.Net Foundation Design Reviews, 16 Jul 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

00:-33:-56 - Approved: Add Encode(ReadOnlySpan(char)) method to TextEncoder for performance https://github.com/dotnet/corefx/issues/39523
00:10:49 - Approved: Deprecate PipeWriter.OnReaderCompleted https://github.com/dotnet/corefx/issues/38362
00:32:15 - Approved: (System.Numerics) Cross Product for Vector2 and Vector4 https://github.com/dotnet/corefx/issues/35434#issuecomment-511933825
01:28:39 - Rejected: Add StringBuilder(Char) https://github.com/dotnet/corefx/issues/17745#issuecomment-511942559

A

A

Sweet, so no, we are live again, so this is the first one that I think he wants to review. So being was a great overview ordered what this is.

B

Okay, yeah yeah Levi is probably actually the best contact for the implementation and um overall shape of this API. But basically, when we're implementing custom, encoders, there's like two code paths, one one for utf-8 and one for utf-16 and I did the utf-8 one on last week and then I got to the utf-16 side and I call that the API that I was using when I was writing.

B

The code to support utf-8 did not exist for utf-16 um for encoding and there's a there's, an a' code for utf-16, but it takes a string and returns a string and we can um call that efficiently because we just have a span of car and we don't want to do any copies of base or you know any kind of Alex. So when I was talking to Levi yesterday, he just suggested that we pretty an issue for this and I, which I did and I copied. The syntax from code utf-8 and instead of byte is car.

B

I don't know if this is gonna be overkill for some of the scenarios in the sense of like this return code, for example, on the the utf-16 version, does not have a error code, it just returns a string but but I'll. Let Levi comment down on whether this is the process.

C

I think I mean Steve actually had a really good overview of it. So there-there are today encode methods which take as input string refers. A string takes as input a char array, writes it to a times later and so on, but there are no methods that take a real loose man of charm and, as Steve alluded to that's difficult for the JSON staff to consume, because the Jason stack is working with read-only span of char everywhere. So this is adding read-only span of char Overland's to the encode method. The signatures it's using spans.

C

The signature is using an operation status pattern and it's simply mirroring the operation status pattern that already exists for the utf-8 overloading that's already checked out.

C

Yeah change flight to chart and remove utf-8 from the parameter names.

A

Yes, so the so the second one makes sense.

C

So the first one is not actually proposed, then the first one's already there and the second it's just showing that we're proposing the same pattern just with different types.

C

So charns said a bike.

C

If you pull up API job nut, you can see what the existing in code yeah.

A

I'm trying to look at this because I'm texting quota there is low in code utf-8. That's why I'm asking in.

C

Code utf-8 watch chapter or maybe api's have done. Letters have taken Bodeen that reflect in every day, I'd like to anymore. Do.

A

D

A

It's in 7/16 vincent's.

D

For stop that yeah.

A

B

You can't bait and encoding.

D

Tfh in the other ones, internal, so their internal, both our internal of them sourced on that, like the update now source cotton that is literally on head.

A

Did you guys make it public and then the chicken I mean yeah.

D

Checked in 25 days ago by you, you checked it in this internal.

E

uh It should have been, he did it yes on it's, not it.

B

Yeah, it would recently to be public just to support um and.

E

Cold yeah, it looks like you did it yesterday: okay, yes, that's what it was. Yes.

B

It was done a couple weeks ago or week ago.

F

Well, the API, maybe but not the way, not.

C

The part of its public see yeah. So the part where it's public is checked into your checked into core of facts. It called trucked in, like you said yesterday, so did source. Don't then maybe eyes have done that just art in.

F

A sense I mean they give it the efforts.

A

Was checking it today, we may have yeah.

A

It's the second one, I mean looking at the signature, makes sense. First, one is a bit odd because we have nothing else that is suffix with anything right now, but I miss the way.

G

C

Are two utf-8 methods all touched encoder by dog trapped in the other day, yeah.

G

I mean when I'm looking at it right now, I did that it's not. This is basically like is of yesterday, as of today, depending on get one. It's it's the first time we have.

A

Ones of suffixes, where I mean that's kind of the way we have it anyways. My wrestle.

C

And then yeah the when we brought this a while back through API review. It was either this or spit up an entirely new utf-8 encoder class yeah.

A

C

We thought was the worst off for me.

A

No I know and I. Could it's yeah? It's not a problem, it's just you look at it and it's like yeah I mean.

C

What else would they call that and.

A

So then I guess you said we call it source destination invites consumed by it's written and his final block yeah. Basically just remove the utf-8 for the parameter names.

A

The other works for me and the I don't know I'm still not 100% certain on the span. A PS is this: the parameter they have vice consumed, 10 bytes written. Yes, then it's final Dhokla.

C

I mean feel free to propose another pattern. No.

A

I'm just asking anyway, honestly I, look, it's been a.

C

So the reason it's not an API is that have that pattern, its operational status, ap yeah.

E

That's right: it's women, because you're with.

C

Operation status, you're working with potentially not the entire set of input, data yeah. So the the out parameters and the return value need to tell you how much progress was made so that you can appropriately slice or can count on the next call.

C

If we had api's that just operated with string as the existing overloads do like you, don't have to worry about any of that. It's just oak string and string out done excellent, but you are taking the hit of the allocations are, which is intentional basics before I'm, going to get faded, name's, poorly, but course destination. Then.

C

It's concerns, which course and expert his final block, a vulture and of operations. How to say too early, yeah I think we haven't been. We haven't been very consistent with the parameter name in here assume then rittany they're consumed in by threatening okay at least we got out.

H

Standardizing on destination is the name of the thing writing.

C

And sources feeding.

C

A

A

What do y'all do anything else, and this is what it is.

C

And this is, it forget, modified while it was on the screen. Yes,.

E

I didn't modify it, I updated you. What did you modify you model.

B

E

Parameter names I get in I did.

C

Am I done so keep in mind, that is an existing chapter. Api. Are you suggesting that we modify that on the existing tract in API as well.

A

D

Thought we had previously said that, for the ones that are read-only span a bite, we wanted to be explicit that they were UTF a input. What.

A

It sort of answer like basically when you look at this stuff here, so it may been asked whether be provable beauty of 8 from the parameter names. Everybody seems to not that's why I didn't know I.

C

Don't remember that question.

C

If you ask you, did you ask it in a YouTube channel? No one can answer no.

A

I asked it blow, it I think.

A

C

Thought you asked, you know: are we being consistent? That was a parameter ordering.

A

Because somebody said removing in the utf-8 suffix and I mean and and I asked from the parameters and then the answer was yes: oh.

C

I did that on the encode method, Oh.

E

The utf-16 one, okay, that makes more sense, yeah yeah,.

D

Because I had thought we previously said in the.

E

Work guys exposure.

D

Enormous magnifier, we wanted to be explicit in the name that it was a utf-8 input would remain read-only span of charting yeah.

I

A

Right so then the method name is encode. Then sources utf-8 source destination is utf-8 destination. Go back to the previous odd! Well,.

C

Destination we just be destination because this is an encoding operations. Oh wait! No we're doing characters together. So, yes, it would be two.

F

Legs looseness.

C

No utf-8 the name of the method, as in the public ref as of last night, is it significant I.

A

Do this, should we remove the utf-8 suffix in the.

C

Mid, oh yes, I made an edit this morning and around like 9:50 a.m. that edit was, as far as I can tell correct, / everyone's naming guidelines.

A

I can actually go back to the other.

C

Like this is heavy, we have a bunch of forts because we really should have added Chari. Please yeah really really really really really wish you ate at a party yeah.

F

H

There are enough videos on the Internet.

A

All right, since they can't go back soon, one more time. What's the method name, encode.

C

You to the first ones: encode utf-8, okay, you could also just copy it from what star Kentucky native.

A

Yeah it's my Google work that he's doing it this way. So then, is this the one we want yes, correct.

A

The second one is just encode source destination. Yes,.

C

Because the user chars.

A

Fine, alright anything else.

A

This is what it is all right. So then, next one this one.

F

C

C

Oh, so this it turns out, we might not actually need for three, oh, but just on the off-chance. We do need it. It would be good to just get at your review cuz it's here.

C

There might be cases and we're not aware of any, where developers really really hate some of the ways that we encode things like double quotes inside the encoding routines.

C

The default encoder is very opinionated in the sense that it's trying to prevent common misuses, and it will do things like it- will escape a double quote: character using /u syntax, instead of just slash double quote, and the reason for this is, if you happen to call this inside, of an HTML attribute or inside of a script block, we don't want malicious inputs to be able to escape that, and then XSS your website, so for developers who really find that particular encoding unpalatable and who really honest-to-god just want.

C

Like slash double quote, we can provide a factory method that creates an encoder that produces that relaxed output. um Now, to clarify, based on what I said earlier like we, we thought that that we were seeing feedback from developers regarding this in the 3o timeframe, but it turns out we weren't. We were accidentally completing it with another issue, so it this might not be necessary for three. Oh, so.

J

Let me, let me add: some corn perfect. Let me add some context to help motivate, because this because I talked to Nick a bit more and turns out in for setup, and the exit is an excellent issue there over escaping the plus and the forward. Slash causes dotnet test to fail and I had thought that the plus was coming from vset for encoded text only which I would thought we could fix separately, but turns out restrict the shop. After that you end up. Writing in the Access file is actually a regular string.

