.Net Foundation Design Reviews, 20 Aug 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

00:00:00 - Approved: Add a Socket.Listen() overload https://github.com/dotnet/corefx/issues/9680#issuecomment-523130997
00:05:45 - Approved: Add ImmutableInterlocked.Update for ImmutableArray https://github.com/dotnet/corefx/issues/19408#issuecomment-523140923
00:30:11 - Approved: Add support for getting current stack into Exception https://github.com/dotnet/corefx/issues/19416#issuecomment-523144443
00:39:55 - Approved: Add 'split' support for ReadOnlySpan(char) similar to string https://github.com/dotnet/corefx/issues/26528#issuecomment-523152939

A

All righty, so does anybody have any preferences on order like there right now, I believe ordered by oldest to newest? Well, we just go down that one.

B

These are all post three all right, there's nothing for three up correct.

A

The three-year-old thing is I believe let me check button, but yesterday, when I looked into it, it was empty. Should.

B

Be empty and if it's not well, we'll make it empty.

C

Yeah, where would you be able to speak to the song one or would you prefer for other people I'm not working to be here I.

D

Might be able to speak to it, I haven't seen it.

C

But maybe I began to interpret it. This.

E

D

The quick review opening, ideally it's either we can say guess, or we can say no, where we punted to a later meeting okay. If it really needs discussion, we should be better about our process.

F

The ones that have one years opened is waiting for the dotnet yeah criminal types, review, they'll.

A

Just open a bunch of chats when we can just pick it look them. Okay, alright! So I guess this! Is it it's just one that doesn't take it back? Look.

B

I think it's perfectly reasonable dad he.

G

He's right that.

B

No one knows what to pass, and so you know a lot of coaches passes in Max value or passes one if it doesn't care.

H

Parameter yes,.

A

There's some of those things where maybe the comparator would be set, because we may actually change that we're.

B

Gonna change it. What we might want to do it based on the underlying OS yeah.

D

Yeah it's currently throwing for negative one. It could make negative one via whatever you think the way the answer is, but probably just a little bit yeah.

A

Shoes event, otherwise that's your decision and.

C

Then we'll just like someone, smart figure out the default value, yeah.

C

We should care about behavior.

B

I don't know, based on emails, comments about his swearing earlier. It seems like we don't.

A

You really have to restrain myself let the sprayer just.

I

Yeah all right.

I

C

You we don't have hexadecimal, but we have bytes to have strings, but it is a hyphen between every single pair yeah yeah. Oh that's the one. Yes.

D

I have a helper method for this in every test of similar than I've ever.

C

Bring everyone has a helper method for this, like I, yeah yeah. This would be souping. This.

B

Though we have I haven't read to see all the various patterns that if this is posing to support, but we have various formats for it. No separators different kinds of separators and square brackets around to each set of bytes no square brackets, you know, is.

C

There an actual concrete proposal in here yeah well, because I know that if I'm remembering the spread correctly, there were lots of different proposals because I think I even contributed one but I, don't know which one is being promoted as the one we should take. Yeah.

B

I agree: we we should add something for this I. Just don't it'd be good to have something concrete to hang on yeah.

A

I think this one, it seems way over like I, think what we should do is even if you have an existing lesson, that is almost what we want, but not quite as soon as no was that an overload it takes either a format string or just an inner alert format, or something like that. What.

F

Would the difference between this API and string dot format, array with separator character, be because we have that strings.

B

On that join one man.

B

Bring straight string join to eat this.

F

One, how would you is that we.

B

Don't have a joint format.

D

Yeah we're doing just blues things together, so you'd have to earth to string each of the individual bytes and then they move them together. This.

F

Potentially performance optimization skipping the get default culture and arse out X to a million times. Well, you guys.

B

It's more than that, like string, dot join just takes an ienumerable of T yeah. So if your T and there's no format applied to each by there's.

G

No string the more Matt an argument to string joints.

B

You'd have to use, link or something and say, string dot. Join. You know, o open, quote close quote: comma bytes dot, select the arrow B dot to string something it's yeah. X2 should.

C

Should we actually be considering there's.

B

Another issue: first for a join format somewhere I mean.

D

Honestly, like that, we're doing hex, we could do all of this, but out intermediate strings. You know if we don't either spring got going. We can't do it without an immediate ring, because that would be the contract to shrink that Troy yeah yeah. So this.

A

What I said honestly to me, it sounds like bit converter to string. This is a reasonable place is just that. The method that we currently have kids to make a hyphenated string well.

D

I mean, and then, but we should you know, is that the discoverable way of getting hexadecimal like it does have a you. Either know it's there or not it. We don't have a way of getting it from hex back into the thing. Presumably, we should span a PHY it as well, using the try right pattern that, like all.

C

The base64 stuffs all really only convert class yeah.

D

If, whatever is at the top here, suppose that seems more complicated than the trippy.

D

I would say, the easiest answer is that it should be next to to base64 from these six people.

D

To have string from country I don't know that we need to care about separators or prefixes or suffixes utilize. Maybe we care about.

B

I mean it depends what our goal is, it might be worth I think it would be worth doing a survey of all the places we do this in the framework itself. Ideally, we shouldn't have to have any custom implementations of this once this method goes in.

D

Can we see what the actual proposal is, how the Levi's special topics this whole game? Well,.

G

Gee I, just called through its who's. All like is lots of discussion, but well.

D

Somebody marked it ready for review. That means they think they know what we're actually doing.

E

We've got several composers.

C

C

Even worked, it is ready for you. Then I probably marked it with my proposal done and then more proposals happens. It's just ok.

D

So this is more send it back of please land on a design and edit a top list.

