.Net Foundation Design Reviews, 29 Jan 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Review and DbDataReader Additions

Description

00:00:00 - Approved: System.Reflection.Emit.Label should implement IEquatable https://github.com/dotnet/corefx/issues/32959
00:02:50 - Approved: Exception to throw when interface dispatch is ambiguous https://github.com/dotnet/corefx/issues/34124#issuecomment-458653253
00:22:45 - Approved: Consider moving several HWIntrinsic instance methods to be extension methods https://github.com/dotnet/corefx/issues/34902
00:29:52 - Approved: Take another look at the `COMISS` and `UCOMISS` hardware intrinisics https://github.com/dotnet/corefx/issues/34881#issuecomment-458659597
00:40:32 - Approved: SafeBuffer Span(T) methods https://github.com/dotnet/corefx/issues/33279
00:46:02 - Approved: Add string overloads to DbDataReader.Get*() https://github.com/dotnet/corefx/issues/31595#issuecomment-458677042

A

Alright, not only alive so I can repeat the last three sentences that I stole.

A

So basically I want to view these two issues because they're marked for a cure I can it for the standard because they either came up in the context of the standard or we just edit something from that area into the standard. Did it make sense to do it wholeheartedly? So basically, this one is an example of, probably don't you know, doc. My favorite V 1, V 1, point 1 constructor.

A

We have a label and that's used in a reflection in mid to annotate, respectively and I'll instruction with a label, and it has an equals that takes obj out for stores, because there comes some object, but it also provided the strongly type equals, but it never implemented ie credible, and so when the API is somebody said this I think it was America said. Maybe I should employ ty equatable who's just proposing that change, but is there any concerns? I would just move.

B

That I'm curious, do we like, if all the code we see, are these ever compared it is they equals.

C

Actually used I.

B

Can tell you in the sec.

A

I would assume at some point.

D

um I mean and actually Loki incest chemo to demonstrate I could.

E

Imagine scenarios where maybe you would want to put this in the dictionary.

A

But it is used across the board and the compact Labatt's 1.4 percent and api afford it's like half a percent or something so people call the strongly-typed equals I mean the other one of a probably not find any use of this, because it's a virtual and then the operators don't seem to have any usage.

A

This is the API shaped by the view of America's just like. If you offer a strongly type equals you might as well I'm tonight, equatable, let's know doesn't seem a downside.

A

Only upset I think.

F

He's been already had this problem tight school spent it and it matches all the behavior that you would expect for. The I equate able yeah.

A

But given that I didn't hear any pushback, all this market has approved the.

D

A

Many API labels.

D

Refresh this, you.

D

A

This is the part of the equal implementation of interfaces area, so he is name is Michal pronounce. It correctly proposes that ye and to use a new exception type. That indicates the share diamond issue when the type load of blows up, because it can decide vegetable implementation to use right.

A

So in this case you have Aybar and I-best both extending I fool, both colliding the implementation for prob and then, when you load the class that inherits all of these interfaces, it's unclear which one you want to you want to get because they are both equally applicable, but equally just.

F

Willing to make sure that I recall correctly the compiler when it can determine act, compilation and that this diamond pattern occurs, will error that desert said yes, so this will only occur for the runtime case, where someone compiles and then upgrades one of the dll and the upgraded version of the dll contains now continue to be Baltimore. Yes,.

A

All the platform, whatever you like, yeah.

F

I think I think having.

A

Yeah I think to me, like I, don't know like its I'm undecided because it doesn't strike me as an exception you would likely handle like I mean. How would you recover from that I I.

F

Don't think it's I, don't think this is a custom exception type or something that you should financially explicitly handle, although you might want to do that for reflection cases, the.

G

Derived the derived interface, could I mean it: it's like an execution exception, and then the derived interface, if it wanted to disambiguate, could implement it great.

F

But I also think this is a case where, having explicit exception, to provide the additional contextual information is worthwhile.

E

Why would this be under one tiger, not reflection, because.

F

It's only thrown by the run, it will also be thrown by the runtime and down reflection.

A

But to me it almost sounds like the typical type load exception right, yeah.

G

And this doesn't feel like it. You need to.

F

It's not a pipe load exception, though the types can be loaded and they can be valid. The only problem is if you try and invoke the method in an ambiguous manner that it will fail. Oh sorry,.

A

So that if we know the class yeah.

F

You can load the class and you can call things and you can disambiguate yourself, but if you happen to cast to the shared interface between both of the implementations and then try and call it, it says: I don't know which default implementation. Okay,.

A

That is much better to recover from what I thought. I thought you cannot know. You cannot actually load class at all. Well, if he okay, if the, if you only blow up in execution minutes, you know it's still bad for the program, but at least you can in principle discover well you can you know higher up. You can decide to do something like if it's a plug-in or something that you want to. Imagine you just so.

F

So like in this case, where you're calling from you could disambiguate by saying oh I, definitely want the Aybar one or I definitely want the iPad one. So you could retry by doing last, UI Bardolph, oh yeah, I, see.

G

Can you scroll down to show the other one, the other exceptions that were considered.

A

The reflection one in because you know I think this one is I- think ambiguous matches. That example for that, because that's not what you want but ambiguous matches. Really, when you invoke the binder and you you know, it cannot decide which of the overloads to pick effectively right.

H

It's a very different error case. Well,.

F

It's basically what's happening here under the covers. It's just not necessarily happening to you, which is the confused.

G

What about something more generic, like execution engine exception or like, can we can we brainstorm some other alternative, existing exceptions and and ruled them out? It's.

A

The execution engine I would rule all because there's one of those that you can't handle right. You know that mean you don't feel you can, but you have to opt in which I don't think they're supporting core thing in court. You cannot handle it at all.

F

Only had the message I would I'm not sure that it's worthwhile I would I. Think I would also want to see additional info on there. Rather than just a message like possibly be. Two types would be portrayed.

E

F

It couldn't be some big Ewing between if.

E

If we did that, and that meant that this exception wouldn't really be usable for other scenarios, it would be locked very specifically to the timing problem effectively, because otherwise the properties wouldn't be populated or would be nonsensical.

F

The message is trying to return yeah.

E

Yeah very clearly kind of well, but that's just a string right like we can always put other strings in there.

H

You know what thrills invented.

E

Program exception.

H

E

Anyway, no I think that's if you try to like load of a 64-bit program in a 32-bit process,.

G

There's there's other cases, I think it's IL validation, okay,.

A

Yeah I think executioner, in the exception, is in the Salah detects. Corrupted stage was definitely fatal and yeah I met a program. I think I've seen my night when I had that I have.

G

I think the Impella program is actually I it's useful to to limit that one I recall it being very specific, so we shouldn't reuse that one for sure, yeah and.

A

I think the one you talked about with something with image right image: oh yeah,.

I

D

I

Something image that image format excel yeah but didn't.

G

I

G

We have type initialization exception, type load exception. We said those are both out of it because there's not type loading happening here right, that's a method, call the types loaded, fine, it's a method called it's family.

G

We could say it's like method, not found exception, except it's the opposite.

J

A

Many methods, you know you could butcher into submission, let's target invocation exception, but I'm not a fan either.

E

Like how you phrase up butchering this again, yeah I.

F

E

F

Think the exception name and the place are probably good. The.

E

System, suspend.

F

E

Time, there's a little weird just because they're so little.

H

Yeah system, one time it so system.

F

Run time, compiler services, it's.

E

Not a compiler that was a long time for the.

F

Deca crew still uses things in compiler services, yeah.

A

Any compiler services meant for things that the compiler invokes, which users never ever have to deal with. Well,.

F

A

F

Also put unsafe in there and method in pulls in there. It's basically things that coordinate between the compiler and the runtime that one or the other needs to discern additional information works.

D

Well, as a few yeah, we did, we should and.

F

I think we even have an actual target. You.