J

That has a basis for component and the whole chain is not a seat for encoded. Okay, so the board, and actually needing this to unblock dotnet test form working for three or so.

C

Called Coursera would call this.

J

Dependency model depends model disease to generate deaths, Jason father and assess product yeah.

C

So there there is a bug in the chase and unescape inverting, that's currently used by Coursera, and the fact that we're escaping of the plus character is exposing that bug. So this API would allow poor setup and others to.

B

Mm-Hmm there is another worker on Levi that that we discussed briefly very ugly, but you can derive from JavaScript encoder, mm-hmm and override some method to say, allow um plus whatever you want. It's not supported, but it is another option. I guess your emergency use only yeah.

C

It's it's not something I would really want to put into a shipping product, though, like it is the definition of a half workaround yeah.

B

We can't really stop people from doing it right, but.

C

I mean, but we can bite, we don't want to promote its usage by including it inside of our source code, though, as something they're able can look out and maybe think as best practice.

C

A

It's ounce, we think and we don't need the API at all, there's someone else know so.

C

So we do because it's course setup and others. We need to call in to us didn't.

A

Just didn't recommend Eric.

C

A

The basics before thing or whatever I was.

J

There is I taught that the data that was being outputted was basically four encoded purely yeah and I escaped escaping the plus in BCD for, but that doesn't fix it because they're actually writing strains not basically for.

A

And so what is the I still understand? What the problem is. Like I mean it's escaped, a nun escaped.

C

When you read it back right, so they know that's the problem, the particular unescape routine that they were using had a bug that couldn't read this, even though it was well-formed racing.

J

Yeah, so ever actually like index or out of range or something if you had escaped, like plus devil escape in your days and payload. So.

C

That's a bunch, so you you can escape a plus, for instance, as slash: u 0, 0, whatever right right but and and the Jason specification says this is totally valid. Every Ryan client jsonparser has to understand this.

C

However, the particular JSON parser that was being run over this had a bug where it couldn't understand this, and it was causing failures and in order to work around that, amongst other things like this, API would help with that by saying: hey, don't escape things like plus, like just let them go, as is that way, you're less likely to to encounter that bug in practice.

C

That bug is not an hour sack the the arena club is not yeah. Otherwise, we would just pass out.

C

Workarounds, for instance, oh you could do a string replace like you could replace / you could. You could use a fancy regex pattern because it would. It would have to be that and then replace / you blah blah blah with plus assuming there's an even number of slashes in front of it, but yeah it gets hairy.

A

Think that a whole enum is necessary. I'm.

H

Open to ideas, because.

A

I mean if, if we think there will be multiple behaviors and we should probably make it flag so that if we X, if because the thing they will probably be independent options, no, oh yeah that shouldn't gonna flowered, sorry, but then even then, like it seems a bit odd. That default, which is zero, would be the values for and whatever it is here.

H

Would be the thing we already had.

A

Does accepted everything foremost in terms, it seems on.

A

Deepa leading zeros, it's all flat basins, it seems weird zero means. Well, it's a particular combination of flags, but it's not the one I'm telling you like, like flags, usually have the default not default, but being none like a base. It means no bit set yeah.

C

A

Which case it would mean? Well, it's not it's 0, but it's not none. It's well.

C

Whatever the behavior is thanks.

A

C

This almost like um each one of these flags represents a different protection, that's being disabled by the default encoder by having the default value of 0. That means every protection is enabled by default and then you're off to now that individual protections on a flag pipeline basis, yeah.

A

Which would basically, logically, all one frame.

C

So you're, in your suggestion, the enum is which protections to turn on, and this is which protections to turn off no.

I

Because yeah this, because.

C

By default, they're all long I.

D

Don't think it's unusual to have an enum where the default may be enabled for some feature. No.

A

That's not on what's all is, if you have an API where, basically, we don't have flags even some legs, to my knowledge, where zero value is not named, none.

H

I hope we could name it. None I think it's a guideline well,.

C

But then you would say, you're escaping behavior is not, which sounds like nothing's escaped.

G

C

Like and then we could call it, let's take lessons, but why.

H

Would you want just to be a flags yeah.

C

What's the next thing, you're gonna have.

H

C

Next thing, the next thing to add, for instance, would be: do you want us to disable vestment mapping, we're always most by the bowl like that? So if you pass in a smart quote, for instance, did we escape that, even if smart quotes otherwise a lot of our game quarter right? But if I say just that and don't say, relax so again, pretend it's five. Does that make sense independent of the relaxed one, because I'm current default we're already already doing it?

C

So that's really only a modifier on relaxed, which means that we're not flags we're modes like we're. This is entirely. You have to name everything of what it does and maybe we'll end up with some ugly combinatoric names over the next 50 years, but you're not all not all combinations with the ballot, which is not a good candidate.

A

I mean honestly, like I mean you can just honestly just make them overloads with rules. I mean it's not great, but at the same time it's not so not horrible. Another thing that you could do.

C

Is you could have instead of having an enum, you could have a static property which says, give me an encoder which has all of this relaxed behavior just automatically set on how.

G

Do you support common etics than that yeah? You.

C

Static property, static property on javascript encoder that has a global that modifies other libraries, no.

H

C

A static, a static property that returns an instance of the encoder, it's a static property factory, so there's already a static property, javascript encoder default, which returns an instance of the encoder which has been instantiated with all default settings. Yeah.

A

That's what I thought you.

C

A

The problem is, you can't support combinations for that right, because, basically, we have to provide. We would aesthetic properties.

C

We would combinations we would say here is the behavior forever? Oh yeah, the relaxed reasons they think are which, honestly, if you consider it a mode of operation, will be totally fine. I.

A

Mean one other option is: this is just making over lots of creative wait. So we add one overload right now that has I, don't know, relax escaping equals false as a default and then, if you ever have to add one more be at another overload before the other one, and we move the default on this one and then that's the way you configure.

C

Security thing: if it's boolean overloads, we can't ever find them, so we can't audit them. We can't possess them. We can't write a tool, that's nice. This needs to be black. Like you know, they pay double attention for this. If it's a name, if it's a native alternative to create and like create relaxed escaping of text encoder settings or if it's the accelerated instance, then then those you can yeah. You can write tools to identify that you're using one of these and it needs the extra space. So.

B

C

Actually discuss that to an extent later in this issue, if, like whatever we name this thing, we actually have to have a similar rule for internal use, that Flags up and warm, because it's not going to STL approved on the time I. Actually.

J

Preferred the named enum prep yeah right now as proposed, because it's more scalable the bull option. If you have two of bulls now you get to compute of language out parameter means what well we're.

A

J

A

On saying for me, their argument will hold water if I can actually extend that in a minute same way and I'm, saying I, don't think that's true right now we characterize.

J

Mods right, like it mutually exclusive modes like default, is one mode and then it'll actually give me another and then, when we exchange it to another one, it will be a complete, separate mode. You.

C

Have to be to be perfectly honest, like I, don't really see relaxed being further modified because it's already relaxed.

C

Like to be absolutely perfectly honest,.

A

Sure but then I think in that case that would sadistically you differently. You know right.

C

What, which is fine, one of the other things I called out in this issue, is I, want a name that properly conveys your turning off protections. Make sure this is really what you intended.

I

So why not have it be a struct.

C

Of options, strucked of options is otherwise known, as in you know,.

D

Right but I mean it's more extensible. In that way, you can still have the bowls. You can differentiate them. You can have the logic that says that these combinations are allowed, and these other ones are I mean.

C

We that would also work I'm, not necessarily married to touch design. Point I mean another thing that was looking at earlier is: why is it's not modifying text encoder settings, because it's not text encoder settings is used by HTML, URL travel script and a few other homes. This is not honored by any of them.

C

That's also why it has relaxed chase in the statement of the name.

B

If you want to complete control over this, um which I don't think we do, you could cuz. You already have a way to specify the allow list right. um We kind of you kinda, but then we we overlaid them for bidding characters. On top of that, yes, which include things like the plus sign and there's no way you can see with that forbidden list is or modify it.

B

If there was a way to do that which I don't think Levi would like at least some of the characters like quote, but if there was then you could do whatever you want. Yeah.

C

So you can always see the for written list because our source codes, public modification, is I, don't think that's a path. They want to go down.

A

Yeah, what do people think then, about this affair?.

A

Basically, it's a bunch of bulls you can set. The default is none which means you do whatever the default. Behavior, let's be awestruck done, not Amina. Well, I, have you know it's a flag syndrome, then, what's.

C

The benefit of making this room, okay, so encoding options, none means just offed me into whatever the default thing is, are.

D

We ever going to have more than 30 to 90.

C

Yeah I mean like, but you know my question of being. The only other option that you have thought of ahead of time is the best bit mapping and what does it mean if you say best bit, mapping and not relaxed, because that effectively the same as done at which point? This is not a good flat.

G

I mean that is also true, although we do have flags we're all combinations of so.

J

I could understand: why do you want to operate four flags. Manik reflects what's the general guidance like making a nose flags versus regular, well.

A

So the current proposal is that these are independent choices, right.

H

A

That's the argument of being scalable right. If it's not independent options, then it's not scalable, because now that means, if you add another option, we basically know every enum value has to somehow represent a combination of other in your members which doesn't fly when you have more than two or three right.

J

But I thought I could regular enums are what make independent like modes. Yes,.