C

That if I marked it ready for review than the one that I wanted to bring, was the one immediately before, where I'm walking. So it looks like their general history yeah. So.

A

Is it fair that I signed it back to you, and then you figure out what the yeah.

C

That's fine! It's fine.

C

You yeah, though, I agree with what was said earlier, though, like we need to audit how we're using this internally and make sure that, whatever proposal, we have can actually work with all of those different.

J

Also, we are talking about hexadecimal here. mmm Should we consider other.

J

Primary impacts in the very least, make sense.

K

Yeah I understand.

J

K

So how else well emission.

C

Could be quickly now that you've.

F

Got C sharp, binary literal expressions so.

C

Hex and base64 kind of go together because they're both ways of representing, like them, binary byte data in a string as a container like you, don't really see, octal used for that octal generally just uses a numeric representation like binary ones and zeroes I, never heard of any request for that for data that came from a binary. Okay,.

B

I'm, just thinking of the framework itself like there are probably I would I would estimate 10 to 20 different places that we did was conversion with hex today and I can't think of a single place that we do it with binary or octal other than the one method on convert that takes a base. Yeah.

C

Binary would be good if you're, if you actually have numbers instead of byte arrays as your data source, because maybe you want to format it for inclusion in a c-sharp document. Again we're talking about numbers yeah in.

A

The spirit of what Jeremy or they said, should we just close this tab and then right.

L

Didn't really do this for more.

M

A

It did it don't even ever miss mommy.

A

I'll stuff, it is from 2018 I think this is something they'll be starting, but we didn't finish.

C

Now we can take as much time as one.

A

If you bought 15 minutes, then I understand why so I.

C

Think there was a consideration in here for what happened like if you were trying to move substitute overlapped with itself.

C

What's a nice read.

F

So today it does a remove and add, is how you do it and move is more efficient, especially for large arrays, because you can. You can avoid an intermediate copy, you're trying to write a really.

C

Fast bubble, sort yeah.

F

So so, basically like today, if you were to say you want to remove an element from like five elements. From the end of the array into the middle of the array, you end up having to remove the five copy, all of the other ones down and then insert the missing five into those empty slots. And you can you can optimize that slightly and then varying positions like if it's at the beginning of the rave and it's really bad, because you remove the first five.

F

Then you have to shift everything over then you have to shift everything from the position you're inserting over towards the end and then insert so.

N

This is not just copying one range to another one right: it's inserting right it's moving, but you can do it more efficiently. Okay,.

J

You can you can skip the my original question about like the scenarios that we need. This I mean. Is it like many scenarios that we wouldn't need this or just.

C

One person area would be like a imagine. Imagine you have like a bunch of pages in a OneNote document. For instance, and imagine you are yeah like you, you select this range of them and you just want to move the marker right like it's. Obviously, if you have a UI based application, you would be using. You know, probably a different data type but I'm similar.

E

It seems like it could be so centrally inefficient that I wonder if this would be the correct data structure to use, in that instance, I think.

F

It's probably more of a there, so they're they're asking for on list collection, extensions and array, flexion extensions covering anything implements I list, but it's it's basically one of the this is a fairly common operation for various scenarios and they want it exposed all rightly, and they just covered the three main ones. So.

C

There was there was a request from Christoph way back when, like 2015 asking for someone, anyone to provide scenarios.

A

Yeah, just waiting for the feedback I mean there's one item here: we're not quickly but the scenarios but.

D

D

A

It seems like to be user of list of tea, a more efficient move with our phones with the public API openness which.

K

I mean it's basically just trying to do something that you can't do without access.

F

To the underlying array.

G

Then, of course, the Exorcist by this man is.

A

We honestly for list of Tina ray I kind of buy that because it seems like a very like it is something is kind of said that.

K

You can't actually off to us and C++ exposes a move method asked you several other languages, Java yeah looks like breast. Has woman too.

A

When the collection extensions, one I'm, not sure, I entirely buy it, but I guess it's nice to have one central part in the friendly, where we can do something smart. Like you know, if it's a list of T and this an array, we can do the efficient one. Otherwise we just do a yeah.

A

The manual thing to the interface right, which, because.

F

We couldn't optimize hwachun extensions, it would just be an insert qualified.

A

All right what you wouldn't mind you working with try cows floor, laying and just up to you I, didn't it don't succeed, you're, probably more 31. Why.

O

Would you we I guess it's you so one.

C

Thing potentially that we could do as a more generic operation would be to have list. Allow you give you access to its underlying away as a spam inside of a call back. So, for example, there is a method right now on list which is call for et, which I know we don't like, but just consider for a minute. So if we had list dot with spam or something like that, you passed in the callback within that callback, it gives you back a span of T, which represents here's.

C

The backing array of a list have fun and then, as soon as that method returns like the spans dead, you can't use anymore, and within that method you obviously can't change the size of the list, but otherwise have fun. Multiplying whatever element you want. You would.

F

Have to also assume that the collection is being modified, even if it isn't so that enumeration doesn't break because enumeration currently anytime, you add and cert, remove anything else that it updates the version counter. Yes, yeah.

A

We could we could. We could chop that.

C

A

Continue to different versions that we could have a mutate method, it gives the biggest bad when they're read in an efficient way that gives a becket we'd only spent, in which case you don't get through. So.

F

K

I think there is another.

F

Proposal somewhere in here for expose a way to get a me know, expand from list and.

C

Others, yes, because he the thing that I had just said like, doesn't try to solve this problem in particular, but it would allow you to write your own inefficient in your own efficient implementation on top of it, because it gives you the tools that you need to are yeah I would almost preferred.

E

That Easton and oh, we see a concrete example yeah. What.

N

People are using this for.

