.Net Foundation Design Reviews, 30 Apr 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

00:00:00 - Approved: Implement global proxy configuration for HttpClient https://github.com/dotnet/corefx/issues/36553
00:34:19 - Approved: HttpClient.DefaultRequestVersion https://github.com/dotnet/corefx/issues/37224
00:47:46 - Approved: add Netlink, Packet and Can to AddressFamily and ProtocolFamily enum https://github.com/dotnet/corefx/issues/35143#issuecomment-488053098

A

All righty, so we have some conference projection issues, so we all look at our laptops right now. You can see the blue screen in the background, so it's not really helping our case, but basically, when I look at to httpclient issues before we go over the other ones, w preference for which one first of the proxy proxy.

B

A

Even I'll call no and I'm Jeremy I, both on the close all.

C

Right tonight, but I can also talk to we get Stephen open up the other one for us.

C

So you want me to summarize yeah there's a lot of information here, but basically, what we're trying to do is replace the fact that we don't have app config anymore. We don't have a way for people to configure a proxy with app config or even say they want to use default credentials.

C

So we have a lot of blocking issues already where people can't even do you get restore you know, because it uses HTTP cuz they're on a corporate network that uses default credentials requiring on a proxy and there's no way to tell any of the components. They should do that so basically we're trying to replace app config, which we don't haven't gotten in court generally, and we're also trying to replace a common pattern that people used in the framework, which was a method called web request.

C

That was the actual type dot default web proxy, which was the static property. They moved away from all of the web request classes because we considered those legacy they're part of our system that requests namespace type and library. So basically, what we need to do is add back something that enables customers to latch on to what what the default proxy is and the default proxy now is cross-platform. So it's not just when I net settings and Windows it can be environment variables.

C

We've also enabled environment variables for Windows as well, so they get the same experience across all the platforms, OS X, Linux, etc. So this kind of completes the rest of the puzzle, oh for global proxy config, and enables them to set just credentials if that's what their environment is or have a hard-coded proxy, which a lot of customers did in the existing app config. We already have an existing schema. That's a framework add for system net, a lot of interesting things and people keep complaining. Where is that?

C

Where is that there's a plan that someone else can write a compat layer that basically will parse the XMP old style, XML file or Jason, follow wherever people want, but we need an API mechanism in core that could enable those scenarios because, of course we don't have a direct config replacement in core so anyway. This is really this property's, all about and kind of completes our enterprise authentication scenarios for proxies so that someone can inject the credentials using the default proxy, there's a credentials property on it or they can inject a separate proxy.

C

It also helps the problem with when they don't control the HTTP client. We have a lot of components being built now, like HTTP client, factory and a whole bunch of other stuff.

C

That's making these HTTP clients for everybody, and they don't have direct control and the one thing you know they want to be able to control as being on a network, that's not under their control, so they need to inject sort of a little bit of configuration, and this is how they're going to do it, and it's very similar to the pattern we have in framework. It's just on a different type, with a slightly different name.

D

So I had a little bit of confusion about what you just said. So you had mentioned that in the past, in full framework, we would be able to throughout config specify proxy settings. Did you say that the intent here was also to allow something similar to that or to allow or or or is the idea going to be that if I'm an application developer I should a tap start read a value from somewhere and then call those properties? No.

C

If you're an app developer, you can use this API directly, okay, but there will be customers that say we live and we have components that just we like defining app config, and you guys, you know we wrote a thing: that'll parse, app, config and kind of emulate everything it did. We just need api's to make it work cook up, and now we have an api to make it work. So.

D

In inside the new Ghia client, for instance, the nuga client would have to have code inside of it which reads the proxy settings from some file and then calls them to safety on. Yes,.

E

Yeah and that will enable them to do this without it they try to say nothing worked, but it sounds like it's going to be on the individual library or.

D

App developer, instead of something that the framework itself will provide as a first-class say. Yes,.

C

Because core no longer provides as a first-class citizen, a control mechanism- okay, so so.

A

What what instance of hope I said it's like I mean it's making.

C

A

So I gave some examples under cname. You would have to implement I that proxy, then right, because we don't ship a public implementation. Is that fair? No.

C

You don't implement it, you do a get and you automatically get something useful. You automatically get these the proxy up, that's defined on the system settings on the platform. It's basically it's an it's an object model of, what's already defined on the platform, and now we have a mechanism like there's two, this two basic use cases I put in the sample. Api usage.