A

But that's basically equivalent of saying we don't have any number, just a different method names, which means you can only ever modify. One thing you cannot say: I want this option and this other option, but that's basically the thing that that enum is by your group of like an say well, I want the will, extend coding and I want this other thing that I want to equate yeah.

C

I I'm, having aside from that one about passeth mapping like I'm having a lot of trouble. Thinking of what other value, whatever the advocacy mt popa policy, which.

A

To me sounds more like at that point: it's an overload with a different name or just.

H

The different accelerator addicts like we have default and.

C

It also works relaxation escaping oh yeah yeah is, can you get directs Jason escaping with the custom map? Does that make sense so right now, you can do that because of the way that the overload is created.

C

If we have, if we have a static property, it would probably be instantiated with the entire BMP range so including Chinese, Japanese Cyrillic. It's all right so I mean, but my question is: would you does it ever make sense to say that you want relaxed and a custom range? If no, then it should just be an accelerated instance. I I, don't know why anyone would do that.

C

I mean you, you could I just don't know why anyone wouldn't practice so I think I'm in favor of doing an accelerated instance, and if we get more demands later, that require more complicated input, model moves and the chemical thing, but that we should drop this truck. We should drop that over to create. We should just have one instance that behaves differently than the rest and and that, like that's, an internal implementation detail, have a good thing. Yeah.

A

That's my plan. You like that so I'm saying to me: scalable only makes sense if you design the you know to be scalable, otherwise, so whatever. If we need to reconsider this in the future, then we can have that. You know at that time.

F

C

We figure out: what do you feel for what.

F

C

We have a third or fourth things, so we can understand what the modeling is, what the combinatoric look like it, or it makes sense of the text encoder settings. That's.

B

The understand.

C

Or the alternative is to create, is it a mode? Is it a struct? Is it Flags into the place, I think right now we just don't know enough about how this would grow or coming. If we don't think it really would grow, then we just be the one. Is this and.

B

The static is nice easier to use and people don't have to worry about cash in it. I think you know, can we use it across different yeah, very difficult, there's.

J

This one question is: how would that compose with custom escape encoders they're, created by the create method? So, if I want to this, not escape Chinese characters but I also wanna relax is just escaping. How would this chatty.

C

Probably work I mean that that's what that's what we were just talking about with Jeremy, it's, the the relaxed one allows not asking by the ball, but allows not last material not by default rest period. So.

J

It would be equivalent to D for JavaScript code or create Unicode ranges, dot all comma relaxation escaping yes,.

C

Gotcha we want something in the middle for that later, then somebody can propose it and we can figure out what to model it as. But but this is you know effectively. This is, is this actually minimal or is it minimal plus like it's? Not it's not minimal to everything. It's not minimal. It's there. It is still somewhat opinionated because it will still escape things that the Chasen spec says. Don't have to be escaped, for instance, see one control, characters, non-printable, whitespace and so on. Emoji.

J

So I like this propose I had a question for the original proposals. Why did you choose to only include this in imprinted on the create method that took taxon core and not the other one, because the.

C

Text encoder Serene's is itself instance, people from a set of Unicode ranges, so it was the more ah it was the more generic one. Okay, more general answer yeah, so you were asking Unicode ranges not all like. Why doesn't it allow those things, but you know I, said well.

H

C

You're escaping emoji yeah, but you also said all Unicode characters, yes, I other than c1 controls and and a few and a few other things like I said this. This is opinionated, so the the reason for that is the settings class passed into. It can't represent anything over 64 K because that's as high as it that map goes so anything beyond that, including emoji. It just says screw it I'm going to encode it. They shouldn't be common enough where anyone should really care about it.

C

All of the characters in common use throughout the world fit very cleanly into the basic multi level, one which the bitmap completely represent.

C

Notes inside there are. There are some very, very rare, one-off characters that are outside that they're, not in common use.

A

But what is insisting on if I really need hustling ranges and the next encoding? What would I do you.

C

Would your first auto for anyway yeah first off, you don't have that scenario? If you think you have that scenario, I invite you to explain why you.

A

Well me, the question is more: like I mean today, you also kind of just extend JavaScript the encoder right with an any use, custom ranges, but you would have to literally write everything is over them.

C

Yeah you, you have to override them, subclass all the abstract, but the poor override all the abstract methods which, by the way, our editor brows will never so good luck, finding them right. Yeah.

A

I'm just asking for day because it's always seems odd. If we make things not independent choices, just because we think probabilities low, because at some point somebody may it doesn't matter too many folks back area but yeah. If, if the whole curve design already, has that then I don't.

C

Care, so so this is what people are circling around. Does this name? Does this name properly suggests? You really don't want to call this unless you know what you're doing.

A

I mean I've to.

H

Be honest, I'm either scenario.

B

A

Somewhat busted, because it's like saying well, somebody cannot add two numbers. So we give you a different overload of subtract that has been overload that says Babu's behavior for compared yeah and that kind of seems like it's the same way like I mean we've, we have a correct jsonparser in reader, yeah, sorry, parser and generate or matter I guess, and somebody else can't meet it.

A

And so it is a bit weird to have an overload that says, make the stuff work with the other thing, which is that by definition, an open-ended set of like people can have an arbitrary box in it. Yeah.

C

Yes, it's not it's not arbitrary, bugs in the json parser that it's trying to defend against the the default. All of the default. Encoder is not just a javascript and HTML URL. Everything else protect against known, viable attacks, for which proof of concepts or proofs of concept exists right, like people the aggressively yeah. So the reason that the JavaScript encoder encodes plus by default is because there is an actual honest-to-god legitimate attack that you can pull against websites. You know it's.

A

Not know I get.

C

A

It yeah well I'm, saying is that if you write a JSON parser yeah and you can't handle in escaping and it's a fundamental bug in your parser right, yes, so that means if we we made a plus that is backslash you whatever right and then somebody else can't read that it seems a bit odd that we could over that says. Well, you know, that's just not overly aggressive. You know things because somebody might not be able to decode that.

C

That's not what I'm concerned about what I'm concerned about is people will just say this gives me prettier output. Therefore, it's and what I'm going to use oh I seriously, but still people will be focused, very myopic, Lee on it's pretty, rather than hey. We actually are trying to protect you against things that you probably haven't taken into account right, so I mean frankly, if the only consumer of this is coming to produce this even further. You know if the only consumer that we need for this for 300 is CLI.

C

It's not I thought going because of them because of a bug in X unit, and they can work around the bug in X unit and not us rewrite the file in a different format. So.

B

C

Would also use this when writing responses, but that's because asp.net controls the entire stack of the response, all the way up and down and can properly audit that stack and say we do not need the extra protections that the default behavior gift and they.

I

C

For what reason curve for perfectly developed poll, mainly readability, lockers, but.

A

Why would what innocence like? Why would they say? Oh we know that we can encode classified asbestos, you of.

C

Something because they do things like set the content, type charset pattern in the responses, so a lot of the attacks that can be levied against this just disappear. Yes,.

A

Yeah that API name- that's not so much that, but also don't know what we would call it before were insecure escaping, and it seems a bit.

C

Rio is feature complete 30 days ago. Yeah Rio is a super-duper feature complete in like 77 minutes yeah, and so you know for three. Oh my answer.

C

I would say we don't leave this and that they they can get a fixed version of X unit that fixes the bug yeah.

A

Was it assuming we just push that opinion? We have already I'm.

J

A bit Jeremy I, don't think we should discuss the timeline here, particularly because I don't want this to go to ship room and come back later anyway, because I mean you are being wok domestic that we can ship without. Without this feature,.

C

But my point is like there are questions that we're asking. If we're saying for three. Oh then there's either no or what's on the screen right now, probably, but we had other questions of future and combat or X, and what name should we give to this? And there are a lot of questions and I think the answer is, we should say hard firm. No, this is not a three Oh, a P I and a P I need work and saying kick it back to Levi from coming back in.

A

Won't come back in November just like.

C

At your lesson, can.

A

You explain what you mean. My mate comes back from ship room like what would be. Are you saying ship who might force us to ship an API that seems backwards? No.

J

What I'm saying is currently dotnet Tess is broken and we need a solution for that and also a spin and has been asking for this type of feature for months now. So I think we should ship. It is required for us to say look it's almost time to close on three or doesn't Ness some. My priorities will come back and say we need it. We need some solution, that's the same.

K

C

So, let's consider, let's consider CLI right now, so we can, as Steve alluded to, we can actually get a workaround for CLR like they can. They can put the hack into their code. It's fine. We would want a huge comment on the actual method itself. Saying like this is not recommended practice. So what would they do when I could move all the API M? What can me they would override JavaScript encoder and they would I'm certain they would subclass it.

C

They would override one very particular method that says we allow us to go through not escaped. Do we have an API said? Yes, that API is already there today? What is the.

H

C

I

Is it a pee I ain't.

F

J

C

Okay, sorry, except asking Jessica and quartered right yeah and then you were. There is an API on it, which is will encode it's an abstract, nothing. You would override it and you were passed in plus and then you just return false and if anything other than plus was returned, then you call into the bitmap directly I see.

A

Well, I looked just the JavaScript in code, I didn't see any Mel to base the house yeah, but.

C

That seems like a very reasonable workaround. It's the the problem was that workaround.

