.Net Foundation Design Reviews, 12 Nov 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

00:00:00 - Approved: HMAC-based Extract-and-Expand Key Derivation Function (HKDF) https://github.com/dotnet/corefx/issues/29660#issuecomment-553037689
00:23:44 - Approved: Add cancellation overloads on HttpClient and HttpContent https://github.com/dotnet/corefx/issues/32615#issuecomment-553050411
00:40:29 - Approved: Add List(T).CopyTo(Span(T)) overloads https://github.com/dotnet/corefx/issues/33006#issuecomment-553078596
01:29:09 - Needs Work: Expose the match length when using String.IndexOf in culture specific mode https://github.com/dotnet/corefx/issues/33543#issuecomment-553082487
01:37:21 - Approved: Hide Thread.VolatileRead and Thread.VolatileWrite https://github.com/dotnet/corefx/issues/33701#issuecomment-553086736
01:48:11 - Approved: ComputeHash Async required https://github.com/dotnet/corefx/issues/36068#issuecomment-553088815

A

And here we are sill. Let's talk about this API I feel like we talked about this one.

B

So if you I don't know if Steve's on the call, but if you scroll down it's not how you think he says you could do this using a regular synchronous API. So it completely works, use the.

C

Synchronous one and just for the tap everybody I think it didn't work on framework, but we made it work on core everybody thinks that doesn't work and then they ask your and like I, will add the method here it it really just called the other one. So.

A

It just I think there's like 2 AP is right, the one so there's one API. That is basically start impersonation and revert or something right.

C

A

That one be killed because basically everybody forgets to call the revert one, and then you have facing the issues with security might versus this. One is here, you possum a delegate, and so the thing will basically do the revert for you. You don't get to run indefinitely and then cross native contacts right, but I thought we already and didn t already add that API I feel like we already have this well method. We have works so.

C

It's are you more KITT use the thread static, so it didn't work as soon as you hit in a way.

C

In poorer has an async logo, so it does.

C

Of people but yeah people believe what people believe in this method is more about ignoring minute.

A

Oh I see what you're saying so this one is basically almost ending the async overload or it didn't see many when.

C

The comments somewhere in the bottom is okay. The implementation of this will just be to all the existing method, but this one says they think the people are happy. This.

D

Is an impersonator task, yeah.

B

It's not all it's not even has not run because it's still single or it's still invokes her on.

B

Because if we return to ask, if you want to give it anything, basically.

D

So what you wanna do sync across no, so you could you.

B

Could call the synchronous version and return a task if you wanted to because the synchronous version returns a tea. This is literally a wrapper around the synchronous, other static.

C

Task of T right impersonated, you think of T, whether Y Eero run impersonated semarchy pants yeah, it's! This is literally we're just putting a sink on the existing thing because he could be saying. Oh, but there's no easy version. By now. There is I.

D

Mean is if the base API is challenging for people to use correctly. Then maybe there's some measure but.

C

I'm gonna put this back to Steve because he didn't comment why it doesn't successfully.

A

So I was about to ask like this is a very long conversation I feel like we don't have the right people in the room to make a call on this one. Well just give this one.

A

I mean father 2017,.

E

So, what's a couple more wait.

E

A

Oh my Jeremy. This is yours clearly, because there uses many words that I don't know what they mean incorrect.

F

This API allows you to.

F

And I'm consist of two metals extracted next month and start this kind of taking that key or what they call it in material and person with reasonable and the expungement intensity they actually, which is Turkey.

F

Or reasonable he's absolutely right under another kiss from that and the derive cannot.

A

So does the color now the size of the of the thing that its extracting.

B

The caller presumably understands what each KDF is, if you scroll down and actually yeah.

A

I'm sorry I'm looking at the api's and it seems like I, mean there's one day, return to bite away and which is fine, but the other one pass them to spend every by to this one year and I'm wondering like to how does the caller know how big that it's bad I see so.

F

Basically, based on the cash I'm doing the use that the office is gonna be different, but you kind of if.

A

At the call that in principle does know what size any to other value to just guess and then catch the exceptions so name, okay,.

G

C

Represent before every other people who are gonna use this idea now your nectar center, never gonna call extractors we're gonna call the right. Private key just does expect to expand for you, but there are different specs that talk about like using the extract that from HK, D, F and so yeah gets the newest public or the people.

A

To me and say: I've no opinion on HK, D F or how popular image as small as asking from the from the normal pattern. Point of view didn't me say like the color has to be able to predict this size right right to.

C

The literature from the cache over the color, the first argument is the hash algorithm name that determines the output. If that determines the number of bytes right so.

A

If I know, if I use sha-1, then I know it's whatever Shawn, sighs, all right and then nice and then the right key.

H

Also gives you something bad, but what does that output? That is the key.

F

It fills the same sorry right, yeah right. The link is basically arbitrary.

F

How do we get some epi is taking bite? Yes, we basically compared mic. 1 1 over log of Z on allocate only.

F

D

It is the reason this.

B

Our guidance is have one that they expand and then have one that they slightly yeah the the fight everyone's very we've seen in practice, tend to be in particularly the ones that return the batteries. You know.

A

There's no point I returning every to only spend it I mean like if you, if you produce a brand new Veda, you might as well give them the the actual type, because it's at that point is there is no no.

D

No I'm not talking about.

B

Yes, city from the from the surveys that we've done, like our our conclusions, have been that most people with array-based.

I

Right but then the question would be: why don't we have ones that Inc we don't expand and then have the bite, wounds or don't expand one, that's what they have here. No we've got one that returns a byte array, hates fine raisin got one that returns and makes read-only spans. We don't have the one that doesn't the turn spec there yeah I.

C

Guess I don't know Roman we.

I

Don't have one that returns a binary but takes in reto I. Think is the question, which means that you either have a slightly failure case path. If you want to use fans or you have to go the allocation, you don't have an in-between of I'm fine with an allocation of the return, but I've got memory exists in native code. Forget.

D

This is a more advanced API. Do we need.

C

So the easy moves are for languages like football, not net, probably can't work with yeah. So this is a people to mcphobe angry I guess this is don't depend on the implicit cast and to span don't like people, think they're working with byte arrays. Looking at the fight arrays, this is a convenience thing. It has no distinct implementation.

C

It's simply there for helping people at all right now you can see we have four buffers, as in the end like the one that could end up with like overloads by following every consider, and so it's the you have to start somewhere. People ask for things. You can do.

C

I, like the derive key one, there's no wrong answer. You can put for a span that it fills because it just does it for span determinant. So what I want to kind of our family be back.

F

Extract on exponent is most of the most likely people just using the drive key, but that's what most of this it's like. This is an option because the spec kind of it.

E

B

So I can't say for I am okay. It's like this back said, o foremost of the crypto stuff. We haven't really gone for friendly names for parameters we've just matched whatever the name of yeah.

J

I believe it is, it.

B

It might be initial, yes, yeah agency,.

A

Question I have is: why does the spam version of expand, she's impossible info and the other one does Maude.

C

K

Do the same same yeah, cuz no over a span is just behavior in this in the algorithm.

F

But we don't have to swap the order at this point of the arguments. My philosophy.

A

But it's already weird right because they expand one that takes a bite away. You have to pass in the output length and then you pass in the in phone, the other one you just pass info and then you pass the output span. The thing actually swapping them would make them more consistent as well I mean the shape is different, but like it's still.

A

Yeah, why is it not optional? Basically right because the other one you basically specially said like yeah, you can omit the info, and here you say no, you have to pass me one and then yeah. If you accept default, that's not visible in the API signature.

A

And let me try to.

I

And people yes.

I

Yeah, the question is just where do you put the and regularly the outputs fan? Okay, as, for example, for extract, you there's no logical place to put it like for expand their heads.

A

What's the same, as with you know, the other one by the output death comes before enforcer, we'll just move the output span before the info.

F

One of the info or sort I can't remember which one like that and a funeral is not exactly equivalent in one of the cases dealing matters and the other one like one of the cases like zeros like Spanish, equivalent e20 0, we have the other case. The moon is kind of different than I, wouldn't think the American.

E

F

25 small, like whatever their I, don't think this implementation should care about because.

C

This bank calls it out, but at least this is an input into HVAC which gets like headed with you. Okay I can.

K

Do nobody just was.