C

Part of this issue right here, there's the case where I just want to use, what's defined already on the machine like I met when I net settings and the IE or an OS X, there's max settings or on Linux I, have environment variables or maybe I have environment variables on the other machines too. But I there's no way for me to say: I want to use default credentials, so I need a place to inject that- and this is the this. Is the mechanism so default proxy? The default value?

C

The initial value of this property is non null and it represents an object that you can set the existing properties on it already exposed that this property exposes the iweb proxy interface, that's already defined in the framework and core, and it has a mechanism that you can, just as you see just set credentials and now once you do that one line now it will use the system proxy, but now it will be able to carry the using credentials and get through of the authenticating proxy. That's the that's the one important use case.

C

The other use case is that I have I want to override that setting- and here is the injection of my class called my custom proxy, which also implements the standard iWeb proxy interface yeah.

D

C

Are the two key use cases that we see everywhere? So that's.

D

Why you have a property setter them not just together, because there.

E

Is a gathers, no they're together? Well, no, no, but I mean this is why you also have a setter.

C

A

I want replace the whole thing. Okay, absolutely intensely. When you bullet points you say on Linux. If no matter goes I said then Alice returns. Yes,.

C

So I was describing so what I was describing here is, irrespective of us doing this API mm-hmm it's. This is what we've now enabled in dotnet core. As of three, oh in the master branch, where we have a consistent mechanism or because environment variables had been a very traditional way on Linux and customers also were asking for I wanted to work the same on Windows two, because I have all these docker containers.

C

So as part of broadening the scenario we made environment variables work in a consistent way, you have environment variables that are certain set defined. It will use that as the basis of the platform, proxy definition, but environment variables alone, don't capture the credentials that you want to inject. So all of us describing here is well. How else is this used with the rest of the mechanisms that we've already enabled in core?

C

So we have a consistent story now, if you have docker containers or whatever, which sort of inject these environment variables for you, and so now we have a very consistent story with how what the definition of the system proxy settings mean even on windows, and now we have this mechanism where you can inject either the cadet into that the credentials or in many in some cases, do a replacement of of it, because customers may have a fixed class that they wrote that they want to inject a proxy game configuration those.

E

Were the two use cases that have traditionally been used in frame, so the environment variables are automatically picked up? Oh yes,.

C

Yeah we already did that change recently. Is there.

F

A way to disable hookup, environment, environment variables for people who don't want global mutating state is that's been a problem with like emma spells and stuff before, where HP went, to find a platform, an environment variable that they shipped on their laptops and then msbuild broke.

F

C

People want to disable the environment variables. Do the environment variables typically are not defined, but by default they're not defined, so we do fun with the name. We.

A

C

The same naming pattern that curl and everyone else use.

A

It's like a TPM but across the APS, and it's where.

C

Proxy no proxy and on all of that so and by default they're not there, but they.

F

C

They can be created there in the process whatever, but.

F

What I was asking is: is there a way to configure HTTP client to say, don't use the environment variables even if they exist so.

C

On so so, um I'll turn that question and say: is there a way to configure HTTP client not to use a system configured proxy? That's the way, I'll turn your question around, because the environment variables are just another way of saying I have when I net settings, which also say use a proxy. So is there a way it needs to be client to say, don't use system proxy settings, and the answer to that is yes.

C

In general, HP client has a mechanism when you create an instance of it to say, use proxy false and it's not going to use any proxy at all, no matter what adding this is also now an additional mechanism, because you could inject a default proxy with the setter and your my proxy custom proxy. Basically is no proxy. You sure that you define the bypass setting and now you don't use it for experiments. Oh yes, and.

F

My other question is: have we checked with asp.net and made sure that this will work sufficient for them with their app settings, Jason that they already have um we've.

C

Talked to AST net in the sense of they can in they can use this mechanism if they want and call it their app settings Jason right now, I mean they define, they may define their own schema they if they want. If, if they want to enable config like things to sort of replicate the XML schema, we had an app config web config. They can, they can define it whatever they want it to be, hopefully it'll be similar and they can use this API to make it work.

G

Early apraxia on occasions.

G

C

Yes, by default, HT Carnot, when you create an instance of httpclient, we use proxy is true and it will use the system proxy no I mean. Is there an iWeb proxy property on e? Yes, there is.

E

H

Proxy property yeah.

D

Steve's online I've made it that far since we don't have the API on the screen. Okay, so this is so. This is the static and then, when you create a new HTTP client the static it's used as the default out there is the default value administer.

