.Net Foundation Design Reviews, 16 Jan 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: Index and Range

Description

https://github.com/dotnet/corefx/issues/34076

A

All righty, so we're live again so sinking chair, I'll be playing musical chairs. Okay, when I see one that doesn't Oh.

B

Try to make me feel small, so.

A

Last time, when we talked about the next range, I was under the impression that shows up in under an hour the whole hour, and we did one review a single, acre, I think so this time you want to really focus on the API shape and maybe also text and all some things that we think are missing. So we can add this to do this, so you just have to me, but they weren't here last time.

A

So when we reviewed this crystal part up, one question which was do we need the types and all- and we talked a little bit about that in the notes for the API review. I've tried to capture his general proposal, which was imagining a low-level from entire black span, where you really don't want to have the overhead that wrench and index impose a new meaning the whole. You know duplicate bounds, checking in particular that he's concerned about the digit cannot get rid of.

A

So the idea would be that if you care about performance for span and other things, then we would always have the in base overloads anyway, that you can call and then he said like that's unfortunate, because it means you cannot use the language syntax at all. Just for the point for doing simple stuff, so he was proposing. Imagine some sort of attribute on the method or the parameter. Where you can say.

A

Well, you can add the cold side kind of do the dot, dot, syntax and the compiler base is flat, demented two arguments right and then afterwards and I said well. The problem would that would be that it doesn't compose. Well of you know, because it's not an expression. You basically have like special syntax on the call side then basically proposed at well.

A

The compiler will just codes, but the call side, meaning the value is of type range or index, no matter where it came from the compiler, just unpacks them at the call side and then the very special case where it's literally between the parenthesis.

A

You can just optimize that I'm taking away you just don't have to type at all, but everything would compose fine if you take the expression assigned to a local that would work use the head operator, it would work, however, of course in use, head operators and other things well, you're now, starting to incur some or the overhead that you try to get rid of at the API, but at least yours pelipper experience doesn't completely deteriorate just because you used one more feature right, but I think I think what we landed last time was that this is kind of what solvable to the types at all, because I think there's a there's, a language aspect to that which has to do with composability and other thing is where types are useful and then the other question is okay.

A

How can we get this to a better place for some performance, oriented, api's and I? Think, given this proposal, I feel like we can have a very smooth transition there. If we feel like we need to do that and for now having the overload seem fine. The only thing is, we add these parameters or this these special attributes.

A

You probably want a rule where we say: if you have them, we prefer them over the range based or index based methods, because otherwise, if we ship span today that takes range or index and later on, we do the other thing. Then, while we're screwed we're still calling the other ones and you don't get the performance benefit, but I think that's more like a designer. How the feature would would work at that point, so I would table this for now.

A

I would say: let's not talk about this today and all because otherwise, if principal Christophe is meeting here and secondly, I feel like that's something we can split off and we can just review index and range in isolation all right. So then, in the.

C

A

D

Wouldn't be able to do that, so we would not be able to do this error yeah. Why Britain came in what, since, like the same syntax, can't call a different method ever if it currently compiles and binds to a specific method, we can never change.

E

D

Always finds the we, the API, if.

B

The API introduces a new method and we prefer that that is a that is known. Then it's no longer a language breaking change. It's a library, breaking change to have a different method, called that the span API would choose to take that.

D

We change overload resolution prefer a different method. Don't we usually consider that a very big break would.

F

Think that you, you would because, if we already exposed both a span, overload that start and and a one that took range and teacher trips, the feature it calls range and then later you add the new syntax that would make it prefer to call the more performant. That would be that's.

A

True, well, I think it would be that you should too, let's say for a second: we ship, both methods but slice method it takes in the size of the range right I. Think we wouldn't introduce new language that you would just say we provide the developer, an attribute that they can apply to the methods which, if they do, that the compiler would now prefer the other overload but there, but the library also has to do something to their code.

A

It's not that you just take an existing code for the same target framework recompiling in a different Bible I.

B

Get that but they're still the there's still the situation where you target that new API, using an older version of c-sharp that didn't know to prefer the new overload right.

A

B

And now it's a breaking change to upgrade to see that.

A

All I'm saying is like the the author of the library base. It has to say these efforts have the same semantics. It's just faster I think it's it's going to be problematic.

G

I think that's why we'll say like that seems.

A

Like something that we'll have to imagine, let's do anybody, but let's I mean so basically you're pointing out. We should review this before we ship three, all right, so that we make sure that if there is a break and change, we can at least decide not to ship certain methods now so there we later, but that's yes,.

D

I wouldn't say that yeah, you know you shouldn't add the method you're worried about the performance shouldn't have the method, because there's a very good chance that we cannot change this yeah.

B

I, don't know, I think this, there's all kinds of problems with that particular solution. Also, you have an income in message: that's different from the usual framework pattern of having having offset comma length. Yes, and how does that clash with you? No it's just.

A

That's why I said like I think this whole proposal wants a very different discussion yeah and we can choose to.

B

Have it you'll.

A

Have to trust everywhere, yourselves.

B

To find a way to solve the problem, if it is, if it turns out to be a problem at that time, I think yeah.

D

So one thing that we talked about briefly in LVM today was, as you can see, all of the well-known members are annotated, but at least four range some of those aren't strictly necessary.

D

The compiler doesn't need all or from start or to end it really just needs create, and it can it can create or something substantially similar, and it can cons up the appropriate indexes indices for each side. So we can remove the well-known Ness of those right.

F

Method, property or a static rivarly field, because it isn't all just basically index start in that class.

F

Yes, 0 0 at 0 yeah, basically, I do not recall because it would seem like it'd, be more patient for the DECT, optimized, etc. You can use it with in parameters.

B

My guess is it's for consistency with the with the other 3 creation, those three factories, but it's a we've kind of factory, because you could return the same, actual well, you're returning the same actual value. Every time I mean it's.

H

Mostly a constant.

A

Yeah I think it's problem. It might be not effective like the way they were originally done, but the obj was a constructor or was some I think we also had someone had Boston arguments in all cases and I think that yeah at this point it should probably be similar to empty or I. Just like just aesthetic field, yeah yep.

D

And so that's something that we couldn't I can remove the compiler that one of the kind of compiler is to news that I'm adding has removed the dependency on these well-known numbers.

H

You support off without that even deserves on my design.

D

B

Yeah, if you want a consistency between the two types in terms of how they're created create, might as well be a constructor I think it's a little lot of one is created with a constructor and the other was a factory. Well.

F

Create presumably needs to call an internal constructor anyways. Thank.

B

F

B

It could it could call it full, constructor and and assign the values in itself, but it's.

F

A read-only strong, so you'd have to do with calls and all kinds of other crazy stuff right yeah. So it would probably.

D

And that's fine, so we can just make it a constructor which is fine from the compiler perspective, we'll just change the signature that we're looking for.

I

So I had a question: I you dropped out for a second or like, as I may have missed it, but one of the right right now there'd be no difference like right now. The implication of ranged on all does new range new index new index, but in theory it could just be stored in the static we'd only field, and we would avoid all of the argument checking and whatnot every time all is called with the constructor or the create method. If the compiler is just generating the equivalent of range not create new index new index.

I

In theory, it would be more expensive than message. It is really good at making them.

H

And it's true yep: well, we could. We can always change all to call an internal private constructor, so I don't I, don't think that would affect the cosine.

I

I, don't mean I'm not worried about all. It was the if the compiler consolidates all of its access to the single instructor or the single create method. Yeah, there's not much. We can do about that other than have the JIT be really really good at right. There you see in cons, you know in whining everything eliminating branches. All that has done so.

H

We did, we did actually run into something similar with the span and we filed a few bugs where the compiler was implicitly creating spam instances on your behalf, passing in minerals for the constructor arguments and the the JIP team was pretty good about blocking all of those out, so the projected Allied all of the events coming I would be surprised if we didn't do the same work there.

I

All right, we should probably just I liked simplifying. You know what the compiler depends on these you're, probably just a bet for the nd or someone yeah. It can be done. It is yeah I.

B

Mean that's a similar thing for from start and to end, but they technically don't need to do the argument. Checking before the end that they're just gonna have a 0 or.

D

Go yeah now right so uh I kind of anticipated, which it wouldn't never seen through a struct constructor because it really should, but.

H

We shouldn't contort the API or the compiler to work another day with it. I.

D

Mainly I was thinking I, don't see a way that this could be observable, except for the fact that the all method would have to exist, but there's nothing wrong with a well-known member, not necessarily existing, so it we can optimistically. Do it right in the same way that we did this for a dot empty and friends that we emit a parens, that's not filled in with array dot empty, but only if a ray dot empty is present. We can check v1 to create.