A

Think of my shows is meant to be like yes, he said like integration within the compiler and the runtime, which is more for plumbing system. That runtime was the new specific first have invented to put the crazy things that runtime folks wanted in there like. There's memory fail point target the patching off the hour, but basically things you never ever use.

H

Yeah, if we do a good job of hiding this from people, because no one's going to run into it for the most hard and.

F

It's one of the exceptions that will probably be classified as you should never throw this. The runtime, which we already warned about on a couple yeah yeah.

A

And that case is the name the best one. Should we have the name interface in there somewhere.

F

I think ambiguous implementation is probably good, because it would, if the compiler ever decided to allow the similar thing for non interfaces as crazy, as that seems it could be read. Would.

E

It make sense to use the word target instead of implementation, because we can't figure out where to dispatch. The whole thing.

A

I hope you in my target eyelid usually refers to the receiver. Like the instance. They say that what this place, you don't think you want.

G

But member ambiguous member exception.

D

But that one would you put you very close to the reflection, one right, yeah vigorous match.

A

As the implementation probably is a good term because that's it really describes the mechanical parts here right.

A

It's concerned that the name is too generic people will yeah run into the type and try to reuse it right. Well,.

F

We can always request that the compiler do the same thing or do whatever we are doing for other types like index out of range exception. If you try and use it, you get a warning saying you shouldn't use this. You should use argument range exceptions, then really yeah, because index out of range is only meant to be thrown by the runtime, whose.

E

Morning is that I mean a good way to make sure people don't throw. These is to remove all public constructors and say only the wrong times ever supposed to throw these things. I mean that I realize that doesn't follow the pattern for exceptions, bub.

A

I mean and that's actually another good point like we should probably have to be flow constructors for those guys as well.

D

And photo gemologist.

F

Looks like it's probably.

A

They're sure that's not one of the endless on the rosin Tebow, because I have never seen this morning. I've never heard of it.

G

You never do the bad thing, maybe.

A

The bad thing, but by the way Eric are you at home or you just do Daisy to walk over the 50 meters. The.

G

G

I'm more efficient, this way, I'm just too lazy to walk over.

A

Yeah I understand that I mean, but in your case you have no excuse really Steven, but it's only like what 3000 miles exactly like I give a tick.

A

Yeah I mean so the other question that I have is like what is our stance on Cilla Dacian constructors, on on exceptions that we're doing them or we're not doing them from the exceptions from.

I

Newark sufferings yeah. What was the one that.

H

We just added four on the switch matching yeah switch Express, which some things which.

A

K

Bust in for adding sterilization constructors to new exceptions. Yes, yes,.

A

And do we market utilizable or not? Yes, just for exceptions. So basically, since aim is always be Mackin. Serializable we implemented fertilization, constructors and pejorative only means you should they add all the other constructors right or the other to really list agree, but the one that takes no arguments and the one that takes the sterilization of two things. Yeah.

B

I think for exceptions would be worthwhile just follow the pattern and the reason we do. The serialization constructor for exceptions is just that. You know you're marshalling exceptions, you have no idea what's being thrown and we don't want to make you you know only marshal ones that are serializable. So, basically, all of our exceptions are sterilized. Oh yeah.

E

Yeah I, don't know what the exception we just added a few weeks ago was because I can't look it up.

H

You know I something-something switch a bit assure us which expression exception or something which they lose. Maybe I.

A

Or maybe some of the title I think that might be it, but I can say it was followed by Neil. If that helps.

A

Yep match made an exception. That's the rival, but I think the name we finally approved was something switch. I believe.

A

Yeah Marian just edited so.

F

Pr, probably I'm not getting it looking for I'm being seizure.

E

So standard exception instructors, market serializable.

E

It should probably put its a plus exception directly on point on. Okay I was thinking, install a system of socialist or I mean.

A

The polymer system exception is that it take the whole system. Application. Exceptional hierarchy is just completely busted anyway, but I mean if you want I, mean either way. The statistic like that you know I'd give the original spirit way like it's, not something that you expect the user to ever, throw.

A

So I can go either way. I, don't think system except one adds anything anymore, but.

E

Yeah 50 class, be sealed normally, exceptions. Aren't some.

F

Of them are especially in the cases where we don't expect them to inherited from yeah.

H

Like that reference exception, then.

F

Runtime wrap, which.

H

Is not see if I can believe in runtime.

F

Wrapper exception it that's another one. That's in our service, yeah.

A

F

L

A

Probably seal I guess in like why would you ever extend this same.

I

Reason that argument exception stuff, like.

E

That is what sauce sealed, because there are multiple different kinds of it: users.

F

Are meant to grow? It's also a non breaking change to unseal, a class and here's a breaking change to see it right. So we could given that this is a runtime exception and only meant to be thrown by the runtime. We could seal it and if we determine it leads to a standard later yeah.

A

Exactly so, I think this one of those kinds where I mean you could imagine that is Turner said well. It's a compiler extends to feature two more scenarios that maybe there's more of a hierarchy behind that right, but I mean I, don't know why anybody ever would otherwise air it from this. Well.

E

Then we come back to what I mentioned a few minutes ago, where, if you want stronger type of properties from you know, Ferguson and now the exception is- is applied to more scenarios to those properties, make sense for those additional scenarios. I think.

F

I think we should review the additional properties as much as I would like them, probably separately and later, when.

M

F

From the language team on, if they think that it would be in rem, I would.

D

Go to the terminal like.

F

I think it would be useful to give a the list of pipes that it couldn't disambiguate from that way. You know which that way you can easily determine which pipes are actually he's the source of the problem, because.

E

You're thinking someone's looking at the exception in a debugger and inspecting the proper.

F

Right and it's easier to do, do it that way, then just reading the message, especially since it doesn't. The message here doesn't look like addicts, but it says like I, bar and iPads are the ones that are a conflicting.

I

And let me more working, you think it is because you have to well it's.

E

Not just it's the method, signature like the interface name is probably going to be the fully qualified type name, which means the exceptional sperm is going to be in terms of language right.

A

Which, I think is probably fine, because if you ever encounter this guy, you probably wouldn't call it going on, like my favorite case, where that this mess is math right with math catalogs explode and you don't get the would exception, and then it takes you like 20 minutes, while in the program that walks the exception hierarchy and they get so the inner of high-blood exceptions and bases just concatenate. The giant string with you can actually reason them out.

E

Oh yeah there's nothing inherently wrong with long or detailed of such a mess. It is it's just we don't want to develop. We want a developer to be able to very quickly reason about something rather than having to turn on word web.

F

But but like I said, I also think it's probably something that we'll want to discuss more with language and on time team to determine. If.

M

They think it's.

F

Feasible and what they think would be appropriate.

E

E

Did that thumbs up just go from two to three while we were talking yes,.

F

E

A

You was the other one, yeah.

F

It was everything John and then I got I.

G

Was convinced I was convinced, so I committed my boat.

G

A

You what I have in old, probably have at least three titles in there.

N

That should be fine.

F

The constructor that takes an inner exceptions are we going to say this probably never happened, interception. He don't need that. One.

E

Well, that's exceptions, as a matter of course have that as a constructor.

F

Right, but if not listed here and I wasn't listening, yeah be.

E

Sure to edit I I figured. This was a proxy for at the standard exception instructors.

E

Okay, that's just add this guy, which means that the serialization constructors body that yeah I guess but we'll just fall about it. Yeah.

A

That's information I've been here the other things that can yeah all.

D

Right API proved.

A

Wait alright now let me go back to the list. So what are the things that you guys want to look at? Oh.

F

Yeah three: four: nine: oh two.

A

I'm pretty sure, that's the only feel that month, looking for three four I.

F

Mean you can just take the PR and philosophers yeah.

D

F

D

You were all happen before nine or two.

F

So the issue here is basically we've determined. We.

