.Net Foundation Design Reviews, 22 Oct 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

00:00:00 - Approved: Attribute for minimal runtime impact during an unmanaged call https://github.com/dotnet/corefx/issues/40740#issuecomment-545076797
00:34:14 - Approved: Provide an `Unsafe.SkipInit` method to allow bypassing definite assignment rules. https://github.com/dotnet/corefx/issues/38585#issuecomment-545079783
00:41:27 - Needs Work: Make Type implement IEquatable(Type) https://github.com/dotnet/corefx/issues/41784#issuecomment-545080947
00:44:18 - Approved: Mark Assembly.CodeBase as obsolete https://github.com/dotnet/corefx/issues/41695#issuecomment-545084027
00:52:07 - Approved: Add HttpContent.SerializeToStreamAsync overload with cancellation https://github.com/dotnet/corefx/issues/9071#issuecomment-545093260
01:15:02 - Approved: SmtpClient.SendMailAsync overload with cancellation https://github.com/dotnet/corefx/issues/24475#issuecomment-545097103
01:25:16 - Needs Work: TryParse for email addresses https://github.com/dotnet/corefx/issues/25295#issuecomment-545099733
01:31:59 - Needs Work: Provide a zero-cost mechanism for going between System.Numerics.Vector and System.Runtime.Intrinsics.Vector types https://github.com/dotnet/corefx/issues/36379#issuecomment-545102767
01:39:11 - Approved: Add new overloads with ReadOnlySpan to System.Net.NetworkInformation.PhysicalAddress https://github.com/dotnet/corefx/issues/29780#issuecomment-545105286

A

A

A

But if you scroll down a little bit more, there is a list of here are some terrible impacts that can occur. If you use this thing correctly, I did validate how this impacts in the development cases mixed mode. Debugging is impacted negatively in certain scenarios. I've spoken to visual studio as well as archival in time diagnostic scheme they're. Okay, with this, this activity is not intended to be used by many. This.

B

Is a very advancement.

C

What happens if you get it wrong.

A

Okay, so let's say the attribute is applied to a function that makes it will actually have something bad any number. You can start the GC Oh, weird hiccups. You could crash absolutely possible if you do something wonky in there. If you try and transition back. So let's say one function takes in a callback, you store it and then you call another native function. You go down through it. Oh I stored that delegate and try to call back into the runtime it'll crash.

A

If we we will propagate an appropriate error, because we do already have mechanisms to check that, so we throw a well-defined exception but you'll around the process. That.

C

Guy said because.

A

He said, suppress just any transitions are not allowed to do this kind of thing right, yeah,.

C

D

A

To so, by entering it without a GC transition, the runtime thinks you're still in the runtime so calling back into the runtimes like what did you get.

E

A

Get have a left. How can I get back in I? Don't understand so that's kind of the.

F

And that's traffic per thread. Yes, okay, just wanted to make sure that you can be aware. You call in and then some other thread called back.

G

The question I asked think it was over here. What happens if there's an H fault, because what the things you say don't do is any type of mayo.

D

Yes, you I can't detect that there's no way to.

A

I, don't know how that would I, don't even know how.

D

I mean nothing would happen that I know of.

G

So what is the guidance around not performing any sort of walking or any type of.

A

A

Those are examp like basically syscalls that can head blocking I/o that, like read actual reading a file from this hitting.

D

A network or any.

D

You scroll down.

A

I've got I did get some feedback from people regarding.

A

Right there, the people method that actually those are really you know. Five thank.

D

A

You're familiar from the runtime side of things, alcohol, this is basically identity, identical to would have call on the run, so anything that you couldn't.

D

Do it enough call.

H

Is anyone other than system private corner basement.

A

Yes, they can't they would like to, as I said, let's say, for instance, they want to it. There is some sort of very specific inline assembly that they want. Does it's doing something they have no idea what it would be. That's with it. It's very. It is very limited kind of what you're getting at there are some win32 api s that I think Germany was found that you would be interested in in consuming Levi's mentioned some things about calling it IC new or ETF.

H

Work because I was gonna start with, should this just be in another property on yellow it only works with the ll. Importantly like this is very advanced. We want to hide.

I

It okay, if likely, will only work.

F

With in the future, yawns already called out, we need to make sure this works with yes and I was actually about.

J

To ask: why doesn't say how to read targets, method, a delegate and then so.

H

Then the continue is, should this be the declarant attribute in this namespace and when you hit the namespace qualified name, you get the behavior and we never make this public. So it's the. If this is you know, one can reason about it. Who doesn't work in this building and don't add it to public API but add the support in the runtime. If it's, the average person can reason about how to use it and maybe in public, and if it's this is a super common thing, then it should be on yellow or yellow important, but.

A

What is that yellow part would make? So that's like a pseudo attribute right, so it's actually built into the metadata. So it's not really an attribute yeah, so it would require changing so that that's me, I would love to do that. Absolutely good! There are bits available.

A

Changing metadata, as probably imagine is a non-trivial exercise, so would love to do it, but nobody there's not much appetite for that. So an attribute was decided upon.

A

Function, pointers is the other side of this. How fun computers are gonna, be represented, isn't obvious and it may, and a delegate implies a lot more than what the function pointer would be. There's no way to put an attribute on natural function, pointer as I understand it right.

J

Now their proposal, not the delegate concept, so the scenario that I'm thinking of is in three: oh, we introduced the native library API, which allows you to basically load library and then get a bunch of function. Pointers out of it. Is it possible for me to say I want to I, have a function pointer, I want to be invoked. I also want to suppress QC transitions during this camera.

J

A

Comment drop, this does not. This does not impact palm Interop. At present, there is, as you can see, there's no reason that you couldn't put this on a phone and clumb control off function. Oh so when I say.

F

Come and drop I'm not talking about and I'm talking about. You manually declare a be table with in pointers that you then marshal to delegate and then vote that way, which is the.

E

A like sharp DX.

F

A

That well, it would be so bright, so it would be by a call I and the problem of the call. So delegates, I, don't know I, don't believe we're going to be able to make that work, because there's a lot of implications. There's a lot of there's a lot of things that are implied in that delegate. You can't necessarily peer through what will be you are calling, so that's very difficult, but the function pointer proposal that is being pushed out.

A

There is no way at present to add an entity, because if there's no concept there so there's been like you want like a mod wreck or a mod off on those things, how do they get projected so attributes wouldn't even make any sense in that case. So solving that particular problem is something that we would have to defer. We can't really do that. So it's this is very much tailored to the deal import. So it's set for delegates. Can we not just.

F

Say it only applies to runs and also have a managed function all attribute on them.

E

That allows you to specify.

F

The calling convention delegate.

F

It's a unmanaged function for every ya system. Runtime you drop services, I manage and it allows you to specify preserve, saying : convention and everything else for the purposes of a marshal function. Pointer, raffinate, elegant.

J

It's basically deal a little important yeah for now, but for forget, Bukka under us.

A

I see I, guess I I guess I would have to look at. It actually know more about it. But how would but that wouldn't, wouldn't necessarily address the deal or enforce enter know.

J

But sit the same afternoon. Come oh yeah, so we'll be attribute to target, stop method or attribute targets about Delaware and then.

F

You could likewise have a comment like this is.

J

F

Applies to DLL import. You could also like. Why say this only applies to delegates with unmanned dysfunction.

A

I, okay I would have to look at that other algae and but okay. That seems soon right now. I, don't know how that translates to the function pointer proposal. That's there's something that that's something that we're gonna have to look at later on how this actually gets. How that actually gets marked I would imagine the compiler thing would need to do MOT off that.

F

Way, it doesn't impact overload, but we can talk to Jared about that, but.

A

From an implementation standpoint that is handled it's just it probably wouldn't be this happy worth it was you don't know where we go like.

H

I assume that the code describing this is looking for. That should be my name and not by I like ye, which so then it's the if this is expected to be extremely rare, is putting it in the public surface area setting people up for failure, at which point B you have to declare this inside your own assembly is just another hurdle that the people who and use it correctly are willing to accept. Therefore, it shouldn't be public, API I, don't.

A

Know how that I, don't necessarily know how that works from a library perspective, mainly because if you declare it in a library that you're consuming- and you also declare out what is the other semantics they were, they were policy.

