.Net Foundation Design Reviews, 1 Oct 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

00:00:00 - Approved: Add Encoding/Decoding APIs for new System.Buffer types https://github.com/dotnet/corefx/issues/41166#issuecomment-537141144
00:15:18 - Approved: Add Encoder/Decoder for new System.Buffer types https://github.com/dotnet/corefx/issues/41326#issuecomment-537150026
00:35:57 - Approved: Add List(T) AsSpan to CollectionsMarshal https://github.com/dotnet/corefx/issues/31597#issuecomment-537151655
00:39:36 - Approved: a new GC API for large array allocation https://github.com/dotnet/corefx/issues/31787#issuecomment-537157276
00:52:58 - Approved: long ReadOnlySequence(T).GetPosition(SequencePosition) https://github.com/dotnet/corefx/issues/31807#issuecomment-537160854
01:01:58 - Rejected: Color should implement FromHsl methods https://github.com/dotnet/corefx/issues/31820#issuecomment-537164275
01:10:15 - Approved: Extend CultureInfo.GetCultureInfo and new CultureInfo to optionally accept predefined cultures only https://github.com/dotnet/corefx/issues/32201#issuecomment-537177127
01:39:50 - Rejected: Add remove Range to sorted List https://github.com/dotnet/corefx/issues/32987#issuecomment-537183526

A

No, a stream all right so almost.

B

Like I've been done before, and so in a nutshell, basically, all of the API is on system text. Encoder today work either with arrays pointers or spends David and the trust rule to have a scenario where they wanted. They wanted these api's to also be able to work with read-only sequence and I buffer writer, we're dealing with really large payloads things that can't fit into a single span.

B

The scenario for this is like I'm reading data off the network I'm the web server, for instance, it's coming in as bytes I- want to push it out to a buffer writer that will generate a sequence of chars. For instance, me. These api's can't live in system, can't live directly on system text encoding because of layering issues, read-only sequence and I buffer writer aren't in an assembly that's reachable from system private :, but they could live as extension methods in the same assembly that I buffer writer and read-only sequence live in.

C

Just a fire by this isn't just for really large payloads right. This is just because we don't Lee sequences or the lingua franca pipelines and so correct.

B

C

B

Its I shouldn't I shouldn't have said the scenarios only for large payloads, it's really for chunked payloads or the whole thing can't be for whatever reason where the whole thing is not representable as a single span. Maybe because it came in as two packets on the network.

B

The fact that these are extension methods means that it is obviously possible to write this code today. It's just a little tricky and you know we're hopefully in the business of making usable APs. We.

D

Don't have the including attentions type today my disease would be a brand new thing. Yes,.

B

The name space would be system text because we always use the same namespace as this parameter and the assembly would be system buffers, I, guess just a memory wherever read-only sequence and I got delayed early.

B

E

Oh Zen petition will be like it should be pretty straightforward, yeah.

B

We can we can optimize the implementation like we can. We can take shortcuts to say: oh, is the read-only sequence, a single span, as opposed to a sequence of spans like? If so, we can just go down a fast path, but otherwise this implementation would handle all of the merging of data itself.

C

I know there's been some discussion about read-only sequence and its size in the past. Is it still of a size that we believe in is correct?.

B

I, don't know the answer to that. However, I know that a bunch of our API is that take read-only sequence. Do take it as spam.

B

For whatever reason source, don't love isn't a source. Dotnet is currently.

C

B

C

Anything outside of court look okay,.

D

So either corporation.

B

What why does it need to be extension methods, as opposed to an instance method yeah, because system text encoding is an assistant, private, core lab and read-only sequence, an ibuffer writer or not.

B

So the extension method would have to live outside of system particle.

C

Or we figure out what the point in time is when we decide that read-only sequence is considered core enough to be in Korolev yeah.

D

I mean I, don't know like this seems to me to me that kind of that. Turning point all right. It seems a bit weird to like introduce these higher-level things. Just with all that interaction when more lower-level things kind of would benefit from it like it seems, I, don't know, I know whether we have other issues. We would like to take it off provider, but I would guesstimate based on the way the type is being used. It's probably not being the only case. We would like to do that time. We.

B

Decide at some point in the future that we come over the extension with identity solutions. We can't ever turn it into an instant instance method, but we can move this entire type down into system player hologram. You want to you.

D

Just have to know yet another time that you can never remove the encoding extensions. You.

B

D

B

We have extension methods hanging off of read-only sequence today, I'm sorry, we have. We have extension methods that take read-only sequence as a this parameter.

F

B

Extension methods appear to be marked as in us, as opposed to just working with us, so it looks like the pattern is still to take it, as in.

C

Yeah I'm just asking because I know some changes were made late in our late in the last release, I think to read only sequence: to reduce its size and I, don't know to what size it was reduced and whether we believe that's below the threshold, since in general, we prefer not to pass things yeah.

B

It looks like right now the size, ins, two references and two integers and.

G

B

G

At least 24 bytes on yeah, and so that's going to be copied by puppy bug failure you copied by reference internally, regardless of how oh yeah so in just avoids an additional copy that happens at the EPI layer. Yeah.

B

The things can look 10 or so the things going to get stacks killed anyway.

E

So we have another assembly which has system texts, including code veggies. You think possess and live there.

B

So you said system text, encoding, yeah.

B

Dark hole: haters. Yes, it doesn't seem like.

E

A writer somewhere myself, but it includes all encodings that we support today, almost nothing except is a only good one.

B

F

B

Seems odd for anyone to take a dependency on that particular assembly just for this extension method that doesn't really have to do with code pages right right, but it's this.

E

Assembly as in box anymore now and it's it will not cost anything new, but I'm just telling about like here. We have this option.

B

And I promise I'm not pushing thing. It is a read-only sequence. An ibuffer writer available in town will like do. We have a nut standard package or something or.

D

A little console me.

B

We'd only sequence, an ibuffer writer I think.

D

I guess we have.

H

To because the editor.

D

We don't listen to an exoteric to r21 I'm.

H

Not sure we actually expose the whole system text, jason has a dependency I'm. Sorry, it might be a memory actually. Okay, yeah.

A

It's the system of memory. The chips is a pseudo memory. Do they probably so.

B

Regardless of extension, method versus instance, method where it lives like, are there any problems with the overall API signatures?

B

We're trying to do I mean they. They are patterned. After the existing instance methods on the encoding class.

D

Well, the current ones get bites. Oh answers, nothing's, right, yeah,.

H

B

You know take effective. The encoding is the first argument. Yes, there are some differences, for example, the get bytes method that writes to an ibuffer writer. Instead of writing to a span returns a long instead of an int, because you could write more than two billion elements.

B

But beyond that, the pattern is intended to be identical.

E

So, do you want to get like good bite counts and.

B

There hasn't been a need for that yet, but if we decided to add those, they would go on the same extension class follow the same pattern.

B

The thing about get bite count and get char count is normally. You would call those before calling get lights and get char so that you know how how much you need to allocate. But in this case I buffer writer is itself an unlimited sink. So you could just start writing without the need to allocate at a time like the writer will do the allocations for you or the rental or whatever it needs to.

E

Yeah I agree: I mean like this is a scenario but like if somebody won't understand about how much data is going to be encoded before heaven, it's not about a location, buddy yeah.

B

Ok, we could decision. If we wanted to, we could add that it would return a lot in both cases.

B

There is already a get bite count that takes a span, so we don't have to worry about that envelope.

B

So get pipe counting get charcoal as well. Have they really? They are existing instance methods on the system text encoding yeah. The idea would be to make them extension methods here as well. Alright, so we should, and so what would they look like so get byte count this encoding and coding in read-only sequence of charge, chars.

B

Basically, you see the one that takes in read-only sequence, char and then writes to my buffer, a terabyte. It's the second function in the extensions, so.

D

You see when the kicks in coding, sequences and bytes the.

B

Second function in the extensions that takes the encoding in sequence and then writes to you later, yeah.

C

B

That to leave the last parameter and rename the method get my count.

A

I get by columns.

A

D

And we don't need that for Shh.

H

People he was longing for char as well I guess, if we'd only spam, you already have to add the correct it already James's right.

A

D

Get your account, which.

H

Takes an encoding only the only sequence of byte by and then returns along.

B

Again, the fact that they were turned long is different than all of the other methods of a tournament, but it seems appropriate for the types that we're taking them.

B

A

A

A

B

Another thing that we could do is we could say, if we're, okay with in general, what the call sites would look like. Then we could have a separate discussion as to whether to bring read only sequence and I've upper right or down in the system private :, but that shouldn't stop us from actually being able to work on this ocean and create the implementation.

