.Net Foundation Design Reviews, 31 Oct 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Reviews: GitHub Triage

Description

00:00:00 - Rejected: Add TransformationStatus enum from corefxlab https://github.com/dotnet/corefx/issues/22412#issuecomment-340833769
00:05:11 - Approved: Productizing APIs for {ReadOnly}Memory(T) and friends https://github.com/dotnet/corefx/issues/23862#issuecomment-340845398
00:41:10 - Approved: Add Overlaps/Within extension methods for ReadOnlySpan(T) https://github.com/dotnet/corefx/issues/24103#issuecomment-340856211
01:15:51 - Approved: Memory and ReadOnlyMemory validation errors not matching https://github.com/dotnet/corefx/issues/24296#issuecomment-340868030

A

Other guys, let's get started without too much about foreign heights problems.

A

Some of the characters are getting close to it now lies good life I'm fortune tellers, not here so I get to decide which issues I review because nobody's.

A

B

Is great, we gone through eight, how many of it? Well. That's all you need to look at the bottom one. We looked at last time, finishing all of the midnight sit. How can it be.

C

Both ready and blocked.

B

It's something you don't start with the pointer.

D

Was in point it hard because it's blocked on the language feature we're deciding we're gonna wait until was what NP native ensure.

E

Interesting table take ready for you.

A

F

E

Same thing here, what is this block on crystal? What what is this one? Oh.

D

That wasn't really blocked because we're gonna wait till we start reviewing the other API, so we're gonna return, this eat them and try to get a consistent. Yes,.

C

I think a DS. It seems piada.

C

It was just renamed to James. Partisan has proved to move on like.

A

C

Propose that we I wonder it like it was not in the one of the reviews of the actual IP is like. Why are we reviewing it? Independent.

D

From everything else originally Steve about this, because when he was at miss panty P has some of them. We asked him to return this for some of the API. Yes, and then we were like. What's this tractor, get the rest of the poor effects left? If you guys are you looking at a holistic, they decide.

F

D

Come in as part of the basics for.

C

Me that's what I thought.

D

C

Don't know what.

D

Do you mean other enums.

C

D

Are the other api's are using other than that? Yes, Oh.

C

C

What was the other one that asan was doing? I can Bonnie there, it doesn't use it.

G

C

They are doing now, parsing and formatting, which don't use it because.

H

C

Like when you have like you know a large batteries and prevent transmission, that's all like we were gonna, so maybe.

D

C

Decide but it's part of the base64. We are in fact that one is not on it's checked in our chip, yet.

D

On enum I think we should look at the two as part of the follow up with. This is look at some of the api's that we exposed it by and you and Harry had strong opinions about potentially getting rid of the email, yeah.

C

Well, so Levi initially had some objections and then he kind of you know last meeting. He said that he conceded after looking at you know with us on at the API. So I don't have a strong opinion, but I in general I find it weird. It says you know it's still open, because it is a dependency on the API that we already approved. Well, this one Steve indentation his API was a check.

G

C

Not surprised that it was opened initially I'm surprised that it's still open, I I, don't think we should be reviewing it. There.

D

Was about to say, like I, don't think, are the value script, the universe on the neighbors matching this, namely rigid yeah.

A

It's probably also don't think interviewing in an all-action consuming it. That is right.

D

C

But this was blocked. Oh yeah.

A

Click on the base64, let.

C

Me just say this is no Frank yeah, it's actually merged. We can even look how the minim enum looks. Let's change it.

D

If we are interested, I want to look at what the final thing looks like you look at the right.

H

There, the bottom just on so it became operation status if now use done destination to small need, more data and invalid data.

D

Now we can't see what it used to look like it's a that was actually the same, except for the different name. It was transformation status, yeah, alright,.

C

Move on, let's move on yeah.

E

Just a time exactly, how did you know that I literally just pull this up? Oh.

A

You have to work on the entrance effect. Them is to make a david keene entrance. You just kick over the door and saying you cannot start and then.

I

What is this one? This type is already checked in, so there's two questions right. The first one is, should we add the bulk operations that we haven't span like fill clear copy to index off, etc. A second is we want to discuss where to use a span. Mit use the property method versus a property for span to memory, convert entire memory dimensions, yeah I, see I.

C

Can't operate I'm kind of torn I understand that it would be convenient to have these operations, but they are potentially a performance trap in a sense that getting calling these operations on memory can be expensive. The better way to do it is to get a span and then call operations on this path. So you pull several of these operations in one Star Trek, it's better to get this fun. If you call just one they are pretty much equivalent, but.

G

Either is there any benefit to if you're just gonna call it one? Is there any benefit? Can we do anything better in these implementations that a call it could do by calling span copy I, don't think so.

C

The especially convenience message to avoid that yeah that exactly exactly they are just convenience methods to get a span and call them. Well, if you yeah, even if you do like multiple of these calls, they will be getting multiple spans, which is equivalent to.

F

C

I think we understand, like these implementations of these methods, would just call span that index off Harmon. Well, there is a harm that they kind of you know a better ways to get a span, have a local variable and then keep calling different methods on the span, not on memory.

J

So some overloads missing, so this sequence, equal for memory, read only and read only read and should be moving.

C

No because this the reason we have read-only memory and memories of the extension methods can be called. Okay, the second parameter. You know it indeed just custom I.

G

Certainly have a strong opinion, my inclination would be to just leave them on span and you just do dot span. If you want.

K

To access them.

A

Otherwise, I only see the worst possible all this when you have like two times, if you paste you have to manually, keep in sync I think that's not great.

C

Yeah, that's another negative of adding these methods is that we will forever have to like whatever we had to and we have to add with the other they get. You know they diverge. Do we have any.

D

Other duplicated members right now on them can we bring up Watson arena.

I

So, no just a spoiler just like that yeah. These are the ideas right now I'm just evening in so slice is common. Where, with slices, it's not.

C

Different calling that span that slice- oh sure, yes, oh.

D

Yeah trying to understand, if we've already got some IPS were duplicated yeah.

K

That's a good question, so I think I to array is probably.

G

One of the only ones like that that exists on both memory and span and there's no difference right, correct yeah,.

D

What about the mutation once you get this course that span here, you're, not mutating the memory. Are you in place human? Well, if you so.

C

That that's an type or the memory type themselves. They are immutable, but the contents that they point to is.

D

Not so if I call memory of T dot span and I write to that span, it's gonna basically reflect that and sharing the same yeah.

C

So I think is Eve is right or whoever said that to memory is the only one.

L

C

Array that, like just calling span that, where I would be equivalent I, mean.

A

Honestly, those seem fine, because it's just another type of memory. Yes, and it's like having them only absolute five minutes wait. Okay, everything else is just yeah I'm.

M

Even the slicing one, you could argue is.

C

Somewhat redundant but like no slicing is very not redundant, because if you can't span that slice you cannot get span which you cannot start the on the and you can't do yourself, you can't you slice back from span to memory exactly.

D

The span already have fill and clear, for example. So if you do memory that span that clear yeah.

C

Okay, so let's look at those, so the only ones I'm thinking. So what is as ma'am? Oh I see yeah. So maybe there the principle is. If, if there are conversions, we.

C

Don't so conversions are as memory I from any array segment as read-only memory do.

G

You really want us to be extensions like so. Every array has an as memory on it. Yeah.

C

That's a good point: I, don't think so. Exit everybody.

I

Has the last man like oh well,.

C

C

Don't know I like to spend more than memory, so maybe I'm, biased, I, don't know what do other people think so.

A

Honestly, if we have to make sense because I mean like I said before, I think of these type is effectively the same thing except one of them allows you to push it on the heat right and it's from. As a result, it has slightly more overhead right. It's.

