.Net Foundation Design Reviews, 21 May 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: Connection Abstractions

Description

Powered by Restream https://restream.io/

A

A

Hello friends we're alive again last week we this week was built. Hopefully you did not watch us on Tuesday if you did I'm sorry for you, but there is a auction everything I'm from home from builders available. If you go to channel 9, you can watch all the sessions which you definitely should, and so meanwhile, we'll continue our discussion on the connection abstractions that we talked I think last Thursday and lots of notes. Let me just scroll down to the notes.

A

I think it's probably faster.

A

Amos here we go.

B

All right so just take it from the top I guess: should the HP interfaces or classes I think that with a advent of DIMMs interfaces, aren't too bad here, I'm, not sure what abstract base classes will get us there, but I'm not opposed to I, guess prototyping, something and seeing how it looks with base classes. So.

A

One word on dims, so keep in mind that many you rely on dims, you may cut off languages that don't support them and then. Secondly, there is this larger work item that we still haven't fully tackled, because nobody has I guess has sit down yet but, like their students is concerned, they don't quite know enough about DIMMs to really understand what the servicing constraints are. So there's still the fear, because we found out in some cases that editing dims will break stuff right, which is the question of like way.

A

You put the default method in interface hierarchy. So there's still this concern that, because it's new, you know you might not think you get the benefits you're actually getting on. So in general, it's.

C

Gonna seem alien generally, what's gonna cause us to learn, but those downsides are I. Think part of it is just we have to prototype it and actually see what what.

A

Actually falls off because, when learning by shipping is not great for customers in highly used abstraction because they may corner of ourselves right, I got.

D

One for you, so so a striker is shipping to things that use them. So we'll tell you in a.

A

Few years, well, that's the thing I like I like, ideally, we would prototype something earlier and then just basically simulate what typical versioning looks like and see whenever there anything breaks like I think like actually, shipping is not great, because, first of all, it takes a long time and it's. Secondly, when we actually break people at that point, you're kind of screwed. You have me can't.

D

A

Back in time, the.

D

Feature silly: it's like telling customers don't use it until we finish the feature which is intended.

A

D

A

What I'm saying I'm saying so we know what the feature does we know how the feature.

D

A

But what happened I'm saying is that using it as a as a way to after the fact change, the interface is very different from just adding them right, because, by the reason why we added them was Java Interop, which has I think the question is just similar to anything that we do in an API design. It's like what does like. How do we version this over time? Right and I mean yeah.

A

We ship default properties right, not even properties default parameters a while ago, and it took us some time to figure out how we can actually use them safely so that we can actually version our API. So after we edit them right. So that's it's kind of an orthogonal item to how the feature works.

A

He's not entirely orthogonal frost but color.

D

Me skeptical, it's fine, I mean I, don't have a huge push back, I I would say like if we can make the base class work and Castro this. That seems, okay,.

A

It's the one thing I'm still not clear on it's like. Can we like, if you have an interface by the same v1, you don't have you don't extend any interface, then in v2 you want to extend another interface, we've DIMMs. Can you safely do that and I'm not sure the answer is yesterday versus we have an abstract base type. Of course you can add another interface to it. Right, like so there's a few things where it's not clear to me that then we get all the benefits as an abstract based I would give you.

E

I'm just curious, the the two examples you cited. Are you shipping new interfaces with dims? Are you adding dims to existing interfaces, I believe the.

D

Verse won't be added to a distant interface and the second one is a brand universe.

A

And that one I have almost no concerns about.

D

Yes, so one was an authentication featuring a peanut horde they'll be added a tattoo.

D

I'll try to find it. Oh yeah.

F

Like I'm not saying, we should never evolve our positions from our guidelines, but I think that we need to have a strong reason to evolve our positions from the guidelines, and it says essentially, if base class works, use a base class and use an interface if it ends enable or otherwise describes partial functionality instead of core functionality.

A

Yeah I mean I will say, though, like even even if DIMMs would work just fine, an abstract base type will always give you more control right because you can add new methods that are non virtual, which you can't redo with an interface. Everything is always virtual. All right, so there's a show me some boss, infidelity. If you go down that path, I mean to me. It's kind.

E

A

There's a I mean three: it the way, interfaces that seems to be an aesthetic argument in favor of them and I think under design guidelines. We generally try to avoid aesthetic arguments. It's it's more like what features.

D

A

You need and wouldn't.

D

A

Based I work for you for.

D

Example, the property, the property by that interface. If you told me that was gonna, be a base, class I would say to know, because we actually want to implement that thing on on the tips that we currently have ready, but doesn't.

A

D

Allocation all right.

A

D

Makes sense to me.

A

D

For the listeners, like those, we do have classes today and casually I'm hoping we could actually implement this interface. Our derive this base class from I'm, hoping that that works. We don't, we don't break consumers, n5o and so I'm fine having the base pass as like as being the planet record for the listener and the connection was it factory and then seeing it that works.

D

If not, then we can discuss like the fact that we left make a break and change, and it's beyond that.

B

Okay, um why don't we tentatively agreed to use base classes, we'll kind of prototype that a little bit and make sure it still works and specifically, like not.

D

For everything for I think for connection to make sense connection and listen there and the factory not for the property bag, yeah.

G

It's also worth noting that for dims you don't have to make the methods virtual. You can put the sealed keyword on them in it prevents them from being virtual awesome.

D

Other question I think you're saying like: could you implement other interfaces with Tim's? Was the question?

D

Could I have like a dispose default interface method that implements AI disposable? Is that the question.

A

The concern there is I mean.

D

A

Tried this with the rotary read-only collections, but where the idea was that we would make the existing interfaces like, let's say, I a collection derived from I would only collection and effectively implement the I, read only collection api's so that it doesn't break explicit interface implementations, but that causes problems because now you're implementing a base, interface method on a more derived one, which means, if you have a shared diamond, you no longer have a best equal implementation and the runtime with one exception. It's kind of the problem that we had.

A

Kind of the concern that we have that you know you might end up in a world where, at one time, it's ambiguous, which one to call.

G

Isn't that only the case for existing code and that that's not been recompiled and only if both are explicitly err, our implicitly implemented, if ones implicit in one's xsplit licit, then it will never conflict.

A

Right but like that's the only speaking, not something you can rely on right, so this other poly.

G

Well, if we make it explicit, then we can rely on it because then both will never be implicit.

A

Not sure I follow, so if you, if you have a, if you have an interface right and then you add another interface to it in order to add another interface to it, you have to implement all methods from the base right.

G

And if you implement them explicitly on the interface, make them explicit, implementations then they'll exist and they will never conflict with anything that the users already defined. It's.

A

Not about what the user defines it's when you have a shared Diamont, the runtime has to define which the wisdom to call right.

G

Which isn't a problem if you've got one, that's explicit and one that's implicit, I'm, pretty sure the runtime handles that. But you.

A

Know why don't you I think that's still the same problem, because I mean the basically the X we have a base. Then you have two interfaces that are both derived from the base. All want to add this interface to its who both have an implementation for that interface, and now you have another type implementing both interfaces. Now you have two dims. Both are explicit, which one would you want? Em call.

G

Okay yeah, in that case, if the user didn't also define one then, and it would conflict right.

A

And that's kind of the thing with you would only interfaces. It might not be that kind of everywhere. We are right that it's it's possible, that it doesn't want a general problem right because it's like it depends on various things. It was just an example where it kind of freaks me out a bit that you have these.

A

What sounds like simple rules, but when you actually compose larger systems with them, you can end up in a world where stuff doesn't fly right, and so the question is: can we get more crisp around what these rules are, and can we basically distill them down to local rules? So we can reason about our versioning bubbles, because if you, if you tell me nobody can have a share diamonds like well, how do I reason about this when I looked just at one interface right, it's thick! That's the problem with glasses.

A

You don't have their problem, because what you kind of move, heritance would kind of given by design there.

D

So, as far as the interfaces versus classes, I, don't think it'll be a problem in Kestrel to implement the listener classes it might require, like extra listener, objects that basically redirect one redirects to another or something more or less but it'll, be fine.

D

Awesome stamp approval all right.

A

What we say to those connection listener and connection factory right, yep.

B

A factory the connection in the listener.

A

Zippy, the property back should be interest.

E

B

B

Okay, ready to move on to the next.

B

Affection property collection should not use the name, collection, I. Think that's completely reasonable.

B

Do we have suggestions on what it might use instead of collection.

F

Hey I'm, not a community in my at the bottom, where I was trying to write down all the things everybody was talking about for yet property, apparently I went with connection property set, which less implies a mathematical set or the set type collection and more just there's some stuff here. I think I certainly see us being a lot fuzzier on the word set than on the word collection.

A

H

Call the bag yeah.

I

I mean yeah legs, all interfaces.

A

What interfaces I mean I was actually seeing a connection property, it's just the plural features.

D

I connection features I, like that, your properties, well, it.

B

Is I connection properties? Okay, what we're talking about is the class that sort of default implements by connection properties for user, whose users, who just want something easy so.

D

People who are lazy want nice and nice api's default next in properties, the properties defalcation piranha.

E

Containing a default is just a reasonable like dictionary based implementation right, yes, yeah I, don't like people in.

D

The name if we go.

E

H

D

For everything other way.

A

If it doesn't have the I it's killing, the default right I mean.

B

A

The color making.

B

D

We just call it condition: properties like the default service. Yeah dictionary based connection property implementation improvement. Is it shaded to I mean it's super helpful to save summer.

F

D

B

F

Thing that people are going to realistically be implementing, or is it a thing that like 3:00 class, is ever gonna do and we don't need to worry about it again. I.

