.Net Foundation Design Reviews, 8 Jan 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

00:00:00 - Approved: Native Library Loader API https://github.com/dotnet/corefx/issues/32015
00:22:27 - Approved: IAsyncEnumerable(T).WithCancellation extension method https://github.com/dotnet/corefx/issues/33909
00:41:18 - Approved: PlaceholderText API addition to TextBox control https://github.com/dotnet/winforms/issues/134#issuecomment-452412351
00:45:02 - Approved: string.Concat(ReadOnlySpan(char), ...) https://github.com/dotnet/corefx/issues/34330

A

What to do alright, so any preference is in order. Otherwise, I would just start with the one that was connecting the yesterday, which is.

A

Any preference is in order, otherwise I would just start with the one that.

B

Yeah and there's one with the fur teasing streams, that's not on the screen, but it's further down, but I would really like to get into preview to possible.

C

D

The dog actually wrote it like just the one I expected the one now.

A

So they say this was the one you reviewed last time. This API should be pending reader and then the only open.

D

I provided some context below as to why it should be another day.

D

The whole point of this API is to give Rick's ability to be easy right, so you should give them the stability of doing something when it actually wasn't imaginative. If.

E

They're actually different, then then no is different than.

D

E

Because I think that was really the question from last time. Is it normally.

D

The same as I guess.

E

D

The other tainted it was like it was running bull and you commented that produced an exception instead of returning Falls. There was something all these days.

C

Is there going to be a way to check if there's a handler already set I mean.

D

You didn't Plato said it and it ain't.

C

Non expensive way of trying to check if it's already set I mean.

D

Does not get it be a mess, because what are you gonna do no.

C

Brain stir like if there's one, if there's one not already registered, then you might yell do something different. Like not crash your App Engine.

E

Is thick: let's just try to catch me if you're, if you want, if you need the resolver to be the one, you want, that's set it and if you don't get that one throws I, don't see a need for exposing. You know: here's a notice that, because most behaviors would be if it's set throw and that's just called that's just on the set method.

F

It's a set rule, not cool. Today right they were just returned false right, yeah.

D

There's one like in the in the beginning, the top of this page I wrote depression of this.

D

Below the flow anything.

C

Yeah- and so that was my question: if it's just going to be void and always throws, then users have no way of determining if something already set yeah.

E

But the expectation on this I believe is that it would be called from Maine. Did you? Did you paste your line twice? Okay, then we'll throw otherwise you're, not Maine, so you shouldn't be calling this Bentley, so I, don't sure like until we get user feedback of what a scenario.

C

Looks like I wouldn't I mean you can also imagine the scenario where someone's got a plug-in library and they want to register the ability to search for plugins for themselves before.

G

The rule assembly further.

C

On this Emily, you.

D

Do that something I mean right.

C

But but let's say that you're, a a library, that's providing.

D

C

Extension and plugin capabilities so like paint on that or something and you want to be able to as a plugin into paint on that be able to search for your own chains that you can load and then perform operations on which is separate from the plugins that paint.net one's to load and so paint.net itself. Critics could expose a mechanism to chain into, but you might want to check and see if someone is already registered. So you know what you want to do.

D

At the end of the day, that's actually what they did today like what the deal is: a plug-ins in the commitments that they can have they don't. What about the tech.

E

You just registered always against yourself I.

D

Mean I I see the point that you're making right like they say you have Newton softer J's and there are multiple entry points and first entry point could be anyone a lot of things, and so you want to have way of setting it from any. One of those points are not visited. That is true like because.

C

It seems cheap to be able to expose a property. Is.

D

They were telling you to do one is. We can externally expose that or we could just say that it's the same dll having multiple when that ele leu in dementia zone check it.

A

Also means this one I can no longer to say it.

D

Is not registered for now.

D

There's no all right or unregistered. We just want to see how it plays out, because with all right, I'm thing like that, like it really need to condition right, you could lead to more bands, so.

C

Is this will return this one, throw.

D

C

The assembly is already registered or any assembly is.

D

Registered for disassembly, okay, only.

E

The answer is: register disassembly that you all know the behavior okay. That makes.

C

A

I think that one tennis example like that's the thing they will fall on the belly eventually right, like I, think it works fine if you're the app right as soon as you have plugins the polymers. What is your assembly like unification, have to at some point and like yeah, three plugins happens to have the same as I'm the internet right, but then at that point your school regardless because well, what? What are the three photos? Do you want to use to those native dependencies right there? It's not that you can like at this point.

A

You have to decide who the who the one is and probably they're only sent to the model as well. Whoever registers the assembly, whoever those see assembly first, that location should probably need for the location, but that might not be something that this API shape will allow you to do, because you kind of want to you basically want to register almost like, if you know them to send over that identity. Whatever that thing returns, that thing should register the callback right by like there's no way to do that.

