Agoric SES, 10 Apr 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SES Meeting: Shadow Realms and Import Assertions

Description

Discussion of Shadow Realms globalThis, its prototype chain, and what host features are available by default. Unlike other Web realms, EventEmitter can be removed from the prototype, as can all ambient authority. Explanation of current state of Import Assertions proposal and how it may interact with SES.

A

Mathieu hofman: all right. welcome to the sess meeting today.. We will have a short of date from kari on a shadow realm, and then have a discussion on import. Assertions, kennedy. there you go.

B

Caridy: yeah,, so it does a very quick one., so we have been talking about the global opted being a an event. Target.

B

Caridy: it it seems that implementers are okay. With with that idea.. They also propose to be more basic, based on existing or prior art.. On these.. They propose that the global application have a prototype.

B

Caridy:, that is mutable,, which is different from what there is today. today. they,, the window prototype, and all the proto chain. All the way to the event target is is not mutable.

B

Caridy: not configurable. so they're proposing that that.

C

Markm: sure markm:, the the reason it was made non mutable was because of a frankly bizarre security concern., but it was a security concern.

C

Markm: has that concern security, concern gone away?.

B

Caridy: no, that I know all.

B

caridy:, I I believe caridy: so., I believe, they.. The security concern is not a problem with shadow wrong,, but I might be grown here.

B

Caridy:, so you can provide more details.. Maybe we can talk about it., but the the assumption that we have what the requirement that we have, or a shadow round is that the.

B

Caridy: global update. it can be used to virtualize environment, and therefore you cannot have any unfortunate.

B

Caridy:, including the caridy: and as a result of that,, you think that they are okay with the global to have a underscores.

B

Caridy:, a an update that has these.

B

Caridy: you can target in 10 my slots.

B

Caridy: and then who's on the score of the score. Proto is the event target for the time.

B

Markm:, that's a.

C

Markm:, okay, good., I think you're right.. Let me let me recall what what little I remember about what the security concern was. It had to do with.

C

Markm: starting a starting?, a you know,, a page or a frame as and person, html.

C

Markm: and the whole browser type sniffing thing.

C

Markm:, I I cannot remember the details of the exploit,, but I think, since shadow, since the shadow realm.

C

Markm: is never markm:. Is, is never even in a browser context. created such that there is an uncontrolled parsing of html before anything happens with the set shadow.

C

Markm:, it's not like creating an I frame.

C

Caridy:, I think I think that the the security concern does not apply to shadow all the security issue. was that because the window remains the same.

B

Caridy: october., the the you changing the prototype,, the on the screen.. That's got all the window proxy.. If you were able to change that, what happened when you not reg. to a new page.

B

Daniel ehrenberg: oh,, it's not about navigating., it's about loading script, well, navigating, might be part of it,, but it's also about having script. Tags, and.

D

Daniel ehrenberg:, which could be in the case of a worker from like import script,.

D

Daniel ehrenberg:, which daniel ehrenberg: you use as a way, or it could be through import. you.. Could you could access this through dynamic import,, this behavior, where you get some cross origin resource and you execute it., and by looking at the error you can see.

D

Daniel ehrenberg:, what that contained.? So I think, if we still allow import to bubble up to host import.

D

Daniel ehrenberg:, this issue applies., but I don't I really I I missed some context: here. why, we're zooming in on this particular issue.

D

Daniel ehrenberg: what?, what is the semantics that we would adopt that that puts this at risk at all.

D

Daniel ehrenberg:, why wouldn't we just make an immutable prototype chain for the global object in the shadow? wrong case?.

B

Caridy: why,, we will not put it as immutable.. You said.

D

Daniel ehrenberg: yeah, yeah,. Why wouldn't we just apply the same thing.

B

Daniel ehrenberg: right?, if, if it is a mutable,, then you kind of virtualize that global often you cannot.

D

Daniel ehrenberg:, let's just do that.

C

Markm:, I would prefer, if we're not., if there's not actually a security concern. and since the whole immutable global object thing was never a javascript.

C

Markm: oriented concern. we always to deal with the browser problem,, the shadow realms, don't have the browser problem.. I would.

C

Daniel ehrenberg:, I don't see how they avoid this problem,, because the problem was all about.

D

Daniel ehrenberg: importing scripts. they were from another origin, and being able to get some of the contents of that script. By looking at the error, throne.

D

Daniel ehrenberg:, because you do the 2 things that aren't scripts., and so this would be a privilege, escalation.

D

Daniel ehrenberg:, because you're not supposed to be able to daniel ehrenberg: just do a fetch to another origin., that's a violation of the same origin. Policy.

C

Markm:, how how does this come?, I mean, shadow?.

D

Daniel ehrenberg:, this is not about parsing html,, the air,, the the security issue., and I know this because I like championed making the prototype chain of the global object immutable for this particular purpose,, so that I can ship proxies.

D

Daniel ehrenberg: in 8.

D

Daniel ehrenberg:, the.

C

Markm:, how does cross origin? script execution arise if you're not parsing, html.

D

Daniel ehrenberg:, through an import, dynamic, import.

B

Caridy: so you doing inside the shadow, room? you're saying because I'm doing a with a string value that points to a cross origin.

B

Caridy: patching and executing something from an order.

B

Caridy: origin,, because I have some.

B

Caridy: mechanism plays that prevent me.

B

Caridy: from that to happen,, I will be able to capture that. what you're saying.