F

And then people always have the ability to write an extension method which is get backing span, fix it up and then return.

C

It you could implement Adana I mean we. We would have to be a little careful here, because we would want to make sure that you can't like, like you said, we want to make sure that you can't do anything that would change at least the size of the list. While the closure is active but like I mean we could be back to spam, feel free to change the elements right.

F

And then people who are modifying the size, at least in terms of decreasing the size they can always modify it and then set capacity afterwards, yes to to remove elements and then for adding they set capacity prior and then modify so yeah I. Think I think that.

C

Would yeah I, don't I, don't want to derail with so I'm just wondering if we're trying to solve to narrow the case? No.

F

I think that makes sense.

C

You this is where Steve goes, closures well,.

B

I, silently did that in my head, and we would need to pass a generic tea that allows you to the house and things like from index and to index as part of a struct yeah.

C

I think we have a pretty well-established.

C

Basically, it would be string or creative effectively in terms of what was that delegate. Signature looks like.

B

Probably even use the I didn't use the same delegate, type, a span action or whatever that's called. Oh.

C

B

Because we do it's, what string Korea uses I forget the actual name. It's something.

B

F

F

It was Ben Adams who proposed list of tea as spam,.

C

That would be difficult to do important. Yes,.

F

But I think it's along the same premises. Let me get a read only span in a span from a list.

A

All right, so you.

A

I have commented on this and I pushed it back to API suggestion.

L

L

Seems like there.

C

Clearly, the method should be mutated and.

A

It's Tunisian I'm feeling lucky.

A

There should be the Rena major get name.

A

A

C

Does anyone assign generic chemical so is anyone here, aside from Steve, even familiar with the immutable? Did you walk to class yeah.

M

I had to fix a mug that all got one in hi like three years ago.

C

B

For those that aren't it's basically just a typical interlocked, the typical wild interlocked update loop, where you get the current value, you then trying to knit a lot compare exchange if it's the same you're done. If it's not you retry over the bride, get the immutable collection types because immutable collections are immutable and some a typical pattern with them is, if you're using them to represent some global state.

B

You create a new version of the immutable whatever and then you interlock compare exchange it back in to be the shared process global one, and so these helpers that do that loop. For you.

E

And in it I think it makes sense to have this surprise. We don't support it for a already.

A

Yeah we have a generic update, one that takes the tea and the funk of tea. Tea recommended.

G

A

That one, but of course it's annoying.

A

I mean it seems reasonable to me like I, it's very much in the realm of what the class was about so I think.

E

If we supported this specific types with other boroughs and there's reason to have feature parity with beautiful array,.

P

Mesa's on the mood will interlock that just taking writable away if.

A

That's him to look I'm very exchange, isn't it looked exchange interlocked, initialize, so I think an update one that takes me with the beret seems very visible. It's just that when we wrote the class I think and we're never gonna. Do this one.

B

Question I do have is the first overload update that takes around yeah yeah? How is that different from update that takes a rough T in a funk t2t.

B

Like the existing update method,.

B

Oh I guess that one's constrained to a class, an immutable array is a struct.

G

A

B

Say right because interlocked doesn't work on arbitrary struck, sone, Leone, known primitives and on reference types, and so this would actually be doing the magic to access the underlying reference type behind the immutable array. Struct yeah.

C

Is in the compiler, the latest version of the compiler can out disambiguate overloads based on constraint. Is that correct? No, you can't.

A

Just by constraint, but in this case it yeah well.

F

Where it will kill constraints that can't be applicable, but I, don't think it allows you to over a little based on that look.

C

On it, so the where T is class simply won't be considered for inclusion, because T is a struct. In this case, immutable right, I believe they'll, explain about it. Yeah I just wanted to make sure that by introducing this over a little like we're, not just gonna cause compiler errors everywhere, I mean the interesting thing.

B

Will be, will these things always have a way of introducing ones you weren't expecting? So whoever does this work assuming we approve, it would need to then look at our tests and if there are any test breaks as a result of adding to the ref, we would need to look at that.

A

What ambiguous he's like this actually be complaining? The world.

B

It just passes an arc through to the delegate yeah.

A

It's basically the the middle version of the other good overload right. Take.

C

Care yeah because this could be written today on top of interlock compare exchange. Yes,.

B

Well, except not because you don't have access together lying in array, no.

C

I mean there there is an interlock compare exchange over load that actually works specifically on a mutable right, yeah yeah.

B

Yeah Mike again wanted me to pull your message.

C

Yeah got it so like a naive implementation. This would just be a loop around yeah. It.

A

To be clear, you can implement this today, you don't need the API.

P

Is but if you do this a lot, it's annoying to write and I'm.

B

Missing out where, where do you see that interlock.

C

I'm sorry, mutable, interlocked, dot, interlocked comparison.

B

Oh I see it then: okay.

A

Yeah I mean these are reasonable to me and anybody any concerns with that.

G

G

G

A

A

A

You all three visits again crazy ideas.

B

Yeah, so there are lots of different places where, especially in our async code, we manufacture an exception in some place in order to like map to some native exception error or whatever native error code, and then we store that exception, someplace and then sometime in the future, on a completely different thread on a different stack. We then throw this thing and we lose any knowledge of where this exception came from. Normally, you know where the exception originated because it was thrown, but you in that pattern it wasn't thrown.

G

B

Just allocated and then stored and so having the ability to actually shove into the experiment like a basely environment stacktrace is useful and we currently lack any such thing. You can try and do something like you know, throwing it and then immediately catching it, but best case. It only includes the current stack frame because it only includes like the distance between where it was thrown and caught, and so you don't actually get environment that stack trace. You just get literally the current frame. If anything, yeah.