D

The other column is like everything the whole point of doing. This would be to avoid having these extension time right, yeah and so there's always remember to not edit and then Vader.

B

D

There's is there anything.

B

Wrong with us doing it in core effects now is an extension and then decided later, hey that really no.

D

There's nothing wrong with it. It's just that it tends to be this kind of that. You forget it. We shifted it. Let me move it to call the membership the extension time ago. That's like the worst of all world, but yeah. So.

H

I, just open up an issue nailed oh yeah I mean that will be a good with a little bit. Nobody think of something else where.

D

All right, everybody else, even if you think otherwise, that's going to be mine if I can find the video.

A

D

One of those mornings the kid: what does.

B

All right, then,.

D

There was one more before we go down right now. What is the ocean away.

B

The issue number I filed is for one for something more.

A

One four: seven, four.

D

D

We continued anything has changed. Yes,.

B

I fixed a tiger on your stuff.

D

I also like how, as soon as you hit enter, it goes back to the new one, and now you lose the one that I just entered. It could have your stuff.

D

I mean enemies again really affected with one.

B

B

You came for a discussion on the system text. Encoding touch right.

D

All right, let's try this again all right, so.

D

Look the other one, the.

B

Other one is 4 1 3, 2 6.

B

B

Basically, the exact same thing, but instead of on the encoding class on the encoder and decoder class,.

B

So if you're, unaware of those classes of the system text, encoding class, is used when you have the entirety of a piece of data to convert- and so you say here is my entire byte array or my entire byte span, or now my entire by read-only sequence, the encoder and decoder classes are when you need to perform something in a chunked fashion, and you don't have all of the data immediately available. So this is used internally by things like screen reader and street weather, because you don't know when the data is going to end.

B

So you just keep calling this over and over and over. Eventually, you pass flush when you're out of data. These api's are also modeled on the existing convert methods on this class. Those existing convert methods are instance, methods.

B

One thing that is a bit stranger is there is a what is it normally? The convert method takes a span of mites and a span of chars, and it will out here's a number of bytes I read here's a number of chars I wrote and here's an out pool whether or not I actually completed my operation here. When you have a read-only sequence of bytes, it doesn't necessarily make sense to out a byte of bytes read. What you probably want to out is the sequence position in stem that way.

B

The caller can immediately slice to that position.

B

Or maybe it makes sense to Alden int and then have the caller call slice int.

B

That seemed more logical, though.

E

I would prefer to keep it consistent ether as it out.

D

I mean what is the very next line, the color right so.

B

Slice and this.

D

Inverse equal solution make a difference. You have to first get a position.

D

B

We don't didn't I agree of Terry. If we do, then that kind of sucks but I don't know about the runtime complexity of the different slice overloads.

B

It's mid-flight already talking about in 64 versus signalization.

F

The sequence position, one is all one: the n64 one is a number of segments you have, even if the controller is positive, that's right on sneer. Only sequency, yes,.

D

Okay, well, you have to like Mack like, depending on the whole largest number, is what it is. That's, basically, all the number of seconds you have to advance right. That's right! Yeah! It's not the number of seconds. You already read right, so yeah.

D

I mean which in practice no means he would have to agree the really large training order finish the way. The first way you're.

B

Just reading stuff off the network edits.

F

A question about the general API: how come you're, not using operation status, style or junk in code reading.

B

Writing because this match is the existing instance methods on the encoding around the encoder and decoder classes. It has nearly the same signature.

F

So, by the same time like you will be added other api's, especially in the encoding space, mmm we did follow the existing instance methods. Buddy followed the original self, no, no, not in the encoding, namespace or not.

B

On the encoding class, they don't expose any operation status methods at all. Okay, the only methods that are operation status or methods that are completely brand-new unrelated to system text, encoding and part of the reason for that is. You can't actually get an operation status back from an encoder or decoder, because there's too much state stored internally and we can't crack that state, open and figure out. What's actually meant.

F

F

D

So, what's the verdict, then, on position races and do we care.

B

So, just as a point of interest, you'll notice, some of these overloads don't out the number of bytes that were consumed. The reason for that is, whenever you're right into an ibuffer writer, you have you consume all of the data, because your sink is unlimited. If your sink is unlimited, you always consume all the data. So that's why they don't out anything.

D

With me, and so.

B

You have you, have the source and you have the destination right.

D

So if the source.

B

Is a span or a sequence, and the destination is a writer. The destination is basically infinite, like the writer will always.

F

B

Write more data I see. That means that the writer is never going to be the thing that limits you, which means you're, always going to read all of your source input. So, in those cases those api's don't out anything for bikes riding because they just read up we're going to the entire sequence: were the entire Spanish consuming make sense?

B

Now what we could say in order to be consistent is we could still out of ice used or bikes, read and always say like this is contractually guaranteed to be the entire banking. The signals given.

E

B

At the last meeting we talked about stuff like grid docks, I write out in 16, yeah, hey.

D

Those is kind of like different, though, because you you, like you, need that number. It's just whether you have that number given to you or whether you just know it on top of your head is the difference and versus he would never use the number. So you basically have an argument that you will never that you have to assign to some local that you will never ever leave from. So even if this guy already takes like six argument, seven arguments and two of them are out I'm, not sure adding another out.

D

It's like going making people's life easier. So.

B

Awesome in in your opinion, like is it more usable to out the position rather than to out the number of bytes right.

F

Maybe but I would have to think about like so you see, for example, if you outing the secret position, then you're forced to slice the really sequence create yesican.

B

Actually, generally, what people do anyway, that's the expected behavior record for smoking.

F

F

Yeah I think that that makes sense primarily because if you pass, if you return an int- and there is no slice, then a lot of time that the slice will not be all one.

B

So you would have the position instead of a longer on it.

B

I

Passing me agreed to the contract of musings.

B

Position would we call it by its used position.

B

It seems like this strange grammar name.

B

What does bytes used in that context? It's normally bytes used, would be the number of bytes that you read from the input source in this case. We're not giving you the number of bytes that we read we're giving you the position where we left off yeah.

D

Which? In which case applied to use relation things.

B

Oh bites used as the parameter name used in every other overloaded list method. That's why it was listed as up and I see.

B

We could literally just call it sequence position what the lowercase awesome.

B

They'll say this is a special case. One of us is really concerned about making it consistent with the other brand names.

D

When by juice is, unfortunately, because it kind of makes it hard to combine or things like position, R and X equals position, bytes use is like what would that mean? You kind of have to be dog.

B

You could also call it a remainder or something like that.

D

That is basically the next position you should read from yeah.

B

Basically, slice here and then call me again.

D

I mean given that it's out I, don't know it's end of the business yeah.

D

Resulting position is the best thing I think I'm up to, but that does not have any big guys. We have support it.

F

Can you really create what it's supposed to mean? What is the vise use position supposed to mean.

B

This is where we left off reading from so the next time. You call the convert method. You should pass me a read-only sequence that is sliced to this position.

H

I

H

Videos, what do you stop? Yeah yeah? It's the next start position right, yeah,.

D

Yeah, it's good of them.

B

Convert okay: when does it not end designate so it convert? What convert will do? Is it will try to take as much data from the source as a cam and transcode it into cars? There are a few reasons that it might not finish. The primary reason is that you're out of space in the destination and they're still source sites remaining, another reason that it might not finish is that the source is split across a multi boundary.

B

In that case, the bikes use position will point to the end of the input but completed what would run false completed basically means there is literally no of my internal state is recycle. There's nothing interesting interesting, because this is normally am the encoder, and decoder classes are state, maintaining they are actual classes that have internal state. So.

F

Questioning their state maintaining, why do we need to return sequence position? Have the user slice that you in a sequence? Why don't you remember what position we were at and just continue from there? Well.

B

Encoder and decoder classes don't have this; the only state they maintain is the last byte or two that they saw. They don't actually know anything about, like the in hertz type, like sequence span. Anything like that yeah, okay,.

F

Finally, for context like utf-8 history returns people's position, but I just called a position yeah. These are some data property. It's.

B

F

Primary name I mean.

B

Musicians, law, that idea and.

D

You want to see I can do either one position.

I

Vectorization yeah.

B

Because there's only going to be one position ever returned by these guys.

G

Hopefully, as cheaply as possible, yes.

D

So when position and.

B

Final position, just by itself, we also have offer they can have buffers yeah.

E

Possibly we can have like efficient buffer consistency. Well,.