D

We don't like just with the constructor with a range constructor and then, if the JIT can't do it, we can add a well-known member for all with whatever it ship size. And then you know, the nice thing is that the core FX can kind of decide what they want. That API to look like and finalize it, and then we can come in later and just kind of accept whatever X.

I

Is P but is right, we'll just sort of image it and then it helps all cases, not just all yeah. Yes,.

B

It again a similar thing for index, actually the creating the zero index and the from end the last element index of it beyond last element index. Those could be static properties as well. If you are worried, but they probably I mean that's probably not worth.

I

It okay, hopefully, will be essentially free and the equivalent of default index and default range to do, and I should.

H

Be clear, like I'm, not speaking for the jet team right I, just giving context based off what they did for us in the past yeah I.

G

Have a question if I understand the conversation here, we're talking about dropping the one aspect of it or we're not talking about dropping those members are, we know.

F

G

Right. Thank you.

F

And just so I understand so from start is effectively a call to create start and then index table last yeah and then to end is effectively created, zero index off first two or whatever. We decided that the property should be last time.

I

His last hat zero or had one hat zero.

D

So our sorry last that's correct last. What we thought was have one because that's the last elements. Yes, yes, yes right! So then so it wouldn't be. It wouldn't be last. It would be a hat zero.

D

It will be end. Yeah.

A

So what is the competitive need from end in vain? You today, like? How do you use these two members.

D

From index, we need to extract the for arrays. We need to be able to emit indexer I, see where we actually tell whether what the front-end and what the actual value are, so that we can emit it in vector when.

I

We do attention if someone writes a race of oh.

I

D

Mabel matter, if they'd like array of hat three, we take the index and then we omit and basically a load element off of the index, but index convert index to an int. Sorry.

I

I'm someone writes hat three with an array, since the compiler is generating all the code. Why do you use index at all.

D

Well, every type is gonna have a term or sorry other way around every term is going to have a type and hat three is just term type as index. It.

B

Might be that somebody stored in a local and use that later so right, okay,.

D

Like you should still be able to, you should always should be able to index an array using an actual system that indexed type sure.

I

But I guess I'll admit it. Maybe this is a you would accept it someone's a minute at PR, where, if you actually typed array, sub hat one to access, the last element like you would be: okay with a PR Denton that, where the resulting il didn't mention index at all, right I.

D

Guess sure I don't know it would be we'd have to see a reason to do it. Given that we have to support like in the compiler, we generally don't try and shortcut these the normal operation just or kind of convenience sake if.

H

The type index doesn't exist solely if the compiler doesn't see it types just about events can I still use this syntax to know. Do you will get a compilers.

D

Hat means construct me an index and if the index type doesn't exist, then you cannot use the Select. We.

I

Did pay me I was very specifically I, totally understand, not shortcoming for things where the operation is defined in the library, but for array where the operation is defined by the language round, tripping through index seems odd to me.

D

But I understand it's just that we, you know it's easier since we have to have a code path for index anyway, it's easier to just go through the same code path and reduce the probability of bugs I mean there's like a lot of I just found the bug actually because of this code. Spit I did where I made like a subtle change, and it turns out that that's observable through ref variables.

D

So right, that's like the kind of thing where you really want to avoid that, because you might introduce a bug and it better be worth it so.

F

The compiler basically won't be doing 4x index hat 3, it won't do load, Len and then some 3 and then just call into the.

E

Access there's, no access will.

D

Do a new index and then index top and then the full index map in order to generate the awesome.

F

Does that does that mean that system memory needs to add a method.

D

No, in the same way that we do not call a method to do array indexing right now or.

A

Additional fatuity, this way I think there's an eye other sentimentalism at the inspection right.

F

But the compiler has to have the knowledge of how to get the length out of array, whereas for every other type the user has to do the conversion from a hat index into appropriate index. Whereas a ray, that's not possible, because array doesn't have well.

A

Logically, the way to think about this right but luxury, is the compiler pretends. That array has an index that it takes an index and just write code. That doesn't do that right. So, but in a normal or in your own custom time you would actually have an index. So that takes an index yeah and then you implementation has to do the math right, so the user never has to do the math yeah so you're, it's only. Then we see whether it has.

F

If from in then like thank.

A

You yes, but they'd, use that they receiver. Does it right, so the person crossing it didn't never have to do any math. Just like.

H

A parlor right now ever can possibly woke up or rained all night as a as a belly.

D

Yeah, well, we don't do a write-off length. We do there's a specific aisle instruction for getting the one we.

A

Do provide those for you really.

D

H

B

Can write each for each loop over a rail turns it into a four loop so that way, you're getting the link that there okay.

D

So if you have like some array- sorry Steven I, don't think you can see this. You can see we can. We can type it up into this problem. Don't worry about. I can guess, okay, so you write on an array, hat 0 or 1/2 through your ever right. So fundamentally, what this is going to do is just going to go through two passes. Go through initial binding 3 is where we see.

D

Okay, you have an array you're doing an array, access and you're passing in the scenes as arguments, one of possibly many so we're going to buy in the array.

D

We're gonna bind the element, access order to bind the index, we're just going to stick them in this bound, the right thing and then, when we come down and we're going to visit, we have a rewriter pass to simplify in our lowering phase and we're going to see this bounder right access and we're going to say: okay is the index, but I'm sorry is the indices sum 0 and index and if so, do let generate a load element, create the bound nodes for a low element access, but do all the index math and the process.

D

So the suggestion of like hat 3 and this statically turning into something different, is essentially saying in your lowering phase. Instead of just taking the normal code path, where you look for an index and then you turn it into a series of operations on that index, statically realized that the thing that you have in hand is a constant value that was written literally pull it apart into its constituent values and then emit specialized code for that constant. So that's not.

D

This isn't like a new way of doing things. It's just like additional code. It's a specialized code map and so the answer for any specialized code path and the compiler is. Is it worth it because this generalized code path works for everything, then the only question will be? Is it worth it and that's a question that we can ask and then answer based on data we.

B

Do do it for another language feature which is tuples where if they occur literally in the in the source code as tuple on either side either as tuple construction expressions or deccan tuple deconstruction, we will try to avoid actually loading the corresponding underlying types and I'll.

B

Try super hard, but we will try they're, well-defined cases where we do so that if you never carry the tool itself around and never refire tuple value, we will we will not require it that has, but that had to do with timing issues around who could get at the tuple types. When the language feature came out, I don't think we want to do something similar here or they just.

I

Rinse, it said it can be done with later. My my question is particularly because those thing about ask you know: common access patterns today and erase of a radar length. Minus 1 is a pretty common thing that you see done this, like you know the last element or something, and so if you people are going to start writing a race so at one instead of a race subarray dot length, one just.

D

You know it occurred to me, ya know it's and it's I'm not saying we won't. Do it I'm, just saying that you know we should consider it as with the full knowledge of what it entails and the trade-offs like, for example. What what if it's hack negative one, give the make sense yeah would throw we want the compiler to generate a throw. That seems really bonkers. You should probably write a compile time error for that, but oh wait. We can't provide compile time, errors and and lowering that needs to be in binding.

D

So we need to hoist this code up to you know, there's there's no flex fee, you know, yeah I mean the question.

A

There is like I think it would be nice, but I, don't think that it's like required to fail for this at compile time. But it's not like anything bad happens. It's just said you have nonsensical code and we cannot detect all usages of nonsensical code anyway, but it's the best effort, I. Think if you at all so I think to me, depression was more like do.

A

We believe that having any implicit conversion from int to index is good or bad because of the negativity issue, I do matter who is doing it, the compiler or the framework right, and if you convince ourselves that the usual guideline up on having implicit conversions that can fail, doesn't really apply here, because the the the code that will blow up that you've seen the debugger.

A

It's basically the line where you, where you make a mistake right, because, regardless of what you do, the receiver, usually when you do it in line with an expression rate index into an array or your call substring. That message would throw the index out of range exception right now, and we just do it at the time you do there.

A

You know the conversion, which is usually the same thing, but it might be earlier if you capture, there's a local variable, but I think that's still very close to where you made the mistake, and so that's I think actually pretty helpful. It's sort.

B

Of the it's the price we pay for not having two extra bits in the instruction right, we could have index completely collect any integer and and have a bit next to it, that is from end or from beginning, and we have two of those adding up to two indexes and like two integers and and two bits in the in the range and then we wouldn't need to do. Then we wouldn't need to check yet we could. We could delay it until, but look at that.

B

We have two extra bits that make the type layout insane in memory and all that kinds of ways. It's sexually we're just throwing a.

G

Little earlier on negative.

B

And and saving those two crucial bits: it's it's a performance.

C

Thing really, but.

