.Net Foundation Design Reviews, 3 Oct 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Reviews: GitHub Triage

Description

00:00:00 - Approved: Add QueueUserWorkItem for local threadpool queues https://github.com/dotnet/corefx/issues/12442
00:36:16 - Approved: Add CancellationTokenRegistration.Token property https://github.com/dotnet/corefx/issues/23828#issuecomment-333935487
01:26:39 - Approved: Support casting ReadOnlyMemory(T) to Memory(T) https://github.com/dotnet/corefx/issues/23879#issuecomment-333939600
01:40:56 - Approved: Add IsGenericTypeParameter and IsGenericMethodParameter to System.Type https://github.com/dotnet/corefx/issues/23883#issuecomment-333940885
01:45:26 - Approved: S.R.CS.Unsafe and read-only references https://github.com/dotnet/corefx/issues/23916#issuecomment-333942166
01:49:46 - Approved: Add MemoryHandle.HasPointer https://github.com/dotnet/corefx/issues/23974#issuecomment-333943725

A

Three two one all right, so, hopefully the audio works according to this screen it does. We will see whether it actually does so. We will spend the meeting essentially to the two things the first. We will take a look at the things that awesome was floating from broken and then the second, hopefully more than how people spend on PR, Jing or.

A

You know all right, so this is the issue that Christophe shared with me like. Should we go over there? Should we just dump into the API? Is the thing ready you, the bulk of that thing right a.

B

Second healed er, yes well, but there is only one type it used to have 20 members. Now it has 16 was.

C

Under discussion, if you know associated discussion about, like you know, extending this to some kind of streams or something or was that closed or not or were like you know, these are just the primitives that we need kind of the higher-level API.

B

C

Just asking on a you know: I'm.

B

Not sure if it's closed, my my position on this is that, yes, we need high level type. It's non-trivial to design I tried it and I'm, not even convinced it's possible to design it such that it does. Everything that we need I mean.

B

I, just don't think that it this type which, to that this.

C

Release next week, Steven, you are probably most proponent together of that. Like you're. Okay, with that state I mean.

D

I I think I follow the opposite side of Christophe I. Don't understand why we can't but I'm not gonna, make a big fuss about it. Okay, let's.

C

A

All right, so this is the EPI. Are you working.

E

This doesn't matter pushing.

B

The issue: no, it should matched I copied and pasted. Oh well, if the type name is changed because.

B

This video, so maybe very quickly, Mike how he ended up with the 60 members. The type used to help just read, write, read big big onion right, big endian, a little endian, and they were over t. We decided that we don't want them over t for very good reason, because if that T is larger than an individual, primitive like it's a struct, let's say with many dudes, it doesn't make sense to reverse all bytes in this chart. It only makes sense to refer three pairs. Bytes like in individual primitives, like inns and logs.

B

Also there's a performance, quite significant performance improvement by making the methods shall be typed, meaning there's you know, ain't and write it. Unfortunately, it exploded in the number of members, but I think it's like that. That's a kind of reason and a pattern to all the Knights. So it's not a huge deal.

D

Chris I've, just a couple high-level questions, so I think there's a still a few open issues around. One of these are extension methods versus instance, methods right, all right, I'm, sorry, an extension versus just plain static right, that's still open yeah! Well, yes, I guess.

E

D

Should any discussion about like further for the read and write methods, whether we actually need int versus our sign vs., unsigned overloads or like? Did you consider just having signed and doing a cast?

D

We didn't consider that we could consider okay, I'm, not sure which is bad. Everybody would, you know, effectively cut the number of methods in half and then there was a discussion. I think Josh mentioned it earlier about get sad or get put versus, read, write. Nothing.

D

Did we decide against that or it just just hasn't been factored in so.

B

That's that's. You know this is such a mechanical change that again I wanted to discuss it in this meeting, and then we just make the change. I don't actually have a very strong opinion on, like which one we should choose. I think you know most people were leaning toward either get said or get booed. I think that was the consensus. I mean there were two jokers me and Ari. While it became Pope.

C

We have actually we have someone come currently asking like if we consider other languages. For example, Python has PI come back. What item has pack and unpack I? Don't think that we have anything like that? Actually any variant VCL so I would probably be against that. But you know just saying the.

D

Only place I can think of where we use that terminology is in the overlap, types.

F

Is the reason that we unfolded the big in being a little bit in the methods instead of using a second argument in the functions I.

B

Think it would be slower if it's a Nina I mean either a boolean or any. No, but that's I mean.

G

There would be a branch involved.

B

Well, there would be a branch in that this is not a parameter passed a limit and given the methods.

H

Are so cheap, it may be not trillium I mean like me. It is very clear what this does.

H

Oh, yes, I'm, visual and I assume purpose. Also. The reason that we have both read-only span and span correct well,.

B

It's not purpose because he's interested method- oh and also they are extension method so like if they were only taking read-only span like they're. In my thoughts, you could not call it on one span with their with the fake instance methods in that and.

I

Maybe mention this: we did settle on in 64-bit, Indian realm they get in in 64, I, don't know if we settled I mean as we want to it seems like. One of them is I. Think those general consensus that ideally the type will be at the end, because it reads better, but the intellisense is less is out and and reading better may be important, because there was some confusion among various people about whether it was Indian of you or Indian at the data. One.

B

Thing that we could consider is like drugging either little-endian or big-endian, like choosing a default. Unfortunately, I think the read big-endian is more common accident, a little idea in fact, I wonder whether we eat the little endian in the first place like.

J

More common because that's the network and genus yes, a.

B

Big-Endian, basically, most of the machines that we run on would be little endian and all of them, and then you receive you know a stream from the network and you call the big endian metals the little endian methods. It would mean that you got a little endian stream on the big endian machine. The likelihood of this happening is quite low. There.

F

Are definitely protocols in particular that were created by windows that they just let it out there structures the network without.

K

F

Html and so there there are places that your deserializing data, that it will say this is a little india, 32-bit and.

B

If you want your code to be portable to describe pathetical.

I

We shouldn't assume that we're always gonna be running on big-endian little-endian, of course, but I'm curious other. Is there anything in the pipeline that would hold us to big endian.

F

I

B

Of it, okay well, but on the other hand, like I, don't know if we are winning Mars by any movie knows, and they are just they just follow the pattern so I know. So we either way I would prefer to eat them. But people think that you should say if I would just Chris.

D

Not separate from the Indian discussion a few minutes ago, I think there was a question about the read: only immersive span versus span and I didn't hear the explanation for why we need both for the reads, because.

J

That's the only way to make the extension methods work, but if these weren't extensions we can effectively eliminate half of the read methods. If they were not extensions, we could eliminate I think there is some cost.

D

I would hope that, given that read-only span and span are identical in layout that we could make the cast like free, yeah.

B

Theoretically, we could whether it's already works in practice or not. We don't know to check so now. You know you wanted to discuss the extension method versus not so here's my take on it I think if we remove the make them non extension methods, the name of the type should be short, but the name of the type is now so generic and there was a suggestion of the previous API review to make it not so generic not not like, as grabbing the you know, the best name for a binary ventilation api's.

B

So in the proposal online I change the type name to be longer. I forget what I like binding.

L

Attention span by name is something.

B

L

Binary extensions or.

B

B

I would prefer quite strongly to keep them as extension methods, because it would be kind of agree to have to call binary span extensions that treat in 32 big-endian blah blah blah.

D

Yes, yes, I understand the desire to keep it short, but at the same time I mean these methods are already so long read in 64 little-endian does adding the word primitives or you know, or binary friends really make it.

M

It doesn't really make a problem at all. Yeah I I think that I would say your dog seems to think that way. The majority.

B

Why I think it's so nice, like today, I looked in the you know code that uses these api's. You just say you know mice, panda tree in turn, 2, big idea or actually in fact, you very often call just span that read into int, because if you kind of don't care about the end guinness, it is called the read method. My.

D

Concern with these being extensions is span of bite is such a general type. These are so specific and if I've included the relevant namespace, yes, intellisense is going to blow up so.

B

I, that's why, again on in the proposal on line I move the type to system buffers that binary namespace.

F

So I mean the like the word fans I think it would be rather annoying I.

G

Completely, that's why I moved it to something so I think I would have been not easy.

A

Well, I would say they they are in a different space. That is very specific, I, don't think it's I mean the problem that we have today is where my for example, Systems Inc, is a dedicated name space, but it's effectively always imported by people, so it's effectively in the core namespace and that results in like 500 method showing up it's rated. That's the kind of idiot.