B

Remember, encoder and decoder are abstract base classes, so we can't assume that any derived implementation would understand this right. This would. If we tried to build knowledge of this into the base, encoder and decoder, we would basically be adding a new abstract method which would break every implementation.

B

So this has to be built solely on two applicable methods.

F

Which is fuzzy, how beautiful do we have use a sample of how this API would look like with the slicing like Howard Collier call it there's.

B

um If you look in the US has for encoder and decoder today, there are samples of calling the convert methods in a loop, there's also samples on MSDN for how to call the convert methods in a loop. In those cases they were dealing with arrays, but you can imagine that they're dealing with freedom, we see quizzes read-only sequences. Instead, you know you just call the slice method. Instead of allocating a new array segment.

J

And so position.

G

Is going to be simply position?

G

Yes, because that's what it looks like slice.

G

B

You mean the name or the type, the type yes, it's gonna be a sequence instead of an H, Dale I think that slow would have to be along. I long.

J

Yeah because that's.

G

What doesn't have to go long, yeah, yeah, yeah cuz? That's that's what we don't need. Sequence skip position, currently return sequence, I! Imagine that both we and returning and you slicing by.

B

The way you'll notice that or the destinations were also outing bites used. Cars used like for the number of elements that I wrote to the writer or the spam I would prefer for those to be call written, but there is precedent that those parameters are called vehicle away. If you look at the other over loads of converts, you know we out int used to represent the number of elements that we wrote instead of calling that parameter written so as much as I, don't like those parameter names, there's precedent, so I'm not going to fight it.

B

To my how much pressing on this particular type, the only precedent.

D

They generally try to be locally consistent when they all liquid system, video.

K

D

More useful, how much well over those busy you said, yeah.

G

D

D

Alright anything else otherwise, finally, epi shake I mean it's the same magnet as before. But if you move this down, then we probably should, if.

B

You move until these would become a non virtual methods we should, or maybe they would be- virtual I- don't know they should have the abstract, though we.

G

Should propose a language feature for a new attribute that we only put on our reference assemblies that says this is alternative names allowed for this particular parameter. So that way, we don't change it to be more correct and still be backwards compatible by.

F

The way, if you're moving these down or considering them, we should talk about a liability, because for the last time when we talk about a span methods or list, we didn't consider another ability and they live in cold egg. Well,.

G

The only no ability here would be ibuffer rider and the input parameters and I think we generally say inputs are not at all yeah.

B

G

There's you can't do anything of riders and all.

B

G

B

Nothing here can be no saying which the previous thing that we will get.

F

Okay, so there's other, we know annotations on on these ApS, then correct.

F

It means that the general principle like if you have something that cannot be none of these two argumentation. We don't market as nullable.

D

If something cannot be now- and we don't do our validation, no.

F

If we do argument validation in the method and to say, if your argument was null, we throw there's a signature. Should the signature not have a question mark, we.

G

Have to do that you you have to check, even if it has the annotation, because you might be calling it from a version of the language. That's not no I.

D

Think what he's asking is like is there a point in marking? Things is nullable when you throw, and it would say no because they're fun, you're lying to the caller I just mean accept not, and you blow up into taking out like that seems backwards. Yeah.

B

And many times were extension numbers going in traffic. You spoil me.

B

A

B

We'll get about a book about always consistently.

D

Basically means they are not annotated, correct, correct and we have to find a way how we do get made it'd be just to the pragma a little any little in the sample code. So we know that people thought about it or something, but otherwise I mean.

B

That would be using enough to do here. Oh yeah,.

D

No, no just for the API VP otherwise like how would we know that? Somebody, though you never thought about it,.

G

Annual mobility consider well make sure it's a color that Carol doesn't like how.

K

B

Change? Yes, though, that's fine, yeah, Billina.

D

All right so then, let me paste again and I guess this one is a PR me proved.

B

That was easy, plus I have 45 minutes of your time for tuition.

K

The reason that we have to pollution all right, I've.

D

Already skipped those guys I think, last time we talked about collection, washing, they did not finish if we continue and like finally,.

G

We did finish that okay.

D

G

Tweak based on feedback that Yan and others gave in the actual implementation that we just need to would make sure we addressed and said: that's okay, namely I was back. We we had previously said the approved. Api was both a span and, as we normally Spanish a span mutating the version, the implementation that was suggested by y'all and due to feedback you might as well always we just have a span and it doesn't increment the version because it doesn't match the underlying semantics of mutating anyways. Yeah I, like.

B

That sorry awesome, you were saying you cut out earlier.

F

I was just saying that I marked it back as ready for review and added a note for what you want. Whatever yeah.

B

Whenever you want list be knowable in this case, uh it can reverse hurt an empty span. I believe this list is no yeah.

G

B

So that would actually be this.

C

And we're doing that for her to behave similarly with both string and race yeah I'm asking. Is that yes case? That's why.

B

Definitely no you're right. This is not. This is not this. It's just regular mr. yeah reach.

G

Out to your extension, that was of our home, but yeah for most of the api's that return spans. Since you can create a span around null, we just say accept null and return people.

D

What was the rationale iPhone watching.

B

Remember you once you get the spam back up, writing to the spam directly, isn't going to increment the version number of the list. So why change the version number at the time when you get this family yeah, because normal less semantics is, though, every change to the underlying data and just getting the spam hasn't mutated. The data.

I

Summa bleah, if you get a writable yeah.

G

That was what I said as well, but Yan and Ben both push back against that the.

B

One of the other arguments that they deployed was yeah.

A

B

That this is an unsafe operation and once you're dealing with unsafe, like why are we checking I.

D

Think I mentioned that last time, but I think it was the only one in the room but I kind of assumed young, but but that was the aunts position. Remember there's one of those API is where like, when you go down their path. There are no more protections for you yeah. You know you could be.

D

All right, so then we are back to this all right.

K

K

A

D

I like how this is always popular on the wrong window, I need to do something about the videos. I hope they make it more Android about the other guys. That's.

H

Right, a little desk, I forgot.

J

I did didn't you just finish, porting it to that core three I did almost faster, asleep. Well,.

D

The funny thing is with the tuning that we demoed at the donut con the tomb in ninety minutes, which is pretty good for ninety projects and approximately four hundred thousand lines of code.

D

And I, so whether this, if you ever movie ever review this issue, I.

B

Think we did previously and I remember what we did is. We said: let's try it out internally. First, do you know one of three? It's courses before we make it public and it's currently used by a ray pull of TDOT shared know that actually.

A

G

One that we should have my own handwritten form well.

D

That's what I'm trying to get it like what change the last time so easy to do. Just return, I, look at Iranian they're, looking for to size.

D

I

So is this just uninitialized our location, that the benefit of we're just saying new and.

B

G

It goes on the large all Turkic, yeah and I think there's some variation of the API that also allows force. Depending of the data. There was a few variations.

K

Holy there's a lot of when.

B

Did it get marked ready for.

D

Review it must be done in the middle there's again, this 24 hidden items thing.

A

D

Young edited in September last year, I.

B

Thought it off: okay, whatever.

D

All right, so let me actually go back to the top, then I think to the only people we had was we wanted initialized and uninitialized to be different methods, and not just a boolean and I think that was pretty much the only feedback we had and looking at the API proposal. That seems to be kind of what it is.

C

D

I can't sigh are you, okay,.

C

Maybe maybe the screen just didn't catch up, yeah.

K

So that seems fine.

C

So mike is most.

I

Of the stubberd isn't currently implemented.

C

Right, like I'm a little concerned about exposing API like this, so.

B

Out uninitialized array is internally implemented right now, yeah.

C

But not all of these, like, if you pass getting true, does that work. If you.

B

Pass generation.

C

Tell you does that work.

B

I believe it does not work. I, don't think the API internally looks like this right.

G

I think for the initial implementation that black mani worked on, they just did a minimal subset of what was your product for internal use, yeah.

A

G

Things like pinning and stuff were also on the table and under consideration. They just weren't needed for Rio.

D

Well, I mean tonight, I guess so: Stephens born, like I, mean I'm, okay with her with approving an API. If you don't have invitation yet I will be concerned. If we approve an API for which you will never have an implementation, that seems bad yeah.

B

We would just need a confirmation from them that they actually plan on getting to all of us.

F

B

In some of these I don't know what the parameters mean, but.

D

I think the length of generation makes sense to me if it makes sense to me it.

B

Doesn't make sense to me.

D

B

Does it mean, while.

D

You get back and way, that's already pinned yes,.

G

What does already pen mean? It means that the GC will never move it.

B

It's like an RBA static.

G

It will never be moved Tyndale.