F

Is there any concern that, for the common case of using a hat, people are going to have to check if it's from end, which will be a bit mask access, then they're going to get value which will be another fit mask and then they're going to ultimately subtract the value when, in the you know, it could have been as simple as a if value is less than zero, then add it to length instead of two instead of just taking value, as is when computing the index, it seems like there's a lot of operations there there's.

I

A lot that those are the kinds of things that I was talking about with regards to fidget and whether it could eliminate stuff or not. But.

A

I mean like the less than zero comparison, is effectively the thing that that the from an index that can do so when you inline everything at your time, it would be probably close to what you would have written manually at that point,.

F

Well, other than value has to do a. They can't just mask out the upper bit and.

B

It does it to if it does it once compliment yeah.

F

Do one's complement mask out the bed and then the user will ultimately end up subtracting that from length when in actuality they could have just taken the original value like since it was already evacuated. That's.

B

A one-off issue: well, it seems like clearly it's because it's once complement and not two's, complement right, you, you need to add it and then either subtract or add one to the result.

H

B

It wouldn't be fewer operations, I think.

H

We would have to news we would have to, and that certainly takes and next works in the framework.

H

B

One thing we had in the inner and earlier version of the type, if I recall correctly, when we, when we tinkered with it on the compiler team, was a static, helper method. Well, maybe it was even an instant helper method to apply it to a given.

B

When you have a collection of a given length, you could passing a length, that's a parameter and you can get an int out which is sort of the computed index from beginning based on a given length. So we could offer the helper method of computing it for you. You don't have to write the logic in your own implementation. Yes,.

D

It was like get offset or something yeah.

B

D

B

A static helper, oh that.

D

You can't get offset on index I think and you like passed in the index in the offset right.

B

Then you don't have to write the logic we give you a helper method that does the logic, and all you have to supply is what's a length that you want to compute relative to and if and if from end is false. We just ignore it and if it's true, we.

J

J

Okay, then close yeah, that's.

B

Yeah and that should just return an int, I guess yeah. It's.

A

The thing we should we should talk about- maybe we should talk about this now, but I mean my other question before we go to range. Is that well get length, sorry that.

D

Get length let's go range. Well, there was the second offset index. How far that could exist. Yeah.

B

It's one for index. Yes,.

A

That you need to so what would it be good get.

B

Offset or something yeah, we compute you the absolute position relative.

A

Given length yes,.

B

Yes, yes, yes, that's fair.

A

So assuming we had that with the compiler call that probably not right, it was to do the Mayo. No, we would be Colin and I opposed to doing the from n question mark yeah.

B

We call it from the array locking solution, yeah.

D

Well, yeah yeah. That seems fine like we rather do we'd rather do a method call Ren.

B

But implementers of their own data types would call it as well and right and nobody is like having hash code helpers. You don't want to implement. Yes, no. This is just a case where the compiler is hopefully implementing the indexer.

H

B

Because the indexer.

H

Doesn't actually exist not far away? Actually, I was playing around that's in c-sharp code, a raves right now also Emily's, a c-sharp compiler allows you to use a long as an indexer into an array of estimate. Would we need to account for that here as well, like your conversion from long to invent, doesn't care for the original discussion on this I? Don't know what the scenarios are pretty using.

F

Long index, well, you can use them with native code, but.

F

H

Only be for indexing yeah.

B

I remember in the beginning, when Christopher also attended some of the design and c-sharp design meetings, and you were there too, we did discuss having an index and a long index having a second type, and we just realized that scenarios for that were probably zero. So, rather than trying to have completeness on that, we would just wait and see if anybody came and said we'd crucially on this, so we were kind of a aware of it, but yeah advocating.

H

Game over water raised. It's just one of the things that came up in yesterday's design meeting was that I have an API that took an int or long or something as a parameter and I was plugging it through to a collection that took an index as a parameter. Yeah.

G

H

Kinds of casts do I need to perform explicitly versus what should I depend on and possibly yeah.

G

H

That we said we want an implicit conversion from into index just to make the code look nicer, giving me the same for long just because a short currently allows it and I.

E

Don't know if they're good.

H

Scenarios- and we don't have we.

K

A

Throw them long to index so they would just be another unpleasant operator, but.

K

From long tween or longer index, if.

B

There's one that would fail on.

H

A

I think that's a different class, but I think developers are kind of used to the fact that in in practice only means half of the thank-yous, because almost there's nothing takes negative values. But I don't know from long to it seems fairly long like I mean because now we're like well, you care about it like index value plus 1 right. The scenarios.

H

Where I've seen it happen and I don't know if these are good scenarios is a kind of what Tanner said. I'm interacting with native code. I have two pointers: I subtract them the result from subtracting. Few pointers across all dawnette platforms is along right, even if I'm, in a 32 the process and now I use that as an index into something else. Yeah.

B

H

Though I know for a fact, it wouldn't fit into an int. It's just it happens to be talked about, but.

A

I mean like, if you might pointer code, then I think in practice. You believe, when I haven't sorted that point anyway and said: okay, you know I'd better be finished with.

D

Lot of the I suppose I've got a scene explicit.

A

D

Yeah yeah. That feels like it's probably better, because if you're wrong, you've been really.

H

B

A

G

You're wrong, but there's.

C

H

An explicit conversion will truncate, whereas an implicit conversion throw in there celebrating something. Well, if you do check well, yeah I was.

E

Gonna say and well dr. snow. Shout no.

A

I'm saying that you would, you would have to be surround the cops that eject in order to get that behavior and.

H

As we discussed.

A

Yesterday, only Jeremy, Jeremy and I are like I feel like. If you do point arithmetic, like I, think you'll pass the point where yeah I would say: convenience is your biggest worry, I think you because Vargas, hopefully they could down pointing to I. Think that you like this was just an example of where you might get along fun. Yeah, that's fair, I, think I.

A

Think there's in the Simpson area came up before was a spin head in the context of pipelines where they basically use buffers that are concatenated so like every index they have must be along because they don't know how long the request goes, but any given point in time. They don't exceed the the window size or next value, but everything takes along I.

D

Should also note that array access currently supports an index pointer access, it's not much where's long, you mean no well, you won't be any higher index. Oh and oh yeah like pointer access, there's! No, because there's no.

H

D

Yeah yeah so I'm just saying like in this scenario, where you're manipulating poincare lengths, but you don't have a pointer but that one's a little shaky, no like I, feel like most people who are manipulating pointer. Lengths. I'm also probably have a point about.

F

Buffers where technically it's a pointer, because the language hasn't updated that you span yet hopefully soon. That's why you know the language the terminus well,.

D

Hopefully, you always know the length of your pointers. No no I, don't know I well, but for fix like a first gonna use. Can you use it, though their pointers for the c-sharp purposes also c-sharp does not if at at.

B

A point where there's bands well des bands and.

J

B

A

How do we feel about general any operators on index like, for example, it? You know plus plus minus minus zero one we discussed I, don't really discuss some pairs, and these would be nice right. Inputs equals not equal to be as we've always.

D

Checked I, just don't think it's possible like how do you compare zero and hat zero yeah.

H

Comparisons of the strains, unless you know because.

A

They're not equal to can do well, there's only.

H

A

There's no W I, guess the for started simplicity anyway, but like so.

B

Plus plus, would would add one if it's from the beginning and subtract one and fits from end. It always goes to the right yeah.

A

Tarek just brought this up like I said like to me: it's not very unconfuse I. It always goes up, but since you just said the opposite very good point, it's like it always it's confusing.

L

A

M

Advances yeah I.

B

Can see that many.

M

M

To advance is to move is to move right so.

H

Your case for them, which.

B

Means subtract, two from the hundred that you had from end it's 98 from end yeah. That makes sense.

M

Full sense because, like you know, if I'm in some index, I know that if this is a mix from the end and I'm saying Possible's,.

M

B

Think the expectation is when I have an index I if I have operators directly in the index, I shouldn't have to care whether it's an abstraction over whether it's from end or from beginning and I mean it. It should have an abstract, meaning that makes sense for all indexes right and the only useful meaning for those would be that it takes you two in one direction or the other direction. This.

M

This would be true if everybody understands this, but I don't think.

A

Say that I give you a value and its height less index I.

A

Don't think it's helpful if, if I do plus one on that one, it might go left or right depending on how it was initialized because like to me the way index works is head or non hat is just the way I give you the video, but it's still the same value. Like means you're still talking about slot number 455 in the array whether I gave it to you relative from end or whether I gave it to you absolute from start all right.

A

It doesn't matter to you right so if I now say plus one, but whatever this.

M

M

B

What they, what the utility, though, is, of half.