B

C

Can't call base you have to delegate to someone else, which means you have to wrap the default encoder I base, because it's abstract and there's no derived JavaScript encoder. The ref should say that there's an over right here, so you can call base to get to that.

C

C

The will and code method, because it's abstract when you call base what is there to call if you derive JavaScript encoder JavaScript encoder overrode will encode it will call the women code on JavaScript encoder, as long as the rest says, is a override. Then it is actually override my own code. It has to it's abstract. No, it doesn't wouldn't be. All of the factories create internal types. I was looking.

B

In the file, yes, I didn't.

C

Down the script encoder with John, the class is involved on.

E

The element go there, but.

C

The thing is become because of methods abstract. You can't call, based on you, have to delegate to someone else, which now means that hopefully, you've overridden every single method and you're delegating to someone else. If we introduce a new virtual method in the future, such as operation status in code of read-only span of char, then you now have potential security bugs inside of your encoder, because you've overwritten some methods and not others, which means you're delegating to some methods, among others.

C

That means that the better fix is JavaScript encoder should over I mean like this means. We don't have a good type hierarchy. We thought.

B

This is classic this I can.

C

I can go on for hours about how the type hierarchy is incorrect and in fact, I did do that for hours back in November and December, where all of these API is kept getting punted back as like needs more info, I know so like yeah. So my if we, if we didn't, have a timeline right now, I would say the answer for this has needs more work, that Levi brought in the name that he's not happy with, because it doesn't convey the security concerns that he has with it.

C

None of us can come up with a better name for it. We don't know if it should be in overload to create. We don't know if it should be just this accelerator, a bunch of stuff, we don't know which means the status on this is API needs work. Now, if you review this from the context of, we have a 300 release, that's coming up 30 days ago, then, if there's something that we need to do right now and I think that the answer for that is also no true of I. Think I.

I

C

Paid on it, it would be nice to have for CLI asp.net can take the higher bytes over the wire now I.

J

Would put back Jeremy, it's not nice to have I mean this is consumer. Goes to my pocket.

J

It's not nice to have a semester is blocking us in.

K

Scenario, no, and also on the body.

J

Component well, yeah sure, but also you cannot force everyone to update next unit. That's not the problem right. There is no reason we'll work around other.

B

Than having to drive.

J

From jaska encoder, and also if you drag or asking at order, your for saying depends: you want to use unsafe code because they, the api, Jascha encoder when you drive from it, has pointed the methods I will also put I, also push back on the concern of the naming, because I don't think this is a very common request from people. I, don't think that security concerned, the gray name, shooting should be so scary that people run away from it. I think the documentation and guidelines can solve that problem.

A

Oh that, for example like what Deva brought up I think is very valid right. Like you, you write a web api right. You test the thing in postman or whatever your favorite rest invocation tool. Is you look at your pelo results? You look at the Java Script like oh, this looks ugly. Then you go Google why's my JavaScript result ugly. Then you find some Stack Overflow post or somebody says you through, like Jason escaping and then it looks. Prettier like great I will do that in all my applications.

A

Now and suddenly you people, don't actually read the remarks section and the docs that says by the way, if you do that, you might be more insecure than you aware and that that is something that you have to convey in a three line code symbol that somebody can put on Stack Overflow like that is just a reality. If we care about this as an attack vector which we call they do because we chipped a bunch of types that protect for this kind of thing right.

D

And then people are going to look at the summary, which is the only thing shown intellisense yeah.

B

D

They're not going to read it if it's more than like three or more than one line on the screen, so so it obviously starts with dangerously and so.

C

We can we can, if we care about, if all we care about is, is fulfilling STL requirements. Another first party code, like we can always fix that, because we can always have a tool that runs over this and says: hey make sure you wanted this. What I care about is making sure that you co system its helps in it yeah.

A

But that's the thing right, I mean I. Think if you talk about an address service that we have to lock down sure they will do the right thing. If we look, if we talk about how many people push from new project with customizations to Azure and hosted well, the none of them will run STL over their corporate.

B

A

And so like for them, it's more like. What's one of the defaults- and you know, is the result reasonable I mean another way of doing. It is just use the different naming convention we always had, which is unsaved, something we could call it unsafe, Jason escaping.

A

I mean we already use the unsafe convention all over the place to convey like this is it. This is an API, but you want to read the dogs and, if I see Stack Overflow posters, somebody says oh yeah turn unsafe Jason escaping on. At least it should give you pause. If that doesn't give you a pause, then well, there's not much more. We can do for you, but I would also be with Jeremy, like in practice.

A

I think a better design would have been if this heal I could just extend Jabra shipping code override this one method and be done with it, because that seems like what the what the API should be allowing us to do, and the.

C

The API does allow that, to a certain extent, because without overriding anything, you can pass in the bed map of characters to encode or modencode. The thing is remember how I said the encoders opinionated. It.

B

C

Override certain entries in that bitmap and plus is one of them. Yeah.

G

That's what it meant like. Maybe we need, but.

C

That's not overwriting the type that's using the create yeah.

G

C

The thing that actually does the encoding the thing that should have been JavaScript encoder is called internal shield class default, JavaScript encoder in US, and probably a better thing to do for this type then, instead of add, this is in the next fix this library, so I attempted like I I, am going to be very salty here for a minute.

B

Because I haven't been salting.

C

For the past material on speak.

B

C

Attempted to do this in October with this API review. I intend to do to do this in November. I attempted to do this in December the API kept getting hunted like the feedback that I got from trying to do. That was uh national and here's where we are trying.

A

To say is that.

C

The people that complain right now, how can you it's I need I need? Not just we don't like this. I need, here's what can be done and that I haven't.

A

Heard so to be fairly fair, like this I mean I. Remember the conversations I just never had enough context to actually give you actionable feedback. Yeah I think like. If you want to fix this time, I mean one of the things. This is basically I. Think one of the things we convinced ourselves in October or whatever December timeframe. It was that most people never ever create these guys. It's by serious infrastructure code, oh- and that has to do that. You.

B

A

Is why in the party never bubbled up right? This is an example where well yeah. If everything works, fine and that's true, but there are bugs or whatever then third parties have to extend those types and that's where the thing falls apart, because the senseability wasn't really a the thing we designed the type hierarchy for it seems, and if we want to fix that then yeah we probably should sit down and circle.

A

Ii hear that you're the extension as you would like people to do like, for example, tweak the behavior of particular characters to be escaped. And then the question is: how many gestures do you have to do to pull that off and today the answer would be well. You extend JavaScript encoder you're reputable script. Encoder, you delegate, you override all the virtuals and then.

C

If all your caramel does plus like it's far easier instead of sub classing, that's tied to do a string replace.

J

I understand to the security concern and shines security concern with not making it easy to relax it. However, this is which is getting in the way of users who want to do this right. If I use like we have we, the default is secure right. We made the default secure, yeah, but.

A

J

Not an argument again if.

A

You go through razor, for example, let's say any sort of templating engine. That is it that spits out HTML right. People do want to string concatenate their secret that one string concatenated the HTML. They want to image JavaScript right in there in the HTML that doesn't make any of that secure and like if you go to on their path. You have fundamental holes in your in your application that just we can't help you to fix so.

C

Like it, amusingly enough things like race or will escape plus for the publication to escape plus and people haven't stopped using razor there well clearly.

A

Not yes, this is. This is also able to say like yeah. If you look at your payload on the wire, like sure would be nice, it would be readable, but but the most part you also optimize for Perth's your minifying. It so I'm not sure that it's that readable to begin with in most of the scenarios right.

J

Keep in mind that if you use relax Jason the pain will be smaller, so there is a part of it. It's not just have a human readability. It's also about the payroll things you.

C

Know speed versus performance by using C instead of C sharp.

K

Performance versus security, the Jeremy.

J

I'm not I'm, not.

K

Making this admit in isolation, shame.

J

Is an isolation I'm talking about in custom context to user scenarios you have to partner teams who are using it. I want.

C

To use it I agree that there is a need I, but I think that the API is not ready. I think we shouldn't rush the API just to meet the need and the partners can work around it themselves, but.

A

Ok, but let's let's go back to one thing, is the one thing I do buy from Lawson was like: let's not block the reasonable, because we don't have the perfect so, for example, in order to rename this to something ugly, we'll be ok with that I mean the scenarios unblock, the API doesn't look attractive. Is that good enough like? Could we call it unsafe Jason? Let's take I mean that seems to me like it satisfy the security concerns, because you clearly say yeah.

A

This is unsafe, it's easy to Aude it and it will unlock the scenario. Now. It's not a generic fix, but the generic fix I think so. The other thing is based on the other on the December and October discussions. I, don't think we even know whether we can't fix the hierarchy right. So there's a lot of complexity to do that in a compatible way, and the answer might be, we can't easily fix the hierarchy, but also modulo those records.

A

It's also not a Primus in our weekly optimizing, for which is why we kept planting the bug, so it's possible that we will never do the right thing, in which case maybe the compromise is the a thing you will ever do, in which case, if it just needs an odd name and it's not renamed, and that that one we can totally do right now, I'm fine with your party sing, relax but unsafe or.

C

Unsafe relaxed or whatever we do have a guideline that says optimize for clarity, a set of brevity.

C

H

And now, since it's more than that for characters, nobody wants to use it they'll just leave you all and.

A

Awesome would this address your concern.

J

Michael, yes, absolutely my concern with a dress.