A

So I think it's probably going to be fun. What's fine in the camera.

C

All right so Ben is there any reason why we don't expose and get yellow import resolver that way someone can change.

D

And that is left up to the initial process. To do for itself, like our model, is just that in a lot of ways like new info. If we just we don't support, changing, explicitly says that process has to do it's not changing, you could add again but easily, but the fact that we don't allow for writing well.

C

Cuz like, if you don't allow overriding, but you do a lot of training, then it works the same way. A lot of our other resolvers have worked, which is that the first, the first delegate in the chain it will either return something or not, and if it doesn't, then the next thing in the chain gets invoked until someone returns. True.

D

Yeah one of the early versions of this API was written such that it actually returns the progresses over. If one is already said. Otherwise, it returns the and said said so. If we think chaining I don't know, there is no known use case for chaining, so we thought of just doing a set. I didn't know what, but it.

F

Is a fine thing to.

D

F

Will we support chaining in the future yeah yeah.

C

You know if you don't sit right in the first, then. What's the signature, another good, it's.

F

E

In presumably like, if the scenario.

G

D

Something I think that.

E

Matt wants to support, complicated, dynamic, binding. Then they would just write their yellow and fork is over themselves to be a they expose, their hooks for complicated.

D

E

You you, as a down level library, shouldn't, be changing the way they look stuff yeah, like that's really my easy belief on this is ain't done that owns paint.net building if they want it to be dynamic and extensible, but that's a feature that they should write, not a feature.

E

Yeah but who's pulling his AP I think that that is calling it think about that yeah. Your application is calling it for your application and you're thinking from the main method. Yeah, your assembly is doing it for your assembly visitor late assembly. You did it in a static finish like.

C

So basically, libraries would never we never do it for, for.

D

The main assembly and Liberty can do it for this. You.

E

Would only essentially, the argument you should pass should always be get executing something if you're doing anything else, you're probably doing something wrong. Yeah.

C

Whoa, but this is for resolving DLL imports for the assembly that contains them, and libraries are probably more frequently the thing that's going to contain: yellow imports in the main app itself so.

E

If the library, so the library somewhere in a static initializers, would call this as, like. Oh you've, woken up a plug-in library. I need to know what I do so in a static initializer that fired in there somewhere before any PM votes ever would have got to touch. It says now, here's the way you resolve things for me. So again it passed this not get executing assembly like who is the line of code right now. That is college, but.

A

You're making the argument that seemed alive with your quotes, if you wanted yeah.

C

The libraries, the libraries probably more often than not going to be in thing calling this mess right, and that means that there's a lot of boilerplate that the main app has to pass down through to the calling library to be able to say here's where I land. Here's where you need to look for for default stuff, in addition to wherever you're going to want to look yourself.

D

But look for it particularly assembly, it's already written in the same DNA like we like. Why would you want to chain it like? It could be so like if.

C

You have an app that exists in the root folder and then you've got plugins in a subfolder you're going to want to tell the root app is going to want to tell the plug-in you want to look in these folders for any additional dependencies. In addition to these other places that you might be looking into by default, nobody.

A

You but the point is it's not arbitrary dependencies, but only native dependencies right.

H

A

What would the assembly itself not know all the locations because it should probably be whatever your stuff is in, because.

C

There might be things that are shipped as shared central resources that aren't necessarily shipped side by side with the other thing. But I don't think that there's.

E

Any sensible way of saying I as a library have these dependencies and you as an a post, have these dependencies and maybe that merge them together like that. Well,.

C

That's what chaining is: if you have training, then the then the route map registers and then the the other one registers as well, and then both can do their searches independently and whichever one returns true, depending on the desired ordering yeah and then when they don't they don't both return true week. So that's order, you know no one returns true and then the second one never gets a home right and.

D

C

A policy in this feature of who got that which one gets called first, which that's up to each app and the library to determine so.

A

It's very much the core model of we don't support any of that in a scheme. You know you can't do this, you would they kill the way it would work.

A

Is that if you, if you are applying for painter and then then paint it might gives you an API that says, give me you know, give locations when they do dependencies right and then there was all the you you register for your plugin just has to do through the for each there, because I think the problem is is if Jeremy said I'm, not sure there is a single policy level always work like. Maybe the plug-in wants to use a different implementation for me, what the whole sermon he has right, like.

A

Maybe you bring a new original, whatever dated module you have, or you might be, because you're running on, like three different versions of paint.net and you're, saying well, certain versions: I want to prefer mine one over the one that they have like I. Don't know like that. That's one of those things where, like there's, not a good answer and.