C

Just me up, like you, know, having one as panel on a ray, it's you know it's additional member on the ready having to and then people having to think which one they want.

G

E

C

G

Nurse as memory and as read-only memory plus the out spam, it's.

C

Actually is four so like so.

H

One can argue that what we want is to get rid of as spam to add that memory that you can do as memory- that's bad or you can call this man constructor all right unless you think that memory is going to be more rare.

A

So the reason I think it's useful to have them as peers is I. Think of this in our air, where somebody has to positive around I didn't have to call an API on it.

A

C

Just I'm thinking so maybe I don't have a strong opinion, but in the code that I wrote, I very often want it. I cut an array and I needed to get a span, because then I can do like some operations on on arrays that you cannot do natively on race.

C

Creating memory is something that like you to once, and somebody just passes you the paper. You touch it and start.

G

It is basically.

C

It time to store something for later, which you don't do very often you don't do it like in every signal routine, that you know you enter a routine. You have an array and you do some operations on it. You'd, like don't cast local arrays to memory. Well,.

J

I mean so that forces you to type new right, which, at least to sleep with first people indicates, is getting me so.

C

For what it's worth I have memory doesn't like if you do memory new memory and passing their a no, he politicians, okay, well,.

G

I have like, with some of the api's we've added, that take a memory and say cancellation, token or something I have given an array. I have passed in the array, but I'm not calling as memory or the equivalent explicitly I'm. Just letting me implicit cast happen. Oh.

C

Because it is a common form array to memory, yeah call any method you just want to convert it. Then they, sir cast so.

H

Rain goes implicitly to both. Does that mean that we need to be careful of having overloads to take.

C

Well, in fact, it probably doesn't Elektra make a lot of sense to control. If you need to start the thing for later, it should take memory. If you don't I.

I

Think the implicit cost for f-sharp, if this is the issues, is saying that we should have explicit methods for every. If this is passed.

A

I

Who has a specific off it licit in similar billows? Somebody comment to me: the code sample would look like.

G

I

G

Providing explosive methods for implicit operators but I, don't know that they have to show up in everyone's face. As you know, extension methods it I.

A

Mean I agree with that in principle, but I find it weird. If you do it for half the time, I. Think honestly, like I, don't think span is more important than memory. Memory is more important. That's right! If we do it for what we should do or we should do it. Neither, but.

C

I'm saying that there are scenarios, there are many scenarios where you want to kind of cast in quotes, cast array to span and then call methods and for memory, it's quite rare big for the reason that we were just discussing. You don't call methods on memory. You shouldn't be doing this. You should call them on span because they are more efficient, so usually like in a local routine. When you have an array, you don't cast it to memory and then start calling methods. So what? What kind of message are you calling? Okay, like.

N

It's a missing.

C

Method require memory right well, but you'd, like you pants them, you don't like inside you. You don't get an array into a nice.

F

C

Memory and then start calling methods on memory doesn't make any sense.

G

Austin Corral's: ask you imagined some roll around. You know equivalent operations for implicit casts, but don't we already have one in spans constructor, which.

I

One, a constructor Kosuke has implicit.

G

Cut well, no thanks fans, there's a constructor for span that takes an array and it's exactly the same as what the implicit cast us yeah extension method is the same as new memory of.

H

Array yeah: what.

I

Was the question.

A

Sure we already have a replacement for the implicit cast instead of doing OPR sprinters that they could just new up this man.

C

Yeah they can, they knew a person and they have to pass three parameters at a segment. That array that offset that link.

G

It just takes the array. Why can't that would be? Is.

H

G

H

If you have the array segment of t, but then that would just argue that we should just add a constructor. That's a very segmented T, sir, if you're.

G

Happy right segment, a you can pass, I mean.

A

What is true, is you? Have a generic like a conversion method is nicely because terms, they say the type again I, don't know what would have decided to type anyway because they infer it or something I.

G

Mean honestly, I I would almost prefer to remove as spam than to add as memory as a read-only memory and whatever else to array an array segment.

H

I'm willing to believe Christoph's belief that memory will be relatively rare and span will be relatively confident. Therefore, adding.

C

So to me it's my argument is not that one is rarer than the other. Calling methods on memory is rarer, for the reason that mine is to get a span out of it. It's basically it's a factory for spans. Therefore, why would you first create a factory and then get a span out of it? You can just get this fan into three squares and.

G

What we're talking about the as memory once all right, correct! So the case we really hit that all the time is you're calling a method and which there will be more and more that accept a memory, and you have.

K

An array yeah: you can just pass an array to a method that.

C

Accepts memory to need to do Co has memory unless.

H

You're just trying to give an analyst we're all constructor call as memory not slice, but.

G

Chris, if I can do the same thing with span, if I have a method that takes a span and I have an array. I can just pass the array so so.

C

For calling methods that take parameters, I agree with you. What I'm saying is very often there are two different groups of scenarios. One group is what you mentioned: you'd have an array and you just want to call a method that accepts either array or a memory. Yes, both of them work the same, and we don't need an API state. Just we have castes there's a second group of scenarios, which is you have an array you want to wrap it in something and then call methods that don't exist on array and either don't whatever.

G

C

G

That's that's a scenario in my experience, thus far, it's very rare for me to to use to use either the span constructor or as span that just takes. You know the single array most often.

O

G

The cases where I'm going from the rate of the span I also have the offset and count I want to use and so I'm forced to use the constructor today, because there is no a span that takes well. Nobody.

I

What you want you to do is you do a span and slice exactly.

H

And that's why I argued at a previous meeting that we should add if we're gonna keep the as span, we should add an overload to it. It takes the offset in the link.

G

ha Span and slice is more expensive than just using the span. Constructor your balance check twice and there's there's multiple temporaries, which we've already seen or hurting us with the jet zeroing. So.

C

I thought we've done like the experiments 5 times over and it's not true that they are expensive I by the way, I was arguing that we should not use first expand and then slits and Yun keeps saying that they are equivalent in performance. When we have one method or two and then we do experiments and yannis right, I will does pretty good job. It's not a sin that you, like, you know stupid things twice in a row and then just gets rid of one of them and.

A

This other one, some high level thing for us, because I hate it to me I really with the principal amount, putting extension methods on fives but I'm like its original purpose types. But I can this feature area like I'm, not convinced this actually would be bad to begin with, I mean if you feel that super controversial like we can just fund the extension methods for now and see you every landing and add the platter, but as a principal it strikes me as like.

A

Arrays people don't call methods on arrays that much so it's not like you. You know thought into something that you're surprised. When you see there, it seems logically erase and span and memory they're, all related, they're, more less the same feature, they're all types of the same thing. Maybe they use a race today.

A

Maybe they should you spend a memory depending on what they do, I mean they seem related versus some of the stuff we talked about here, like extension, methods on the street, Spano buys all of the string that one I will see more of a problem. In this case, it seems like it raised substantially low level enough and related enough that I, don't think I'm just generally hesitant to do like almost a probabilistic approach.

A

Where we say you know, half the API is on the other half or not that that seems like Wiz I don't like it is because it makes it harder to grow the feature it is just based on all.

A

But it's still like you know. How do you reason about the teacher you think of these as peers and if so, that also.

F

Here as peers in the geyser hood I agree with you, this is the main.

C

Negative of not having these methods.

C

But is so that we have three choices right with our one choice. We just add these methods. We'd have them both ones on. You know, took us to span and took us to array. We have a choice of just leave it as is meaning.

C

There are methods to cast to span, but not to memory, and then we have a choice of removing this from span right, I'm, really not a fan of the lot the ladder, because unless we do what Jeremy is proposing, which we don't have a cast to span, we have a slice on the right, but we've been there. We changed our mind on it. Unwinding it now would be hard because the pattern of toning, like men, you know like first you have to cast before you can slice inside that.