C

Regardless of what word I'm saying, yeah, we should probably have relaxed him or somewhere, because that's what it's actually yeah yeah.

A

Is that an actual term that is no no I mean.

C

It's it's Max and.

H

C

I'm going to stay like steer far away from the whole unsafe station, I mean like you're the one who brought up that you wanted the name to indicate to people that this is not a thing that you want to use without thinking about it, and we have two words and dotnet that mean that one is on safety, others dangerous, unsafe, then.

A

You have slightly more a shorter or a bit, you know on saved, then we have on dangers and dangers. We only had a few kids who used to.

H

This dangerous or safe handle uses dangerous, but where thread pool uses unsafe, it means this will.

C

Be faster but may introduce a bug if you don't understand what it means yeah and here unsafe means. You have pointed that this is a foot gun should be aware. It's loaded.

C

I guess is what all our on safety. This is I mean.

A

I'm not entirely a fan of the unsafe convention either, but like it's the one we have, and we have so far failed for doing anything else. So I mean.

C

It's a little better than do-not-call relax, Jason's yeah.

A

You can also prove its application. Targeting net for five I mean that also works. We've done this in the few places a family all right, so they let me copy and paste this game, and we all swallow our pride and take a shower after this meeting and.

A

What can I say, I.

H

Mean this is what happens when you just make a decision with 58 minutes to go to release yeah.

C

Oh, this clearly isn't going to be in 58 minutes like clearly.

A

Alright, actually, let me not paces them, because the whole thing is referred to many times in this whole thing. Let me just add a comment inmates.

A

After some discussion.

J

So it's it's somewhat unfortunate that the colors of this API nor in this meeting I, would expect some pushback on the unsafe, because sp9 will now call an API the hell save in the name. Have you looked for source code?

J

That's fair! All right, I'll retract! My statement now.

A

I mean I, give you that the level of opinions on that but I think that's. This I mean I, don't want.

B

A

Your argument, but let me that's kind of Jeremy's point- is that like, if you don't have enough time to discuss it properly, then maybe be a correct answer. It's not doing anything. Yeah.

C

So in an interesting point to that- and this is something that I myself have grappled with to like the word unsafe- doesn't necessarily mean you are shooting yourself in the foot. It just means there's a very high potential to make sure that you're handling the gun correctly. It is a book gun, yeah.

H

C

F1 read what it has to say about itself: yeah, oh yeah! No just call this and think it's fine like to know what you're doing before you do it right, like the the unsafe class, for instance, is actually a really good example of this, because the the api's on there like there there are legitimate reasons to call those api's and we don't want to discourage people from calling them if they find themselves having those scenarios. But we want to discourage I, don't know we want to discourage arbitrary use of it like we.

C

We want to make sure that if you are using it, you have thought long and hard about your scenario and that you've taken the appropriate. That's.

B

Standing apart and.

C

Essentially, the unsafe keywords in C sharp means I promise. I know what I'm doing like it's. That's all it means is the compiler lets you write coach because you said you mean what you were doing yeah wisely. That's the word I'm looking for. We don't want people to use it lightly. That's their look yeah! So again with asp.net it would you know the comments on this, hopefully would say something like if you are in a content described payloads such as charset utf-8, and this is fine and they'd, be like oh we've.

C

Just we did this after we picked the char set equal, utf-8 header, both yeah.

B

C

And you're, not in spite of another envelope like HTML or or script.

C

And you're not putting this into a database, it has a bar chart, column right and a few other things yeah, so you'll get the right postdocs! Oh I get your parking yeah, okay, you're the one who can explain to people when they when they could possibly use this.

C

And if again, my opinion would be, if we can't explain when you should use this, then we should kind of yet.

C

Ma'am I'm just trying to push back on that. If we think that if we think that we can meet the needs and the disdain is okay and then we can write the documentation for it from this release,.

B

C

Be comfortable with the documentation, and it is what it is like to be fair I think we didn't even discover the PX. You know bug until like a week ago, yeah but I like this, a lot better than that. You know yeah.

A

All right, so let me actually show you what I just wrote up, so you can tell me whether you agree or disagree with mine.

F

A

What is the the rationale between property versus decremented I mean.

C

A property is a factor enough. It's just a factory enough. That returns.

H

So I've been stood on a single thing, yeah.

C

Give me the same one every time.

H

It's convenience factory well,.

C

Because the created ones all taking arguments yeah because the default property actually just falls into the create method and then caches in resolving this- would do something very simple by exposing it via static, complicated civic object returned from this one wincing, the otherwise don't say thanks, Vince Oh, actually I believe they're talking about it. If I take that the toka starts a you is the type itself once it's created is immutable and I think we actually have documented on the text encoder base class. All derived classes should be thread safe.

C

I've been saying it's another difference that it up no.

E

H

E

C

Singleton is.

A

If you return Aesthetica, but a bit yeah yeah.

C

The the intent is that these types refer to because they're there they have a constructor and then they're completely mean evolve once they've been instantiated. So.

D

Lazy is thread-safe true.

C

It's not lazy. Actually, Steve Steve actually made a fairly nifty perf, optimization to the class that uses a little bit of laziness under the covers, but it's still bright side.

A

One person that shed dongok curcuit says ultimately foot gun. Jason escaping is the clear winner here.

A

Alright, so then, but.

C

We need relax and then it would need to be put in relax. Jason and unsafe is a shorter treasure, so Mary Webster suggests that we use dodgy or perilous doji I like.

G

Da da da da Jie escaping.

C

To me sounds like it works.

D

Half the time you might have a security hole, the time you might not know half the time it works for you and half the time it works against. You.

C

A

Thought she would be great as a prefix for API I would totally and we need a dodgy attribute. That's what we need.

C

Yeah well then, when someone files above more like.

D

An alternate account on April 1st.

A

Alright, so then we are, as far as I can tell none of the three old stuff. Thank.

C

You everyone, what.

A

Do you mean this is the back look now.

C

A

Yeah I know where to go.

C

A

C

I do appreciate that you know people took an hour out here with you guys, oh yeah,.

A

We can totally bikes it for more than an hour if you really really want them today. I can.

H

Say no like nobody's business. Alright,.

A

So then, any preferences for order here.

I

C

I

That you could say.

C

Are there any that we won't understand that you just want to hop assertively for dinner and then I.

D

Tried that last week.

C

So that one that you're hovering over actually should be relatively straightforward, they search locally yeah.

D

I think the ones that I've been asked in pink the most about from the community are the ones we've said need dedicated sessions like.

B

D

Want to support more types for a vector, two, three, four common things of that know, and so we're not doing this today, should you have.

C

Members from like ml, your when we discussed those or they go ones kind of community members, not.

A

C

Of us, okay, so.

F

What is this one here? Can you talk to this because they see the same straightforward or not.

F

I

D

Not prepared go.

G

To another one first.

C

Stick then we believe us one, so this is I. Have an array and I want to get a reference to where the first element of the array or to where the element at index zero of the array would be. Basically it's it's very similar to saying ref my array sub-zero, but it won't throw if the array happens to be zero, like.

A

We have an API for that. I feel like there wasn't, and so did these angry man. What.

C

You can do right now is you can take the array, turn it into a span and then call and then call memory marshal dot and get pinnable reference from spam or dangerous cannibal or whatever the API? Is this just collapses all of those together? So why does.

D

This give you a reference to the non-existent.

C

Memory, because it is where this is, where the first element in the array would have been had the array that nonzero length and the reason that this is interesting is because, if you're, if you're, trying to make very high-performance methods, you're going to be doing things like reference address comparison. So whenever you reach the end of the array, you need an address to compare your current pointer to, and this is that address. If it happens to be a zero like right.

C

The way that you would do this is like breath of T start equals, get raw or a data. Myarray ref of T end equals unsafe, don't add, ref start call out a radon link and then inside the loop and right.

A

So what is odds already colored unsafe helper class? Is it just cool I'd, say no.

C

A

H

C

Safer memory, Marshall yeah, any ones which have universal appeal yeah. um So there was some back and forth on this on where the actual method belongs, unsaved memory marshal or marshal, and if I recall correctly, it may have been young I. Think who suggested memory emotionalism I, don't have it up on my screen right now, so I can but.

A

I, wouldn't we call it as Reve.

C

As rough there's already an aswear MapR here, it's a yeah.

A

But why would you just have one that is an overloaded fix theory, because.

C

As ref I believe takes an object as a parameter, you.

A

Know it takes span and only send as rough yep does it seem that that's what you're doing right.

C

I might be misunderstanding.

C

There's no Ezra API, then we Marshall dog has read, remember I'm. Looking at unsaved.

C

A

A

And you want this API preserver to not froward, actually give you the address on the object, but rather the offset is to get to the actual data right.

C

Well, as ruff is also contingent on to your struct.

C

Right, oh yeah, yes, I'm! Sorry, as well as ref takes a bite spam. Basically.

B

We're all a raw memory over and.

C

Projects it as a struct. This gives you a rough to the actual element in the array. It doesn't.

F

A

So you want to overload to get reference.

C

A

But why can't you just call get reference because it takes a span of teacher to be able to plug in there all right because get reference. There's it blow.

F

Up and assume no, it doesn't blow up.

C

And value the empty array is their element for.

A

Only one of them and.

C

What you said earlier.

A

Yeah, that's why I'm trying to understand what how this a key, because I I'm pretty sure we had an API like that.