F

A

Only thing, I would say is like I'm concerned. If you make statements like the app has to decide right, because, because I'm in practice, like it maybe I like this I agree with tanner like the most likely color will be. The library like, like the app will not call these methods ever right. So that means you need to live in a world where you have a bunch of it assemblies that are loaded into your process.

A

Each of them happen to call this API, because that does honestly the only sensible model like if you write an manage wrapper over over native module. You don't want the consumer having to do any work. It gets the whole point like.

E

Just thinking of how, if I, would ever use this sorry its if I had a bunch of my own helper libraries and they're, not nuking packages, they're, just libraries that I use for me right and I've decided in my a post thing: I'm moving all of the native assets into a subdirectory I in main, find all of the assemblies that I know that I have is part of my thing and say: don't look in that directory yeah.

A

But good somehow anybody works like this, but that's.

E

The ethanol, no, but one is that it's not.

H

A

Works like if you, if you write a console, app or web right and let's say you had a nuclear package dependency on sequel, I'd, right, yeah sequel, like gives you an x86 folder, it gives you an X. 64 folder has some content items in your project and when you build your app, your app happens to have two sub directories for the x64 and x86 byte arrays right and then all you do.

A

You just call manage api's against the sequel, client, API and then somewhere deep in the balls of this thing, this one called this API to make sure that white ones are being loaded right. Whatever.

E

The app deployment model is put that file somewhere and it has to say where I put it the yellow, doesn't know yeah.

A

That's what video but given my point is that almost anybody who do who provides native assets today does not require the app water to do any work. That's.

D

Typically, how you do like, if the plug-in has its own native libraries may not doesn't even have to know about it like it. Only if the boat somehow happened to deal with the same native library, then videos listening and that case is not typical.

C

I think well, I mean anytime you're dependent on something like DirectX. Then you immediately have a shared model where you can either be using the one and the system root, you'd be using one that was shipped side-by-side with the app or you could use that shape side-by-side with the plugin I mean.

D

If, for native it's already global I mean our story, nobody goes right right, most likely to a polluted and it's already loaded. We don't actually get into any of this like if you like libraries, already loaded, you can only load one version of the native library, then, if the app and the plug-in don't agree.

C

On the same, that's not.

D

C

I know, but what.

D

I'm saying is, but what I'm saying is? If they don't agree, then we have bigger problems then like where we're.

C

Looking for well, what about the case for assembly unloading and made it beyond motivating.

D

And I don't know what so do you go to bed you do that I'm doing is I believe without support. Yeah.

C

So we can't this is that we can't unregister and resolver. How is that going to work when assembly unloading happens and people are able to unload assemblies? Then they load it again, and then they want to set a new resolver, because it's a possibly different, it's the same assembly, but they need to do initialization, logic, etc. I.

D

Mean if it's a board unloading I think we can visit I'm sitting.

E

Today, line unload that clears the resolver, because you unloaded the assembly and it was always bound to the assembly.

D

That it just where it's per that I have you object something out the back case, I'm good. If you unload the native library I, don't even know what happens. We don't support it now, but like.

E

Barring you should be careful saying that we need something more complicated, I think if this is one it is the most simple approach and we can explain it later well,.

C

Iii think that, like just having an additional either, is it already said, or even a public static, DLL import was over to get DLL import resolver would satisfy extensibility the ability to determine if it's already said they were allowed chaining. If users want or need it doesn't watch it. Yes, it does. If you expose get DLL importers over, then you can train because you can change elegance. It.

G

Would that would be kind of you.

A

Don't know you.

F

A

You you would at least require a second, more thorough.

C

A

How would you do your II think, there's only one register and how do you, how you add, 1 to that combined.

E

Returns a new delegate, yeah.

G

And then you would and then you would write the existing delegate. You wouldn't have so what you would do is you would get the existing delegate and you would insert your new hook wherever in that chain and you need to go and then you would replace the entire chain right, but this doesn't allow overwrite exactly.

A

E

Saying you would against neither.

A

E

That you control yeah again.

A

Out to be the pest.

E

Is useless because there's until we have a pattern where we would understand what someone's going to do with that other than Pro, we shouldn't data I! Think.

A

It's something visually I said that gave you one chaining I think the most logical API can do the deal and.

G

A

G

Only what you add and remove your own thing: they don't let you reorder and they don't let you inspect. What else is there and I? Don't know if that's an important business trend so.

E

G

E

Because it does it's very clear in what it does and if we need a different model later, we can write a different modeling. We.

G

Could we could always make another another delegate, that's just more powerful than a husband.

H

G

Then just have an overload of the stuff that it should be a very full change.