C

This is the pattern that you want, and there are many ideas that work like this.

I

Think the point was to make you, for maybe the console, slicing and slice is free. This is the dryer coach, but ever you want you slice, maybe.

H

But like I find it extremely annoying the only time I'm ever calling as span and so I follow it with the word slice, because I tried to get at my array.

H

So I wanted to say as like the fact that I'm writing the slice again means I'm, probably go into my project start just adding an extension that the DES fan takes the count. So I don't get rid of here.

C

So I I think that we do have a bit of a problem. The code is very busy like I agree with Jeremy when I write code, it's exactly the same, it's always as found slots and then you're trying to eat. Like you know, you want code to like when you read code to just have information about like the like what the code is for denim.

I

I mean that, like you, cook, Ian, Paice, you're, sure, they're local right, if you're slicing over over again or if you're, taking a span insider argument as an argument that would be on anokhin method that you were sliced once or yet. You can dispense once you know there do a spent slice over and over again have you asked that we got once great well.

C

No, no usually I, think the most common parent is, has spanned slice, and then you keep slicing the thing that shows that from the person so.

E

Even then, it's.

I

C

Yeah and but still the code kind of looks a bit too busy, so I think it's a problem now I, don't know whether the fix is to just add as memory, but maybe anyway anyway. So those are the three options like.

A

A

I'm more concerned with consistency between the two I say F in having a regular exposure is worse. I would say: let's either at that I mean.

C

So I I don't like removing the one from span the other two options: I'm okay with either I, would slightly still I think if they're not where they want to took us to memory, because I think the scenarios are almost non-existent but I don't mind.

G

C

Didn't steal you kind of implied that it's over your dead body that we had eight custom methods to array out.

G

Of my dead body, that's that's extreme I! Just I! Don't like cluttering up core types like that. If what what namespace is memory extensions in.

O

C

Get it by default all those methods, yeah.

A

I mean I, like all those is also stretch I. Have you already killed half of them right, so how.

G

Do those well those I, don't care about ones on memory or read only memories, much I'm concerned more about the ones on array, five and Ray and I dot into it and I see a bunch more stuff, but whatever I.

K

A

I think copy two seems fine, because it's over, though you don't actually add a new member, so the actual new member is as memory as we'd only done we're the only two members but.

C

But the copy tool, all the other ones. We are not adding correct. We said when a.

G

Sequence, equal.

C

You you want to get spam first and then call it is.

D

A

You think we're index off like this is not a conversion right, so like I would say, memory is not.

H

The same boat as any other.

A

H

Multislice back, what's.

A

Memory still contagious, yes,.

H

Because there was the other thing, the disjoint object.

D

Memory of T is exactly the same as deep span span. No.

C

D

C

Yeah but it is.

D

Memory Jackson make sure you can put them well.

C

Memory can be other, he I.

D

Know Americans.

C

Well span well, they're compiler tricks does.

D

G

C

Have a bit of it, so we do have a bit of a problem. But it's unrelated to this review that you can put saucepan on the heap some.

F

C

Clever coding that I will not, while.

D

C

Recorded people.

M

All right so which one we want to so like we said like, would move feeling clear.

D

We would happy to feels a lot like a conversion, so nice, a.

A

Copy I can see, on the other hand, or copy tools like you were ending an operation. You know starting one right, yeah, you already done with it essentially at that point. So that's why I'm I know.

F

Compared to working over well.

H

Yeah, these are fine, maybe I mean, but isn't it just going to say spam that copy.

D

You were trying to reduce that future span data copy. How did you get the memory to get back out from? He has a passage into constructor where.

H

You can well, so you have a memory when you're trying to copy it to another memory yeah, so it would be bender.

D

H

That opportunity- that's there mm-hmm.

I

Talking about clever or like that, I think that a lot of cancer- this is with this copy and then there's there. So, yes I, we could say.

P

That this one is that we're doing it because it's compacting and allowing.

C

P

C

Many of those would be compacting, I would say like one of the principles that somebody proposed is that if it's a conversion in the broad meaning of the world and it's fine, otherwise, it's not because we do have many of these types and like, depending on your scenario, you want to kind of convert, move data between them once you like start doing like something with the data like the index off and sequence, and so on, you should use expand. That could be the principle.

C

A

Whole thing that seems fine right, which is conversion and copy to above, like universal things right versus the other ones, are literally messing with individual elements. Well, sequence may be.

M

A

Well, I think equally cool is probably in practice rare enough they're going to spend on sequencing, cobalt. It wouldn't be. Today. Do we add.

F

D

Because sharp to another, they specifically the first sequence, equals for their things. Well, how do what you call it meaty sharp I mean we don't have all the double equals to it. Quite this.

A

D

Language, the sharp language uses these sequence equals in some cases of language like compiled code into sequence, equal calls they usually they had the other using interface. I. Remember we had to add some of those methods for tuple for the thing.

M

You talking about structurally.

D

Credible, which is the racial.

M

Okay, so like this one already we so this one is easy. So what about sequence? Equal back? Let's.

F

A

Think it's polymer.

F

Q

F

Honestly, this I could look at that set. That seems sound so.

C

One additional thing that I would do if we keep the methods and it will also highlight the benefit of just not having them at all. I think we want to memo it to extensions types for both span and memory. They are close enough. We should put everything on like you know, span extensions yep and that memory is a kind of span. Can we yes.

H

C

H

It like the complicated ooh, slash, not frozen, so like spending memory already said they have one feature. The other one feature expanding census, lock dish it is like is the span extensions class itself. You know in a bad staff.

C

Of doing that, the memory treasures that type is no, it is some kind of magic types in system namespace that nobody ever uses, because the only thing that they are using is extension methods that appear is instance, methods on these types.

C

You know like from a c-sharp language perspective extension method, holder classes could be like hidden like well.

H

I mean like merging them also has the benefit of, if we're going to add things to them in parallel, like you're, not like together, you edit the one file to do both things. Instead of it always be editing, two files and they're always looking like. Oh did you do. The other I mean.

A

The programmatic benefit is also that you see them logically as overloads.

A

So you see the tie like if you didn't separate times you kind of have to like do. The combination in your head ya know season developer. Only us yeah, that's what I meant.

E

C

Or we can just like remove all these methods and don't have this problem.

A

Yeah, but you were the one saying you don't want to get rid of so.

H

If we're getting rid of method, why would we keep a memory copy to memory but get rid of array copy differently? Because an array is a memory and is a span and so for.

C

It's fine, I think a copy to emo said it will show up as an overload. It doesn't have like.

H

That's not joy, become a skin on memory of T. You know like you can drop to the span he's saying.

H

See, if you don't solve the pop your system, you.

P

G

But I'll go back to.

H

This previous principle and like just grab the spans.

H

You're saying kill that one as well then, and then I'd kill the instance members to like essentially Christoph. Certainly your archives and Christoph's earlier argument. Ire delete everything on the screen because it's just faster to go through this band, or at least it's more informative to this man. Now the other argument that C sharp is a very descriptive cluttered language, and so are these worth big disagrees.

R

Descriptive enough to to be worth adding the extra code yeah so so, who five the world motivates adding these members.

I

The buck appears were added because why to have parity between span and memory, I think emo in our previous description mentioned that we should say feed both us on the same. These will versus for completion that all other API. If he hasn't spent transition memory as well and as memory ones were or F sure like. If this explicit operations for implicit.