D

Daniel ehrenberg: yeah., so sometimes the other origin might be configured., even with all the course stuff.

D

Daniel ehrenberg: that html does correctly for fetching scripts.. It was considered that sometimes another origin might leak information in like a random text. File, fetch.

D

Daniel ehrenberg: and that you could address any kind of script thing to.

D

Daniel ehrenberg: to to that.

D

Daniel ehrenberg: I think. oh, wait!. Maybe this was unique to.

D

Daniel ehrenberg: oh, sorry., this was unique to script.

D

daniel ehrenberg: backing up a lot.. This is unique to script tags and not to imports,, because.

D

Daniel ehrenberg:, when you import a module., it takes the mime type, and when you use a script run, it does not take the mime type.

D

Daniel ehrenberg: all right,, but that's, but this is.

D

Daniel ehrenberg: okay. daniel ehrenberg:, that was the crux of it. sorry for my confusion.. There were several different issues that about html integration, right?. We we should go over like all of them, while we're talking about this.

C

Caridy: that would be great, because I'm. completely unfamiliar with the html side of shadow realms.

D

Daniel ehrenberg:, like one of the issues, was which things are we going to expose? and what is our rationale for that?.

D

Caridy: yeah, but that., but maybe it's that's, maybe a step.. So the first one is that we have any concerns about what they are proposing to do,, which is a a proto chain that is configurable, and the.

B

Caridy: intermediate update was going to be the one that will have it to a stream.

B

Caridy: behavior to print out, shut around global update. If you try to make it to.

B

Caridy: and then that one will inherit from a event target prototype. do we have any?.

B

Caridy: yes,, I can link it to it.. You've got to find it.

C

Markm: walker is doing that. I just want to make sure that I'm understanding the conclusion of that previous discussion that it does have to do with parceling, html and particular script tags because of the sniffing behavior.

C

Markm: and that because markm: dynamic import only does modules, not scripts.

C

Markm: modules, ha! are are not,, are based on lyme types,, not sniffed..

C

Markm: not sniffing at content.

C

daniel ehrenberg:. So therefore the concern does not arise. is that the conclusion? be clear?, we're not talking about sniffing that doesn't come into this. we're just talking about not taking the mime type.

D

Daniel ehrenberg:, the when you have a script tag it doesn't decide what to do,. Based on like reading some of the contents of the script., it just ignores the mime type.

D

Daniel ehrenberg:, I see., I see there are other things where sniffing comes into play. okay.

C

Markm: okay., and so so in this case it just comes down to scripts,, ignore mind. Types,, modules,, don't, and imports, can only import modules,, not scripts,, and therefore.

C

Markm: the security issue does not apply to shadow around.. Is that the conclusion?.

C

Daniel ehrenberg:, I think so., but I also like the idea of making a global object still immutable.

D

Daniel ehrenberg:, let me let me catch up on this thread.

B

Caridy:, this is the one., so all the comments below it's just saying that this is the way to go.

B

Caridy: but basically caridy:, which is a who's for on the score of the will point to shadow round global scope for the.

B

Caridy: who's on profits will be constructed, and the to strengthen.

B

Caridy:- and you will in point to on the score to the event target, with all the things that we have today, and then that we had it from.

B

Caridy:, that's kind of caridy:, the the the the reason why we need this intermediate one.

B

Caridy: is because otherwise, we'll have a a global update that we need to do to string on it it it. It will print out event target.. So it's like this one.

D

Daniel ehrenberg: yeah,, you know,. I think we should do the web platform usual pattern.

D

Daniel ehrenberg:, so I agree with the other people here.

D

Daniel ehrenberg:- and we could define this this with web idl, where we make an actual show room global.

D

Daniel ehrenberg: interface, daniel ehrenberg:, that we say just do the steps. for you, know, initializing, an instance of that interface.

B

Caridy: I was trying to to work with daniel on that., but I didn't really have any time.

B

Recycled to work on this caridy:, the last few weeks, and I don't inspect, I will have any on the next few weeks.

B

Caridy:, so any any objective comments on this particular topic.

D

Mathieu hofman: nope. caridy:, okay, perfect.

C

Markm:, so the conclusion is that.

C

Markm:, the prototype chain is mutable, which enables emulation of hosts.

C

Markm: that do not ha that that have other natures of global object.

C

Go ahead. okay.

A

Mathieu hofman: yeah,, it means you can get rid of all the even target stuff. If you want.

C

Markm: great caridy:, yup.

D

Daniel ehrenberg: so.

D

Daniel ehrenberg: good., so that's one aspect. we were also.. Does this mean that when you have an unhandled rejection happening within the shadow realm,, then it gets triggers.? That.

D

Daniel ehrenberg: shadow realms an annual rejection, event.

A

Mathieu hofman:, so what we,? I think, the last discussion, and correctly, I'm wrong already.. But the last part of the discussion was that unhandled rejection propagation is currently broken. Anyway. for example, between a worker and.

A

Mathieu hofman: and the main thread and shadow realm is not gonna. Try to fix it,, because there are. there are issues about the promise identity that that are needed for the the handlers, that we cannot provide it through the boundary right now,, so we'll need to figure out a way to handle this.

A

Mathieu hofman: appropriate it.

D

Daniel ehrenberg:, sorry, when you say broken,, you mean it, doesn't. propagate from one video.

D

Daniel ehrenberg: correct that doesn't sound broken to me.. I think, if we shadow around being an event target means that it can have its own unhandled rejection handler, and that the promises rejected from inside the shadow room would go to that,, and that seems like.