C

So who's who's supposed to call this method, then, at the point where it's caught at the point where it's really the.

B

Point where it's created so like ago, I can't I can't find some examples, but basically we there are bunch, plays and we'll just take a quick new win, 3-2 exception or new HTTP HP request exception or whatever, and then we store it somewhere, we're not actually throwing it and then later on, it's thrown as part of an intern exception or its propagated, with exception, dispatch info dot, throw or some such thing. But we don't have the original stack trace from where the actual error occurred. Okay,.

C

So at the point where you're creating the EDI, then is it? Is it typical to create an EDI from a new exception without having from the exception at all, it's.

B

Super common, sometimes directly, sometimes indirectly like if you used like so a very common case, would be task: completion source, not try, set exception or set exception right, you're, storing an exception object, so you might manufacture a new exception and you store it into the task. Internally. Task is saying: exception. Dispatch info capture with the provided, exception, object, and then we later await the task. It does exception, dispatch info dot row so anytime, you see code that does set exception new exception, or you know some such thing.

B

It's hitting what this would be meant to address, and so instead of saying set exception new exception, you would say new exception and then whatever the method is a pen stack, trace and then some exception. So.

Q

This be an overload on the existing data. Api I just had a bool or capture saying append constructed it.

B

Could but then it would only be applicable to exception dispatch info itself. There were places as well, where we like there for better or worse. There are some places in I think SSL string, maybe where the exception is just stored into some field and then at some point later whatever's in that field, this store is an inner exception. Actually do this new code today for shipping new, it's for you right there, whatever exception that we get somewhere from an error occurring, like you know, we're an exception.

G

Wasn't thrown in caught rather.

B

We we got an error code from the server, and so we manufacture a protocol exception object for that error code. We published that exception to a field and then at some point later on another thread. A new wrapper exception is allocated and that original exception is stored as the inner and so there's no exception dispatch info in the picture anywhere. It's just an exciting, inter exception that lacks a stack trace.

Q

B

System runtime exception services, so.

Q

You might use for the static route it swing up and you get it or never see, address.

B

Yeah the static is a misnomer that so.

C

Should this potentially be an instance method on exception, rather than a static method under EDI, it.

B

Could be I I didn't know where to put it in fact, I thought I had some alternatives listed here, but I guess not I, it's sort of advanced and so I was trying to bury it.

C

This doesn't seem like an API. We want people to use exception, extensions under systematic selection statistics, I.

B

Mean I sort of put in I put it in EDI because with the idea that it's EDI is already doing this kind of thing like it's already appending.

G

All the old stack.

B

Trace with the new one, and so it kind of seemed to make sense to me to say, let's put the same functionality there, but I don't really care I. Also.

G

A

My other question is whether dependent on said.

G

B

Weak, it could be said if you were dealing with a brand new exception object. The difference between set and append for a brand new exception object also wouldn't be any different, and yet.

G

B

Could be used with an existing exception, although I don't actually have scenarios for that like that would make sense if someone had thrown you an exception, and you want to append the current stack trace. I have never think.

A

That's logically,.

B

That's logically, the functionality that exception dispatch info already has sort of. So that's why I said append, but it could be set again. I, don't know that it really matters.

J

And it'd be like set on to have somebody on telling if you want to bend over just set leader.

A

To me, it seems like, given the scenario you described like said seems to be the more natural mapping. It's fine! If you actually append a.

B

Nice added into Tareq's idea: we haven't really needed the append in the future. We get out of going.

Q

Back to dr. dr. Zhang, so static method on PDI, why can't we add this as a pool to capture rather and you may die, because.

B

Edi might not be involved well, I, don't.

A

Want to create an exception, this magenta.

C

Where you may just want to deal with an exception, object and EDI as a clutch just happens to be a convenient navigation.

A

This is an aesthetic method, well, you're, not actually creating a media and.

B

I messed up with the static on the class, but the static on the method is crude. Must.

Q

We do this or because this thing this mess is returning void. So you don't get the yeah no.

A

Literally, just replace the way, I said the thickness of material. It.

C

It it has nothing to do with video.

A

L

C

Scenarios that Steve gave- and it's often used in conjunction you.

Q

Need to modify just the exceptions for yourself.

B

Basically, it's just exposed it's it's allowing you to basically take a spring and shove it into exception. Dot, stacktrace.

B

With potentially the ability to be a little smarter about I mean it's possible that the runtime could do things like actually capture the relevant Watson bucket and store that as well.

A

So these sectors that you would append would include whoever calls append currents neckties, but it wouldn't include apparent sections itself right.

B

You mean what would a pen current stacktrace show up at the top of the stack yeah I, don't know I kind of don't care, although maybe I'm being too flippant about it. I mean the reason this is relevant is simply because there's information that we're currently losing that helps to track down where things came from. If.

P

B

Is additional information there like an extra stack frame and there I'm not sure it really matters, but there might be some people who care so we could think about it. Whatever.

C

Makes sense to have this as a static method on stacks race instead of media I'd.

B

Be fine with that too I, don't.

C

Really know how that would work done, I.

B

Mean it's a static method. They both live in the same core lived assembly like yeah. Doesn't any static method anywhere could do the trick yeah.

C

We're probably just funny I.

A

Wanna see my key idea to put it on the eye because basically you're putting it in a square that says years advanced stuff, you could do with exceptions, and that seems more in line. Yeah.

J

All right so then, just for the behavior only have been dolls ie. Your recordings that we used to have like, like dashed line between the two abandon Thanks.