I

So copy and secret sequel etc will all be pocketed in the same reason, which is very deep in Spanish abs. Take, for example, somebody mentioned how they copy from memory to another memory and because the codecs exceed the API is the solution was can expand. You can do it, so maybe that's a learning curve, but they might show how to do that and the way to do that. Like.

H

I'm personally, fine with the we always make them look essentially the same all the thing everyone's doing is saying you know internally, it's dropped to span and do the thing like that's fine. We just need principle in the principle of saying we're, not gonna have sequence equal cuz, it's faster to go up. This fan says that we should do want this.

C

So there's another one which is: if it's a conversion, then it's okay. Otherwise it's not but.

I

It's just one is truly about performance like Khan, expand again and then calling these methods it's going faster because you gonna have it once, but for copy and some of the other methods. You probably only do it once so. Maybe I show that you have those.

H

All the copy to memory is written as its span back to the station.

H

C

Not a negative performance, the negative is, if you keep calling these methods in the same frame.

C

The Jade, of course, will in line the inner cold, but it will be calling get span multiple times and get span is very expensive in, like relatively it's expensive, to get a span out of memory. Looking at like span.

A

Extensions, if you just merge these numbers over that it would look horrible like I mean, like the the the type is already pretty much a grab bag of things and.

C

Yet has chain as well, okay. So what about this? We keep the copy to and try coffee to on memory a Steve mentioned. They are on memory. So that's like they are not it's fine and they are like conversion so like moving data around.

C

If we want to have copy tool that extends array, I would just add it to span extensions, and then the s memory has read only memory methods. Can we confirm from f-sharp team that they don't want to use the constructor visit.

I

I think I even open report come.

A

Up coming, the problem is like even in c-sharp all new constructors is you have to specify the title if you don't get any sort of benefit from generics.

H

A new memory of light.

A

Which gets annoying very quickly because you're right, the tough stuff versus you say, the dot is like it's very short. Yes,.

C

But just so you see like I, don't even remember, and it called that I wrote and I needed this method. Sure and Steve who wrote a lot of code, probably even more code that uses memory of T than I did also kind of doesn't like adding them and doesn't think that is worth it sure.

G

Also, it's a valid argument around the type T that said, I would expect that 99.9% of code- that's written, uses these type. These types T is either into what is either int char or byte.

A

Or think I'm not I'm, not saying it's like it's it's just annoying. To spell all because you have to have the break, is your spend, you know specify it in like it's just not nearly as nice, but yeah.

A

So so that means we actually do this one you these guys, the other ones.

F

I

F

I

The accession class to like some.

C

Of the more general you want to, you think that we should rename the existing extensions yeah.

I

Do not just by officials.

C

A

Could call it buffer extensions.

J

Work, there's anyone pay attention to the name of the extensions class, no like.

A

He's never visible in user code. What is visible? It's documented, you can directly call it so.

B

A

Mean Europe I mean it's not that you call code wood but like Italy and other things exposed items all the time. The.

B

Whole feature area is called span, that's my hand, wavy explanation: can you market fascist, editor, invisible this.

P

Is the discourses I have no idea.

G

What memory extension.

P

Extension classes, browsable false I.

G

P

You're, actually not a crazy idea.

A

Because the thing the way I know it is a cold, wet stuff.

A

Which may actually be not true, like.

D

So all the reference in was generated with the dish Mike.

H

Eristic, not adding something to memory is, was something like sequence: equal. You think somebody might call it two or three times it could. Therefore they should have snapped. This fan wants to avoid the conversion.

C

They're gonna break anyway, so what's the name of the extension type memory channels, so we do not rename the call type code span extensions to memory extensions, but all the extensions are for span no.

D

He said merger was fan extensions. No, we will now merge Spanish, since it's insist.

C

It's a bit weird, but whatever Theory.

M

C

In the racing extensions, it's a nice form of memory now, okay, emo is arguing that already some method.

D

I bet money, there's no web Spanish census operation. Oh well, because all the.

H

Devil breaks unquote, like all the calls are to span extensions like you're calling it is an you you're, calling it as an instance method, but it's really a static, coldest panic system.

D

H

Next week saying, why is this this package update.

D

Not work and I'm gonna be like because you've made names for any extensions. In spite, we should I think.

C

Another reason to rename it is Steve has been asking for us to design extensions on spans of char, like, like other extension methods and I, think that memory extension kind of can contain lots of extension. That I think we will be adding more and more extension methods and it's good to have one type. That is a grab bag. For you know similar extension methods, I.

P

Am a bit concerned with some like break state I. Would love to see your codification of that.

P

Type for all your exits that you ever everything can be rationalized after the fact.

A

Of the extensions that's where the party is alright, all that's contains.

D

What's in the room class.

R

There's overlaps yeah.

C

This one is so I think we need these methods. Methods like this I just think that a people wanted to know more complicated things and whether to.

I

There are several things actually.

C

I

More than this style destroyed all of the options Shoal down this. These requests covered all six scenarios, a to e C.

F

I

But contains the array like the name because it sometimes implies like the contents are contained, not span itself is.

J

H

Oh, the purine Shetty I see second from.

A

The table the table intersection.

H

First is second.

I

C

But you have to call potential, you have to call to api's to be able to tell the cases. Are there two.

I

Separate scenarios, one is with let's hear from full moratoria to the span back, the very back. You want check if it's empty or within the other. One is like for copying and scepter. You want checkups overlapping, so.

C

They actually say I want to know whether I'm in situation, D I, have to call both overlap.

N

And within no no at these scenarios can be inferred from either from either the.

H

Overlaps is true means you're in BCD and that F, G or H you didn't need to call within to figure out which one you're it also.

G

H

Are guys different letters, but they have the same truth value, so you can't tell which is which.

I

But the combination of all of them is on unique. You have like Indian, I am J. Both it unfolds right. It's not I'm, not saying that it's not overlapped everything. What don't have to be called for you to figure out what scenario where but they're, not sufficient right.

H

Well, I mean what they're saying right now is that you do need to call both of them, and you need to call some other method. Don't exist. Well, if you need, if you need to care like, if F will destroy your model, you don't know if you're not sure, but I can.

A

Both in such a solution, because your problem, because I mean do we even know that you need to have this as an atomic operation like how common are these scenarios? Do it on that, because I mean obviously things like overlap seams I can see that being very common. The other one.

A

I

Calling within allows checking for which buffer the DMS span is within all.

H

Right so I guess a whatever aid in the anti AR are irrelevant, yeah quite a bit. Well, a is. This entire span exists in before that entire span and I. Don't I can't think of a time that that's useful somebody who's, not a hacker. So the old crackers I can't think of a time that special to somebody who's not trying to actively break your code.

H

Simply knowing these are have no impact on each other I.

H

Mean I guess maybe you complete their content, that they are some spare so.

C

What is the other thing? We wanted a method that like gives you all this information atomically. Is there like what would the design event be, because maybe we have no alternative? Maybe this is the best we can do any as there are. You have to call it twice and there are some.

A

C

Flags enum would have a table and MSDN that shows the ABCD the options.

H

I, don't think it would be flags, AJ or whatever separate states. This right, I.

A

Think everything I would think that there's probably an indicator for how the Stars relates the other starting out the end.

E

I

Have the element of second element upset all right for that they have far.

C

Into the first planets the other, you know what I have an idea: what about the following? Api, you pass two spans and it returns you to ins that tell you the offset from between the begins and the ends. I.

I

Think a ball that was proposed by somebody else, whereas you have to Winchester 112 Penrose me and they.

C

Are in they are literally in so if it's zero, they are. Basically, this starts at the same location if or you some hair star.

P

If it's minus 10.

C

P

Number right so.