A

Yeah I think giving more complicated at spaces. There's a few extensions going to give people the better off yeah.

E

And then, if we get concrete feedback of what isn't working for someone, then we can. Then we can react to that. But while we're being speculative I think.

G

We can say yeah, but we we agree, though it is. We think we can extend this in the future without too much changed our life, probably without too much pain on top of well.

A

I mean like the most simplistic thing: if you own, this is a poor chaining is you would just make set.

F

Nor throw one second set: you would just expose together, okay, like man, that would be so. We have. We have a plan for the future, but if need be.

A

It's better than real time, good.

A

All right so then, I think we're good with this one right.

I

A

Thanks for coming by.

D

A

Right Stephen, you said something about please scroll.

B

Down an upstream date to a sink the last one.

B

Okay, so yeah, so we added this cancellation, token parameter to optional parameter to get a sink a numerator, and the idea is that if you want to write a wait for each, you do a wait for each of our item in source and if you want to add a cancellation token, there's no language support for saying. Take this token and pass it into the get async enumerator method the compiler generates. So we need an API that'll, basically sort of funnel it through, and so we need some sort of talk with cancellation.

B

So you can say a wait for each of our item and sore spot with cancellation. Token, that's sort of the general design that we agreed on with the c-sharp language design team as well be scrolled down a little bit.

B

It sounds like.

G

This is almost like configure await in spirit well,.

B

That's what the main bulk of this issue is is we now have we now have with cancellation and configure wait that are inaudible off of the innumerable, and so, if you scroll down, we we need to basically be able to combine them, and so previously we keep scrolling down email to the red green.

B

So previously, we'd approved adding configure weight via basically an extension method online, a signal source that returns a configured async enumerable like this, you know, custom struct, but now we need one that also can accommodate cancellation, and so you need to be able to do like duck, configure await down with cancellation, or you know, Don, with cancellation, configure, wait, etc, which really means you kind of need them. You kind of want them to both be operating on the same struct.

B

So what I'm proposing is basically getting rid of the current one that we added, so they scroll down a little bit further and having the single struck. That's returned both with cancellation and configure, wait both return it and then that thing has instance, methods as well. That return. You know another of itself, that's been configured appropriately. The upside of this is allows you to say dot with cancellation dot, configure a way that, with cancellation, dot, configure await or dot configure await dot with cancellation. So you kind of get you know all possible combinations.

B

Someone might want to write we're not locking you in the downside is it allows you to do something nonsensical, like docking with cancellation dot with cancellation down with cancellation, which is just gonna, keep replacing the previous one that was included and we could choose to throw if we wanted in that situation, but I don't have a better solution in mind other than having multiple types to encode sort of the notion of which API is you've.

G

Already called we have other places in the framework where we use fluent syntax, like this main counter string, string, builder, I, guess like configure way full of like I, wouldn't really say, link is fluent syntax to because it's actually it's returning I guess this is returning a different object. Every time is left organic.

F

Is having a BS 300, effectively I.

E

Take about a little bit, we generally don't do it, but in this case it's one extension like configure weights and existing method. Dallas is configured with the extension method, tacking on the same paradigm that we have in the same object space. So it seems to fit so.

A

It's even like the the live cancellation, one that one configure wait, one that taken I see a new mobile. They also handle the case when it was a configured cancel blazing anymore. There was, there was cost to the interface right, no.

B

That would be coffee the instance method on the struct.

A

This is intentional that this one cannot be converted into phase at all. Yes,.

B

If you converted it to an interface, it wouldn't do what you wanted it to do like configure await works because it returns something that strongly typed. That has a move next async that strongly typed took a return. It configured value tasks available.

B

If you ended up casting this via the interface you, your move, next async would just return a value task, a bool and then configure await, wouldn't actually take any and have any effect because configure wait, works via pattern, matching or yellow strongly type pattern from the compiler, picking up the right configured, value tasks available and therefore picking up the right. A wait pattern on that. Oh.

A

Because? Because the cancellation is part of the of the for each.

E

Which gets lost.

E

B

It's so imma get II think in America, it's optional, and so in the case of this struct we're not specifying it. The compiler will match, as if no cancellation token was being passed in so it'll match, either again async a numerator, that's parameterless or again async a numerator that takes an optional token and in this case, because the only for using this whole API is with is that you've passed the cancellation token into with cancellation.

B

We've defined, get a Sinkin over here to be parameter list, because you should never be calling this get async, no matter where the token.

A

It's effectively deliver one denies thinking. You know that it's canceled all right, because you want this to be it. Whoever is actually relating it doing this.

B

Is purely about the call site further, you wait for a.