D

Daniel ehrenberg:, how we would want it to work.

D

Daniel ehrenberg:, I was. mathieu hofman: yes,, what's that?.

A

Mathieu hofman: was that gonna be possible? by the way,. It wasn't going to be possible if it wasn't an event. Target.

D

Daniel ehrenberg:, because then they error be reported to it.. I mean,. This also applies for exceptions.

D

Daniel ehrenberg: unhandled, the exceptions thrown in the shadow room.

B

Caridy:, that one will those who will find and those will work fine with the semantics,, but they will, they will.. They will propagate.

D

Daniel ehrenberg:, if they're not handled., will it call window that on air on the shadow, room, or we'll call window that on air. on the closing thing it will first call.? I I believe it's going to be the same behavior as worker,, which is, it will first call the window on there in the shadow.

A

Mathieu hofman:, and if that doesn't handle it,, it will provide it up to the parents. Well,.

A

Daniel ehrenberg: or to that. no, no,, we doesn't go through our list., it just goes back immediately.. The.

B

Caridy:- and there is an issue with that. yes,- it's written down that issue.

A

Mathieu hofman: yeah.

A

Mathieu hofman:, it's been a few months,, so I'm a little bit details., but.

D

Daniel ehrenberg:, so this is your comment from november 20 ninth,, where.

D

Daniel ehrenberg:, I don't., I don't see where the conclusion about this is written.

D

Daniel ehrenberg: okay, anyway,, if we agree on that,, then that sounds good.

B

Caridy: yeah., so I I have a comment on december 14.

B

Caridy: that those were the notes from the discussion here about that particular aspect of it.

B

Caridy: but no conclusion on that one, but this.

D

Daniel ehrenberg: so below your comment from that there. dominique then suggests that there should be this intermediate prototype object.

D

Daniel ehrenberg:, which is like the shadow room.

D

Daniel ehrenberg: type of thing, prototype., daniel ehrenberg:, and that makes sense to me,, and that's where, like things like structured phone, would live.

D

Daniel ehrenberg: does that seem reasonable to you?.

D

Daniel ehrenberg: or I mean,- maybe not structured clone,, but 2 string tag at least.

B

Caridy:, the string tide. yes,, the I would think about that. what?. Why do you think that that will be there? oh,? Because, just because I got myself confused.

D

Daniel ehrenberg: so.

D

Daniel ehrenberg: okay, great., I'm happy with this design.

B

Caridy: then, the last thing I think the daniel mentioned is that we need to figure a better way to explain.

B

Caridy:, what should be added to the shadow wrong on what should not be added to the channel, and the main discussion between that and me, and being about fetch. I.

D

Daniel ehrenberg:, the example that anna brought up wasn't about fitch.. It was about different event, types.

D

Daniel ehrenberg: with, and it kind of making the argument. either we should include all the events, or none of them., but the event constructor itself doesn't do any I. o. so. the question was,. Why? In the current draft. it included some, but not others.,.

C

Markm: and to clarify, we're talking about daniel ehrenberg:, the graph of the html integration of shadow realm,, not the graph,, not the ecograph yeah,, the excellent draft everyone's happy with,. I think. okay, great thanks.

B

Caridy: that that we discussed this here. and then it was not here at that time.

B

Caridy: we, with our position. was.

B

Caridy:, we really don't care if it is there.. If it should be,, it should be mutable.. That's all we really care about, and if they want to, whatever they want to add. that's fine., but daniel has some.. I just think we have to come up with a concrete suggestion.

D

Daniel ehrenberg: and write down the logic around it., that's what they're asking us to do.

B

Caridy: and part of the discussion with daniel was about prior these questions around I o,.

B

Caridy: and how we want to prevent I/o caridy: from happening in the shadow around at some point.. I think we dropped our bones: in. well, it is. and.

D

Daniel ehrenberg: yeah,, so I heard that last time you were discussing this, you were going to include fetch and all this stuff.

D

Daniel ehrenberg:- and that sounds really broad in a way that I don't think anybody's asking for. I think. anna, when I talked to him in the past. I can't remember if it was in a threat or not.

D

Daniel ehrenberg:, he seemed., he seems sympathetic with the idea that these things would not do. I/o,, that that would be the organizing principle,, and so we would include all the classes that don't do. I/o makes good all the ones that do do. I/O?.

D

Daniel ehrenberg: and I think that's pretty easy to to find and work out.. The problem was that.

D

Daniel ehrenberg:, what what leo had made before excluded a bunch of things that don't do, I/o? and there was no clear, organizing principle around it.

B

Caridy: yeah., and at least caridy: part of the discussion that we have out in the past is that if you exclude those things.

B

Caridy:, you should have a very strong reason for doing that, because it will hinder their ability to use the shadow wrong.

B

Daniel ehrenberg:, I so that's what I was confused: by,, because.

D

Daniel ehrenberg: that because daniel ehrenberg: for fetch,, I thought it would make sense to have a membrane around the fitch that the outer thing does, and that's how you would.

D

Daniel ehrenberg: well,, I think you can't get very far with shadow realms without a membrane.

D

Daniel ehrenberg:, that's the consequence of the you know strong boundary design. We adopted.

B

Caridy: fetch very soon dynamic, import.

D

Caridy:, but you were saying that they are not the same, and we should not,, because one of the things that we mentioned in the past at least, that I remember mark.