C

I think we have cases where this happens, but basically the guidance would be they would have to copy and paste effectively actually good and make it internal. Well.

F

That would also mean that the e/m, during type loading now has to check every attribute, rather than just looking for attributes that existed in private Cortland was.

K

F

By me, well, it's doing fine by name based on corland.

K

There's some systems in Portland right, I, don't know, I, don't believe that that's true no.

A

K

Not the way the decision.

F

It just looks for every attribute attached to that seems expensive for the type of learner and yawns previously push back.

K

Against, what's doing that, I mean that's how every light-up feature works, those it does.

F

It works backwards. That's how the c-sharp compiler works. That's not how the gist been working. If Jaden explicitly like the intrinsic attribute, you can't find it anywhere. It has to be an intrinsic attribute in correlate because type look up on every activity, every assembly anywhere. It's expected.

K

Well, you wouldn't do that, but you will just look at all that it's about plug to the method and compare by right, which is expensive to do because you're doing the string comparison rather than it does this.

C

Right but it's, but you can't catch it once you have it right and.

A

This is only yeah I agree with everything that you're you're, saying Karen, I, guess. The thing that I'm I guess I would mention is that that is a one-time. It cost like I'm, going to be doing this anyways. Usually a PM book is going to be expensive and so basically sacrificing putting yes a little more time on that lookup. But it's going to be faster at the steady state level. So that's great feedback, that's a completely legitimate right man I mean.

C

My concern is well, but I think these experience might be even thank you I would tell what's they would say in general, I would prefer having centralized API sort of problematic rather than distributed, problematic, API nice, because then you can also probe for them more easily right like it.

C

If people like copy and paste the offending stuff everywhere, then I think we're in a worse world I mean the question really is that do you need anybody outside of koala bright now, because I mean to do to Jeremy's point right if we like, once we like I, mean generally speaking, wimpy features like this, where you kind of want to have people using it for a while to actually get a sense for the you know. Does it work like what are the caveats? How do we document this, and especially, did we already brought up function?

C

Pointers which unclear how this would work at all, maybe there's it once we get closer with function, pointers in C, sharp 9 or whatever the version of ends up being. Maybe then it's better time to make it public, like I mean, do you need an API right now, because you already know of an excellent consumer. We.

A

So external isn't as a like stuff like well.

G

L

Had a realistic.

A

To call it probably yeah I, don't know where he was or where Germany is I thought.

F

We had query performance frequency, very performance counter which are on diagnostic stopwatch, which are not in Cortland today,.

J

The private core, yeah I, think I would believe there would be places outside of that.

A

Yeah, like all the more effects there.

J

Will be places we didn't want to do it, but core FX is special because we can reference these APRs and if it's it's their homeland.

H

You know we could have it public and or a little bit, not in the rest that you know, but the but.

C

Do we compiling in school I thought we compiled you.

H

Lazy to know you if your inbox yeah.

J

But that's why I said like I: that's why I don't see but.

H

Anything that person.

J

H

Extra set can always just declare the attribute itself internally. It's.

F

Also not uncommon to want to write your own, for example, or a performance counter and frequency api's, because up until three, oh, the implementation from system stopwatch wasn't the most efficient, and so there people would likely want to be able to use this in the same rap.

H

But if my point is, if we don't expect, very many people are gonna, do it and all the people are gonna? Do it are hellaciously smart people and it's being internal to zero Hospital.

J

So I mean there is. There is discussion on the original issue about games, scripting engines potentially using this, but I don't have enough unity.

H

J

Had is, would they be interested in this or in the hall I work, that's being done or both I think both this has been this.

A

Is one of those things where it's people had talked about the actual idea? Because it is an F Hall.

A

So itself has already said that they are waiting.

B

Diplomat because they're interesting that they're they have customers as well right. So that is a that is.

J

A customer, especially if they would, as if, if they would just go slowly, to call I instead of this, then that's less of an organ for adding this. But if you think they're going to support but they're going to want both yeah.

A

Because Kali doesn't Kali does not stop the frame that school is going to put that on there right. This is further optimization. On top of the call you know, I could I.

H

Could see the extra those money listen like if we think it needs to be a public hearing like this is a scary thing is just.

J

Absolutely spirits to big long scary and this.

H

Is the we can leave it at that, but.

J

It's it is under Interop services, which already defines TLO import, which would self is scary, so scary, in the sense that, like how many times have you gotten DLL import along in their life. Okay, this seems much less clear in usage compared to deal with.

G

Look I really wish.

J

That we would just take the the ACMA hit and htlm for club. We.

C

My other question is: do we have other cases license applying to places like is, and do we have any other custom attributes that that's special? That basically indicates the runtime special Marshman behavior when you call into a be in vogue.

A

Well, there's I customer for it's an interface that you can implement on how the more for many.

C

Like literally custom attributes so basically and the other areas are, there are other things like it with referencing or referring. The metadata would be.

F

Sensitive library loader to address that and that didn't necessarily need it, I think the big thing that isn't necessarily addressed today, that is, that has the embedded metadata today, but where we need to expand, it is calling conventions and that's the long-standing issue and I think John has had existing ideas about that. Sorry.

J

Can you guys hear me? Yes, I've heard some people express the desire for this to be on yellow import and also saying the teal import seems safer, but I question that I mean DLL import. You can get so many things wrong with it and maybe not GC starvation, but you can cause runtime termination. You can cause stack overflow, you can call it. You know, data corruption.

J

You can cause all these things super easily just by having a type wrong in your signature and putting this as an attribute on as a property on DLL or what I might be an actually make it more prominent someone's gonna type DLL import, open, paren and it's gonna immediately be one of the options for them to add, whereas you have to kind of think to go and add this attribute in addition to the other. So it's not in-your-face making it more likely for you to use an accident.

L

J

What do you love? Poor people often will go through and say like for each thing, they'll go and figure out what they want to set it to, and this is one of the other options. Oh I need to set that to true or I guess: okay,.

C

Yeah I mean to me there's a likelihood of getting it wrong or right, it's more like avoiding like basically having like Franken designs, just because we can't do the right thing. So was just one whether there's other instances or we just didn't- want to bite the bullet of revving the metadata, because if we have enough for them and we batch them up, then maybe that's not a big of a deal but like like for just one thing. It doesn't make sense to just do the extraordinary effort to like put another bit in metadata right now.

C

A

Don't know of anything else, other than mentions I mean.

C

The other question is: what do we do? Move on, manage rob of sort before all function, pointers like, but that's but.

F

That's the question: that's the general question of how do you attach attributes to function, pointers and the answer is probably going to be mod or some or some metadata in private implementation, details and that's future or and but if it's going to be an attribute, it would probably end up being you know just attaching this happy yeah.

A

Thing dislike very specifically, tells me exactly what we're doing so, if you scroll up a little, there are two proposals that I a little bit more time right bit, give a little bit different take on it. Basically, what are you trying to convey rather than what you want, and that is the trivial and managed method attribute or the limited on management as far as naming conventions, because this one tells you this is what we're doing, not necessarily what you're trying to so that's another consideration. Oh whatever.

H

The first version like if we called this trivial unmanaged method attribute- and we found like three things- to spoil into it for five release and anything beyond five. We would be afraid to add to it. So we should just be I, think, okay, clarity of suppressing GC and then, whatever other five things we come up with with like fine just echo, your head, I mean.

C

Four losses: you standpoint I, like the idea that you have a feature that you can evolve over time or go on having to expose more API surface but I think for for mashing in particular. Do you honestly think that there's other things in the one time that we could do like you know? Let you optimize for stuff that we would turn this on. Just because you said all this is a trivial API, or would you rather have people like specifically say?

C

Oh I know what the trade-offs are long, what the limitations are and I'd specifically tether. One time don't do this. Do this? Don't do that? I I that's.

A

Why I may be the way that I did I? Don't like the fact that I didn't want to hide it behind something, but those were other options that yeah so I think if we think, if there's an application.

J

To making public the first easy transition sounds like a reasonable name, and if we, if we did want to make this work with unmanaged function, pointer routes with you in the future, we would just change, attribute targets to include delegate as well, and we don't have to bring this back to the review.