H

What, if so, you can definitely always do anything artificial. You feel like doing if based on pointers, which y-coordinates yell at you and then, if they're, both back they're, not if they're, both backed by an array but they're not backed by the same array. What are you gonna answer.

C

What do you mean back by array, but not the same array? Then they cannot overlap, see right. So what are your inception.

H

So this is array.

C

1 yeah: this is array all right and it will compute the offset from this to this and that's it would be 1,000 so, and it will be all from this to this, and let's say if the length of this one is 100, but of this one is 200, it will be.

H

So but so you're gonna drop the array to pointers, because I think that by the time that you're asking you compare both back by managed to raise and they're not the same managed to raid, the answer is question makes sense. Well, does the current take the ID balls? It.

R

Does not overlap different different arrays.

L

Okay, I see if the reference believes it moving around right, so yeah.

H

And then then GC compaction happens and now the pointers are pointing in in different places.

I

They don't store.

H

I

Isn't the Delta Kondylis be large and it should be long so like.

P

Yeah yeah, so I, don't think I, don't know that we, you know, would actually give you everything.

J

That you need it does what if you ever starts within an end with it and then overlaps is starts within, but not ends within, or vice versa, and within is both give you both a everything. Workbook loss.

A

In their text span, which is basically an offset in the length determining whether one is opening or the other distribu, because he just happens, but if you write the conditions, which is what you still have to write with these ins, you always have to stare at this like for, like 30 seconds to a minute to determine, did I get the monster correct because it's not intuitive. It's like oh star, less than like it's or less like you read the conditioning like it took me. Always five tries to get the overlaps into secresy.

C

So one thing we could do is, for example, just hyperbola, it's very easy to understand and then have this method for all the other cases.

A

Because I think so.

C

A

Like it, when we want to talk about, have to have the message I mean like people could do it to do, which method like what you would do is basically a tribute you just you just extract the pointer into the Delta right like how hard is that for the caller. It is hard and it's unsafe code. Oh we can. I yeah so.

C

The the reason like one thing we should talk about this with the scenarios are so overlap. Like the moment, you have talked elapsed, pants and you want to move data between them. You need to be careful how you do it if they are not overlapping, then you kind of like you know the code is you can write naive code, then within I, wonder whether like what the scenarios are even.

A

C

How many people have to write that code? Because, wouldn't you just call copy? Well, it depends what you should do, because sometimes you kind of we only have copy methods that copy like involved. But maybe you kind of want to, like you know, poke at some indices.

H

Handles from a1 to any of a two through five, it completely handles like oh you're, the exact same array and the exact same destination.

F

I'm going to do.

H

No work and then it handles like oh you're copying into like two bytes to the left of you. I'm gonna shift everything that way and it handles you're copying everything two bytes to the right I'm going to handle app. So these two cases you definitely need to know, and then certainly it can just pick an algorithm between these two of like a you're off on your off and never never let that I don't care what you do.

H

But, yes, when you need to know the overlaps, if overlap says true, you you probably want to know, is it left or is it the same? It is a lefty overlap. Your righty overlap see because that's gonna change do next right or left like that, so I, just so, I think that overlaps with overlaps returns false you're in a good state. The moment returns true you're screwed, unless we give you more information for a lot of that's what I was.

C

Arguing that's like in majority of the code that I imagine people writing. They just want to know whether it overlaps if it overlaps, they need to make a defensive copy and if it doesn't overlap, they can call like base64 encoding. Like all the like transformation methods, you can just really call them the moment it overlaps.

C

Yes, you can be very clever about your algorithm, but most people would probably just do like a defensive code and I mean that defensive copy, like a stack, our log transformed to the stack, a look and then copy, so they don't have to worry about overwriting data, the other old cases.

C

Maybe we just expose those ins and then you know tell people you are now dealing is with some very complicated, maybe unnecessary.

H

Right and when we have the myth, the the safety of the managed knowledge that you're talking about different arrays I, don't know that it makes sense to talk about how far apart they are. I, don't think once it's, if it's a it's a pointer, then there's really nothing. You could do, but.

C

I, just don't know how to design an API that like returns, you you know something different, then those integers. If they happened to be an array, I mean that's even more difficult, API to use I mean and if they have managed right. Yes, the integers would be very far away from each other in the algorithm. Your higher level algorithm will tell you, don't worry. Yet they are different memory locations. You can just don't transfer. Maybe.

H

Any more about, why don't one answer is you ask, you know, try to get deltas or whatever it's try to get deltas and it returns false if they don't overlap and then out zero like I'm, just not answering, because no.

I

But I think that's what the parameter would move over lapse rate. We return some non like a zero or something well.

H

Like I, don't know what to do with element, I guess.

H

You have like the say us covert. I, see that's fair, because you know the length of both things, and so you just need to know how different their starts are right, so so overlaps false, always out zero and- and if it's true it outs, the left distance and you can compute the right distance. If you need it so.

I

Then see it says, element up says only valid to take a question returns true, if it's flawlessly yeah. So then within.

H

You could compute based off of overlaps outing. The number.

H

Because that would be if second dot length minus element offset is less than first stop links, then it's with it I think I need to look at the subtraction formula, so easy.

C

For you guys to understand, like you're, looking at the API that returns a limit offset like what this number is, I think.

S

We we probably want to rename it.

C

I, don't because mean like this, like those two integers like yeah, it's like you need to see a drawing on the board, but after that and I think it's very clear, right, I think that would be a madness to it would be this number ready.

C

H

No, you know this number with a bully and in this case it would return. It would be zero. This is nothing yeah and in this case it's you know. I like and true of this was nine positive. Nine off of this one I see this one is negative or okay.

C

I

It's basically how far the first element- yes index of the start in seconds, man of it.

H

Yes, I would call it relative offset instead of element offset.

H

hmm Because you're not talking about elements you talking about well, the start mark is okay.

C

So you basically can get this one, and then you can compute the offset of the end by just computing, something with lengths. You.

H

Have the links? Yes, okay,.

A

Okay, well, I, guess. The reason that said element is because it's one number by its red its number elements in between them. So.

P

A

About blue is a relative.

H

C

What about within.

H

Is within within also you can compute correct within would be is, is it a positive number and if B dot length plus this is less than 800? Okay, then.

C

We can just add those two overlap: methods.

H

C

What does within do? Does it say this? Is we don't need it.

H

Someone can compute it given just the overlaps that returns the value for.

I

Bidden was there different, optimization paths for the dimension of with it, so it will make within faster, and you can make this in faster than overlaps. So if somebody only cares about with it, they call this method, so it is only about with it.

I

There you already transplant back any one of its buffer. It goes into that part which buffer from your buckles I do so you just want care about me. Then you could keep going with it quickly and you get your response and you didn't return it. Why.

C

Would the overlap network slower.

I

Input it could be slower that the denominator would be different from this optimization I, say C, so.

C

Slow because you so check whether.

H

It's assumed, like would be sure, yeah like how can you write within faster than the same in the first quarter and then X checking out you.

A

Have to compare both sides of the buffer right, they can and I think if you have a buffer pool, you know you never have overlapping.

F

A

Know you don't rent offers that are overlapping right so, like give.

H

Its overlapping, its with it I think you know that said Italy, because you can't just say if be downlinked is less than eight it like. If be downlinked is bigger than it out links. You know it's not a or it's not nested, but you don't. But in order to determine the true, you have to go through the all the same work as overlaps, so the false path can be faster, but nonetheless.

I

If the length is not, if one length is smaller than the other, you just proved one check. You'll.

D

Be able to do the link checks first, so it's basically the same thing. You do it after you to do over that. Yeah.

H

Turns out if somebody wants to optimize their within that way, they can check the links. First, okay, like there's magic, that we can do because we're we have the span and the pointers and stuff I can't think of any way that you can do it without a double bounce check so and.