M

F

The helper methods to be on the vector types themselves, as instance, methods. We have since determined that, because their instance methods we actually have a perforation and it causes the JIT to hit in liming limits faster because it forces the c-sharp compiler to emit additional locals and then load the address of that local. To support of this parameter. The suggested solution was, we should take all of these instance methods and make some extension methods on the existing non-generic vector classes, to avoid the to avoid the problem, but still allow users to write the ODIs.

B

And do we think this is my? Is this the right solution or is the right solution to fix whatever it's polishing? It.

F

That the way to fix it is true is to make the this parameter no longer pass by rat, which is making an extension. Sorry.

C

Earlier, your frizz, you mean they're.

B

Fixing the c-sharp compiler, whichever would be more feasible, I, mean the answer. Those are impossible or if the answer is, we would have designed it this way, as extension methods anyway, then great but I think it.

C

Feels weaker to.

B

Change the API to work around a current issue.

F

Fixing the C sharp compilers not possible, it has to emit it as an address and then for the JIT at least. The comments that Andy had made indicated that it was extremely non-trivial. They would have to look ahead during the il importation phase to determine that the address can actually be folded away and they may have the special case the methods to support, all of which would be non-trivial to do.

F

And it's also likely not general-purpose. It would be specific to the interest.

E

I see yeah, Tanner and I had a discussion about this thing. Yesterday like it's, it would be nice to do the jet fix, but it's just way too much work right now,.

B

Alright feels weird, but it's the best solution.

A

But what would that look like you have like a list of API yeah.

F

Yeah, if you scroll down a little bit, it was small and it.

A

Was sleeping on the same time, it would just mark them as extensions. Is it.

F

No, no so right now, they're on vector 128 of T, as instance method. They would move to vector 128 as static extension method, which is where we put all the other static. Constructor method like create and like that.

E

We have an inductor.

F

On back twenty-eight of tea, we we have get element in with element, because, because it's read-only right.

E

But we didn't put a read on the index or e. We just have to get nothing right.

F

Okay, I believe that's because we said the standard pattern for this stuff was to do it this way. Okay, so we don't have to worry about that them.

F

Yeah and I didn't explicitly call out that, in particular, the ones that are the perk concern are as to scaler to vector, unsafe and get lower. The other ones don't have the perk concern because they're not actual runtime intrinsics, and so users could work around it if needed, but it's probably just best.

D

You well I, was in, would they be different, homeless boys so.

F

When you have a when you have an instance method, the first parameter is hidden and it is implicitly wrath of struct right when you have an extension method, it's just a struct passed by value type, unless you explicitly say wrap it. Oh so you know, so the compiler no longer has to create a local and then load the address of that local, which means that the JIT will no longer hit the MM limit when it hits 512 locals.

F

That's quickly, because.

M

F

Heavily for converting between, like the various teas.

B

And we're not gonna find that there are cases where it actually ends up, resulting in a copy of a 256 byte structure or something like that.

F

No, it sure these are all read-only structs and when they're passed by value, they're, implicitly not mutable, and in the case of the JIT we actually elide the abs entirely and just mess with the type system to get it to be the right thing.

F

The problem is that the JIT still can't just alight the filo feel that it was creating.

E

Is if you know, if you recall how il works right like the MS il is a staff based system when you're passing a parameter by value? You just push a value onto the stack and then call the method or, if the values already on the stack, you could just do the bit and then I'm all about that right. If you're passing something by rath, you have to pass the address, but you can't take the address of something that's on this tab.

E

You have to take the address of a local, which is what Tanner was talking about now. The compiler is the emitting, tons and tons of locals into the method.

E

Was lesson of the job.

L

A

When you could imagine the other things that we give you a way to specify what you want this to be reimagined rather than yeah.

H

Create extending by hand but yeah yeah.

E

But now we're talking one time changes, compiler channel changes now I get it it's unfortunate.

A

Yeah, this is absolute I, think one of those times where I'm not super conservative, like the most usable IP answer. It should be good enough. Well,.

F

I think the API services, the usability of it, doesn't change unless you're trying to invoke these be a reflection, yeah I.

L

F

Think people should be moving these, my reflection, and if they do, we can at least tell them it's centralized with all the other open air effectively I.

H

A

Extension methods remain second-class members right like that that it's not in your face quite as badly as aesthetic would be, but it's still somewhat in your face at times and if.

E

I'm in a debugger and I'm Indian media window, I can still use substantial methods or I trust you working almost in two hours, yeah.

A

But what using, as you get I, think you're the you might get them from the current file or something like this she's, probably with you.

E

It's good enough for you, yeah and we'll miss scenario would work, because, if you're using vector, 128 you've already imported the namespace with all of your subtitles you this.

F

Is just a case of like, like we did with span with some of the cases where we didn't want it to.

F

Hostage it to explode four types, we decide to make them extension instead,.

A

All right, so anything else. What was your chance every do? You want to agree to anything during me.

A

Other message in there random things, I've.

F

Got one more that I think was pretty trivial. That's good huh good one. Three, four, eight eight one, three.

D

B

Strictly Remo you're asking about any others of importance before we go through the general list: yeah, okay,.

F

Basically, for this, we currently are exposing some of the comparison intrinsics as compare the equal, ordered scalar and compare equal unordered scalar. It was raised that this might be confusing, as it might imply that the that the result of the comparison is unordered, whereas the actual functionality is that the comparison itself is unordered. So it would suggest to definitely reword this to either be unordered. Compare or compare on, ordered, basically just moving the ordered unordered keyword before the equal, rather than that.

F

How does it change that, because now you're saying that, instead of the quality being water grounded in comparison.

F

You saying compare unordered rather than equal.

E

E

We do have precedent in brain working for like string methods we have like equals. Ordinal compare ordinal stuff like that, where the the modifier occurs after the compare verb in the second bullet you have here, but people mistakenly believe that the scaler itself is ordered or unordered. That doesn't really make sense. Conceptually.

E

So that, like I'll, be a huge concern, you.

F

Mean in the existing water room, where it's underwater in scaler, no.

E

So I'm, looking at the you see those those two bullet points where you have a and a or b n.b I'm, looking at where you have compared unordered and compared ordered in that order, is it possible that people might get confused and think that the the equal scaler itself as a noun is the thing that is ordered or unordered I? Don't think that makes sense conceptually? So people probably wouldn't have that understanding right.

F

And I think, given the the context of the ordering of the existing names, where we put scaler it's doing a scaler operation, it's much like a sink in that regards ok, I. Think I. Think that Clara at that part has.

E

Given the other api's that are in the intrinsics types, almost all of them have the verb at the front, and it would be nice to keep all of the compares together. Rather than have some of them with compare and some with like unordered, compare right.

F

Which is why my preferences second here.

A

F

A

Page, but do we have compare an equal? Why don't we just have equals?

A

How what do you mean like? Why do we say, compare equal? What I was just saying equals.

F

Because all the it all maps to a simple instruction, which is compare it's the commas in you- comments instructions.

F

Which is compared on over scalars and err ordered scalars and there's multiple operations for that instruction, equality greater than less than.

F

A

This one does only equality, but yes,.

F

So there's you get a bull.

A

For a limit or something you.

F

Get a mask of so.

A

It's basically a boolean result, project.

F

Li, well it it's a you, it's a vector, 128 result, yeah.

M

F

And it's not a pool because you can have like all ones, all zeros, all ones, all zeros forgiven, which says that the first two were equal and for the first one third were equal. But the second and fourth elements were not right.

A

But anything you get per person per Vader, you think if we get a boolean, is that what occurred right? Yes, but yeah like as opposed to somebody, indicates which one was louder, which one was smaller right right so, which is why I said like in? If you look in the family that you only have compared to indicate basically the in based resolve, it's like larger or smaller or equal, and then we have equals, which is basically just the boolean.

A

That's the only reason I'm asking it seems odd that they would they compare equals because also makes it longer all right.

A

But yeah and much of the other one reason much that's the same problem.