I

Football for just we can't move the span before before info. It's going to cause hugs codes, because bytes implicitly convertible to read only span. So if someone has a a read only span and or if they've got like, because because arrays are implicitly convertible to read-only spans and because the last parameter of each would be office optional, you could end up passing in to a read-only span and an array meaning to be the salt and end up with the wrong overload using yourself one.

I

What you meant to call problem of having two overloads that have optional parameters and especially when you've got implicit conversions between, which is why we say don't take emphasis. There.

C

More than one overload the popular brand.

A

Maybe I'm not seeing it but like one, would basically have an additional int parameter and now that will that expand and derive key, you can do it.

C

On but extracting yeah.

I

Tracks would be problematic and let's.

C

The sample in the automatic, but the rive has a length in the everyone that it doesn't have, because this family knows expand that week. I see that now the donkey is unambiguous default arguments that extract.

A

I mean at that point maybe Jesus. You just said that the disband ones are the more advanced ones and just nothing is offering you after we go, but then I would at least still aligned the in the output length and the and the info. 1 I would also agree the length of the end or put info before.

A

But you can't do info before because it doesn't report.

C

So yeah that so expands can be hash, algorithm, name, PRK, output info.

C

Okay, it could be in in everything except the switchbacks.

C

Salt would have to be.

B

Allow people hustling you.

I

I think if we just talk to me, probably wasn't PRK stand for Nikki. Okay I was hoping it wasn't pointer to results.

C

I

J

C

Stands for PRK. Is it the web perspective so.

B

Does that mean salt is actually by bracket question.

C

No because this library wasn't happy.

B

And I believe he was saying we should have no little enable at the top of all of the women.

A

um Well, at least to make it clear that you have thought about it right, but now I would assume you didn't, but based on the fact that it's defaulted suggest its well being the dutiful ones. Yes, what I would I would I would yeah yeah good point till you did not because it was. There would have been some question whatever.

C

Project ref CSS itself. There is up-to-date when you prototype with it's written right there yeah so.

B

Number used to have a guideline saying if a name is an acronym three letters or more, if only the first letter is patois uh-huh. That would look a bit strange here, although I.

A

Think in cooker we have given up when they, given the overwhelming majority of api's love follows that I make shall one.

C

But those wheeze yeah I.

A

Mean it's just a clumsy observer like.

C

I will say him in the book: I did add an annotation, which is people think crypto has an exception. No every single one of these is wrong. No.

A

L

I think it's just at this.

A

Point is just starting that collation company's phase also does not add, and it's.

C

Like a EES is capital, a lowercase T lowercase s, DSA is big, big ECB and little s little a short like all of the new things we have happened. Doing it right. The problem is HK. Df is technically. It stands for H Mac basics back to expand either of ation function, but it's just called hkpf, and so we can write it as big a little a little B little left, but that felt so much man you'll hear so, but.

A

All the different from Shawn, it's the same thing right, nobody says Shah, even though it's an abbreviation clearly same again,.

C

Like that, one was sure.

A

Like all I'm saying, is that like to me that's what I'm saying the thing crypto the reason I don't care is that it is already more or less arbitrary in the nature, surface right I would care much more and let's say collections generic or something is like. No, we have been very consistent here. We should you should keep that.

A

But that's why I'm saying in crypto, like yeah I mean we can go one or the other I would not I would not ask you to make it all caps, but I would also not like kill somebody over. Let's make that the rest of the framework consistent because there's no matter what you do is always going to go. Kill areas right like that right, I.

C

Mean we could expand it I call it h max the derivation function. We call it a.

C

Should probably just keep the capitalization honestly like HK.

K

Definitely looks.

C

Nicer, if it was.

C

We don't use an acronym.

C

He be Katie of one we call password derived fights. It's a cost for graphite, violates the beautiful yep once back and Medicaid, who we call RFC 2898. There are fights. You know both student.

B

Atf's, dependable things back so yes, they don't like anestine on that. Actually has a class called pbkdf2 or a method call to get media to only the fuse couple thing.

D

I need to be in so, but could this be like each Mac Drive bites something like that or what about is not extending the right place? What.

I

K

Capital K yeah.

B

C

Had this entire.

B

Argument in my head in the Trib time.

C

Because you know I like I, can't because that's my other two rules, the correct rules, I think look terrible, I, can't think of a better thing to do. There are no words, looking really glue on the end, because we couldn't put the right bites on it: cuz it's not at the right fights, Hey and.

B

Keep in mind use this.

C

One too, because they're.

B

All terrible these columns aren't also like set in stone. We must follow them like under punished. Under penalty of the API, looks as easy as possible.

K

They kind of obvious yeah this.

B

F

K

Ice on which fuses all Lord.

B

Know that you know, if we have a case said if we have a case insensitive.

M

No thank you or BB o recommend. We remove.

A

All right, so we have to remove one because I don't think we will find a better naming convention inside of crypto. That seems like a PhD thesis worth of material.

C

Hey yes, but if we called it RFC 58 69 key derivation. No one will find it.

A

Alright I'm closing the snap.

H

This is probably also something where crystal anarchist of evil be helpful. So let's just keep this one. If you would know what there.

D

Anybody's speak to this. This is just a feature that we have done. Basically, all of our acing EPX I back some sort, allowing us to turn off execution context flowing permits reasons socket a sync event. Few places.

A

I will be event.

D

This event hugs as a callback every year socket operation, it's cold. These.

B

Things are created by the runtime. No, no, this one's exposed, specifically for, as in like you, expect application code to construct the arms type.

N

Maybe I don't understand by that me know if you.

A

N

Mean the call beg receives the arts right. So how does it.

A

Help you at that point.

D

This one it's modeled this way, because it's it kind of closely wraps autocompletion ports which needed to derive from.

D

Yeah this this thing makes you call it for any of the acing methods the static, socket doc. Connect async only supports this method.

B

Yeah whoo and I guess coming back to the original question, like what application code would ever construct an instance of.

C

Two things.

B

That have got an eight that would.

C

D

Who's looking for.

O

Six crystal instructors, but yes, I suppose.

D

C

D

C

D

C

D

It there's a public, you see, you are argument using the festival is to make.

C

The boolean also point where.

A

Source darknet right now is that is it gone or is it just.

P

B

That's up working yesterday, I'm looking at it right now. My.

A

Brother is showing me the donut I will.

C

A

Funny enough API, so there it is also not comes race, actually is its back baby.

A

Okay, now, finally, there all right baby, just my machine being overloaded with including the video.

C

Q

C

Someone outside this Thai food.

C

Like it's weird, for you hope, that's not the one raising the event ever create the event. Yeah.

D

It basically just prevents having to allocate, but about Jackson all them. So if users are now allowed.

C

To specify this parameter that it'd be exposed to the property.

D

Another something that Stephen brought up is that we typically make safe.

D

Disable and that's so a property or attic method is static.

C

Method for that it like, instead of a constructor making it be a.

B

Factory, so you can stick the word on thinking.

C

It makes it but mistakes, operation.

B

There are subclass types at least internally. Yes, you can make this constructor protected and you can do that. You know that.

C

C

I

With no, but at that point, why? Just why not just make this public and then name the parameter, unsafe.

B

I

B

That it's the father we but everyone can. We want people to write. We literally want people to go on the keyboard and type unsa generally.

B

D

B

So it's good for showing up.

D

During like code reviews right reviewing each.

F

Other's selected right.

C

But again that's changing somebody's.

A

Yeah, can we just make this a property.

C

Right now, you know really want.

A

This in it only you don't.

C

Really want it in while I would have been this fireman. Somebody can change.

C

I

I suggest make.

C

This protected because inheritance inheritance in this area for this type and then they go aesthetically, is the help of method and then general guidances. If you are accepting something via a constructor or a factory that used to give it back to the user, so they can factor. This is actually the one.

D

Three method doesn't arrive it and you, basically you you can say new or you can.

D

But it makes it weird to follow the same pattern.

C

It turns out the pressing execution context is a pattern that we don't want to be easy.

R

Is it weird to have public instructors and public create methods on the same class? Yes,.

C

Well, let me do it sometimes it's not unprecedented, but again the the the problem would be. If you just see new socket a sync event, ARDS of true in a code review, we might be like I wonder what that boolean is, but then like I, probably don't care, and then you move on and but if it says, socket event, args dot create unsafe. Whether you're like what are you doing, it could also make it.

B

Really wanted to have an instructor of a stomach.

A

Because that's not a bad idea, because then we can name the you know, member and say whatever, and then that's actually in need. That's actually a good idea, you're, not using anywhere.