E

C

Effectively, yes, unless you override the proxy property with something else at the instance layer, yeah, so.

E

What's the behavior, if you set the proper digital, oh I, don't want this property set to no okay, so you shouldn't. Have it returned? No.

C

One way it will never return Norrell, even if you so, if you didn't define any environment variables, um it will still return this object, but the object will have sort of the side-effect of its bypass.

E

True, always okay, effectively I see just your your statement here at least the issue statement, the default value of this property.

D

On Linux environment variables are not set and always return. Yeah.

B

C

um I guess what I meant ain't.

E

C

um You can so we can decide that was sort of an implementation behavior. We were trying to like figure out what made sense, because either way the issue with Linux right now is Linux is the one OS platform where we don't have two choices of system: proxy settings on Windows. We have environment variables and if they're, not there, we'll use IE settings which of course, IE settings may result in no proxy either and then the same for OSX. We can still choose.

C

This is sort about what we think is confusing or not. If you get the null I guess is the issue: do we want to make sure if developers try to set a credentials property on it and it's really no it'll it'll throw when we sort of just want it to be a no op I.

D

Would say if your senator is going to reject know your dinner is never allowed yeah.

C

D

We can definitely and.

C

We can definitely do that and we can return an object. That's not null, but its effect is that for every uri you want to go to you're effectively bypassing a proxy and that's fine and that's a consistent story, and I can roll that in there as well.

C

But that makes sense yeah if you can't set it to now. You don't want. Yes.

E

So essentially, when.

C

No ability yeah it's around to this one, just yeah.

E

Say that it is not dog yeah, because it.

C

Really is the same thing even on even on Windows. At the framework, the default proxy and web request was non null. Even if IE settings had all the boxes, unchecked or IE settings said I'd like to discover a pac file and of course there wasn't one on the network, so it would still return a non null object, but the effect was you never went through a proxy, so that's fine I can make thing. We can do that and then then we have a consistent story.

C

It's never know when you get it and you can't set it to know so.

A

How much work is it for me to implement the interface there's only three members, but how do I implement them like we don't have a default, one you can derive from and you just have to literally implement it. No.

C

We do actually we have a web proxy class that we can use and it's a sort of a simple kind of class, and you can give it some interesting little parameters when you create it. It actually has a history of there's some brokenness in it, and I didn't really even want to attach anything more to that class. We have it for compatibility.

C

We have some other weird methods related to it that return instances of web proxy instead of an interface and I web proxy, which we didn't want to do, but there's a lot of interesting historic, fifteen plus year old comments and framework that say wow. This was really broken, blah blah blah, but anyway, that answers your question. Yeah. If developers don't even want to implement the three methods of iWeb proxy for custom proxy shorts, they can just uses the web proxy class for a quick conversion.

H

Console if they wanted to be able to delegate to existing behavior in some cases they could always grab default proxies from its getter, and you know wrap that in their own iWeb proxy. So if they wanted.

F

H

Of the settings, except when I'm going to URL blah, then they can say if URL equals blah else. The captured default proxy dot get proxy yeah.

C

C

Everything we've seen historically when people write their own glass that implements I, read proxy. It's pretty simple and something you know it's like ten lines and that's your problem. Building their own little one.

A

I mean one person is asking the chat: if it's not null, how do I determine that I'm using a proxy or not.

C

C

I mean that's a good question in the sense of.

C

In general, you don't know for sure when you're, if you're using a proxy, unless you call the get proxy method of the proxy with a destination URI, you know and it returns, whether you're being bypassed or not. We're we're basically returning an object model of something that represents the settings that tells you like. This is how you can figure out if your destination URI actually goes through a proxy many times.

C

If you know you're on a corporate network, there's a JavaScript pack file that got fetched and it's got a complicated algorithm for figuring out for certain. You are eyes you're going through proxy one for certain your eyes, you're going through proxy two for other your eyes, you're going direct and so to answer the question about whether you're going through a proxy depends on what URI you're providing. So it's it's really a function, but it's an argument passed to a function and but you could tell if you are actually since we're always returning a non-null object.

C

You can call the is bypassed method of it with your destination. Uri and you'll know whether or not, if you're, using the system proxy, whether or not that URI goes through a proxy or not. That's the answer to that question. Of course, this API.

H

Also, doesn't really change that that answer I mean like.

C

F

H

Create a basic line today, the proxy property is initialized to something, and all this API is doing is allow you to override what that something is Yeah.