B

Caridy: was that? well,, there is a you already., you do, and you can., I mean you can do, import, and you can interact with some of the timers.

B

Caridy:, you can do a bunch of things,, so.

B

Daniel ehrenberg: draw the line between what you can do or not.

D

Daniel ehrenberg: in in general,, I guess I'm not a big fan of that.

D

Daniel ehrenberg: style of argument, daniel ehrenberg:, like oh,, you know it's a slippery slope, anyway., I don't know,. I think, saying the only I o that you're doing is import, and all other I, o. is prohibited. Is is a clear thing, and timers are not the same as I/o.

D

Markm: I'm sorry. where? where?. Where do timers arise in this cotton.? You just mentioned it.

D

Daniel ehrenberg: so.

D

Daniel ehrenberg: of of being included in shadow realms., so I think they don't.

D

Daniel ehrenberg: sure they're the kind of thing that you might walk differently,, but but they're not.

D

Daniel ehrenberg: side affecting the world in the same way.

D

markm:, it's just because very quick. timers are included in the html. Integration of shadow rooms are not included. In.

C

Daniel ehrenberg:, I think the integration in the h, email, integration.

C

Daniel ehrenberg: okay, and definitely not in the javascript by itself. version.

D

Daniel ehrenberg: okay., so daniel ehrenberg:, I don't know there was previous.. There was previously kind of a coherent goal.

D

Daniel ehrenberg: of we're not going to allow any I, except for in that we can port.. It is a pretty restricted form of I/o,.

D

Daniel ehrenberg:, even if it's not that restricted it's it's less powerful than exposing all the fetch.

D

Daniel ehrenberg:, so I thought it made sense to draw that line.

D

Daniel ehrenberg: and we didn't get pushed back from browsers on that they seem to these.. I had the impression that they that they thought that that was sensible.

D

Daniel ehrenberg:, so I'm not sure why we would go reverse it. Now.

D

Daniel ehrenberg:, if you think it's really important to have this convenience of using fetch in a case where you don't have a membrane.

D

Daniel ehrenberg:, I would like to understand more about that. usage pattern, I'll I'll just. I'll just insert my 2 cents here,, which is.

C

Markm:, I'm surprised that everybody was okay, omitting fetch., but.

C

Markm:, given that they were, I'm happier.

C

Markm: to go with the omission.

C

Markm:, I I expected to be forced to to accept, fetch.. So if we can go forward, omitting fetch, that's great.

A

Mathieu hofman:: what is the benefit of meaning any I/o capabilities,, since we can remove everything anyway,.

D

Daniel ehrenberg: it's.

D

Daniel ehrenberg:, I guess the idea is that this has the safe defaults., which is maybe the same extremely high level idea behind having the the strong boundary. in the first place.

D

Daniel ehrenberg: that then you build something to communicate over as opposed to needing to build the boundary with a bunch of work.. That's good.

B

Caridy: yeah,, I I don't., I don't remember really the conversation., but all the details of the conversation. from what I remember. The conclusion at the time was that. well.

B

Caridy:, it's too late to try to.. Those are the I o.

B

Caridy:, in the sense that you you'll be able to do some I/o already.

D

Caridy: and then obviously, we know the downside of not having fetch.

B

Caridy:, we know what that is with more of all. well.

B

Caridy:, it's fine, because you will be able to remove it.. If you want to know it.

C

Markm: okay,, I I buy. I buy both of daniel's arguments, here,, both the rationale for why you should default to to to omitting it.

C

Markm: and why membrane base is a perfectly fine way to add it back in.

A

It's pretty heavy call,, I mean I'm I'm., I'm thinking, for example,, like.

A

Mathieu hofman: a testing framework, or something like that that I might want to use shut around.. It's pretty heavy cost to us to come up with a membrane to be able to add back.

A

Mathieu hofman:, I I you.

D

Daniel ehrenberg:, so let's, let's continue this thread. I'd, really like to understand how people want to use shadow room without a memory.

A

Daniel ehrenberg:, I mean that that replicates like if you're in node in.

A

Mathieu hofman: creating a testing environment, that within the browser in the creating that setting environment, that's in another environment.

A

Mathieu hofman: that replicates mathieu hofman:, it's like that., if that other environment has a.

A

Mathieu hofman: me mathieu hofman:, most of the time,. You don't need to go through the the global boundary for for some of the things you can just virtualize on top of the cabinet that we're present in the shadow.

A

Mathieu hofman: a testing, a a testing environment might not need to go back and forth between between the realm and the in the shadow.

D

Daniel ehrenberg:, wouldn't you want nodes, I/o capabilities to be exposed to the testing environment.

D

Daniel ehrenberg:, which are different from the ones like., are you?.

D

Daniel ehrenberg: are you expecting that the node built in will also be available?.

A

Mathieu hofman: no;, but you,. If you leave them in., if you leave the, if you leave the host capabilities and you implement your I/o capabilities that you're emulating on top of them, hiding the the original one.

A

Mathieu hofman:, you don't need to go through the callable binary.

A

Daniel ehrenberg: and you don't need a membrane.

D

Daniel ehrenberg:, what's the goal of using.

D

Daniel ehrenberg: shadow realms? in the first place for this use case, where you should run your test and your you can go and and un your your whole.

A

Mathieu hofman: all the intrinsic, and then just kill it again,, like you know that. That's why testing framework is always creating your own.