B

Yeah- and we still do the exception- discussion for ads I mean it might be. We could say that set the the scenario will be limited to could be four brand new exceptions, and so it could even fail if the exception already contained a stack trace. If we wanted it to and then it would only be for the brand new case, and we could deal with things like what's the separator between the Pens right in the future, if we ever needed the append functionality well,.

Q

We want to bend only because extension court have as long as that. This that's.

B

What I was saying is.

R

A

R

A

So I added a comment just kept it over to say and.

R

Then I think we are there, but we didn't set.

A

That was once addressing and if you ever need an overload that takes, they me if I need a candy, would add novel of the tears a pen, yeah so second override.

Q

The current status for the healthy attention or.

B

Elite fan, let that was yeah.

Q

Over to assailant, the exception had already been through.

A

Which is what I like said over in it, because if you ever change your mind and makes it more sense to just remove the error check and add an overload right.

A

All right so then,.

R

Oh, these ready roars, I think.

F

We said for this, and possibly the next.

K

To that they're waiting for the.net 5, primitive types, yeah, that's nothing! Well, then Sigma this going here.

F

And that one is one we said: we need to dedicate a new session because.

O

O

L

We are sorry when that comes which one the enter pointer, stop.

F

You want me to forward the meeting invite to you.

A

And you say things no way: yeah wait until.

M

You get like 12 meter, all right, the runtime, the framework and the language team, all I.

C

Want to ask is: will I be able to write this code, and just have people tell me yes and then you know, waves of happiness will watch over there. When did this ever happen to waves of happiness yeah not since I started doing security, a good.

A

Job seems to be not best set up for the peanuts, an area, but there's a reason why Barry's grumpy all the time right.

A

All right, you proposed to us yeah.

Q

I went to a lot of changes, Kathy I'm late. It's me I eat more than yours, because, right here, yes, it's a useful API to have. Basically the change was more round three one. How do you want the enumerator to look like? Are we new and expands out on a bicycle spans?

Q

Previously, it was more like returning a pistol memory of here and giving that. But that's this is ready. Daniel on I mean the objective is to have significant ideas, strained or split, but for Spanish art since innerspring allocates. This is giving back in the numerator that he did for each over the slices that you are sitting and where would you put this.

R

B

Feel, like we talked about this previously, but strings split options being used on something: that's not necessarily tied to strings.

C

We use string, compare on things like read-only Spanish are frequently so, but isn't that charm, because it's meant to be string well, this.

O

Q

Is not righty, no I'm, sorry.

Q

Yeah the title: can you know that was an initial motivation to add. It was only four units but I'm, sorry, but there is nothing anything else for expanding to any kill, as I could.

E

Oh, that this is a any.

S

Tea should we have something you can pass in a comparative. um This thing already equip.

Q

Heat as I feel exactly equals on what are the key elements on yeah.

E

But what about a custom, yeah.

F

There's a lot of like most of our IP is even if they have an overload. The tix I equatable also have an overload that takes in.

T

Combined, compare if there's a need, foot I, you could add it, but I would move. That's everything. Yeah drink doesn't have a interesting one, doesn't have.

C

One doesn't have the ability to take in a comparable potatoes, don't compare them with its.

L

That might have here's to be orbital.

J

Yeah, it could be the split here organization, but do we have any more load text? The separator is a string. Yes, so probably separator st is not enough. You could have like multiple items.

C

At two-speed here, I'm checking to see if that's what you're sensitive.

C

B

I guess my my feedback would be that we need to be careful about scope creep because the the primary need for this is using like doing string splitting, but with spans right and- and it's almost always not always, but almost always with a single character. It's a comma! It's a semicolon. It's some single character.

J

Don't know what this thing is: okay, I.

B

Realized there's another overload, but if you look at the use cake-like the use cases for them. The 90% case is for the one that takes I either just pure white space like because you can take a separator yeah or that or that uses the single character, because it's usually some you're splitting tab, delimited or common delimited or colon to limit or some kind of.

F

Delimited text, the reason we're going to have to expose one that takes at the very least a read-only span of car is because not all Unicode characters are representable with a single charge, so you have to support globalization and therefore Unicode separator characters that are too surrogates.

B

Well, there's there's a different we have to and like what scenarios really.

Q

B

Q

Common scenario is for single cars that are yeah.

J

But Tonetta is that it I mean, like you know, if you have a string and you want to, you, want to search for a surrogate there.

C

J

Yeah we had a data.

C

Type in dotnet core that represented any arbitrary Unicode scale or value, including you know things in the astral plane at.

E

The same time, what want to split on a graphene yesterday right so like.

Q

E

Far, do you want to go down yeah.

Q

So you could come up in a scenario that you'll need those multibyte, splits or Marty's character. Splits is that this is a simpler, simple, more heavy use here. So that's protein inside work. The problem is that of the strings per ton. There's overloads.

A

That take multiple strings and you want to pull charge. Yeah yeah.

C

What are they actually very commonly? What I would pretend on them? Anything in the what what I would probably prefer is that we have overloaded specifically for charm and those overloads also take things like string comparison and so on inside of them, because when you're doing when you're doing a culture aware split, it's actually really non-trivial component off core. So if we have an API that could take a string comparison as a parameter, then we could all of the logic in there.

U

C

U

Say, don't guess we do? Is we actually sorry do.

B

We actually have concrete requests for anything other than char, or we just doing it. Opportunistically I think.

Q

You had I think what somebody who's like shameful bike, yeah in the in to tear itself, and then he used to discuss this and it was party approved. He heard it needs to believe that there is no point to constraining each other and.

B

And for bite there was a need for multi multi, byte sequences, string, split options and all that stuff. Mr.

Q

Jolly was single bite.