A

Right, you think the question from Fredrik was probably more along the lines of like you. If you're in Linux, no environment variables, I could figured the environment settings are basically saying: don't use the proxy, then the instance we're giving. You is basically in null instance by the base. It just says: don't do anything and how would you know that and the answer from you would be you shouldn't special case that you should just generally assume yeah.

C

Right, you would get, you would get that property say and you call is bypassed, which returns a boolean and you give it a URI and every you are our you give it. It returns. True.

A

All righty is any more concerns with this one. My.

H

Only question is purely about naming and I: don't have a agenda. Here's just a question: every other property on httpclient controls that instances behavior, whereas I think this is the first static we're talking about adding HTTP clients. So we should be sure that this is where we actually want to put it. And then you know from a naming perspective default probably makes sense, but the next api we're looking at I'm proposed that it also be prefixed with the halt.

H

But it's actually the pros as an instance and it's affecting something where, where that also makes sense. But now we're gonna have two defaults that both make sense for their own thing, but one of them being a static and one of them being an instance. So I just want to make leave. Maybe we should like I mean.

E

H

D

Like there's already an instance, starting with people, there.

C

Is this the default request headers on on HTTP client I mean if we don't like default proxy I used the word default because historically we had default web proxy on the web request method. You know we could also use system proxy. You know or platform platforms kind of, not a great word. I know systems. A great word I mean.

H

I'm, actually, okay, with default I, just want to make sure that holistically. When we look at anything at all I.

C

Thought it was I thought it was the least bad of all names and and not and not and not. It was also discoverable Morton, because people would be looking to ask about web request default web proxy and you know we're telling them a web. Hd client is the thing you should be thinking about using now and and so it's I thought it was more discoverable.

D

Yeah I mean I think the only thing that I would stay at all based off of Steve's raising the point. Eight, this be on a separate class called something like HTTP client defaults, so that, if you have other static defaults in a sort of these are the Global's. And then these are the instances so and yeah.

C

So I I thought about that and it sort of brought me back to the horror service point manager, which was the other sort of global class that then everything else depended on and I sort of thought like I didn't want to go there, and so that's why I thought that it was more discoverable to just put it on the thing. That's ninety nine percent you're using it with. So that's why I wanted it on HP client.

A

Would it be kinda fact we make more sense or lessons HP.

C

Klein factories, not a thought net core class, it's something that got created grow as a workaround for the fact that our HTTP stack doesn't do true. Ttl timeouts of you know on DNS and connections, and it was created as a nice workaround, but eventually we expect HP client factory not to be needed. Once we improve our stack to be more aware of like time to live on, DNS records and stuff, we.

A

Don't really have aesthetic type then there's like some ambient state or anything like that right, so in that sense is probably the best place to put and as you want to cook up a new type, but that seems a little bit over the top for one property. I I didn't.

C

Yeah and I didn't want to create a new replacement for service point manager right.

I

D

I mean just the the point being that there's already the default request, headers, which is instance, he wants to add equal request version this instance, and then we have default proxy, which is static.

D

Now you can't like when you, because intellisense doesn't as far as like well doesn't that really distinguish between instance and static you're on an either you're on an instance of the thing and you hit dot, you see all the defaults, but what you really wish is that it only showed you the static ones, on the static type in I. Believe.

A

That citation pleasure their daily another Taylor leads to call setting methods. Okay, yeah I mean isn't that some we can states with torn I guess if I think.

E

C

E

A

Aesthetic on an instant, oh yeah, but don't think I think it's always about the filters.

A

No now I think the problem is more if you want to have an instance method and a static method with the same name, because that does the fly as I all doesn't support them because they're not different instances a mile, but thinking too sharp they're, separate namespaces, but I agree with Stephan, like from the point of view like it's the first one you know do we want to have said that I mean what if we do other things like before proxy like this is a good place to put it. If so, then, let's do it.

A

If we believe it's not, then it's not because it's the first time you do this, but I yeah I, don't think you have a better place to put it yeah.

F

I think it's fine yeah, it gives you explicitly. Member X cannot be accessed with instance. Reference qualify it with the type name instead didn't tell is in suggested. No. There.

A

Will be bad intellisense? He right.

D

Now, what we in intellisense.

A

D

Have the S thing on it? When you did it, then I tried exactly change the left side to give it like I thought it was trying to be overly helpful.