A

Daniel ehrenberg:, so it's basically like a lightweight process. That's not going to interact with the javascript that long shit at all just interact with the rest of the world via I/o.

D

Daniel ehrenberg: is that the usage scenario you're picturing.

D

Mathieu hofman: yeah, and reporting results back through the callable binary.

A

Mathieu hofman:, so the only thing that needs to be reported back to the whole with all is the result. Test.

D

Daniel ehrenberg:, so yeah, so reporting testing results could be done in a way. That's lighter weight than having a full membrane.

D

Mathieu hofman: okay.

D

Daniel ehrenberg: sh daniel ehrenberg: it's. it's hard to picture this, because I think.

D

Daniel ehrenberg:, a testing library, would want to do more I/o than the web platform provides.

D

Daniel ehrenberg: wouldn't it.

B

Caridy:, I'm not., I guess you're talking if you're walking the things that you're catching, or something like that. sure you can., you can do that., but.

B

Caridy: most most cases just give access to.

B

Caridy: getting getting the content directly, like a local content.

B

Caridy: I mean, in general, like it's just caridy: you.. You can do many things that are self contained inside the shadow room and then.

B

Caridy: share results or not necessarily test yourself,, but share data. That is.

B

Caridy: for us. they call the boundary.. So all the operations are happening there like a plugin system for you to do calculations or whatever you're doing.. You do it there and then eventually.

B

Caridy: we'll be with the inquiry around.

B

Caridy: so the other,, the other form of I,, that's.

C

Markm:, even more universal than fetch., much more is consolid.. What has been said about console?.

C

B

Caridy: in the in the current, in the current implementation., when you look at the the.

B

Caridy: behind the.

B

Caridy:, but if you don't have console,, then how do you, when my console is not simple?, it's not that simple.

B

Caridy:, you need just to do console.. You need to have a membrane.

B

Caridy:, you just want to run some code there that does something.

D

Daniel ehrenberg:, I'm more sympathetic to that in console than I am to fetch.

D

Caridy: yeah, I. this is exactly the same conversation that we have in the past.. We start from., you have been port, and then one of our console.. What about this other thing? and say, yeah, sure to block that.

B

Daniel ehrenberg:- I I you know, I'm not the champion of this proposal., it's up to you what to propose., but I just.

D

Daniel ehrenberg:, but it was. it's clear that we developed this.

D

Daniel ehrenberg: I. my understanding was that we had developed this cross browser consensus on this organizing principle that in general we were going to omit. I. o,. Maybe we're happy with that.. If you want to have a new principle., then you'll need to state that principle.

D

Daniel ehrenberg: and have that discussion with the with the browsers and get them to agree to it.

D

Daniel ehrenberg: and that's really doable,, but it requires like.

D

Daniel ehrenberg:, thanks for explaining yourself about what the concrete proposal is rather than just saying we're: okay, with anything.

D

Daniel ehrenberg:, because you're, I don't think the I/o wasn't a stated principal,, but.

B

Caridy: but sure we can. we can.. If we agree that we will allow any kind of I.

B

Caridy: sure we can., we can do that.

D

Daniel ehrenberg:, I mean what, whatever you want to state as a.

D

Daniel ehrenberg: conclusion,, whether it's including fetch or excluding it., both the principal and the listing of the apis, are the kind of the deliverables needed.

C

Markm: okay, I'm:, I'm still prefer to see us exclude it., but I also.

C

Markm:, you know, admit that as long as you know, since since it's.

C

Markm:, since, whichever way you default it,, you can build the other one.

C

Markm: out of it, and markm:, the and it's,, and the issue is absent from the javascript standard. anyway.

C

Markm:, I don't.

C

Markm:, I don't have to to weigh in here in a strong way.. I do. also.

C

Daniel ehrenberg: does the does the web standards, standardize, console part of web standards.. It's there in what week.

C

Markm: it is. okay. daniel ehrenberg:, so.

D

Daniel ehrenberg: by the way,, when you write an explanation to this group,, it should remain free of.

D

Daniel ehrenberg:, the daniel ehrenberg:, the sort of technical language that's used in these discussions shouldn't use the word unforgible.

D

Daniel ehrenberg: for example. daniel ehrenberg:, because that's a statement of the goals of this group rather than the goals of them., I mean just.

D

Daniel ehrenberg:, if you want to say like, there shouldn't be any properties that are not deleteable or something.

D

Daniel ehrenberg: I and then we're okay.. I don't know just just to enable communication here.

A

Mathieu hofman: right? mathieu, hofman: oh.

A

mathieu hofman: great!. Is there anything else on this topic? or should we switch to the topic of in 4 assertions?.

C

Markm:, I just want to say that with that,, not being deleteable and being unforgeable, are distinct, concepts, and the first requirement should be precision.

D

Daniel ehrenberg: yeah., the word unforgeable, has been used in multiple ways. By this group.- I think.

D

Daniel ehrenberg: so.

D

Daniel ehrenberg:, I agree that things have to be precise,, but also terms like unforgible.. Just have to be not used.

C

Markm:, unless what we're talking about is unforgeability, that has nothing to do with unrelatability.

C

Markm:, a fresh, a a fresh anonymous symbol is unforgeable.

C

D

Daniel ehrenberg: concept., but there was a different concept used about things like window top.

D

Daniel ehrenberg:, where some other piece of jargon was used daniel ehrenberg:, to refer to the undelete of that.