D

S

Yeah and it's a little wet but re do.

I

You have it as public anywhere else. There's a.

S

Little there's.

I

A champion language proposal that enums will eventually become context-sensitive, so you don't have to use the name of the type to access the members of the enum, provided it's being used directly or a parameter or assignment I. Think Neil's. The one that's champion that, if that it then and you're gonna be in the same boat in which case just eat the bullet. No.

A

Because you still have to app, you can still call the member unsaved blood flow execution context. Cuz. That would be certain type knitting, remember yeah. It would not name the type something else that.

F

You do that what they're using yeah.

D

I think that's not a bad idea that just the consistency.

C

As far as I know, this is the one place that you're using that in them you know the public API I.

A

Mean in general, like every time somebody has a public pool on sorry in a pool and a parameter like generally speaking, we try to avoid that because it's always unclear what true Falls means from the cold side right I mean in my own code. I usually use named parameters so that at least it's somewhat clear or I sign a local variable of some meaningful thing. Otherwise, if you call a method, you said true, comma false, like oh, that's, just great I'm,.

T

A

We have a ton of enums in the VCR they'd only have to memorize, because we just wanted to name them. I mean that's. Not it's not there's enough priority for that. I think.

R

Value test source on completed flakes, as this is an e new. It's a Flags Inu, there's a use, scheduling context as one value and then flow execution context.

R

So it's not unprecedented that this is represented by any new yeah.

A

I mean if you already have an enum, maybe we should use an existing one, but I don't think we have an existing one that actually mirrors what we want to do here.

A

G

C

Have a very yeah.

R

This is for value tasks, so.

C

It'd be pretty weird, but it's not get extinct event or use a thing of value. Hat yeah, yeah.

A

I wasn't talking more generally so yeah.

R

My point was just that that that's how it's represented a horse publicly.

J

J

J

J

A

A

A

All right, oh I, will use me have not really decided what the name is I will just in a placeholder name. That makes it very clear that it needs a name and I will bump those beggars needs book. Then somebody can think about her name talking under school. Please underscore and give underscore me underscore.

E

A

Where's, your taste I ain't fixing that yeah.

D

We have these methods that are kind of important content class that don't have cancer.

D

We have these open methods of all solutions. This.

A

After all, these methods exist, but none of them take a cancellation, trophy yeah, they all exist and why would we default them Anatoli then there's an.

H

Overloaded taste now right.

M

C

A zero Argo of the zero argument overload with him.

A

Still seems weird, though, thank you like it's optional, but you can never call anymore I already.

E

A

I already have other cases where we have patch async and post acing. They basically has an overload given without cancellation, to open without them being optional. Sorry, if in the Tiber or it would be consistent anyway, just.

C

Let me make I think the only difference. This would be better if you are marking the existing ones as editor browsable.

I

um Editor browsers will never it doesn't impact compiled arrests.

H

C

No ambiguity better than we would just be a better man, but an intelligence would be like in this supports inflation, confusing.

C

Because if you make the other method, just call this from the big argument, then in most applications won't show up to do build this early.

C

They invited recalled get by the raids.

U

So we recall cancellation token. Do we call it token under require consideration cancellation.

B

So they see the content, has a Thursday read.

E

D

The client house for the issue, yeah I, think some use.

D

Oh, you mean the birdcage on the HTML content.

D

Turning by Therese for everything.

B

It seems like just.

E

C

Although it's on content, yes, there's a retouch freed I, think.

P

But instead of get, it should be read as or the Asian content.

P

P

He's a biter, a stream Rosie other ones.

A

Lauren Sienna: did you see it for the first 15 minutes.

A

H

A

O

O

E

D

C

B

They're not more.

O

Of this public.

B

Alert sounds like you want in that room for a source chain.

B

The process, because.

C

In CS never come back, we had it once before. Ray array.

C

Oh, the 17 food was because of our race.

C

At 18 she's gonna come in yeah, it's that wouldn't be charged. That would be another 17 over.

C

Because it would be ugh screams Oh more than that to get utf-8 string and utf-8 spam and an attorney. Oh, it's like 34. More of those all right.

O

So way to kill a.

B

Futures doing just fine on something doesn't be good over there play.

N

Like four times all right, Eric is on.

R

Yeah, so this one came about when I was refactoring, some ml dotnet stuff to use spans and I noticed that list of TS copy twos that take in arrays, but they don't have one thing taking spans and like this, this may have been nullified since we've added the get span API, but that one's like in a hidden memory, Marshall place I, remember directly.

R

B

Copy to I think is an implementation of the I collection. I wonder if maybe this could be a use case for default interface methods.

R

Right Santi Santi asked the same question: I must have missed it. The problem is I list. You can't get at the underlying expand inside of your I list right because.

I

R

What you want to do here is a span copy to span when.

B

She's gonna be.

R

The fastest Jim.

B

Like a Jim would allocate if we just had a default method or we go or a wreath or whatever, but I mean list of he could implement it and say: oh I know how.

C

Essentially, if we think of what we would do, if we additive is a default interface method on a collection of tea or a copy, and we can do it here and then later decide that we want it back. I, look to the collection so.

D

The default implementation.

R

Imitation just loop over the thing right from index to count that's true. Yes, we will.

M

Make one element.

R

At a time putting it into this band, I mean.

D

Most people assume the best it'd be confusing yeah.

B

That's that's a fair point, so he may be a.

C

Positive, adding.

B

A new primitive, so maybe maybe we can add this to list of T and some other very specific collections. If we can just make a bioscan.

A

Normal collection, sizes, Dalek would element-wise and a copy really show up, but most collections are like eight elements or less and I fully follow band and Men copy is about the same. But this.

R

Was or this was being used in ml Dannette was it was building up like a list of floats woofer machine learning right and then copying it into its buffer when it's done, and so it's not at least the use case where I bumped into it was in a case where the list was relatively you.

A

Have seen adding that on the concrete type that list of TV we know what were Foreman's prefer.

A

This makes sense I'm just wondering how much value we get out of getting some default implementation on an interface I mean the only benefit we could have is saying, like you know, for most collection types element wise, you know, writing it into the span is probably going to be about the same performance as you could have done it anyways and then, but there might be some types that specialize this and then just do it faster, but you're not getting worse. Performance right, which is the double copy, would make things worse. I.

D

Think it would be okay so long as we also said that we're going to build that information on the build yeah, they stall, the.

C

Important, like oh yeah, those three isn't really going to be able to do better than that I'm doing its own walk and yeah yeah.

R

F

Is the personal first gay party.

O

This suggestion.

A

I mean list of tea is the work of thing right. I mean we like this stuff. She has received a lot of like ok lang spent, for example as well. Like I mean it. Is the high performance like build a buffer kind of thing like I'm like that's? How do we start with least of T? Was the first yeah.

C

I was about to say it, I get more.

A

C

Here, if we think about what we would want to look like on the interface, we had it here and then again still reserve the right to go back and add it on there. Yeah minimalistic bound by have anything.

M

E

Let's talk about that first.

M

Over line, what's the company yep.

R

So it like, if you look at the existing copy, to UM overload rate like the difference here between our normal span to span api's is normally you you get your span sliced and counted your source span sliced and counted to where you want it to be, and then you copy it into a destination, write button list, you don't have a slice operator or you can't get like you can't get where you want, and so that's why this overload exists.

R

Is that if the index is the index into the list and the count is how many elements to count and if you just have a list and then a destination span, there's no way of seeing both of those things, but.

I

Why do we need count? Why can't this? Why can't the user just place the spam to the number.

I

So what was that.

R

Because if you want to do a sub.

I

Right so I was I, understand the spam. I understand int index spam t span I, don't understand why we also need to count because you could just have index and span. Then the user can slice the span to be appropriate count. You just feel what.

R

T

Where you have.

R

A list is 10 elements long. The span is 5 elements, yeah.

B

Right, it seems more like in heroine day well, I mean right, like.

C

Yeah, so one of the things is: if you're filling a span, you must always return the number of things written to it, so they can't be void. They need either out the answer or return it in probably a tournament, and then the question is: do we say that the short span is I want less than you're like this?

C

Is the buffer I'm working with just fill it and return the count, if there's still more data, or is it an exception, as if it's an exception that when you try copy to so like there's a whole bunch of behavior? That needs to be defined for this method before we can understand, but it needs to do this also.

R

This banjo properly.

T

C

T

One yep for the first one.