D

A

Maybe all right so then, let's to prove this guy, if I find the right tap on my little disaster, that I stood up myself here.

A

A

One down all right now: let's talk about HTTP client default request version actually.

A

Surprisingly difficult top, you talked about.

G

This before maybe not.

H

I, don't think so. No.

E

H

Something similar so the issue here. The idea here is HP client is a it's a thin wrapper around an HP client or an HP message, invoker, which this provides the send a sync method and you give it an HP request message and it sends it off.

H

Hp client wraps that, with some helpful functionality, there's also a send async that just kind of goes directly through to the handler, but it layers on functionality like some some nice helpers for just saying just get string async like get me the whole contents of the response, as a string, so just a one-shot call or get bytes async, or it allows you to have it handle offering the response. If you don't want to do it yourself and so on.

H

All those helpers, though, because there's sort of this one-shot thing layered on top of that send async they end up. Creating that HTTP request message for you, so you know they rather than you saying: VAR M equals new HP request message configure it. However, you want, and instead fast office and async HP client. Does that, on your behalf, HP request message: has a version property on it and that controls the protocol version that the underlying handler uses.

H

So if you specify one point, one you'll be using HP one point one: if you specify 2.0 you'll you're asking for the handler to do hb2, but with HP client creating that message for you, you have no way to control what version it sets it to and that's a pretty important thing to be able to control. So what this would.

F

H

Like HP client has default request, headers or whatever the property is called where, basically, when it creates one of these message hit. One of these request messages. It populates the headers, with whatever you specified in that collection, ihb client. This does the same thing when you, when it creates a request message for you, it will populate the version with the version you specify. Now there is one small difference and it's worth talking about, and that is for better or worse, the way that that default, request, headers property is implemented today.

H

It adds those to every message that you send, regardless of whether it creates the message for you or not, whereas this default request version would really be a default. If it's creating the message for you, it's going to set it, but if you just give it a nature, you request it wouldn't touch the version at all.

H

We could talk about that. There are other options, but that's how I currently have it defined.

A

Know everything you just said sounds really familiar. I swear to God. We have talked about setting the default version. Oh yeah,.

C

So I must tell you the background of this. Is we thought a year ago? Maybe we were gonna come here with opt-in, opt-out, we're adding h-2b to feature, and we thought maybe we'd have a version and then actually we had a conversation about it at the time and we actually had it in a different format and it was decided we're not ready to have this API, because we don't know, you know why we adding it. Is it because we want off tin, we want opt-out, and so it was decided.

C

Then, if you just want to play around with the experiment of the experimental, then at the time HB 2 then use app context using environment variable just allow people to try it out and come back to us later. When you have an API. Now we're coming back with an API and that's kind of why there's a DejaVu going on here about this yeah.

A

I thought there was some issue that I think they discussed at some point where somebody suggested adding an default member to HTTP version, which would basically be in max video, and then it would just use the highest version. That's available on an open platform which might be too low for someone. One fatherís and I think we've rejected that one, and that might be because we talked about that one. But I have definitely.

A

Yeah I think I have no problems of that. That seems like a reasonable way to do it. Yeah.

C

So I just want say the other thing that we've learned with using http/2 and having customers try it out. Is we don't want to have the default as 2.0, so one of the things that we did last year or the year before?

C

Is we tweaked one of our classes, HP, request, method to feed, request message and we said: let's make the default version, property be to O and then maybe stuff will just light up and it actually works that way for you, 80 and we're gonna leave that behavior for you AP and master, and what we discovered is that well there's some things that we don't want to make HP to the default. You know the NT authentication kind of works in some differences, and so we decided that we really the default.

C

The initial value should be 1:1, but we need an easy mechanism to allow people to turn it on when they need it for G, RPC and other reasons, and so this has sort of been the learnings that live up to what Stephen row here in the proposal for for this property. Allow them in easy mechanism to turn on HTTP, to the reason why we want a property on HP client. Is that many of its methods? You don't pass in an HTP request message.

C

You just say: I want to do a get a sink to this URI, and so the question is well. Who is creating the internal HTTP request message? They have the version property. Well, it's HTTP client, it's doing the convenience for you, so for customers that just use the less convenient way they use a send async call of HP client. Where you create an HP request message: they can create it with whatever they want.

C

It defaults to say one one, but they can simply set the version property and for that request it's going to use whatever version they want, but this property is mainly useful for the 80% case that use the convenience methods of HP client, where they're not creating a separate request message and we need a way or HP client to do a good job at deciding what that default should be since they're, not creating it directly right and that's the motivation for this property. An HTTP client instance didn't you I, think.