A

Now the other hand, like it I, don't think many people who invoke press its on Stan that are not actually doing anything interesting and also like the visa extension methods are actually in the sub namespace that is dedicated for parsing binary a person chances are. You were effectively wanted, Russia to see them in school. It's a thing, that's the concern of extension method is that's the right target, audience, see them in scope and, secondly, does the wrong idea of audience see them installed right and I. Think if we designed.

B

A

B

Avoid you know false positives, essentially, and also in general, I kind of today. We have many of the ideas that manipulates pans in the same namespace as this one system. Batteries I actually think that we should just move them all out because right now, if you kind of like I used, you know I di you reviewer tool, I put all the dll's in it and there's no like this kind of walking in space that contains the very basic primitives.

B

Like the you know, all the buffer right move this and there and then bunch of kind of random methods. Some up for manipulating tests. Some are for manipulating binary. I think it would be nice if the root name space system buffers help derivatives, and then there were some namespaces for either text or binary manipulation. I think it would be nicer, so I kind of wanted to move this type out kind of anyway. It also solves the extension, the devolution of the intellisense method. So, like you.

F

Know I guess, with an extension method, you always have the option of collagen static. You include the include the using because you need it in one method and unless, unless you feel like carving it up to a partial file, just to avoid the extension now the rest of this class that you're working with now, you start seeing the loaded have.

D

Have we thought about how we would want this to show up in Docs right now, if you look at like Docs that Microsoft comm extension methods on a type show up on the documentation page for that type, so all these are gonna.

M

Show up for spam yeah! That's a good point! I! Don't like it! I, don't know why we are doing this or that is a.

A

Column, that is at the dock team as well I mean we have solved this intelligence. No. No, there. You have seen the spread like since I think a couple of releases, no kurosu Dominion oka stands at. The very bottom has filters, so you can say only show me events or only show me message and you can turn off extension methods on flights which I've used quite a bit to find actual the things I'm looking for be able to look at sample code. That is I, think the difference as well I mean.

A

If you look at the different ways you can call it I think there's one that meets quite nicely, which is this one here, because you have to keep in mind that most of the time you have to slice the data before you can actually call these extension methods. I! Think! Oh you don't so I, don't just reach the first four bytes. So how do I plant afterwards then.

A

D

The argument for why I was pushing for rewriter instead of these methods, but.

B

Well, so, basically, the real codeine signal, Aires they dust, lies or call my leg, because the protocol is length length prefixed after yeah, this section actual code.

N

That's not right, it would be the size of that in yeah.

D

Called a slice like that, and you did before comma dated length for.

F

Saying the length identifies how many bytes are part of the spirit yeah.

F

D

Well, no, but then presumably you're gonna, okay, presumably you're gonna do something with the data that comes after the thing you're reading. Unless you're and that's this method is purely about like if you have multiple things in this span, which is often the case. You're not gonna, want to overwrite data with.

B

Actually does that and then you overwrite the data and you pass the data somewhere else, but.

F

Then, presumably something else has data from before you render link so.

B

It can now advanced or yeah; no, no, it doesn't need to advance. You pass the sliced data and it just does like some additional parking. It's very good right, but.

F

If you had like a deal that was, you know four bytes long, so you're gonna have a link to field that says for then you have four pedlow device and then presumably there's another field after that, the point prepended. So somebody needs to know how to get to the next field exactly and literary prizes inside of it. So you have a structure that has two fields that are vocally pretended effect. Like your serializing two abilities, somebody needs to still have the outer street or the bigger span to know.

F

Oh, we read it four bytes for a length which then that a value to so I need just get six bytes to get to the second feed right. You do basically say you're, saying we keep.

B

Going yeah I thinking, I, don't remember how it's in signal are I, think it basically, they just received a bathroom buffer tells and then they are like done with this, but it's possible what.

A

You said it's it's possible, so these methods effectively you can they don't care. How many bytes are remaining so long, its minimum current, what they, what they have. So they want to reach four bytes. It has to be four bytes or more, but they do very fine score by just write any credit.

B

Please throw an exception, is you don't really know, and then there are the try? Metals. Don't know.

B

Who holds the reverse for its? If you just first bit plate extract, and then you realize that your you have different idea Ness, then you leave.

C

Variants by fueled by so the question was on the chatters about the bite is for completeness and does not. It does nothing so.

B

The idea somebody suggested I forgot, who it was anyway, believe I suggested that we had bite, because if you have a shot with use, you can just always sleep errors. This one is in vogue and probably gets optimized out, but then, if you change a few to some other type that keeps working.

D

It's a little strange to me. It.

J

Presumably you'd.

D

Be special casing each field based on its type in order to have the strongly type nature of it like if you're going through all the fields you know if field is by call reverse by the field is short call, reverse short, I.

F

Mean I'm guessing that it's you'd have, you know, pounded big endian or you know. If whatever dot is big endian, whoever we exposed that value, then you would say you know, foo, equal, reverse of who are equal, reverse, afar, etc. I want change the type of food. Why does my code not compile them.

G

Well, no, no, it's compile and install well. If we add bite, they also don't have bite or.

F

If you change it to bite the worst one, auger sister, if you change it, divided with with now turned up one bite before light, which then would fail on the assignment.

G

I'm saying if it.

B

Used to be a bite, you don't call this better because it doesn't exist. Then you change, you do it now. Your code is incorrect because you forgot to add database, so that was nice but I I love. You very well.

C

Speaking of which reverse like because it's on byte short and ends like member or an extension middle that you know, should we call it like rivers and then nests or something I was honestly if I'm looking at byte like what's in the reverse, what does it mean.

F

A

We did remove the reverse of team right because.

A

So take what is reverse to us. I think it doesn't digitally reverse device, but it will swap them in place. Correct.

I

It's a no bite.

A

H

Actually, reverse it.

A

O

H

Switched around, isn't it isn't a long.

O

Not just reverse the byte order, because it's more it was busy. Thank you.

P

O

Reverse, which one is left clip of the pipes as long is like three two one: zero, seven, six, five four.

A

Isn't it anyway, it doesn't matter who it does because because, if it's not an actual reverse, we should rename the method. No. But what do you need? It's not very verbs, so.

K

A

K

A

First straight, if it is so, let me put this way: that's what it does. I think the method is. We should reverse.

C

And Ian, for example, change. That's.

A

Actually, fully reversed screen and I think the message may make sense, but the case it's not like byte wise, reverse.

B

I think it's I think it's full reverse, but I still think that we should it's such a rare method to call, and it would just make it clear.

A

Q

C

Only indian Nessus, whatever I, don't care.

B

C

Suggest it on the on the channel by Chris, King you're.

A

Proposing Los Angeles.

B

C

India, whatever, as.

B

I said this method is not something that would be very commonly called, and it just makes it worse, because I think it's a fine idea right.

I

I

Think you're right, 1, 2, 3, 4, 5, 6 s, 0 1, 2, 3, 4, 5, 6 7 turns into six seven four five, two 301.

R

That also true in 32-bit I'm.

I

Just looking what we do for IP addresses.

B

Well idea: justice made special, for example, with suppose B is special. You don't ever eat with. Maybe.

I

Units because it started in Windows.

B

Maybe we should add a novel because it doesn't actually reverse well.

I

Big endian machines, I presently reverse them know that this.

S

Is special case even aristocratic, so.

A

I think most of the stuff was noticed: I guess it can. We make the engine is overload. Somehow I think we said no preferences.

A

The other one was: can we collapse the number of methods? I don't see.

B

A way I think the only way like the only one that I know is.

B

But this may be short-sighted. We may find ourselves having to add them back at some point. It's just.

S

Any reason why we have.

S

Yes, because the generic one is bit rusty, oh.

H

Because the drej one cannot do in unity, yes does.

F

H

F

B

Okay, so here's an issue we throw correct if it's, if you pass and non-primitive shot, okay, so here's what we wanted I'm on daemon --thread went the seizure team. They said that they are willing to add, concentrate in this release to constraint. That is here to primitive times.

B

I need to write a proposal or language design meeting and- and we should throw but I say it's a super trainer feature and they wanted to do it so tomorrow, I will write a proposal and basically the tea in all these dry write, write and read methods is actually has an attribute called primitive. It's not enforced now by the compiler of the idea. It would be, and then the teacher is correct without this enforcement, the teacher is a bit tricky because.