A

If you presume the answers already to a number of cases, but in which case you she then has a past.

H

False off of overlaps, so they can't get to true any faster. All.

C

Right so what about this? We had the first two methods and then, unless somebody provides a concrete like justification, why the second one is needed yeah, it seems like those two do we need the first, because we are.

H

We just happy without underscore for don't care about what well c-sharp added underscore for there's enough parameter here and I. Don't care, oh, so what is it could call a method without? Well, if you have to write on the overlaps, a I don't care if.

R

We think that that's ugly, which it kind of is, oh, you mean.

C

R

It the overload that doesn't take that parameter. Yeah yeah.

C

I think because, in a lot of cases you just interested whether it overlaps or not, I mean that's the majority of the case. Okay, if it overlaps, I have a problem if it doesn't I just call methods. Okay,.

H

C

G

H

G

What our current framework design guideline.

H

Rules are so that we're consistent and then what do we think about the names of the oh?

H

We had first, second element offset so I think element offset should be relative, offset.

C

Either fine sounds.

N

R

N

But first in second take one.

D

This morning, when I'm chefs about to start it's clear that its basic.

D

Dental start start off.

G

C

But it's an offset of either start or or the second or the third. They all did have the same, offset this sharp invaded beginning. So to me the main difficulty is understanding. Is it offset from first to second dollar from second first.

P

May be, the name of the parameter should be just the equation should be first, second, all other, so I think yeah I think.

H

It to me logically feels like second first and then this would be if I do. First of I, plus this value I get the exact same value that I would get a second to buy because it is the relative offset of their outlets. It's.

I

On a sec, look good with things overlaps so.

H

I think it's just relative offset.

N

Oh, this throwback.

C

So one additional thing: if we still are thinking, brainstorming, the name of the parameter and final relative awesome, why is case I and J, not something that you can distinguish with this API? We look what I am jr.

E

D

H

They are empty. Well, it's we don't overlap, we don't tell you if they don't overlap because and the non overlap is before or the not overlap, disaster I think that's because a and E I think not overlapping yeah. So.

M

When it doesn't.

H

Overlap, we don't tell you what the relative up positions are, especially if it's I managed to rain. It can change.

C

So why don't we always return instead of like when we return false? We still give you the awesome, because.

H

I, don't think we want to tell you how far apart two minutes, Teresa they're not same, why not cuz? Why? What? What can you do with this information.

A

Well, for starters, interesting we move and do the conditioning your code or you just compare their efforts if you don't care for the whole father a party. But you cares about the not having to deal with the fact whether they are a part or the just write the condition, but.

C

That's truly needed it's because the boolean is we return, the boolean means today overlap or not and the end we always return. You I.

E

Think it's I, don't think I think it's gotta be within the same thing, because.

I

The point is, it means nothing for non-overlapping. You got make sense of it. No, this would solve this problem, but why do you need an ambiguity of the other case right? What.

C

I

Is the case where you need.

C

To know that's.

I

C

Good question I'm just saying: I the only main statement I made this. It would solve this problem. I'm, not sure whether the problem is a has a practical, I believe.

H

You don't want to so I do not want to solve their album of telling people with the current layout of memory is when the whole point of that that is shouldn't know. Unless you're dealing with a very very specific thing, you don't expose the actual.

H

F

C

Oh yeah, you can cook some medicine, okay, fine, that's actually I buy this argument. You can, you can distinguish, you can distinguish between I and J by basically pinning the the thing is yeah.

H

And then your answer is valid for as long as it what you really want to be created, you could just return.

A

Minus one and plus one and then.

H

Just you get the yeah I think so, because the offset I think if these wasn't a negative, if the two arrays are not. If we have a raise on both do not reference equal, we just return false.

I

You already exploding both negative positive on the range, so you can't use a negative one, positive one for that. I.

H

Think it overlaps is false. Relative offset is zero.

A

Return false Oh, false lashes, because then you think that now we know you can use.

H

If somebody wants to complicate, we know the layout of memory of a reference object. They need to write unsafe and, if they're, in a place where they can't write.

C

So another argument for always returning the offset we don't have met it like the parameter is called the relative offset. If they are not overlapping, they still have an offset. Why would we suddenly, like return, zero from a parameter.

H

You can't pass it, you can't do a I plus this option. It's also only like for.

S

C

Know for a try methods. We basically say that, like if the method returns false, then we do like special things with other output. You know all parameters, we don't do it for normal methods. This this method is like are they arise, overlapping and by the way, here's an offset. You.

D

Know if you do that, you can even necessarily get any velocity. The size of the elements are different size. Now, an officer isn't both.

S

D

Of the tea is the same. Yes, you got two different unmanaged things you've dealt. It may not even be a multiple team. Yes, the pursuing that starting fortitude on related.

H

Memory points that offset may not even they're, not necessarily pointer, aligned and there they.

S

G

S

Kind of think that it can be not a cheese, it can.

M

S

For even a lot of a lot e ones, unfortunately, now with the shame, T yeah,.

C

With the same T,.

S

D

C

You can basically cast it to cast a span of of T some large struct to span of bite slice by one and then do dangerous. Castis span out this chart correct. That's why it's for dangerous, because you just what will we then return from this method?

C

Well, you give.

H

C

S

Semantics, like.

C

D

Already perhaps.

E

State or any of the spending we.

D

Know that we have bed.

E

Space I'm just saying.

D

I need to decide with.

E

Between this case,.

I

Right so, regarding.

E

I

Well, the special will be with Troy for if there's misalignment between I think a few of the time, I measure.

I

The first, if ii spent, is ahmed line of the first man I'll get restaurant.

H

I think it's so that Christoph's question of don't call me what you're doing dangerous stuff if you've intentionally shifted into the stupid land.

H

I

The point is you can't answer the question, that's what you because.

H

You're one tenth of a value offset from the earth I'm.

C

Pretty sure there are scenarios for like casting an array of in-store, I mean spinal pains to so.

H

What is the question.

C

So you have a span, it could not be a whole number if you compute the.

C

A

C

Usually has to say like: are they other? Like? Is the.

M

Big, you know you, can you can.

C

The boolean you can always reliability. The element of set is weird if they are misaligned. Yes,.

C

Persistence, just don't do yes, like consistency, it basically tells people don't do these things that Jeremy called stupid land or something like you cast this panel. You know some struck to span of bite, then you offset by one bite and then you cast it back this panel. So the reason why.

A

The only thing.

H

You care about it's like well that if.

A

You took a span of T.

H

Where she is a multi-line thing and then you turn into a span of pipe to intentionally misalign I, don't care what.

C

This, it's not, you, don't have to be a careful when you slice, you have to be careful when you cast from span of bytes to span of something larger and that's why the method is called dangerous cast it's actually it has it in do.

I

We have a dangerous cut or you need you just create no.

C

No from the to cast bronze panel by to spell something else on put.

O

C

Non-Portable sorry yeah, it's not called dangerous. It.

F

C

Way, throw if you ever did up cast kind of up in a sense that cuts to something larger, so.

H

You move a struct into bytes, you move by a non multiple of the size of the scratch when you turn it back into a span at that struct, the relative offset just doesn't make sense.

R

Was a nothing since? No? No, we could return the float.

C

We need a national number action will not give you the right answer all the time.

H

Anyway, so the reason that I don't think that we would want false to try returning is in a 64-bit process. Two arrays can be more.

H

A

Could return in law right if you really cared about it like I, think what is argument it stronger because, like there is no number, you can't return really, because I mean they could.

D

Be two bytes apart in you not even intentionally try to do anything harm. It's got two different reasons. Where did you see put them? There's not aligned.