J

But what? If it would do what? So, if you scroll down yeah, it has attributes targets method right now, and that says, if we wanted to make this attribute work with unmanaged function, pointer attribute as well, then we could have attribute target stock method or attribute targets delegate. We could just make that change without bringing it back to radio, but.

C

I, don't think they're considered delegates no.

F

No, this is for delegates marked with unmanaged function, foreign or attribute dos, which is respected during marshal dot Guinea, yet delegate policy.

M

F

Is separate from actual function, pointer.

J

Does also means it's it's the way that today you do be invoked. Okay,.

C

Let's ignore function, one is from quorum long and.

J

Then, but what I'm saying is I'm saying, there's very likely to be a change. That's going to require that attribute to change. Yes,.

C

All I'm saying is like if, okay, let's Dead Sea Cadets ignore function one at the moment this for the I managed function pointer with it already exists in delegates. Should we enable this thing? Yes or no I.

M

J

Think it doesn't make sense for to work with DLL import only and not also on manage punching pointer attribute, because everything else that we have worked with I think is. We need to invented that could actually.

H

Work in the code actually be in the field.

I

Would be a super scary.

H

Thing, if we put the attribute here and didn't have backing support. Oh.

I

Yeah then, a few down version.

H

Turned it on because now things that never had this behavior started.

J

Right what I'm, what I'm, what I'm getting that is it's it's! This would be unique and that we would be introducing a new public API yeah in the purports to be symmetric. Good only works a lot. Yeah I mean I, think that if, if.

H

The my opinion, if this can be proven to work on that thing as well and the features button adding the delegate view, is a.

A

If, let's call it telmo okay well, there is, there is a let's say, another caveat to the delegate funds. This. Doesn't this isn't a.

A

This could cause like a propagation of those types of particular about delegates. So let's say I declare my fee in bulk like this and then I have a delegate, ID name and then I assign my peon book to that. I can't do that an instance of that delegate. I can't peer through my delegate and see that I have assigned at a p.m. book from, and so from that perspective he would require the propagation of this GC attribute, and so you would need to put it on your doubt.

A

You need to put on your team vote put on your delegate, at which point, if you didn't put it on your vote now, your delegate says I'm going to be pointing at this and you can assign any DLL import to it and I will remove it at which point you're kind of hiding the actual implementation, because.

L

A

Have a delegate tight well, is it any type any PM any DLL import that I assigned to it now, I'm trumping, the actual implementation, which is love so.

J

Your experience might be different, but in my experience when people declare a delegate type to represent a P info, that's not known until runtime, like they really only ever assigned one thing in one thing: yeah.

A

That's true, but in this case you could assign any implementation. Well, you don't necessarily believe.

F

A

F

A

F

Way it works in practice. Is you don't assign a peon vote to a delegate? You have a function that returns an int pointer and you call Marshall it delegate for function pointer type of T, where that T is a delegate, that's been nurtured, a manage function, corner and that's great practice, beginning at.

A

Is you assign a peon book? That's a manage call to the manage stuff that goes to the PMO agreed, but you are still hiding what I can assign to and you are give you are applying semantics to.

J

A indeterminate thing: it's not any different than declaring like a calling convention on the delegate itself. You.

A

Can assign anything to that's fair, that's actually fair, it's just I guess what I'm getting at is you're stepping away from the actual I've declared. This I know exactly how this thing works. I know exactly this implementation by putting it on a delegate, you're stepping weasel I.

F

Guess, I guess there's a little Turnitin, you could say, extend Marshall, get delegate for function pointer to take a pool, suppress GC attribute. Since that's what is that? What is what actually does the suppression and the generation of the stuff? That's.

H

What I would say, probably should be fine flags, enum.

A

But taking it at that, integer at the end of a I would be far more yeah yeah.

F

But we could probably discuss that more after and we can open an issue.

L

F

From this and tag y'all and Aaron, and everyone on how should we impress it for this scenario? Wouldn't that's.

J

As long as the scenarios at table, whatever isn't.

A

So with the native library, because that word, that meant: that's it right that word: okay, so you're not talking you're, not talking about the function! Pointers you're, really just talking about the delegates coming from Marshall dot; yes right, because that's how you do.

F

Function, pointers today without I'll be writing.

J

All the library call get Cracow address, call Marshall getting delegate.

B

Yeah I would be more inclined to say that should be.

H

I would say that that one we even if we think, even if there's only ever one it defined in it, that that seems very likely on this boolean will get more options in the future. Say binary though I would have to say it.

C

Which Bridget bishop of why they are talking about? What does the building here? If.

A

We go down the case so, instead of putting the attribute on the delegate, we were to put it on the API that gives you.

H

Yeah, so it's another, it's it's a downstream of this and it's a future they kept ago. You know it's.

J

Propagating in a different way, but it directly falls out of the south.

A

If they're a list, unless you can come up with it like, if there's a there's, a pressing scenario right now,.

J

Well, I mean that's probably going to be the game scenario.

A

J

Different things, probably based on what LS your life you.

D

L

Mean that's the entire reason that people are in that first.

J

Place because we need to learn different dll's based on right, what would function in us would function with function planners. Does that change that.

A

D

C

C

But I'm hearing as we're not holding off then well I can prove the api's. Is you believe there is reason for this API to be public.

N

Just think is there earlier.

N

C

Oh I, don't I also take their compiler guys on the sign.

A

A

I would this: how would I check this okay I, don't know how to exit idly.

C

You are in the happy position to get to touch the core Simmons, which means you have to do this dance. Where you add the API and implementation, then at the API and the EPO is already doing yeah.

F

So is it used to be public in Portland that needs to flow to core effects? You know to update the relevant reference assembly that contains scale I. Think it's just a long time.

F

Once that's been updated in core effects, you have to wait for it to flow back to core CR and all of that, the core set up and yeah.

J

Well, you basically once when you do it on the new people, it's really easy. So it's already, it's already checked into.

M

J

Which means a core effects already has a build with us. So all we have to do this. It was eight days ago ci has been flowing. My last.

J

If you, if you need assistance, just ping anyhow, so make it back so.

A

We're okay, so the other thing, then, is the documentation which I just put up there, because that's how I yeah that's how.

I

C

That yeah, okay.

A

C

Ends up happening is that becomes the templates for word of a dog writer wants to tackle that. But, given that wrong, just retired I'm not clear who the next person will be the dead.

A

I have been getting a.

H

Bunch of after wrong this will go and then because it's in runtime I'll get in and I'll go. It looks like the feature or the person who wrote the feature or thought so approved. So this will be the public document. Yeah.

C

Then you're done you're welcome to stick around, but thank you.

J

You can watch this rathole and Tanner's. Oh.

H

J

I

H

I think that's new record slow.

C

What's a one-time change, I don't get too bad.

H

It's GC change. Oh.

C

All right so then,.

F

We should look at a three eight five, eight five.

H

I had to say no, no, no.

C

This is really important because otherwise they would rather go old extremists, because we keep not doing that. Well,.

F

The the first oldest ones are ones that they're also mine, and we keep on pushing them back. No.

C

The in PDR I don't care bro because I'm something that.

F

Was actually unblocked now so.

J

This one three, eight five, eight five, we had already tentatively approved a pending customer scenarios and we have customers there. Yes,.

F

Young had given the scenarios of what we're going to use in climb.

J

Like we should just bite the bullet and say yes.

C

I can literally paste in what you just said in the movie yeah.

J

Like I would be fine with us.

F

Basically, the only question is: should we make this where T is a managed.

F

Technically so, if it's a struct and it's not unmanaged, the JIT will still zero out any of the reference fields. So I managed is much more accurate and what this will action we previously said: don't do it, because it only works on c-sharp like seven three and later, but now that we're not putting it guys back to older targets and we're saying new or language versions versions are only officially supported on or anyways.

H

Yeah, let's make it more constrained.

J

Yeah, especially since you would never use this with Erics, like you always know, the Capri type going in there right and.

I

And you need to because you need to know that, so this is just. This is the thing that you call that the Jets gonna just be like this is stupid and believe it right.

F

Yeah, the methods completely empty is literally just a rat instruction, so it should be fully alighted and dropped so.