D

Mean so this whole feature set is for is pretty low level, so a number of people isn't the concern per se like for testing its super useful to create a mock connection and have a mock property set to add some up to it. So you can test your fake memory version.

B

This so so what I found when implementing this is that you can usually find some way to sneak the allocation in for the eye connection, properties, interface somewhere else. So if we're doing something, efficient we'd probably do that. But if you just want something super easy, you have a factory you want to call connect.

B

This makes it much easier to reason about you know you just create a collection. You add your properties that you want to that, and you call connect you don't have to create a new type or anything like that. Yeah.

D

Okay, wait hold on this thing is passed. Interconnect yeah forget passed into everything, so next.

E

To my she's, the.

B

Same type, it's.

D

The same features as we do to pass: okay, yeah.

G

H

D

Connection screen in it's an argument: yeah.

A

Yeah I mean connection property back connection, property is said, seems both fine Nexen property.

I

A

Might be problematic, but anything short seems to use in the world.

D

I thought: there's consensus on connection properties, less like it's for.

A

An interface I think it's fine for a class, a plural name is kind of weak. It.

B

Is jetty cloth.

A

I'd, rather have shed.

B

A

B

A

Think bad kind of carries the connotation. We won right. Here's a bunch of random stuff! You can.

F

In no description of unique or no assumption of uniqueness or linearity or well.

A

F

A

F

You it's stuff like it's, not it's not it's true.

A

I

Not a structured.

F

Connection or a structured collection, it's it's a stuff stuff goes in stuff, comes out, yeah.

C

C

A

Connection properties and would be I like.

I

A

B

Move forward with bag, unless you doin has a strong objection. Okay, we.

A

Have AI connection property bag and connection property back I can.

D

As connection properties.

B

So we're renaming.

A

The interface as well a connection property bug I mean it seems certainly easier if you deport implementation is the thing with all the I sure.

A

And that's what most people will try to instantiate it.

B

Okay, that's fine.

I

I

B

My connection stream should not use the name stream.

B

We have both a stream and a duplex pipe on this. On this interface, your stream is probably not the best name to begin with.

C

To swallow up a latte.

A

I

B

B

A

So connection stream is really thing that has as a stream or a pipe and a connection is basically.

B

Yeah, so it's lit this way because of anticipation of future quick api's. Basically, a quick stream does not have an end point on it, so we could, in theory, merge these two interfaces. It would just look kind of weird and quick right.

F

And I think last time we discussed doing that because since that end points are already nullable like there's already the notion that they don't have to exist yeah. My primary worry.

B

With that is like, as is it important to draw a distinction between the endpoint, not existing or more, of a usage problem like you're asking for an endpoint when you shouldn't be right, I mean.

E

B

It was a fundamental.

E

Difference between those like promote from what I do with it perspective, what do I do differently in those two cases when it's Nolan both.

A

I guess one would indicate a buckin you're already asking.

B

Yeah, so with quick, eastream itself does not have an end point on it. The overall connection, which can have mini streams, is what has the end points.

D

But I know it's not necessarily important to design this interface, so maybe I'm digressing, but could the endpoints not be copied from the connection to the quick stream.

B

It could be it's totally one possibility.

D

But I do think they should be merged and the fact that they're all built, like so I, agree with Jeremy on this I'm. Sorry, if I interrupted your questions, even no.

E

Not you're agreeing with these so I'm.

B

And I'm fine merging these I, don't think it will be too bad if they're, no or if they just copy whatever came from the connection, it.

A

Seems like I mean honestly like if you look at the VCR did the general way we do. All of these things is that we eventually may have an enum or something on the connection stream. That says, if it's a connection or not, and we just documented, if it's returning quick connection, then properties don't make sense, because if you have too many different API shapes, then usually the end problem. That this creates is that you have now to expose more api's to actually contain the type as you're passing this.

A

It's supposed to everybody can exchange the you know, I connection stream or the connection stream and then maybe add new concepts. You don't have to change a bunch of api's to take the other thing.

A

Otherwise, you, if you have different type shapes, then UN looks having to effectively every time you add a concept.

B

F

Also talked about having this one again be an abstract class, because the there's the note here that interface implementers should make the stream and pipe be equivalent and with the base class that could be the constructor takes either the stream or the pipe, and it does the automatic wrapping of the other one yeah yeah.

B

I don't know if we would do that because there's some benefit in late, constructing them, but probably some protected method to create them. Sure.

A

The point still.

I

Stands you know and effectively.

A

Nfs, the other.

B

So interesting thing here is like with quick. You have a connection and then, on that connection you open multiple streams. So it'll look a little weird. If we have this like I multiplexed connection for quick that returns, something called I connection, yeah.

D

So we had the same debate in cash flow right for for doing a htv3 and we actually introduced a new class in the hierarchy that didn't have the stream in the pipe, but just had like the abort on other information, but they don't have the actual there I think I think we called it. Batum is the bad name, basic base connection context.

D

So I think we don't want to actually, but the stream and the pipe on the thing that isn't the connection or we're gonna have to have something in a hierarchy that doesn't have a stream in the pipe.

D

Yeah we we, we spent a lot of time debating this this this end-to-end in Castro, and we ended up with this with a different type hierarchy about.

E

Calling David, are you saying that the only thing that would exist on AI connection stream is connection properties and abort.

D

Oh yeah, abort properties and.

F

Basically, yeah undisclosed, yeah, and so then you are supposed to blind cast or, as casts to get to the thing where you can consume it. So we have two different type hierarchies want to.

D

Get E and so nothing ever returns the AI connection. By itself, you either get a connection that has that so you didn't get both best connection or a connection, and if you want to do sharing between those two you can take in the base. So you can create types that like ask that pass through the same base type because those derived from a bit of base connection context but to actually get data from the from the actual connection. You have to either have a multiplex one or connection yeah.

D

So to put things like a little bit more concretely, transports return either a multiplex connection, context or a regular connection context. Both return, the base, connection, context and kestrel right or both derive from sorry neurotics and petrol uses the base class for lifetime stuff so for managing connection lifetime shutdown. Things like that, if you're not a server.

D

It becomes less useful, but it has helped avoid a bit of code, duplication and Castrol for sure.

B

So another option of this emo. If you want to search for future quick amendments, you'll see kind of where I mocked that what a quick API might look like in.

A

This issue yeah.

D

And I think one of the one of the problems was that, once you have a quick connection, you end up accepting more more connections from that thing or more streams, and the streams are the things that can actually read every data.

B

D

We have it look.

B

At it, if you look at the the interfaces, I defined there, the I multiplex connection and the factories and listener one option that I came up with to kind of merge, quick and TCP into one set of interfaces would just be to have this I multiplex connection be the I connection, so with TCP you would be able to call open stream or accept stream once and it would start throwing after that, whereas with quick, you could just call open stream as many times as you want.

I

Actually I, like.

D

So the downside with this is how in maybe there's a way to do this, but how do you associate the streams on the same connection like? Is that some sort of property that we're looking at.

B

Should be perfect in each eye connection stream, it can have a property saying what connection it came from I guess. One downside of this is that someone might completely reasonably call open stream twice, expecting it to work, and it would just wouldn't sometimes like if you're on TCP wait.

B

I, don't I, don't know if that is necessarily a terrible thing, because quick and TCP are so different that you really can't reasonably write code that works for both at the same time, so I think you would know from the get-go like either you only have one stream per connection, or you have many.

D

So we had the same debate and cached. We spent a lot of time trying to figure out how to be white, a blonde type hierarchy so to reduce all the duplication. But then it ended up the code and it being so different that we ended up with two Iraqis because of the same reason, corey mentions I mean.

F

To to actually throw an interface at this, if the only thing that you really need the two things to unify for is for an abort method, then maybe that's like I aboard able connection, and then you put it on multiplex connection and connection and multiplex connection is just connection factory.

A

Yeah but I think that might not work right because the the point is not that you want to call the abort method. But the point is that if the base you just accept, is you don't actually open connections? You up, you just accept any connect as that around.

F

A

F

The case because, as it is.

I

F

You're doing dynamic up and down casting and if the it sounded like. The only thing that you were holding on to the base class for was to call abort.

D

So yeah one quick question Cory with the I multiplex connection interface. Is that supposed to also represent like a normal TCP connection? Is that what you're saying and then you just call, except stream month, yeah.

B

So that's that's one option. Is you take this multiplex connection interface? We say this is the connection interface and with TCP. You can only open it once so.

D

So here's my hot Tec having done this like lifetime logic and kestrel I, don't think we need to merge the multiplex connection and connection interfaces. I think we just need to make sure the stream and the like. So basically, the I connections, stream that you get from a night multi-plex connection is the same type that you get for like a normal ccp connection.

D

As long as the connections themselves are the same like it's not that hard that create a type that has like two constructors, one that takes like a multi-plex connection and one that takes a normal connection and then like, let's, you abort them both like and use that in our like lifetime management code, like there's not that many servers that would need to do that, but.

A

Yeah, but it kinda has the problem you're pointing out right. Basically, it order to avoid casting all the time you either have to have more loads or have to do some addresses invitation to hold on to.

D

I would argue that I mean this table is saying this is pretty low-level feature to begin with, like there aren't a ton of devs that are going to be programming against this, but I'm like.

I

D

Bun, those like those people are not going to be like even less they're gonna be managing the lifetime to both multiplex and regular connections. Yeah.

B

I think casting sounds really bad to me. I I, don't like that. Ideally, these features to just work with the things that you just have to know, but so.