A

That is like, so the reason why I think operators might be useful is that if you don't have them like very common operations, you do with indices, basically stopped working right, because as soon as you as soon as you have, let's say you had five and now we want to say you know minus ten or something on it. You can only do that now. You have to like, if, because there's no arithmetic to find on those guys.

F

From and then value otherwise, plus 4, plus plus yeah and.

B

We think it is common to do. Cuz I was thinking about okay, I'm writing my own for loop, but if I'm writing the for loop I need to have an if I'm writing a loop. I need to have an in condition of some sort right, which means I. Have a collection. I have a concrete collection that I'm falling over and that one has a length and I'm, probably better, off, converting and converting it to an int and and taking that up to.

A

Them I guess that's fair, but the thing is like the way I see those like when you do parsing code. But very often you say you know first name, comma last name I didn't see you say well. The first name ends there, but then I know I have a comma and I called you. I want to plus equals two on that one out, but no it's like well I have.

F

To write the business I would also think that most of the market- well, most people, will already be exposing overloads the take in and so they'll add the index overloads for convenience and those methods will surely call into the intent overload by realizing the index to concrete values. So.

A

The receiver doesn't know I'm talking about that. I mean the person that uses string either just called substring that takes index right or substrate that takes range. They want to do their computation on or the holy Express, the first name, how to express the last name: Auto Express, the comma right and at some point in there you do arithmetic, where you say: yeah nos man starts to two things. After the first name, it seems.

D

Reasonable, especially for range to be able to like say, yeah I, want to move the start position in the range a little bit forward or a little bit back right and for that to work. You have to have arithmetic on index, because otherwise, because it's based on index, so it has to be able to move it right forward.

F

That's basically the concept of instead of start, and you have start length right. You want to start here and you want to move this many digits from the current position. Whatever.

A

It is well an end versus length is just how you want to express the rich idea, but at the end of the day, you still wanted to say to what undress. That's what.

F

I'm saying is you're saying that there you have the case where you have the starting position. You know you want to advance two characters right and you don't want to yourself have to realize the index, because you might not have all the info right.

D

But then you will then there's. The second point, which is like the plus operator would have to throw you.

C

Know circumstances.

D

And the sense that you're going past the end or before the beginning, yeah, but.

A

Index wouldn't know that right, the.

D

Earth would wouldn't.

A

Only be able to W do half.

D

0 plus 1. What should it say? Oh ok,.

A

To mention throw ok, so that's the same situation with like 0 minus 1. Exactly so like yeah, it's good at the cases where you can validate you should. Yes, that's probably true.

F

And then you also have the Katy. So unless the compiler has special knowledge been for literal values, you will never get the error codes.

D

Which is about.

C

This like, when I, ran it through of.

D

All of these operators throughout yeah.

A

Seems for math, though that's kind of unfortunate, because you really want to do the operation and then check where you land inside you don't want to do it before you do the operation. That's very account intuitive yeah. One thing forecasting it's kind of different because you you may not cast when you just call the method and then the cast happens as part of the method replication. So that's not the point that you expect any anything to happen. But, like look arithmetic, it seems odd. That plus would throw.

F

And there's probably cases where the user statically knows that what they're doing is safe and we have to do checked internally and they want to alive those check. Contacts by.

D

The way I think, maybe that statement is true for where we've ended up, but I think that from a design perspective, that is absolutely not true. That I would expect fixed length integers when you overflow or underflow to swirl by point and that, in fact, all my preference, if we were redesigning everything from scratch, would be that every single one of our arithmetic operators throws on out of balance values. So I would agree with that setting, but I.

A

Think the problem is an ins. Is that if you think of index as a more convenient way of in in 32, and then you wouldn't expect it to throw negative by this, you wouldn't expect this to throw on you know your venue is 25, because the collection is 25 and then you add one and now you're out of balance, but you would generally expect to do the arithmetic you get away. Then you compare that value to whatever bounds you have, rather than the opposite right, but the problem is in our essence.

A

We can't represent certain images. We can represent negative radius and we can represent. You know past the end right. So the only the only option we have is through. So that sense indexes is not like implicit. Would we could add, try methods, I, just don't think it's worth it. I prefer no I mean at that point. Like you know, when you have a try method, you have effectively defeated the point of having operators right, because no, this really gets ugly to say, plus 1/2 I add one twice: we move, one I was.

D

Saying you could have it in addition right you could have the operator and then you could have a try method like triad.

B

I think the operators are for situations where you know that there is that much left menu, maneuvering space left in the collection like you're. You know that it's a first name comma last name.

G

B

It's not for so much for looping situations where you're already doing a bounced check.

D

Yeah I should I should say this is roaming in every single instance, where you go out of bounce. This is a logic. It is a bug in your program. There's no situation in which this is for something. That's not yeah. All.

A

I'm saying I said, like the I think for the conversion: I convinced myself, that for the most part, the difference between your using the value versus your contracting, the value is within the same vicinity, I. Think of arithmetic. It's not kind of like that, cuz that, with the way you write normal code as you do arithmetic you get values out, they may be invalid, but then you do your checks right and then you call them nothing. That's right see, but generally intermingling them as part of the athletic operation.

A

I think we were surprised it to people.

H

So I've written on the board, since we were talking about the example of splitting less than comma first name so right now, if you call them index, if you turn on index up searching for calling you're going to get back in and 32 as a friend as a return value, do we see any api's? It might return index as a return value, as opposed to simply hit ball sense like do we only see index is something that's going to be passed in, to be honest rather than from.

D

Actually kind of dislike returning and index because it's it's no longer specific to the it doesn't provide as much information, because it's divorced from the length of the collection, you've actually lost information.

H

So it's not telling you anything more than a regular and thirties didn't work. Yeah. Thank.

B

You telling you less.

H

Because if if I was going to argue that, if we saw an API isn't returned in index at that point, it definitely makes sense to add operators, because now you're going to have local variables of type index that you're going to need to advance on or reverse on. But if we don't anticipate people having locals of type index, the operators are a lot less used. Oh.

B

Wow, well, you might solicit component expressions instead of thing right. You might still get the message for him/her yeah.

H

Yeah but didn't Daniel if.

B

H

The message forever and we expect people to immediately call the get any actual offset from the underlying content. Then you only have them for exchange purposes and everything falls down to in 32 other, which means they're not really sticking around as locals. Right here, yeah you take it as parameters and then we're all okay.

M

So we don't have to expose any now we see if somebody leaving we can just edit it.

M

A

Should definitely have equals equals because they are well I, mean I, think there already exist and structurally flawed right, you have to write them. I think there was the conclusion. Yes them and the family of limitation is not as efficient, as is what I always thought that when you do equals instruct without defining yourself, it's probably performance will appear like this long. Okay, it's effectively a mem comb.

A

The thing is that, for offers of a survey see everyone, we don't want those guys. So then, for the the method that gives you the the absolute offset. What would that look like we say, get offset with passing length yeah we have you of the names. Also thank.

D

God I think it's position is used for stream. If you want any greater than and.

B

I'll get musician in.

D

Today, yeah I mean offset, seems fine, I don't know, but if you want to give the sense that this is an absolute position and.

A

The reason I like offset is kind of it implies it's from its relative to zero. Well,.

D

Then why it's relative to something in the same way that when you say offset in arrays, you don't always.

D

Like the array, copying or something like I can give you an offset, and you can apply it to something like the start of the array right or started the span before we had it spans.

B

Hey guys I guess the right word is the index, but.

A

That's kind of ticket yeah.

K

L

Don't get indexed what.

K

Are you guys smoking.

L

That's, why is it absolutely right.

D

Yeah offsets I mean it's fine, I, don't think it'll be confusing. The dot comment clarifies it's fine I, don't.

A

And then the person think oh length, not count, not size. Length of things seems to be the right. Oh.

A

Yeah I'm not sure what container would add, because I mean again like when you see things I, don't know what else would be.

B

A

Be a little I mean it's.

B

Not self-explanatory here right I mean.

H

But there's like a five people implementing things like elections and I would imagine that that category developer is going to be going through some of our tutorial and sample code. So I would hope for that. Let's be familiar to happen.

D

Well, the document the document will make a lot of sense and then the collection length, if we name in collection length, that's fine if we name it length- and it's explained in the talk comment: that's fine too yeah I think it's all fine, all right, so negative, get off for now.

A

So I think that index is done or any other comments or index.

M

A

Fact we message which ones we propose.

M

From start zero index yeah, we have from.

M

I'm just saying, but you can actually want to create like zero in this I.

M

Don't see any big value, but just using that one.

F

Yeah I'm about.

F

I mean couldn't you just make front-end optional at that point, and then you have a constructor that works. So if you just say zero, but he start and falls.