H

Well, I mean obviously pointer a lot folks. I. Just think that we would just say that, if we're returning false okay.

C

If we have a run into limitation here, we can, we can do something.

M

C

It offset relative but kind of isn't that the definition of offset okay- let's call it just didn't.

I

The relative offset, but the point was to make element rather than bytes, but also community never bites being offset yeah.

H

C

I

H

C

This is a good point, but.

M

A kilometer was called.

C

Element offset so it's clear that it's in that in T is the currency not bite. We could call it byte offset and then we could return it over. That.

G

So what was the answer earlier about what it means in terms which is first yeah? Second,.

H

D

That's not how they PS. When you do a subtraction, you do it. First, you read the docs.

C

It's a second first.

A

A

How would you implement it, how will you beat it like? What would it be if you say first dot within second mm-hmm.

F

A

What does it mean? Is it logically equivalent to first contains? Second, no.

C

A

I think, if you think about it, not a purely logical way to ask the question: it seems backwards.

F

A

Like almost all, if you guys we have dinner around, you ask the larger moment that contains the smaller ones.

I

Write the name tint cause the flip exactly.

A

But like this I'm saying, like the only the only reason that the name change was to reflect the fact that we don't okay, do you have like.

Q

A

Still do logically, the same check you ask whether this this.

Q

H

C

Is different, I think.

P

Yeah, it compares.

C

A

Two of them were tested for the teacher.

A

All right, thank you. No I'm, happy with which thing the comparators return. Sorry compare this.

H

Comparative tb- area might be.

C

If first is smaller than the second, its minus one.

H

So that's a mighty.

H

And because if yeah.

A

If a is 0 B.

H

F

H

F

Poli we dropped off. Of course we got it off again.

H

Yeah, the answer: is you? Try it once it returns the pump sign your program crashes, you flip it until you understand it. We.

C

Don't crash that easy you're on a program just does something else. Well,.

H

If you have good tests, when you're testing.

H

The overlap is correctly, sir. We had this from memory.

I

Fk the last method: yes.

C

That span overlap.

C

In fact, implementing it on memory, any other way would be so tricky that we wouldn't be able to do it in any other way.

H

C

Because memory can be like many different things, including an abstraction that gives you like some numbers that don't even exist, I, don't know, memory is an abstraction. It could not be even in memory. So.

M

Let anybody follow this.

G

M

G

One I can speak to this came about because Tim had put up a PR where he was relying on memories, constructor to validate the array, offset and count arguments which makes sense except they're named differently. So if they fail validation, the exception that gets thrown contained the wrong parameter names, so I had suggest that he should still be validating at the call site, the the inputs to throw the right exceptions. But then you know we're duplicating checks, which is unfortunate, so this came about as well.

G

Should we rename the arguments in memories, constructor from array start and length to be array, offset and count to match sort of the predominant arguments used in similar situations elsewhere, or should we have a constructor or factory that a that doesn't do any validation and you you're?

G

You know you do the validation yourself so that it still just happens once or should we not care, and we just do the double validation or should we not care- and you know it doesn't matter what the argument names are you get an exception and you may not be able to correlate it. It's done the except did parameter name.

G

What was the question are.

A

You saying the duplicating checks purely to get the string for the parameter name correctly and the exception message: yep.

H

Yep or if we rename all the parameters to be the same, then it just happens yeah it seems like an almost no-brainer to fix it. No.

C

No, we cannot rename the parameter. Why not? Because we would have a parameter called count and the property we would on span would be called counts. So now it's not as DeRay where the property is called blame, like a.

A

Cycle on the cpu.

C

Every single time you construct an here, no I would like if we wanted to solve this problem, so one I think solution is to just punt this problem like yes, you you have to do the check twice and you know because you're adapting, like some existing code to new code, you have some inefficiencies. The second one is I kind of like the proposal to add a method to create span without validating, especially for arrays, because it's not like memory unsafe I would just not put it directly on span.

C

I would put it in some like bury it somewhere else.

I

C

So in interrupt services we had a class called span no present, but okay, I did not say yeah, it is very I would just call the type something else. There's.

A

A bad name which.

C

S

C

Yeah Marshall, maybe either right yeah I would not burn the name span because at some point maybe we want a span type to avoid the genetic.

F

Holy uses, aetiology.

D

You call this, you don't get any checks now.

A

I mean that the full code, like I, mean.

G

What goes so, I look at the beginning of any method today that takes you know any stream read a sink or anything. It takes a byte array and an offset in account. There's like four checks that happen up front where it says, is the array will noooo throw argument null exception if offset less than zero, throw argument of range, exception, etc, and then it says you know new memory passing in those same arguments and today, new memory does those exact same well very similar checks.

G

So if we were to go down this path of having this Dangerous Method you'd do all those same checks and then, instead of calling memories, constructor you'd call dangerous, create.

I

So here's Bechet- if you do this checks already, then why would you get the exception for many to new memory? Many call a contractor you're.

C

Not gonna get exceptions but you're gonna. Do the checks again.

C

So Steve, do you believe that, like the you know, double checking is a significant performance problem.

G

I doubt it that's what my suggestion was to Tim was to just have the checks. We want I.

A

Mean we have to edit cinema saturation and immutable array of every basically decided that the indexer doesn't know the date. The argument just index into the array we.

D

Have double checks on replacing stream to.

A

D

A

Not necessarily marvels of efficiency of today.

O

A

Mean the middle array was designed to for perfectly said you having you like, you get a slightly different exception. Only we get it next out of engine and yellow I get actually I. Think it's an accelerated hope. What was the thing? I wonder. I think this nitrous. What's not the moment, we're happy with I mean they said screw it like it's it's better to do it this way towards the double-checking but honestly indexing. You do way more than you do. You know poke operations or restraint, so I mean hopefully.

I

Those things are like I want to do that, there's other scenarios for dangerous, create that the ones that we don't have very end up, selling something check. You know other it guys. Maybe it's tribution spell you do the industry so, but we already are sure that the end excellent are already valid.

C

Yeah but then probably some other some others, it's a common problem that we do you know double and triple and quadruple checking the the question I would have is like in comparison to the cost like. How often do you create spans?

C

Well, I mean a lot right that you know the cost of these checks. Mattered, I, don't know so.

A

C

One is a bit different. Well, why did we put it there as.

F

A postal else in a dropper they.

C

Think there was a issue files to move them move it isn't this the method that is hidden, so people cannot find it I.

C

Thought there was an issue to move it.

C

No interesting I think.

Q

If someone will do 17 messages, changes I.

G

Would I don't have everything that was dangerous, whatever some span marshal or memory marshal class and there's no issue for that. Yeah.

C

Do you remember we had that discussion about this correct and I think we decided to move it especially.

F

C

In the you know, in the future, I hope that we're gonna get syntax for pinning spans in the language and then like. It would be nice to not have these metals there, because majority of the scenarios you won't have to do it. You just pin.

A

So what will we call the time? Because it seems to me I can into I've seen before it would be.

C

Dangerous memory, extensions: we have memory, extensions and dangerous memory exceptions. We move all these methods, they're dangerous memory. Extensions in system run time, whatever interrupts services or.

C

Steve, what I thoughts on this I.

G

Mean my my initial inclination just looking at these specific create methods is we probably don't need them, and we just you know, and if we find that there actually is a measurable cost to having the the extra checks, then we can add something later, but right now, I've, yet to see evidence that there is I mean you know. Part of me doesn't want to burn any extra CPU for any reason.

K

G

At the same time, like I I doubt it's really measurable in these cases give.

A

It done okay, but I would say we could decide not to do these two methods, but like do you want to move? The existing I was just to come up with a name for that. Guy.