I

Just continue confuses the compiler; it makes it PPA stay right. There, I'm.

C

Confused at how to use the safe here.

F

Unsafe not skipping it out. I struct.

H

Definite assignment to satisfied without you assigning default and paying the bureau cost before the field, which.

F

Avoids multiple rights for union types where you're using field offset, but.

C

That means it all need works for locals, but it doesn't know how much thing of right well.

F

Technically, it also works for fields of structs in making the JIT thing or in making the. Why only think, it's been definitely a sign, so.

L

You could do out.

F

This um or the union type in the constructor, if you've got a method that says I'm only initializing. This particular tag of.

H

That's clever, but that's the intent. Yeah use the unmanaged constraint.

C

What does the mean again? It means that the thing doesn't contain any reference stuff, yeah.

F

Unmanage means it contains no reference types, but.

C

It can be instructive industry.

F

Yeah, it can be any structure pending any other structs car in float.

F

C

C

Does I manage to imply scrap Dorothy is a struct, it requires, but you can just say, Betty cool on there.

C

C

What happens if we ever find out that we wanted for regular struck with reference types? Is it that we can't do it anyway so or it's the we.

J

Just don't want to what would happen is it would get zero initialized, so this method would be pointless.

F

And also as I called out so there's the skip locals in it attribute, which is coming, which avoids zeroing a struct in general, whereas this one is meant for union types where the jet currently thinks it has to write each field. Even though fields overlap in memory, you can't overlap reference types or generics with field offset, so this doesn't really apply to those.

C

This be more useful for a medium.

F

C

F

You have a union type with right offset. We might have four or five fields that all overlap at offset zero. Currently, the jet and the c-sharp compiler think it has to write. Each field is zero. It out, and so you end up- writing bytes zero through four five times to 0. It is 0 at each time. This allows you to avoid that Oh.

F

O

F

To only write the fields you're actually using because it's a because you know how.

J

You utilizing the Union. We had spoken with the trip folks and we were wondering like if, if they would just be able to make this change directly so that we wouldn't necessarily need this API, but apparently doing that work is untenable.

I

What's the simplex you shot, maybe should have a motion for they don't tell us to have to understand.

F

Offsets and then, if they can't really because when you get in the cross architecture and honor.

C

Should not, this should not directly understand the layouts but like having a first stuff representation, for this is either this bad should be pretty straightforward to.

J

All Senate, but we like we could. We could review the conversation for last time. We review this because we got all these same discussions yeah and the conclusion was tentatively approved. Pending the customers, names yeah.

C

I mean I yeah I mean we're not getting a c-sharp subjects anytime soon. For this, though, but even if we do it's just one method, it's no! No, no.

C

Right so then done was there four.

H

One seven eight.

N

The motive reveal might only take 40.

H

O

C

You make that movement I quit who's, that and.

H

Y'all, you marked it ready for you, so let go.

J

Oh yeah, it was strange that it didn't do this already. I think it was because I want to say the reason for this is because there is an actual hierarchy inside reflection, so type implement or type inherits like member inherits, mom and wall, and do you make everything all the way up and down the chain why equatable of tea? So just member info become my equatable of member.

J

N

Baby, the moraine yeah because I stick with it.

J

F

J

Info become my equatable number, like that's stuff, like that.

C

I mean in general, you would we.

M

Implement the interface in the hierarchy, because otherwise you have weird semantics: I've, never.

J

Seen a equatable on.

J

That's the thing that makes it that's not a reason not to do it. It's just hey.

H

There a question so bounce this back to whoever under question, which is more important than me. Yeah, apparently it was gonna, be easy, either easy, yes or easy.

C

I mean reflection realistically has not seen a lot of the generic actually.

J

Member info doesn't even have an equals member info, so it wouldn't be I cleanable number info. So at that point, I, don't care like make it I make this highway toll type money, Schumer.

C

What about method info, like all the other week, no, never notice it I! Think it's a fair question of like what happens there. Nothing else, I, assume.

H

This question is that I already has like that equal time and details. So what this this duck type needs the interface. Can we just declare it as me.

H

Only eight guys have done that would be.

C

You time works way, you're hitting the roads now other times in the memory.

N

A fight that equals.

H

I said it's X no bubble for a nullable tight, so I don't know what that does with the interface, not.

L

J

I guess we said a few.

F

But Yohn also commented asking why we did I manage. He thinks we should just keep it. This truck.

F

Why he says it makes the method less useful for no good reason, nothing like solemn, safest, constrain that way.

H

I, don't see anybody needing to reason getting to use it, but we can change the derrick pilot. You want to go inside.

H

My answer is unsafe should be as unusable as possible, so yeah.

C

C

Yacht is a very simple philosophy like make things useful or make them unsavory ones here, announce a.

L

Which is breaking change to remove the constraint yeah.

C

I was thinking about that. It's not a breaking change, because you can't pull what the constraint you basically like. If you call the method adjustment, your generic yourself, it just means you have to declare yourself as unmanaged and then this way, if we remove it, while you just overly constraint on the call side now but I expect I mean there's a deck. It's not a breaking change to move a constraint on a non-virtual method like that.

F

It's horse-breaking.

I

Managed and yeah.

F

I don't know this teacher. This is your compiler in some scenarios omits unmanaged as a maghreb I. Don't that might just be virtual method. I'm gonna show it vegetables.

C

Because then I guess this way you.

H

All right come on. The point is that strums there are struck that you can pass it into that. It's yeah like, if we're not doing anything so like understanding me, this is having no effect on your type, is alright.

F

H

F

They only zero out the reference pipes and not all fields, in which case it would still be applicable to those types. If you had something where the value types were overlapped in the reference text for.

M

Or 165 approve all right, so we have not.

C

Obsolete this stuff, realistically so I think no the problems of observation for a moment. They, like just from a practical standpoint, anything wrong with doing that. I mean.

J

You have a clear replacement, API yeah.

C

I think it's the one-to-one replacement.

J

And co-pays it's just bad. Does it come back in the same format like an actual path as opposed to a final URL or something you.

C

Mean location, yeah location was always an actual path. Co-Pays was always the one that had the via URL because it handled things on that and other ways.

J

J

Or support living assemblies from your network isn't like I, have a manifest and I need to download a module over like HTTP or whatever a thing.

C

In this case, you just have the same semantics that you get from a similar form of ID array. Just on don't have a location, it's basically empty and like the only reason why we care about a location in this case was for security, so that we can either be know that it was loaded from the internet rather which we don't care about right, whatever in assembly, how you got it, we don't you.

C

H

Barry said Johnson.

C

Well, I mean obsoleting is more like Oracle I, think the the the the more problematic calm water ish abilities, remove, api's, which.

C

Very school things we can remove stuff, but.

J

I obsoleted trained our coffee in Rio and I mean these. The people actually and.

C

Still find nobody call this honor.

C

Only concern that I have a homeland how many people use it actually Oh.

C

It's 30% so that has been to begin warning but I think that's fine! Well,.

J

The one library, the 30% of applications, use one more yeah.

C

That's a fair question: I mean honestly, though, people do crazy stuff.

C

With co-pays, usually it's like we're. Locating things in the output phone all right, a.

J

30% of all dotnet applications they guarantee do not holla something like 30% of our developer audience. Timon yeah,.

O

O

C

Somebody's asking other nice standard rules from mark something is obsolete, has marking something on support, not a coin absolutely considered.

J

I know there were questions about how it affects standard.

C

Well, I, don't think he's asking for dinner, he's just saying standard, /, rule I, guess then, and as.

H

A practical they if we're marking something, is obviously we want that also mark these headed a browsable.

H

C

Me, the board will never never.

H

Advanced whatever or do we leave it, and then you write that code Basin, where you see it in intellisense, you call it minute cuz what.

J

What I would probably recommend is we go to the VST and say: can you change intellisense not to recommend there's an app or possibly just pretend they're rolling around another? That way we don't have to attribute person. Can.

C

Do that too, but I mean.

H

That's it for the practical thing: we've not done too many yet now.

C

I think this is a good example that we probably should I mean the problem of also leading things is that I think for members? I can probably be convinced that marking my data possible is not too bad. The problem nu billion mark types it gets annoying because if you still have to use them and then they're always hidden that sucks and then another question is inside of a region where you said you know, Park must suppress whatever the warning indicator. Falsely goes. Do I see this thing now or not. What.