A

Very saying, whatever single well, that just takes it, didn't make it negative.

F

No, no, he would he was I was saying you could just make from an optional. Oh, it's.

B

F

Instructor yeah in the constructor and then you can have a single constructor that works for the common case of.

M

G

To do this on the head, could somebody summarize, before we finish index, how the final shape looks like compared to what we're looking at from the preview one release.

D

When they cut this point, we're just adding stuff all the stuff up there and Christy.

G

Would definitely valid.

D

On the Walker layer.

G

Which which things would be I'll just quickly review so.

A

The one thing maybe added mass index I'll get all set. Sorry like. Actually let me just pretend project this I guess this easier. Let me just preface this.

D

How do you propose to her for something I.

N

Just on the walk over here when I was trying to find my dad joke, that's when I came up with.

A

So we already know him yeah. You.

D

Just completely you're, halfway into the meeting today, right.

N

Yeah the range of time I'm in the meeting is half.

D

You didn't figure out the whole faculty: I don't have an index top end representation, its independence of air, that's dependent upon hmm so.

N

My bytes here is not one I tried to make a syntactic joke and I succeeded.

O

I did all right Kelly, you see.

G

Both of them together, oh sorry,.

A

You're asking ringing our questions today: I am.

G

Demanding I am demanding today, you're.

A

Very demanding today, I am.

D

Yeah, so that seems I, don't think the from skirt from.

B

End is a bit problematic in that that's already a public member called from end right.

P

A

She's seen of this community, although.

G

I sir I will say that I was imagining it looking kind of pretty with the front started from end in a language that doesn't support language elements. Yeah.

I

G

I think we gave live without it, but if we change that boolean, which is probably gonna, have much less usage by real humans to a different name. If that was possible in the for the for the building property, then I thought the form. End was kind of sweet as like is a factory matter. I.

B

Mean maybe maybe we can come up with a better name for the property currently called from end? Yes,.

N

Thank you, that's what I was thinking actually not bad.

D

But yeah it actually that's perfect. It should actually do that anyway. I just.

G

Find that to be feet, I'm imagining it and you know some sort of complex code and the vestry factory methods. Just look prettier! That's where.

O

G

Gonna obviously appear inside parentheses inside some other junk, and those constructors are gonna just result in a lot of syntheses and a lot of fantastic, apparent nesting. Visual next look well, I! Guess we both have as money for the suit so see. You cancel that argument, but I still like the factory methods. The factory method specifically named for men, yeah.

B

That does look good correct and what we.

K

Call so manthan is.

B

K

B

It's backwards.

N

If we have, we've talked sorry I'm late in last and I only did half but we've as long as we have like end and start I would say, it's like don't go, don't I would say is from end is more line than its reverse.

P

Yeah, oh, did we event yeah.

N

Because I'm gonna have the start and end in this little setup, stuff, I'm.

G

Happy with that, that's what I wanted to get a visual visualization of this I think ice cream and you.

D

Want to advocate doesn't yet have language.

G

Support out there doesn't ever have like your sport that you deleted from.

D

So that's a chrome starting from ends don't exist right. Oh.

N

But from so from started is that's equal, equal.

N

But index so from start 1 is equal equal happen because you know it's not from end is equal to equal to hat one right. Yeah.

B

D

From its hat value, yeah.

N

But he had tuned one there for a second ago. I was right. He what static, no until.

A

The aesthetic on it and then.

F

The property from end are we keeping it me from end rather than is from end, and that's the property I think it means better is just from in here. I want to make sure.

G

F

G

Parameter should also be instrument. Yeah and a capital are.

A

Very demanding to that Kathleen I.

C

A

To just grab the keyboard there's a mute button here somewhere. Let me find it.

A

Can you call the rice it please I, don't know why it doesn't color wise I think there is a I'm, just I think I have a bug in the regex I think if I remove the semicolon, there I think it would work. Did you use most of line using sublime there I think this school is a bit more forgiving I think this works the thing yeah we.

N

Do use sublime when you're talking in the vs killing people gonna be pretty boss, I I. Do this all the time to piss people off.

A

Let me make friends very quickly in here.

F

And so get offset for for a not from, and so so if you just had index 0 from the start and.

F

Just return 0 right or would have return length in that case, so.

D

F

D

Offset lengths would return what you think. Yes, okay,.

F

So so you're saying get me the offset from length for the current value get.

B

Me the offset for a collection of length for the current value. Yeah, that's where length is a little length, is a little confusing you like that. If it was collection length or something like that, then it will be more self-explanatory. I walk up to my specific index and I get its offset in this collection yeah, but it's also right.

B

D

The collection of this length I, don't want to bring this up what, but you will both nonzero base arrays. This is not for, though we decided, we don't know all.

F

Right this is not for those. So what happens when you have an array that is nonzero base and you can pilot uses the syntax yeah it's illegal, but what if the compiler doesn't know, don't care? Look.

L

At me, no Penner doesn't.

N

L

You know I said different types right. You know.

N

You're gonna have a true, you know what happens when you did. The compiler are true with the numeric value too. We don't give a yeah.

F

Well, you can have an array like a regular canvas.

D

You can't give up a balloon right: what has value in brigly.

F

Always we always know whether or not the type is an.

D

Scri or a multi-dimensional array right now, this doesn't work on constant event, always of any type okay, so any it only works on SC arrays, which has the name.

F

D

The answer is, it'll will probably get there, but we are.

A

You doing it the camera you're just I'm.

N

Pretty people should look at me. Who's gonna go.

C

A

A section of the video strapped.

L

A

Of that I'm sure you know, as was Jay Thomas, what people.

D

So the answer is, we put in the common get the offset given a zero index base container.

A

We could also call the length and offset right and then it's the same thing. I.

A

Don't know yes, you're right, it's the same.

A

Yeah I wish we keep I would.

C

You want to call the connection right because.

B

I would be willing to burn a longer name if we have one that we think is really good here, because.

B

There's no, but now there's a collection indictable.

C

Collection I know.

B

A collection is not.

C

B

Collection has a count property, not a slinked.

A

Yeah anything that is indexable heather has a according to interfaces, has account, and if you could go back in time, I would not make a race link. Yes, that's it so I think if you use the term collection I think we want to use today income, because that only is any collection that doesn't have comments.

F

Also applicable, what we're saying some collection, it's.

A

Not a collection right.

D

That's why that's why I always really just had links yeah like.

A

I think that's.

D

Yeah you're right but I, know unfortunately.

A

All right yeah, it's not really a collection, it's just that all.

D

Human speaks for the one time, doesn't.

A

Know, oh alright, are we going with the next I think we're done with a next I think we have participated this today.

B

F

Challenge accep, sorry, one more thing you forgot to add: the properties we said would exist: the static property for start start and last yeah.

D

And maybe end as well.

A

Yes, so that's a good point so, starting last, so my concern is somebody says, start dog lost and they will be surprised by the output.

F

D

We also had n, well Romanian thing or maybe link well and actually works in the sense that the end is hat zero. Yes, then, it would look.

F

Maybe calling it 0.

F

N

A

That's different than it I mean it's an index. An index can it'll be linked and then.

P

It's like, like.

A

You have to buck in your your mental model at that point, so that's the way it statics in next start and then the other one was static index. What do we say end and then in.

L

P

C

A

G

Not saying you're tired, is there a reason for that yeah.

A

Because I I'm not a masochist type first and then I copy and paste it over because because.

P

Of you, people.

A

On the car, not.

P

N

And I'm not helping, we.

G

Freed, what are you.

N

G

Figured out what index start actually does. If someone wants to explain that to me. Well, emo types I.

D

G

D

That's it it's just a shorthand for new index 0. It helps.

B

G

No I, like it fine, but.

B

Not a lot because there's also a conversion.

G

N

Maybe Persian.

G

Is the one I'm seeing that's useful there.

N

What are we keeping the convert? So sorry I'm late? What happened? Did we keep the conversion? Yes,.

G

N

End conversion.

G

I'd like to keep the conversion.

F

So pretty, rather than a read-only field, that's if it's a read-only field, people can pass it yeah! Well,.

A

We'll be aesthetic I, guess, I think yeah I, don't next! Oh my pants I think that was a I. Don't know what the solution was, but I guess you could believe this.

P

A

N

Gonna be read-only structure, doesn't matter yeah, yeah.

A

It doesn't it's a real.

N

Instruction, probably more efficient, uh so just not to take a look and gift horse in the mouth, but with the implicit conversion. The people who were strongest against it aren't here are we gonna be revisiting this again when they read the note, we.

C

N

With other people who were Thomas Lee against it, aren't in the room.