C

Clearly, if I give anything.

R

M

R

Does even the current parameters yeah.

B

I would expect this to behave identically to the existing array-based overloads like if I give you something it's too small or throws money yeah.

I

But then there's also the the case of you: don't want to copy an entire large list with span. You might want to just copy out segments of it, in which case, ideally, there would be an overload that allows you to do that with it returning the number of elements copied into it, if there's less than filled the span and potentially and just having that, you can then say how many elements are left in the in the array from the index, you can say how much of the span was used, how much of it still available.

I

So the looking.

C

At the one on list, the one it takes just an array and then offset the middle one on the existing that one who calls array dot copy, passing the list size into the fit aside. So if you give it to smaller, go through the top and, of course, respects the top. One question is which one do we think that the span? One is inherently array index and count. So it's the top one or is it the middle? But.

B

What little be even better is, if you are intending to call the first very, very first method and you accidentally leave off the last parameter. You get this panel.

O

C

If you had the second one, you added the index and now you're doing nothin yeah.

E

R

Mean we could rename it right and.

E

R

Overloads with the hurry of you.

R

A

You I mean if, but would the change semantics or do the same thing which.

B

Instrument, it depends on if one of them frozen the other one right, that's data, because the the semantics will change, because you say index comma or a comma array index, meaning I'm intending to start writing to the array. At this index you accidentally leave off count, and now the array index parameter that you passed in is treated as it because it matters.

F

These you just over hold.

B

F

That the first envelope with all that comes from this one and index and.

N

They're learner-centered IQ scenario, that can you say so: coffee 2 is 0, comma new.

B

Int array of 10 in hama, I yeah, which one of those 5 overlays.

A

But you're saying the switch from array index to count yeah.

B

Is the I have I have passed you a t array, but I've only provided three total arguments to the array is implicitly convertible spanha to you so yeah, but is count.

I

Optional: okay,.

C

Well, but it's the in the top one you have to. There is no three argument. You would get a compile error, you don't get a compiler when writing this, if the other one. So it's not this, isn't that, like you recompile in your behavior changes, this is from time you walk out along Pat, so this actually violates the rule of you have the same type in the same position of the differences. So this is not a legal over call.

D

The first one copy.

T

Or a copy to span both of them copy to span.

C

And they were gonna make that order be whatever we want. Instead of trying to match.

R

Did we settle on a lot of whining both an index and account overload to this method.

R

Still questionnaire of whether we should have this at all, what.

U

Did they remove cone from the span overloads what why does it not work again, but.

D

It doesn't expand as.

I

Basically, the question is: do you want to require the span to be at least as large as the offer, or do you want the user to have to explicitly say account you want if the.

C

List has length 10 and you give it a span of a 35. Yes, does it return 5 for 30? Second, one.

A

That kids, it would have to just feel fine, but it would.

H

Be like what about that's.

I

Not oh the array, one yeah, but that is how most span stuff work today.

B

R

Partly because you can slice the source down right and all dispensed up today, because you can slice the source down to how they won't want it, then you can make it you know copy into some destination. But here you can't slice it a list right.

B

If we were able to slice a list, then we would definitely say Oh list of speed all sliced I'll call being cold today, right.

I

But the question is: do you want to force the user to allocate a span? That's big enough and in the default case it throws or do you wanted to fill by default and say how many were written and then the user can say husker an index number written might.

B

Think- and you would use abilities sake, I think this should probably match the span of T dot coffee, which is I'm going to copy everything we're in going down, and the user can specify parameters as proposed Bureau to say no I really just wanted some and.

R

Also, how this is a teat copy to array work. You.

S

E

R

T copy to array will copy everything in the list to that array and.

B

E

R

Not big enough I'm gonna say it's not big enough. Yeah.

B

I think the common case is going to be. You want to copy the whole thing out, and it's probably unintentionally, the destination. That's too small, they think that's gonna, be in common agree.

R

I think it's except I get this. You made an error when, if you did that, if you, if you gave out a span, that's too small.

A

And just adding the ban index, we don't want to do.

R

See there you can slice that span down yeah yeah.

A

You can but I'm saying like. If we do that, then the the ambiguity goes away. We don't even name the method. Yeah.

B

But we don't have any other API us to take a spat out because we expected.

H

It's just what's birthright.

B

This is an interesting problem, because this is potentially defining a pattern, but we're likely to use throughout the I.

A

Just don't like the renaming the matter yeah, we don't have to.

U

Well, if we don't, then yeah.

A

Steamed sleep weird, though, I mean.

B

It's we could just take the hit of the potential ambiguity due to overloads I mean what's weird on this is actually the top.

C

Copy the top copy, too, should have been array array index comma list index I'm a kept.

O

Like that index is just in the wrong place, because.

C

A

It an almost worth comma destination.

B

Well, what we could do if we wanted to avoid the overload madness is we could do int index in count. Spam I know it's the long order, but there's no ambiguity. What about the one that listen thinking eat? He said, let him be with sanity.

I

B

You're right yeah, they have the exact same behavior, so I go.

C

Yeah and if it's through Georgia throw and if we wanted the heart at zero and go to this family when you come, but the span copy, twos I mean the irate ones already should have returned to the engine.

C

Camisa go, but our proper.

R

C

Because it's you.

R

Copies all of everything in the list right, unless you you you unless you've explicitly said the count it copies. Everything in them I mean.

A

You know how many would call be very clean right by just looking at the call sign, I.

G

C

Modulo reading, but yeah yeah well yeah turning is involved. Oh but um I, just like our buffer guiding sense, I guess the count must be determined. So if it's yeah.

I

The count properly but.

C

Like we're burning as far as I understand method call patterns, we burn the a register by calling a method. Anyway, we may as well just put the end.

B

Yeah I mean once we come span like.

A

Know what what do you think of the rest? Addition in the Nexen counts man that.

B

Seems fine, though I mean I'm I, wouldn't I wouldn't do it as standard practice? The only reason I even suggested it is because it solves the Ambika, a downside.

E

B

If we ever wanted to introduce this pattern in the future, like to other collections, we would hopefully follow a more sane system or a more sane parameter.

A

I mean I, think everybody everybody who does copy probably has the three overloads.

B

Pacifica list.

C

A

Some of the interface I'm saying, if you want a specific type in your probe or Ian tonight, wouldn't be shocked if that exists out there. That's very first overload with those same.

B

Parameters in that same order well.

A

I'm not sure what the order at.

B

Certain electronic guys in that meant no.

A

You cannot it's.

P

Right there, but then bad things, and you just try coffee how many possible heat the ice could have a more than to anything.

B

uh I, don't think I, don't know that I don't think so on.

I

The eyes of dotnet there's only like 30 to 50 thing.

A

I shot 200, so it's clearly less than that. No I think that's the total list of things. We're yeah.

B

I think this is just someone that wire, hey 300 and that's probably what this would be.

A

But then are you fine with that, then.

B

And if we did want to make a like an icon coffee to spam interface on list, we would just have an explicit.

B

H

C

List doesn't have like a list.

C

The copy series right now come from my collection. My collection doesn't have one that takes off.

A

We want index in counts, pan common Indian excrement come.

B

Spam comma int index comma in Celtic, looks with those second.

B

It does because, in the other, when.

A

You one would have the exact same engine when you were tied. This.

R

Intact this index is the index into the list, and the one up above the array index is.

C

Indexing, that's the writing. Offset.

H

H

C

Because the semantics that we have with the array, one is the one that doesn't account for you have just not changing the semantics. You have to have a sound that doesn't work because it is you.

G

C

Let's duck copy to the list is 10. The array is 5. That grows an exception. Look at your destination. It is basically.

A

The end with the rationalists, if you give me a buffer without specifying sizes, the assumption is the buffer has to have enough elements to store the entire list. And if you, if you don't want that, if you prepare to get partial results anymore, you do the math. You tell us how many things you want to have. How.

B

Often, is it that you're going to be talking partial list so supposed to be a by your listing whenever no.

I

Word in the list.

L

I

N

I

You want to trunk processing of it, I mean you, you X does this? Sometimes you just want to parallel lies or data.

B

What I'm, trying to reason about is, if, if we could say use collection marshal without actually driving a significant number of customers, because the typhus on her.

A

I mean I would say the reason people use Spain here is because they clearly wanted you some sort of bulk processing the performance matters right. At that point you want to chunk or believe in.