B

If we have her add this constraint in the future, throwing an exception from this api's, she is not a primitive it's not enough. You can catch the exception and something else, and then in the future code blue will break so the only heart that I came up with in case the language team does not support, for it is to all fail fast. So if in these metals, will basically I think that he is not a primitive, we would fail fast and therefore you won't be able to actually have programs.

C

That run, you don't have to do it, even if language does that, because you want this as OOP down level.

B

Because we could basically do what span is doing which, like he'll the type or another compilers.

F

Some primitive or include presumably well decimal and double, and we don't we presumably don't want any of those type stores.

B

No, so primitive would.

F

B

Check that that checks, whether it's a striped and it doesn't have a ref reference type field, that's the definition. Okay sounds.

Q

Like let me lick the bowl or something so so you wanted to work on any portable.

A

C

So, do you want to put primitive and not writable or.

C

T

C

If you're introducing this, you know alternative names for the same thing, because that's confusing so.

F

This is the keyword, types string all.

A

A

Think that anything else to say to be.

N

One of course, on the namespace and what said.

B

System partners binary it's a son named space of the pipe the namespace of contains, like the you know, own memory and friends yeah and the name of the type now I put it. This binary primitives but I find.

A

It hard to because I think it's it would be easier if we could see you know the entirety of this bag. Teacher works with all the namespaces and I can tell.

B

You so the other one will be system buffers text and that's it because the other idea is basically that, like similar to these, except they manipulate text and basically forward, go there yeah and base64 daughter, because Levi's arguing that it's a tax for minutes. So we have sister buffering.

T

H

Chose either this one or energy.

F

Yes, that's why? Because make 64 decoding, so we've got wherever you're putting you know like to utf-8 string so.

H

B

Do the encoding the thing cause bytes to text, not the other way.

F

U

We have somebody to community so what we have detects all the encoding services.

B

U

To torture form a basis for.

B

Parsing formatting and then buffers contains and buffer, contains, like the primitives, like old memory and some it doesn't contain span. It contains like helper types, like my man, I, regain abilities.

B

B

It's basically in system we have the very core pivot is, then we can system buffers which is like a gue for lifetime management, mainly, and then we have these- that are operations on spans.

N

B

Not because I can imagine the end overloads for memory.

N

B

Other things in there you.

N

Had an overload for all.

N

C

It's you know, so, just a heads up understanding, email, but we talked with us yesterday about the open kind stuff and basically even for static things we have. If we would this and if we put it in dotnet core, we have to freeze the API servers, you cannot ever add anything can do ever well.

B

So we were on email thread that we said that if we change the net standard version that we can change.

C

It will have a special meeting for that. We cannot add that like technically, because it would break some platform, some combinations that we're gonna walk you through the technical details, I just saying that, so from that perspective, you know imagine like if you will do in future at something you want to make it as well. You will have to you know, add one more time anyway, no matter what, let's.

A

Discuss this show.

C

A

Think you given it this time, seems to be designed for holding extension methods related to binary, parsing and spans I. Think the name extensions make the most sense. You don't want I mean unless you mean there's only really one case. We have to call this aesthetic subjects using a crazy language to support extension method or you are having some weird conflict with their body library, so you're proposing binary extensions.

F

Then yeah personally I would make it not the extension methods and if somebody wants the extension syntax, they can add all.

A

I mean I, don't see, it also can be used action methods, I mean I'm just talking. If I get.

H

Mean our goal is not to ship the minimum possible API set.

F

H

It's the way that I've used it and.

F

Granted right now, what I'm doing is weird, because I have a common file or effectively unwilling to break it up into even one. Only one class in the file, but I have a multiple thousand line. Long thing right now that every usage of span would be contaminated, having the same space, if I ever used it yeah.

N

You feel included a dude generalizes it only if we would have actually put them, as instance, methods on the type. There were some reason. So would we ever put these as directly follow this I? Don't.

B

And I don't think this guy I think it's actually good gun. You don't want it to show up as an S. You shouldn't have too many.

N

Did you just say namespace.

B

Is the time yeah then I fight that? But this one is nothing, that's right, just to be able to show your game things first other. What should.

N

I do I think the guy that says.

B

If it's in the same namespace like.

F

My expectation of this that it's a thing that you kind of wish wasn't in submit and I. Don't I think that here we're just liking the usage of extension method for reducing the number of characters per line and to me.

K

That's not what exception.

F

It and if somebody wants that you can do it well, he didn't the same. Probably a.

N

Good point just doing the study.

N

You can do anything, nobody, you can eliminate the type name by you.

B

Oh, you mean you just you still call it with the static syntax, but you don't have to type the technique.

H

Michigan imports is in fact one character smaller than in the extension just.

B

So we understand all the implications of this like then calling the methods get would not make spec to make any sense whatsoever, because it would basically have get data. You know like.

N

We can have funny about I think it's good is reading right, we're getting my.

D

Concern with reader right is only when they're instance methods when they're extension methods. It's obvious, I mean at least if you understand span it's obvious, that it's being passed by value, not by reference, and then you have to worry about any concerns that someone would think the span was somehow being advanced.

A

But if you actually call it through the instance method, subtext and you.

E

Know, like you I'm saying, if they're.

D

Static methods and they're, not extension, that that's my concern around the read/write goes off of all.

A

Something you were just making the point of extension versus instance, but yeah that would be also to first engine, see. Okay, then.

N

You bring up your sample code again so.

N

Like if you had to using statement of finding primitives at the top that finding permits doctors are like to yeah, you can do this.

V

N

B

You need to take.

N

Now you show the best in the pen: oh yeah,.

W

S

Like that, so I'm just say thank Allah, that's right, short yeah, but.

B

Then I think it's not bad as long as people know, the syntax of using static system. Parker is fine there. It does say what we call the type name, probably binary, so we need to kill it. So you need to connect some names.

B

Like in system buffers directly, there are some like new types that almost you know like not many people use.

D

If they're, if they're, not extension methods, what's the rationale for pushing it down into a sub namespace I mean I, don't really care I'm just curious, so the I don't think the.

J

Reason is like.

B

You know super super strong, but the time step today we have in system buffers namespace kind of like plumbing types like you almost never use them. They are used for extending the holder memory like time management system.

D

So just saying like system buffers is first sort of the Entertainer exchange types or whatever, and the sub namespaces are for all the operations over them.

J

Yeah, that was.

B

That was kind of what I was saying. Those are the two reasons to have this app namespace. Now they using static statement kind of slow, but maybe it's okay, I thought I I'm not totally opposed to it. People think this is better that.

N

Will definitely help with the of the potion of intelligence, I.

A

Don't have strong opinion, I get I would say to me it's as I said before it means that's about you know, minimalism and more about you know if I use this feature commonly, what makes my life easier, I would I would sort of sort that extension method to make my life easier. If I call them I am, if I'm very really calling them, then it doesn't matter and it's just one file but I do some parts thing and then you have to just think of it.

A

Bob I think it's fine, but I think it would be expectable. Did you spend use api's commonly I? Don't see the point study but I, don't even know explain to the teacher to make that determination. Well,.

L

Given that you're slicing anyway and if I she feels like reading multiple fields, if CTC is better to have a reaction, because you could do Dara dot, read, there's a slice tarot reader, a slice like this I mean what is judo is we can we.

A

Could decide to make an extension that is later, if you make the type static today that some of you may also have so.

N

No great change.

A

Makes attachment so because people start and being you know, static methods you have to import, or we could say or later on. You know based on effective extension. That's it.

A

Okay, so then what I'm hearing is we go with one extension that is for now.

E

Anybody a product I designed in a way where we're comfortable with that being.

J

B

J

The only thing that we.

B

Would do differently whether they're extension methods or not is the name of the time. So if we change it, some at some point to extension methods where the type will stay pioneered riveters, which is not ideal. But maybe it's not a big deal.

N

That's not going to work I think the read/write versus get set is a bigger deal because.

F

We can always just add the extra binary extensions with get set or whatever, which changes the verb and makes it in Excel. Another idea of what you can call the type name, something with and eNOS.

H

And eNOS well because then you need to move your Reed of team right into you.

T

K

Default into the Machine and eNOS, or we did default selected converter and.

P

So like bit converter.

K

Is not great because it does default to like the Machine end eNOS now from time you want to read like something from the network, which is always.