A

Well, I think yesterday, even yesterday, I think we reached the conclusion that it doesn't really matter whether the belief do my thing or not, because it has to happen at some point. Okay, that's right! I just want I, just didn't want to say it's just a question of whether Emily raised the chords like yeah. Is it in the framework like it? Also, it doesn't even change where the exception occurs, but it's doing the same spot so in practice, I, don't think it matters sounds good. You want to make sure that we.

K

Come back another day so.

B

Start, if not very useful for VB but end might be do.

A

We start even in c-sharp, I mean I would say start an end up. Just it's like similar to like zero second seems ought to have end, but more start, but it says yeah.

B

A

B

I'm fine with having them both I'm, just saying well actually, for symmetry reasons you might use start instead of zero and VB as well. You might rely on that, rather than the conversion image it'll.

G

Be a little more efficient woman if you use the beam only struct morning.

B

G

B

Think it is once it gets in there. It probably won't matter.

F

My rap no, it's a word.

G

Size, it's actually going to.

F

Be more expensive right, because it's a single age well at least it will be less expensive after the gym finishes. The zero possible traction, but still.

D

It's not going to be more, it's not gonna, be more efficient registers. Don't get Norfolk I need.

F

You, on the other hand, range is separate yeah because of that will only pass to us and we want to start off along system. D architectures I should say.

P

D

And are we just leaving the plus/minus or B 2, plus minus N 1 index yeah.

D

Because you might want to be able to like move and index well, but there's no arithmetic operation to do it and it's actually not trivial. Based on whether or not it's from the end or from the start like bets. That's a reason to do it, but that's a I mean that's a corvex.

N

Decision, that's on it, my newest issue, at that right.

N

So one thing guide I'm. Once again. Sorry, if you guys talked about this, is there a reason we don't have a deconstruct method on range. There's.

A

My cadets a good point: that's headed in my head, I didn't ask about it. He hasn't.

L

Started Ranchettes.

A

Up so, but even on United Knicks will be. Will we do that on.

N

Index- oh that's a good point because we can deconstruct that to a bullet in it here it doesn't seem good for range. I. Think ot, construct method makes a lot of sense. Yeah. Are you done.

A

There we just record the segue deconstruct.

K

B

Deconstruct I wish you would let start an end in the opposite order.

F

N

B

As long as as long as the Deconstructor has to start first then I'm happier.

K

It's both ends, but.

K

We don't do without overloading to construct yet but yeah yeah by.

O

C

Well, you can overload it as much as you like, but we're not going to be able to bind. It is.

P

It very big around okay.

C

D

And it's actually.

K

For rain shouldn't.

D

We have a method, it's a ways you have decided yet so like it's. If you talk a little instruct operation, you can know.

N

That what Deacon started the race well, there's two things interesting with range: one is potentially deconstructive. If anything, it's a logical to fold, but the other one think the interesting that's not up there is we're saying that we're buying essentially range is an abstract range. It has no concreteness.

N

If you have like a from in value yeah, we.

D

Should we have a blind ribbon, all returns, a tuple email would like to scroll down there close. So that's.

B

Similar to the closer work, but not so.

N

B

Get a range close.

N

Should return up to hold.

N

Because you're double exceptionally at that point, because every time you do because what you want, when you close is you have you no longer have an index ever closes, always giving the to go? That's yeah, but.

F

Then we have the problem of putting a tuple in the public API and.

D

I think we have all the people. We need to prove that decision right here. This is an API review. Let's just do it so that it's thick as long as it at a time.

A

So I think they be treating return ranges that favor. Why.

D

A

Seems dumb to do that, because the whole point of binding is down the end yeah. The only other suggestion that fuller who it has was. We would have one method that gives you a start and and length as out, and you just discard the portion that you.

N

Don't want we can't do that I. Think. There's some level of one thing to consider is get someone more intelligent than me to talk about this. So.

L

Anybody would do okay, we.

N

Establish everything you're hired.

L

N

N

The difference between a three-word passing, something which is three words in two words, can be significant. Oh it's gonna, say and so I know parameter passing two and three words can do big difference in performance. I have the mini returns are similar, but we should ask someone. The terms are effectively equivalent.

F

To an out parameter of destruct, except that we don't have to do the bright areas.

A

So then, the next question is three different methods. While we have three different.

K

Matters good start, good, end, good length. You call them once appear all right. Just think of the.

B

K

B

All of them, you have to passing the size of their collection, yeah. That's.

A

The Dan said: that's the thing else: I don't like.

N

But why would you need the length because length is already a parameter too close, so why would you return the input well.

B

What is the length of the collection, the other is the length of the range of the range.

N

If Stephen was here, you would perhaps even some few.

L

All he dropped out because, when you walked in he's like that's.

L

C

Could have infringed what you put it, what inch range type.

K

C

K

We could, in.

F

30 execute.

A

Properly you mean that get type, that's gonna need to be close there. Anything yes, well.

F

Isn't it guaranteed to be closed as long as and and start or both not from it? No because you, when you do if.

N

You pass close a link if you pass range lock closed returns that great no.

F

But I was saying: I was saying that a range could be considered closed as long as neither start nor and it's from n, because then you can statically determine that there ever a quick relevant to each other etc. So it would still be valid to return range. As long as we say night, the range return here will have yes.

N

But then it's static versus runtime checking because who, like do I, have to reclose some range every time they use it like well, I mean you.

A

Could I mean I, give you if it let's say you might think. Oh right, like I, think it just says, gets very ugly in color. Can you put something up here.

D

Because I'm not even sure what we're talking about this point.

A

So let's say you are in the.

F

Can you do this? Can I save you? We couldn't have any is closed property and the check would just be or start and ends. Internal values together and say is less than zero, but.

N

Then that just means you're adding more dynamic checking, because you have, because you end up having essentially like a method which expects a close range folks says: if not is close, throw argument of range exception, then close and cool and then get in and get link, which are also going to check once again is a closed. Is it close like it's it's how much dynamic taking you it.

B

Occurs to me that the the API pattern today for people to do range e things is to have an and index or like an offset and and a length right.

B

And it almost would be the most beautiful thing to return an offset and a length from this right, bring it back to the two to two int values that are the currency in the API today. Presumably people would do this would often turn around and call something with that father.

B

So you know. Ideally, we would return it at a table with those two things in yet I mean.

P

The other auction is like what you could have this, but you could say lost then that would be range.

P

B

Just range Oh or the int cap right and then you could say.

A

Something like that I can just extract it in the way you want it through yeah.

B

G

I'm, like an emo sir.

C

What Jimmy.

G

It have the parameter named end on line 18.

C

Did I have what named in on.

G

Line: 8 million. What is the parameter? Name correct? Yes,.

B

The parameter is the length of the two lengths here, which is very confusing. The parameter is the length of the collection that we're yeah. Applying this to and the length coming out is the length of the same. You.

F

Could call it rings.

G

Got it got it that confused me but I think I'm a bit easily confused on this particular stuff, you're.

B

Probably not alone yeah.

G

Yeah, it's me just a little bit it's you know. Some these words will end up repeated a bit.

B

Hey I don't know I I feel like the we've decided long ago, that when you do this with intz, the preeminent values in question are the offset and the length right. It's like. Where do you start and how far do you go.

F

B

So we already have patterns around that, and so it would feel consistent with that, as well as probably be useful in connection with that. If we just settled on that and then yes, okay, I, wanted, oh I wanted the nth absolute and position well, Adam!

B

That's like what you had to do before right, essentially I'm cashing out of the new abstractions and.

A

So you're saying that.

A

B

End version of it.

P

B

Saying that that's a innit in a sense, that's a decision that that the doctor in work made long ago, like back in version, 1 or whatever, that right they we don't have overloads for start coming length and start command and and all that we just have start in length and that's how you do this Vince. That's how you express these ranges with Ames and we've been very consistent throughout the framework about that. So we shouldn't have overloads either. Here is my argument. We should just have the one that corresponds to what we've always done.

B

Unfortunately, that's the confusing one where there are two lengths involved, so I do I, do get that.

B

Well, he has two versions of it: no.

F

No, it's all I was saying you could have the second one that doesn't return there to boil, but it returns int. It only has out by their legs, so it returns start and then has length returned as ya plan, which I think is more common for. When you return to values, we return one directly and then have the other video, but.

G

Which they have equal importance so I, don't know how we pick.

F

Start generally comes first, so return start.

B

You know it seems worse than both mean yet time today, I mean there. There's yeah wait.

G

Why do we want the out version? There's scenario that that's gonna work better your.

O

Cannot agree on nobody name tuples, it's gonna.

G

Piss boned a terrible discussion.

G

Okay, no travel discussions that.

D

We can, we could have.

B

The terrible disco yeah.