B

Yeah so my my preference basically be since I believe the vast majority of use cases here are and and also the complexity comes from strings right. The multi character sequences is primarily about Unicode, not entirely, but primarily Springs string. Split options is primarily about strings, although you could theoretically say I want a room with MTM trees regardless, but what is empty need.

B

The it seems like if we just had split based on read-only span of char, with whatever overloads we wanted that map, as close to one-to-one with the current overload, so that we have just replacing string with read-only span char and not having the ones that take arrays of strings. But that would get us 90% of the way there we'd get 9% of the remaining 10% of the way there by having one that was just we'd only span of of tea, with a tea separator spell PA.

B

Our rather than PE are no string, split options, no compare and then there'd be the remaining 1% that we can deal with in the future. If meat really presents itself.

C

So it sounds like your proposal was to have not a generic version only to have the driver or, if I misunderstand you, the.

B

Nine percent for the last piece would be it would be the generic it would just be split with taking this read-only span of tea and a tea separator and.

C

Maybe also a comparer or is that even I.

B

Was saying just just leave it as a TI equatable I gave its the only other scenario we know about his bite.

B

That's my suggestion: isn't it you have the one for char that covers like all the complicated string stuff and then the more generic one that allows you to do it bites me integers wherever else, but.

F

Isn't not supporting Unicode characters initially going to completely prevent it from being used for parsing, for example, emojis, which is I.

B

Don't know why why would the 90% that I was talking about was based on read, only span of char yeah.

C

Okay, we already have the api's of let you do that. It's just not.

Q

What you could build wrong, but most of the time in the artists pan API is there good. Indistinct was based on concrete scenarios and even further 90% case I think those days you mention, obviously exists. What people can do you can use use, string, AP is, and whatever is they're doing now. You know, unless you did enough evidence to say, look people actually using not asking questions after separation or emojis or might be by it I well,.

B

I, don't think I.

Q

B

Like I feel like I may have been confusing.

B

So let me let me try again what I was proposing is two sets of overloads one set of overload that Maps basically one-to-one with the existing string methods we have and that are implemented with we'd only span char, there's no T, so we'd have one overload that so we'd only span char and that's it and there's no separator, because it's based on whitespace and we'd have one overload that was read only Spanish R and a T separator a charge separator and one overload that was read only a span char and then the separator was a read, only span char right, yeah, it's or maybe it's a string, I don't know we could decide that, but it's and that's the one that ends up supporting the you know: multi char jars and then separate from that grouping.

B

We have a single overload that is read only span T that takes a T separator and that's it and that's only note. The Antonio.

Q

Q

B

Because it's for multiple reasons, one, it's not clear to me what it necessarily means and two most importantly, it seems like it's the wrong domain to have something that's relevant to strings applying to something else.

O

Second, one is just separate terror.

F

Make it clear that's interpreted as a single separator string, it's.

A

Honestly, though, that's one of those things every would differ from the other ones, no, because the other one, basically that we have a char, a runaway the strain one, but in the char a one is basically an alternative and then the string, one is the one. That's the sequence, I think.

Q

Yeah we got yeah.

B

Ie MIMO, are you saying we have two overloads one that takes the char where this it's treated as individual character, separators and another one that takes a string where its treated as a single separator I mean.

A

That's basically, the decision you would have to make all I'm saying is that this one string there's a difference between the one that takes the Chara and the one that takes the string. Yes, one.

B

A

Alternatives and these three and one is considered a sequence: I think that.

B

I would then I would say that the second overload here should take a string.

Q

Yeah decided that if you wired the Chara support, it will be split. Any everyone, no do not over the same thing you named differently, I mean.

P

We could do that too.

B

Problem, though, is if, if someone's refactoring their code, and they just change the type to span instead of string and then they end up binding accidentally to this new overload, where their second argument is treated differently. I think that's a problem and.

C

That second argument is almost always a literal yeah.

F

So maybe if we wanted to have a second split that has a read-only span of char, that's treated as a string rather than is separate, we should put that wonderful, I, feel and I. Think at that point. Split.

T

Any as as you have proposed.

Q

To people so, but what if I'm kinda stupid on us things like I would have rather good thing: you're, not delicate I,.

T

C

Think that's a common scenario.

Q

Yeah the whole part of this one, immigration after was doing news allegation. Oh we.

B

Are are, we are reducing allocation? The most common place that you allocate is for the first argument, not for the second I.

A

Want to say that name isn't never the second one is they do if I get a slice, just not a little like I'm, pretty sure like 99% of use cases are the truth. Okay, because it's a comma or as a semicolon. It's the pipe right back.

B

hmm So anyway,.

P

B

It were me, I would want another overload at the top here that didn't take a separator at all.

C

Little splits on white space yeah, so we don't currently have an overload of string da split, but does, though, maybe we shouldn't introduce out at the same time you do those we.

B

Do don't we we.

C

Have one that takes up ramza.

B

J

The one that doesn't think it.

A

Happens, if you call this one bargaining, 3, sorry, you're, right, I think this is hopefully handle the presentation by saying, if empty I know, then could you be yeah.

B

It's it takes, takes a params, char, Prince, char array, rather I'm gonna.

F

Treat that as no replaces looks like no.

B

F

Split internal separators: if separator list length is 0 return, new string array, this.

F

Blossom reading the code wrong or missing something I.

P

B

Sorry, where are you.

N

Where are you looking.

F

So, let's not net I went to strings split prams, that just directly calls split internal yeah and then separators is made into a separator list. If the separator lists length is zero, then it returns a new string array containing the original input string. So that's separator.

B

That separator list is something entirely different. That separator list is finding all of the separators that were specified like look at make separator list. It has a case for 0, where it then uses chardot is whitespace and in what's generated by make separate or list is the list of alcohol.