C

Markm: right, and that and the reason not to use the term unforgeable, for that is because unforgeable is the wrong term. For that it is not an unforgeable issue.. It's not the issue.. It's not unfortunate.

A

Mathieu hofman:, I believe, this.

A

Mathieu hofman: the it has to be configurable, and the configuration has to succeed, or something like that, like it's, not like.,.

D

Daniel ehrenberg: okay,, that's that's great. yeah,! As long as we stick to like.

D

Daniel ehrenberg: describing the observable behavior as opposed to words that describe the goals.

D

Daniel ehrenberg:, which then have to be like,, decoded and worked through.

D

Daniel ehrenberg:, then I will. it'll work.

A

Mathieu hofman:, I think? I asked., I I don't know if it is now,, but I I think I had to make sure that., but the success of the configuration,, because if it's an exotic object can be like,, okay,, it's configurable. and then yeah, actually doesn't change.. If you try to reconfigure it.

A

Daniel ehrenberg:, all right.

A

mathieu hofman: all right,, let's switch to it for assertions. who wants to drive this conversation, mark,. You have to put this on the calendar.

A

Markm:, I'm not the one who put it on., I just the one who keeps, you know.

C

Markm: postponing it until we have critical mass to discuss, which we finally have today.

C

Daniel ehrenberg: but yeah, I can., I can discuss it. it'll be better if we had.

D

Daniel ehrenberg: niccolo here., but I've been participating in some of the discussions,, so.

D

Daniel ehrenberg:, let me daniel ehrenberg:, write him to see if he's available to join.

D

Daniel ehrenberg: okay. well,, I wrote him,, but I can. I can start explaining.. So in the last meeting we we discussed how it's not enough to have import assertions.. We have to be able to.

D

Daniel ehrenberg: modify what the module does,, or at least how it's fetched,, which changes what it,, what it might be.

D

Daniel ehrenberg: to meet. each email's requirements. so.

D

Daniel ehrenberg:, there are kind of 3 3 questions from there about what the in expressiveness of this should be. one question is,. Should we maintain the arbitrary key value nature.

D

Daniel ehrenberg: of import attributes.

D

daniel ehrenberg:, or should we limit it to just one key?, and I think that the j. me. and group's opinion is,. We should maintain the multiple key aspects of it.. The multiple attribute aspects of it.

D

Daniel ehrenberg: another question: is.

D

Daniel ehrenberg: should.

D

Daniel ehrenberg:, what should happen when unsupported assertions or or unsupported attributes are are given.

D

Daniel ehrenberg: I mean currently in imported search and unsupported attributes, through. and so we're we're saying, this. still,. It makes even more sense for import attributes, because they could change the interpretation, and you wouldn't want to kind of lose that.

D

Daniel ehrenberg: interpretation change?.

D

Daniel ehrenberg: and then the question is,, are they assertions like,? Should we use the assert syntax, and the proposal of the champion group is to switch to a syntax based on with, instead of assert.

D

Daniel ehrenberg:, where the the idea is,, we would keep, assert in as in line,, normative optional, comma legacy,, where we somehow start to define legacy.

D

Daniel ehrenberg:, with the goal that, if it's possible,, maybe we could delete it in the future., but it's included there. Now.

D

Daniel ehrenberg: and, to the extent that normative optional is like an all or nothing thing.. This would be acceptable as kind of piecemeal omitted,, so we would definitely not be asking from, or or node,, or anybody to unship.

D

Daniel ehrenberg: import assert., but daniel ehrenberg:. We would encourage new things to use import with instead.

D

Daniel ehrenberg: so.

D

Daniel ehrenberg:, so this is the proposal.

D

Daniel ehrenberg: it would not., we would simply delete the paragraph that says that you can't interpret a module based on the import descriptions.

D

Daniel ehrenberg:, based on the import, attributes., now,, instead, they're, just part of the cash key. and with romulo's refactor of.

D

Daniel ehrenberg:, with romulus refactor., it's not ramulos nicola's refactor of the way that modules work. there is a cash at the javascript level. In addition to at the.

D

Daniel ehrenberg: host level,, so that c to the javascript level would be key by both.

D

Daniel ehrenberg:, the specifier and all the key value pairs for the attributes.

A

Mathieu hofman: so mathieu hofman:. I guess that's where I was going to go for the question.. I thought mathieu hofman:. There were discussions about having so much reviews being.

A

Mathieu hofman:, not part of the cache. and having like.

A

Mathieu hofman:, basically for for extensions or things like that., so.

D

Daniel ehrenberg: for that aspect.

D

Daniel ehrenberg:, the daniel ehrenberg:, the attributes after the with would all be part of the cash key. however,. Other things like.

D

Daniel ehrenberg: deferred imports, or., you know, import module, like the the different kind of modes for importing something,, so that it wouldn't be executed.. Those would be not done through.

D

Daniel ehrenberg: import attributes,, but instead done through a keyword that comes.

D

Daniel ehrenberg:, at the beginning, right after the infant keyword. and the reason for that is because those are not part of the cash key., but instead about how javascript runs the module.. The other thing is that those.

D

Daniel ehrenberg: effect the syntax of the module import statement,, so they would be added as individual syntactic entities.

D

Daniel ehrenberg: and those would all be defined in javascript.

D

Daniel ehrenberg: and that's how that's, how import module or import source or important for would work, then, for.

D

Daniel ehrenberg: extensions like., if you have a tool, daniel ehrenberg: or a runtime environment, that wants to define its own attributes.