A

It's simple in the meeting I said I think you said like, because I don't get to accept the default. We should have yeah.

C

If we look back at the recording, I'm I think you said something.

A

The only question I have is like do we think that, like instance, is the only scope that makes sense, because that we don't want the global setting to say if you have an HP client? What do you want to need people for that to be? We.

C

Discussed it I discussed it I kind of proposed it originally with Stephen and our smaller team meeting, and then we kind of went back and forth on it and goes well. The problem with a global setting, like that is, you would be affecting, like maybe a library, that's creating HTTP these messages and that could really change things. The way the library maybe isn't tested for that it's different than like injecting proxy settings which are environmental. This is injected.

C

This is kind of changing what the library wants to create and so injecting it at the static layer. We decided that not a good idea, so this is why it's an instance method. That's.

D

A good point because to what David just said, you would actually be implicitly changing the libraries contract with the servers right, yeah.

A

It to be clear, I'm, a sadist in doing all I'm saying, is that if we take the property, as is if we.

F

A

Want to add the static layer. Well, we ought of I think.

H

I think if we were, if we wanted to add the static layer, we would add it and, as David had previously proposed to HP request message itself, that.

C

Was what I proposed I originally proposed as part of this having a static property on request message itself, which sort of just changes every request message that anybody or any layer makes use this default instead? But you know so we still have that ability if you choose to go down that route, but this so adding this property as an instance method, HP client doesn't mean we can't do the other thing which we may not want to do it. Whether.

A

You want to have it if you want to say that if a new initial peak line I want the value for the instance default, the first version to be that one. But that's that may be different from saying everybody who creates a message defaults to that one right. That's not necessarily the same. Yes,.

C

But you're still injecting something which is totally unexpected. ah Yeah.

A

I'm, not just anything I'm saying is that I think if you want the other policy that I think probably want I, don't think when I put it on the other tile, they would put it at the same time, but under a different name. Okay,.

H

B I think my what I thought we would enter doing if we did that is. We would basically just say that HP client, its instance default request version is initialized to HTTP request messages, static default version. Oh.

I

Yeah I just come to this yeah. That's ago, yep yeah became quick full what it does because then the innocent.

D

Still wins, it's just yeah use the other type for it's. That's great whole people yeah, because this honestly seems like a good compromise for what did you say around the 80% case. Most.

F

D

If we don't see a good reason right now to worry about having a global static and like I, this one really should get the better name. I. Think.

C

And the name was chosen because we have a default request: headers on as an instance, property of HP, client and so default request version as a similar name as an instance for things seems the right name.

G

You asked Sarah to use if you something HTTP to request the server doesn't support it community. Oh yes,.

C

Yeah that that's in general, how H to be to works it its first actually does the TLS connection and in in the ALP n, it's kind of saying: hey I want to do a TLS connection, because I want to connect the h-2b tube. So it's passing that little h2 and then the server you know at the TLS layer says: I, don't understand this no way and then, by the time the TLS connection comes back. Everyone knows whether or not they're gonna be happy talking, hb2 or not it's something like requesting them or something.

G

They requested merchants learning. If you have the stuff, though the HDD play and talk request for them.

D

They request immersion doesn't its since it came down great that that's protocol created on the messaging. You call it like. This is just deciding what it does when it creates. The other object. That already has this property well default names. Great cuz I mean Kent and to server negotiating 101 stud. Also, even if you say one one, oh sure yeah, so it's never truly honored anyway. Yeah.

A

You never required.

D

A

Yeah I think the default is basically what what? What is the? What do you request by default right right instead of request right, so you're not getting guarantee, but.

B

That's already in the nursing, I.

A

C

Thought the name being similar to default request headers just made more sense when you kind of bring it up on intellisense I think default makes more sense in our foreign policy.

A

Yeah also kind of conveys that it's New York into the overwrite it right. One of her requests, aces, which kind of conveys then yeah I, don't I, don't have all of the name. Location seems fine.

A

But I think that so right now we say defaults to one one: what happens if he really ever change this to tool, or is this more like? Well, the bed is too broken. If you do that, the thought.

D

A

You if he defaults all right we're.