F

K

F

K

I think really is obviously is a machine.

F

P

Or is it a fault I'm.

B

K

Endian always things.

B

Like you're just generating in this of this, it seemed that you got you need to know the machine that DNS we know. So we thought magic number for real right size right.

X

B

Since you didn't tell us what your stream is, technology.

K

Change which is usually a bug and someone's could I.

F

Assume that read of T is just interpret the data unsafe interpreted as a t-star and in return star well in a lot.

B

Of cases they're, the three of t would use to optimize performance because it's better to this blade and then he persecutes one. What I want bristles instantiate discharged and keep reading this bad one. If you.

F

Bite you write soft is that your implementation is essentially let's compile our interpreters, T pointer and then be reference to keep order to copy it back out. So it's just take a pointer and turn it into a plated. Yeah.

D

Josh, what are you suggesting as the alternative.

K

The alternative is you, I was asking if we've just picked a default like we said that, like we're having the conversation before of, should we drop the like big Indian from from like the the read write methods, just so like I have a default kind of thing of.

D

Is if you pick either little-endian or big-endian, and that differs from the machine? We will force a reversal of the entirety which doesn't make sense so.

F

Honestly, I'm thinking that read of T is I, guess.

K

F

K

Is that actually true, your statement Stephen so like so, if I'm, if I'm working on something that I've read across the network right and it's in the protocol and the protocol says it's always, you know network order, it's always in big and yet do I didn't have to still reverse it.

D

If it's, you know, inter long, if it's, if it's for long fields to properly apply ending this tree, would be reversing each individual fields, not the opaque. You know, bytes that represent.

E

All for long fields together, what it sounds like is the reading right of T. You can.

A

Only use if you base it to stuff a memory like as soon as you persist stuff that maybe.

B

A

What I said so.

B

That the correct and most common reason to use those would be I have a struct and my prayer is then I check and E&S. And if it's different than my machine, then I refers field by field. Otherwise I believe that's the most performant way to basically TC arises chart but I think.

F

Most of our users are going to forget that it.

B

That may be correct.

F

So they'll write, it it'll, be read of tea, it'll work and then they'll go to arm 64 or whatever platform that we're working on that is and then now all of a sudden files that they create on this system works, but they can't exchange files with. But they don't like, because a proposal.

B

To come to it may.

D

Be the only option or two options: either you don't have the methods or you have them and their machine, and in this and.

F

It specifically says like I, think having it either it needs to be. You know: dangerous, read, machine, Indian, Asst or read big Indian, which is NOAA like, which is a training.

B

I think it's actually reasonable I like what Steve suggested more than dangerous, because it's not dangerous like this one.

B

To strength tensions, we could be user ID on't. No, we are. Are we inventing like a new term.

D

Current and system, at near machine and being the fault Indian current is not bad.

F

Yeah, like current, does it suggests, exchange mm-hmm.

K

F

System reporting so.

B

Another thing that we could consider that people kind of don't like the length should we just use le and de instead of expanding little-endian.

B

N

Now you will know, oh, you need to have something you could hook up and.

N

B

We just get a discussion. Whether long is it. Eight bytes are events or.

A

You should think of a name for read/write that indicates that it uses machine Ian.

W

W

Imitates machine ID like um so one question that was.

A

Asked in the chat actually the one who we hear that either let me copy them essentially, then.

B

He passes fan by there, you should they be red, so this we've done asan probably couldn't speak better to it, but we've done some measurements and it's not clear that passing span by that is better. The.

D

No pass passing the span. It's all beam under the coverage. That's already passed by Rev, like the the G, is generating passing the span, essentially by Rev.

B

That's one you've done some measurements. There were some differences about.

L

You know I have not.

A

Actually modifies this truck what well.

D

No, my interesting from a valuable be for the in 32 and 64 etcetera. But for the try read I mean you could imagine it just being a reinterpret cast and returning a ref T, which would potentially matter for.

E

Large T's woman.

A

That's a simple thing there and trying to stood at the first one, so they say you plus a span. The receiver is writing to the to the parameter which they can the receiver posted by Rev. Doesn't that mean that the receiver now is also 22 difference that no.

B

Because so they said the date underneath is beautiful, but the span itself is not beautiful, so the span isn't. But if you pass the location of the Spanish, whether it pass it by value or by rare it points to the same location. So it doesn't matter. You always want to find the same location.

P

Let's say, let's say the declaration.

R

V

A

The caller than Austin the data, no, that car is like forehead, it's correct, so the Jim under the covers is causing this thing by Ram. Doesn't that effectively mean if the caller does that it's no livable? On the other side now.

D

I mean it's not exactly like it, if not as if you just passed by reference, the chin is playing some tricks. It's an implementation detail, that's not observable. Okay,.

B

But the point is I: I know that from I think Russ I was sending me some some performance, comparisons and I think there were like it doesn't matter you and in fact he was saying like it depends on your context. Sometimes it's better to pass it by 12, performance-wise yeah, probably because of the tricks that they play. The.

A

B

A

Eb already has extension: that's it's pretty sharp for some reason, but.

V

A

Right so then I think we covered everything and then I think Stephen, just one of another one for Reed tea white tea. Should those take reps or return reps cuz, if you actually how.

D

Do we view on this these Reed methods versus unsafe, as.

A

Well, let's say these would still verify the balance of the baton. Right versus unsay would not do that.

D

So this is just adding a length check.

A

Well, I think, like I mean usability wise. It's just nicer. If you can call the same, API is but I think from a writing standpoint. What's a nice, did you get some because you don't magically no big deal it like you, know, pasta and all right? Well, you just.

E

Took a bounce checking from the.

A

Spam do we have unsafe methods that actually operate over spam? I thought you liked unsafe, maybe you're right, I think it just takes a.

E

Rest to you believe, okay,.

Y

D

And we have methods on spam that cast from like a span of t1 to a span of T, yes, but some of them are non portable.

B

Which way this is the reason not to do it by grep, so I was thinking about. This is a good like for the right method, maybe that you should be by that it made. It will be cheaper for a large scraps. On the other side, when we read, we return a virus, we may have some like ongoing performance penalties. If you don't, if you are not aligned properly. So here's the issue when today the method is not by read like the return value, it's not the reason of the returns T.

B

So basically, what happens is actually like. There's a new T value created on the start and we copy the data from the span into the new value. The new value will be alive because always because we always like when we create the locals on the stack, there will be a line property right all right, but.

J

They may not be if you access arbitrary locations and they reduce location, they return them by rest.

B

Then, if you misaligned, you can keep paying performance penalty every time you do some with decided, because.

A

Yeah return would effectively point most likely because.

D

Z

D

Even he I mean it, even if it was on the stack somewhere it you might have might be an unusual offset that.

E

You're reading from yeah, because yeah, if you give a span of by a.

E

D

Example would be, you know, you have a 2-byte length, followed by the data yeah, that's gonna push you off from pretty much any. You know eight byte alignment or something yeah.

B

But we should consider the right methods to be by that. Take the T by that I.

D

Assume are even, though, that the shortest methods, the read and write methods, are the kind of more niche need, like they're they're. Only for big trucks, where you've already handled reversing the fields and blah blah blah.

AA

Comments, what is big, because all four fields start becoming, let's perform work. It.

L

Was a big structure? What.

S

Is a big shot yeah? How many fields three point same.

B

But the number is changing over time, depending on how.

Z

D

Mean it's really, it's less of I, think about the number of fields and more about the overall size, I. Think, historically, the guidance had been. You know if it's more than sixteen bytes years, you're sort of quote big- it's probably grown to be larger than that, but you know, certainly in the 60s or more as big.

Z

D

The other issue and enough folks are yawn, commented again on the on the issue thread about the assembly. Oh.

B

This empty to me in testimony I was thinking about system batteries of yellow.

D

Yeah and his comment was he doesn't understand why we're spreading spam related stuff across multiple assemblies? Why not? Just having you know all system memory, for example, is.

L

It specific to accession methods like he, the one Spanish amethyst to be in difference of this.

D

Well, I think you could comment but I think it's more about. We experimented with having very fine-grained deployment, and it was you know, to some extent a failed experiment, but sadness exists. So this argument, doesn't it like you? Well, sister memory exists to why continues reading things in multiple places. I think is the.

E

Comment well, I mean I. Think Kristen's arm is a very impulsive. Seven.

A

Decision but increasing the number of Senators he just picking, which ones to put this up.