D

Daniel ehrenberg:, you would be strongly encouraged to use a namespacing scheme.

D

Which might look like daniel ehrenberg: start with an underscore daniel ehrenberg: to indicate that it's kind of private and then right the name of your kind of namespace,. So it might be node, and then capital like in camel case,. The name of the attribute.

D

Daniel ehrenberg:, so you might do like.

A

Mathieu hofman: and those would become part of the cash key as well.

D

Daniel ehrenberg: yeah,, but they would be rejected at runtime if they ever made it into a final program.

D

Daniel ehrenberg: I, but they would be part of the logical cache.

D

Daniel ehrenberg:, so we would make this strong encouragement. Just like people are encouraged to use data, dash, attributes.

D

Daniel ehrenberg: for for custom elements or things like that.

D

Daniel ehrenberg:, for when using import, attributes, you'd, be strongly encouraged.

D

Daniel ehrenberg: to choose a name that is not likely to be a future name based on it., starting with an underscore, and then starting with, like the name of your project.

A

Mathieu hofman:, but so that that means these, this source files will not be.

A

Mathieu hofman: executable without being transformed.

D

Daniel ehrenberg:, that's right. yeah. daniel ehrenberg:, so this is airing on the side of safety.

A

Mathieu hofman:- and I know justin was a big proponent of being able to support this used. Case.

A

Daniel ehrenberg:, yeah. mathieu hofman: so, and that that is something just is comfortable. With.

D

Daniel ehrenberg: yeah. justin's been very much part of these discussions, and even proposed this concrete naming convention.. I would have imagine something more like.

D

Daniel ehrenberg: and we're more like with a colon or something like that.

D

Mathieu hofman: or to to catch those, or how does import hooks?? We definitely have to receive.

D

Daniel ehrenberg:, the all the key value pairs of the import attributes.

D

Mathieu hofman: okay,, so it's possible the source would be able to be evaluated.. It's just that. You need a runtime in port work that can do the appropriate.

A

Daniel ehrenberg: yes, exactly.

D

Daniel ehrenberg:, if you're using the daniel ehrenberg: module constructor, then, yeah,, you could., you could emulate all this stuff.

D

Daniel ehrenberg: and there would be no particular difference in terms of the import hook about whether the names started with an underscore or not.. That's purely a convention.

D

Daniel ehrenberg: yeah,, but we would have a promise going forward that all the built in things would not start with another square. Just like html has the promise that it's not going to have any attributes with the dash in them.

B

Mathieu hofman: okay.

D

Daniel ehrenberg:, how does it send to you?.

A

Mathieu hofman:, I don't know it sounds fine,, but I I have not been involved enough in the conversations to.

A

Mathieu hofman: no., if I'm missing mathieu hofman: some some of the arguments. so.

A

Mathieu hofman:, it sounds reasonable.

D

Daniel ehrenberg: okay, that,, thanks to any other people, have thoughts on this.

D

Daniel ehrenberg:, or did I go too fast?? Should we go into something in more detail?.

C

Markm: I want?. I want to explicitly solicit kurti's thoughts on this.

B

Caridy:, I don't have a. I was on the on the in a meeting as well in your group, and.

B

Markm:, but in particular, about how it interacts with the whole module loader.

C

B

Caridy: at least the way I caridy: I think about this is.

B

Caridy: I I specifically about layers 0 like a a module constructor more the social structure.

B

Caridy:, if you're gonna get caridy: access to that update with the with.

B

Caridy:, I I feel that we have a lot of power that we can exercise. there is,, I don't believe., but at this point I don't believe that the browser.

B

Caridy: will ever implement caridy: a bunch of different 4 months.

B

Caridy: and it. it. it. It feels to me that the.

B

Caridy: providing the the user they, the low-level api, so users can do that kind of thing,, especially bunders and tooling like,, and rely on this,. We gave you to.

B

Caridy: provide information caridy: through the import statement to the loader. That would rely on this information to do the proper resolution, and the super creation of a.

B

Caridy: as a module infants based on this information.

B

Caridy: and the module hope.

B

caridy:- I think that's very powerful. caridy:, but every time that I talk about that, I didn't get much traction.

B

Daniel ehrenberg:, so when you say powerful,, do you mean that as a good thing or a bad thing?, okay, cool.? I think I I I see.

B

Caridy: implementers of bombers and tools caridy: asking for the browser to do that for them like.. Let's have a.

B

Caridy: new api that you're going to specify what kind of things that we're going to load, and then they will do all the the allegiance to get it ready for them.. I don't think it's going to happen.. I did that and it it directly,, but he was saying,. You know there is the interest on that., but I I I don't buy that we on it.. So I think.

B

Caridy:, if we do something that will not be caridy:, that will not provide it. There, that that that powerful, api, or in order to.

B

Caridy:, do all these things and use a line,, we'll probably make a mistake., but.

B

Markm: again, like not not not really much much there markm: so.

D

Daniel ehrenberg: powerful. and I guess what the the champion group is proposing that we add this power.

D

Daniel ehrenberg: so to the extent that you're saying we're not getting much traction. okay.

D

Daniel ehrenberg: yeah,, it's not., it's I'm not getting one fraction. Every time I talk about.. Maybe I just not articulated.

B

Caridy:, so it's like the optic company who gets that okay decides what to do with it. and it's primary.