B

Highlight, or we discussed further is I wanted to just briefly read this I don't have a strong opinion, but we previously put this on to task extensions, which is an existing class, and maybe we worth revisiting whether we actually want to do that or create a new class.

B

The reasons why we want to create something new are potentially minor, the main one being if we did want to enable someone us or someone else to release something you know how to band these extension methods would end up being on a different type so which, after using extensions and acts, doesn't matter at all, but if you were you're using the full syntax, just the static method, syntax you'd have a change in your code.

B

I think someone else raised the question of whether it's also kind of strange to be on task extensions for with cancellation, because you're no longer like configure when you kind of wave your hands and say alright. Well, it's about configuring. The await on the task, but with cancellation really isn't about the task at all.

B

It's just sort of about the general for eaching.

B

Assistant collections, generic.

J

And it's quite possible.

E

Which makes them a seka dimmable extensions.

B

Hey I'm something else to discuss. If you're gonna do a change, we should change it. At the same time,.

A

B

H

A

Honesty, I'm fine with that being in the other namespace, because they look at system collections, generic ease and, as just do of collections. This one really has to do with you, know, cancellation and how you consume a synchronous of it, which I'm.

E

Just going with like, if somebody's writing there for each like yes system, threading tasks, I think is in the default template from BS but like if you, if you are in a context where you don't have this use it like? Is it reasonable to be doing a cinema bow and not have a system threading tasks using because if so, then you get the funk I'm writing this line of code. It says in the sample, but even you can't find it though I do think. Yes is better at like sniffing for missing using snail.

A

Well, it definitely gives you mean if you type it correctly, they give you the right suggestion. The problem is still that you don't get like if it yeah. If you don't know the name exactly then, but the question is like I mean if you use the cancellation technically ever using full system of threading, because that's what cancellation token is, then then figure out, wait, I, don't know if you do configure wait, you're likely to be using for anything.

A

J

Cancellation tokenism system, turning not system turning test.

A

Fair enough, although like yeah I mean I'm finally putting in system setting I wouldn't put in system collections in air because that's relatively clean, there's a namespace, yeah and I would not put in some some stuff that has to do with cancellation and configure a way that does not seem to be long way at all. So would either put it in system threading of system, fitting tasks so I, don't know, I mean creating a new type, seems a bit I, don't know, I mean tasks.

A

Extension might not be the best type to put it on at the end of like what would be call the other one. Like a singing you mobile extensions.

A

E

General, don't we put extensions type in the same namespace as the type of six and we.

A

Try to collective problem is that I think you also need to consider the nice, because you put it in right. Thank you. If the namespace is already super clean and very small, like adding a type in there, that is just a grab. Bag for stuff seems wrong right and, like you, can look at a new mobile, for example, in your mobile. It's not in the same age position I anymore and.

G

You're, you are, you are going to be using types from system threatening to houses anyway, as you write code which iterates over this, because presumably your method returns tasks, yeah you're very likely to have it uses yeah.

E

I mean unless you're Stephen, you know that you can write a customer with yeah.

E

Yes, I, but you could be sure.

B

E

B

This API you're writing and a weight for each write for each you're inside of an async enumerable.

B

Well, but no, that's not returning a task. That's returning! An async is innumerable.

E

J

E

Method, you're calling it front would be async and unless you're anything for your probably async task, very sick values fast. So.

B

For a stick or your and innumerable yourself right.

B

Your preference is to leave awareness, yeah.

A

Man, unless somebody has a brilliant idea like it doesn't seem like I, think toxic tension is probably close enough. Yeah.

E

I think that it just comes back to Steve's question of it. If we think that we're moving async enumerable, then this needs to be out of different types of canoes with.

E

But we can always move that right before jet yeah.

F

That sure was like sports Pamela.

B

Okay, so leave leave those, as is I, don't think I have any other questions you scroll down did I, say anything else below.

E

There's a word cancelable here: are we sure how many elves belong in at this time, so.

B

In general, we've never been 100% consistent, but since I've been involved, anything new we've been consistent on is that cancellation has two L's. Everything else has one okay, so canceled canceled cancelable all have one L cancellation has okay.

B

A

You is that all English works is just because you it up for cancellation. That's.

E

For American English works, Wiktionary calls 1l the u.s. alternative spelling of two L's for cancelable.

C

And I think I think our framework design guideline says: prefer the 1l version.

E

If only someone had a copy of the book right, there I knew that's cute. The pig was.

A

J

I can keep looking what Bart said. I forget what his question was.

J

The type as well.

A

So what is the clashes of her until.