D

But one of them already contains fan stuff and one of them doesn't.

A

Sure I mean I mean I would say if you would come up with a thirty something here that haven't seen it. But if you already have two or seven is the deployment standpoint you're not changing anything so.

B

I guess I'm also thinking they're.

A

Taking the stuff.

B

So system memory contains those primitives. We will have to do some careful engineering to make sure that we do everything right in terms of holding these api's. They are just static methods. I think it will be either simpler in terms of cooling or we may never like it's entirely possible that we don't ever add these to the bathroom they are permanently, which is not the case for system memory for the other. So I have a question here. So.

A

How many ascends to become they have like an effort to a system memory system, buffers, I, guess, system, numbers or span? Is it fair? So that means this one will eventually be a boss, because you know it well. These you mean when your memory yeah.

S

A

And then system buffers that, what's what's in there.

V

F

The way look like Christa Lena suggests that you should not use a strap that you will be 16, bytes or bigger. Oh.

F

That's still what we have, or rather one top-selling.

B

Idea, that's what it was. We got 32-bit Smith machines at the time.

A

So the new feature, like all the things we do for spend, are there any types to go into system buffers already or is this? The first are the first one. Well.

L

On memory, let's watch that relate with.

S

Coordinate systems- oh that's MDA, I,.

A

Mean the only argument at IC book, like Yann sting and now from our cindy feature, there's a there's, a desire on the runtime team like if you want to optimize things, you know have to cheat like understand certain patterns and it is easier for them as a whole, it's harder for that some random assembly. So if, if I think this is also the matter of point of like you know, yawns arguing, you know, deployment is one thing, but also from the runtime perspective. How many things we have to worry about?

A

So is there any reason why the old should be something like, because that would to me indicate that maybe we should maybe.

B

But I honestly, so you know what I just said: I think we should consider it not ever in boxing needs so I take the problem of that will be.

T

The same thing.

B

I think that that's kind of what you're saying like is it how confident yeah.

N

Maybe we making a market for this and I think we can I could probably even agree with you. Maybe we should never boxes, but I don't know about the other stuffs, insisting so.

D

The only discussions about surfacing a ray pool in low-level API is, if that happens, that's got to be in the box and the rayful's.

J

Already in core I mean it's already in the doxa core, but then so if we wanted to just inbox a table, we.

B

Would move a table and forward from system buffers that the other I'm saying these all these studies.

D

From that perspective, aren't both system memory and system buffers aren't equal corneum box. Since there are built-in implementations of span memory, an array pool all in system private Cortlandt, so it some people who are speaking to you can do it be a ray pool. Span area called span are both in system private Courland.

D

Doesn't that mean, then that system buffers dll and system memory DLL are effectively already cocoa inbox and.

C

It's if it's a trend box in walnuts, yeah.

A

So let's say: if it's already in box on tour, the likelihood of us adding it to net standard recently be higher and then same point if it has to be so one thing we should consider.

B

It like should because so, if that's the.

A

B

We consider this guy put it in an assembly that we say we don't plan on in boxing.

B

Because it's a tricky to inbox types and I, don't think we need to in box.

N

C

But that's you know, honestly. Better I would actually, in that case, vote for even like explaining in the in the namespace making Microsoft that we never actually excellent a little bit.

AB

Let me just me.

A

So yeah, so it seems like there's no Sam down here so for now, instead of doing a third, let's just leave it as bright red is right. Now the lease is the least offensive thing that about the way to leave an assistant poppers. We should probably have that part of the gentleman conversation whether we should change that and I look at the overall footprint they'd be fine on adding into one.

L

Alright, so then I think we're are we kidding the name? The names are right and read. Courses.

A

Discussion yeah so so personally, I have to agree with like what people said on the net. But for me we didn't write and flies. There is a curse or somewhere that you and that's get it in place. I, don't think they.

AC

Don't to dad's really in place, they don't change.

A

Of course, people.

AD

You know you know we just never in place. Yeah, yeah, I, almost think about replay stuff. You know me and change the method.

B

Name to get I think get is reasonable, get in 32. Little again said: I kind of don't like that is so weird you put I put is fine.

K

There is, there is the negative of food, but like I just cannot like mean we can use it. It's just may be the only thing jollies you know before you read it just right.

B

To the data go like add, a method called I mean a call Oh butts.

R

Are on HTTP.

B

So yeah, so this one is not horrible, but what's much worse is the static methods in this so go to the last.

AC

Nana I'll be joined up.

U

C

Don't have anywhere else so what do a month.

P

F

You'd get so prostatic for a pure static function. We have Marshall as the thing that already does is for pointers and we have Marshall. The whatever, which takes a pointer, does not advance the pointer, and this gives you back the data it's interpreting at that address. So then it uses right. So if it's a non extension or a non instance method, I think the reading right at the right works it matches what we have in the framework it matches with people who are already.

G

F

This or used to- and so it's the lowest cognitive so even.

B

Metal 3 needs to say, send in 32 little endian think about nine.

V

K

That is my ground abilities and Adams on the chat suggests. We just drop the word India on everything. So to say we little.

AE

Factor crystals comment that you need Indian in the will: a joke from bat, so sorry bond.

C

R

Not sure, what's it.

F

Hard to read, one type name was Indian, then dot read in 32 little okay, hey.

P

To me, this is effectively.

A

Like I would argue that, like as soon as you have mutating methods and your important aesthetic, that's a general, if you would she call this put all right weird, no, because you.

S

A

It the first thing that the only sensible thing.

B

Honestly I think there's a problem here. Like I, don't know I.

D

To me, as long as they're, not extensions, read and write or a slam, dunk I mean if we agree that they're not extensions, then you go with read and write. You get rid of the the span, read methods and it kind of nicely cleans up the the reads once you make them extensions, and you worry about having said or get or put, you've got multiple things that do exactly the same thing for span versus we'd only span because of how extension methods work.

D

It's just easy for them to not be extensions, and then all these other issues seem to go away. I.

B

Think for me, if they are not extension, methods said, does not or I wouldn't like laundry. You know like say I it but like it's just so we set into little endian data like I, say.

B

Three, I I'm fine.

N

Would be done? No, it doesn't work it doesn't. It doesn't.

A

Really yeah Pano's made the point it's like you know. People can obviously write their own extension vector, but if you actually want to write potential methods, they do it. Thank you or would you call it.

B

One of first extension methods that advanced would have to be by that span. Byron I, don't know how they would even work by that span. Yeah it's a bit I think we are now discussing something that, like somebody, would need to go I think through ok, now we're now we're.

AD

F

If right is always string, basic things or like, if we had a thing that took a bite or a great.

K

A

What I think will be going on the unsafe type, I think reading and writing from memory? You know you could argue that they also want advancing anything because you're only passing one point when you write to that location, there's no pressure either I think they're following the years that span of T is like kind of like both like a buffering. It's like you know, super lightweight, screams, I, think that's where the implication of like you know what happens to this kind of I.

A

Mean personally, is it like the only thing they behalf of the friend isn't as far as naming goes, is read/write and gets it you're the only to think we have I'm not aware anything else, so I think you get to decide a finger to it. It seems like between the two. We would lead to what we drive.

V

B

I think that's all they are sending questions. You can poke I'm telling you guys you're missing such a big opportunity make a history. Nobody can ever use these names.

B

We can peek yeah Oh in the only understand you, you see it, it's exactly the same thing. We don't have hope. This day we have poked, the bear bear got poked, we ready, ok,.

I

B

Going to advance suggestion of adding Microsoft Visual Basic extensions with beacon, Pope I mean this. It's an opportunity. We cannot just pass.

AC

Ok, some on two minutes on one sugar.

Q

70 we started late all right, so I.

C

Have numbers sorted by old issues, so 2 2 for 17, which is add, converts and based api's from most even.

P

By age, that's.

C

P

Z

Convert span based api's down down or.

A

C

When they were moved to a PRF or of you.

Q

That's one of the but.

C

Due to 4:17.

D

So these were separated out from that master. You know long long list of API as we talked about, because there was a whole question around this transformation status thing and we were going to hold off and talk about transformation status in general.

X

Well, what do you mean by this well, are we ready for tea.

B

So well, we have a definition: yeah we first transformation status.

X

You know we can look at it now. We're cover clap.

A

X

Call operation status, Andy, yeah.

F

The question on this is: we want a transformation saying.