C

And desist amount.

B

C

So you're right it is, it is.

Q

B

If we wanted to, we could do the same thing here. We can make this a params. We could take a params char. Alright, you want to do.

A

But that would mean you're located on the common path right right. Well,.

F

Until C sharp 9, where this is.

A

F

Anyways with on at 5 and we might have prams or a crime span or a by then so.

B

It depends like, if you don't pass anything you're gonna.

G

Get a raid on empty like the.

B

C sharp compiler is gonna generate a raid on empty, so it's not gonna allocate on that common path. It would allocate if you did comma a comma, B, comma C like.

C

Honestly, I do like the idea of those just being two different methods entirely, like I feel like string dots, which should have been that way, because one method would be documented as I split on one space, and one method would be documented as place flip on exactly when you tell me this, but on I. Think.

I

It was just the fallout of the fact that there was a parents want. You know the parents, when you can always call with you. I came again.

Q

And this I think this is. This is a point so again it's to the ad overload. Then I go in for a second one. Christie span teacher separator.

A

I mean this is this: is the link all right? That's all my point of view. The difference between this and everything is generic is is close to zero. The only one where this is funky is this one yeah right, it would be a bit weird to say your partner may be the only span of byte, and then you specify a string separator like what does that mean? Also.

R

What's one fortunately, there's a light space, but.

A

The first one, it's very clearly to talk about utf-16 right well,.

Q

If T, what does it mean in terms of what.

A

I'm saying I think if you, if you replace charm of T here I, think everything makes sense except the solve alone also goes away. Yeah.

Q

The person I don't understand how the course will make sense. What.

A

Would be whitespace right.

R

It doesn't make sense for T yeah, oh I, see you're, saying the only the say.

G

Only the middle one.

R

Would have the generic one yeah I? Guess that's fair I guess this. They also.

E

Want overloads for great.

C

Yeah, so at that point, I think we should introduce to remove ambiguity, who are assonance and spaghetti as in if you identify any of these separators. That would be for that addition.

Q

Until except impossible, so the last one don't.

J

You think that can be used with UTA.

C

It would be probably Burchard AJ.

J

C

Your yeah, if you're treating the data, is asking it would work if you're treating it as anything other than a ski you're gonna run into problems. Maybe.

F

We should flag some of these api's, which were reviewing for dotnet 5, as if we're adding char 8 for dotnet, 5 and C sharp 9 then come back, do a pass over all of them and add char and overloads as appropriate. I.

Q

Think we would do that for all Australians manage our ideas anyway, right.

F

But but this is, but we did that for the last one, knowing that we needed to add that we'd only span defied overload in this case, it's saying there for the next release. If we add tar, eight only add the charité overload, otherwise consider adding we'd only span a fight that way we don't have to unnecessarily duplicate in the next year. We'll work my.

J

Other question about: if somebody today, you want to use it to do TV yeah. It's thing: bites I,.

C

Flagged the issue as there on hex utf-8.

J

Yeah but I'm hitting just a photo also for secreto yourself. Little doesn't doesn't make sense to be one bite, that's correct, yeah. At that point we.

C

Would either use ruin utf8 stance all right guys? Who is that it look.

L

Good, yes, no and you create the label. I did I created the label. Yeah you get into.

C

The conscious create enables like good.

L

Why it wasn't prevented from.

C

Creating the label, it's.

F

Okay, we're merging me.

B

F

Changing all the other.

B

Options because we could just stick with the char ones and punt on the T ones. Until we have a better sense of.

G

B

Like the person who was asking for bite was that for utf-8 or something else, it.

C

Was probably for asking specifically I.

Q

Don't remember.

Q

If you look at the illuminator self I.

B

Mean yeah before we do it just to highlight, so we haven't factored in the string split options into the first three, but personally, I. Think that's fine, because we're talking about something relatively low level here, we're not allocating anything. You're gonna get your calling move! Next move next move next and each time you're gonna get back a range and if the range is empty, you could choose to skip it.

Q

B

Well, whereas with string the string ones today, it actually affects how big an array is allocated.

B

So I, like the first three I, think we could choose to punt on the last one. If I wanted to I.

J

B

Sorry, Austin, you were asking to look at the enumerator. Oh yes,.

U

If people can scream.

C

To change the integrator into the range based.

Q

That's not good.

Q

A

A

And so my first question is, will be still global, for generic wanna will be also hot. Cordless is that the char you're, probably better off making this guy generically and.

O

That why not we ensure than the future yeah.

A

Making what renu-it stands to the numerator we've got was on what the over those are, because that means there's.

B

A neat that doesn't need the constraint.

E

I think for performance, you probably make it different persons, a 10-step, two types, two two types of walk: one for read on the span of char: if you're using those special overloads and one for the generic 15, what would it do differently.

A

Made logically, what I would expect is the space that enumerator would I, don't know how you would implement it honestly, like us and I. How do you implement this business like this? Guy has to do the parsing incrementally, as you como connects right, we would have to have different implementations or it needs modes or something on it right, so sharing the same time could be hard.

F

A

O

F

Type and specialized.

C

Based on what your news yeah, based on what points to be supported, not necessarily because how do you do like white space versus list or versus strength or stuff like that, I mean you, you could do it it just. It means that there's a lot of state contained within the Union is. This is one of the problems I had with all of the Rooney enumeration AP is like I, actually had to introduce like four different, a new minor types, because they all have slightly different logic. Yeah.

A

I mean, quite frankly, I don't care too much, because this is not a tie. They accept people to exchange at the extent people to just buy for each right and whatever the type is, is retired. Let's all somebody make them s the type, because we don't expect anybody to ever put them anywhere right now, so long, the stay. This is something that I think should be good. Mobile, instrumentation and you know see what works I want.