B

Caridy: objective is to create a launch of linkedin in whatever shape or form that modelington's needs to be. Created., represent a cs. to represent some of the format, or represent a awesome module.. Whatever it is, it's representing. we can talk about that,, but.

B

Caridy: it it it to me that rely on the user agent for all these., it's just not going to have.

D

Daniel ehrenberg: well,, what about us as well?

D

daniel ehrenberg: okay.? So what about the bundlers?? Because we will initially wouldn't have the module of the structure?.

C

Markm:, what about the goal of of amyo being able to emulate one host on another house.? Is that all? well supported by.

C

Markm:, what we're saying here?.

D

Daniel ehrenberg:, I think this would. this would permit those things.

D

Daniel ehrenberg: okay., in conjunction with the module constructor.

D

Daniel ehrenberg:, so I think the the hesitation that we might hear from engines is around, whether the module constructor should be supported short of having the module construct, training, emulation or power would have to be done by the bundler.

D

Daniel ehrenberg: right? okay.

B

Caridy:, but if we don't have the monitor constructor,, the calendar is kind of a little more, either. well, the bundles would just not use modules., they would just., they would do something out, they can., they could do things they just,, but that, I think that's that's really a good point, like if we don't have a module, constructor.

B

Caridy: then they will continue to what they do to pay. if we give them the module constructor,, why even bother to implement all these other things?? I I do want to be able to make the argument for import attributes in a way. That's independent of the module constructor.

D

Daniel ehrenberg:, because I think we don't have consensus in the committee on whether the module constructor should exist. yet, and hesitation about the module constructor shouldn't.

D

Daniel ehrenberg:, the the same as his station about import attributes.. They just separate proposals., even if we want them both to happen.

B

Caridy: yeah,, you don't., you don't have that. and the only consumer of bill is that it will be the they use. The agent. we're fine.

D

Daniel ehrenberg: yeah,, the user agent and tools., I think the tools part is- is important.

D

Daniel ehrenberg:, so about about the import hook., there's a thing where we would have to pass an object for all the key value pairs into the import hook., and I guess this would get a fresh identity for each..

D

Daniel ehrenberg:, you know, import statement.

D

daniel ehrenberg:, and there was some idea at some point. Did this be tied into records and tuples.

D

Daniel ehrenberg:, that you would get a record there., but since we got hesitant feedback from browsers on records and tuples.

D

Daniel ehrenberg:, I don't think we should necessarily count on that, and we should try to.

D

Daniel ehrenberg: convince ourselves that we're okay, with this being based on.

D

Daniel ehrenberg:, it's being based on an object with those keys.

D

Daniel ehrenberg:, so having I'm fresh., what are the external values for those things? Get can be like objects and, like you, can't really have anything., they can only be strings.. The values can only be strings. okay.

B

Caridy: yeah., but if, if you, you, you not observing that.

B

Caridy: identity, user, lab.

D

Daniel ehrenberg: no,, you could.. If you have the total constructor,, you can't observe that.. So we go back to the the the constructive thing. conversation, like the identity. Thing is only if you have a monitor contract,, you don't have that many.

D

Daniel ehrenberg:, so I'm just talking about daniel ehrenberg:, something that we would observe with the interaction of the module constructor. and this proposed.

B

Caridy: yeah,, I I would say that sure represent. Google looks really good for that case,, but.

B

Caridy: they have absence of that. I I feel that the the module hope,, whoever is implemented and more of should not rely on that.

B

Caridy: but the volumes associated to that up to the.

A

Mathieu hofman:, I see just inside.

A

Mathieu hofman: you're muted, justin.

E

Justin ridgewell: alright., so there are a few topics, one bundle. That's fully supported.. With this, we can do everything essentially.. You also asked about emulation.. The only thing that's not directly emulatable is the module attribute.

E

Justin ridgewell:, which gives you an uninstantiated module instead of an instantiated module,, and that's because it's not a part of the attributes.. It's a part of the import declaration. Itself. on this topic of.

E

Justin ridgewell: the justin ridgewell:, I lost my place.. What were we just talking about module constructor, and the identity? You would. The attributes themselves are a single-depth object. All the key values are primitives., the keys are all strings. The values are our numbers,. I guess. no, they're strings.

E

Justin ridgewell: they're only strings., so you just need a single depth: deep. the quality matcher. In order to check your cash key here.

B

D

Caridy:, I think what you're saying is that identity is not super important, that we,. So as long as we make the well defined, namely, that each import statement gets its own object, identity for the attributes.

D

Daniel ehrenberg: and it's kind of useless in the first place.

D

Mathieu hofman: or we can say that the the keys are ordered lexically, and then you can just jason's 3, the 5 thing, and.

A

Mathieu hofman:, that's for kashy.

E

Justin ridgewell: yeah, that would work justin ridgewell:. That would that would aid the if you're, storing this on a map as well,, because you you're guaranteed an order,, and you know the cash. he would be equivalent to the string of by that would be pretty convenient..

D

Daniel ehrenberg: yeah,, I guess we can decide whether or not the keys should be sorted when defining the import hook right and defining the model of instructor.

B

Justin ridgewell: yep.

D

Daniel ehrenberg: okay., so do people have any concerns about.

D

Daniel ehrenberg:, this direction for import assertions, or, if you any further comments.

A

Mathieu hofman:, so all right. mathieu, hofman: well,. I guess we are at time so good timing. We didn't overflow today.

A

Mathieu hofman:, so thanks everyone.