B

I think he's referring to the fact that we're gonna have an async enumerable for link, probably living not not in cortex it'll be and likes. You know, reactive somewhere, they'll be async, innumerable dot, select and asynchrony, and so I create a type called async enumerable that has configure await and what on it, but we're not planning to define. You know all the various link methods and those would probably be on an async in a normal type.

A

Yeah, this is an ever-growing list of things. That's do a candidate for for later or maybe never knew.

A

All right, then, it seems like great, so she has.

B

Rather than that, were there any other thoughts on the api's about how to any other change them I'd love to get this done this week, it possible not to get you.

A

To I it's nonsensical to chain them in in various ways. I agree with that. That's unfortunate, but we have said many other areas, happy much anything that's flu and you can always change it in like weird way. It's right or any instance, method that returns itself, but I mean, like my clinic example, looks like to string and string. While you can two stranded four times and it's nonsensical, but also there's nothing prevents that and.

B

It's there any I can go either way on this from a design perspective. Is there any benefit to detecting that you've done something nonsensical and throwing or we don't care like you can imagine the struct, storing whether it is called weather with cancellation has been used and whether the figure weight has been used, and it it's true, and you call that same method throw well. We care just over.

A

It what's the behavior that I would get I would assume the last one would win. Alright last one would win yeah, which seems, and what you would expect I connected to this being useful in debugging right when you just want to add something quickly to the end, to hole right, something is it earlier and then he gets to remove it. I mean.

E

If it was returning the thing that could be, you know, sit back to the interface, then I would say it definitely should prevent the the hoops behavior, but the fact that this pulls you out of interface land and pretty much locks you into the for each statement now means you, you literally wrote it twice on the same line or you you understood that you had a configured can sablée sync enumerable. The only exception would be if you barred it and I have opinions on bar well,.

G

It's not all that different than like you say that you have a in options: object where you set the cancellation token property twice like we're, not ever going to stop that right. The last one you said mr. Wright's personal I, totally.

B

Mind that yeah.

J

F

True, actually, if you can do yeah, you can just look.

A

Over them as a sense.

E

The only question I would have for this Steve is: do you think it's common enough to want to set both that you should have an accelerator for it? Yes,.

B

Oh I mean do we need like a combined with your cancellation.

E

It's worth avoiding the second copy.

B

You're about to do in a wait for each.

E

G

Honestly, if it, if it shows up like we, can address out of that Chuck indeed, just humor me, how would you name that method in your away, probably.

B

Actually, don't think we should do that, and the reason is we've always talked about like you know what, if you wanted to, if you are awaiting a task and you wanted to cancel you know, you can't that point.

I

B

Can't control how the task is operating, but you could choose to cancel the wait operation. You can call dot wait cancellation token today, which cancels the wait, not the thing you're waiting. So if you imagine, we wanted to be able to do a weight task in the past. We've always talked about that as doing a weight task, dot, configure away cancellation token, in which case the cancellation token is about me because, if you're configuring, the await to be cancelable by here, that's not at all what the cancellation token would be. I say.

G

So you're actually flowing it through to the operation yeah right, that's better!.

F

Booya no I have to do work. The question is well.

A

All right, so we handled that one I.

B

Totally think we should do Jason document, but I just want to make sure. Are there any others that we need to do to in this time? Because presumably this one's going to take the rest of it.

B

Famous last words.

A

You can't let this on Steven is that I.

B

Care about it or maybe I think.

A

A

B

Care about both that one and the one two below it for three, oh but they're, not pressing so I'd. Rather, we do Jason document which is pressing, yeah I.

A

Think there's nothing else. You should probably do this one here really clip, because it's a small and this one given phones, guys.

A

And it should be fast one. The.

G

Bottom, one thumbs up yeah. The bottom one seems like, hopefully, that's fast. Okay.

K

Next, maybe do these two.

A

A

Basically, text boxes get this thing here. You.

D

A

Enter your name and then you as soon as you focus it or as soon as there's text inside of the what they call the the the placeholder text disappears right it so that all they need to do is basically explore the property on on text box. It says: what do you want the text to be order, that's typical and give it right. We.

F

B

Babysit WPA is very different on. Does it not have placeholder text at all today, or it's just hard coded? No.

A

It doesn't exist today, it's literally like custom, drawing that's the feature. Oh maybe.

F

It is only on WTF funny, but.

A

It's a very common thing that almost all modern things have, except when from clever headed.

B

H

B

Think it's a great thing. That is the question: is this placeholder the right term and are there other kind of text things on text box or other API? Is that we give this sort of follows the same naming pattern? So when RT has it, then they code that placeholder text? Okay, but.

E

You said you think that.

A

Guys thought ICP before it might be: WinRT actually I'm, not sure yeah.