D

Let's do it we're gonna means what we're forced to do it anyway. Let's do it just.

P

I think we should.

D

B

I'm just telling Kathleen what.

G

The terrible discussion, no.

F

No, we should just have the discussion. Yes, the former is nicer for he construction, because you could presumably pass into a method that takes start offset the call the tuple version and have the compiler be constructively direct. All right. We.

D

Don't sort of you, we don't really white people.

F

So you don't last for arguments right now. That's absurd! That's.

B

It we could one day it's been proposed, but you know: we've had a couple proposals around: it's not terrible.

D

Basically, I just think the too bold version looks better because I don't like the void, return of get offset like yeah I, don't I mean I, don't.

A

I, don't think.

D

We should return one inch, I think that the void return is just nasty I'm.

G

Just wondering the code in which it's a preferable way to do it because it's the chip, oh, you know we always have tickles available when we have this feature. So I did like the named range length I'm just saying you just change that name. I thought it was a little bit helpful to have them have different names. I'm.

A

Just trying to make samples comparable, so they don't compare. What's our answer, the same they're basically use the same variable name so that we fill.

G

A

Because it actually is shorter, but I.

G

Mean can anybody describe code where the where the out variable one actually looks better? No I can see code where they're equivalent in code, where the first is better yeah, essentially.

B

I think the only place where out variables are still better is in try patterns we're getting just a pullback. Let's you put the call directly in a conditional context and getting additional values out through our parameters.

G

But this is not that.

F

Because it currently does that one in each day, I kind.

D

Of suspected due to the lack of a necessary, but very that the simple version will eventually turn out better well,.

F

Right Barry example: I did a lot of times when you're in lining and then the variables day, whereas it doesn't get alighted as often force trucks. Today. What's your two bullets, it's.

B

A point in time, yeah.

F

It's a thing as.

B

You know if, if we start putting tuples in the in the framework, they will get more use and then you know yeah she'd pull up to Mai's more.

G

Yeah, nobody is actually going to bat for the offset, except for some pretty minor things. Does anybody feel strongly that we I mean so the out? Does anybody feel strongly that we should that we do need that out? Overload no.

F

No not at this room, I think the only concern is the discussion of for the public signature. Is it going to be upper or lower case for the start and range link parameters, level, return.

A

Because that was the issue last time, so out of curiosity, if get offset length, would return a key value pair this. It would be equivalent some text files in the consumption side right you'd be construct, keeping up. They.

D

Probably, do parent tells not somebody.

A

Construct, nothing I'm sure.

K

As far as I met with this off dude, we talked about anything yeah.

L

K

He would have said I, don't think. Let's.

B

K

B

D

We a domestic thank you meant, it may mean I, don't think so. I know we did right in our own code, we out of it but I'm, not sure that he ever out of it directly. Oh.

F

There is one on thank God and one on dictionary entry, but.

B

A

B

This is not a key and a value, though I mean you know, I'm just saying that.

A

Community doesn't matter, but it's a temple type or not by the owner, better selectivity construct method and then that call set would work. Is it in the same way right yeah.

D

Really then we might as well just say, Value table yes, but cept. It doesn't matter. There's gonna meet debate matter. What we went through this before it doesn't matter what type you return there will be a bit well if we make a destructor to certain public. You know there is debate about that too, but we'll part about. If you make your own struct with the constructors, there will be than me about whether or not it should have accessors for the individual elements and then those accessors. It will be debate on the names.

D

Okay, there is no winning risk. I promise you the only way to win it. Is they actually have a discussion and just settle on so.

A

I mean I, don't know what I'm saying is that like I think nobody like is the second one clearly like what so the other question is assuming.

M

About like why you need to return false and independents, because.

A

You need to close both of them. Yeah.

F

Well, you have to because more often than not you're going to have a method that takes a range and then you're going to want to call the overload that takes a start and an offset. Because.

A

You have to do this why'd, you have to say, return. Substring,.

A

Start coming right, that's the that's! The actual thing right.

A

B

That's just a little unfortunate. They both have to take the same argument. Yeah.

A

Okay, so to be clear, you're proposing this, why do base is saying instead of doing this, you would say unit length, R, star.

P

Equals good start.

P

He's fast at all,.

A

Well, that means a little bit but I guess it's just the way. This I think there's must work.

F

M

Most of the time, people don't have a stuff input value, so then they will have named their what they don't have to start with the time, as we know that the started from stuff in it. How would how would substrate now that.

F

Substrate has to check because someone could has arranged a mess. I had.

D

Three missus, just this is just strictly more typing right, like it's longer,.

B

D

B

Computation work because get length has to compute the start once more and yeah. It's.

F

More work for the JIT in the J and the user in all cases, yeah.

B

Like it's, a agreeance employment I mean.

A

I would say, like I would prefer this one over at this one for sure, hey so I would say like let's delete this one. Is it like jelly everything and then I think it really boils down to these two and what I don't like about the second one is the fact that now this is really confusing right get links, and then you ask you possibly in the length is that whoa? What's wrong with you like I, don't know.

C

Yeah, that is my.

A

Unless we don't give you a length, if you just say you have to do the neck, the computation yourself but I, don't like that either yeah, because that's the whole plus one minus one disaster that is about to happen fairly buddy No, yeah.

F

I prefer that the top one which one is the top one, the two simple base one and that will work nice in languages that do arguments planning all right.

N

It work nice well and also, if we ever do it, but if we ever did, if we ever did argument, splatting I think we would probably watch me in the allowing destructive methods to participate in spotting. Otherwise, you have to declare intermediate to pool just to do spotting.

N

A

Support anything it has a deconstructed yeah yeah.

N

Because I, don't that's the logical out, so if we did a tight or two would both work sure, but so out, Baroness want to talk about names. Well,.

F

Out variables, work in F sharp because they automatically convert that into a tuple only if you have one out Bernie so.

N

O

N

F

N

This will be going back.

O

N

Code and Donna's.

O

N

Day, you can go, tell them ask of a person, I didn't get stuff, but.

D

N

General shortly yeah, he did.

D

Yelled about that people interface, not this before.

D

I'm not sure why he's wasting more so we.

P

Can consolidate.

D

It on the way to battle get like yeah.

A

So so I think we really wanted to get offset length method well,.

B

Yeah we could discuss the name of that as well, whether there's something more conceptual well.

A

I would just say it probably should start with awesome because you're on it next, we call it off set it. Kinda makes sense at all. It all said here and on start, but I don't feel super strongly about that. I'm.

B

Just told me to apply and we used to we talked about close within you know, there's close, but.

P

A

B

We could consider, even though it doesn't mutate, you still consider having verbs that are more.

A

Concise, so the consistency I think the turn off. This thing happen like I, think originally we had like clothes or something like it doesn't, including the VCR clothes means something very different. It seems like just odd. Like I mean it makes anything, but it is an open interval or something, but it doesn't really make sense for like the thing you want to get out is the offset in the length right so generally like calling the neck that what it does made much more sense than trying to introduce a term that will not be used.

A

That broadly and do.

B

We is that what we call them elsewhere. Do we call them those two ends? We call them offset and length. We.

A

Don't really like, but ever even in the frame or good thing we are I. Think at some point we looked it up. I think the most commonly used names are index and start.

D

But both of them are not great names here, oh you mean start at for the first element. Yeah, okay, I wouldn't call the index because it would be really weird get start as long as you use, because they're indexed I'm not get start and like that's a reasonable I, mean.

B

But we all here's length.

A

But the only thing that I don't like start is because they have a star property and that to translate next yeah.

D

I'm, not that's, it's the same value represented as an independent index, so if they aren't actually different they're, just right.

A

But logically web doesn't matter: does it cause.

D

Index don't get off standard.

A

I know I know I just like to do to be used to trim there, but like I, don't yeah yeah I'm, not I'm.

B

I'm more concerned about length yeah both an input and an output. Yes, um this is so we kind of avoided having a better longer more descriptive name when we talked about indexed of get offset, but now we have the same parameter again in the context where we also want to return the length. Yes, what we can call it count the ones count one length and yeah totally clear, which.

D

Is which it's never clear, especially if you're using a mutable array, you just type the wrong one every time, and then you correct it. This is what it is call it decide things dead.

A

Spam one more time, like one more term that you can screw up so.

D

The one thing that I really like to settle before is that last thing: where.

A

We Emily know the truth of this. We're different, meaning, I think there, because there are the people that offended. We were offended, but some others are not here either. I would rather say, like we can't send on what this thing returns differently like for us agree on the name and like what the shape is right. I think that seems more like is anything else, but this method now on you know, image images need to discuss yeah.

D

That helper, that I really really need I need to help her in the compiler to generate a new array from an array range and back. Sir, was.