D

B

D

Saying at all, I'm not saying that you would as as anywhere right. So what I'm saying is that you would have an AI connection thing that represents both say like a quick stream and a TCP connection, you don't really need to have shared interfaces between quick and TCP at the transporter listener later and there wouldn't be any casting like. You would just have some sort of like type that this represents a connection lifetime that like could handle either a multiplexed or regular connection. But.

A

What does they know and say I do the thing that Corey proposed because I mean I can his model. You still have the same benefit right. You don't.

D

Have to do it makes the API ugly immediately ugly or where the case is uncommon, TCP, yeah, exactly so to make multiplex better like now, TCP you have this extra kind of step you need to take, they call accept stream. You can only call it once if you can't call it open stream and it's kind of it's not the way you would design it if it weren't for multi-plug. So just add some friction. So basically,.

A

For TCP I guess the idea would be. You only have a stream property that good one right.

D

Thank you and I want to say in TCP, but but it's basically at any transport, so the name pipe TCP like WebSockets and those are good article examples like quinton TCP, but obviously there are a lot of analogues to a streaming transport.

B

Now we all know why I split my connection and my connection stream.

B

Do we still want to merge those interfaces? Yes, yeah again.

F

The multiplex can still just copy the values up onto the connection I'm a bit confused as what was being proposed.

D

It's like okay, so the thing that we're gonna get back the connection object is gonna. Have the thing that reads: data right! That's gonna, be this year: primitive right, yeah,.

B

So when you open a you'll open, a multiplex connection for quick and then you'll open, just a regular eye connection to get each stream right so that.

D

Means kind of weird.

B

D

It makes sense and and then that thing will have like a board, an endpoint and all the nine point, abort and the pipe in the stream and the property right well,.

B

The board votes. Yes, if we decide to make an important API, that's just that's later.

B

F

Isn't to not okay.

B

Yeah we have a very tcp centric view of things right now when we started using quick, we realized. Oh, you can't just abort, you have to have an error code. So do we pretend that zero is what happens when you call a board? Do we expose a way to set that elsewhere? It seems like just having an abort parameterless. Api is kind of it's very TCP focused.

D

Not a focus so I think we need to get get over the fact that this thing cannot be pure. If it was pure, it wouldn't be. Maybe we couldn't do it all, so we have to make some assumptions that some things will be unclean and then for more specific things that are transport specific will need to have different features because so, for example, like we have so in its Banette core HP context has a board right.

D

If you one has a board and on HB 2, you can pass in a code for the abort and guess what that it doesn't work, but for the and we default to someone to some argument, but that isn't a reason to not have a portal. It just means. If you want to do the more specific thing for HB 2, you have to basically try to see what what protocol you're on and then do. The protocol specific, behavior I think we need to figure out or.

B

Basically, it essentially means that, if you're using quick when you create a quick connection, you'll have to tell us what to use. If you just call a board sure.

D

I see I mean that's sure as a default right, yep yeah, but is not.

B

In my opinion, I was taken aback when I saw that quick did not have some default like I. Don't know why I'm aborting, I'm, just aborting, yeah and and that.

I

Kind of informed.

B

My decision here, if like oh maybe we shouldn't, have an abort, be first class because it's different in other places, yeah.

D

That that's a fair first hit, but I would assume that we could do something clever or passing a default or assume something or throw I. Think a board is common enough and these networking AP items that you want something like that and in our experience in like castro for the last couple years at least and quick.

E

Doesn't have any kind of something went wrong. I have no idea, one generic Eric I know.

D

Yeah there's a big discussion alias and they don't feel tempted to change it right now. We're technically violating the spec in Castro ridiculous.

B

It requires close integration between quick and the protocol on top of it like HTTP 3 has defaults, but quick does not.

E

B

E

Just gonna end up like she was saying or and America that doesn't say this one I think.

D

You're right and that's been trying to today, I can link you to the conversation.

E

I'll be enhanced or calm or whatever. That was from a tree.

F

I'm, a teapot all right so.

B

We agreed to merge a connection stream and a connection putting keeping the knowable endpoints on a connection. Everything else the same.

A

B

Whatever little.

A

Bit air losses like visualizing the hierarchy because Billy.

A

Actually tried to sketch it here and editor, really quick, and then we can see whether you as we was.

I

A

Because it seemed like for the the multiplexed one.

A

I

I think you're sharing just a window of course I'm just staring at window. Oh it's.

A

I

A

I

F

A

And so I can't share the home. Monitor, saw I guess what I should have done is clip the monitor, but that doesn't work in teams. Yeah.

I

A

Here we go well, the people in the stream can't see it, but I'm not going to mess with OBS.

A

I hope, roughly what you said earlier: the merge connection and connection stream, but then we said we want to differentiate between the multiplex and the normal two plex, one which kind of means to me that stream and pipe thing should be properties in the one case, and then there should be open stream and except stream. In the other case, I'd have no idea what this guy will turn, but we could probably say this grabber transit connection, I guess.

B

Basically, just kind of remixes the hierarchy I have already yeah we're not getting anything new from it. I think that the plan was to have an AI connection that has the stream and pipe on it as well. Okay, so, and.

F

Then get rid of the I.

A

It's so basically, maybe I don't understand that so basically, you're saying you want to merge these two. This is basically what we currently have right so like what do you? What does merging.

B

A

Two endpoints.

B

Under connection Stream and then a multiplexed connection will have instead of returning well it'll return a connection stream, essentially yeah. So it's correct how it.

F

B

F

Interface eye connection stream becomes possibly abstract class connection.

A

But it'll be clear, then that means this guy doesn't have properties, and then you have duplication on whatever these guys. Alright, oh I'm multiplex.

B

Connection would have properties itself as well.

A

Yeah yeah, so that means basically these three here or no, that's, basically, the thing you get from the other.

B

A

B

And remote endpoint would either be dull or just return a copy from the multiplex connection. The connection properties might actually be different. They're, probably not gonna, be cloned. Yes,.

D

Different, yes, okay sure has a similar hurry. It has a most best connection based connection context and connection context.

D

But but that's more of a type hierarchy: there's no interface inheritance here, our inheritance here, deities and then.

F

Did the connection class when, when multiplex connections get added, will it gain a property of what the multiplex connection, parent or source or whatever the name of the property is because you said some way to tie them back to each other, I mean.

B

It could I I'm not convinced that there's a need to do that, but we haven't probably do that by having a connection property to expose.

I

That hierarchy.

F

B

F

I guess that would be you wish the multiplex connection out of your connection properties, a pretty bag, so yeah.

D

So it's okay! So can you change the interface to a base class.

B

Worse yeah, so I can.

F

Exit strange you.

B

Can just assume in space class, okay.

B

F

B

F

Then it's just connection because it's not.

I

F

Stream, it's, it is a stream factory, but it's also a pipe Factory in a and.

C

D

C

No, you don't really affect.

D

And we don't want to have something in between both like a base class that has the shared whatever and.

A

So that's my other question. So ever the board go on because it's.

D

A

If you know how it's often I have to add, you know a public void aboard here, and then we duplicate this as well. At that point, I would say you really really want something in between do not have to duplicate this thing. Everything, yes,.

B

This thing is with quick, you can abort the connection and you can abort individual streams, so it it actually kind of works to have it on both. There.

A

But I mean if you look at the thing that I'm proposing here, you have the same thing right, because the multiplex connection itself is a connection and the.

I

A

Return is also a connection, but you, but you only have one class hierarchy where you define all this stuff, you don't have to basically add the same member in two places.

D

I think it's nice. It certainly makes some things easier for Custer all, but I'm okay, either way, I think it is pretty rare. The people who will want to deal with both the connection in a multiplex connection and the same use of those generics Kestrel does the that would help.

D

Casual mixes, both generics and and the inherited turkey to do co-chairing for.

F

These two things I mean it could be abstract class connection. Then a concrete, non, sealed, probably class of simple connection in a class of multiplex connection, but seeing simple pop up all over the place would be a little weird I. Think.

A

It's just what I tried to bottle here is the concept not the naming like.

I

A

It's the it's basically, this.

I

A

Of a common base, you have both derivatives and then the one here is basically it's chaining, its of inning all.

F

Right and then I guess those could be going back to words. You've used before that. It's connection, context, connection and multiplex connection. I mean it is weird that connection multiplex connection will be a connection context instead of a connection, but.

D

Another weird one is equal, I single connection, it'd, be weird that you know. Nice single connection is an AI connection, but we kind of you.

F

Know correlate.

D

F

Connectors so connection single connection and.

F

Multiplex connection and the multiplex connection can open single connections. I mean it's not quite right, but.

B

So if we have this distinction between a single and a multiplex connection now, then we should put the endpoints.

B

Outside of the I single connection, no.

D

B

Don't care that much, but it looks.

D

B

Kind of changing things a little bit, but not really gaining anything from the proposal here, I'll tell.

D

You what like local and remote point our access so commonly that it feels I mean so you can put it on the base hierarchy, but I feel like it needs to be on the base and explosives properties. Okay, I wouldn't want someone doing like TCP to have to fish out those two properties from the prop from the bag. Is there so commonly accessed connected in the bag?.

B

I see I'm just wondering what this is getting us versus what is in the proposal, it's kind of just a little remix, but I'm not seeing benefit.

A

Maybe maybe I'm missing the point, but to me the value of this is my logical understanding of what you described is this like, because this to me model is the actual relationship you have versus the other. One doesn't seem to do that as a board would go on yeah.

F

The basically he'd ever made.

A

So so again, I'm not I'm, not married to the names I'm just married to the I guess D we had a stuff live and who is deriving from whom right? That's just to me the the model, the thing you described right.