A

It just makes the second one I, don't know it reads more complicated now, why'd you say come I would agree with you that I know that, in order that the first words seems bad, you won't ever.

F

A

The first one, but the second one still meets really really say compared unordered equal scalar. So it's like you need to basically slowly walk in your head, the secret, the is going on a baby. That's the way of these intrinsic sin, general yeah I can go either way. I can leave it, as is I can also like over than you want. I, don't feel super strongly. What other people feel about that.

E

This says as a member question, so it looks like this particular instruction also sets three flags. Would we consider, maybe in the future, having methods of just say equals of order.

E

F

In this scalar case, we do return ugly if I want to imitate them. Okay got it and there was a PR app to make the cool comparison to the optimal thing which sets the scene.

E

Trying to like I'm reading the documentation trying to wrap my mind what this instruction does. It's.

F

One of those instructions where it's a single instruction that has like 32 operations but luckily they're comparison. So it's easier to reason about that.

M

F

String manipulation.

A

Interesting well, Ben was suggesting you pull the skater forward, so you have compared scaler, say: compare scaler, unordered, equal and compare scaler ordered equal, particularly they are all like prefix by compares gala I.

F

Think if we were to move it forward, I would want it to be compare, ordered, scalar and chimera unordered scalar, which matches the naming will be a purple arrow and then put the modifier of the comparison operation. It's been like rolls left.

A

Actually that makes the double to me yeah, because now you basically have a prefix or two prefixes effectively, one for the one fun ordered and then all the combinations at the end I think that's actually the nicest way to group them.

F

I think that would also match with the name we have to you for ABX to wear.

F

Rather than having the individual equal overload, we do it as a union value because there's too many, it was individual.

A

Yes, let me just record this seems like we have fun the thing we all kind of like.

D

Oh I won't have it here.

E

I'm glad all this work is getting done as painful as this process has been because of how much work there is yeah.

F

My good we're basically feature complete in our it's just people using them and saying: oh, we broke perfect. These changes and these names don't make sense. So as long as we can get that prying get those fix, proportion idea.

E

But the pain has absolutely been worked at what this news feature is enormously. The.

F

Remaining issue is on what should we call a pest, the z-test sees and test not see and not see, whoa.

G

F

I've not come up with proposal for what motion.

E

You know if someone's going to a final up, I'm saying we want direct access to. He pledged forty stores from seafoam.

A

It seems people always want more access unless they realized that the three.

D

Releases of doing that, the code isn't right.

F

But I was also a Jew. This.

A

Is so I think at this point the streamin review next week? Db we do a review today.

D

Yeah I got and eopns died over the other side.

D

This is something that we have enough context on Stephen. Do you have context on this.

B

Yeah, right correctly, there are some existing methods on safe buffer that for reading and writing, and this just creates span based versions of invasive.

A

Hold the safe over live in wherever API, no.

B

It's like dotnet, 200 or something along those ones.

A

B

A

What do you use? Because it's forever.

G

Yeah we brought it in here in that neck or why now I believe if.

E

You guys have done that shows that is for pairs for framework for oh actually, ok, that's recently, I mean 10 years, but yeah.

H

Well, the APS do we have.

A

A

I guess we explain and read: spam will be consistent with reader away and right away right.

A

You ever have a fundamental concern in general seems like a type that would almost go out of my guess, not really because it's save his life time right.

E

Well, this type is supposed to be a safe view.

G

Yeah, it's a disposable thing, so this has to this have to be copies.

D

How do you get the sizing that guy, oh.

E

Yeah has a pipeline poverty, you can say no, it's all once these things are constructed, they're immutable in terms of the properties that they have even.

A

How do you construct it with the sine of U.

E

Sub classic, oh you.

C

Do oh I see you're supposed to do it from that I see and you call initialize to specify how long it is.

D

Okay now make sense.

E

And the reason you have to subclass, it is because you need to give it a disposable.

A

Biggis you're stupid: yes, we decided how you wanted manage memory of a wall all right, any concerns, then otherwise our disapproval.

B

I had one question: are there constraints that we've added since Donna Faro? That would be reasonable to apply the teeth like? Should it be unmanaged or something like that or we just wanted to be struct.

F

Be unmanaged I guess we could here because they poppers not itself, isn't.

E

Drink like didn't, we decide against using that for the api's because of our difficulties to use do.

F

You mean because it's limited to c-sharp, 7, 3 and above and it has the mod rack constraint so compilers that in languages that don't understand it can't quit involved. It's.

E

Not your snap but say to say that I'm, currently using safe buffer right and I have a I, have a tea array inside my own code and then I call Reed array or white array. I now would not be able to call Reed span and right spam because, even though I, even though I have existing code, that has constrained to you to be a struct t, is not further than trying to be unmanaged and it's kind of a viral constraint. Yes,.

A

Yeah, the only reason I wouldn't add the constraint here is because all the other things seem to be constrained to where T struct and it seems a bit weird. If you want to move from window ready to you know, reach span. What would you have a different constraint? Well,.

E

We probably would have used on manage and we developed this type wall and managed was available short, but yeah.

A

You're, given the as I said, given the violence, it seems odd to have difficult friends and members at the same time or and.

F

Then the new unmanaged construct intact.

A

Right without I mean all constraints are somewhat viral right. If you would write generic message operating over a safe buffer and you have mismatches in the constraints, you're you're forced to forward them to your series as well.

A

A

A

Yeah, it seems everybody seems.

D

To be generally in support of it, alright, let's do this API.

D

D

E

All this doesn't work. Is there an API here.

D

Don't think so.

E

We might be taking item for the Terry. Also I saw that Steve near the top left to call them to with an API in it. But I don't know yes, no, that's the yeah.

A

And that one I think it even for the beckretary to come back with an action.

I

No, we don't have time for us: okay, yeah, it's easy, but with this one we should have time for this one. Yes,.

B

So the reason is besides this mean we're actually considering using obsolete. Now.

A

Would be the economy have used obsolete in the past? It's just that we well I should say I generally, don't believe it, or that observation is the best way to defecate. Api is in this case, though, for individual members, it's not horrible, it's more like for. If you want to say, if you want to defecate a feature, it's usually harder, because you have to find the API, as you want to officer lead in many cases that it's just noisy, but not actually the actually usage and.

E

You can suppress it easily, the anyway for context. The reason that I had proposed officer leaving this API is because it doesn't play well with the strongly duplication feature, that's being worked on and.

A

So the usage is fairly high. No I think we, you and I had this discussion on like what the usage means. So I didn't any of us as a white hypothesis on this. But you know the net effects competitive usages. The thing about 20%, if you have four, is about 10% and the other ones are I, think the highest ones we have was from some Android apps, which was like 60% or something.

A

But that seems more like a the template issue or one popular library, because they really have a hard time seeing 60% have developers calling this API.

A

I mean the control generally, making things obsolete that isn't that is used. You know five or ten percent is like it's super noisy right. Maybe.

F

This is one of those cases where, rather than obsoleting it, especially given how uses we should add it to the core FX analyzer, it guys we have, and maybe we should do a blog post to help raise the visibility that we have for FX, API analyzers. That say you shouldn't use these api's and.

F

You should consider doing this pattern when you're doing when you're running this energy.

E

So the reason this is a little more interesting than that is sometimes when we obsolete methods. It's because something better has come along. The reason I put obsolete on here as a proposal is because, if you use this API you will will not might will be broken when this feature comes online. Yeah.

B

I agree with that, at the same time, the moment this says, obsolete people are gonna, go in search of a workaround and they're just gonna say new string. You know slash zero with the count and then just copy that you know fill in that string and they'll be equally broken. We.

E

Had actually proposed making that particular scenario work but yeah how.

A

Would you make that work? You would return? The same string instance know.

E

During deduplication, we would look at null strings and what do you do for me, but.

B

It wouldn't be, it wouldn't be now all the time they only do. You know some of the time right. I, don't see enough work, yeah.