C

Doing stuff you want to operate on like five things at a time you can do it yourself with the enumerators.

B

But this saves you a bunch of Airy lets. You be more to Claire m2 and in those scenarios you specifically want to copy list data to another span for processing, rather than just get a span. That represents the data already in the list. Well,.

I

B

You're supposed to use.

I

The unsaved class, if they don't have to- and they could be doing, they could be having updates happen to itself.

B

Dates happening to the simultaneously.

A

And it's kind of listen here and if more than one Fed is involved, there things with yeah.

B

As what what I, what I'm afraid of, is that we're going to introduce an API that doesn't actually solve the performance needs of people and that everyone's going to use collection Marshall anyway, yeah I, think? Why did we introduce us.

A

I think this one I mean era can talk more to it, but I mean there's plenty of scenarios where you really want to copy the data right, because we basically you're just using lists of T as a builder and then you end result goes elsewhere right, in which case copy to make sense. But then the target you may not have any of it as a T array right.

R

That's exactly what happens in ml done it. Okay. Is we have these things? We have this thing called a B buffer that exposes its data as as a spam, so yeah you're, building up I'm sort of data and then now and I'm done building it. I want to put it into the B buffer.

A

Okay, but I think you have a point right. If there is a way to get the span out of the list of T, then you don't have to add any of these methods, because.

B

I, don't want to I, don't want to drive people toward the Marshall classes, just as a matter of course, yeah.

B

I think we all agree that logically.

R

It feels it feels like this. It feels logically like this should be here right, like it does yeah almost everywhere, I mean not everywhere else, but in a lot of other places that took an array before and then array index and count. We were like, oh let's add an overload for span yeah.

B

So what we could do, I think we're all in agreement that last overload makes sense coffee to just stand with no other parameters. The only one that we're really torn on it sounds like is coffee that has into spam hints what we could do is we could rename that one very specific method to something else like coffee, partial I, don't know, and then we say okay, then.

I

B

And that method now has different semantics, including, like you, tell us exactly what you're attempting to coffee will make a best effort to do so will return the number of things that we copied right.

C

B

That one, the one that only takes up spam, is the.

C

Only one that we would possibly do is the dim yeah. If we're looking at this from my collection, that's the only one, that's a candidate and when we're looking at it from list, then we're having the awkward overload, which means that maybe that would get at the method so.

B

What if we took that first of the first method in the new proposal and just renamed it, for instance, coffee too partial and said that API, like you, give it the offset that you want to coffee from you, give it the spam that you want to copy to and we'll return the number of elements of like it's. The method name is different. It has a different behavior than the rest of the coffee to it, and you just have to know what the toys bring the dogs.

D

G

But was it collusion religion standing.

B

So it's not a conclusion, just a proposal so.

G

B

The first of the two new methods, it's being proposed coffee to into span int, rename that method to coffee to partial, remove the last parameter just have a starting offset a span and it returns a hint saying. This is how many elements I copied it doesn't have the same, throwing behavior as the other opportunity. It could.

K

Be essentially endless, I.

B

Mean it marks a minimum.

K

Minimum value of family is in. We we've.

B

We filmed the buffer as best we could. This is a number of elements we were basically Manohar go yeah like.

E

Is that we have.

C

B

C

Call destination.

B

And in any other way, yeah in this scenario index is the index within the list and way to.

C

Begin I would yeah.

B

We call it list index there, liftoff.

B

And the only time this method will ever grow is if list index is beyond everything, gallon list.

U

F

But I still don't get why it should be done anything they dislike, it's not worth it it. It makes it easier. I guess.

B

I understand that.

F

If we didn't know what the number.

B

Of elements were in the list only do we do it, it's just a convenience or, if you're doing this little move, you can just add it to this like happen to you is.

C

Already gonna have completed.

B

Them in, if you want to know how many got.

C

Copied why are you repeating over like return.

B

The answer it's already sitting in register.

C

We're already learning to a register because we call the function.

B

So Eric does that actually solve the scenario that you had I know that I know that you might have to do to have some consideration for the coffee, too partial didn't actually complete fully. Is that acceptable.

B

If you might be on mute, we can't hear you if you're talking, no.

R

I'm I'm still thinking thinking.

H

R

I mean at least in ml dotnet in the way what we have is is a list source, ASP and destination and then an optional count.

R

And so the okay, and then we always say the destination has to be smaller or the destination has to be big enough to fill the whole thing right right.

C

But then you would just call this one with course index comma destination, that slice, 0, comma count.

B

Proven that will throw an exception.

C

Destinations, yeah.

O

Now you validated.

C

Your counter-argument, are you validated your destination against again the exception types and may need to repeat the work too tight.

R

In the case, where you're trying to do actual buffering.

F

As I think it's whatever.

R

It's either the whole. This is the whole list.

C

If you want the whole list, you call the one that just takes a spam. The only time you'd really want to the one with an offset is when you, when you have a partial phone, which is why the one that thinks offset from Reyes also takes account in the other ones, so that you can accept a smaller array. This is the same thing: we're just we're changing the method name, to indicate why the count the count is spammed ugly, yeah.

B

Because he, the conversation earlier, was like doing it, and can we use the span length to determine how many elements we should coffee and I think the hold up on that was we shouldn't give it the same name, because people would expect the same semantics. So if we change the name, that's.

I

Would we expose it as poppy to which, which is fan destination comma in source index, equals zero and then Compu partial? With the same ordering.

B

So death overload that you just mentioned flicks with tra it all right which has different meaning for the.

B

If stat of T and tra weren't convertible with each other, it wouldn't really be a huge.

C

Like I think that this is the best.

B

I mean to be perfectly fair, like this does have some smell to it, but it seems like.

A

The least thing there that functions that will take me a while to figure out what code am I supposed to write in here. That's like it's not clear to me that if the span that I pass in is large enough that it's a one-time shot, we can also add.

C

Happy to into source index inspectors, we just dump that one doesn't get that how? How far do you want to go? Because that's where the ambiguity.

A

Yeah, which no means how do you ever get to something to this.

A

Yeah yeah, but then you need two methods, in which case you might as well just I.

C

Think the answer is: if you're working with fans you starting to figure it out.

B

Yeah, but didn't in general, though we want to. We still don't want to force people to do to write that math themselves, because you know the more lines of code you write around this, the more likely you are to make this plus for the bit of and so on, yeah.

A

I mean I mean I'm generally a fan of saying you have to pass in the count right because that's a general pattern. We happen to be CL, it's just normally. It's just an overload that accepts the count. It's not like. Oh now, I have to write some merry-go-round.

B

Out in account here is holy fun.

R

Have the count added and this returnable.

B

For some ability, the only reason I proposed this is because it says we don't have to have the same semantics as coffee like. If we want to add a counselor I, don't think. Even if we have.

I

It take account we can have it return void unless say, throw if there's not at least count elements to copy I. Think.

B

If there, that would be the proposal, though right yeah, yes,.

B

You so you tellin, you must coffee exactly this many elements, starting from this index the destination, and if we can't do that, for whatever reason we just fail, you do that.

C

R

You told us how many things.

C

Default into negative one, which we use, the family I'm.

B

I, wouldn't um they don't select, why would I ever use it destination? I just always heard.

O

Something API of a destination comma destination that wait so.

B

The really destination the.

M

Reason, I wouldn't use.

B

A magic value for count here is, if I'm performing arithmetic in order to pollinate the count property I could end up producing negative one and at which point I would actually want.

F

Like why just not just those destination that slice whatever you want well.

A

It's always about the the. What is your intent right if you call copy and then you pass in an end and you're passing the spam, did you ever accept? Did you ever expect that that span might be smaller than what the list is? Why do you do not everyone like? Oh, we don't have.

N

A single a behind.

A

The framework where, if you pass and you just.

F

Put in a radiant each.

B

C

Is very special.

B

It's a filly, yeah yeah uses this.

C

Family billet, this is, we have an existing API. You semantic like.

N

C

Cat bites, for instance, populate.

A

Everything I mean any API that we have the base is to supposed to be one. You know a short API, you know, unless you give us the bounds for both source and destination, we assume you always wanna copy everything. Now we don't have things like all of the the destination is smaller. You only give you a partial data like we don't do that anywhere else or.

B

At the count, parameter and.

A

That but then partially doesn't make much sense right. Does it.

R

But it solves the ambiguity with the overload above it. You know.

A

I mean like I, think at that point, I think we are very used to be my very dissidents.