B

It's still G seeable, so you still need to put it inside of a pic statement. If you want to get a pointer to it and not have the GC reclaim it now from under you.

G

I think it's. If you want to write c-sharp code, you would need to put a fixed statement, but you could technically safely use unsafe as pointer. Ooh means.

B

You need to know that the array was already generated using this one very particular flag. What's.

D

Wrong easing when they have it is going to internal.

B

D

We provide a pin heat yeah when they allocate they put on different heap. So that's why they don't want me, but.

B

My I guess my problem with that particular parameter is the entity which creates the array and passes him. True is likely to be a different entity than the thing that actually consumes. The array yeah.

F

B

How does the consumer know what parameter the Creator passed in for that argument?

B

Well because it seems like the consumer is always going to have to say fixed byte star, my array equals array, and at that point, what's the purpose of the parameter.

G

Well, I think it's more like in the cases where you would use this you're not meant to be taking it. It's for internal usage, not necessarily for public use. So, like you know, if this is an array coming from your internal data structure, you can know it's pin and you can rely on the implementation detail and.

F

G

Like if you're ml and you need to allocate you know a two gigabyte array of floats because you're gonna, because you can walk, then you want that, pin you don't want it move, you don't want to deal with the native memory allocation there and you can rely on. You can call unsafe at point. So.

D

What I would I don't think it's about the you know, avoiding the get you know calling the API that does depending I think it's more about, because you're reading all did you will pin this thing eventually, you don't really want to be okay than a different heat, so.

B

I would say, the parameter should probably not be called pinned they'd, probably be called something like immovable or something along those lines, because pinned has a very defined meaning from like c-sharp and everything else. Like Bend basis, you see it won't understand, like Ben does, does GC actually have a predefined concept, called like a pinned heap or a pend oreille bits.

E

B

Yeah, it happened, it has a concept of pinned object, but that pinned object just exists in the standard array right, just a special handle to it right. Well, then, the standard heap sorry this would.

G

Still exist in the GC, not an external hit, though.

B

G

The heap it lives and doesn't matter: yes, that's only important to the GC itself. It was less than GC. No, this is going to be pinned, it needs to be pinned long-term. Yes,.

B

G

Don't worry about like trying to rationalize moving alright.

B

And that's kind of the point I'm getting that like the there is already an understanding of what the word pinned means with regard to like language. The GC everything else and please put this into a completely different heap- does not mesh with everyone's existing understanding of the word pattern. But.

D

What would so it's what I'm trying to get his neck? How does it make a difference to you? It's just predator. That's know I'm Eric. How would the resulting usage like what assumptions would the user make? That would cause incorrect usage on his side, because in order for you to get, you know a fixed, you still have to put it in a fixed statement, so you still have to do whatever the transitions are that you have to do today to get to a pinned, object.

D

I, probably wouldn't do the unsafe cast that I suggested because like well, you know, I, don't know that it.

B

Sounds like what you really want is that parameters actually saying please put me on a heap that will never cause fragmentation. Do the juicing basically.

I

Is it about causing fragmentation or just I? Don't want this removed if I'm going to use buffer for some IO, it could be an unofficial just to say I'm, never, I never want this move because I'm gonna keep no ending it over and over again. Well.

G

That doesn't matter, though, the exact thing that made me put in the proposal is we'll provide a pen teeth that are only for objects in via this API, its person areas where you have control over the allocation of the object you want to pin, and you want to pin it as long as it's alive. Yes, it will not become. We will not be compacting this heap or we yeah. You will not be keen compacting. Some fragmentation may be a problem, as with normal pinning, it should be used with caution.

G

Yeah I think it like from the from the perspective of the external user. It is pinned and it doesn't matter that it's in a new feed. You.

I

G

May only is comfortable with it being in a separate heap on her and being a VC expert, then I think it's probably fine from the internal implementation, but listen.

E

Things depend on Bennett, yep, okay, so maybe yes, a brompton can be like.

B

Goodman Andy I've been with what the perimeter, what the parameter really means is I. Want you to put this on a special heap. The standard objects don't have anything like that's really what the parlor games well.

G

Then she also put I would like limit T for the pinning case to be unmanaged. References, yes, so, which makes sense.

G

I mean if you could technically today create an object. I'll use, GC handle to pin it and then keep it. Pin for the lifetime of the auto yeah.

B

I, do that all the time no different yeah so, but it exists any standard heap, though, which means that the GC now has to work extra hard to kind of shuffle stuff around it.

G

Right, but to the to everyone consuming it, they don't know that it spins right and things like that, but it becomes an internal implementation detail at that point, and so I think this is more. It's called in to give context to the user you're effectively pinning this for the for the lifetime. You know, no one else will know this unless you've told them this so.

B

For ended, ism.

E

Oh anyone will remind me again: I worked like why we have to one initialize an English life, because.

D

L okay terrain gives you bacon zoo initialize to reallocate enrich the uninitialized away, will not do that. It will just give you a long memory as if you already know the way why.

E

D

The reason being that you originally did they proposed a single API with a beta just have another parameter. That says, is an accession yeah, true false the reason we don't like it, because we look from the cold side. One of them is substantially more unsafe than the other, and it should stand on more than just it would true false.

D

Possibly it's just about that, because if you call allocate uninitialized array that sounds more dangerous than a locator ray and then somewhere the fourth item, when you pass forces, you're gonna have to know that all that means now. You get understand every step. What we should care about, what I'm doing next?

D

That was basically the rationale.

B

All right, nobody, nuts, a little oyster right, violates type safety is what it comes down to look. That's why right.

D

The only other thing that we only said is like maybe when you pass in in true, you may not be able to to have a tea that has reference types in them. That seems fine, but that's not.

G

That would cover the majority of scenarios or it would cover basically everything except for having other arrays that exist within D in you. Couldn't, for example, have a struct of float arrays in. Are you AOS yeah, but she said she was open to further discussion, so maybe that particular case would be allowed I.

B

Would imagine it was just not wanting the GC to have to walk that entire heap yeah.

G

There's a similar discussion. We.

B

Had make sense.

A

D

D

We on fire three issues and decorating mildly, but.

G

More of a slow smolder, I guess the one thing that might be worth saying is: if we are going to restrict end to be managed effectively, and if it's always going to be restricted to that, then maybe you could have an explicit overload of I locate an uninitialized array where t is: oh man, if you wanted that to be explicit or just tell people, you know this will throw at one time if it's not yeah.

B

Well, that's a current behavior right, so one.

D

Of these kind of things is that it has that Commendatore X right right.

G

And and if we decide to allow in the future, then yeah yeah.

B

Also, if position just returns the element index in the sequence, something.

G

Like this already exists, I thought we I thought. I saw something.

G

G

That's not the position that takes a home and return to sequence position. Yes,.

D

Maybe you wonder long from the position right I mean so, as father suggested, you could do slice, 0, comma position, dot length. That's logically the same thing.

G

This covers the case where you have a sequence position and you want to get B the long position out of it. There's another overload forget position that takes long offset sequence position: origin, which gives you an offset from a position origin as a new sequence position. Would you do that as good.

J

G

Get position offset origin, sir.

D

Where do you see this API? Only you don't a sequence, yeah.

G

I'm gonna go missing. It.

D

Was to overloads and get position right now.

D

But that's different right. It's kind of like offset plus origin origin plus offset.

G

Right but I was saying if this one's also called get position, and someone wanted that same thing. We couldn't overload based on that. We could only expose this and you have to do two nested calls what will be the runtime.

B

Complexity of this function.

B

Of which one the admission of love, anyone yeah there.

D

Should be pretty much a one, because you just.

I

Kidding the bass, fishin.

D

Listen, you get that from the sequence. No, the note from the object data into the position and then you just lost in the solution.

D

I mean lung Holly. What I'm really for sure.

D

Just sooner sent you over local, sir well.

G

So there's two overloads for gate position today. This one effectively provides a gate position overload for the first one. We couldn't provide an overload for the second one ever because it's because then the third D conflict for the info.

G

G

So do we want to call give this a different name or do we want to say usually.

J

G

So there's two overloads of get position today. Yep one thing so long offset returns a secret position. The.

F

Other one takes.

G

A long offset and a sequence position start and returns a new sequence position. That's it. This is effectively the opposite of the first. It takes the secret position and returns the overall offset. Yes,.

A

G

We wanted to also have, and we could not also provide a corresponding version for this. Of the second overload right.

B

Disappear and there's a big flipped. That's the only way we would be able to do that right. Okay,.

G

So do we want to give this a potentially different name like get offset, or do we want to yeah.