E

So what does copy dude? Is it actually coffee? It literally makes a deep copy of a shirt, add more terms of new instance, because.

H

E

Wants it, as it goes alone, which returns us yeah.

E

I

I'm going to type this object, Jesus Christ! What movie thinking back then, because it's like one of all.

E

Oh, no, really, you know my clinical returns object. You yeah. We.

A

Could have sent a minute explicitly and heard an actual method. It returns the string, yeah.

H

But that was like dotnet one, oh yeah, they had X. Isn't his implementations in you want.

A

Actually, what actually, because someone I don't know, maybe they were broken, there was a bug, something.

F

Maybe it was also at the point at the point where that would have been considered. We determined that clone was just evil in general, shallow or deep or yeah.

E

So, but by the way, Steve, like your point, is well-taken people might just work around them. Hopefully, the message on the obsolete attribute, which says to do what you were trying to do do this instead, hopefully that would work.

B

What are we saying to do instead,.

E

Here I have used string, guitar array or, if you're doing inter off, you is one of the Marshall types. Maybe.

F

A new string to star that way: it's not limited.

F

Any of the others.

E

Without I guess.

B

My concern is, if someone's using this I mean I, don't.

G

B

A sense of how this API is used, if someone's using it, are they doing it because in bhimli they're, then using unsafe code using on yeah maybe take care of performance if they care about performance, they're gonna, you know cry when they see to.

E

Chartering, no so the acid we write be the typical use cases I've seen this used for are I want to make a duplicate of a string and then use unsafe code to change a handful of characters. Inside of them. You.

F

In the recommendation than be to use stringbuilder.

I

But let me put the email address in the in the mail or sleep message by email address, so they.

E

Can talk working.

I

On the DD paper, I.

A

Think that should be the Bob Bob solution in general, like if you willin to put my dress in the message, you feel that strongly, then you can possible. I could put enough to be linked in here.

L

Like that, okay.

J

Which one they have W legal, the email address.

A

Yeah I mean I think this is something we should consider for it like in general. If you should have Bobby have a border content. You know discussion on I got any other API Simone, also natively over. That makes sense. Making a one-off decision. I am I. Think Stephen has a point that we have not obsoleted things in a long time, and you know I think I have talked extensively about how broken off solution in general is, but.

F

A

Also, don't have a great answer for this I mean the only counter argument I would have, for your case would be if the workaround is more convoluted, it might be harder to find when you have box as a result. Right, if you literally just all this- is one method that probably indicates a bug that you probably use when you do this crazy workaround there at least you can find those things so like that might actually be better. You know better to the API.

A

In another thing we could do was like hide the API all right, I really don't know what a clone or copy should really be used for anybody would assume it's. It's.

E

Not that because I I want to cause noise in code when people recompile right, like I, don't I, don't just want to market editor browse will never like I want developers when they recompile. You have a warning light flashing in their face, saying you're calling this method this will have on it like this will have undefined behavior. Don't show that.

A

You still would also not get you cook using it. Man, no.

E

But people who are migrating their home or people who just have existing code and then they have like a nightly of CI, build where they say hey. My library is now compiled against three own set of two.

E

F

I specifically meant for four to one: that's energy I mean.

E

That's people to drug that fit or I.

A

Mean if we were to make the change about blue this way, I think we should consider it. Okay,.

E

So well, what are? What do it sounds like we're, gonna mark this. Those needs review them. Yeah, I, think I. Think.

A

You I think it would like to have a few more people. Don't have thoughts on this, but I. Think in general like are you opposed to hiding it or you're? Just saying, hiding itself is not good enough. Hiding in itself is not at all okay, but you also would be opposed if you would also hide the ink that we.

E

Should be doing that in general anyway, with the empty eyes that are nonsensical.

E

Oh, we can't hear whoever's talking right now, so.

A

You're very, very quiet.

D

D

I think we'll go to the second topic, okay, which I hopefully can find the yeah. For ah here we go.

D

Sit down, though,.

I

No reason to leave, you can supervise your opinion. I have started.

E

The other thing remember, we decided my opinions were strongly held incorrect. It.

A

Makes you join the other call, sorry guys it would drop you off so makes with everybody else run. The other call that we have everybody on the same call.

D

Great thing with gives me.

A

Here, would you mind joining.

O

L

Going going going going, join, join.

E

Join join, join, join.

D

L

Like the same technology is like this, can.

A

I can I, is there restated we figure out to we? Would these guys.

N

D

N

D

D

A

Is this light on PC? This is an action running. A piece is a surface. Oh my god.

N

But it's not as awesome.

A

What do you mean there are? No noise services is not an hour.

N

Unitedly, have you out there.

A

All right, sorry, guys we're waiting for the conference room to be would.

H

Lexi story: yes, not starting yeah.

D

D

A

A

We're moving the conference room with Non Aligned your views.

D

Couldn't sign into Microsoft themes.

D

D

How do you try in yeah there's no K but I'll retry.

D

Oh boy, oh here we go.

A

A

Nope stainless you.

D

D

A

Can you guys hear me.

F

A

So I cannot join the conference rooms. I join my laptop now, so hopefully everybody in the room has to speak up a little bit. Otherwise, it's not gonna work, but hey we're flexible.

A

All right will you have people from the F team here, so the ego is here and your name is again shy. You just joined right, that's right, yeah, so.

M

Do you want to walk us through this? Yes, so this is what this is a relatively non major issue in the API for ad or net. The idea is to make in general type a bit more useable, maybe.

F

M

Talk about it more later, so this attacks a very specific point, which is that we have many accessors on DB data reader, which is the main way for accessing database results. X DB data reader already has a lot of API is, but the problem is that all all field access is currently ordinal, it's only by index.

M

So when you're getting your results back, you need to know how your columns and the results are ordered and you need to specify an index, and there is a utility method called get ordinal which gets you the index given a column name, so you can chain those two things together, you can say, give me the integer for the syllabus, so the idea here is to simply add overloads that directly received the column name and provide provide the result. The Calma.

M

Been some discussion whether there should go inside or extension methods if we scroll down but I'm, not sure that's relevant anymore? Why so? There was. There was some discussion around not putting this inside dotnet standard, but rather having. This is an extension with M you get because the proposal, the proposal. The proposal is for these to be non virtual, specifically non-virtual, because there should be a reason for any provider to actually override them. They should have the exact same behavior as if you are you don't want to allow any sort of behavior variation there.

M

They should literally internally each one of these are just to get order not to get the order. The number for the name and then called the overload of the same method that accepts an order where you want to lock that time.

A

Yeah it's hard to imagine, however, implementation could do better.

M

Than that performance wise right, yet there was a question of doing any sort of caching, so caching from the transition from the name to the ordinal, but that's already supposed to happen inside the ordinal, which I would be wrapping here. The performance happens at a lower level anyway, yeah.

M

That's that's basically. Yes,.

M

You there isn't there's another issue like a somewhat orthogonal one about allowing overloads or some or other versions that return value tasks. The issue is that at least for so we have 22.2 acid methyl here.

M

Basically, the idea of this the scope of this one is to provide matching methods for the methods that already exist with their. So it's only to solve the problem of inter string. Then, if we want to also solve the problem of value task, then kind of considered and I think it's an orthogonal question. So.

N

Would you recommend to use value task even if existing overload that takes a the ordinal, because.

N

N

I I would say the only reason we haven't considered it. As for this that we have a another issue that we haven't figured out yet how we're going to fold it the risk about a lean body task of the versions of this idea and the main problem. Is that all the reason our names for a px are already taken and they return tasks here. We need to come up with a new naming pattern for, for these methods, return value, pokey, and if we read that, then I would agree.

N

It's, let's only add the methods that return values at 50 and not iranian-born, lack of the contract of feedback to produce the returns of that and reduce intentionally reduce the scope of BC should lead us. The things that we already know how we'll do right.