J

Like that mean, this is probably rattling, but the recommendation that I would give to the vs team is like. If someone has used a member or a type somewhere in the project, its marked obsolete, show it everywhere, intellisense, because they're obviously still using it. If they haven't referenced it at a hole in the project. Id affirmative, it was interrogating.

H

We don't Brown that we don't mark anything right.

J

As editor brows, well, never is all dis intellisense anyway and yeah we control intelligence. We can always make this better. You like without attributes.

J

Yeah all right, it's a it's! A good consideration.

C

Didn't think about, like I, had all my list to make up solution better for Fito I already have a proposal written up and actually intelligence interests. There is an interesting issue, but, like yeah, we from our point of view that the biggest problem we have solution is that it's pretty heavy hammer. You can't really target Lille press it today, which makes it annoying yeah.

J

There are also like various degrees of obsolete right. Like there's observation is and like this is an API that we want to steer people away from yeah versus this is an API that will not work in that. We really need to get people off of it. I mean.

C

In general, like obsoleting, like particular methods, is relatively straightforward right because the only people that are affected are the actual people using this one API right way, it really starts to suck is then you want to obsolete technologies like let's say WCF or you know.

C

Let's say you know, binary civilization right, which is the upset of tides right, and some of them are used in 101 scenario, someone that were only used in this and then you when you mark the whole thing, then the problem is now you get it because you have to you know, perhaps these type 3 projects like this- that's not great all right. So let me refresh this list here.

C

All right, let's it's trying to go three years, knowing that we have um not easily because they're gonna give up on my screen just start reading.

J

Numbers 173.

C

But what this one year, that's the oldest one account I have I'm Steven. You know, I still, love us know and don't tackle this one because TV is gone say, do you need Steven? Well, he found it so, but can you just do it? But if you can, if you want to talk to that sure sure.

L

So this issue basically blocks our ability to do to plug streaming and canceling properly. So this method HP sleep, Opie streams between, but because we don't have a public API to pass the consolation through the consolation doesn't work.

L

It excites so if we, basically, if you needed some hex or 300, make it work for the built-in streams, but if you basically derive your custom content scheme, basically fast consolation because there's a step very basic use, the consummation token, the proposal is just to add consolation, so we can propagate it properly. All the way down to the strings from HDPE makes.

C

Complete sense so.

H

The guidance that we have for a sink is cancellation, token should be defaulted.

H

And then I would take the existing abstract API and make it virtual to all the other women and make the virtual imitation stuff.

J

Like that, I've never been a fan of because now they're.

H

Going we don't.

J

Want people to see if they should override the one that doesn't think actually no.

H

No I wrong: it I.

J

Understand that but I'm thinking back to the same problems we had when we introduced data annotations, like back in the forest timeframe, where people didn't know which, over which method to override, because when they go to visual studio, say implement abstract class like it doesn't override any of them anymore, because nothing's, abstract anymore, and now all of a sudden, like you f5 your project, and it just fails it like it. It becomes very weird what so anybody who's already implemented.

H

It will get like they've already overridden the thing and that's what gets called and that's what it's yeah they'll have to, but leaving the existing method abstract. This means that somebody is always forced to write this defers from the other one yeah, which is a shame so.

J

We should just write that one.

H

For them and then the only thing that the baseball I mean the bass, one can do the weight. I mean it can do the reflection test. If did you derive this? Like? Did you derive the previous one? Okay, fine I'll call you and ignore the cancellation token, which is what we've done in most of the screen things. If each layer figures out which one is supposed to call.

C

So you basically but I, don't think we can implement the the abstract one is throwing because it will be pretty bad because when you force anybody to override, though for eternity no.

H

So the abstract one should just call the other one. The the old abstract becomes virtual call vanilla the new one that either is just implemented as throw or it's implemented, as if the other one is implemented at a level. That is not a should be content than invoke it, otherwise them yeah that won't make sense, but.

C

Logically, you want people to derive HP kind of moving forward to just all override the one that takes consolation to great so.

H

Make this one virtual editor browsable, you know whether that's advanced for never like.

C

Well, what's the benefit of making it opposable, hey.

H

You don't see it when you write override yeah, true.

C

But how is it different from calling another API where it's decoded a calling, the API seems fine? Oh, it's.

H

We don't want you to over run so it's protected and that that's not public. So it's the only time you would see this is when you write the word override on us, and so we want to tell you. This is not the one to override.

J

But I'm still thinking of scenario where, like a public public class, my HTTP content, :, HTTP content right click implement abstract class like the template now does not include everything that I actually need to implement. In order to be successful, you do.

H

That is if we make the new one extract like this convinces failure. You know this week this we should not feel we have to take the breaking change and move the app there make the abstract a translations open. We take the break and change and make the new thing the abstract, or we do what we've done for every other type and it's this is better than screen, because no one should be extending this. Is she be content?

H

I've extended was before, huh but this is seems way less likely to be extended in the stream and stream. We've had to do all these gymnastics in before the answer. Is you override the other one.

J

Well, would it make sense just to have a protected cancellation, token property that you could query from your rounded if you cared.

G

Real ugly does.

J

But means you don't you don't have the question of which one should I override.

H

Like it's gonna be very easy to lose it and flow analysis. We're gonna tell you like. If you override the abstractly, like you'd, have to take any analyzer tool that says: wait, you had a cancellation, took it in and call the thing. That's like a cancellation. Don't we need an asset like now. You're gonna need to teach it this particular type as they protect the property that you feel like.

C

Well, I mean like you could have an analyzer that basically does like in the same way that it tracks within a method. You could have one that that M Eliza's the symbol influences your overriding. Only.

H

One that you have to go like special cases, no.

C

You would just say you're overriding a method, but the the base type defines the same method with the cancellation talking with also virtually. You should override that one. Instead, like I mean we could handle this analyzer, but.

H

Unless we do but but if we do, if we leave this one abstract, then we you're forced to implement it, and then the analyzer tells you you should have overridden the other way. You.

C

Think I'm fine with the virtual one. What scares me a bit? Is this dance around them? How do we avoid typos right? That's.

J

Right, that's what data annotations.

H

J

H

As soon as someone implements I realize the stream a snake to take the cancellation token, it doesn't call the dance one I mean you know so any type that we build, we're going to override the good one, and now the virtual one or the one without cancellation token just forwards. The call to the one with cancellation token without token, and the vertical one doesn't need. A reflection is the only time you get. It is if you literally you're you're, not one of our types and you've extended. This, then we'll do the right thing.

H

That's the same reason. We said it was okay in strength because all the strings, we provide we over the thing and aborted their reflection, but.

C

Stream was complicated for the dreamers.

H

Name is complicated because it has gone through every evolution that we've ever.

C

Done at home, fermentation.

C

So let me get this right, so you're, basically saying the one that takes cancelation token is virtual. The base implementation would just call the other one yeah that makes the API merely usable, and then you would say in the existing one. We would make it virtual and we would say basically zero code. You know if to realize to scream a sink.

H

The that reflection code goes in the one that takes it so the new method we we won't make abstract, because you have.

C

A presentation, so the new method is virtual. It's.

H

Virtual implementation is if or go ask where the other one is defined if it's defined on this type thrown out implemented, exception else, and then the one that is Mb comes virtual and then calls the one with cancellation open up.

C

You think it's a good idea, I think.

N

It's what we do for free.

C

I mean that insistently you have a high chance of. Basically, if you want to pass it on the cancellation talking in my throat.

H

If you have an existing one, you've already overwritten the one that doesn't take it no.

C

I'm saying like I'm, an ace pin it right like I, take a resume, should be context into content and that I don't know. Yeah and I know safely called Co lines to stream a.

H

Second, cosmic installation: yes, because the if they didn't override that one, the default implementation is reflection, dispatch back to the other one, with the belief that it's not going to cause the cycle.

C

Singing the one that hates the cancellation token, the bass implementation. Yes, we'll call the other one well.