A

A

You know this is opinion, so to put it.

C

I would be fine with your reference.

A

Yeah basic, this is what the guys doing it to is. That house is any different from what that, when you should be able to do just call memory marshal delicate.

C

I mean that's basically.

H

What it does under.

C

The collars, well it's in codes as well it. The only thing, is a memory marshal dog. Yet reference of span and read only span is guaranteed. Never to fail. The method will never throw an exception under any circumstance. Memory Marshall get reference of T array could know whatever it doesn't seem like reverse anymore. Are you no never know where y'all actually talking about that.

C

Because absolutely it would be, it would be a breaking change. That's what that's what it would be. It would be a source breaking change so right now today, because t array is implicitly convertible to span up to you. If you call get reference of t array right now, the methods guaranteed never to fail, even if you pass involved with this overload, that would now be the better match and you would get a you would get a little rough coming out of it, but he's basically saying don't do that right.

C

No, so he's saying don't do the null check at the beginning of the method, just like an aura propagator.

A

You know I get this so basic, so it's Amy. We remove this line here right.

A

If the thing only does this, then why do you need this API at all? Why paint you just literally call this API this.

C

Api will this API will is contracted to give you the address of where the first element in the array would be if the arrays null the API is meaningless, and we have to throw an exception of some type, a null reference argument or whatever, because we literally can't give you the address of where the first data would be.

C

Okay, but I. Think emos question is, if it's implemented as remove the if and just call the return yeah.

B

C

That will not throw because new span of null array is defined, that, yes, all spam correct means it would give you the ref to zero besides it's doing currently so, obviously. Obviously, the implementation does not look like this. If they, no reference, exception is coming from somewhere. So where is the null reference exception? Coming? The no reference exception would come from the the dereference of the length property where it it's.

C

The method wouldn't look like what's written here, because this method has a null check on the input parameter and, if we're removing that null check, we have to do it implicitly through access in the light property.

A

I thought to me: I'm still not entirely sure, so why would you call like what? Why would we add this API at all? Why can't the caller literally just say memory Marshall get reference and passing in the array, because any array is implicitly convertible to the span. Yes,.

C

Including the Dollaway, including.

A

Another rain, which would give you just an IRA right. It would.

C

It would return back a ref of null and which, which is totally valid. If you read it or write to it, which is fine, which is totally fine, because you you're you're, still able you're the CLR guarantees absolutely guarantees that you can talk about the reference to the first element in the array, even if the arrays of length zero, you just can't ever dereference it, but you can talk about the reference in this particular case.

C

If we said here's an array that happens to be mole- and you give me back the retinal saying here's where the first element of the array would be. You are now talking about a nonsensical address, like your you've, now basically violated the contract of the method, but.

D

Memory marks will be reference already returns, no rat.

C

It done swings, it does only for nullus fans and mullrine all these fans and when I say no I mean they literally have a point of it on one side, if you create a span or a read-only span around an empty array like there actually is a memory address, error right.

D

Yeah right so I think what mo was saying is that we could just tell people to use memory, Marshall get reference and that will create a ref, a no ref. If you pass in a nolle array, if you pass in an empty array, it will give you the address of the where the first element would be. So exactly what you want to forget all right data, and then you don't need the if array equals null check, because you'll just allow the implicit null to happen. Yeah, yes, so.

C

Is there? Is there any better justification inside of this work item than what I'm saying like it anywhere in the comments? Is there a better justification for why we need it at.

A

A fundamentally mean sounds like: is this yeah exactly so? This is yeah. Basically, what this guy's saying, no.

C

It sounds like what's being argued on both sides. Here is: why do we even need the method at all right.

D

Because you can just get reference.

C

And what I'm trying to figure out is, if there is better justification inside here, something anyone at all times.

D

Yeah I think the only benefit would be a people that get reference and took the TA. You avoid lead for any kind of conversion to span in the case where you already have an array, but other than that I think you can achieve exactly this using the existing get reference.

A

I mean with no overhead, it sounds more like the only thing you can't do is you can't say, get reference open, paren now close Quran, because, like that wouldn't work, but right but I mean I mean it would seem. You.

H

E

Know as tier I would work at what d4 would pick.

H

Some sort of team yeah.

A

A

Yeah I mean to me: I became just mark it as like you had somebody explain to us why we would need to.

C

Have a punter pack them! Oh, let me actually just try to write out with they don't like thought. That would be easier. As with always all.

D

Right I'm up to date, with the with the issue, then I said alright, no calm, I'm.

C

Glad I bought you time.

C

D

We've got another one that would be good, be unsafe, skipping it that we talked about.

C

C

So one I guess one difference between this and new span of tea is that new span of tea does array variant chucks on its input, whereas this wouldn't oh I,.

A

Think I know what the problem is.

C

So if you have an object array, that's really a string array. This method would still work.

A

So what happens when you have span of T and we'd only spend of Tears overloads in your past in the T array and they're, both convertible I thought.

C

C sharp said one of them was braver. Ankle. Sure I think that arrays only convertible ban of T, but that span is convertible to redouble expands so since span you get there in one. No c-sharp never does double implicit conversion. Okay,.

D

A

The way C chef works as it compares to conversions from spent with only span of T, and since we only span of T as a convert since span of T, has a conversion to with only span of T. We only span of T would be considered.

A

No just enough people be preferred because it's more specific right.

D

But I was saying that if you've got it, what I was saying is: if, if span, has an implicit conversion from T array and read-only span didn't see sharp, won't, allow implicit conversion of T array or.

B

Toledo's, that's.

F

Right because it.

D

Doesn't do double hakka both of them have an implicitly it.

C

F

C

It for primitive times in certain circumstances, which is super apparently span and read all new span art allowed by tried out node, so I can't test it right now, which I guess makes sense.

C

But he first man would be a lot, but again things like. What do you mean it's not allowed system doll? Spam cannot be found. You need to change off t9, not to nine, like exit c4. How.

D

Do I do that in try, don't know a little drop-down work? Oh no dude! Don't do try it on that. Go to sharpen. Why not know is not a sophisticated for the purposes of ilt compiling sure.

D

D

Okay, so it fits the span overload, yeah fluff mix this.

C

Which means, if you have again, if you have a local typed as object array and it's pointing to a string array which is totally possible because of array variants. You'll actually get an exception at one time, because you can't create a new span of object around a stringer right, for instance,.

C

I, don't know if that's an important scenario, but it is an example of what would fail.

C

A

All right so I put it back, and then you can think about this offline and probably Yan can do it in five seconds, but I think it's an area for him. What applies our young? What was the.

A

G

Six one two, three all right! Thank you all right. You.

A

D

What's yours again, okay, so basically in normal mathematics, a cross-product only applies to a three dimensional vector, but it's very common to have overloads that also do a cross product on two and four dimensional vectors, because it's basically just a simplification of the algorithm. And so the request is that we expose those on system, numeric, spectre-2 and four okay.

A

That seems like it. You know.

A

C

Does this cross product formula have an industry accepted definition.

D

It's the three-dimensional, so the cross product applies to three seven and other dimensions and windows for those dimensions. The algorithm is all well defined, and so the simplification is basically saying because the unused elements are zero. You can drop those because the subtraction of the subtraction of zero returns the original result. So this is basically just saying, given the three dimensional cross product drop, the third dimension, because Z times anything will be zero, and so the subtraction will be zero speakers by the hands of Jeremy's.

I

Question is that how people interpret to possible.

D

That's how a direct next math exposes it and the mother won't know much but yeah like on the vector.

C

Oh I, guess it is so this is, if you assume the Z is zero, then a cross B produces a thing that is perfectly aligned with the z axis and that's why you're getting a point like I, don't understand how the cross is producing a single value with a point. Well.

D

That's just what cross product does cross product always produces a vector? No, yes is.

C

A cross B is the right hand, tool they cross. B is oh yeah. Your your dot products produce a single dot. Products produce a scalar cross product is a vector, so I don't understand how vector prospector is a book.

I

How would you comment on there that explains it? You scroll up a little bit. The video the Z coordinate is from the cross product yeah.

D

Yeah, it only produces the p-value since the inspector to exists in an XY plane so effectively what you were saying: yeah yeah.

C

Because because the X goes to zero and the Y goes to zero.

C

I mean it's all under checking out a plane, but like I, just wonder if if we need a more complicated version cross, but that's where I'm asking the question is: is this a if we went to MATLAB what it understands? This version of a of a vector, 2, vector 2, turns into a scalar as a cross-product I.

D

Don't know if MATLAB would but, for example, the DirectX math library, which is what the vector 2 3 & 4 implementations, were originally based off of exposes these methods, because they're commonly used in various graphics applications in gaming.

D

A

Well, my position is: if that's what sense of happening that seems fine to me. It.

D

Is worth noting that the one difference is that DirectX math does still return it back there, but with the value commuted to all results, which is what we do for some other methods on vector, T, 3 and 4 as well. So DirectX would vector.

C

To prospector, 2 produces a vector 2, where x and y have the same value. Yes,.

D

So it's a question of: do you want to return a vector and permute, or do you want the user to explicitly have to prove it to all result?.

F

So then, what's before doing it returning back to school right.

H

C

But I don't know I, don't know how to right-hand-rule five dimensions. So I, don't know what vector four is losing over director five. So, like I, don't know enough about multi spatial geometry. To understand, like the question I'm really asking is: are we are we modeling bad behavior or our.