F

But I mean I, don't know that whether it be that I connection is a base class of these two classes or it's an interface that two classes adhere to I, don't know that it's actually like the only time that we would really need the the extra type here for unification is, if somebody's really expected to reasonably work on both otherwise having the two api's that just look the same because they represent the same concept, isn't terrible yeah.

B

F

Agree so I don't know that we need connection I, think it's just sorry. We need a connection, but single connection and connection. Merge and multiplex connection is just its own thing. That has a bunch of things that look the same.

D

Not having any sharing like I guess, your fourths.

F

Duplicating character, parity, it's just a question that, like yeah.

D

F

Where I came back to who's who's,.

B

Gonna use polymorphism.

F

Scenario, or is it just like oh yeah? Well, both of them are things that can be aborted, but the abort means a different thing on the individual stream level versus the multiplex level and yeah. They both have end points, but they could mean different concept like if they're just.

I

F

Same word, but they don't necessarily mean the same thing. The interfaces is really just duck typing. At that point,.

D

Yeah, so in touch where we do use abort polymorphically for like ungraceful shutdown of both like quick multiplex connections and like regular single connections and I.

F

D

Yeah, exactly and graceful shutdown do that matter.

D

The the use cases there I would say it's not going to be the most common use case that people like writing middleware, but.

H

Over common, we should have.

I

D

Yeah, even in middleware, if you want to like traits like all new connections, whether it's like you know, quick connections or TCP connections like you can see it. If you just want to, like you know, log the endpoints right.

F

And so, if there's a scenario than carving out the whatever reasonable name for a base class or saying you know what these four members are the interface and we have the two classes that do it. That is reasonable, like essentially, nothing would return the interface there just for, if you want to have.

D

F

Polymorphic unification at your local level, if.

D

You wrote middleware it's the log, this for all connections whose price they like hook it up for your multiplex transport and your regular transport. But you could have like a shared function that, like does the work yeah.

B

I mean there aren't too many times where you'd want to abort either a stream or a connection was quickly like. There are two very different operations right, but but we have shared code.

D

Only we literally started with two different class car keys and the code duplication like looks small at first, and then it got insane, and then we said: okay, let's, let's bring it back in by adding the base and then having having a ton of shared code and actually helped a lot with this yeah.

D

We didn't do this in the abstraction and we would do this ourselves more or less in Kestrel, like we'd, have to like build something on top of it. That kind of had this hierarchy, but anyway, we're.

F

Out allocating.

D

Per connection yep, is there a basic message designer faces.

F

D

It could be a struct.

F

And it just holds both of them, and it's just two pointers instead of one everywhere, but.

D

B

D

Like I, like the poster having.

B

Like a base eye connection, that has the properties and a board and the endpoints on it, if we want to try that out,.

D

I have to go soon that me.

D

Was there any other words a here at all out while we're gone, yeah.

F

It's gonna be super easy right, so I'm part of like so you know on this one. Now you have the is connection as using the words emo has here. Not as that final words is connection, a class because there is actually reasonable code sharing between it and single connection and multiplayer between single connection and multiplex connection on these, or is this one actually an interface that single connection, which gets renamed a connection and multiplex connection both adhere to of they both have an endpoint they're aboard able and they have some properties.

F

But if you wanted to do anything interesting with them, like you know, look at data now. You need to actually know what kind of thing you're dealing with.

B

I think the primary thing we'd be enabling is maybe some form of logging and then just maybe like a generic abort feature. So.

B

I, don't think it would be bad to have both of these implement an interface.

F

Yeah so I mean like it either ways. Fine I mean like it's so I guess really it's. These are the three things that will exist and a multiplex connection and the single connection are probably both classes. The other thing is another thing, and it's either a class that we figure out a good level name for it or it's a interface and, and they both adhere to it. I, in this case, I think that this would be an interface describing interface enos, which is different than the where the connection is like.

F

Well, you have a lot of like extreme in the pipe need to interact in this certain way, which really says that you want to sort of control the implementation on it right.

D

I'm fine with this for what it's worth, we do use just base class and kestrel, but I think making it interface. It's fine yeah.

F

Like I, don't have a strong preference one way or the other I mean the general. The general guideline that we have.

I

F

Boils down to with a lot of hyperbole if a base class works use, it.

D

Can somebody's decide it's a base. Class Ben I mean.

A

Okay, that would be my that's why we just label it as a base class. The only way.

I

A

Jeremy's model was that you have fewer names right. You can you basically the way we're calling one I connection, 11 connection, true.

D

A

Really, the only minor section.

D

A

D

A

D

Base and Jeremy, please, we merely come in.

F

J

Discussion is going in the right direction.

B

All right, why don't we try with the base class? You can revisit interface if it's needed for some time so.

F

I would I would say that the thing currently labeled a single connection using the rule of give the best name to the thing people are going to work with the most. It should be connection and then.

I

F

Now we need connection base as as a suffix, because we can't think of any other suffix like I would say personally, I would say that means that we considered things and there there was no better name and and that's it and most people aren't going to work with it at the base level anyway, but I'll, let Christophe explain his grimace and he can propose any.

J

Question about maybe I missed it. Open stream opens stream for for sending data and except is basically for receiving. Is that correct? Well, either one can be bi-directional, but this is essentially.

B

Socket connect and socket dot, except but on a multiplexed connection, I see.

F

Mm-Hmm all right, so unless somebody has a better name for the unified type I think we have abstract class connection base, probably abstract class connection, because you'll need the open, pipe and open stream protected methods on it and then the multiplex connection, which will do whatever it does whenever we get around to having multiplex connections. If.

B

It just to be clear: we're not proposing the multiplexed API as part of this review right learning for.

F

It we're just making sure that we're not boxing ourselves out of the ability to make it later exactly.

A

All right so then, for clarity. This is what we're proposing right. Is this the name? Still you see I think you said nice.

B

Connection base and then single connection becomes just connection.

D

I was thinking a baby, better names for connection base. It's hard like a connection better than about really a connection lifetime yeah right.

F

Anyway, anybody who can come up with a replacement name.

J

I'll tell you what we used on in Midori, which was initially when I saw it. I was like geez. This is weird over time. I'm kind of and I was fighting it and over time, I kind of had warming up. We we used their prefix a.

J

It's kind of interesting because it also it means both abstract and also. You know a word in the English language, that kind of describes actually what it is. Yeah.

F

When I was in office, we had the same convention if it was a if it essentially if the class was being declared with the abstract, modifier I got an a, which is. This is probably not the thing you're looking for, or at least it got, we used an a if if it was regularly going to be. This is not the class you're looking for of its here as part of a making a tree work, but you should really never accept one of these or.

A

F

A certain n.

A

As well when the next letter is e, U or something.

J

Sir to you, because.

A

In English you, you say, you know, you would say they user context.

F

No sorry, you is the weirdest one: hey elephant versus an elephant.

J

Yes, yes, when, when it starts yeah, that would be a bit of a problem, but so one one thing about the bass: suffix are we, our users ever gonna be see like? Is it gonna, be something returned from api's the election base?

J

I think this is the main problem with bass. Bass makes sense if you implement it. What only makes sense kind of doesn't make sense when users, you know, get this tied back return from something like bass. What the heck yeah.

F

It sounded like to me, and so we can see if Steven and kori agree with this assertion, that the only people who would use the base class are complicated, hosts like Kestrel that deal with a bunch of different things or code that simply wants to log a new connection was made, and so they they say great I, don't care if it was multiplex or simplex simplex.

F

Actually, the wrong word, but I just want to see a thing happened, and here it's two endpoints and and so it's it's so super minority case, it's never a return type. Probably we would never have anything public that took it as a parameter. It's just user code and infrastructure yeah. It's.

D

Like it's a parameter like your own, like project, might yeah, but.

J

If it's, if it's basically not a in public surface area of call, the users will be in it, you know calling and I think it's fine, okay.

D

Couple things: okay, not sure, I've been properly introduced for everyone, but it's Stephan, so you can distinguish me from Steven tube and no worries and the secondly David Fowler is messaging me from outside the meeting to point out that we want this to work in that standard. Oh I, don't think, there's anything stopping us since we've moved away from dims, but for signal are we want to use these types?

D

So if we can make it like a you know, nougat, and that standard Oh compatible that'd be great.

A

That's a great document against them. um Yes, but I mean other than that. I don't see a problem with their like unless you want them to be implemented on some existing type. That should.

A

I think your name is not Stefan, because you're younger than Steven, your name is no Steven base. I think that's what we agreed.

A

That's what I'm.

I

A

It's totally better than a Steven right.

D

B

We ready to move on to the next. Yes, I'm.

A

Just copy and paste so basically what we now see on screen. We agree on right. So let me just copy and paste this as the proposal.

B

The only thing that would be added would be some protected methods on connection base. Sorry on connection to create the pipe and stream right.

F

Like because connection base will need something protected, sorry, it will need something virtual in order to make a board work. So it's either public virtual void abort because it has no parameters, so it doesn't need template or it gets protected, void. Abort core I think.

B

Connection and multiplex connection here would both be abstract as well. Yeah.

F

And so connection is, would be abstract class connection with protected virtual stream, open stream and protected virtual I duplex pipe open, pipe or because then the class is going to manage, making the wrapper over the whichever one got called first for the one that gets called second, but I sell that right, close enough. Yes,.

A

B

Not the street not the base, maybe.

A

Just can we have right now because IV I think, like nobody, I can mechanics I think the hard part was getting the shape. Yes,.

F

There are some there's some details to be filled in, but I think we got through the hard part. So.

D

Specifically about a port I think there's some hard part left like the one thing that came up the last API review about this is that in cash from at least, we want a way to initiate a graceless shutdown which would mean like draining the output buffer yeah.