H

Ask reflection: where is this: where is this other method actually defined? Yes, if it is this tight thrown out of all that, I said, which means they didn't overwrite hydaburg, if they've overridden the one that doesn't think a cancellation token reflection will say, this is implemented on a derive type and then we'll call. We don't have a cheap.

J

Way right now we're in reflection to ask: where is this, but.

C

Again, it doesn't have to be cheaper, because, basically it is that you would override one of them right, but.

J

If you override the wrong one, because it again did.

H

We did industry dream is much more sensitive than this than a should, because.

J

I, don't you yeah I, where, where is their reflection, codon stream? If I've read that takes a span, you.

H

Taste, I'm pretty sure we did this this dream. I know we did it North a.

J

Know what that does is it. uh It says if I'm not overwritten them I, basically rent a buffer, read into the buffer and then copy the buffer into his fan, which is even slower. It's not reflection. Okay,.

H

Well, that one was capable of working so that I know we've done it in I know: we've done it in places when we've taken an FM thing in a virtual. You.

G

E

H

It's the sideways dancing, no I'd, like you seen that code before just right. So again, once we start having, if we have the analyzer that says, you've overridden the thing that didn't take cancellation, hoping you should have overridden the other one, then your core problem goes away underneath.

J

Like what we, what we still end up with, though, is we still end up in the end, with a type where you absolutely have to override some method, but the methods not abstract but and the types going to fail unless you override it.

H

J

That's that's not a tenable position for the framework itself to the end and.

H

We say no to this and we never read cancellation. Those are the only two options, because if we leave this one abstract and I mean you can boilerplate have to implement the other one. It's just gonna be.

J

A good I would honestly rather not do this, then then make people have great difficulty, sub-classified or I'll or I would rather say, like there's an H to be content to or something where we actually say, like no serialized to trance to scream a sink with cancellation. Token is the abstract method, and we make this the one that doesn't take a cancellation token sealed we're saying that's breaking K, no fur, it should be context 2 or it should be content to like we introduce a new type that actually hasn't right.

J

Sir, like that's, I would rather generate honestly, then that way, when, whichever one of the two you subclass like there is one true abstract method that you avoid I mean I. Think.

C

This approach doesn't fly my opinion because, basically I think the vast majority of pads like this will be implemented by basically what amounts to first parties, and it has way more callers than it has implementers right. They're, not exposing things for callers, because something new matters have trouble seems the wrong axis.

H

When you define a new type in the throws, you go oh now you write override and solve the probably the throw will tell you override this method. Thank because.

C

I mean like I, mean like: do we want our networking stack to properly handle cancellation? Yes, yeah. That means there's a few people that have to do extra work, because we just can't express it in the type system it's unfortunate, but also maybe that's an area where we can actually generalize it to a to an analyzer and say like.

C

Maybe we have an attribute that we can put on our honor on a on a method that basically these soft abstract every time you see somebody overriding it see somebody deriving from this and they don't override it. Add it to a quick. You know quick fix a list and say you should override this method. I would.

L

Also tend to a cleavage emote it, it seems to be rare for somebody to do it. I, don't think it's like mainstream pass and you seem like GRP see using it right. So I could say that most of the implementations would be, you know, comes parties or people actually know what I do and they sort of average users would be stuck with the process we actually use, rightly or ef-1, for content, and we have stream content as implementations and they work correctly.

L

But if it gets too complicated, maybe we do should pay for Stephen sort of a tin or I totally strong will be easily and apparently so.

C

I think I didn't honestly the the did like I didn't. Actually the proposal like it's. This is a very easy API from the point of view that all you want is passing and cancellation. What makes us complicated is the fact that it's virtual and we had to spawn them in other cases, in the framework where, when you add an overload to of to an to an abstract type, it's generally fine, because the idea is that people, you know that the quality is in the isn't the base.

C

But this is an example where you extend effectively what the what the core functions. You should have been right and that's something that right, it's just fundamentally harder to express in the framework. We don't have a good answer for this, but at the same time, I think like blocking these features seems the wrong.

H

C

Wrong ways that I know that we.

H

Did this before at least, is a framework of RSA pkcs1 key exchange people matter of it asks which.

H

I going to do magic or am I just going to call the right thing, and because we we changed the type hierarchy and the right answer, change and the base in the base method for the right answer grows. So if you had a classic the extended RSA, we broke yeah. If this was a type that everybody generally extended, then I might feel differently there. The notion of, like you, add a new thing and yeah. It doesn't work right out of the box.

H

But as long as we give a useful exception of hey, you realize for HTTP content, you have to actually write serialize to screen right.

J

So if, if very few people actually implement this, then why are we even doing a reflection, dance in the first place.

C

I mean we don't want, that's man.

J

No, no I mean why don't we just say the abstract method stays abstract the virtual method, delegates to the abstract method and throws away the cancellation trophy and because we control all the implementers are going to override off throwing away a cancellation, token yeah. But we control the implement.

H

Or so who cares? Okay, then, when we make it, if the cancellation token is cancelable, throw I'm sorry, this type doesn't support cancellation because we should not, but somebody who has a followed the abstract previously ignore the cancellation token, because they didn't know it.

C

Yeah, that's what it expressed earlier, but I don't like about this. Is that basically, you know create a reverse incentives to pass down cancellation tokens because, depending on what instance, you are passing, it might go up. Yeah.

J

But we do this all the time. One way: a stream stream, for instance, like the api's I, take a cancellation token, just throw it away if they have to call a PM, api's nothing's.

C

Throwing away it's fine, whatever, but I, think it's problematic is saying the API that takes the cancellation token if went over and throws an exception because then the problem is, if I know it's a lie by author or middleware in HP net. It has to decide whether I call this method. I'm like well. This API might throw so I'm less inclined to call it. I'd rather give the EPI cancellation token, and if it is able to forward it, it will forward it and if it doesn't well, then drops.

L

C

The floor, which is still better than throwing I what.

L

Do we do right now, right, you have a layer of error, you pass consolation to HTTP request and dump it on the floor, some in the middle, because we didn't have a Pio pasco.

J

Call all the implementations and we're going to do the right thing regardless and at that point, why even do a reflection, dance well,.

C

I wouldn't say you wouldn't say you control them all in the in the way that we ship them all at once right. So what I'm saying is that there's a very small set and it's effectively the first party, but it's it- might still be an open-ended set of we provide, but it's it's kind of like you know the the users by far outweigh the consumer at the producer right and so like not giving the consumers a feature, because the producers might actually want be able to do it fast enough. I think that's a no-go.

J

M

J

Give them the future say, add the API I have the virtual one, all the abstract one and then say no reflection, dance at all like if you, if you.

G

Want to honor that.

J

Fact leave the abstract one abstract. Oh now, you're saying everybody. The correct answer is always override book. Yes, there were a few implementations, so that's not a hard shot.

C

I mean in fairness, though, like the performance reason important to do is anyway right because there's you the reflection, dance, then that's probably not a beautiful position right there, better.

H

Getting like part of this I'm.

C

Time of the two, because, basically, how all like my hypothetical soft abstract thing would catch that and you would say: yeah you overwrite, both of them right, yeah, I'm, okay, with it.

J

It is, it is weird that I mean are what we've been justifying to ourselves as we're willing to make implementer each on floor hoops, yeah, I, think and if, if we're willing to make them jump for groups override those methods.

C

Would have been on the call he would have been able to participate in a conversation than me. Actually, I can try to begin.

C

All right, let me actually change my ruling. You should come to the main news.

J

Just going for simplicity,.

N

Leave it EPS transfer I mean.

H

There are many places that I wish that we would take in the wrap the bracelet change of marching something abstract like you, move platforms, override this method, yeah I agree and the implementation can say virtual. It would be nice if the reps to start forcing people like this is the right one. This is the wrong one. Stop like we can extreme. Hopefully, this fanboys are actual.

H

N

H

All right, this all comes back to virtue the party. Yes, the guidance for virtual is somebody.

F

Major version, and not only that it's two version jump so make all these changes this. This.

J

Is the thing that people don't necessarily understand about extensibility like when you design for one extensibility model, you can't really extend the extensive working model without introducing lots and lots of various if.