C

Leaving okay, so we're not talking about this property in particular, but we're talking about what is the deep? What is the default? The initial value when I create an HTTP request message and we, like I, said we changed it last year to oh. We had already changed it three years ago for UAP, because we didn't have this property on HP client and we wanted a mechanism that would light up what we had done for you, ap in the wininet stack and the WinRT stack, which was oh. We wanted the majority of customers building apps.

C

You know for the Windows Store app store when they used the winner when they used the system that HP client, which used the WinRT api's. We wanted them to have an instant light up for, oh because that was deemed the the right choice, and so that's motivated the Oh default or a request message, and then we generalize that last year, as we were doing, h-2b to work for the rest of the guttacore stack.

C

Now we're kind of reverting part of that decision in the HP request message to go back to 1/1, but but we'll still leave the if death you ap for 200, because we didn't want to touch that. Okay and that's kind of how the history of the wine wine versus 2l default goes back and forth.

A

Got this is another thing you have to think of other than default to change between flat phone. So that's thing that thing that see yeah.

C

And also we're talking about now, you know we're releasing. This is for net core 300, it's a major version. We thought like. If we're going to do these kind of things. This is the the better time to do it.

C

When you change a major version and like I said by introducing this property, you know, in addition to us changing the HTTP request message default back to one one: by introducing this property, it's a softer landing for anybody that needs HTTP, client being 200 if they want to use this property on their instance of HTTP client and said it the tool. So it's kind of the right the right way to migrate people to 300 and give us a clean way of making me to be to work for the G RPC scenarios and everything else. The.

A

Other one guys asking like heavy thought of where it should be three slash quick, like is this like I, don't know how far the keys listen, but we even if they should be find in that world, it would be different.

C

Even you might be.

H

I think we we would our ideal would be. We would still use a should, be client and we would have yet another implementation in socket HP handler. That's supported, no quick yeah. That would be the ideal. We haven't actually implemented a means to know whether there's anything that would prevent that.

H

Although it's the the surface area is such that it's hard to see something that would be problematic whether we would change the default request version to ever be 3 instead of 1 1 I, don't know, I mean what we've seen with 200 is hb2 has been out for how long, and still only as a minority of the web supports it and there's lots of differences in implementation, lots of corner cases, I.

F

Think could be a long.

H

Time before we felt comfortable saying everyone has high enough quality on on quick that were HD 3, rather that we're going to you know effectively make it force it upon everyone yeah, but.

C

With this property being added to HP client, we have a very quick mechanism for any other stack consuming court to change. You know.

A

C

Them and then the rest of their code will does work at I. Think.

A

The person was just asking: have you thought about this in API design, I think it's one of two things, but either will just be another HTTP version that you can set, in which case the API will work, or it should be client, as maybe I doesn't work, in which case it does yeah.

C

I mean our goal is definitely the client should work. The same I mean all of these new transports. Like you know, hb2 and HP 3. They really don't change the fundamental like way that it still h2p 1:1 under the covers the really the messaging works and the headers there's optimizations, there's compression, etc. You know and with HTTP 3 with quick.

C

You know: that's like UDP, instead of TCP, so there's this stuff at that lower networking layer, that's mostly optimizations, but the overall concepts about HTTP and methods and headers, and that's all the same and we believe the object model of HTTP client will still will still live throughout.

C

All of that, we may add little optimizations for things I'm, even in the gr PC world with me getting duplex streaming working or like with when you think about push notifications, there's all these little really expert-level things that are interesting and if we need to enhance the HP client API for those we will. But we think HP client is general-purpose enough, that it will. It will continue because the fundamental building blocks of what you're sending is aren't the same. They haven't changed.

C

Even with h-2b and they're not really changing with HP 3, so they're still headers, there's still methods there and that's all what HTTP is about.

A

A

Somebody says we have a quick way to implement, no pun attend so I think that's so they're both approve now so then it was like seven minutes left. Is there any quick one? We can look at.

G

The one with the comment in the name, the one with the come in the name: ok, thank you for stopping by yeah, nothing pocket another one I see what tsunami.

A

D

It's just adding value theorem. Oh yeah,.

A

Netlink a kid.

A

G

Actually, lineage suppose a couple of which we can basically give users of the images of those values that specifically mostly care about. Can a must-stop basically allows us to do? Subject sockets to like you know,.

D

In the description they said, non-compliant operating systems just thrown existing good exceptions is.

A

It is protocol family and ever similarly, basically the exact same anything not.

G

Exactly what I mean.

A

They give had top of them. They literally show the exact same to the members.

D

I have ipv6 monochord.