D

If that doesn't succeed within, say a certain amount or rate limit, we would want to be able to abort. Basically, the connection after we decided to send it in you know it sees TCP pilots, sure you're talking about.

B

Adding a closed, gracefully async.

D

Right is a case, those if we want to abort, like not at least you know, falling good design guide well right, so.

F

I mean that sounds sort of like connection basis, I disposable, an ia sink disposable and then you would a sink, dispose with a timeout and then move into abort. Of course, now you're in a troublesome situation of you've already started disposed, and then you called abort so.

B

D

When I was saying, that's where less of casual does today.

F

So yeah, so maybe you want abort and I, don't know if it's Boyd, abort or past board a sink or yeah. It would be the task. Of course you know the.

I

New guidelines are.

F

You should have synchronous API to mirror your asynchronous API and in the right. So maybe you want I, don't know if that means you want graceful clothes which is different than disposed of, but that's the thing to think about something like an almost if you're gonna populate you know, socket names.

B

There are three operations here that I can think of. Abort is like gracefully aboard make sure the other side knows send whatever error codes. You need and then there's closed gracefully, which is also like an I/o operation and then there's disposed, which is don't, do anything just stop and release any resources, so the other side might timeout something like that. Is that kind of what you're thinking Stefan yeah.

D

I think you know making it AI disposable like after you've, completely gracefully, shut things down or abort is perfectly fine, but yeah. We should have a way to do both before it is supposedly.

D

Another question: like Jeremy: you pointed out that, like the guidelines you're not to have synchronous API for all async, api's I know we're doing that for like AI, async, disposable and disposable here for like HP client scenarios, but I, don't think we're doing that for anything else and I hope not like for like yeah. It.

B

Was a conscious chemistry.

D

Yeah I hope for this kind of low-level networking stuff. We just it's new code like it shouldn't be, you shouldn't be blocking I'm totally.

F

I mean right so you you say that, but you know there are people who, like you're, just writing a simple console. App. You don't care about like you're gonna. Do everything in order you don't care about all the weights. You just say do this and when you're finished removing on the next thing, like low-level code, especially needs to be hyper flexible here, because you don't know what your caller wants to do. That's why we have the guideline of you should offer synchronous to mirror your asynchronous if.

D

I would argue if you're, in a case where it was acceptable, to call like a synchronous version of accept, then calling that result wouldn't be that much worse, like you're still blocking a thread for a very long time. I mean.

J

I guess that's a big difference so.

F

J

F

Is special because you're now literally waiting on work well.

J

So I think there's a very big difference between calling that result and and these IP ice or this hypothetical sink IP is especially that they aren't sockets sockets actually have a native synchronous. Api is in operating systems, and if you call that result there's this soft deadlock issue that you know we we wrote lots of articles about and yeah so and, and you know, natively implemented. Synchronous api's would actually not be subject to the sub deadlock. Yes, they are inefficient from you know: machine resource consumption perspective, but at least they don't dead, look I'm.

D

Not denying that there is the difference between, like you know, sync api's down the OS to something that would need to be completed by another thread. That would you know potentially do it off dead blocks, but I'm saying if it's acceptable to block on accept you better, not have many threads doing that. So hopefully you have other threads available. The coming.

F

Because right so accept being the state of like I'm now going to sit here and just twiddle, my thumb's, until somebody wants to talk to me like I, could buy an argument that there's very little scenario at all for that being a a synchronous flow but B I'm, going to connect to something and send in this data and then shut down like that seems like a thing that someone could do. That's could end up being doing in code.

F

That's already contracted to be synchronous and they can't change it and now they're doing they sync over or they're doing like sync over a sink over sync over and it's like call the method that represents the flow you're in. So we.

B

F

B

Before bringing this to API review, the that argument was the first one to get called out was like if we want existing users to adopt this instead, like replace their sockets code with this, and what are they going to do for other synchronous api's, um the sort of ultimate decision on the NCL team and ASP teams was, if you're, using this connection, abstraction. We expect you to have an async workflow and if someone calls your sync api's, just throw.

A

Kanessa can I propose early table the a sink sink discussion for another time, because I think like let's get the rest of the proposal, maybe into shape cuz. It seems like there's a lot of controversy around many other areas here. So that's just picked some battles here. The reconnection.

B

A

Kind of a thorn, alright I think once we have the shape for a sink I, think agreeing on whether we do sink and if so what it looks like it's probably easier. Then simply we want to type.

D

Out what we're making versus shut down and abort like, should we add some like placeholder? Yes, I'm, fine,.

B

Having an abort a sink, a it was gracefully facing.

B

Yeah I get c-type.

D

That sounds good nation.

B

Token as well and return value task yeah, it all seems fine to me. You.

A

Lost the screen.

I

You should be back now right, yep, yep, yeah, so I'm over the board. The way.

D

Abort async takes a cancelation took in return to value tasks.

F

I mean: is it value that you expect it to complete synchronously.

B

Potentially yeah for sure I mean some.

F

E

B

Always synchronous for circuits other things. It may not be okay,.

F

All right because abort is just tell the OSR it'll it'll flush a fin packet at its, at whatever rate it feels like doing it and you're done talking to the OS. At that point. Okay,.

D

Basically yeah like if we receive any more packets, you know the lesson back in rst kind of, but.

F

At that point, it's all the OS is problem in your. Your flow was synchronously, tell the OS I'm done and then move yeah.

D

I mean the the.

F

Process could shut down the alright.

F

But then shut down, gracefully or graceful shutdown or whatever you want to call it. Presumably that one is tasks, because the whole intent here is it's going to do. Work, that's not going to complete synchronously right.

D

Like casual this case, specifically like we actually wait for our application buffers to flush, you know when we dispose.

B

Yeah and then there's just the os calls of it'll call shut down and then close the socket. So those are both synchronous, so they could very well complete synchronously for a lot of workflows, good.

F

But that but first you're gonna, your graceful shutdown is gonna, first flush. All your buffers which yes.

B

F

Be synchronous operations, maybe pop earth or bloody, so.

B

The this OS has its own send buffer. So for a lot of workflows, actually, all the writes you do to a socket or synchronous anyway,.

D

Yep exactly it and like even when the process shuts down, the OS will try for a while to flush your buffers for you. If you try to like gracefully, close and stuff okay, just the.

F

Value task, is it solves some problems at the expense of others, and the only time it should really be used is when you expect, in the average case that it will have completed synchronously, essentially.

I

F

Test should call a value task thing: an assert, completed synchronously and if that is false, then you've designed the API wrong.

B

We're intending it here more for optimization thinking that there's no reason you would ever want to cash. One of these tasks, for instance, and.

I

B

Low level enough API that we think people can use it successfully without you know the nasty bits that come with value tasks.

F

You know I mean the worst. The worst possible thing you can do with value tasks is be optimal and have the completion handler, because that's when really all the dangerous bugs a value test come into play, and so it's it's make sure that we're like yeah. The flow that we put for people is, if you expect it to complete. Synchronously and value task is an okay thing to consider, and then we didn't even talk about, though the fact that you can custom handle the async state.

D

What I would say for these networking API is like there's so much buffering. That goes on, like these things, complete synchronous, a lot or complete immediately a lot.

F

Presumably, flush is an operation that you expect, like is. Does that tell make the process wait until all the send buffers have drained nope that.

D

Just makes sure it waits until the buffer gets long enough, at least in casuals kiss, and it's usually synchronous early immediate. That's.

B

Basically, wait for all of the data to be handed off to the OS.

F

I mean okay, so essentially, I mean essentially it's anywhere that the word value task appears.

F

The the question will be, and you expect this to complete synchronously, and certainly my notion of that, like from an application author, my expectation is blush means it's waiting until all the data has actually been put on the foot on the wire, which means that it probably is not a synchronous operation.

F

The funny thing is: there's no way to know that, like typical, socket API, that's fair, I'm, just the level of what the what the but the functionality describes itself is doing, and what the and such an emo there's usually an a an async yeah again.

H

F

To value test and.

D

Then close, oh and.

F

If, if abort doesn't earn a response, then you shouldn't need value tasks at all, because you can just return test that completed tasks, and you still have no allocation, so value tasks would only mean that you're explicitly intending to do the dangerous thing.

D

There ends up being kind of a consistency argument.

F

D

D

We in asp.net we've definitely been leaning on value tasks, there's the ability to pool them, which is nice for performance improvements that I know the board path, obviously not very important right and yeah.

F

I mean the the avoiding allocation you can still get by like you have to do manual work around your task flow and then understand that you can return tasks that completed tasks, but right, yeah,.

D

I'm saying asynchronous, but yeah a state machine.

F

Generator will do that for you, if you're, if your first walkthrough the thing ends up hitting return, it returns tasks that completed tasks with no allocation and.

B

Value tasks, depending on what your options will also have its state machine pooled right, so there's still some benefit there, even if you're, using an async method.

F

But yeah our guideline on that one is avoid the non-generic value task.

J

By the way, is it something that users would actually weight, or it's a more like a async fire-and-forget? You.

B

I

Are three cases I.

B

Outlined you would have a board which is like. Actually you know if you need to do I owe to tell the other side your boarded then do that and then dispose would be distinctive from that by saying, don't do any IO just fail as quickly as possible. Those.

J

Are the other side of.

B

The connection might be timing out or something yeah.

J

I understand that what I'm asking is in typically use would people actually await this abort or they will just hold it and say I, you know like I, don't carry when it actually completes.

B

um We would await this at least four HTTP three I think it's gonna be very specific to what every library is using this well.

F

I mean I, guess the restate Christoph's question: what's the scenario for carrying that a board is done, yes,.