B

If you want to implement, for example, fall back on writing code, this.

A

Really I kind of remember that they were like taking like war arguments. Maybe there was the point, maybe.

D

Maybe when I suggest this one's probably gonna end up taking the rest of our time, maybe we do other things. Okay,.

Z

F

Going to say, having worked with suspense or some spam stuff, since we talked about this I think that we should probably just do what the existing convert does, which is it returns to the right answer. Rivers and illegal data is.

B

We then cannot implement for bucks, but.

F

If you're adding a base64 transform in the system that buffers not text, then that would be the one that can do operations. This is on system that convert where we're taking all the fighter, great things and adding a span based on the existing one throws this. It.

Y

F

Just match it semantics.

B

So you say: okay, you're, basically saying unconverted, we would have simple api's that don't let you do like conversions of discontinuous buffers, it's the same API as occurred in the past, which is you have to pass the buffer and it converts the whole thing either succeeds or fails. If we need api's or discontinuous buffers, then it's a very it's a new advanced feature kind of like the one that we just discussed. It's advanced for advanced developers. It's a bit more complexities. Okay,.

J

Yeah I think that's, that's! That's actually what we originally.

D

Had in 21 to 81, these were basically try methods and they would do what we've done elsewhere. If the destination isn't big enough, it returns false. If the data can't be converted, it throws yeah and.

F

I think I'm, the one who was pushing for transformation status and now say.

AE

W

C

F

Since the I've since internalized that try me there is one problem that we report him is common open.

C

The depends on that one probably bottom of us could not.

F

C

F

I know internalized that it should just don't want invalid and.

B

I also think that so you know like we have, this incur effects lab the transformation start and safety is and like we did it like a pre review a couple of months ago about this, like it, it's very difficult to write code to actually use the status correctly, because you have this multiple batteries. You need to know how to then slice them recall. The idea again like this is not any general purpose. Api for the you know everything about open. It just wants to play 64 encode. So this.

A

One here what dependents will be actually have from our stuff on that almost caught the only well.

B

So there there are AP is in correct exam that we will be coming to the review and they basically depend on because they have to support like I, have multiple batteries they and that contiguous in memory and I need to be able to like transport them. For example, it won't basically become part of cortex. They would be. The plan was to be in system partners that the owner, just like the idea that suggesting you, except they are in the text 7 namespace.

B

So that would mean that those guys are this guy's, just one inner portion, zoom only optical system, public, say if we, if we decide to put all these ideas in system buffers or force.

B

All the time so, okay I'm trying to avoid describing too much complexity. So there is this existing assembly called system text. Well, encoders right. We were planning to not add new core effects: lab api's to do URL, encoding to a new assembly, new type just and augment the existing types to be able to do. What, like castra needs to deprive if this type doesn't live somewhere, low, low, undefended and.

A

B

System web encoders would need to depend on system batteries that DLL, which I don't know if we want so zoom like that, wouldn't cover both the enum as well as the interface right I think we should postpone it through the next video when we actually break api's depend on it and discuss all these issues together.

A

It seems weird to move the England about moving the interface, because why would you need the energy body, because even static methods return? This? Oh yeah, that's right!.

A

Alright, so then, let's postpone this one units, well, the other one. Okay, so.

C

Any order now add cancellation, token registration, token property to free h28.

D

Hey a common pattern: is you register with a cancellation like rate synchronous operations? You register with the cancellation token, and then you store a long. You know you're sort of asynchronous state. You store the registration so that you can dispose of it later also end up, storing often the cancellation token itself, so that if you need to throw a cancellation exception, you can do it based off of that token, and that means you're, typically storing both the cancellation, token registration and the cancellation token. Even.

K

D

Cancellation token registration already knows what the token is, in fact, there's an internal property on cancellation hook and registration called pokken. It's just internal I'm, just suggesting we make it public. Please step back to me.

E

B

See some people can design simulate the eyes the big ten seconds this.

T

A

Today, alright next issue, all.

AB

Right do free eight, seven, nine for.

AC

The eight seven nine.

AB

Pieces costing buried only memory to memory D hum.

D

Scroll down, because this one we've got back and forth on around it ties in with other things around how we expose unsafe for dangerous operations. Oh yeah.

D

Did you break out no I'm? Sorry I stopped talking so right now, and so the actual proposal was that we need to have a way to basically treat a read-only memory as a memory when we know it's safe. So the example that I ran into is sock and a sink event. Darhk's supports both read and effectively read and write operations and it stores the memory you want to use, but for the write operations you want to be able to pass in a read-only memory, which means you.

D

It needs to be able to treat or read-only memory as a memory. Now we can do that with unsafe. We can say unsafe as and cast from one to the other, which requires us to know that it's actually safe to do because the layout of read-only memory memory are the same.

D

The better thing would be if there were api's in the platform that did the conversion for us such that it wasn't so fragile, but it's still sort of you know, logically and unsafe for a dangerous operation or a marshaling operation like we have you know elsewhere in the platform there are other quote, quote: dangers or unsafe methods currently on span and on memory, so, for example, from a read-only memory getting the writable array that it wraps or we have these dangerous, get pinnable reference methods on span which are used generally to get a reference that you can use fixed-width to get a pointer, and so the suggestion was find someplace.

D

To put all these things together and I was suggesting we do it in system run time interrupt services on some Marshall class Christophe suggested I think maybe just putting them all together. I think someone else suggested, maybe just putting them all on the existing Marshall I and suggest to two different classes, but it doesn't really matter it. Would we just need to pick a place I.

B

Would prefer to not actually have them in system memory, and maybe we can afford that because we don't need these immediately like if you don't have them in the package, then you just you have to use the you have to really implement that which you can use it on set.

D

Say sorry: first off your tongue is specifically about the cast, but you might get animo reference. That's.

J

Necessary yeah so and I think the only one that.

B

Is using directly its get Beneful reference? I, don't think! Oh, maybe they also use the try get today.

N

I feel a little bit odd separating these from the actual memory in Spanish, because they are making assumptions about the layout when they're doing unsafe, stuff forecast and if we ever did change the layer I had to do. It would be nice that these were lockstep of that.

B

Which one in particular, you think depends on as.

N

Memory thing making assertions to the taken read-only memory, how its structured.

N

I, don't actually do like the idea of hiding these dangerous areas under these Marshall statics I think does it go India I think they should probably be living with types next to the type of situation simply similar.

B

Except the types live in two places, so I don't know what means next to the dice I mean at the one of the wise or nice. This one.

N

So if they need to what is the espero yeah, there would be implemented at this panel next to that span and.

B

Anything we immediately forward.

N

If you yeah, if you want to yeah granny key moments in this system, memory like somebody will just merely for them now, then they cannot be an existing type. Marshall. That's one indication. That's right! Yeah we would I wouldn't promote Marshall, so we could put them on a span. Marshall combined these two potentially I would marshal the.

V

A

The only option is you, you know you guys.

A

To do what I mean what Stephen suggested, but you want to be able to do this happened. There were pictures like. Oh, this is not the thing way like Oh. Clearly, there needs to run time feature so I have to wait until the thing is in box right, there's, no reason why the Autobots were standing. The.

B

S memory Steve needs to use it in socket event. Arts right there is no other usage, so I kind of like would prefer not to move it because the only kind that means to use it is an internal API. It's not even doesn't mean this method doesn't ever even need to be public.

D

Make it public- and it's just me part of the platform I can just use unsafe as and we just need to you know, comment it heavily that those API is better keep the same layout Bing.

A

Gives hell- let's not loop unless we have to well but I mean I would agree with that. But I would say that if the openness feature I think one one issue, they run into complexity. The trouble was that we had disjoint feature sets from the get-go, because that makes things more complicated because my department, depending on the Evo and.

K

A

Actually using because you have access to the future, they don't have so we say if a feature is I, think it's better to say it's OOP and then the whole feature is available like unless there is like massive positive, the feature about being able to, but I mean it seems like. This is a very simple API.

A

The only downside is it's only new time which I'm honestly using the Interop services is not necessarily an area where, if you open up the namespace you like, although the beautiful set of API guys we have like I, don't think that's that's actually letting spend any words, but it does help of our like general strategy of like having the feature set available about places. So.

B

Another argument for disposition is that, like some of them, we could technically not will not make the public will do. Something then gets been able reference, there's no escape from it and probably get try get array. We could maybe have it, but, like there's, been a boo-boo references, you're not.