A

Api catalog they're literally the same content. It might be that they're subsets of an opponent through the two platforms, but the union of all the I don't know specific.

G

But they will most be the same. What if somebody I was in one unit or any different.

A

Well, it is what it is: I mean it. That's that's design of a genome that we currently have then adding it to both seem to isn't.

B

Where is protocol family used.

G

That's for no address.

H

Address family is used in, like all the socket API is but protocol family sources.

D

First I thought that that says protocol families never used I'm checking reference source as well. Well, they able to view that uh-huh so who family is used in or for proper reference source looks like some DNS lookup stuff.

H

At all, internal use, right, yeah,.

D

That's all internal! Why.

I

Do we have the in the middle, but we could be even.

D

But they stopped supporting, it was probably some windows come happening originally, as I swear, like windows actually has two different internal. Yes,.

G

Everyone did suck at a Linux. You have to bus like this community and now.

H

You know you don't pass protocol family.

G

H

Are but not that it's protocol family isn't one of them well,.

A

Seems agree that we should maybe defecate it's.

H

It address family socket, tighten protocol type.

A

But it does shipping system that socket since OH, so don't think it was coming from the competitor.

A

In any case, I think like that seems fine I mean like I don't have forms of the I mean in the protocols are called meddling packet in can, but.

I

It's not calm tan was, it called. Can all.

D

Caps, the name of this controller area network, and we should just call it control where you know, because we don't overly acronym things.

A

That seems fine with me, like I, don't care phone and there's a bunch of other a quota codes that seem abbreviations already so.

D

If I mean shouldn't, we name it the same thing that a like other C libraries have named it just a moment what its name is, but if c libraries call it just can well for one they probably naming it can and all caps, and we would say, don't have three capital letters, because I'm sure, like I I, think that this is just a the name of its control area network or you see the cam and all caps.

D

Or you see the Pascal case for tan, and if you know that you're looking for you would find it.

F

D

Don't want it, you don't see it, but if it just says like address family that can like I can certainly given that that is a word in English. That means this is a possibility. I can see a lot of people being like, oh, maybe that's a Discover or something so. The fact that that works out to be a word is yeah, so I think it would just call it controller area network, and if somebody doesn't like that word being long, they can they can make their own pasta. I mean.

A

Looking at the other entries I think either one would work whether I think they're there. The concern that is actually legal, English word is totally put it there. But, let's just you, let me just tonight this in and if a fundament window again.

A

D

Entirely sold on packet, but I don't see a better name for it.

I

They're all packaged this one of these is more packet than the other. Well.

A

What is it called like? I mean it's certainly unexpected, but what does it mean? Is that an actual thing or is that? Is it just a convention or something this is an extra portable I, don't know so.

D

Let's posit a packet address family packet, the packet sockets are used to receive or send raw packets at the device driver level. So I don't know if we should call it raw packet or we can't call it packet socket because it's only a packet socket after you use the address to have a packet and I don't have a Linux packet, but it seems a very unfortunate name for whoever chose it. Then I guess I. Don't.

B

I'll tell them tonight.

I

Right now in Miami.

D

Didn't just improve my mother.

D

You I'm looking at windsock on HR now to try to figure out where this came.

A

From all right, so then, is there anything else, so we're ASIS, and it's so packaged to be renamed as you believe it is. This I can't.

D

Think of a better different name from it, but that's just from like 30 seconds of looking at a man, page I think it's an unfortunately generic name. If that may be simply the name of it but like searching for AF packet, you get you know. What's the difference between a half packet and a high net, they put two packets.

I

Alright, then, let's go with that. One.

A

Let me approve this guy and then I think we're done for today.

A

Yeah already I have.

H

Want one quick question: yeah.

B

H

It is it a breaking change to make a parameter to give up an existing parameter, a default value.

A

No, it could be a source breaking change if they're, unfortunately conflicting, but you have forgot to binary breaking change and I mean even the source side. If the source breaking change is rare enough, I wouldn't care, I would double take a call away. I can.

D

Select any mode, do you mean if there was a one parameter shorter over a little bit of it? That was.

A

If you have normal nose, it's definitely not sauce breaking changes, it's more like if you have existing other overloads and somebody could have passed.

A

Basically, the other one would know the like. It depends on what the prefix effectively is. So, if you know amid this weather, which one no wins out.

I

Alright, then,.

D

That's it for today, yeah I'm, going through the the windows, header files and it says protocol family same as that.