D

Here's in the scenario, the connection you know owns the pipe there wants to make sure the pipe is completed in some background thread by the time aboard AC completes so that you're, like you, know, kestrels connection lifetime and you're after it aborts, like all the connections, notice that all blocks are returned to the memory pool and therefore it's safe to dispose the memory pool. But.

F

That sounds like it's just part of the implementation of the virtual abort: that's not a collar of the abort method. The caller.

D

Wants to know when the task complete, so they can know when to dispose the memory pool.

J

The color will be responsible for disposing the pool in the case of petrol.

F

I mean wait, couldn't you use an.

D

Internal API I mean.

F

If you're, it's, if you're putting the the notion of when you're done, sounds like it belongs to the connection, and so that would just be the last line and a board a sink is I'm done, go put this memory back.

F

Like somehow the connection got the memory pool memory so.

D

Well, what if you're like returning stuff in dispose or whatever like I'm seeing this is like you call away to port async, and then you call this pose. I mean right.

F

So if this, if the scenario is that you need that you should not call dispose while abort still running or that the abort may not or if you call dispose well abort still running, it may not finish, and you want it to finished and that's that's. The scenario is: is not getting to dispose yet I know weight.

D

Pads like an async, abort API right, like for streams yeah, it does that's where we would use this well.

B

Awaiting deport.

J

The reason I was asking is I wonder whether if there is a difference in you know guideline that we should have for those iron and forget methods. Maybe it's okay to use. You know non generic value tasks because it's like in a lot of cases, the user kind of doesn't care. They will not have wait. You don't have many of these problems. We value tasks.

F

But what's the difference at that point from value tasks and task is a value task. If you use the async state machine generator, if it goes async, it returns a task and if it doesn't go async, it returns tasks that complete a task. So it well.

J

What the implementation of the abort method would return value tasks, which is, you know, one less allocation and then, since nobody awaits nobody will you know basically the state machine won't be generated? For you know, if you don't await correct the.

F

State machines generated in the assembly that defines the method. It doesn't know what the caller is gonna. Do it it's going to it to allocate the task object if it goes async no.

J

F

J

The in the method that you know the caller is calling if the method is async and actually awaits, but if nobody ever waits on this one like basically the whole method that calls their board is not not async, you don't have you basically don't have an allocation.

B

Can I propose table English until we have Stephen around because I know he knows all the ins and outs of failure, task issues, sure at least better than me.

F

That's fair I can tell you the flow that he told me to write in the book, which is, if you're, using non generic value tasks. You're Stephen.

D

You know we can wait for him to get blessing or not so there you go.

F

Since Stefan has identified that he is not Stephen he's not granted Stephen authority a little bit more derived before I can.

H

A

To Stephen base I think that will solve the problem.

B

Emo I think to add we would add a pose, gracefully or shut down, gracefully async API here, hopefully.

I

Don't call it like.

B

That yeah I hope.

I

B

Call it something else, too, graceful.

F

Shutdown I mean graceful. Shutdown sounds more like a method named and shut down, gracefully really.

D

F

D

Like a verb and the other, it's like no but yeah.

F

That's what I was thinking.

B

F

Because graceful is an adjective on or I guess is an adverb on shutdown when used as a prefix. So it's still there's still the verb, which is shutdown, methods have to have verbs. They don't have to start yeah.

D

I guess, shut down, suppose a verb and a noun which makes it confusing yeah add the adverb and it sounds like an adjective. But it's really good in the leads.

A

What about complete.

D

Yeah, that's kind of similar ish, the pipes.

B

D

B

It also uses it to complete with an error right. It does yeah. You.

D

Could follow a similar pattern and make you know complete method that takes an exception, and that makes it abortive.

J

So I may contradict myself, but to me close is basically it's a graceful shutdown, yeah.

B

But we can't have closed and disposed doing different things, yeah.

H

J

Time I said that we shouldn't close especially doing different things.

A

Fortunately, I'm pretty sure david keene is not watching, but if I will send him notes that says he will make this pose call a board and close to something else and see how he react.

D

All right, but we want something we do need like a cancelable.

B

Like graceful way that graceful shutdown nation, we can yeah inch the name later if we French something better, he.

A

Said so long I mean like do you expect many people to call this, because that's I'm, also okay, with a shutdown.

H

A

About to say, I mean that don't seems already better than a board all.

B

Right, let's do shutdown.

D

A shake like I'm when I shut down on my computer. That's a lot more graceful than you know when I pulled by word. Usually.

F

One leads to the other.

F

Frequently during a message it says please don't turn off your PC.

A

When you phone symmetry, we could call a board shut up a thing: yep shudder.

B

I

B

Surface looks fine for me. Oh, you have to add a sink disposable. You.

D

B

D

And then okay, so there are a couple options here we talked about complete, like kind of copying the pipelines option. You can also just say, like I'll, shut down a sink with like a pre cancelled token, and it becomes aborted and that's a little weird yeah.

F

D

F

Following with the pre canceled token means it should throw a canceller, it should throw operation canceled exception. Yeah I can agree with it.

D

A

D

Logical thing, at least, let's kind of brainstorming there we've done that unfairly in place. I agree, that's probably bad practice.

J

So what about the following? If we, if we had a parameter when I thought the parameter, would be graceful or not, and then could the metal be cold close like? Would anybody get confused there? It's different semantics than this pose if it had a parameter like it would be close async and takes a parameter called gracefully I think.

H

I like that, the bath, though yeah, but then what a general rule saying.

B

Don't have a method take a parameter if the parameter changes the essence of what it's doing.

J

It's just a definition of assets like.

F

Openssl in their library, for you know, and for a TLS stream, they they do have just the one shut down and then there's they have a state on the thing of if shutdown is abortive or graceful. So if, if we wanted to have closed and then figure out what sort of boolean parameter for if this is close or abort, because they're both close the connections, just one of them is and don't talk to anybody and the other one is and and and say goodbye politely before you leave okay, the.

A

Problem that I have with that is like let's say we take a boolean parameter. Are we really seeing or not defaulting this thing as soon as you're defaulting at the argument goes out? The window.

J

We could not default it, but you want everybody.

A

B

A

B

Right, the only speaking right I would almost want an enum rather than a bool. Well,.

J

By the way, and also like what is so dead dispose semantics are not graceful shutdown. It's not.

B

Disposed disposed is don't do any IO, just close the resource, so instead, basically it's abort, so instructed would abort on quick. It would cause the other side to timeout.

B

So it would be neither abort or a shutdown.

J

The problem, because we have basically three ways to shut it down and everybody would confuse which one is which and.

F

That could be what Corey said of an enum and then the enum is shut down. Immediate shutdown, graceful shutdown, abort and dispose is shutdown immediate, which.

A

Is don't don't.

F

Talk to anyone and and just be done if.

A

The problem is: if it's quick versus non quick, then it's harder to do when it's on the base, like a Kenya on quick, even doing aboard. Basically.

J

D

Work for anything.

J

I think also I think it from a perspective of you know what what's happening. On the other side, we may have three different semantics from a perspective of the caller I think they only have two cases am I willing to wait or not correct.

B

Well, yeah, if you boil it down enough, it's kind of that, but abort in the quick case is still gonna, send something right. Yeah and the problem is, is you don't have the OS doing that sending when your process closes? So you have to wait if you want, because the fin to get to your side, because.

F

There there are three different things: I assuming I understand what this means, and so maybe I'm wrong. There's do you want to not wait on anything? Do you want to possibly wait on not messing up the other end of the connection and there's do you want to send any outstanding data before telling the other end of the connection that you're going away because abort versus shut down I think would not flush any other buffers? It's just going to immediately send B by the way I'm leaving not.

D

F

J

Just saying between those first two cases, I wonder whether there is actually a practical difference and developers would, you know, want wha one or the other or it's just you know, I'm not willing to wait, shut down or set it try to send the data. And yes, there are implications of you know, choosing the first one and they are different depending on the protocol.

F

And the notion of uh shut down immediately and don't send anything / which really just comes down to I, don't want to wait. It's just no.

J

F

F

So I don't know things to consider.

B

At least four HTTP trees sake, there is a difference between all three they're all being used right now, so.

J

We, yes, there is a difference. My kind of thinking is, there is a difference, but will anybody literally care, like will? Will you ever find the developer on this side of the wire I want one, but not the other, or they will be just thinking. Do I want to wait or not that's all day, I think they will be thinking yeah.

B

I'm, not thinking from you know, HTTP the protocol perspective I'm thinking of our actual implementation, so we've actually used each three ways to close a connection in our current implementation of it. So they're right there I can tell you that we found our use well.

J

You you know saying you: there are three different ways and you expose them in three different api's I'm saying: will the colors ever call one of those three api's? Yes,.

B

You know the client would be the color of this. That would expect to have three different ApS. Really, oh goodness,.

J

Okay, well, in this case, we need to expose all three.

A

I agree with Cory, then innum sounds attractive to me.

D

Yeah I think close, a sync with an enum and a cancellation token. The the tri-state Inu it's good I was gonna joke about adding like an eye connection property options to this. But it's a joke.

J

Yeah I think the moment we had enum it solves the problem of people getting confused how it's related to do dispose yet.

B

B

F

So those two collapse down into clothes a sink and then they take a connection, clothes type, which is a immediate abort and graceful shutdown and probably graceful shutdown gets to zero because it's the probably the most common case. So.

J

One one thing that we, it would be good I, don't have a concrete proposal, but it would be good if one of the names of the enum values kind of pointed people in the direction that this is the same. Behavior is disposed.

J

You I mean yeah.

D

It would just be like the suppose, or something or a graceful.

