of cool questions about that, because, like,. I think you mentioned the other one yesterday as well as like.. Are we then losing control of virtualization?? If now you have code that does this? This.

F

Guy bedford: kind of linking,, and you want to then virtualize that code. you can't., can her long intersect, it.

F

Guy bedford:, the the things that are, you know, programmatically in this way.

F

Guy bedford: and so.

F

Guy bedford: yeah, some interesting questions. and then,, as I say,. The third thing was like.

F

Guy bedford: actually.

F

guy bedford: also,, possibly thinking about the the work, a linking again, and that's a wider conversation that that you know.. We probably need to still.

F

Guy bedford: continue discussing separately, and.

F

You know. guy bedford: between all the specs.

F

Guy bedford: but yeah,, and there there was one other guy bedford: one other minor proposal here.. If you scroll down to the bottom of this page.

F

Guy bedford:- and this was also in a similar vein.

F

Guy bedford: exposing important that to and and bringing up the discussion again about about whether that should be. you know, reflectable on on the instance.

F

Guy bedford:, as a way to again at a per module, level, being able to either inspect it. or also being able to mutate it and provide instrumentation, or or whatever it might be. Required.

F

Guy bedford: and thinking of this in the context, that for the module, expressions and module declarations, proposals., we will need an instance, object., a module instance.

F

Guy bedford: spec. in some phone.

F

guy bedford: and for the import reflection, proposal, we.. We are happy to align with these proposals, and.

F

Guy bedford: look at reflecting a module instance as well. and so,. If we are reflecting this module, instance, thinking along the lines of,, should we.

F

Guy bedford: try and make this module instance more incrementally powerful. Where you know small features can.

F

Guy bedford: to to enable these virtualization used cases.

F

Guy bedford:, as we work up to getting the full module source stuff together. so also like.

F

Guy bedford: thinking about like the kind of progressive virtualizations as well, kind of angle, and and and also just to.

G

Caridy:, my my only p. back there is that utilization.

G

Caridy: these api is not for utilization.

G

Caridy:, the visualization.: what is about the default behavior?.

G

Caridy: they, the host and you replace the so it's not going to happen at this level,, so you can have.

G

Caridy: the same program that was written this way with the doll link, or with the import, or whatever is the thing that you use.

G

Guy bedford:, you should be able to get that program and run an individual environment.

G

Caridy:, it's not like you.

F

Guy bedford: yeah., the import matter, object that is created is the one that's provided by the host or the host took., and so in this model, it's just exposing it on the instance, as the object which it already is, a a normal javascript object.. So.

F

Guy bedford:, this isn't virtualization, and there has to be a virtualization layer where you can.

F

Guy bedford:, properly guy bedford: have that top down view of everything.

F

Guy bedford:, I guess I'm trying to,, because with all of the specs we have to meet in the middle somewhere.

F

Guy bedford:, just thinking of ways in which we can.

F

Guy bedford: try and break things down, or or maybe.

F

Guy bedford: see, see about how things can can link together.

F

Guy bedford:, I'm not into it.

G

Caridy: yeah, for me. it's., it's the important bit., the a high out a bit is that we disconnect this api from retail.. It's not about utilization.

G

Caridy: and if that's not clear with them,, we can., we can chat about it.. But my my opinion. That's the most important bit like this is not about.

G

Caridy:, the all reflection of modules and then visualization.. We can have those discussions once we get there,, which is that they host level container controlling the host level books.

F

Guy bedford: yeah.

G

Caridy:, I think nicole, did the work of getting the meta.

G

Caridy: to use the proper hook, on, so on., so eventually we are running in an environment that is visualized.. If we have the only to virtualize the hopes it will., it will work fine. it will.. You will be able to still create modules, linkedin still getting a meta, that's coming from the from the browser, and so on.

G

Guy bedford: yeah.

F

Guy bedford: yeah., so this this is more sort of ergonomics trying to broaden.

F

Guy bedford: the usability to guy bedford: javascript developers, who want to just quickly.

F

Guy bedford: create a you know,, a virtual module with with a custom, mark, or something like that, and and have.

F

Guy bedford: the ability to access some kind of I. I don't know what we're gonna call this, if it to to distinguish it from virtualization.. But perhaps just some reflection is overloaded., but reflection will be the.

F

Guy bedford: right sort of term.

F

some some basic higher higher level.

F

caridy: yeah.. So on that note. and I know we're on time.. But on that note these api is very similar to the in ibi from 10 years, ago, from the herman.

G

Caridy:, where there was a reflection. api,, where you you need access to.

G

Caridy: once you parse the thing you need to access to all the details of the parser,, so you can then act upon the intents and do some linkage, and do some some of the things,. Then obviously.

G

Caridy: the other kind of problems, like, okay? well,. What happened if I try to import a module,, but the linkage is not complete,, so it fails., so how the user knows that it needs to do certain things prior to the the resolution,. So it's.

G

Caridy: there. a bunch of this.. This api is obviously a lot more complicated than what we have.

G

Caridy: a lot more api's.. You need a lot more care to do the right thing when it comes to linkage.

G

Caridy: and- and then on top of that you have to have your api anyway for the dynamic one.. So it seems like there.

F

Guy bedford: yeah., so I guess it's kind of reflecting what you know. no js or something does today,, which is, you have this global import, dynamically., look. and then you've also got this sort of static linkage which gets applied on the actual object.. So you know it., it's quite similar to what we are doing in in in the you know, in the engine.

F

Guy bedford: and guy bedford:- I I guess there are some most of questions there like.

F

Guy bedford:, the way I I would imagine it right now is that if you don't link anything., you would get the current behavior today., which is it?. It just uses the import hook for everything.

G

Caridy:, so what one important out on these is that you you, you call to implement.

G

Caridy: this api., you use a lot, the toll link.. You can do it in use a lot.

G

Caridy: there's no problem with doing this., because what it does is creating a that,, then the hope will use the, but not the other way around.

F

Guy bedford: right, so you you, you could.! You could implement this on top of the the basic import cook.

F

Guy bedford:, assuming that we don't get a.

F

Guy bedford: import dot sync function at some point in future. This would be able to.

F

Guy bedford:, assuming we never get a synchronous import,, which currently there is no.

F

Guy bedford: and their plans for. guy bedford:, but just in terms of thinking, about.

F

Guy bedford: yeah, that that's it. thanks for putting some. and I look forward to and.

F

Guy bedford:, seeing how modules with this next week.

A

Mathieu hofman: hi, everyone.

A

mathieu hofman:, thanks for mathieu hofman: thank you.

B

Ashley claymore: thanks, pat.