D

Well, good, oh, it's possible I've dynasty to service by combining these all together like this. There's that a separate discussion going on about do. We want to be able to support kind of fixed with arbitrary types, in which case there might be some sort of pattern, in which case we do you know, would we need this?

D

Gannett get pinnable reference thing actually back on span, that's an instance method rather or as an extension method, rather than rather than just as a plain static, and if we, you know, rule those out of the discussion, and we say we don't want to publish as memory, because there's only one use case for it so far, and it's internal that just leaves try get array and do we want to introduce a.

E

New class, for it I mean the scenario online.

A

Steven I mean I, see that it's only using one place, but it strikes, and if you want to write efficient and small objects and there's nothing fundamentally thinner than vh1, like other people may actually run into that, and other people might be asked actually maybe have to pull something. That is, you know in a different layer, then.

A

So that's why I would I would avoid doing this thing as an internal matter API we can it.

B

Seems, like you know, inviting complexity down the road eventually, but I think what Steve said is very interesting. We we have discussions about this big statement, which would mean this whole test been a bowl or reference made literally be an instant legacy. That is true.

A

But what is the reason we can't do them on top.

B

Because they they have a accessed span. Well, the argument that steep rate was, let's imagine, C 15, it says you can, you can add, support for pinning to an arbitrary time. The way you do it, you add some method. Let's say it's called get been able to reference to your time. Oh well, we can imagine them doing it, for instance, methods for static methods.

B

They cannot do it, I mean they would have to like search I, don't know what they would search to find all static methods that match the name and, like it's super dangerous for them to start searching the whole space for all metals.

B

They would I, don't think they would do it. They know.

A

What they do is when you say away something they say they see. What has to compile this. Like the expression, you await their basic to get a waiter on it and they say we don't care I'll get it.

A

We only subscribe that there's a certain shape of inputs and outputs that we don't have, and the name has to point so technically pulling the exception. That is okay, but I mean the question really is like go with the stuff.

A

Live I mean personally for stuff, like that, like 808 er, which nobody ever calls directly, they basically live in compiler services, where, whether you have you know one type or feature you have this one grab bag of type 400 features, I, don't think anybody else arose, I mean that would be the way why it may not matter at that place. So what about this.

B

Weight with this API I'm gonna explain this discussion to Janet and tell him that like this is something asking whether they would like be able to use this, because, if not like literally shouldn't, be adding this idea. Well. I know another instance of a would like to have that which.

A

Is the middle array.

A

D

I think my guess is just from some offline conversations with Jared I think he was in general against the idea of having these dangerous api's, as instance, methods. So I think he would, in general, being fan of them being separated off somewhere. I would agree with that.

A

And then have some magic language, syntax, like things.

Y

But it'd be interesting to see what he says to your email, Christophe.

A

Yeah my typing yeah.

A

So if it just okay, the bottom part then I'll be happy with that, because personally.

P

I don't see involved with that both type name as well.

A

Its location, I would disagree both ways. I would put it wherever men with it disagree with me, no I, wouldn't we should put forever.

B

Hopefully yeah can you just name the time some some called substates like let's say the get been able reference? It's okay as the static method we just added to the same type.

B

So it cannot be called memory matter because it's over span, so it should be called partial yeah. My.

N

Name is Marshall.

V

Z

Overturns such as chain right.

B

So I would call it buffers in the buffer because buffering like to me, it implies that it's about some specific type of buffer and buffers is kind of like you know. There are some like buffer related ideas.

L

Cuz I mean we get rid of the dangerous track array on read-only memory. Yes,.

B

We basically made lots of Dangerous Ideas, not dangerous,.

B

Since they are in compiler services, nobody will find it. They are not the babies.

L

But interrupt services here, Ike well, these ones are going in top stresses. So what about mostly employed in.

D

AA

Interop is inherently quote: quote on the safe, so should we add.

L

To this API also like Google like star just adding these also removing the radio safe guys from memory. Yes,.

B

That's implied yeah.

A

B

Never leave these eight guys or they exist. They are own different types. This proposal.

C

Iranian ship.

B

Right so they didn't ship, they were just about the I incur if X now as approved api's, so we are DC re. Basically, yes,.

A

Any yeah to me they said that supplied by the way. It's really good.

B

By the way we now we have the cold in light, you know to repose and then Kestrel depends on it and then signal are like any of these changes takes like two weeks to I mean this is actually a case study for, like too many repose just add so much cost all.

A

Right, it's not to free.

Z

A

Z

Two two three, eight eight three, two.

X

Z

Eight eight three, two two.

I

Z

Eight three issues: oh good.

C

So this is the stuff we had a few weeks ago with at eight inventing we were, we invented the method parameter generic method parameter. So it's basically.

A

C

But this is just like you know: I think you know thing that will allow you to take that which one it is right. You know to distinguish the Pindos.

V

Little baton, where is it it's in very bad movie,.

A

A

Oh yes! Yes! So this is the faculty and I! Think I! Don't think we designed this part of it, but you.

C

Know they didn't know about.

P

A

And somebody's going, you Oscar.

C

A

And against a reversal, because everything else is virtual cats seems like never doing Eric enough, that they don't have to be virtual right. I mean this. One is already virtual I suppose it has to be because it's over and it's just the semantics. We have.

X

Good point: ask why.

A

This nation lady, has midwives easily because that's frustrating I just scanned for virtual, like otherwise I'm sweating. If you write it up and then people don't read my stuff.

C

He has a actually at the end yep. You know right about the naming the last sentence.

A

Is that actually first message expect right? Oh.

R

P

X

P

Football yeah I'm.

X

A

P

Know, let's ask you know.

V

It's discussed change.

C

Just the URL 23 916.

AC

You know, proposals I fold, it that's nice, I.

B

See this is basically casting how to read um yeah.

A

The same thing, that's even was just.

P

U

We only house it I.

A

Mean rest is no.

A

AC

Don't think is fun, you know what is called just my tribute.

C

Even the ref is yeah, that's not that's, not the matter. A touch hasn't been there in judo, zero right, it's a brand-new stuff. You didn't extend the metadata and steal bits.

U

Or Filipino.

C

AA

Are previous like.

C

You know three variants of that stuff right. You know so.

G

Those are breaking teams.

C

You big, they pick the third one I.

A

Mean in general it seems like funny: I mean it's on the right type and seems to have my signature, but I can tell.

A

So it's a it's.

B

A advanced API, so I, don't know how much we want to make it. It's like you use it over, but it doesn't like the method. Name doesn't tell you that it's casting from read on the earth like it. Would you know when you see it, it's like Ezra like from Walt, so if we want to make it more in like intuitive, maybe we should consider something like probably from read on the earth to read.

A

Well, you guys have.

V

None say so I think they're all in order.

A

V

Azeroth well for Wichita I,.

A

Mean not really, but you know considering what this type does. You know it is effectively a fancy way to okay. Given this, we already have as well I. Think.

AC

A

AC

Of good one more point casting oh yeah, this is actually as well yeah.

B

Yeah yeah, yeah I think thing yeah. Yes, you know what I'm saying like I already have a red eye kind of almost like didn't notice it suite on the ref and then icon.

K

B

But I guess it's fine.

B

Well, unless the C sharp team, real names read on the earth for a few, don't read only two in which I think they are doing that, then the confusion would not be as as large, because you're converting from in to that I feel only become to him. I think that's the kind proposed election now that I remember it.

B

I would only say no we're in this meeting where we decided to call it referee normally and then Vladimir sorry to implement it, and he discovered a bunch of serious issues and I think the current proposed are is to call it in.

B

A

Sorry next 1 2.

C

Z

B

It's basically thing that lets you unpin and.

B

Like on memory can call pin, it will give you one of those I think you can get a pointer from it option if you actually keep in.

B

It don't complain, you cut everything and it has a parameter called pain. So if you pass pain, it has a pointer. Otherwise it does it if it doesn't have a pointer is basically just a way to that count. So.

S

The idea, then,.

B

Is that you can actually, after the fact, discover what the original person did.

S

B

Suggested this that we had this, because there's lots of code today that, like does validation when the method enters, and it would be nice to do the validation here using this API, because without this API, the only way to check whether this handle has a pointer is to actually access the field, the pointer field, which makes your code unsay. So now you have a method that wants to just validate whether passing something to it. It would need to know the unsafe method right and it actually doesn't use the pointer it just wants to validate that.