B

But yeah I guess that all makes sense. My question would just be: let's say we don't figure out. Oh wait, I'm sure we can. But let's say we don't figure out a way to do that, and so we don't have any of those existing methods, overloaded to return value tasks of T. That means that we wouldn't have them for these two new ones being added either and is. Are we are we more concerned with the performance or a more concern with the consistency.

N

I said this version of the metric supposed to just delegate to the existing one. I mean this is not a little method. All providers are going to get it.

N

B

Sick you're saying that these are necessarily going to only target existing public virtual surface area.

B

These are only going to these are just wrappers around existing public virtual surface area. Then there's absolutely no.

P

Bennet, it's a returning value task, do not add the the acing ones, and we say what we wait: that we have.

N

The other story so I mean.

A

You could make them virtual right, I mean you could say they get feel the baby icing and it's DB, not racing you make them virtual make the default. Implementation goes through the old path and then say: if you really want to optimize this path, the provider can override it and then the other set from that. So it's.

N

Just interesting, but you really care about performance. This would not be using this method anyway. Do.

A

N

It should be- let's say, for instance, you do have an RN and you are building the materializer about this product or and what you do is. Basically, you create a hard coded map of the organelles anything that you know, and then you invoke the holy hell version of this ideas.

N

Otherwise you are always going to go through that in direction from the name to the organelle, while you are paying the price for every single field of every single material. So but this is not an API that we are adding for performance, we are in it for incivility. Only oh I would prefer to just wait until we solve the other problem before we have the facing ones, or we just add this and we say well. This is just to wear over exiting API and doesn't matter we will you solve.

N

The other problem is going to carry different naming pattern anyway or.

M

I'll just add that, with the other, the other problem, so the the new methods there's other problems that we want to fix like the problem with handling null values. Coming back, so there's other problems when you want to solve, and in any case and that's why we haven't yet made any sort of proposal and then there's, of course, the argument unrelated there's the argument of consistency of having these you know having everything here, be the exact mirror of the diversion that's getting an ordinal except for get field value async, which would be a different one.

M

It would be virtual all of a sudden, it would be, it would be different in its signature. So it's also an API consistency. Question.

N

So she already said this, but this that's already has too many methods adding and an overlap. A lot of are accessing with little variation. I think it's okay, it doesn't add noise, but if we like complicated thing, we're trying to walk a line, basically between adding value and how does it know, yeah.

A

I mean it seems like it is beneficial to think about how we could use very task on the reader in general and then figure out what that would look like in terms of API patterns right because you it's yeah this problem, almost everybody else in the paragraph is extreme, but every single time you have an extension point and you have derivatives. You introduce a new core concept like how do you float this to the system and not break people right? That's it's generally hard.

N

If you have any suggestions by the way for the name, your father, who like what somebody does.

A

One thing we have this already is like Steven. Do we already have a case? If we did this in the forgetting right now,.

B

Well, yes, but we haven't had to deal with the naming issue, because we also changed arguments so they're they're, true overloads that return value tasks so, for example, like streamed on read async in addition to having a new over the term value tasks. That overload also took a memory or read-only memory instead of a byte array. So we didn't have to figure out a naming convention because we didn't need a new name. I see.

M

The positive thing is that, as I said, that there's gonna be another feature anyway, maybe over the new. So the new method wouldn't only be about returning values. Aspect would also be about other things like and maybe in a better way. That's another problem that we're trying to fix, and maybe we would use that as a way to solve our naming problem yeah and we want.

N

To take a media producer of the same day, you saw what it asked I mean if we are.

O

Going to be done by the past for his left words, this should be be useful to go, find ways to make it more. Efficient I will be an eternal principle, namely.

A

We new video-test is already available struct right, so you don't have to okay.

B

Means in done in court to one we added the ability to back a value task by an interface rather than just a te or a task of T, which allows you to provide your own interface based implementation that can be reused for multiple asynchronously completing operations.

N

Another thing that we're after now, we are talking about the whole the whole set of balance that we carry in this idea. We are going to return a value sold by the. You should probably consider an API that kind of conveys.

N

Change your fitness.

N

B

But there there are plenty of places where we basically said with her written in stone. That value task has this property in general and there's only a few things that you should do with.

F

B

And so you know, like certain streamed, read: async methods, return, reusable value tasks, and if you were to do anything on that stream again, you would invalidate it so I wouldn't be too concerned about having this surface that in the API. We're basically saying that one of the ways that value test differs from tasks is that there are only certain operations you can do on it. You can directly await it or you know directly.

B

You check if it's already completed an asset access, its value, but any sort of caching, reusing after other operations and so on. Those are all invalid uses. It.

P

Wouldn't be necessarily a symbolic that is valid, something that we need to.

N

Know that you are whenever I roll all right.

B

Sorry, I don't understand the question so.

N

For instance, in is equal numeral, the the body task of food that your attorney you can't await. It is the same body task that you can awaken to the time right, correct, okay, so so, in this case you would be the same. It would be that we don't want to embody that about you. Sorry I.

B

Mean rephrase if, if you try to you, will get back a value task of t. If you then call move next a sink and try to await the original value test, you get back and get an exception.

N

But I wanted to be more like.

N

B

N

B

Step, if you, if you try and await that value task after you've, so if that represents an actually asynchronously acknowledged and next to get it causes something else, turn a sink. You trying to wait that original value task, it's going to give you throw an exception. It's gonna say this invalid operation, because the value task itself stores a token in it that indicates kind of what version it is and that has to match with the thing that provided the token and every time you call move next, that version gets incremented I, see. Okay,.

N

B

Still, it's still the same object over and over and over, but we try and be because there are cases where, if you have, if you were using that value task- and you were trying to you know if the from a consumers perspective, it's not reusable, it's one shot and that's always the case, and we want that to always be assumed, and so we do our best within the limits of what we can performance wise to say. These are the only ways that you can use this thing.

B

Anything else is invalid and will try and throw for as many of those as we possibly can.

Q

Okay, I need to go back to to understand exactly how it worked. I I thought that the whole point was the greatness of the. What exactly is correct right, so yeah.

B

So there's two things here: there's the value task of T and there's the I value task source of T. The I value test source of T is reusable and create as many different things as you want from it. You know one of the time the value task of T. The struct is not reasonable. It's just a wrapper for that. I value, test, source of T and effectively an integer.

M

And maybe one comment: that's not that doesn't necessarily have any impact over this, but just as a reminder. In many many cases the user is supposed to know that there's no value and using the async overloads because there's not going to be any I/o. So just as a reminder, the way this API works is when you execute the query, you're going to tell it whether you want sequential access or not, which basically means is the row buffer. Then memory or not, there's there's an API for that in IDEO gannett.

M

So if you don't specify this sequential access, basically you're telling me do dotnet that you want the road to be buffered. You want to be able to read columns in any order, which means basically, that the entire row is in memory and no column. Nexus is ever going to perform. Iowa just a reminder how this, how this API works, so good feels very acing. So I was another person.

A

That you, when you called move next, you don't have the rule. Yet you might have to sort of fetch me and saying that's only the case when you said I wants sequential access for the world.

M

But otherwise the entire rule will be fetch. That is a memory, and then you have random access. It may be the contractors that you can yeah if you can do random access when accessing fields. That basically means that you're forcing the whole road to be buffered or maybe it means that every column access actually goes to the server, but I've never seen a database driver. Do that because it's very inefficient in.

K

Theory anything.

M

Is possible with thank you so I'm just saying that at the async overloads that the guideline is usually too used, and only if you're using the sequential access mode coverage may be a little bit more rare. But it's talking the.

A

Most of the bulk of the work here, when you would do async would be just to move next to a singers, but we just wait for the next road to be fetched and then afterwards it's yep random access.

A

The other option that I see is that you could imagine that, instead of having get filled video a single easty banal icing, you just have one worker method that does both of them and that returns very attached. So that way you minded to likely to try get video a sing and returns false. If it's nice.