F

Graceful all that down abort and dispose.

J

Maybe just suffix, you know and add and disposed to the original names that you said.

D

So Cory in the HP pine code, it's calling abort is that taking a big like abort status code, so that would actually be like the one reason to take something like AI connection properties. And so you can still use that method. But it doesn't yeah sure feel great.

B

Clinic requires a status code. Http 3 is very strict on the status code, so we can't just send a default, so we do need some way to specify it here. Well,.

F

But presumably in your HTTP 3 stream generator, you would just build it where you know that abort means you're, gonna send 762 or I have.

B

No idea what HP.

F

3 looks like so yeah.

B

They're they're, like 20 different codes that you can abort with. We only use like 5 of them, but.

D

The other thing is like this connection, you know, will be application protocol agnostic right so like if you want to set specifically like an HTTP 3. You know err status code like the transport trip, after, like the creator of the connection, should have to know about that. Yeah.

B

I think I would be okay, either saying add a connection properties to the close async method or say that or quick if it needs something other than you know the three defaults you need to have this I, quick, affordable property that you pull out of the connection.

D

So here we would delete that anything so that those are the two possibilities. Yeah you either add a specific feature to like the connection connection properties before you ate basically the parameter in the connection province bag. To close this thing, so, if dispose.

F

Is going to be literally just drop, everything on the floor and walk away, then, presumably.

I

F

You don't want to be a sync disposable, because your your you don't have an async notion of dispo like and then that would tell people like if you really, if you want to do the graceful shutdown. The last line before the end of your using is a weight, close async connection, close method, graceful shutdown, yeah.

D

If it was gonna be graceful, it would be nice to be like. Oh, you just put the connection and it using, and that was on the thin. But if it's gonna be immediate, that's.

F

Where it comes down to which behavior of disposed, are you trying to, or which.

D

F

Behavior is disposed, gonna get immediate.

D

Is fine, I think that's, probably the most like logically consistent, so just having I disposable is good like as long as you have this close. They think they can do like all the things yeah.

B

I think we can get rid of icing disposable here.

A

Alright, so now I'm confused, so which one is immediate. Now this.

F

Pose disposes just call all the native release. This handle things and walk away. You.

A

Need not match what.

F

We do on like I'll writer or file stream of oh, you had some stuff that hadn't been flushed yet flush it and then close the handle this goes on. This is just walk away, so.

B

Based on your current code, I see on the screen mo you're gonna get rid of shutdown, async and IA. Some disposable.

A

And then the email members are ordered a bored graceful shutdown, immediate bracelet down would.

D

Be first like the most common thing.

B

Just in case someone is.

F

If someone writes the word default there, let's let graceful.

A

Shutdown be the default. If you do graceful shutdown, a boarding immediate, it seems like we do, the due progressively less work, which means it's probably safer right, even though borcyz better than just walking away right. At least let the caller know that yeah I mean.

F

The color is gonna, see it at intellisense in alphabetical order. The the only real question is, which one gets: the zero value and I think graceful shutdown should get to zero.

A

You worry about me Nikki in Docs, it is ordered by you know, maybe right and they said I guess the order still has value. Even for the non-zero case. Point yeah.

H

A

Make it meaningful yeah.

F

So graceful than abort and dispose.

A

Oh and then a sports recall, immediate right. That was.

F

But this boat, like Christoph, had the statement that it would be good to be able to easily identify which method disposed is using and so I think. That's why we said just call it dispose. I mean you argue lis.

D

Don't need this photos because there's the suppose and suppose doesn't need a cancellation token, because it's out immediate so yeah should we just get rid of the dispose part of the you know. I kinda want to do.

F

We have disposing.

D

I said make good choices not like this.

D

Take with the super nice pattern like using a weight like don't.

B

D

Saying take away dispose.

B

D

B

D

All right, beautiful weight using this is a here's. This isn't that nice. Here's! The problem in this nutshell is that to have disposed like a great silly close, the connection, which is what you do it if you're using and using like that it has to flush stuff in some situations that isn't cancelable, because I dispose a sink.

D

Our I supposedly doesn't have that right and we can't call any methods after we start calling dispose a sink because, like that's a rule, so that means that like dispose, gracefully shutting down the connection, is this isn't oh, go what I don't love it, but we have a thing called icing as possible. It has this behavior literally everywhere in the VCL. They.

F

Say we can't have clothes because it's confusing? No, we just removed isync disposable, because the statement was disposed is not going to flush. It's just going to drop the connection and walk away. Have it flush and.

D

Then have clothes be alright to do other stuff.

F

Well, you can't abort your dispose, then you decided the flush is taking too long and you want to call the board now cause. You called dispose already lady. This.

D

Is an issue for everything in the BC? How are we not like worried about that? I.

A

D

A

Mean one of the problems in the VCL is that clothes and dispose? Don't do the same thing, and it certainly has a problem that flushing is also not always what you want right because, for example like, if soon as you nest, for example, stream, readers or so extreme writers and streams OH, if you dispose the inner one, it also disposes the outer one. So there's definitely an issue where sometimes you just want to flush what you don't want to dispose so.

I

A

These two titles.

D

A

Not being a winning proposition, I would argue, but.

D

We're gonna solve it here in this new abstraction and leave it broken everywhere else that just like this thing is gonna wrap things under the cover. It's called the split, a sink and just hang and can be cancelled. I, don't understand how facing it on this abstraction is, is like I'll, be anything well.

A

I mean so depends on what you want to do right so like if you want to say what so I guess from for my point with like what the question is, what do you want to do when somebody calls dispose on the connection base? What do you want baby to be.

D

Thanks right and what's.

A

D

H

D

Just do it keep it as bad as it is today, because that's the best, a choice we made the VCL and then add a clause as an override or abort or or whatever is, and why is it weren't like not forget, was it I guess that's like wig yeah.

B

D

B

D

The process shuts down like you, don't await it like it might just like it's that cuz, it's like its application library. What, if you don't care about the result using tasks, then you.

I

Don't await it.

D

D

A

So, okay, you just said you want dispose synchronous version to effectively block and whatever it takes to get rid of the data and then close, gracefully again.

I

A

Applies for dispose async, then, except that you would do an asynchronous flush that.

D

A

Get being to be canceled.

D

Because that behavior is what we have today: okay,.

F

Cuz cuz yeah. What and what was said while you were gone, is that disposed would just drop the resources and not send an explicit aboard for protocols that require it and not flush any outstanding buffers, and if that's the case, because that's why, in our close method up here, we had graceful, abort and disposed and suppose means just walk away right. Tell the OS I've done with the socket move on with life. What.

A

Is that locating for us that this pose is great for shutdown, yeah.

D

Yeah, absolutely this is what he's advocating for and to make that work today in Kestrel and it does work today we will sometimes call a board after we call disposed.

F

And certainly there's a model where it's fine, you call closed async, graceful shutdown, cancellation, source from I'm, 5 or whatever yeah.

I

F

Then in that aborts, then now you call clothes, async, immediate or clothes, async, abort and move on with life. The problem is all of these are actually all disposing operations, but the fact that it's not the word dispose means it. It needs to internally manage its own stage. I.

B

Think that the distinction here would be, you would use icing disposable if you just want graceful shutdown. But if you want to be able to cancel that you would call closed a sink with a cancellation token, and then your dispose would just be a no up right.

D

And it's fair to say that, like if you're managing, like both sides of accidents stuff, there are a lot of times like you're, not worried about cancelling. But.

F

So yeah, so it it just comes back to what you want the behavior to be, though, if, if it's graceful shutdown, then it's I think disposable. Again, if it's drop everything on the floor, then the lack of icing disposable tells someone. You probably want to call a weight, graceful shutdown.

D

So let me let me try to reason about this again. So the current this current pattern is I call close async if I want an abort, that's what abort says: kill okay and I. Don't don't wait, go flush as.

B

Well, graceful shutdown is flush. The barrette.

I

B

Do any IO to tell the other side that you're aborting like sending error codes dispose, is you know totally ungraceful we're just going to close resources and not send anything yeah the.

F

B

As the process shutting.

F

Down, but with about the process of shutting down and everything can be canceled. Oh.

F

B

Not canceled Malia because.

F

All it's doing is calling you know close handle and this moves on with life, but right.

D

But what you're suggesting David- and it seems like we're getting some buy-in for this- is that this pose will be graceful. It will flush. It's not cancelable deal with it like by calling close async. If you really care yeah.

F

So dispose would then be closed. Async graceful, shutdown, CT, none right, yeah, exactly.

D

That does seem nicer for the average user and I.

A

Think we did also kind of solve our problem that hit earlier because now literally disposed closes, like the basic dispose, calls close yeah right, yeah.

F

We could basically have.

A

A close, I think right.

F

So you could leave connection. Close method is being graceful, abort and immediate, like we had previously discussed and just dispose as graceful instead of immediate yeah yeah.

F

Because that gives you the the way that you can say, look I want you to terminate yourself and you're not allowed to send IO, because whatever reason like so I, don't care that the protocol says to Sendai. Oh, let's just pretend: we've died and move on mm-hmm.

D

The other spicy conversation you missed out on David is that you want to provide a you know, quick. You know abort status code, as you call you know closely async abort. What's what's the way to do that? Do we add, like connection property options, the close.

I

I think property.

D

F

Mean I think the answer and that one is you either have polymorphism and you're casting and you know you're doing quick, and so you have a quick connection or if you're, using the property bag, I think Stefan. You suggested this of you fish out the quick connection status, property bag, which has a mutable. What's my clothes code, yeah yep, but at that point, you're working with the protocol and something more complicated has to happen. Yeah.