G

Because I, don't I, don't see it in WPF, I've watches. Okay, that's my.

E

Twin forms: datagridview placeholder cell style and placeholder string format, but I don't see placeholders if.

C

You search for placeholder text. It comes up with a ton of Stack Overflow results for exactly this. So.

H

D

Yeah seems good thing: yeah.

G

And you know in HTML it's called placeholder, so yeah the name has been yeah I. Think that's.

A

I think that what some discussion of Unicode is null text, but that what's not something anybody like yeah all.

F

Right then, that was look. We actually yeah.

G

81 Oh should the should it return empty string or not by default. I. Think.

A

They Larry find that somewhere.

F

Is not an orientation.

G

He still has to be documented, which means it's part of the API design.

G

Probably know if it's what's up the.

C

Native API calls it the set Q banner of.

A

Course, because.

A

If you go I've seen this well.

C

I'm trying to see what the native API actually does for the default, if anything, because I imagine that's just what the default behavior should be, whatever I.

G

Mean what is what is text returned when there's no text there's all these types of eclipses bulbs by definition,.

B

So then, presumably there's always placeholder text as well means presumably of eternity I.

A

Would probably make it return empty because knows that usually the thing that kills more stuff and I mean empty seems fine. But if there's no place or text no I.

E

Mean that, but then that means that we should throw an argument either either were doing argument no exception or normalizing. If you said no I think.

A

When things move to does normalizing yeah thank.

F

You for said, yeah I think it's even.

A

Legal to say, text box, dot x equals knowledge. You just doesn't have to string. Why don't we show.

B

If only the code for windows forms were available somewhere to check WinForms.

E

Text box text normalizes to empty on returns that accepts Nola's input. Okay, so now.

G

E

The return is returned s, equal, equal, no question mark empty. Other else gets.

D

How we do those so they have.

E

A pattern they never return null and they accept null because their necks like that.

G

Let's asp.net, which is like what form supposed to say no yeah.

C

And winter between ox, don't say anything about what the default value is or whether it throws on no. So.

G

Presumably you're not calling the API in passing another parameter well,.

C

I mean it may be like it's 32 API, so I would.

D

Completely expect that thanks all.

C

G

E

The win32 API, no because they'll just say like if something goes wrong, call get last error, yeah and.

G

Then then, you get like e argument order: yeah II.

C

Bad art yeah well in this case there is no get last error, it's and it returns true or false, based on whether it succeeded nothing else. Oh.

H

C

No because it goes something it's part of a dispatch master, Fung.

A

All right, you want concat version to take something, only span of char and.

E

Cool, why did you stop it? For okay, that's a good reason ship. It.

E

Great well I think that the original, like the C sharp to language specification, call that uptick for yeah.

G

And they are called store, 0 store.

E

G

New yeah I remember.

E

When writing the script, our compiler, that I got really annoyed, that CSC went up to like 6 and it blind call that they didn't basket the framework support of it. It's like well CSC. You just violated the language spec. So then I added one but yeah. We could have more when we keep them and we certainly can't find the prince cause. You can't blocks the same yep.

E

This seems straightforward, and- and this is only four straight calls you don't like- we don't expect. The compiler came to be adding Plus on read-only Spanish. Alright, okay, that's good. We.

A

Can never do a span of span crank right.

G

Great, maybe in three studies.

F

A person who can.

G

Have an equivalent of what's granite spring, whatever, whatever like a very otic, equivalent expand one well.

E

I mean right now you can't have a span of spend because span can't be a generic way of it right. There argument you know with the polite, exhibit I.

A

Mean like it would be to a nice if it could have a parents that is basically spam there, and then at that point you at school here, because there's no way you can ripping them down. Yeah yeah.

E

Yes, I do wish that the language would add the parens span and then and then I think yawn sort of wants to add span of this span.

E

Let's fan be a generic argument when it's reasonable and safe to do so. I think the problem is they're having trouble deciding that that means yeah, which is why it hasn't happened yeah. This is this is a good start. Alright,.

A

When the mix is mama for Jake honey, if you like that, look good with just one yeah did.

G

We actually approve the previous one. No no I mean did we mark the the shiny can count as a proof. Yeah you'll get great great yeah. You just didn't see that okay, okay.

E

So the difference is from last time. So if we start up at the adjacent document part, would you just throw up top there's now a parse that takes a sequence, yeah, so I added sequence, a bike in this API proposal? And so this is contrasting a little bit with what Asselin has the PR right now invasion.

E

Writer I use utf-8 is a prefix when we're talking about mics, because encoding and biting together and I do not use them in prefix when we're talking about chart, because that's a concept, type short so and agree only sequence, a bite for Christophe scenarios.