I

I think mission has kind of a defined yes meeting with in sequence, so I would probably get offset or get index will.

B

This be get offset from origin specifically or yeah.

G

I think this one is effectively get offset from origin or position. It's.

D

Like the the launcher could overload to the to the one that takes two arguments would be either taking too long sore taking two positions. No, no.

G

Because you want an offset from so so this one here is saying: get me the offset of this sequence position so you're, basically getting the secret position start and then computing the total offset into the sequence. You see that the second one is saying give me an offset from this sequence position, so you're offsetting into the sequence position a little bit and getting that total offset. That's.

B

Just an arithmetic operation now like I, would we actually want to know the worst about nothing, yeah.

K

Also, the Novolog basically is. This is already to see best position. You think it's it's! It's.

D

The it's the it's, the resulting layer right.

G

Well, it's an additional operation right, but here's like this, so this would also do like long offset sequence position. Origin also handles the case where offset is greater than or greater than the origin life right, yeah in cases like that. So it's not just like ah it's ad with some additional checking and object because.

B

That's going to be.

A

All these I would.

B

Probably call it get offset anyway, because the existing overload of get position is dim position long offset in returns of position so get off set sequence position position returns long offset like there's a natural Signet radar.

D

You know, does he can get us. It makes one but more for the item that you did yes and.

B

If we wanted to add that other over load in the future, we could- and we wouldn't have the conflict. The Tanner mentioned I, don't know if we would ever add that other overload, because you could just write a plus then go about your business.

B

There's no need for us to have an API that represents ad.

A

Yeah agree with that, but.

G

A

A

A

A

A

Is also sauce overload problem that are we going on.

J

J

J

You should have said that only canadensis, this.

J

It's impossible, it doesn't exist.

A

A

You all right another one.

K

A lot of color expert but I.

J

Kind of in API.

K

J

Think this is the one that I know the community. Member of that HSL.

I

Is very frequently used, but it's it's never really a good.

H

Explain about that not.

I

So so, as the hue changes, the actual luminance of the color result changes too. So the Q plus luminosity Lumosity here, like it's, it's kind of a lie. If you implement a color wheel, you'll see it like they've, dark and darken I mean it's frequently used. I think it could be good, but representing some other color spaces might be, but.

D

Is that, because of the way our colors are represented, that they were not precise enough to not have the cue, so the velocity change, or is it just a fundamental thing of the color space yeah.

I

It's a four-hour they just.

G

I see, but this is an operation that most color API is end up, providing yeah. We should likely be proud of it as well. So I would say, though I would rather see this potentially be alpha, equals write, white dot, max value.

I

In the midst of the Myrtle's move, why is alpha.

B

So right now on color RGB and a are all bite like right now. Color is a 32-bit structure and I can't imagine us changing life. There is a new separate perusal.

G

Asking for us to represent to sport, other expectant colors show.

D

That will be different time, but that would not change this one, so I mean Oh like what hands up happening. Basically, so when you say.

D

Oh yeah, the audio again the human situation of low-velocity out of this color there's.

B

Already a get you get saturation on an API who's, not getting an awesome.

D

Interesting we're already halfway there yeah, we.

B

Have we can literally have api's, forget you and get saturation ogive right now? So, yes, we call a pregnant snot move, it I say of course, but yes, we. We have api's to get all three folks and they have.

D

Returned floats, they do one decades. It seems like a gentleman goodness, because you're already like weird, unfortunately called it- is the standard turn to mow city or likeness standard terms.

H

D

I

That's why it's HS, also HSL, is different from HSB, brightness and luminosity would technically be do.

I

What are we doing.

B

We are, we are returning, I, believe the average of our GMB and.

D

I think there's an emoji for what the current behavior is. It's like this. Oh no.

B

We're not returning the average we're doing I, don't know what we're doing.

G

B

Horse but I can't imagine it changed this yeah. It.

I

Looks like we're just returning yeah I'm, not really sure what this we're.

B

Returning something that resembles an average.

B

A

A good averages Peninsula and.

B

The maximum component- yes, we we take the two extremes: we throw away the middle one and then we have which the territory it has.

D

A name for them, I'm, not actually joking I, think I saw I, think there's actually a term for that. So very confident.

B

So again, if there is a proper way to compute the SAP or to compute the luminosity, and they turn out like, we actually add that as an API is all oh yeah. If.

I

Or we should call this from HSB, it depends on what brightness actually means. I can't remember the algorithm up and tune and like.

B

Maybe this algorithm is correct: I don't know enough about for spaces I.

B

Know they're not taking into account that the human eye perceives the different colors brightness is different.

D

Do you say HSB is not the same as Asia so.

I

Pick it out a LSH, you other space.

B

Simple, if we want a different paolo time.

B

Actually, just a list of floats where the float is the wavelength of a constituent of the color.

D

A

They make sure let's in stone, is.

I

Anyone commented with this don't search already, no.

D

I just searched for blinds I couldn't find anything with that. I didn't search for HSB, Oh No, but looking at the sample code like if you look down it makes you fall down here somewhere, except like this here. If you search for brightness like it seems that people use the brightness okay, which seems to suggest what you said earlier, it's like they really mean HSB, because I just found another Tacoma flu answer. But someone said the same thing.

D

You said: there's HSL and then there's HSB, which is the same as HSV, which is basically brightness or I. Think he's also HSI, which is something similar apparently.

A

Hey sounds like.

G

Yeah we're just doing the standing loss of the equation and call the brightness different.

B

So we call it leave it off or we call it brightness, but it's really live at all. So that's.

G

What it looks like I'm, comparing to the known good HSL implementation from the DirectX math library, okay, but that's also vectorize codes. It's a bit hard to compare wonder one.

B

Does it we had said that there was a proposal for like an extended color type or something like that like what it just makes sense to make all of these changes, and only that type I think that's what, as is.

G

Maybe but there's a the extended color type you might want to be part of a bike, bigger library, maybe because there's been people wanting something: that's not system drawing for a while yeah. um Since this drawing isn't consistent cross-platform and it's a GDI base plus there's the issue of some people want color to be for float. Some people wanted to be four bytes there's you know if you look at both inter DirectX or any other API, there's well over 200 formats that you can pick from you know.

G

So really we might need a color color RGB of T and color, our DBA of T or color, three and four of T, where you can then define what those are in front or operational.

I

You know I think just looking at in terms of scoping on color. It makes sense to have some parity here. I know this is a common operation.

I

E

B

D

All right, I just got you back. I just wrote down, basically what we just said and then let the person come back with opinions or friend, Jimmy Bob Squarepants, because he wants image processing whether he has an opinion on that believe that, as it is all right, so no, where is the window again.

K

Then culture in from their culture for a new culture to actually accept three different cultures. Oh boy.

I

B

Does that mean you can pass yeah? You can have like two known root culture, anything and sometimes it will just fabric here, a culture out of NL based.

E

On the words, no, it's nothing. We understand they Wendy's team started to support any any culture, name that it's it's confirmed to the what they call it bcp 47, and so it doesn't mean, like you know, windows carry any data for this culture, but when just try to fake some data for it, okay, so as yes proposal, I think they want to make sure that, when the request some specific culture, do you want to ensure that the windows or the operating system has people.

I

Would this be a breaking change, or would we want like a get exact.

E

It's now we'll return, exact culture if it is absorbed waiting system healthy, have a data for this cash, real data, yeah and therefore 0. If it is today, we do not show us when we speak.

B

For instance, I'm calling their culture and filling just passing them two characters, two characters. No, and it's like I, don't know what you mean so just as a moment yeah. The reason that this is interesting also is because a lot of web servers cache this information.

B

So, for example, like looking up culture, information was a bit expensive, so a lot of web servers will say well, let's look at the be requested culture. That's coming in in the request, header and I'm going to you know, look at the cache that I've been building up and if I don't see it I'm going to call get culture and phone then populate the cache.

B

Generally speaking, they rely on vegan culture info to throw an exception, but this would allow somebody to just like blow up the cache by populating it with form elements, but not for billion, but an enormous number of elements. So.

D

Basically, the idea is that if you call gun culture in when you see a predefined, only true, then it will just return now, if you can't find it is that the idea visible so anything.

I

Yeah this is, this, doesn't seem like it would blow up the cache, but to what you're saying is asking me yeah.

B

Basically, people who would be catching these need a way to say.

F

B

Need a way to tell you gave me back something we faked, instead of something you actually knew about. Just 250 or 26 to the power of four is around five hundred thousand, so catching five hundred thousand things is obviously a non-starter. Can you scroll down? What is the proposed API here.