D

You can definitely use the connection properties on connection base that, like sets on state or actually do the abort. If you really want to.

A

So what is this edition? Then? We add the enum back and have a different. Yes,.

B

Of the enum, with slightly different names and then say, dispose async, we'll just call graceful shutdown with.

F

No cancellation.

A

What so, then, one thing at a time up.

B

A

B

A

I

That mean that.

B

Dispose we'll call clothes, async and just ditch the task. Well,.

F

I mean no return that, oh sorry, dispose the synchronous method, yeah.

D

Or Blanca will it bluntly, bought lock right, yeah I mean or we could remove it.

E

D

Want a non-blocking dispose, you call clothes, the async immediate, and then the space will be in it. When you follow.

F

That so yeah, so then the the enum emo is I, think graceful, shutdown, abort and immediate and then dispose does the default of that, which is graceful shutdown. The super spicy food for thought for the future, but we're.

D

Gonna need this feature on a bunch of other things, maybe like longer-term. We should we should reconsider, like thinking about streams and stream, writers and those kinds of api's. How do we do the same behavior for those things, I? Don't.

I

Wanna flush, I, don't wanna, hang yeah.

B

F

I mean the easiest way. Is you just drop your reference to it? Yep like just stop talking about it. Eventually the GC eats it and it doesn't do anything so hopefully Oregon's. Finally,.

B

Okay, this looks good to me.

B

Yeah me, of course. Oh, do we defaulted because then you're kind of confusing close with dispose well.

A

I mean if you default it like this I mean basically people what what suppose does this was called close, a single, no argument. You.

B

Know it's true.

D

You guys discuss half close or is that more uh there.

E

Won't be another.

B

Property banishing.

D

Yes, today, like we used to pipe to be well, we intended to you to pipe this signal, half those, because it has that kind of notion of losing input or output, but maybe it's worth like discussing that I mean not. No, it's fine I mean.

F

If that's just more modes on connection close method, then that's just more modes on connection close method, yeah.

B

F

B

um I think it's the same problem as a quick abort code. Is that you now now you're in TCP land and you should pull out a TCP specific interface.

D

Will be I supports I can't ask the port half those connections.

D

Is awesome, progress I'm super happy if you're and everything.

D

You camera I would okay. Can you guys hear me I would compare the hapless thing to almost like remote and local endpoint like? Is it common enough for streams? You know and transports to be paths closeable that it's something we should make first class right and in assessments.

B

Have opposable, but only kind of so be another one of those features that would be just be very opinionated about TCP, workflow.

D

And then we have a bass classic. We can always add it in the future. hahaha It's right Christoph. We are allowed to add anxiety, noobs right, that's not breaking correct. You can't renumber things.

F

But you can add yeah.

B

Although other people will be presumably implementing this close a sync and if we pass in a random new value, who knows what'll happen.

D

That is a problem like you, you don't like yeah. What should they do.

B

Okay were basically out of time again, I think what to schedule a third of these.

A

Converging on something I mean that yeah.

B

It's it's you're, making progress, I.

F

Mean we did resolve of looking at what before we had on of the issues its we took a the unresolved discussion about interfaces versus classes and I. Think we've landed somewhere that we're all at least content with. We renamed connection property collection, we renamed connection stream. We collapsed some interfaces. We figured out how to add abort there's, do we do disposable factories and are they one type or two? We didn't talk about yet, but but you know we didn't make Nick. We made positive progress.

D

B

A

It's really good all right, I'll committal his notes and then I'll send them out. One more round make.

I

Sure everybody have one more I think it's to.

D

Come back and jump in Emily, yes, we're almost finished or be like what the heck are you doing.

D

J

Know I compliment you on the microphone and the lights and everything. Oh. Thank you very much.

I

J

Need some tips from you, especially that I'm teaching a class next week at the whatever learning day and and here I have like I, really don't have any lights and I'm all like yellow on this screen, because.

A

J

Have like wood panels here, so it's like all horrible lightning like so.

A

Crystal you may not like it, but like I'm, a Kenan fanboys similar to you. But if you want to be your camera, you want a Sony, the canons a bit much.

J

Yeah I'm, usually.

A

Yeah I've ever sony a 5100 which is not if it's it like it's a dollar, all the less. At this point you get.

J

A new system where they.

J

A

It's a it's a DSLR right but like unless the Canon Sony has a better model like cannabase, you protect the high-end video cameras right so that don't give you certain features and one feature that Sony gives. You is clean HDMI. You can basically just literally plug in the camera to capture card, and then you get effectively a webcam on your computer.

J

That's on there but Sony, the whatever seven are mark three yeah.

A

So then you have a high-end camera so that that works, yeah and.

F

I had the same thing with my Fuji X, the XT 20 worked fine I just needed to get an HDMI or a yeah cap HDMI to USB capture.

F

B

Announced you can plug in just USB on their camera and it works as a webcam. Now, do you see that one of the manufacturers put something out yeah? There was.

A

J

A

The best-connected review is B and the problem for most Canon cameras is that they were really cheap on the USB connection. The data right did you get via the USB is not high enough, so you don't get.

I

A

You get 720p in many kids, don't even 720p you get so sometimes even what is it? 520 people over the next lower one is or.

J

A

Yeah and then you don't even get like you, don't even get fulfillment.

G

A

Sometimes they'll to drop frames so like it's, it's not great. It's.

J

The best capture device that I should get, but.

A

The one that I bought is the gambling from I got. Oh, the problem of the cam links is, they are always bought out. So it's really hard to get them right. Now you can get them on eBay. They usually cost 150 bucks new, but you can get them on eBay for only $350. If you're willing to shell out that much money.

A

F

Got the cam like 4k and when I ordered mine and wasn't on any sort of delay, I mean it wasn't during quarantine.

J

A

$300 and if I'm saying not new, it's $140.

J

Okay and what about the light, what like do you have? Can I see it.

A

That would be a lot of work right now actually get the lights, but these are cheap, LED lights from coming from Amazon like 20 bucks or something Oh how's.

J

The ambience, like the lights in the back sorry.

A

That's the one it's at the ones I have like the ones I have here like the toy, like super cheap, IKEA lights. These are the ones Oh. What.

E

About the mic, yeah.

A

The mic is not cheap, the microwave to.

E

Where I say 70, it's.

A

A four dollar mic from my extend and a XLR input for.

D

Mic yeah: that's awesome. The.

J

Problem is my exist. They need to be very close to you. Correct yeah,.

A

So that's the downside. So basically, so all you I didn't want the mic that is close, but the basically you have two options. You can either go with a ruffle, lavalier microphone right, the lapel mics that you did attach to your shirt or something I have done this for a while. The problem of those is, they are effectively Universal pickup pattern right, so they pick up everything. They pick up your keyboard. If you cpu fan goes up, they they pick that up and the other problem is your wired in.

A

So god forbid, you forget, and you get up to go to the restroom, and then you yank out your whole computer equipment with you I've done that too. What only now has is. She just has a shotgun microphone that we mounted on top of the webcam and that one is a bit more directional and this one is basically pointed at her. You know face basically well, ideally the mouse right and then this one picks up a lot less noise, but it still picks up more than these guys do right.

A

So this one is a dynamic microphone, so it basically has a very aggressive fall out. It almost picks up no ambient noise whatsoever. It only picks up me, but if I move slightly back Elihu me now because, as he picks up nothing right so yes, the downside is, they are more in the frame so like. If you don't like a microphone in the frame. That's that's a problem, but originally I wasn't that cam, but now I don't mind it anymore, like I, just really liked the voice and the fact that there's a microphone.

B

If you want a dynamic mic.

B

A

D

You have one to you, people that have these super fine setups have.

A

You seen the tweet from what is his name: uhh Jason I'm blanking on his name right now like he was basically saying he was working remotely for like ten games.

I

Ya know Microsoft.

A

People are working remotely for like two weeks now and they all order, like DSLRs in green screens, they're like what's.

B

Wrong with people.

A

No, but it's good like I'm, actually quite in okay, the only blonde said know is that my computer is not fast enough, like I'm you're, dropping a ton of frames, because team then Tobias at the same time are not good partners. Yeah.

J

And the internet seems to be I, don't know if it's internet or something else but you're. Now your screen is Larry and not anybody else. Well, maybe Jeremy also has a crappy internet. You.

A

Know it's not internet and mike is it's a cpu like I'm, a hundred percent CPU and I'm 100 percent GPU, and so I.

F

A

Frames at this point like this I I.

F

Got tired of having my I xt 20 mounted on a tripod, but how behind my screen, because I kept using it so when all this started, I ordered anything any webcam I could get that shipped from the United States and was not backordered. So I have a 480p manual-focus webcam that sits on a pole three feet away from me.

F

So mine is camera crystal.

A

That's the other thing why I bought a dedicated camera I like even if the Canon would have worked, I, don't like the mountain and Martin unmounting, part of it, but I just have a dedicated DSLR. Now that's permanently mounted so it's connected once you take care of the cables and the quality is good enough. I! Think honestly, like even if you have a high-end, Sony I, don't think you get much more out of it, because the full-frame doesn't buy you that much for this.

A

The only thing is: I have a nice lens as well over 1.4, which you know Louis what the background quite nicely, and so the whole setup here is I, think I, don't know seven hundred bucks or something so, if you're willing to shell a three and a half K for your camera exam and it K for your webcam seems reasonable.

A

J

Don't know the problem is, I was shelled out so much.

A

This conversation, all right, crystal I, will come later today. By the way, are you around yeah.

J

Because you have one, you have very nice lesson: it's not yours, but.

A

I like how much I paid for their learning.

I

A

I'll call you later.

J

I