B

You pass the right hand.

A

B

Like not just a bit more complicated like you, have to make the methods, or at least block and say, which is.

W

V

A

B

That's a good question.

A

All the properties you better ahead well.

B

A

Maybe you asking questions: should this property it just called the pointer, well I, think my question I mean first of all, I think, just whatever we call the other thing has should be matching this one and then the next question is okay, given that the curling name pin pointer that the right name is social meaning, the other one. So.

B

I agree with your first month for sure they should match the second question: whether can you get unpin pointers you shouldn't, but we have so many unsafe api's. The problem you can, like this whole unsafe type, probably do some things to get an interior point where there is none.

B

To simplify this API review, what about you just say has been pointer and I can go offline and talk to young, whether unsafe API scan. Let you create unpin.

L

I think that's best. We're about saying like as the commercial was is point pin butters, not another because has before it does not indicate if it's checking for another.

A

I mean I suppose it would be now right if it's not a pointer right, so so.

B

A

B

A

I suppose, with you construct an ever handle without the pin pointer, then the existing property would return. Hopefully now, yes, so I I think that means this API hasn't applied, not check.

A

It's just that you can't do the converting or the check well being.

A

Yeah this this.

B

Api basic is pinpointer, equals zero.

L

Right, should we not say the name be pinpointed is not nothing. It's been put enough supplies that.

P

L

Point or not, I think.

B

It just makes it longer I, don't.

A

I think, if you understand detective, doesn't my shakers kind of like I? Can me Nick? That's the only sensible thing and then the question I mean. If, first of all, we don't use a knowledge, we would have to say in that reference I'm. Indeed, there becomes you.

K

A

Is none no reference in pointer or something it's like? Well, that's a mouthful for what it does.

X

All right, good.

A

Four minutes: yeah a good one for.

C

K, zero, three nine Stephen you're up. What are you 24? Sorry, zero. Three, nine twenty-four no just lead to forty two four: zero: three: nine, oh okay, I'll, have to learn how to tell you numbers yeah,.

D

Okay, so the issue here is I'd send an email of us months ago we added read a sink or, you know, read: write, read a sink right, a sink methods to stream that work with span and memory.

D

We want to override those methods on our stream derived classes like memory stream, but when we do so, we really need to delegate to other methods on the type we the delegate, other methods on the type, because someone may have already derived from these types and then, if we add a new method that doesn't respect, you know the other methods they've already overridden, we can end up doing the wrong thing. So, for example, if somebody introduced a.

K

D

Memory stream that overrode all the right methods and ensured that you would you know you didn't end up writing more than a hundred bytes to the stream. If we then introduce this new overload or- and we don't call the existing ones, we wouldn't be using their constraint.

D

So that's all well and good, except that we actually don't want to have to call those custom overrides, because we can often do more efficient things so, for example, memories stream, dot, dot, read spam we'd like to just read directly from the memory stream into the spam, but if we have to call the override, we have to actually get an array from like the array pool, call the array over and then copy all the data back into the spam, so that sucks.

D

So what we do is we do a type check since that's sort of the best thing at our disposal- and we say if the type, if the current type is a memory, a concrete memory span do the fast thing else: it's a derived type. Do the slow thing called the override, etc. Okay, so that's a workaround. It helps the 90% case, but now let's say you are that limit memory stream that derived type you've overridden reads: you've overridden the read method that takes a span.

D

You apply your check, and now you want to call to the base method to do the actual logic, so you call base dot, read and you pass your span memory does stream, does its check and says: oh I'm, not a memory stream I'm a limit memory stream. Therefore, I need to delegate to the less efficient thing, even though it was the drive type. That was actually calling to me to say no really I want you to do your efficient thing. Does that all make sense? Ok, so what I'm proposing is?

D

We add some protected methods to cases where we see this pop up and in particular, starting with memories streaming on manage memory stream, where we add protected methods that do the efficient thing non-virtuous.

D

We call these from our fast paths and we also make them protected so that derived classes can call these directly if they need to delegate. So why don't we do it on the base stream type? Why.

J

Don't we do it on the face? I, don't understand a common email address, yeah those two protected methods to base stream. Don't.

B

What they do, they will basically do the inefficient implementation today, but you can override them and you can basically have the span path. Optimized for an artistry.

B

A

Think they were mean and why overload that takes in spam celexa. But the problem is the other existing read and write want to take either a there's to do something sensible, which means the whole problem. Is you ever devised time, basically today, only the existing ones that don't operate us and so the foam that we now have in our efficient ones like what we wanted to do like they have to do something sensible. So.

K

That the call the.

A

Inefficient ones and then the polymers that well, if we do it this way, then the problem. This is inefficient for the 80% scenario. That's why we decided in memory stream and I managed memory stream to not call the inefficient ones, but to do something smarter. If you are that time.

B

And saying that why just those two times, why not and any arbitrary stream could basically do this optimization so that the hockey coach Bob doesn't kick in ever well, because there's no way.

A

For you to mean this, it would have to do the second thing on every device type, because the only sensible thing you can want to drive time is to say if you are at that time right, so that there's no way to say like there's, no reason to say in the base stream, if the type is essentially a stream, because you cannot construct the screen, you know you can only ever construct it the right side.

A

We have to sprinkle that thing across all our spare of our sweet types, and so we only pick these two, because while there are the ones that are people using the goes where it's sensible, the ones that we already very alright the right things.

D

Chris Todd, let me see if I can rephrase your question are.

K

D

That, in addition to the new read span, right span virtual methods as I read and write methods that take a span and read only span. The weave out of this virtual is the stream.

D

We also add protected virtual read, span right span, methods that would and their default implementation would be to you know, do the array stuff but streams.

J

Could override those? Yes well, my understanding is that the ones that you.

B

Added to stream base, they are not going to read, span rights and they are just called to read it right. Correct, that's, correct yeah, and so we would add those that are kind of like the core version, so they would probably be called. You know they could be called right span reads and or they could be called right, core right side call.

J

Read you know span core what, with the default implementation of these, do exactly what your read and write methods that takes FanDuel today they mean Kathy, emre and Allegheny, wanna, yes, and then.

B

All the right types can basically optimize further internal implementation.

B

Using the same technique, you always just override the span taking methods, let the coordinates.

J

So this is okay.

B

And I think the only reason from practical perspective is that if you want to optimize the span for your for the span methods, you always do it in the same place. You always do the same way and all types can do it the same way.

Y

D

So he we basically be.

D

Right, you're just saying add these two all streams and since writing them to all streams, put them on the base stream class.

A

Basically, that means guidance. Every time you overwrite meet the Texas fan. You are never supposed to cooperate, so they're always supposed to call these guys. There.

B

No, no, maybe I'm missing something. There reads that they expend will be non-virtual. Oh.

D

The read that X fan would be non-virtual.

B

Why does it need to be birthed? Well, it's only cuz these methods. That's all it does yeah exactly. They are protected. Oh yeah. They would be protected. Ritual okay, then I think I'm, not fast enough. Depreciation, yeah, I.

S

Mean it's entirely.

B

Possible, don't get me wrong that I'm missing some, because there are so many like chains already here of like Oberon is an abstract members. You know maybe ice that I may be missing, something so understanding.

D

What what does having so you're saying we have a non-virtual Reid delegate to the protected virtual read spam, but then what is that been? What benefit does that provide versus having just the virtual.

J

Read span because if I understood correctly, you.

B

Now have other api's that called in spender expand. They may be inefficient and there's no way to override them if you could write, but if I, if we.

D

Did that like, if we can't override read that takes a span on memory stream, then we can't do the special case for memory stream, which means the read method always needs to delegate to the read span protected method, which means the read span. Protected method becomes the one that needs to do the type check.

A

So it seems like, in order for us to understand it, probably to spend another 15 minutes on it. At least you know we.

C

Don't do a lot on issues so we made it done. I.

B

Mean it probably thought about this problem a bit more yeah.

A

All right guys, yeah.

D

So just say: I think it's worth thinking about how to generalize it I. Don't people would suggest it works, but maybe there's a riff on it. That does in.

A

Any of these effectively making New Mexico, maybe that helps then ensure that actually the case, because, no matter what we do, we have a new virtual. The people can't override. We want the people, interpretation I, think, no matter what we do. We always end up having to do what you currently do with the special type checks. I, don't know what that.

A

Alright guys online, thank you and then hopefully we see you next time.