K

E

Doesn't nicely, though interesting thing is an invitation here, so I'm not sure, but maybe in Windows we can detectives is that the culture is really vehicle.

B

It looks like a very particular code pages given back for the fabricated culture, so we can check out my windows in one.

E

Was they they create like the English name? For it we put something I, don't remember the name, but they put something like you know, feeling like this is not really difficult right.

B

Well, there's the there's the locale idea, 4,096 fortified by also.

E

Yeah, but what we have really some some really many cultures with that ID yeah. So the question is in Linux.

E

Because also I see you have something called the alias, I'm sure which is like you know, it's just alias to some other culture. That's all. Is this really culture or not.

I

And sexually I think it's not a bad idea, but because we would need to see if it can be implemented.

E

I think in a poll abortion is the team.

I

As a predefined only parameter what would be like different.

D

Also, I don't like: why would you want this to throw an exception? Rosen.

E

Is no exception.

D

I get that, but we nurses have been overloaded like I get culture, my name or something yeah.

E

D

Would be much better? Actually, it.

E

D

A blue back to say.

E

Like you know, I just.

D

Kept exception to validate like.

E

Romanticism, tell me I mean, because sometimes we need to create culture. Doesn't.

D

I

F

I

Get culture already.

D

I've been peeping on Twitter, he said check it check your chat log.

G

Hold up, some have been in YouTube chat, pulled up additional references for where we're using pinned already in the surface area like, for example, memory marshal, create from pinned array and other places where we're using app integrate, yeah, basically permanently pinned objects. So.

D

It's effective is I would probably call this wool steady boy, so I get predefined.

D

No I'll try it.

E

Under we use it in.

B

General and anyway, it's not for this specific scenario, so the behavior of troika culture would be that we don't allow fabricating cultures under the covers. No, we.

E

Still have this prompt encompassing so try and get culture in formally thickness the culture name, and this bullion feeling like I.

I

Actually don't know if I like that, because, typically you would expect the opposite have tried to throw we're returning a dummy culture and if you don't have that symmetry that you get all over there well.

B

It could still fail if you pass on something: that's not a legal right.

G

But I think I think the difference is normally. We would do it such that the non-driver gene is effectively if try returns false yeah. That's.

F

G

How we implement it that wouldn't be the case here, which may be confusing to users yeah.

D

That's what I said right? We defined culture yeah can't.

I

Cause you always create alignment, so we have the same time that person it makes sense. I get the exact culture, something like that. Like.

D

A culturally exactly it would be the, but it seems like I understand.

E

Complex onion, because I'm hoping that this is your V I we can use it and even in minute scenario it doesn't mean, like somebody really leaves capture or not.

D

So maybe I had recently said, but I thought. The idea of the API was that the name you pass in must be in some predefined table and if it's not, he returned fault. Is that not what they asked? Is this.

E

What, but today, as I mention about, like you know, I mean if you best name that it's not in the table at all for Windows, Azure or even my see you can create a culture for me and I'll give it to you right, and this is the maintainer. This is what's amazing,.

D

When subarrays my question is, there is two ways implement the API. One is I only check against the predefined entries or I, basically go down the other path and say well.

D

If you pass me a culture that didn't exist, I will add it to this to this table and when you say to I, get value I, basically check against the predefined things, and whatever else you created before what I was saying, try get culture will only ever use of the things that were basically oblate like how well-known effectively no matter what get culture calls you had before.

D

Would you expect to say, get culture, food get culture bar and then I can say, try get culture bar and then would return true, because you already registered it or would you expect it so.

E

Forget about registration, this is in.

D

E

I mean we don't we just call the OS telling like you know, I have this caption name. Please give me a culture data for it and the windows today you to give you data anyway, right and at that time we will check the data to see if it is fake or not, if 200 this case so.

D

I guess what I'm asking is when you do the ignoring APR, if you say, get culture food today, yes and I could get culture food again. It's only a few is not a predefined culture. Do I get two different instances. I do I get the same answers on the same. If you say get.

E

Cash on hand for yeah I will get the same instance. If you say new culture info, you will get a new end and also there is another API which is called the clear cache which is can't leave the cache and end of you know. If you will like get culture info, you will get instance, here's the cash get culture, and so you get a noise.

D

So then the question is: try get culture.

D

Would you expect it to return true when I went when I provide a name that improvement that that to get culture name, what we're trying to predefined instance for rather than creating a new one, so.

E

If we have should I get culture info, it should return true. If, okay, if we don't, if we just passing the culture in falling, it should return true, if you can create this culture, it doesn't matter. If you say it's a real data or not for it.

E

This is my skin, but if we can add this parameter to that tribe, yet culture info, which is sailing like we defined only oh the tribe, yet culture will take the cultural name and another parameter is called we defined only at that time we returned true only if we can get a culture that have read it so when.

B

Predefined, only your defining to mean exactly right. So if I, if I try getting the culture info for fr, US, for instance like there is, there is no such culture to find, but the culture, that's fabricated will have aspects of French as its base because enzmann inherits the parent. So is there not still utility and being able to fabricate something like that out of it? Oh yeah windows can.

E

Procure kids, that, yes- and this is what I'm I was telling about like indentation wise right- we need to figure out about like it's a way to understand which is refabricated over.

B

Ivory, so this this is, it sounds almost like. If you want to get down to it, we actually need an enum, not a goal for that last parameter where that enum is give me only an exact matter, if you can find it or allow fabricating something that goes up to the correct parent or just fabricate something out of thin air.

B

So, for example, if you, if you passed in FR us and FR us, doesn't exist, but FR does, then you could return just FR by itself and you would get at least the right parent culture information if.

E

I need best, if our less you guys. Yes, we don't have a father. Yes, you know.

B

But it isn't that it well I guess I was argument. It sounds like there's room for a more general get culture info API that takes an enum or how things where the enum.

E

B

That level in areas where there.

E

B

E

Scenario specifically, we will seem like you know. If one does carry this real cultural data, a complete cultural data, only I mean this is the scenario.

B

You can about okay and then, if the caller wants to allow going up to the parent and the caller just rip stuff off on the room right and figures out.

D

So, okay, so then I guess what my question is when they say predefined only. What do they mean by that? Do they mean like in exact.

F

D

Fr us would be not predefined, because there's no such thing. Yes, what.

E

About another essaouira, it won't talk to you about. We have get cultures which is in weight or cultures on the system. This, in rate, only is ability, I mean the cultures that have to be data and windows yeah. It will not have a and, like fr, cos right, yeah right. This means that, if all that she was is not really the documentation.

I

For this, it will only return a culture that exists right if this other mother yeah.

B

The behavior of this today is that when they fabricate a culture, they think of a culture is like a nested hierarchy. Basically, but if you can't find FR jack us they'll, fabricate FR us but they'll, give it the behaviors of just regular a farm, so it would have the it would have things like the the numeric representation and other textual elements of french, as opposed to the invariant culture, which is english because well FR us doesn't exist, but FR does so.

B

The thing that we fabricate will be based off of that that's the current behavior as far as I can tell, and so.

D

There is basically get culture info or new culture info. They have the same behavior except the new content, but will always fit in you, instant know what I want right right, and so is it unfair to say like the overload that he really wants or she really wants? Is they suggest a way to confine the name to this build in list? What.

G

Happens when a user uses the system API stored, register their own custom culture at the OS level, and how does this work on UNIX so on Windows that we could enumerate it properly? So we would have says.

G

Part because that's realistically no different from the user going and creating a custom culture on the.net side in passing, I.

D

Would say if you, if you go through the trouble of register a system-wide if.

B

You're, a minister on this, your.

D

Application is going on right, I. Think at that point you consider the build anything because somebody went out of their way to make it though thing I, don't.

E

Know about ICU yeah. There is a ways that you can extend.

E

What should I see you can?

E

Actually people can build their own culture, more culture, so.

B

What we nothing will prevent them. So what will be circling around? Try get culture with a pool or try get exactly.

D

Yeah, meaning it to me, it's like to me: it's the! If, because there are the apnea Department, that would be called through the pundits leavened with try3 to find culture, because it seems like that's what they asked us for, but I also have no trouble. We're saying, try get culture info and then say, string name, comma will predefined only and then our culture, anyone I, can do either one right. So if.

B

You follow reflection, which has some parameters to take a rule. Some of those ap, some of those parameter names are things like the exact match, one light so that might influence our decision here. Yeah.

D

It's me I, don't know we predefined our exact match. Our I think predefined makes more sense with any exact matches. They might response more like parsing or something, but.