M

A

Be in a new API method,.

M

That we're trying to think about which absorbs the value tasks off the ability saw in and so on. These are the two big problems that we'd like the song and it could be like just a new get method. Maybe it's called get and it, and maybe it interacts with none of the Lydian somewhere. We haven't designed it I.

A

Mean like I, think that the general gist is like in order for to use value tasks and not have weird mismatches between the current virtual method and this one. You quality wanna introduce a new virtual method that you will expect provide us to override to add to them, so they actually get to a synchronicity and not just a wrapper around the task of tea, in which case you need a new worker method anyway, and then at that point you might be.

A

Actually you know better off to just combine these two things also seems weird to say. First, you do an async operation to determine whether it's Nylund. If it's not now you call it another one. That is a so. It seems like weird right. You would probably want to fetch them over at the same time. This is just it basically is another bool like you need to transport some.

F

They're not actually.

R

This is super old, yes, super super old stuff, in that they really was P.

F

For the ones where you're, just forwarding cancellation so can I think you've been preferring to just rameters. Lately, that's fine.

M

Then we can just equals the only the only possible argument against it, which is not real argument, is that in other areas of a do net you already have this pattern, and so it would be a little bit inconsistent, but I don't think we right here. We cannot believe for without it.

F

Yeah yeah, we can't, we can't believe the old one yeah, but for the new cases we can avoid duplication. Languages can have the same effectiveness if you're by yourself.

A

And so it seemed like, even when we just said I would I would be leaning towards approving the API, as is except for the Antigua I would say for the async ones think about what you want to do with them and that you want to have them at all or well, then you want to do a different pattern. May you somehow figure out how you would do? What would you unify the current once they return task of tea with a new task and then come back with that proposal?

A

Isolation, the other ones, seem all very straightforward right. They all want to one reference.

N

N

They're, not the.

D

Only one that's all here is they get provider specific wages of an.

M

Object array: what what is that Oh, an idio demand provider, has the option of having its own dialect. It's for the provider, specific type, for example, and I, don't know in Postgres, you have a daytime value, that's not in the same precision and range as dotnet daytime. So by default it's going to return daytime, but you can ask it for a provider specific type, which would be some other type like a post with specific CLR type that represents it, and these.

M

This is an overload specifics of the specific values returns it as an array of values. Again this is an older API. You have the option of you know interacting like this. With a do down there, I don't think we'd add today.

Q

Takes I know they've already it's it's again it this one.

M

Is equal to this thing, yeah.

Q

D

Already exists.

M

That's a mistake: okay, they make sure they move it right. There there's! No there's! No! There's no like ordinal around there anyway respect you.

A

Forget about a specific value which takes it in 32 and that's the one that you want to do overload for.

A

Yeah any concerns anybody on the call.

D

M

Easy you would prefer to leave out the async methods, at least for now, yeah.

A

I would just decouple them because it's easier to for us if you can just approve all the rest and then it's just you do to just think about what you want to prevent your task and these acing operations, because I think I agree with Diego like in general, like you want to have a story that is holistic and not just solving it for the case where you pass, and it's praying that's good, but you want to want to have a story that, when you're passing and in you still get these new performance right and especially because when you do a normai present seems like you want to go through the in base ones anyways, in which case you really want the story.

A

N

And these methods we are already using for the world valued I cannot.

O

N

I think we need a new class. Aikido is more.

M

A

To be fair, though, like I, don't think this class is substantially harder than stream from when hierarchy stand. One like you have the same probability yards. You have different models over virtuals. We have to think about what they mean, so the hierarchy in to be clear, it's not pretty but like introducing a new stream table, was considered worse than where we can do to the type. So I think this might be in a similar camber. It might be worthwhile.

N

For providers sure I.

Q

A

It's not just well how many lakes we use the type, because everybody knows the type I'm not accepts it. That gets to the range right.

M

That's what I mean you couldn't leave the gives us in five I'm, not in Richard, not unanswered in any way, and just tell people if you want you want like the latest inverse API, then you just switched to this new one. Instead, that.

A

Is generally true, but in practice that often means you need an adapter in between, in which case your sources are the same. That's fine, yeah, in which case you are like they expect to square one of like how do I make it the old day. Maybe we don't talk to the new one where they have a Miss mentioned and the way this stuff is combined so yeah, which is why I'm saying like, if you really want to have it, fly to the ecosystem?

A

What constrains is the number of people using the type and, if you end up having to write an adapter I, don't see this is substantially easier to solve than involving this I mean it.

M

Might be or might be not but I, don't think, there's a substantial. The one advantage is that at the end of the process, you end up with a type that sir, that's a clean API like the one. That's the newer one, maybe.

A

You, oh you mean everybody, has two tabs forever.

A

Like I mean I always say that this is one of those areas where you know HP. That was quite successful of like cutting the ties to the past, but for anything that is more in between between the VCR between an F model like you, I, don't know where the line is, but there's a line somewhere and like as soon as you end up in this wall, you have to constantly have to carry your own bow forever, like I'm, not sure, that's a better model like no. You have just more complexity.

M

We have so this is like this is one small issue, but there's actually a plan to work on several others on a do not net. So on encore, FX, repo, there's there's a handful of issues there. Some of them are actually quite important performance, wise and.

F

M

Those unlike this issue and the question, was mainly what kind of I don't know how fast should we be? What would be the deadline for getting any sort of thing.

M

A

One would be alieve faster.

A

How many are we talking about there's we main issues.

M

A

A

Even though the room has left not show me what's on the call, it's Dniester with me, I'm.

B

Still here, oh.

A

Great, so really some people.

B

Stay partners on the line: brennan, connor, ryan, eric's and john and all their.

D

Alright, so that mean then just copy paste and what we just said, something you need to make some progress.

D

We say I believe these guys off. This was a mistake which he says was just this time.

D

D

In the APA.

A

Let's mark this for senators well, I suppose, because you probably want to get these changes into, send it as well. I guess.

D

A

Give me some keywords: what would be in the API be.

Q

Something I have the numbers.

M

For you wanted: okay, oh yeah, even better, what's the numbers, so one is three: six eight eight yield another one is eight nine five, five.

D

The reason they don't work.

M

He said three, maybe start with eight nine five five, it's a bit easier to eight one: eight, nine, five, five without oh yeah, oh yeah, yeah, they're, all very old.

M

This is one thing we'd like to solve. Currently, when you want to write anything to the database, you use something called a DB parameter and it's a non-generic class. It's never been made. There's never been a generic story here, you're any hint that you want to write to a database. It's gonna have to be boxed.

M

You don't like that.

M

So it's hard to review this in you, don't, after the other before me, somewhere down not yet. This is not in this in this yeah. So there's there's a discussion about design below like that yeah right there. But how do we do it? Do we go for your base class interface? There's some discussion here, but it's not mature, not ready for review.

M

This is an example of the kind of thing we'd like to put to get into 2.1 as possible.

A

Yeah, if you can get a design, that's really would not by end of week end of week and then it could be viewed next week. Otherwise it will get a harder like I mean, depending on how much churn there is. You might be able to improve this later, but I, don't know what the the lockdown will be for the standard.

A

The problem is, we have two core had this change across various parties, whether like the the problem on our end is we have to like in general, we try to avoid like adding new code and getting it under standard at the same time, because it's hard, because you have to basically get three four parties to all: go either this implementation. Yet you want you, don't have confident that the implementation will work before you actually implemented it all right. So you die newly.

A

You have an implementation in decent McCoy, you're, happy with the implementation, and then it's just a matter of getting in tune to the standard which is kind of what we did revising a new move right. We first wrote the code in corvex. We kind of convinces it to shape you wander, then you just dip it against the standard, and then you add the api's, which is generally what you want and I think that in order to do that, I think you would have to be able to do this in the next four or five weeks.