F

If I pass this panel for vice, oh, the expert provides to be complete unless there's.

B

No expert thank.

F

B

Think most people have an expectation and coffee methods work similar to span of tea or coffee to solving.

F

B

Or nothing well, thank you ordered this destination, spawn kind of implies what the count is going.

F

R

The source the source applies, what the count is gonna, be oh great, I'm. Taking this thing and I'm saying copying it to this other place. Yeah right pick the source.

F

R

And that's an exception. That's it's not valid to do that.

B

It's pretty it's pretty common during buffer processing to have a destination that you keep slicing from the left over and over and over Emily um often you'll pass oversized buffers to api's like expand on coffee to us. Oversize destiny. Indonesia slice are based on how much data you know I mean.

F

I'm fine, with the current shape, but uh I I personally find them. There was a just boy understand, Oh,.

C

Like any boy, if the count is just confusing boy and destination with no index it's already, yes, you say: oh I mean we need to rename it the destiny. That was that one solid, it's the one that takes a source offset.

R

I think something.

I

Put it be that in.

R

The world, if would it be the end of the world if we change this, this span one to have int ban and then to intz. Like yes, I know everywhere else we don't take up a span index when we take a span overload, but at least then it matches hey one. That's.

A

What I suggested as well yeah I, think that's yeah.

R

A

Makes I think it simplifies to design greatly, because the only odd thing is now oh yeah I could have sliced it in past in zero short, but at least it's self consistent. You have no ambiguities. Yeah.

I

They don't like having family not count for anything. I would rather us have the he in span inked or the into the count, and then users who want to be partial copies have to pass in that one extra parameter. We.

B

Only tend to use up full members, though yes, like I, mean I, mean I'm.

C

With Tanner as I and Christopher, I would really like. I want coffee partially, but it can't be copied a partial semantics while called copy, and if people are upset with making a new method group, then in spann intend to the best where they get, because when you at least this nothing's ambiguous and you just keep passing the zero and it's.

I

It might also be more readable as copy to in source index int source count and then the span destination.

C

Yeah, but now like what we're doing is we're overloading the course method. So all we should do is change the bracket bracket. This fan of T think about that is that it's going to be.

B

The middle east.

I

Yeah yeah, but then that's just going to be confusing the users where the span length means nothing mule. Oh, it means something just like a railing. It.

R

Means the exact same thing is the array lay yeah.

A

I

I

Arrays, you can't slice and the array index. Is there that way? Yeah specifically because you can't slice arrays, but you can slice pans, so it doesn't make sense this in context. We can do both.

O

Hey it wouldn't of.

T

Course what their counterparts expanded.

C

And dance papi Department, but.

R

Yeah I guess I would.

B

Right now, what I'm afraid of is that someone's going like, while Wallace changes being made someone's going to write the array-based one in terms of his fan base? One and.

C

B

Should be named, yeah.

A

I wasn't what to say if it gets this second one is you.

B

Guys other than one name to write is that's actually legal.

H

This one is this:.

I

Is honestly, my least favorite.

I

Preferred their like original proposal.

I

A

Prefer the original proposal a point.

C

But I buy the error. The trouble is the two ends that are together means like we have overall guidance, which is don't let the same type in the same position have a difference in it, especially.

B

Because I wasn't a realist, is such a commonly used type like I? Don't I'm I would be definitely afraid of building anything to list that could cause people accidentally and.

I

All the cop we're just gonna call.

B

The simple copy, too,.

I

But then the copy to partial avoids us from needing to pass around too many arguments, especially for the span case, which is meant to be generally more efficient. It avoids the ambiguity and people who want to do a copy starting from some index and doing the entire length can either write their own simple, helper Klaas. The does the additional check- or you know, add their own assertion- object assertion that Spain's at least the length of Rey.

C

I, like reducing the two arguments in England and bill semantics, but it's not like we're blowing past anybody's all convention.

Q

G

I

It's bad as to argument right and on Linux fan gets passed into registers, not one and on Windows, it's passed by reference on the stack. That's implementation. Five is still before the no I think it's ORS maxvill.

I

O

I think we should all think I'm attacking the various we shut up. This is fan.

C

And means caring about things like the number of registers that are going into a bend is actually important. Oh.

I

C

Don't forget this.

I

Is five because then this parameter is passed as well.

B

Implementation details, I I, was actually arguing.

C

6 is the Linux thing.

I

Which means this is not sector yes from Windows at will, because Linux is four and then this parameter is also Linux. Is it no Windows? Is you said you said, ok well, because Linux is 6. Ok, well, wind windows.

C

Is a windows is.

M

I

So you've got this parameter, plus the four others, so you boarded up. Stocks fell in one of them and then on Linux. If it's six you'll end up exactly meeting to count.

B

No copy of articles exactly not that we're also assuming that stands.

I

While in expand won't be because it's passed into register on Windows, it will be, but windows only has four registers for passing and you've got this parameter. So you've got five parameters total, in which case well, because this.

C

Is really about the? Can we avoid people calling the collection marshal to do that? They want.

I

And I think copy too partial and copy tomb. Just the span is probably simplest weight and to provide all the science me. Okay,.

A

Let's step back from on, let's say we go over this proposal. I'll be caught. Ivo came of making this general pattern for any collection, tired that one Spencer no.

B

Did that that I got a copy to method that we just added like if list of T did not already have that first coffee to method like we would never had enough, and you only read literally the only reason we're doing it is because there's precedent. No, we were.

A

Asking I'm asking for one.

H

Thing about coffee partial! So if we don't do those, if you only do this, are you okay with making that the general pattern you don't.

R

Want I would still we still need a copy to me. It's.

C

Only some things are taken. Take me under great. There are only two copy tubes that start with a source index and all of the framework do.

A

I get that drink. No. What I'm saying is that what I don't want to do is I, don't want to say. Oh and this stuff's tvf copy positive of a weird signature everywhere else you just have copied to that, takes an end index to span destination and an income. That will be very weird, in my opinion, like we've.

C

I mean I: guess we have the opportunity to fix it here and what's ready for that, I think that it should be destination.

H

C

Then it's a proper overloaded.

A

You can I can't hold on okay. Let me fix this first leg. Okay, let's first agree on what we want is overlooked. Look like you would make it up. Look like how you would say: I would just say, destination.

N

For a fee, is it all source related arguments or all dust, yeah, wonderful to say, I thought well scams become supporters to nation I mean you wrote the guidance you should know in.

C

Case we use a source, it comes before destination, but then it's kind of hand. Wavy on is the but I did have an example specifically where, when you have the source, the problem is the sources of this, and that's messing everything up. Every everything that we reviewed when we did the buffer stuff, I'm talking about source and destination or static method, a source was always an argument and source index. That means the source which.

B

Would still count it's.

C

B

C

Don't see this well.

B

I mean it's: it's.

C

Before the dot in copy work, but if we were to write this like, we have copy right and it fills the span and copy with a slightly different semantics of coding. At this point really well.

U

But you bring up an interesting it that alternative right, we don't have to add them. It's this. This method, we're gonna, make it a static map.

A

You couldn't make the resetting Matic anywhere else and then we can order the argument. Probably all you want I, don't.

C

Know but it will put in by like you, have static list of teeth so just below the list. We have a list class, but we couldn't.

J

Like everything buffer.

K

J

K

J

Could have pinup, but it could.

B

What would it look like, or maybe there should be taken off well.

A

I depends on where you put it. I would pick you same parliament's, yet I would make it sound overload on whatever else this type already has. It would be copy list I'm their.

C

List source I'm a horse index.

C

A

As she's gonna, calm.

C

Account or we would say destination that.

A

I think ounces of non-course, Bubba June, but I would honestly say the countable would be source count and then you just have to span at the end, because I guess I said because.

C

If you were doing like I want to I want to digest part of the range yes, this part goes for this part yeah, but maybe the weird part is copy. Who should have been in Dexcom account come out already come on, it was copy, it was varying at the distant interesting.

O

Sources, sources.

C

Sources, the destination.

I

You were saying: Aaron I think it tried to come in I, don't think we only want copy partial I thought we wanted also the copy to span array, no other. What's your destination, yeah I.

B

I

Think we're set on.

B

Doing that regardless I, don't I, don't think that's ever been.

B

So well, the second one man, the woman just took his family yeah, like I I, haven't heard any pushback against that at all. That's.