H

You are async, take a cancelation, even if you can't do anything with it that nothing. That's why our guidance is fake. It defaulted, so a caller can ignore it if they want to oh yeah. The only question remaining is, if we say we should take cancelation token. Okay, our guidance says cancellation, token should be a name cancellation, token, what's the big t and that it should be default, but that was assuming it was public I, don't know how we care I, don't know for care about protected.

H

If we're matching the same syntax there, then you do what I read like.

J

Says protected because this is meant to be some cool, because this is meant to be implemented like I, would imagine it should flow through visits.

H

The workhorse method, right so cancellation token, should be named cancellation token and it should equal or equal people so very timely.

H

We review the ASIC stuff like a month ago. This is this is old news now and Depot holders.

H

Or, as I also said, don't use default forever in every format months. So like we get to pick what your own you're.

J

Allowed to have conflicts.

H

I'm good with cognitive.

J

H

O

H

Is literally the only valid default for cancellation token it's default is the only compile time plug that you can put there.

F

All right can we go back to the unsafe, open it real, quick, so Yan says this should not be constrained, doesn't manage nothing else on the safe is constrained that way and the constraint just makes Matt the method less useful for no reason constraining it as struct is also unnecessary. The problem with unnecessary nice to have constraints is that it makes the method impossible to use in other generic code. That does not have to be, but.

J

Like when would you ever pass a generic argument to those.

F

Apparently Yan.

H

Has the near you I mean if you know what the PR the question is, is he talking practically or is he talking specifically because if it's specific, then I feel that if you're using this on an unknown team yeah something.

F

I asked: maybe we should look at addressing this by having like some 50, is value type, and now you can college an arithmetic, constrain destructon. His response was that would be nice. The number of things with a number of things would be implemented differently if this existed until this language feature exists. The API designed to optimize for lack of its existence, where's.

E

Their composition go to four one: nine, nine five, which is the P R, so you know I have and then you.

J

Know now that you're not on the yeah there, isn't it like there is no harm to removing it like no harm whatsoever to removing it. It's it chose, doesn't do what it purports to do. Well, it skips initialization, which means it.

E

Does not like well.

F

It's you're still gonna zero in debt. Well, you get to zero in it from skip locals a minute yeah, but that's yeah, that's the JIT doing entity or reference types, and so if there was already a handle value in there. Presumably this does nothing and therefore the previous value for me.

F

So if you did bring s equals new string skipping it s, you would still have the new string value in yes, I mean remove the constraint sure whatever I mean I'm Mike, it was yeah, I mean I, I I'm, not pretty gluing strong either with y'all has scenarios where he thinks it would be useful. Yeah I generally defer to his judgment on yeah.

J

I agree: I, just don't see those scenarios offhand because I'm I can't imagine anyone passing an unknown type in here.

J

F

We have exactly one API, that's constrained to where T is struct on unsafe today and.

L

That's unboxed.

F

Where I album fire is that constraint for Khalid.

F

J

And it sounds like this follows his overall philosophy about like why'd, you lock, it.

F

Yeah there's no safety security issues, we'll do the right thing. It'll.

M

Be clear, T will be completely unconstrained. Yeah, it will be unconstrained. It.

J

Will just be it's just, you should only use a say, VI or T as on it, because that's the only scenario, much of this API makes.

I

O

I updated our ruling. Thank you.

J

I, like that, one.

C

Anybody has any context of this. One who owns I think if it is no way, no more it'll, be all she gained. You have you read this. Do you know what this is.

L

Basically, very similar, there is a cent method which currently we don't have until we don't have a layout ask conservation in. So if you start Sonic email, there's no way how you can answer it essentially right.

C

Did we, like my understanding, was SMTV Club of all those ApS? If you consider legacy, and we don't want to touch it anymore, is that true or have we changed our mind on.

L

That economy item several classes. Seventy people sworn an HTTP listener was all wrong, but instead of counterpoint that, as we see, people are using them, be one make hydration easier and be talk about it with the group and we fell. This kinda makes sense yeah. It's not actually that difficult to do so.

L

The current plan is that we would use the new people to do it as a training exercise right. So basically, if this gets approved, it will get on whatever Colonel maintains a list for easy things to implement that we build Prague or somebody to chew on it and basically get it done. And since we have four people, we actually might spend more time, fixing bugs and sort of supporting the legacy classes and fixing up signals, and we had in a past because it wasn't priority. But now we can cannot afford it.

L

So we'll try to fix some things or 18 fortress.

C

Yeah I would say looking at the API services. The proposal makes sense to me if you think it's it's actually implementable without breaking in to jail, and you see value in doing that as a general exercise. The thing that makes sense anybody else in the room, I'd leave an opinion. I'm.

M

Looking through the code right now, it looks usable.

L

So since we fixed all the streams and the lower layers do actually so so the counselor she actually does work which I didn't in the past and I think it makes sense and to be useful.

L

C

But my only question: is the police only do it for sent mail and one for send a Singh.

J

They always think it's that's. What's in the proposal, no.

C

There's a there's also, another method called send async. It takes a mail message in an object or it takes effectively six individual values.

C

It was all to be done. Oh.

J

So send async mail message. Object is actually the workhorse method, so that would be enlightened developers. The suicune.

C

So the send a Singh one calls them or send.

J

Mail async calls send a sink. Oh I, see so regardless you would have to. You would have to have a send async method. That's what the cancellation cram token parameter anyway, so may as well make that public.

L

It makes sense.

N

Let me just add this to the list.

C

You so the existing sent async is what you're trying to ask.

J

Probably a pap nothing but.

C

How to sent me lacing, oh, maybe that's. Why didn't propose.

C

You it's a bit weird that they sent me lacing this implemented over the other one, the other one doesn't return taxes.

M

J

A sink fires, a send completed event: oh I.

C

See what you're saying and then sent me a lacing is basically, logically the you know: vegetable, the task-based one. Well, in that case, I, don't care. Actually, the organization.

L

It's actually sent Asian cancel, that's maybe how it's better to get our business on a single.

J

Should be straightforward,.

J

N

You all right just do.

N

N

G

G

M

I'm hearing it strong away.

J

Why the best way to see if an email address is valid is to send mails.

F

There is, and even that's, not necessarily good, because not every email server will say this was invalid yeah, so the only real way is to send a mail and then wait for them to click, a link and say yes,.

J

This is me, yes, that is honest to god. The best way to see if an email address is legitimate, email address or so.

L

We already bought parchment other nothing to go for this. One was so we don't need to throw. We can basically invalidation as best as we can. It's also a milk.

J

Like how, how good is the is email address or is a mail address type like? Does it actually understand, like mail addresses that people type into systems nowadays, because doesn't it just have.

F

To be at least three characters where one is an at and then there's at least one character on either others.

P

Yeah, looking at email address at Tribune.

C

Used to be a Rutgers, it basically now just says: do you have an ad in there.

C

First, one.

L

Cannot be the last one, that's pretty much so like that's aside from sort of the usefulness or our safety correctness or implementation for parsing right. It all was to not having to throw you.

C

Know I get that like it's just that email addresses is one of those hardest things in the world. The question was more like whether the current mail dot mail address even make sense is even the right API to call to begin with. Yes,.

E

F

If it's really, that is email, it is possibly an email.

C

The omen interest Parsa, is pretty involved.

C

It's the reason I looked at the data validation attribute is because those are the ones that are used by asp net pretty much throughout and have seen probably a lot more updates than the system net mail mail address parser.

C

So I'm just wondering whether we would even recommend people to use mail address thought cars at all.

C

Yeah, that's the question right because clearly, betta annotations could have called into it and they decided not to a probably because they would reject too many valid emails.

C

You so I mean I. Think the API in general seems fine, like it follows the normal patterns. It's just. Do we even want to bother with promoting that address.

F

And you will select for mayor.

F

E

F

Yeah, which means that if we explode tryparse and we follow the design guidelines, also.

L

F

And I would just say: no: if this isn't what people are supposed to be yeah.

C

That's kind of the gist.

P

P

P

O

O

C

What does this thing called again.

D

C

C

All right so then lets me just go back here and let's.

C

C

C

C

C

All right, so we have still what 15 minutes left I, don't one that might be.

E

Considering the people that are here in the room right now.

J

Three escora right: we can shape the the framework anyways so.