D

Remodeling standard behavior we're modeling behavior, that's commonly used in games by anyone other than DirectX well I mean no. One else really exposes a math library. Unity has similar functions, I believe I looked like there's, OpenGL have a similar thing: OpenGL does not have a math library, okay, not not a standard one. At least it was.

D

Okay, no unity, no or X, and that a or FN a provide cross for vector two and for DirectX does and I thought there was another one that did but I did not list it.

A

I mean it seems like most people just consider it not defiance, and that sense is not conflicting. It just means we have a behavior that somebody has and if you don't want to use it and don't use it.

A

H

Then that is worried. We are going to teach the children wrong. We probably are but we'd.

C

Already get yelled at.

H

C

That's there's. Yes, we just to come. Don't you slope.

H

Then I've been trying to say that you can bring it up.

D

Especially for the cases where this is intended for use.

A

I

C

You're vector2, if you cross yourself, you end up with the nonzero answer.

D

If you cross with yourself into no, you shouldn't, because X Marvel 0.

C

Well, it's the definition of cross-product says: if you cross with yourself, he produced the zero bender right.

D

So you should be producing zero as well. Yeah I'm, just I, don't know, Matt that we added there would so in four dimensions. The hochschule of a buy vector is 2-dimensional. Another oriented, plane element.

C

You said words confidently I, like that.

A

I like how you at least be honest about what your criteria I didn't hear what you said, I only heard the way you said it I mean the hoggish rule of the buy vector is a something.

D

Yeah it basically rhetoric returns an oriented plane.

C

I gave up on eigenvalues in college, so whatever yeah yeah I agree, there's no plan, so my only concern is, if we're using a word in a way that it doesn't hold in industry that yeah, it's not good.

D

C

That it's it's.

D

The Wikipedia page has an entire section on, like true cross products exist in three 7-eleven and other dimensions. That I can't remember what is your hand? Oh, because I stopped caring after like three yeah. What I can only do this yeah? That's this I, don't know how to do a cross bar and then.

C

D

There's well divine mathematics for others, but it doesn't it's not like a true cross product, as you would consider appropriately dimension once my need exists in higher.

C

Kind in mathematics, so I guess the mi my other way of phrasing. That is if, if there are multiple interpretations of what a cross-product could mean and in even space and we're choosing one that we should be conscious about choosing one instead of simply deciding on a whim, and if there are two, then maybe we need more complicated names, I'm like right, cross, whatever and cross whatever, so that we can both implementations or all 27 or whatever.

D

There should not be two: the algorithm is well defined for any dimension right. All.

A

Right but it seems that we are there's.

D

A there's another one, which is three eight, five, eight five, which might be simple.

D

C

Killed it Oh. Does it involve the I whatever to be fun? Was it was.

D

That the one three eight five, eight five, no.

I

No, that 3:8 585 b8.

H

D

Clearly, already on vacation.

I

I don't know why but I'm really been listening to number.

D

One yeah adding a new method to unsafe, basically c-sharp has definitely assignment rules. You can bypass those in a lot of cases, but not all of them. This is saying we should expose a new method called skipping it, which takes an out T and literally, does nothing as its implementation, allowing you to bypass definite sign meant for any value, whether it's this or anything else.

D

The primary use case for this would be things like when you have union types where c-sharp requires you to initialize every field and the overhead of pinning might be too high.

C

I am fundamentally opposed to this, in combination with the fact that the CLR supports none, initializing or not initializing.

D

With this does not apply to this parameter and to out variables that you know are definitely assigned, but the c-sharp compiler has trouble analyzing appropriately, yes, but in addition.

C

To stack alack, which can give you previous stack that was not cleared. If you have the 0 or no 0 in it or whatever the flag is set, then you do this on a large struct and it's now populated itself, with whatever was previously back right instead of zeroing itself,.

D

Out so I, which is.

E

Really fundamentally,.

D

Believed we should not add this ever, which is why this would be on unsafe, because you can do the same thing by pinning explicitly but there's overhead from pinning, which may be unacceptable in some hyper scenarios. What do you mean? You can do the same thing if you take the address of a variable using unsafe t-shirt, joints all definite silent rules, but painting has some implicit overhead that may be undesirable in some scenarios, and so this basically gives you an out for that saying.

D

This is a zero cost way of saying I know what I'm doing I know I'm definitely signing this appropriately.

D

Let me do so. The homeland is API being used commonly if you're in a constructor, for example, of a union type with explicit field play out everything else. The first thing you would do is, you would say skipping it out this, and then you can freely assign only the union field. That is the correct type for the current instructor and you would not initialize everything else, and so all memory would be appropriately initialized.

D

But you would not incur the zero allocation cost of all the other fields and you wouldn't be forced to assign all other fields, and we are sure, there's no call being made in.

I

D

All right, the JIT would see it as a rat, so it would inline it to nothing.

D

So, there's literally no call added, there's no I are blocks added. It's completely alighted.

C

Is there something we would be able to do that would kind of help alleviate some of your concerns get.

H

C

Nose ear, would it feature kind CLR at which point I don't care?

C

Would it be elided in tears here? Oh, it wouldn't right.

D

I think in tier 0 it would depend on whether or not the JIT recognized it as an intrinsic as it does with many of the other unsafe methods. I would.

C

Think it would be alighted because we we would need two special cases method in the JIT anyway, so I don't see why it wasn't me alighted.

D

You wouldn't necessarily have to special case it, in which case in tier zero I think it would still be emitted as a call. That does nothing. Ok, but if you did special case it, then you could just have the JIT drop it from Co gen entirely in the importer. Okay,.

C

Read it would see that ahead, no yeah I mean depending on if first it's already seen this a trillion times or maybe it would argue, yeah.

D

Because the tier 0 doesn't try to inline anything that isn't marked as aggressive, optimization, okay, don't think about it.

C

Yeah I mean think you can do the same thing already by invoking into a thing that doesn't actually pay attention to the thing, so it rises already yeah, it's possibly the most minimal overhead that Judas would avoid. Do we have code pads in core sealer or core effects right now? That would be able to benefit from us.

D

I'm not sure if we do in core effects in horsie alarm I know those places where we take advantage of the fact that we strip locals in it and I know: we've got measurable, perfect crease from that, but I don't think we have many union types today are.

C

They more common than things like graphics,.

D

Processing in a row in our Interop code in general is a lot of C code uses unions, especially in the Windows layer.

C

Would it make more sense, just devil's advocate here if we could somehow make the jet recognize that we're setting burn through setting very specific fields of a union struct we're basically overwriting every single bite in it anyway? So it shouldn't bother I, guess it's not the kid! It's a c-sharp compiler! That's complained.

D

Yeah, in this case, it complains and forces you to do so. Yeah in.

C

D

C

Like I I would want to see this added only when we had like honest to gosh cooked that we had, that would benefit from it like I would want to see this check in at the same time that we consumed it in places where measurement showed. It was valuable. Well,.

D

I've got measurements that can show it's valuable in Interop code, I've written, that's external to the framework, which is really what this is intended for is like not the framework itself, but Interop code outside the framework. I I would buy that argument, actually right now I'm trying to do this with IO rewriting, which my is up pdbs and things like that and I'd rather have a built-in mechanism on the existing unsafe class. We have for things like this to make that easier, so.

C

I wonder if maybe we should add too, if you're willing to share that code. If we could add to this issue like the before and after for the kind of code that you have to rely, I guess it before and after is really just my struct T equals default or not. But if you right yeah, we can do.

B

It, let's understand.

C

Like what the, what what you have to do to work around that today and what the actual throughput games are yeah? No, but we could. We could actually still do this because, if, if in order to make c-sharp copy, if you write my struct T equals default and then T dot field, 1 equals T dog field, 2 equals blah blah blah. If we could somehow get the jet itself to realize that every single overlapping field was actually being set appropriately, could the gypped just elide the equals default at the very beginning, I.

D

Think the analysis for that is expensive, ok, but.

C

I'm not positive on that, because I was just thinking. If we could somehow do that as an alternative to this, then even long Union trucks should benefit from.

D

Right well, I think in the case of non mediums, drops you're only signing each field once so. It doesn't matter it's in the case, where you're being forced to assign these same memory multiple times, because it's overlapping yeah and the c-sharp compiler has no knowledge of like hey. These fields are overlapping but I'm only wanting to access these right.

C

And that's where the equals default comes in at the beginning, right.

D

C

Just clears out everybody.

D

Yeah you either have two equals default, which forces potentially triple zero initialization yeah, in addition to the stack, zero and dash it does, or you have to pin, which involves some overhead and both of those in tight looms can be significant, like if you're doing, DirectX, 12 and drop you might have render targets which you might have like eight of, and all of that is built up of multiple Union layers, yeah and so you're, going to want to clear clear some of those and set up logic and everything and you're gonna have to loop through that and each time you look through you're, going to have to clear out fields that you're not touching, and you know reinitialize them.

D

C

If what if, as we said like take, take some of the code that you've written put it in here, just a show, but also if you can somehow jump to the before and after code run, I think that would probably really solve us. Yeah I can do that. Does that seem reasonable Steve.

E

Yeah I just want to I'd like to see you know real code benefiting it from before. We add something that works against the language like this.