A

The one if we were dead yeah, this one seems fine I, mean know what it's a good name, so yeah I think. But what about this shape? This? Is it the anyway as an opinion? To that shame, because I mean that's what we just said: I.

I

Had suggested that earlier and I thought someone said no no.

B

Doesn't that that does not conflate.

R

About this, one is if we did have a slice.

I

R

The way you would have written it right, you would have said list slice index count and then copy to this destination right. It's in that order. That's the thing I like about it.

C

I would still put the index as a bear where you see yes,.

N

C

Q

As a bank account.

C

I'm just from both, so it's the amount of work you're.

A

Doing yeah, but this cone here has to be that length right so yeah! Well, that's just about it! Oh.

B

It tells you the lesson: yeah sorry.

C

This man could be too big short.

B

A

I

A

Guess what I'm saying is that to me logically, this whole thing, this plus source index plus count is the source yeah. It's why I would pull the coded source code.

H

I

I don't know, and so if we had those first few I, don't think we need papi partial. It's just a nice to have yeah.

H

I mean I would I would suggest this I would deal with the deed this year.

R

Okay, for me, too, I still kind of randomly question source count versus count, but that's in.

C

General I think we have a counter that's applying to both things. We call it council, okay, yeah.

A

I, don't have strong opinions if people prefer a counter controlled help.

C

Feels weird to me, but I can't explain why and I won't complain.

A

It's right, it's shorter.

P

And I: don't since.

A

Everybody seems to think onus, better I go.

P

Of count all right so they're not.

H

A

I

You Eric I like source code higher, but I don't want to fight about it. So well.

T

R

Source count better yeah.

I

I like source, not better, but I, don't want to fight about it. I.

R

Didn't really want to fight about it either.

B

Next week, actually just spring-like number one.

H

Perfect but it is it a cupcake or is it okay.

B

Something else no like that spray get regular cupcakes, then we'll all you know go into a food. No one will be able to give whether the array of cupcakes of the disband last I tried cupcakes.

U

So I wrote this I'll be okay with that.

T

A

L

A

For me, good all right, then Wow 30 minutes, that's a new low for an API review, but really and under the isotherm.

I

Api proposal to add similar overloads to the immutable think and ego key had been frozen. It only reason I ask is because immutable list of T has the same copy to overload as list of T, probably I.

A

Don't think anybody cares about immutable this, but immutable array is certainly one thing that would care about high-performance immutable. This is horrible. It is basically.

G

A

That one is like, if you care about perfect data density, that's not the five I'm, still somewhat sad that it did not occur to me five years ago or whenever we did it immutable.

R

Range, the exact same signature.

A

That's why I was kind of keen on establishing a pattern that we like, because I don't want to have to discussion every single time as you.

A

Alright, then, let's close this one here.

B

This one can no longer be done. We've lost our window for doing this world I've.

U

Been breaking change things yeah.

B

They begin to move in 500, go by bookkeeping system text, Kayson and so many other components now have a dependency on this type. Like I. Think we've literally lost momentum.

A

That are close, it and say I mean not knowing I.

B

Mean if like if we're willing to say we would consider a breaking change to this API surface, even though so many additional projects are now taking dependencies on it, then we could literally, but until that point like anything.

E

We haven't hit for you yeah.

U

What's the breaking change policy, you saw the VCR for informs right, so I would say it seems fine. Just.

B

Let me what's purpose yeah, let me read you this proposal. Okay, I mean.

A

There's it totally.

B

A

Time, I, suppose and but I mean I, think we convinced ourselves only that the impact seems relatively lower, because the only parties that would have to change code, if you wouldn't extend that which is doesn't basically, is only first party right.

P

And the only shipping.

A

In the whole and SP- and this is to get the tool, but they also ship- let.

C

P

C

Pointer base, one.

B

No longer add and order, or not, basically deleting all of the existing out so yeah. The fact that the current abstract is takes a pointer is.

C

Known great, essentially.

I

Well, it's it's not just that.

B

It takes a pointer, it's that the method, so the methods abstract, makes a pointer and marked out of a browsable number, which means that if you were to try some.

M

A

I think only you get is cool and wasn't me just control dot and it writes the method for you.

B

And yes and then you're gonna leave and then you get a compiler error because it's like you got allowance a point.

C

Override and it's like now, there's nothing you can over. Oh, you want to write, equals yeah, I'm.

U

Not sure override completion is a little.

A

But it's actually like: let's give this one here, is it assigned to you, it is assigned to you good.

H

A

Right expose match length and using string index. Yes,.

B

Please please do it.

A

I like this is a totally German. I can see that from here, yeah lots of streets going on.

B

So Mostafa, so you know I didn't know. Shelly I can actually speak to this. In a nutshell, um when you're doing string comparison.

R

Guys got muted by the way Oh what.

B

J

A

Oh, what happened.

O

I don't know, but.

B

Really go marry now. Thank you in a nutshell, we.

U

Were a life though, like we were not muted on YouTube, you.

B

Should hop, oh just on YouTube only on when, when we're doing cultural, where string comparisons, including the invariant culture, because the invariant culture is still a culture and what I say culture aware I mean as opposed to Oracle, it is possible or matches to be of different links than the things that they're matching. For example, the German asset is a single car.

B

However, it also matches the two R's lowercase s lowercase s. That means that if you were to, for instance, error a string which just contain the SS, its length is 1 and you were to compare a string which is to lowercase s's like to it would compare as equal under German current culture. Let me say this is a member of the German.

N

Culture, yes, some people apparently do not want software written for them, because I was anybody supposed to eat this right so by.

C

The way you see if a string is frantically dsm-iv.

U

Ok with that, because tell me like how do you reasonable.

B

Rules like that, like that's impossible, so so let me, let me also say, like the proposal API surface I have for utf-8 screen. Does not have this, isn't it for various reasons: I, don't knock it into alright, but.

U

You just don't have any.

B

U

Off or like what I don't have index on, oh good yeah, that's fine! If there's one way to solve the problem, looks like ya, be done. You don't antic like so what I would actually do.

B

And is the proposal that I have for utf-8 Stern I would actually bring that forward to string um what that is is instead of having index of it has a method which is called try, fine and that method returns a bool and out a range which is this is where I found the match. If you want everything before you found them out, substring ranged off again. If you want everything after the match, substring range and stuff like that, yeah I.

N

Would agree with that if the index also isn't the most convenient to call, then you contrast in comparison and then also dealing with the match.

B

Name feels like because it's not just here, you also need to make changes to things like the the culture of text in system globalization. Whatever you have try finding try find life, yes, yeah I'm cool with traffic. Instead.

C

Of doing this,.

H

It's I think that makes more sense. Yeah I.

B

Mean the downside to that is like we're now exposing range as a first class concept through these API, so as a type rather than just English semantics um yeah. Maybe index of is easier for people to reason about, because it's just a method.

H

So how about this that I assigned this to you? Okay and then you can post what you want to do there, because it seems that we get to differently a proposal from what we currently have here.

B

I mean all putting range is probably fine yeah this this API proposal does work that I have been experimenting with other things and that I think would would be more suited for sometimes like really.

C

It's the continent, if you were already solving this problem in a different way in utf-8 string. You know, as we want the two of them to agree before we do yeah.

G

Didn't mean into this here also because this.

B

Actually has impact beyond this trend and beyond just memories: it's like it touches system, globalization. It.

G

You and let me do API needs work.

G

B

Luigi hates me because of different.

A

All right, so then we have that one. Let me even a few minutes left I I.

R

Hate to bring back up the copy to think, but after we after we moved away, I found out that stringbuilder as the copy to within span int, along with copy to int array, intent.

R

So, basically, the room why we switched it in surround a stringbuilder already has that pattern is.

B

Stringbuilder coffee, too, called anywhere near as commonly as like list of tea.

P

G

A

R

That ones we feel in the next one right we were talking about setting patterns x1. We could look at stringbuilder and be like well. They use this pattern and then nice to Polly's I think.

C

The answer is, were content to say: stringbuilder got it well now.

B

That would be by myself as well here.

A

C

The framework guidelines are written, for we did this and we didn't like it so now it says: do not I need.

B

That kind of makes sense, because we don't realize that we need a guide line until we regret our decisions right.

A

I mean funny enough: the copy to that takes in array array index and count on list of T is used less than the corresponding stream delivered. That's what Trimble is 7.7% and the other one is 5.8.

C

Yeah, but this you already had good alternatives for how you can use the data out.