I

D

I

Know so something Tanner said: there's predefined mean what was built into Windows or what.

F

I

Defined after predefined list, so it doesn't quite and.

D

You can call it system defined.

B

By a predefined in this case would mean anything that you can query the OS about yeah or your information.

G

Is predefined and find the wrong thing if they? If, if the program starts, then they call an API which creates the culture and then all this yeah.

I

That's exactly I feel, like exact, is a little more meaningful to what we want it to be, whereas predefined his questions, I, don't know who will ask those questions? What exactly.

I

Does its follow.

E

Yeah, exactly will not be that obvious for anybody reading. Thank you.

B

Okay, so so not try get predefined, but try get culture taking predefined as a ball. We.

E

Can say like 15.

E

D

Maybe that's what I currently just picked up, but I have no like the one thing that I think would be bad. Is this overload right? Let's move all this because it kind of implies it's the it's the counterpart to get culture. This should only return folder where culture currently flows, yeah.

F

D

Is not what we want right, so you need to do either a different name or some additional parameter that allows you to say. I really only want some subset of that. The.

B

The reason that I instead exact previously, is that, like, if I pass in Fr us I, don't want you to get me on like I, want you to give me a 4. She was warned of it. Yeah.

G

Iii think exact is probably thing that best explains I want something that exactly matches name or nothing. So.

D

Then my other question is what happens with like things like get culture when you replicate fr us before you do that, and then you call this a dad. You expect it to return well.

B

Cuz, it's not exact.

D

But you cook to the thing: a : get culture yeah, let's get it back.

I

Is it a fabrication you get when you call.

D

B

I

The first attempt the second.

D

One it returns they exist and since apparently it.

B

Does yes, but that's because it keeps an internal cache of everything that it's created or seen, but.

D

That's just a caching behavior. It's not guarantee it's probably a guarantee, but well.

G

I, don't think this is some would register it as a culture.

B

G

So you should still be able to the tech that FRUs is a fabricated.

B

J

Yeah, if you pass.

B

In true here, I wouldn't expect this API to return true for anything that had fabricated or anything that it didn't actually get from the oh yeah.

D

Basically, the predefined really only applies to the US, so it doesn't apply to what don't net things is in its cache. So that's why yeah to me exact, match or predefined. These are them convey that right, like that's. Why yeah.

F

D

Define make the most sense to me, but I also don't like acting away the way, because it seems like we need to bad Doc's. Anybody explain what what this thing does and so I definitely might as well. We.

B

Already call it today installed cultures.

B

We actually have flags on culture info today that tell you where the culture came from and those flags are installed: win32, culture's framework cultures, neutral cultures- oh my god,.

G

When I came here, Muslims no significant right looks like the power show back in October of last year. Added reading parameters to their get culture command lit, one is name, the second is no user overrides and the third is what's available, so they call their they there's effectively. No use your override, no user.

E

Override this different things.

D

Like 600 different flavors of the same thing, user.

E

Override is like the user can go and properties. Like you know, current control we can go to insane like I. Don't want, like that's my point. I needed calm right, I thought.

J

E

It's not it's immutable, but the user override is the system. If the user I mean it gets torn invalid. Just so, when we call the the OS to get the data, you can tell you the data. Do you want to? Those are all right.

E

I

Haven't eaten already tell you what.

B

So so, just just to get the behavior of the way, so the behavior would be if that second parameter is true. The culture that you get back is going to be exactly that or nothing. If the parameter that you house is false, the culture you get back could be the closest approximation. The system could come up with right or nothing which is the same thing as good country today, okay yeah, but if you pass in false there's no guarantee as to what you get back. You.

G

B

Get back nothing or you could get back and approximate.

D

It do we even have anything when you say name, comma walls, it's the same behavior of get culture except then get culture would have thrown. This poly will transform yeah.

B

What this method throw, if name, is malformed.

G

I think pastry date here in San guidelines, yes, means.

B

What meant for me that malformed is in like XYZ ABC, like that this is a PCP for seven standard.

E

Get culture only.

D

E

These, like you know, if you say, I'm saying, die whatever question mark.

B

Yeah yeah, if you, if you pass that in the get culture info today like it, will respect them because it's like this is completely meaningless. Right, like I, can't even begin to parse upside I. Think people often use try methods.

I

Or parsing user group yeah, so I wouldn't want it. We just wanted to repulse yeah.

B

This is why I'm trying to get some of these, so this will also return false if the input is battle for yes,.

G

Right which I think I finally most Rowland's on guideline, as it does, because the design guidelines say that the pool should only represent one failure case, so should it represent either invalid input or it sure to validate represent. We couldn't find an exact match. Yes, which would mean we would need to call this. Try get exact and not having boil parameter to respect the framework, decide where it is Jeremy.

E

Nothing I dare to have like 20 eyes in the set of one like one. Much is trying get culture takes names and and the other one we can't call it like try it.

D

Yeah I mean honestly like you: how common is parsing culture names so do we care at all and maybe because it could.

B

So, like you, you could have this method, throw for all I care. It doesn't have to be a tripod like this is not going to exist in any other part. So.

D

You actually would suggest even go a step further I'm just making this get cultural, or would you suggest to only for for any other names.

D

Sounded like the way the what the caller was asking is basically doing this in a case where you have a lot of those instances right.

E

But yes, that's what he suggested because like we have here, is our load for the get action but but I like the idea to have tried, because we don't have any ways that you try to create.

B

Culture and then you would want probably an return code instead of a pool to tell you why I failed, it failed either, because then we couldn't parse a name operation. It.

E

Doesn't matter I mean because we don't know either I mean? How do you know that you know we don't part of the game is about well for me, we can check it. Do you want to put the name of them? I mean.

B

We already do in certain cases. No, we don't not for this particular. If you either, we have what you can the framework. The positive, no.

E

We don't have for.

K

B

What yeah? Well it's fun, Union, it's a SP, double asp.net full framework had this logic, because we had, we actually were relying on this API, throwing a while back like in previous versions of the operating system, and what we would do is we would actually like throw off stuff after the and keep trying it over and over and over again until we go to success, we had longer diversion album. Look.

E

This is a specific for innovative methods. This is yeah yeah.

B

It's right: okay,.

E

Sheriff ffff Yahtzee well.

G

If we do have a try, API that returns pool, then the framework design guidelines also say we need the non drivers and this let's take it, that that takes the same. That does throw.

D

In which case, let's just started with the long driver yeah just.

K

Even anybody ever care.

D

Yeah yeah, it seems like a good starting point. So then the only question is: how do we call the parameter system to find only exact match? I take her free to fight all my work, something whatever I.

G

Prefer exact personally but I mean probably, and everyone uses API myself so I do have a strong opinion. Well.

D

Maybe need intrinsics for this one.

G

D

Or arm arena next time, I talk to the armful.

G

Yeah an order, sport, globalization better. We need you to build a translator into all of your CPU chips with culture, support.

D

I would throw or any valid names.

B

There's near some weirdness around and.

E

Hotter as well simplified.

B

In traditional chinese, yes,.

E

Windows, one was invented, culture is called like zhtw CH, TN, yes, and, and they make it the parent, for it is called magazine at CHT. Yes, yes, yes, I'm, okay, I see that watching.

B

E

Of those are not in standing in line right all right, oh yeah, so when you run into anything like you will get the edge hands. Yes, yeah.

B

That's that's what I just said: CH hands SIA right: okay, which is a parent to does eh ta, JA, ok, yeah, yeah, ch,.

E

Ch as simplified CST is Turkish government, oh yeah, that's all confusing. Okay, all.

K

K

Odorless remove range.

G

Wasn't this the one that we said? Oh no, this is sorcerer yeah I thought this was the one where we said we could handle it yeah. They should take a range.

D

And all the one who do you think I was moved yeah.

G

B

Move yeah move widow said expose the spell I mean.

D

Remove range seems fine, and we have other AP is already approved. Did we move range seems reasonable.

E

Collinson up you're moving to ensure.

B

It is in UI, applications, I, don't know about and list of T specifically, this is a.

D

I mean one of those things that you can do it. The polymers is the person point out here: it's not quite as nice to do yeah, because, basically you would have to move over that and then I don't exactly know how sort of this is implemented.

B

But so imagine imagine you have a UI based application and you have a list box of like 300 items and then you select this block of fifties and then facility, like that's where they were eight days away. So how.

G

Does this more given that sort of list uses an arbitrary T as its object, it doesn't do an indexing account. It's more like a dictionary yeah.

I

Is there an indexer there's.