C

To talk about the range exactly like it's this: is it going to be a common scenario that someone's going to ask? Is this element of the enumeration empty, and how do you get that answer if you're, given a range, the closest thing I could think of is spanned sub range both length equals equals zero, exposing range so.

B

The benefit of the range is, you can then split things like memory, so I can pass it. I can say, split memory span and then use the ranges that I get back to get ranges from the memory.

B

I can't do that if current returns a spam, oh because.

A

It can go back on the stand to the moon, yeah.

A

Yeah Abbas API was never safe right, so that was one of those things was like well before they have to have specialized overloads anyway to want to make that work.

Q

It is a couple of educational sporting, all in good that I have it again, look at it. You know I mean.

B

Q

The other thing.

B

It's it's the general idea that you can. Basically here's told the indexes that yours you know coming from into. You can then use that to do other stuff, rather than just get a spam yeah.

C

So each bar range, in whatever dot, split into that bar segment, equals original span some brain yeah for original memory, submarine.

A

Okay, I think, once you have it, it's me I'm pretty sure most people that wanted to get back around to the prototype in this bag. You confused over the compiler.

C

Error for more women men, woman ice- the other nice thing is: it allows you to use regular spending to breed almost family.

C

That is true, so we don't end up with the type this working.

A

Yeah, let me just add.

B

So if I understood correctly were saying, the shape looks like this, although I still question the wear, TI equatable and then based on implementation, we decide whether we need to change it.

C

So we could get away without the TI equatable as long as the split method itself still has a concurrent, because a split method could just get the correct, Herald fatigue and then swirled out evening I mean it would be a little slow because there's no need you love me.

Q

Only the benefit of not having the constraint on the enumerator I just justified that it doesn't make sense. Rather, you can doesn't make.

C

Sense data, so if you, if you even if you have an overload in the future of split that, takes an icomparer of T as a parameter, so T no long or ie quality compared routine, he no longer has to be ugly.

Q

Let me loose in genetic exchange ever if.

C

So we can well you.

Q

G

C

Because if this generic constraint is being added into another method to satisfy generic constraint, even though you know how important we add a new.

E

Overload than we could return it different type of place.

Q

Was previously a key cradle and then it's known alert available, that's what they contain. Is it I think in the opposite direction, I mean.

A

This is a constraint on the input. I, don't see how we're moving their constrain. Is it going to change no.

C

You're right, it's it a constraint on the output would be a breaking change with convicts.

C

Well, if it returned to read-only sanity, that would ask pull the card.

A

Properly I mean like even the chaining, is not a problem, because it depends what what would the calling methods he is constrained by a and that's independent? What this method is constrained by right, like the only case, removing a constraint can be a breaking changes if it's. If it's used in overload resolution for betterness right and now you remove the constraint, no more things show up in the Nathan to give it its bed yeah, that's in the normal realm of like adding an ethic and organization anyway.

A

So my other question is: is the only reason why we use return range is so that we don't have to provide three overloads the tape memory. So that's what we've done in the past of what people having to requires domestics to go back from spending I mean it.

Q

Wouldn't just be, and.

A

Q

Was reviewing span, but here yeah it.

C

Wouldn't just be over long as those overloads would have to work on different enumerator types.

C

You would have memory of tea. We go. May.

E

Still get to you stand.

C

E

Us I think it's probably going to be common for parsers to want to operate on a spend for efficiency, but then translate that into occurring their own data.

C

Links that there was an API proposal earlier or like slicing a memory of tea to a span of tea.

F

Right I think we've reviewed it like three times now. Each time we've said.

A

Fifteen minutes over so like, let's try to walk this guy in and.

J

You keep using the range and we just which isn't like.

T

Range between house it directly into the indexer. Yes,.

Q

Their API is the big rage on span and at.

Q

R

Supposed to do.

Q

Nothing, so what would it take to add back the generic split I mean a use case.

A

Is anything I haven't captured.

E

I think any I deserves from there, because that's super common to use.

S

Q

I just think the don't get in for the dinner I expect, based on user scenarios and requirements, you right, yeah.

F

Yeah I feel like we're going to end up needing to add it, though, because the those overloads are more frequently than these singular overloads.

F

Just looking at API is starting that usage.

A

And it was the impression that had us a little bit like I, think some of it is also the shape of the overlooks that we have yeah.

F

That might be part of it.

C

Nested types on one.

I

I'm memory extinction so I did they get. The numerators are nested I, never extension, rather than being down pub level.

F

Oh point: zero: two percent for split string, strings, full options versus 34 percent, for split string array, string, split options, so the other one basically has no usage.

B

Yeah that, but the string splitter a I would I I would bet money that the length of that array 99% at the time is one yeah.

C

That might be the case.

F

C

Further about those people are probably cussing. My parrot disgrace.

F

Oh right, that's because we added the one that takes one in life, thought minute or two.

F

Yep thought that core to is when we added the one. That's got no usage so that we only had the allocating version before.

L

A

It's a coupe of you, curly, hair.

Q

Funny how this this API proposal change based on my brother was because he approved it. It was and then let me come back a different it. Yet honestly,.

B

I think we're gonna find based on implementation that we're gonna come back here. What we'll see.

E

I could have used the multiple separator split, inserted, I, see.

R

B

I mean the flip side of all this is it's just a helper right? This is, you can do this with index of and index of any loop.

C

Until we make colder ones and then everyone just cries, I don't have a culture or split.

A

No because it needs to do culture way of parsing.

A

And ideally designed I personally find, hopefully all in the legal case or something like that.

A

Alright, thanks bye, Steven.

B

B