C

I'm, just walking them top to bottom right now, as long as you have something urgent well,.

F

It's one that would unlock work on system numeric. Mind be like that between three three six, three, seven, nine. You didn't.

E

Even have some advice.

E

Oh so basically always.

F

So we've had the existing system, numerix vector two three four invective, tea types. For you know, almost ten years now we just exposed the hardware in forensics there's currently, both of them internally in the JIT are represented by the same underlying type Cindy, but there's currently no way to do a zero cost conversion between them. These are just proposing extension methods for those types that allow zero cost.

J

So what would these extension methods do? For instance, if dr. ot is 256 bits and you call as dr. 128.

F

That's a that's listed a little bit further down and the issue for if it's 256 bits and you call 128, then it truncates the upper and if it's 128 bits- and you call the 256 bit overload it just 0 extends so basically what you would expect to happen. Nd same behavior that happens for vector 128 in vector 256 today.

F

Well, the one note is that this would live in system numerix, because vector 2, 3 & 4 are not on Courland. If they were in Cortland, I would have proposed that they be exposed on the vector the existing vector, 128, 256 classes or other similar methods.

F

J

So vectors two three and four just take the year: their internal X, Y, Z, W values and throw them into vector one. That's.

F

What the sophomore fall back would do, but for the actual intrinsic implementation of basic Olivia and Noah for vector two and three. If they were in memory, it would do inefficient load. So it would be a normal vector to load and then a convert which is the no op and vice versa. But if it's already in register, then it's literally a know-all. So.

C

Why wouldn't we just push the time stone to call it? We.

E

Could if y'all and others don't see a problem with that, would.

C

Rather, do this then add another type of extension methods right and so, especially because you know another question I: should they be marked as a in Corazon boy, because you're effectively, no.

F

We've said that you don't mark them as because taking them by RAF, the jet doesn't emit optimal code because it considers it address bacon, and it can't be fully alighted. It's the same thing that we ended up sure.

C

Weather, but what does that mean that effectively instance methods? This would be no menno. What more efficient, no.

F

We instance methods are less efficient than extension methods in this case, because instance, methods have this as factors taken. Remember.

J

We added a whole bunch of static extension methods to vector 128 and vector 256 before three or shift because they used to be instance, methods, I, believe yeah.

L

J

Then it was determined, it was more efficient to make them static. Take this as a brand like this. These api's make sense to me. Yeah, like I've, done effectively stuff like this in, like where.

F

Do these are they extension methods move? So all the existing extension methods either live in system numerix, vectors in core FX or in system private core live under system on cam intrinsics, depending on where the underlying type slit yeah. If we can get Yan and Carol to say it's, okay to move vector, 2, 3 4, as well as quaternion, matrix 4 by 4 and the other types that and on it down to Portland, then I, don't waste a movie other ones.

P

J

Yeah you're right, they can just me- only have to move there, two three four, so we had so if we move these down to Korolev than we would put these extension methods on the existing vector, 128 vector, 256 classes, right thanks.

C

Yeah we have intrinsic sterling living on sort of Kolob anywhere vector.

J

2 3 &, 4 vector t used to, and then we moved it down it didn't realize those were actually enlightened by the the Joe vector, 2, 3, 4 and vector T are all specialized in the day. Well vector or T is in private core lab so that that at least is understandable.

P

J

For yeah right, yes,.

C

J

C

Sounds to me that we should just bite the bullet, because I mean yeah. It seems beneficial to just have the specialized types all and call them anyways yeah. That's.

J

Fine I mean I can yeah. That would be my vote as well. Move move those in the core Lib removed after extensions and put them on use the synthetic types.

J

So does that mean this just needs to come back around in like a week or two or we could just approve today and say: here's a shame, move the text that one what types would need to get moved down aside from vector two three, four three, so just back to Q 3. That seems reasonable to me.

F

That would also I believe, allow us, through the special casing in the gin that says, look for intrinsic types in either correlated or Superman.

J

Visit types, the roughest, the reference for the type of still insisting numerous vectors you can even prevent to rotate that's fine.

F

F

Consistent American factor all.

F

J

I would go to prove.

J

Of Jeff bring those down.

C

Yeah maca does needs more I mean if you can't move it down, but I think it would take a little bit. Okay,.

F

So it's approved if we can move down and otherwise each work. Basically what I kind of Express that right, I.

K

Know you could.

J

Just type it out.

K

J

Type, it all mystery out well.

C

Just follow the other one then or edit the proposal, whatever like is easier because getting good bit up.

C

Pathetic loser right because they did I, but that a good changes. And then this.

J

Time I am running into this a little bit and what I'm working on as well, because I'm I'm working on some encoding stuff, that is maybe these api's will change. If we bring stuff down to bring stuff down into particle.

F

About an hour and 37 minutes, and that this is a pretty if we get to move packs down, hey.

C

You don't have to include the timeline because but once I click the send email button, the more you become cover.

C

We in the vicinity.

O

Right so I think.

J

Seven minutes we can do it.

C

The meaning one that.

C

C

Now, what we take a look at this one.

C

C

And you overloaded, with only spent, assists information, physical.

M

For as it was physical address,.

J

Being MAC address, that's what it is. Oh yeah out. Physical address, math, okay,.

L

You this one came from a PowerShell that they are some work files and they said performance is somewhat akin for a more important forever.

L

E

And I updated my proposal with a comment at the bottom, showing if the taxi move down.

J

What I, what I would say here is the same thing that we've been doing for other parse methods. We already have a parse method that takes a string and returns a physical address so have an overload. The ticks read only Spanish are Shore make sense for tryparse. We would probably have two overloads one that takes a string, one that takes a read only spam and the out parameter will be named value just to match all of all other tryparse methods we have and that.

M

J

Connected on so, we already have today upon even a parse method that takes a string, address and returns a physical address yeah, so the parse read only Spanish just an analogue or whatever. There is no try parse method today. Yet so, if we added a tri parse method, we would probably do two things. First, we would have an overload that takes a string in addition to an overload that takes a real in Spanish on and the out parameter will be name value us to match the existing.

J

The existing guidelines oh I, see but like beyond that I don't see any changes.

L

Yes, it makes sense, I agree and you have a guy who is actually working on it from community. There are some formats we didn't support. What not so they didn't require.

L

Api changes for the F champion, who's willing to work in fluid, so at Acadia to the first party interested in getting this done so to be easy and she for us. You know.

L

It should keep yours linked or comments for Bob.

C

You would have string address and then out. Physical MAC address majors that right.

J

You like there might be some other considerations like do we want do we want to get a dress, fights that returns a real nice mana bite, in addition to a byte array like we could consider that if you sorry one minute biter daylight, so we have. We also have a method. Get' address, bytes that returns in light array. Do we need an equivalent that returns a span like? Maybe we just made four people ask yeah.

N

J

This makes me think, like I'm, looking at all of the code here and how we have so much tax parsing in the framework like we should just add, like char dot, hours hat or something even a three-lap 0 through 9 interest returns. Speaking.

C

Of which didn't we already add a method that gives me a hex encoded string of our hyphens, because bit converter tune string always gives me the hyphens and always do don't replace with empty yeah. But I thought with it's something we sauna with you, and somebody proposed a yeah.

F

I thought we discussed something like this one or two months ago, yeah yeah.

P

Did we ever deviate the very ship, the API, because.

J

That one yeah we had talked about having an operation status space to API for stuff, like that, don't.

C

Know I want to string developer, I, don't wanna be.

J

C

Getstring not too strong. Well, there's that too I'm a might have been caught actually gets me I, don't know no.

J

It's not alright,.

C

So we have two minutes left to there's no point and plus I wanna run to another medium. All right then I would say this is it for today um yeah. Look this time. I got everybody's, hold it in the video, but now I actually include the comments in the video itself: Oh prosperity, okay,.

F

I found the issue: zero zero one three and you commented it seems we should survey what internal implementations we have all.

C

F

Places or something similar, it's.

C

Still, it's still open right, yeah.

F

It's you can't get from ready for review to suggestion.

C

M

F

M

So you're the reason that I have good call tool replace like oh my.

J

Github notification emails go into a spam box, yeah I know, I know that from.

E

C

Right, bye, bye, guys.