G

Index of tiki yeah, but you can't do it as proposed, which is int, urge or int index count, because there's like it's a it's. Basically, you.

D

Know they also have get by index and they have set by index where you, basically whatever, is in place.

A

I

Looking at the map or.

D

The district T I'm, looking at the baton.

G

You're, looking at I'm, not looking at sorted list, which is TT p-value, what it's supposed yeah.

D

That's looking at yeah, that's what.

G

You are what I don't see how this could be implemented as though our new brain.

G

The closest they have is remove at which takes an index. um You.

J

See, in which case it's just doing a four loop. Oh no.

D

She said: here's the confusion he's literally talking about mr. Fergus of T he's looking. He started with the non-generic.

K

G

He had both peacefully.

D

Baby well on the non-generic. Well, it seems finer because that one egg has index based operations already, but even sorted list of t communitary has a remove ad.

G

Right but then you're just we're providing a helper function for something. That's probably not ideal for the data structure.

D

Well, it's implemented, is it actually it's.

B

Actually, it's actually equivalent to listening to you under the covers, like it's actually just an array, and when you remove at an index, they just remove everything else up on in.

D

Which case you move range is actually pretty efficient, yeah, because.

B

Get by index is also a little one. It's just go into the array on your right right, but if we ever change this in the future, didn't we say we're not changing any of these I.

G

Don't think sorted list has the same requirements as listed there. Yeah.

D

On the other hand, though, like things like sorted list, I mean they're, clearly helpers or particular scenarios where, like this one, it seems I.

I

Would feel way more comfortable if we had some concept of iterators.

G

Well, it's called I knew more what they say.

G

We just don't have a reverse one, not.

I

In terms of C++ iterators, Bruce Lewis.

D

Place our head: are there a nice? Are there simpler, you.

B

Know the reason that we don't have an indexer on the tapes and into indexes, probably because the indexes keep changing every single time, something about it. So it's it makes sense that we would never really want to expose that. Well,.

G

And it it looks like all of the base index of stuff that we had exposed yeah.

G

Oh, it looks like we kind of removed a lot of that from the generic version. Yeah.

A

D

G

We have index of key and index of value, I. Think.

D

Part of the problem is also what happens if you say it's for the list of income or something else. What.

B

You would actually want to do is instead of taking indexing count. You would actually want to take start with this item and end with this item like remove everything between them, because that's naturally how you access stuff inside of a sorted list like item not layin ducks, but.

D

You can tell we do this by just index of key right, yeah.

G

But how do you know that, then it's correct like what you're removing and how do you know that removing something is it going to then reorder the list.

B

And moving will absolutely reorder list has to right.

D

Yeah, you mean it logically like when you, when you talk about arranged like if you look at the DUI scenario, it's kind of what Levi said right. You know how many items you want to remove I think going over a keys, but not natural.

G

Well right, but so because this works on on arbitrary objects and those objects implement I, comparable and I comparables used for sorting something which has a dynamic sort, algorithm.

G

We're like let's say if you have under ten elements, it's sorted this way and if you've got over ten elements, it's sorted another way because it's more efficient, then if you do a range where you're saying move from like nine through eleven you remove index nine. That brings you down to ten elements, causes the list to be resorted, and now you try and remove indexes ten and eleven and they're no longer the things that you meant to remove.

G

That's something we have to consider for the a comparable.

K

These were bullies, but I.

G

Mean we have custom sorting logic like that for string where, based on the number of elements, there are a number of strings in a dictionary, we change our sorting different for a hashing algorithm.

G

So it's not completely bogus it's like a real-world case, be it.

D

Is bogus in the sense that the only party that can reasonably do that is the part of the only data structure like as soon as you're positive, so try with them into an arbitrary data structure. It's pretty right.

B

But what was this sort thing? Do you stash in consign them? No, this is just comparable, not equality.

G

Well, that that's an implementation detail on how dotnet is implementing it, not necessarily for the contract that sorted list exposes.

B

Yes, oh the contract that sort of most exposes is basically a sort of dictionary. You can only have one value associated with any given key, so he can't exist twice in the same list.

G

And it was also called out that remove and move need to be careful with individual operations to be compatible with observables.

G

So, like you can't necessarily say, remove and or you can't have a move range and do it optimize algorithm? If, because that might impact how how observation happens, but there is an observation: that's loading this you get. These are virtuals. You can configure your observation on them.

A

D

With I, observable or I notify property change watching it, and how would you watch it? You must have to call the element. There is no modification, you can subscribe there. No records like how are you.

B

Sorted list has virtuals no.

D

I'm looking at it you're looking at depending on generic one, what the non-generic one, the non-generic one, nobody care, they're all a give a list and total list of all men that you extend them directly and for the generic once we splitted them. Our buddy said, like the core collection types are meant to be super cheap, there's, no survey observer observer ability- if you want that, you have to use the object model once the collection of T and T collection of T KT value, which are basically the ones that you are supposed to wrap.

D

So in this case there is no observation like it's the toy justit's as chief of data structures, you can make it to be so. In that sense, I would say it's a win, the spirit of what you want to do there, but.

I

Yeah, there's the differences and sort of dictionary that in excess this one by indexing as well as well.

G

The generic version doesn't allow you to access by anything. You know, okay,.

I

G

I

Here is wrong with it. Well,.

B

There is, there, isn't an indexer that takes incomes, but there is a gift by index. Yeah.

D

I never quitters the difference. Re.

B

G

Generic version exposes index of key and index of value and they kept and they could be different, correct.

B

There is yes, the generic one has no get by index, which means, if you want to remove a range, you actually have to take the starting and ending keys. That would be the only way to do it and.

G

Then you'd have to anyway to enumerate through the keys, which might not be possible, which.

B

Internally, we actually have that way like we actually have the ability to do that. Control. No.

G

No I mean like if you have an arbitrary T key, where it's string. For example, not. You can't enumerate from start key to end key.

B

Well, we can, because we have a list of keys internally, like we literally have a property which is I. Listen to you, oh right, but then you.

G

Then you fall into the mutating. The list could change the internal sorting of the list. Oh yeah.

B

Chase numerating.

G

Is now invalid because integration rule yes.

B

So what we would do is we would our remove range method would look at the list. I mean it's basically a binary search, but we look at the list. We get the index of the we get the index of the last element in it. We nuke that part of a listen. Then we move everything back out right things, expensive and mom drew you. It is expensive, but it's cheaper than calling remove end times in a row.

K

Cheaper, though,.

G

So event, instead of events where and how often does it happen, that this becomes.

D

Or that yeah I mean like how big is hole? Did this I mean in practice.

D

I mean yeah Owen, squared where n is 16.

B

I think I still think the common case for this is going to be, and you like those applications so you're, looking at like a list of files or a list of images or something which again goes back to and you'll have a few hundred facts. You know you're speaking.

D

Yeah yeah and I mean I'm inclined to say no at this point. I thought the dictionary by the way, the difference between so the distance or the dictionary is just so. The dictionary is a tree under the covers. It's the only difference. That's the same. Maybe I had same API same same semantics, which is kind of funny.

D

Then I guess our conclusion is not motivating enough to think through all the education.

D

D

There are some issues around Oh.

D

B

I want to add a remove. Ranger overload the tips on range now.

G

Was of their proposal to improve the language support for index and range, though that's.

B

The index race yeah- that's already been done only very nice yeah, but it would be cool if I could do like remove range. Three dog, you know carrot one or something even.

G

B

Through the first marker, existing methods, and then the compiler just there's like indexing count, I know what to pass in there.

G

You so github supports multi line comments now, but now you can finally select like three lines and say this comment apply stuff: okay,.

D

What's funny is the content API always had the support for it, because they're busy give you back individual items.

G

I wonder if this works first suggestion replacement now his previous.

K

Age and so now can.

G

I say: take these four lines and replace it with this new text. For those cases I don't.

D

Know the answer to that it was extra work, so no I.

B

E

Render threads.

B

When you have like overlapping in lines same way, that when word reviews do it and code flavor, these feel like they just have of the comments or one underneath another, because.

G

You can already do you can start multiple comment, threads for the same line already today, so I. Imagine that they're just going to order than based on starting line yeah, because.

J

G

Tools kind of like this one.

B

So if you wanted to improve remove range, if only we had, you know someone who's infiltrated the language team um that.

G

Would be Steven? No, not me he's on out the end. Yeah.

B

But is he still on the fly.

I

I read they listening so.

G

He's purposefully staying quiet.

G

D

D

All right, I think we're done for today.

K

Produce them all for us, we did go through the regional market.