A

K

A

I mean like I, don't know what the preview timelines are but like it gets close to your suffering, lockdown, so faster, it's better, but I mean I would generally say like also if you miss the book four to one nothing bad happens, I mean like the you're missing full framework anyway, at this point, because they're no longer implementing that standard, and so are everybody else like orzammar and they move recently be fast, so I would expect.

A

If we get another api's together for two two, then there will be a at some point, all right, so you're not locked out from that. It just means you catch you later, but exactly.

M

This is one you're interested three six, eight, eight even older. Oh.

K

Yeah, why are you earlier? It's been there for a while.

M

This is about this is about providing a batching, an actual batching API, so a way to execute one command. That has multiple statements right now. The way for doing that is to simply pack them in with semi-colons. The problem was that with that is that at the protocol level, in many databases, including even sequel, server, that's a that's an not a good idea that doesn't work that way, because you have to send separate protocol messages, which, of course is a driver to parse the s.

M

We ought to find that semicolon to split on it and send two different things. This is one of two issues designed to stop parsing SQL at the client side. This is like the ultimate goal here: I, don't think that an sq up like an idea, dotnet driver, is supposed to be parsing, SQL and rewriting it, but that's going to be necessary because of this, and another issue, plus even I, know that the SQ of this in SQL Server does work, but it's quite inefficient. So if.

M

Not mind your way to the matching list, thanks.

A

Though the proposal would be that you change the object policy you can effectively submit, they give like a batch class that you can have some ants to and then you just said: execute batteries underneath.

M

A

M

Command set there's something already that exists like that internally in SQL points on when you adopt this idea. Somehow.

N

You say: wait an internal API that that button there is no perfect ice.

D

A

That seems a bit I mean yeah.

A

I mean I think getting this into three Oh is probably much more reason when they're getting in it and it, but we can try I mean if you, if, once you have an implementation, we can. We can see where we are, but I would not rush the implementation to catch. This and I would say: yeah don't slow down artificially, but I would say you know, shoot for three Oh get it get it into the shape with video, and then once we know about the API is.

A

Is it's pretty straightforward for us to just add it to this 92.

O

A

Then, if that doesn't work anymore, then it's to I mean given mmm your conversion code very aggressively moves and I'm, not sure all easy. It is for them to pick it up ad about net stuff, but I would think, given that for that code Brooks, it's probably one of the easier ones for them to pick up compared to let's say, methods on string, so I think we couldn't probably make that happen, but yeah don't take too much.

A

Boom anything else and you break your face, particular.

O

A

Are you plans for both racing disposable and IA, see how we use them in the airport? Slash any other any plans. It's also.

N

In the area that later, we find a discussion of something that this is or whether, because something a version of three days and running.

N

At the core level, we are definitely.

A

You basically give you me icing removal of customer, let's say in the neck, catch them.

N

That is the way it works in ENIAC is basically you decide when you are consuming ready, whatever you are going to consume them? Oh, it's very sensitive right for each other. I'm right, wait for the turn.

A

Both work, the your. What is that your entity types icing, implement both an immobilizing, our prototypes.

A

We do have cancellation the.

N

The same dirty way you do the yeah.

A

Forgot recently, we had those conversations even I, think suggested. Change somewhere is one of the collection times where we base you said well. Instead of making the collection type implement I using a new mobile, we decided to give you a method and I think that was because the would have implications on immutability of the result.

B

Even the rumor that the one we were just discussing was channels, yes, yeah.

A

B

The discussion was around: do you expect an innumerable to mutate, the source you know like with a sequential enumerator you sort of like? If you enumerate over a queue, you don't expect to drain the queue and, if you're a numerating over a channel, it's draining it as your numerating. So the question was around: do we you know created? We just allow you to say a wait for each or a sink for each or do we force you to call a method to say like we doll or something we ended up on the ladder.

B

Interestingly, after the fact, someone commented on the original issue highlighting an existing enumerator that actually mutates.

H

Q

M

Q

M

Q

Record so it's awful.

M

But in theory, if we really wanted we could we could do something. An idea without net worth I think I know more, but that would make sense to but that's fun, discussion yeah. So.

N

I have like two different explanations for why later here doesn't is in whatever sorry I in wearable and I. Now why shouldn't I one is what it was possibly invented before I enrolled, but the other one is basically the same thing: Sunday Internet guys don't want to implement it anywhere, which is I, don't want to allocate the current for every time, and that's that is actually the case with the interview right now is like as an allocation as an apology processes like we know right or some data they offer.

A

You, the reader, is effectively the indium right or and and innumerable combined in one thing, but which is actually fairly efficient way to do most of the stuff. It's just I mean. The question also is for a deal done that, if you think about it right, like there I don't know for eaching, doesn't buy you that much because you're, only the bulk of your code is not the move. Next, the bulk of your code is getting the Raiders or other the same, because the thing you are getting is that the thing you want right.

A

The thing you want is a person or something else here you still have to exit individual fields and get the data out right, in which case I think in a do, though, any F. It makes a lot more sense because you already get the object out, so you might as well just collapse that into yeah.

N

Yeah so I tend to think about it as a lower level begin block. That doesn't mean I.

M

Think it's worth. For example, we could imagine some sort of a Canadian up net, that materializes value tuples and and then it makes a lot of sense for them to be in the middle. Then it could be a very nice.

N

H

Like tuples are great just casing, the can't agree on what the.

A

Case no I mean that tuples have other problems is like the the the biggest problem of tuples. Is you have to decide how many fields you have in you better? Never change your mind. Oh no, the one thing we said in the BCA. What we probably will do moving forward is give you more deconstruct methods where it's easier to just take the results and I can sign it to the locals, because you can overload those guys by parity, so you could say yep and you won.

A

You only had two fields and then we won me at three, but because the way you write, the deconstruction is in C sharp. The compiler knows want to pick up, so we can always add to that, and that seems gently fine. But if tapas you, you really have to be sure that you never need more than whatever you decided that and you better never need any other members. I can method or something else I mean you could in principle, do an extension method over a toughie, but then I'll specific.

H

I gonna notice over all of these all of the time yeah, which is you.

A

Know it's apples are tablets are really great for privates, but if we just have working methods, you don't feel like doing rating times for them, but probably API, is you probably better off defining them.

N

Also, like something I mean super Lambie was, and it's like a that's a ring, but I have our the negotiator would apply to couples to body tuples or you could apply to a work that the internet guys have to do them. In many cases, do you need the teeth that is like this, our PST, but it adds this property and I wish. There was like a way to express it and and constraint that says this T based on this RT, but that's the property Spurs in the rest of your code to twenty four o'clock property.

H

Not really, it sounds like the, but you want extension everything that the belt-fed no.

N

No I mean super c'mon that that you are composing over some more than her, like something padding like something like an incorporated. It's going to take this world thirsty and it's going to add a property to hit that contain some data that you massage like calculated property over something that was already there from now on. There were severe cold to us on this article that is like T one, but without having to go without containing.

Q

N

How do you believe is from you type API or something like a member that, because we.

A

Basically want to be added, you basically want to have the type of presentation of kind of like a fake derive tired, but they basically have T to that is effectively one remove one more property or something yeah. Yes, that.

G

Shapes is that, like the shapes proposal, if.

A

It sounds a little bit like it, although I'm not really sure what the mechanics of the shapes one was. It was a bit more generic, no pun intended, but I think it was a bit more. You I think you could. You could define more complicated operations of that.

N

Like a long time ago, the shape sometimes I, don't know yeah.

A

Because they probably talked about higher order on tours and Combinator s, they usually don't last very long.

D

N

Because all right, it seemed like I read that I need that, for.

N

A

Right so seems like we have exhausted all questions. You guys got everything out of it that you wanted right, yeah and you're. Only in this week, you're going.

M

Back next one back to Berlin: that's it next week, there's something else. I wish.

H

A total of beers to speak.

A

All right, don't let me first finish the online streaming and I'll find a button again. Alright, bye, bye, guys.