K

G

Well, so if, for that separate issue, we might want to go look at it, because we already had an issue that we talked about about where we put these things, we talked about having some sort of span marshal or memory marshal, or even just adding them to the existing marshal class and I would be fine putting them there. The.

O

G

Idea where there was that marshal itself has sort of already thought of as an unsafe thing or a dangerous thing.

G

And I think it's number two, three, eight, seven, nine! Well it's the thing: it's the support. Casting we'd, only memory to memory of teeth.

M

It's just me, but like I.

A

F

A

As one tired, sometimes yeah,.

C

It should be just one type, but then you kind of want to have.

C

Your memory mark shot just like the other type. We call memory extensions this one. We would call memory marshal or dangerous memories. Since then they wouldn't be extensions, correct. So that's the thing I'd.

A

Like to pick, if you, if you get away with them, having well I, guess it depends what you do with it like those are taking those guys, I did what so they could go either way, but like the the dangerous ones that I've just looked at earlier, they were factories. Let's you don't know what they're.

A

And you don't know what did whether they return this one or that one.

A

C

A

The only poem that you have is it's making it one tiredness. You need to be able to tell whether you want to get a memory or you want to get a span, all of them right. My.

G

Being called span of tiny dangerous, create and be called memory marshal that creates fan either.

Q

Specify the name and the method or something make sure you have a unique.

F

A

Would work like I wouldn't do the dangerous span? Extensions, for example, because then it's like dangerous fine extensions dot danger is something and these they are dangerous to shooting the extensions. Yes.

C

So the only thing it like, but that's a separate issue like if language support. If we have a language, support this specific to span, they can call whatever method.

C

To pin things they probably will want to have some convention, how you know how the method works and I don't know if they like, like some other static method, another in another type, because it's difficult like how do you tie them together? How do you know, instance, method is very convenient. We just have a name and when you call fixed this method, which is in general, hidden only which we would hide, or it's even today, hidden on I, don't I, don't remember: I hid what or we hid one, but not the other correct I. Think.

I

C

Some dangerous ideas, but not all.

A

So then, we're saying that for every async read would be eight to ten checks.

M

So they are next an issue of checking. It's not one check issue, roof.

A

Checking is it's one to.

F

Five checks that.

A

Are done twice.

F

So it's eight to ten and what's the cost of pacing greet yeah, that's thing.

A

I, like once you go down that path. Maybe it doesn't count.

A

Q

Neighbor this we.

A

Can't decide for now we have a pattern for what we would do that we would have. Was it again memory marshal memory marshal and does it put the other guys so do we have a tracking issue now.

M

Like I think this is a syllable.

C

Yeah, what is the status of this one? Why is why are we? Why is it not showing on the list or any minute? Oh okay, where's Marshall, our first Marshall came up with that yeah, let's, let's rename it, let's improve memory Marshall, because we already said that, like we gonna have like. If we have those grab bags in this assembly, we use the memory as a previous just.

Q

Change your mind.

C

So the moment we have memory.

C

So the moment we have this method, I kind of don't mind these constructors- that don't do the checks, especially that they take erase so I assume that checks will happen at some point anyway correct, but.

I

The colors item area, if they, unless they don't actually happen,.

C

Yeah, you know the checks may not be like. If we don't do the checks in the constructor, they may not ever happen correct, because the moment an array is put in a span, its Empire.

C

We we need to make sure that the checks at least happen at some point lazily and not just like. If you call this dangerous method now, you kind of can write and read over all over the memory. Why not I mean literally called an API called danger? I'm dead I mean dangerous means, like you know, but you only had the dangerous.

I

Create right with the whole object the object exactly if you are.

A

I

A

You take an object, you convert it to a span and they done writing to it like I mean now. Is that say well with that? One.

G

Right, don't don't the.

K

G

Checks would happen as part of accessing the span property on memory right like when you go to get a span. It'll do its own checks in the span constructor on memory- yes, but not just a few.

I

Saucepan, yes, so look if you have an object. Oh.

C

And you also have length and the dangerous create that we have today you can pass arbitrary like for it. We don't do any validation of like what the object is, because the thing is with this API you can pass an array and it's empty and you can pass length 10,000. Now you can kind of write over some memory. That is not even part of the array. Yeah.

A

I can hold rain is create, pass in the string, wasn't offset ten million and then write to.

C

That so that's a question whether this dangerous create object, actually validates it may actually validate there like I, am trying to say yeah I'm, probably but I, don't know.

A

Very sharp okay, so you certificate is every single type. When you accept an object like you would have to know as a string to even do any sort of bounce validation.

F

A

C

Any bet 10 bucks by the year, hey what are the odds that you're giving cuz I'm, not willing to bet 10 bucks for your 10 bucks.

G

C

I bet $1 that you give me 10 if you, if, if I win no.

I

That's that's. How, like you know, betting works. The odds are not obvious one to one yeah, it doesn't do any checks. The only track it does. It's length is greater than zero. Then.

C

Just just scotch, which one.

A

Okay, now, what do we do with this? One we'd be punt this guy for now and say: if you would ever do it, it would edit to the other guy well.

C

If we already have dangerous create from object that take takes arbitrary.

A

So this one seems to me like the case of premature optimization, we're not convinced the checks are adding. The API, sir seems like the only reason would have them right now is for those cases which we don't believe or not, which.

C

Method, this one or the other ones, because this one is not just optimization. So what is this one for that? This is, it is sorry I I meant the danger is create that already exists. Well,.

F

C

One so there's these guys here, maturity right, there's also dangerous, create that we already have- and it's not currently on memory.

M

And I think this is this is the whole discussion on on both it went to like five iterations on this, so.

C

We move those okay to Marshall, now you're saying that the new methods that are proposed- it's premature, optimization.

C

Ban is saying there: it's a lot of checks.

C

Ban is it relative? You can see the code, the cost of those methods or it's a lot of checks, kind of in the abstract.

C

If we're this way like how expensive are the methods that I think methods that you would just call subsequently.

A

C

Ban online he's.

F

C

Him, how expensive are the async methods that he called.

A

But you just ask them right, because real life I, don't.

C

Know what kind of delay you have.

A

That is thinking back so again you.

C

Know he's counting the instructions.

Q

A

Q

A

Cases is you complete some currency, in which case you are actually quite as.

C

Well, okay, let me say: I don't mind Eddie like since we have this Marshall type, I, don't mind any and we already have a method called dangerous created. It creates it even it's even probably more dangerous that now that we discover it doesn't do any tricks.

C

You're saying sure yeah at least this one takes an array so like potentially at some point, you know like we have all the information. We are not doing checks, the other one is like taking. If you, if you add new analyzer to check so you defeat the point. The whole feature of there's not a chance because they were have. They are equally dangerous. So by not adding them, we are not buying. Much I can actually say.

A

I'm saying we should have this I don't mind having.

C

A

C

As long as they are Marshall type hidden' in either observation.

C

So a bit of a side question: can you use the existing dangerous create to do exactly this? You just need to know the ref.

C

You can pass array as the object and the ref offset. You can compute using unsafe, API correct offset to the first element from the array, and then you basically can do this using the existing API. So.

K

These are just fine right.

C

Okay, so we only have it for spam. Okay,.

A

A

C

We still have 10 minutes, yeah.

F

A

C

The proposal starts with rationale. What is the proposal? There's.

I

The example in github says rational and usage force, then tuple. If yeah I, don't understand why.

A

Well because, usually like the restaurants, I need to feature X and there's x5 on an API. So you start with the 500 api's I'm like what the you might achieve in the first place. So.

C

We need to help conversion api's Oh.

C

Basically, the memory from memory.

C