C

Because I'm, ideally like ideally I, would really like to have legit just understand it. But if, as you say, it's impractically difficult and this might be a decent alternative.

D

Is it really working against the language, though, considering you can already do it? It just has minimal overhead from requiring me to pin yes all right.

D

C

C

As I have no doubt, you'll actually be able to show brains, it's just I think visualizing. Those would help us I hope you noticed yeah.

D

I've got an existing app, a demo app of DirectX Interop code that runs on xbox, be a UW key and I can just I can I think I actually push that to get Xavier so I can just do.

C

The IO, rewriting I feel the difference along these same lines like did. We ever actually make public the API which allows you to set skip locals in it. No.

D

The c-sharp language is the the initial prototype was done by an intern and it's in a branch, but it requires further review and cleanup before it can be checked into master.

D

And I think it's not as high priority right now, because for FX and cor cor are using the aisle link or flag that we had it instead got a which I only writing. Maybe.

I

Where that's going.

I

I

This one's look using the simple.

F

We want the same: what to control instructor.

C

D

Yeah, what's the difference between this and public stringbuilder, one I think this is the entity.

I

F

What's the wrong thing, that's one.

A

I think that's actually fairly good point.

D

Theoretically, it could be a breaking change, so.

C

What we, what we might want to do is always.

H

C

The ASCII letter, as the capacity, if you wanted to do this, there's already a constructor on the string type itself. It takes a char and a repetition count. There's not necessarily. Oh there's, also an overload of shrinko we're all dependent that does the same thing. Append char int repeat count having a constructor that just initializes that would would provide symmetry.

A

It to me the argument is more people write code that looks reasonable, but doesn't do when you think it is right.

D

And then that could be a silent source breaking change. It could.

G

C

On the other hand, that seems far-fetched the only real value that would conflict that people are likely to pass in is the literal, zero and zero is always going to match to end before it conscious to char, but.

A

Then inspect expected right. If you have a literal, zero there I don't think you would think this is a right if.

C

You have a literal, zero you're, probably calling the capacity based overload.

I

Seems odd to do that, but yes.

C

Well, you could reasonably expect that you could pass zero for initial capacity and like 2k, for max capacity.

F

C

Stretching a little bit but I could see it probably reasonable. Yeah that particular overload by the late intent is nine percent usage across api's I've done there were so there's an intent overload on stringbuilder initial capacity max capacity. That particular constructor has nine percent usage across a Piazza.

A

The capacity in max capacity.

C

Unmask yeah, the overload that takes just capacity has fairly substantial usage, which kind of makes sense, because people when they're creating string builders, often have a good estimate of how large.

A

Well, they're, pausing, the video, that's a good estimate or not, and what should help with them. I'm full fighting that, but.

C

But anyway, long story short. If we wanted to do this, I would probably do char repeat count rather than.

G

Well, but that would defeat the purpose right, not.

C

Really like you would be because.

A

The point is not that they wanted to do that. The point is that it doesn't do what they think it does. We will always have to have an overload that just takes charge too, so that this is calling that one not calling the one that takes the pass today.

A

But yeah I, don't yeah I mean to me, is what the reasonable to say. You just do the capacity when you say a pen sharp of the repeat count. If you really want to do that.

C

By so this is what are we going? This is where I make everyone sad by saying you know. C++ has this awesome feature where for individual methods you can.

B

C

To forbid implicit conversions from data type, a to data type B, when there's nothing to being to note I think.

A

You have a point, though, because the same area plans to the other one is wrong.

H

Don't know if anyone.

H

H

C

Also, does this only happen with literals.

F

F

That is my understanding.

F

F

F

C

C

There is an implicit conversion from char, don't.

C

A

So I think we should just add these two and I call it today: I'm not opposed to No it's fairly straightforward I mean this.

C

Would be you've got one for the community I pick up two yeah.

A

I think this one is a pretty simplistic yeah. There's.

C

A there's a flip side to this, which is someone we add this API. Someone now starts using this constructor copy paste the code into something that doesn't have the API and now they're gonna get weird behavior yeah.

A

But how many things do you know that would similar just training where you possible.

D

By targeting framework in core all right, multi targeting framework in core they're using this API.

C

In our source, common folder, right mm-hmm or any similar situation or again, the stack overflow copy/paste, as mentioned earlier you're, calling a different available honestly, like I, understand the concern, but this doesn't seem. There are plenty of places where you implicitly do I think the nape yeah it doesn't exist. You call it thinking that it does and you end up implicitly casting to something else like it doesn't I get that it's a pain but I'm not sure it's worth it. It.

D

Seems like we need an analyzer in either scenario either to either we don't have these api's and we need one for both core and framework to say: hey, you're, passing a character here. We don't have an overload that accepts it or we add the API and the analyzer is only for framework, but I think we need to analyze it for this case, when.

C

People actually get the full framework analyzers like how would think that those analyzers point in the package manually.

D

It's a it's honestly, a case I think that's chicken-and-egg problem of like how do you get people to use analyzers when they don't reference? Nuget packages might involve dinner like there's broadcasting the fact that they're there and blog post saying this is why you should probably have it and trying to get people to have the best practice of including analyzers is probably the only fix, but this would.

C

Be things along these lines would be a hood bird now. This is just making me sad that we have a charter, implicit conversion.

C

To be fair, it's not something! That's going to surprise you in a production application, you're going to find it immediately that.

A

Yeah I think I just edit. This comment, but I honestly think like at this point, I think Stephen has two good points. One of them is I suggest, adding the api's may make it actually worse and then the other problem. Is it it's just how c-sharp works right like if we add the api's here now.

A

We may have to add this in plenty of other places where we have similar patterns, where it's just becomes this never-ending thing right and if you said like yeah, you will find the movie like it's not like a life ending event. It's like you run the code once doesn't do what you want fix your code, your.

C

Some customers test: okay, well,.

A

I mean if they don't then realistically, there's nothing. We can do because there's like a million other things, they can do.

A

So in that case, that would just say you know what, let's just close it it's a good point, but.

C

So just resolve above yeah.

C

Yeah I get a little bit of anxiety when we talk about warts in the framework like this, but I I. Guess you and Steve and others kind of accepted this as a fact of life better than I have well.

A

I mean to me feel like I mean if this point mean it's, it looks pretty bad on the surface right, but how many times has anybody ever heard of this issue? I mean I've, never heard of it. You've probably never heard of it like so that kind of proves to me like, while the problem looks bad, it can't be that bad in practice, but compared to let's say, bracelets right where we know that if you don't have braces in the flock you can cause plenty of bugs, but yeah.

D

We do well that's.

F

D

Know the guideline says that it is not required for single line, but you can't have some code pattern following there's healing like that, we should uh we should.

C

Probably just turn on some type of style input. I know we already have it for the Native time. We.

A

Have three minutes left anyone discussed our deck that discuss what this calf coding style yeah all.

F

C

Is the one in the middle of the screen short? There were two move method. Is that one short yeah.

F

That's what every minutes.

C

So this is moving an array or a list in place. It's basic moving one way to the other right or swapping to is it two elements or okay, so move element and then move elements.

D

How does this work.

D

Listen, how does it work for the first overload if from preceeds to is the first over? Basically, oh, no, that's that's just swapping one. Isn't it it's not doing account.

C

As I originally thought that this was swamp well,.

D

The first ones, the first one is a move, not a swan and then the second one is also a move, not a swamp, but for a range.

D

But I would maybe expect coffee not.

I

Move well, I thought of coffee. You you, you don't increase the number of elements in the array. You're.

D

Copying a range from one location to the other to another location in the same just like so you already have it. We already have copy over loads that take two arrays where you can give the same, and you can give source and destination index and things like that. The move might imply zeroing yeah. But to me it's like you have a.

A

If I index zero, it's not a copy right. You basically have me right. No.

D

That that would be a swamp, not a move. Well, it's a shift right, but as well. Well, that's also not a move move would be, you would have B, C C would be moved or potentially.

A

No, that will be a copy. I mean that's I'm, saying to me: moving flies as a yeah.

C

I'm I'm looking to see what the actual proposal is I'm having that's.

A

What I saying I don't think it's copied, because I think this is a copy. I would agree with you. This would be BCC right.

D

Like if you're thinking of it in terms of most mood semantics like files rather things like that it, it generally involves moving and then deleting the old index so bc0 or leaving the original BCC yeah. Exactly what you just said. But if you do.

A

What you just said, it could look at from a file standpoint, yeah.

A

Bc a BC: they need to leave this one, whether you end up exactly what it is that yeah.

C

The the action, if you read between the lines on the proposal there they're proposing a shift inside I, don't think move is the best name for.

D

C

So if you're, if you're moving later in the list, then everything between from in to gets shifted to the left by one the original element gets inserted immediately after that shift, and then everything after that remains unchanged if you're moving an element before its current position in the list, everything in between gets shifted right by one and then the original element gets the place right there. It's that code right there.

C

So it really is a glorified wherever this element happens, to be remove it right now and then insert it at the index. I'm telling you to yeah.

A

It basically logically remove full of bun and yeah now by insert I should say, but, like.

I

The users they need a method for that. It.

A

Seems like a very perfect thank you, I mean.

D

We can consider it next time. Yeah I would be interesting. I would think that the potential benefit would be if you do insert and remove. You might incur more copying than required or more shifting, but.

A

All right, so we have to lose in the room. Let me just say: I have no.