A

It was the run tag, help of thing yeah. This will be a proof that didn't I think.

D

We need to just tell if they don't I, just don't think we have a I need a signature and a location yeah. We should I just need.

A

To know what they look for, I just need a fully qualified signature. Maybe they could ego. This is I think this was, you know, I think you're, right, I, think I put it into the designs repo, but I'd. Never.

P

Range, the range.

A

Yeah such a for compilers really helpful.

K

I can never tell what they are told in or not I was gonna say. Should we make range in and.

F

So that, basically slices range out of the array.

D

No, it creates it allocates a new array and company and Commons, which is the decision that was made wait. Do we need one for string two or are we chose that should be added directly on the string? They implement eight, like everything there is just the string type, and so they should handle that. But all this is compiler one. That sounds.

F

A

So yeah the open issues we have funded through this attachable heavy nexus with me. We didn't talk about that. Miguel had enough time to talk abou tax and next time is a business on my measurements. I think we had that consistent, no.

A

Don't kill me, should we have one yeah.

D

I think we settled on moving to a constructor for range, a public to argument constructor for range instead of the create nothing. Oh.

A

Yes, there was a create method to be new. That's right so.

K

D

Have to keep right, you drop the create and they have it to worry that and the compiler is going to draw well-known.

D

It's gonna drop all the well-known members except.

G

D

For the constructor and value and from end in the index.

A

And the constructor takes two to in two indexes right: yes, yes, and then we will probably still have the effective methods, I guess for people that don't have language support right to recognize.

P

Yeah yeah convenience.

A

But never compilers.

K

Watch at the factory methods, if you have a constructor, because.

B

N

Cleaner, if you're on BB range doc great where's, this new range, some different, sir. They take the same arguments.

B

No, no, it's not to create it's. The there are other factory methods that take only one of zero arguments. That's fine! Yes, yeah! Those are convenient. Factories that are useful in deviations and range thought all can be a static.

A

What do we have all from star animals get on to end to end yeah.

M

Should they just because.

G

From some of this, some of this gets my my head. A little bit confused from start means fundamentally logically, the same thing on index and range they're. Not opposite. Are they no.

K

Percent, alright.

G

A

In a sense right, I mean one gives you a range and you've seen index, but yes, yeah yeah pointed the same start. Yes, so what was the you think? That's you or to say something should.

K

They just be called from and to, instead of from starting to end I.

N

Would too and I think I feel storm about the two end.

D

I I dislike ramen too, simply because that sounds like conversions.

A

And when they said all should be a field or property right. What.

G

Did that start out and in there.

G

I can live with it, but from a to do seem a bit odd here so.

A

What was this additions that are from? You would do what start.

G

At and into act.

P

Like a live with that.

B

This marking system.

G

It's not something that you know that I get really loud about, but it is something I prefer it.

D

Seems reasonable, that's good.

G

G

Do you have this type? Are you trying somewhere, where you just try to get from me cuz, you ain't got Union no.

A

It's not that it's because I have duplicated displays and it's very hard to type. Then you have to turn your head every single time. Oh I.

G

Understand sorry about that, the! Why don't me to harass you in my well yeah I could probably do bring her up to you.

A

I was about to say you're lying, but that's okay,.

G

Hey it's late here in our country, so I get to be a little bit grumpy. So.

A

Where are you right now, I'm.

G

In Tennessee waiting to go spend time with my mom and as soon as I finish with you guys, I goes and some time with my mom at the hospital we're.

N

G

Creswell, do you know what that is we're outside Crossville, it's pleasant hill, but it's the halfway between Nashville and Knoxville. If you want a general idea, yeah.

N

I grew up in Nashville, so I know, I, know roughly where Pleasant Hill is yeah.

G

It's a yeah, it's not past best part. I trick villains before you get to take Russell well,.

N

Some would say: there's a bunch of people from Tennessee now that are here and I'm the only one. That's not an Alabama fan.

G

D

The real life is Eastern or Western can't see Central.

N

Nashville this.

G

N

Dead center yeah.

D

And the time zone changes halfway through so I still divide people, it's a very cleaned alive bomb in central.

G

Time zone, but maybe not that far but I, am in central science I.

L

Think all the operations we address, except.

A

For the very first one come.

C

P

G

We can't finalize finalize that out without more people being able to go right well,.

A

Like I think, first, we need to talk about index and range. Then sales before we talk about what API is take. It right, I think that seems like a sensible thing. Oh.

G

I see I'm sorry I misread your I misread your point. I.

N

Mean do we want to just I, don't want to finalize, so we wanted to start with an a final and initial list, so they he already had the as and I think you already have a.

A

Pr for that in there I.

A

Have it in the spec I, don't think of any for that? It's basically.

G

It would actually be good if I could go ahead and take off so unless you guys are going to make any decisions that that I mean if we're dad's looking at AP, Isaac I think I may take off. If that's, okay, absolutely you miss me.

L

Let's say yes, but you know what the real miss.

C

G

I think we got some stuff done, and so hopefully it's moving. This thing to the finish line. Okay sounds.

A

Great no later so, okay.

G

A

Yeah I think there was still some concerns like Stephen brought this up a few times, I think on the on the PR, for this thing that what we said in the ldm already right, that you know there's this concern that we may make it easy to use this fan base. The API is now located sorry this and that we make it easier to allocate by accident knowing that okay, this is the what that I think.

A

The key argument from them was, if you say, span off by it and then you're invited inside you happen to have an array, and you say, index, serve of range they're, creating an array that you're gonna carve down to a span of T, but the source code looks clean.

A

Sorry yeah, but I think that my counter argument for this is that pretty much thought that makes it already very easy to allocate that I don't feel obligated to making it harder in the sense that, if you really where performance oriented code, you can underline the language to block you from allocating like you have to do some measurements, like you can box in any given point in time like this is not the the thing that breaks in my opinion and.

N

The case that they're talking about is very analyzer friendly spot. Yes, take.

P

Him, if you take an array.

N

And you pass it arranged and you assign it to in its targeting the span, like that's a analyzers like wet dream like this, like what analyzers do? It's like? Nothing, just panic.

D

Like don't use the array range indexer in carfax code, it seems easy enough right. Well,.

N

That's one way to do it like an analyzer, even when it's like it, yeah.

L

A

N

A very straightforward thing to say that in hyper code we do this not this right.

A

I think the challenge today is that I think you know part of my team was involved with FX cop and they have wrote the line. We can fix it in FX clock too many lines too many times right so I think the problem is we don't have a product guys and let us set for core FX. So everybody believes that this is a pie in the sky thing where other than reality so I think that's where all these things are not necessarily convincing.

N

Ones and I would say that yeah that's the first day. It's.

A

N

Is is we have all it is Frankston corvex part is just bike investment and he said we have very specific patterns that we ban yep Rozlyn. We have a set of analyzers that are excellent at enforcing them like we. We require configure ate everywhere to our demise and I. Think.

A

Your concern from for the people, my team, is not that they're cool shitty is this that the the narrative of like oh, we have to add analyzer for ourselves, and so will our customers in order to not watch shitty code. That's the thing that that causes you know, but I think that what a totally.

D

A

D

Is not representative of the world.

A

It's what I tell people is like the the thing is right now, the last two years we have hired people on the team that have only ever seen performance tuning right.

A

That's the only thing in their career that they've done here right, so they're, very biased for oh, this is a number one thing we work on, but in reality is that no, we Ted ten years before fifteen years before we did other things and like, but the customer segment is still spending the whole thing and so I think that's something that just as sometimes harder to sell but dynamic.

D

If we were highly focused on that's a bad.

N

Link, let's make a feature that we might use on a daily basis, I think.

D

That a lot of people he's done.

N

Based on bug reports, a lot of people use done actually only basis, I'm just gotta, be emailed today, because the the BS team is upset that I use dynamic and Rozlyn, because I use dynamic as a cheap way of reflecting into the BS private api's.

N

D

Should feel bad I do not at.

N

All yeah my board that one, the dev team, with all the other private, it's actually public reflection- well, do do an assembly load casts.

D

Sometimes the dynamic calls enough okay, so it almost seems fine all right. So then I think we are pretty much done.

A

With the issues, memory does not have one because the idea of memory is, you have to get the spend. If we do anything. In fact, the only thing we give you is the ability to get a smaller number, so you.

B

Can slice, but you can't in dec, correct.

A

Because it is once your index, you should get a span out of it first, because it's more efficient.

C

As a shorthand.

A

C

Know, thank you. Do that no.

A

Because, once you do that, you write the code like for blah and I mean.

C

We want to make people only do it once yes, Oh, fair enough.

C

But shouldn't you still have.

C