C

Better, the stringbuilder one is here because it's a very specific API on that you also yeah I mean.

A

You could always numerate this jungle, this well hey. We only recently added the chunk based enumeration, but yeah I mean my key levels at this point. I think low. That gives me feel like we talked it to death. I can go either way. At this point I I think I buy the argument and I think it's unfortunate that we made that mistake on string builder, but stream bulla is also not the type that all up is. Is thankful.

C

For me, it's possible I.

C

Think I think we did a lot better on less than me, I mean.

A

This table is used more, but yeah I mean to be clear when I say more list of T is like 90% and stringbuilder is like 73% so like they are both in the yeah it sucks. If you get, was wrong territory.

C

But I feel okay, another his most common thing. Is you call it a mission accomplishment that.

O

Is also true, it's.

C

A very different okay.

A

I mean I think the question earlier still stands right like if we were to make this a general pattern. What would you rather do and I think the shape be currently have you feel more warm and fuzzy about than the other one, because.

C

If you look at it from the pure fan perspective, it would be effectively this not slice, offset count, close paren dot copy to destination, and that's the signature that right up there, even contra came up with. We just remove the close parentheses in about.

R

We can pull her fruit for real close it. No I just wanted to make sure everybody was okay with ya.

A

Know it's a good point. Priority is important.

A

All right, so almost a few minutes left I'm funny enough. I was expecting.

H

How about this one here.

E

E

N

Because we should be using volatile Dalton study, yeah be okay with that I would maybe even go a step further and say. Maybe she helps me them so.

B

Remember how we were talking a few days ago about like various levels of off solution. One beam like this would work. It's just not the way to right, lock, maybe.

A

U

Like that, the.

A

Point honestly.

U

This is also like something that would be hilariously easy to write a fixer for alright, it's, like you literally call the other API well.

C

Yeah, you I, guess you removed the thing left of the dot you added off in the middle, yes, I mean there's literally a string replace fixer.

A

Here would make sense to Joe.

C

M

F

M

F

We're updating it.

C

We should have interbred whenever, if you don't see it again in like.

B

The the method works and will continue to work. It's not. It's only obsoleted, because there's a better replacement, yeah.

A

I mean, let me let me ask a different matter: I mean I, think the generalist think calculation is. Our solution is meant to get the developer a heads up, right and I think if it's just not idiomatic code, I think just fixing just shipping a fixer that was turn the you know. The one code side to the other code side seems fine, because you basically the squiggle as soon as you edited when you open it, you get a squiggle, that's fine!

A

Now the question is what we actually want to give people a build warning for that and I think the answer for that would be. It was literally like losing API calls of the API there's nothing wrong with the API I think it's probably not worth it. The only case where I could see this being still usefulness. If we actually consider yanking out the API will clean up, but then again like these api's are probably used a fuckton, so I'm not sure we will convince ourselves that making an API braking change forward.

N

Amongst to be an aesthetic issue, it will be worth I think there was a desire to get people off of the thread class.

B

More generally, because in it's not always possible, especially when we're talking about new execution environments to talk about friends, either static, these aren't these are statics. But if, if we're thinking of removing- and if we're thinking that there might be environments where we just don't have the concept of a thread at all, then maybe the entire type goes away. I, don't ever.

I

Say that because the language or the the runtime spec already says that system threading thread is not an OS thread, it's a managed thread which could be a fiber. It could be anything so there's no environment where you can't not have other, but if it's something meaningless.

B

I

An environment well.

B

I

B

Have things like thread local.

I

You still have volatile, reads and writes in in those environments. If we were to specialize for them, then we.

M

These don't actually dissolve how reason right a.

I

Normal reason, right after that's functionally safe from a single in an environment where there is only right.

U

I mean looking at you said you should seems to be relatively low, for those API is like less.

A

Than one percent so I could see us actually yanking them out.

B

A

B

I know that we were going to talk at some point about, like which API is. We should consider having analyzers for orienting like we should put that on the file now.

A

How do we do that, like we have it live.

H

In labor for this bull in our way.

B

I might actually leave it ready for review and just back to it later.

A

Let me actually let me actually, though, let me actually know what you know. Let me just mention the term analyzer and then I have an easier time finding it later.

A

Or taking the hotel.

B

We could add labels, you know one will yell at us. The.

I

Reapers are about to merge anyways, it doesn't matter yeah.

A

Nobody kind of eliminated. This seems like.

C

I

We don't lose the ability to edit code, we can still do. We still have access to the repo internally new, but it.

C

Won't necessarily be building, he shouldn't really be a.

A

So Lulu bean. So let me ask you a different question: would we ever consider just putting editor browsable on it because it seems reasonable to just hide the api's for now at least I think.

C

That, even if we were you, know obsolete it, we would edit around will never do. That now is yeah.

A

I would I mean thing, I think we can't improve. This is proposed and it's like.

B

Without having something like an analyzer or just telling people, this is why we put under the browsable number on every might just the 1%.

I

Of AP of people.

A

That call this API and they probably know what they're doing.

I

It because it well nothing for people already using it. It will cause a problem.

B

It it won't cause a problem for existing code and as far as compilation goes, but if you have to maintain existing code, it might cause problems. Yes,.

I

But we have an entire year until these API, so we have an entire year in which to get an analyzer or update the documentation or anything, and as long as we're tracking that to get done, we're saying dogs need to be updated. As part of you know, this well.

K

I'm kind of like in not in favor of the browser, but it's like.

F

Let's say I like this API and I, don't know that you wanna exist now. I want to find this API and I. Don't know what the new honest I, don't even like I'm.

K

Typing, nothing doesn't show up at all.

F

But but a picture then.

K

You go look at the docs in the docs and what I cannot find the API used to go to a browser, and you.

C

Call it about read instead.

B

But you don't know what got deleted because you type thread dog is like I. Don't remember what this used to be called because show off.

K

And then I should see.

B

The error? Okay, whatever this.

K

B

An opportunity or like intellisense, more generally, to help with these sort of it.

I

Maybe an editor browsable never had a call this API instead and then it completes to that instead, like it shows it, but when you tab it, it yeah.

B

There, because one of the one of the things that I have wanted is like, it doesn't show up in intellisense unless you call the API somewhere else in your code and then at that point, I mean they really. What we wanted in that case is a special one. I talked about parameter.

I

Aliases, thus that way we can go and fix all is defined by National guru them.

C

mmm A sink region called pulse, half score the same as the same one that works for me. It's not virtual leader, okay, oh.

B

Okay, the other thing they did not virtual yeah, but.

A

They may just copy and paste the whole thing room and yes,.

C

A

C

That method in December.

G

B

Yeah I don't know why he wrote first, one.

O

B

A good thing one might be business.

C

B

Existing ones on virtual, they don't know why they make that virtual.

F

Okay, the new person.

B

There is a very long female thread about that. They, the in theory you're supposed to be able to, but apparently whenever.

A

You get destroyed, we would say it's non-virtual tasks apply to compute hash, a saying it still takes a notional conservation token and the second one. What three.

B

C

B

Think, yes, current cancellation token altered at the end.

B

I think the reason I had proposed based.

B

You could all it's passing in multiple different frames like if you had like some type of rear line, not be a valid.

I

I think the reason you made a virtual is a Schwartz touch quarters personal yeah, so you probably it has to be yeah. She probably made both virtual due to that.

B

Yeah, because house core is the actual workhorse and definitely we could always out hash for a second No. Okay, that's fine I, like.

G

C

All right, so the only thing you need to do async is read through there's: nothing like every every implementation we ever have to this POS.

C

B

I guess the the scenario was like again: if I'm, just if I'm pulling data from multiple sources and making it house grates.

L

On all the issues.

B

Like this computer.

L

Service, you can't.

B

L

One- and it is.

I

Now unblocked, do we everything scheduled for next week or can I invite the compiler folks that want this yeah.

A

I wanted to ask you like where we are with.

I

Them because, like we made.

Q

U

Q

I

Should be completely, and we just so.

A

What happened to the half? One I thought the half one was shut down, no.

I

Proposals yeah, but we've not decided what basically that's still pending. What do we need to do to expose it if the language is court operators in okay, but it sounds like they're not doing language support or the type or.

F

That's good right.

Q

Poisonous today well, it says that today, there's a like did.

J

You not gently H mark.

Q

No I did I just think refresh.

M

L

I

They were done last week, they're duplicates of other issues. We review. That's.

A

Good, all right, so do you guys online.