E

Yeah, so that's the only change is based in document and.

B

Jason documents still immutable.

E

It's still only self mutable.

H

H

E

It holds some buffers and stuff when you dispose it, it returns all the stuff to the airport, oh yeah, but be okay, but if a caller cannot change the dock, so the gonna the DOM is, the Dom is fixed. Okay and then the type is yeah. So it's still a read-only representation. It just holds it for.

H

E

E

In the Jason element, the string based property indexers are gone and they're now just a method. This was because when I was talking about something with emo, he got upset that the property indexer was order in times in being the number of properties times, M being the number of characters in the string, you're looking up operation, and so it moved to a method to make it be more obvious that it's medium expensive instead of like expensive.

B

Okay, come back adjacent document for a second just ask another question: so I previously questioned this, the I, disposable and I I just want to question it one more time, and then we can move on so I'm to two questions about it.

B

One is, you know, other places in the framework where we have xx document they're, not I, disposable and so I'm, a little concerned that we sort of established that you you don't need to be concerned about disposing these things, but I could get over like I could I get to see the argument being made? Well, this is just a new thing and people have to learn. The second thing, though, is I'm a little concerned.

B

This is ID supposable because we're pulling a raise out of the pool right yep. That's the reason, this design suppose a little yes.

E

B

A little concerned that this might be coming in for some usage scenarios, this might become a resource hog, and might you know you might create a long live JSON document and it's just gonna hold on to arrays from the pool for a indeterminate amount of time? Is that a valid concern or am I just making trouble I mean.

E

All so, unless I misunderstand how the rifle works, all that that would mean is that the next call to array pool rinse. It has to allocate a new array which that has this slowly cycle its way into Jim too, but that's just true of any parallel load factor and eventually, when you're done with this and you dispose it, it go back in and now there's just the next identity buffer. They write us.

B

Unless they at that point, you've already filled up the array pool again and so it'll dump the Gentoo buffer I.

E

Don't like the expected scenarios, where we've seen people use this, it's you know a light. They want a lighter version than deserialize on on reading a payload. They read it short they're, holding it like less than a less than a half a second, and then it's done. Okay,.

B

So it's like you would expect this to be per request, not cached in any way. Yeah. Okay,.

E

Based on based on the existing usages that we've seen from Jason net, it's mostly just people use it. They want to quick I'm. Looking for you know this sort of J path, type thing, and now I read it and I'm done with the document get rid of it all.

C

K

It's like I'm, either except more convenient yeah.

K

Yeah I'm actually using now, because we don't have the document I'm using the reader to write some yeah.

K

It's really bad.

G

No, it's it's not bad. It's just not intended for your particular scenario. Very.

K

G

K

It's not. That is bad. It's just the model where you have to have our own state machine and check what you've just read and which is gonna, be scary. Yeah.

E

Okay, so then the Jason element, the type, is no longer the Jason token type. It's now a new thing called Jason value type, which is a copy of there's a second type with without in the Ray and end object and start a range start object or just call the right object. Okay, great so there's the new enum for telling you what you're at any worried, object at anymore array now, which are different enumerators.

E

Or anymore enumerate Horrible's.

G

Oh and you have the obstruction types to find too. Okay.

K

So once why do we have JSON document? And then you have to do brute element when JSON document is not the root element, I mean we could.

I

Copy yet so we every member of the elements, tamper documents, because things: what is the purpose of the JSON document? Type them? White pines, doesn't return the root element because it returns the thing you put any using statements and when you're done all the buffers go back to the Oracle.

K

And the root element London all.

E

Elements point back to the document that owns them. The document is the only thing that knows. What's on elements is just a cursor.

E

And so element is what holds the parched the well? It's not a part retreat is parsed flat element is what holds there's. Our document is, what holds everything element only has one or two fields, one is what's the document I came from and number two is what's my what's.

G

My position ID here XML document, has a somewhat similar pattern by the way what you get the root element from the document and then you even right off the bat yeah the documents, one Jamin elements, if any be.

B

Very clear that people should never do JSON document dot, parse carbon element, I.

G

Mean they can that just stole from the array cool I mean we could bring, we could just have a look, the code, analyzer rule or something that says you know your. This is an inefficient way of calling this. You.

E

Call the method that returned in disposable and then you didn't dispose it. We already have them: okay, yes, you're right, okay,.

E

F

Not okay, what.

E

H

Can find my muscle.

E

All right, I added single, because that came up last time, I also added you. You went 32 in addition to you at 64 cuz that came up last time.

E

The Tri get values all got expanded to try or value that replace with the name of the type. So it's getting 32 and try again at 32.

E

So the the previous enumerator is now a write, a numerator, nothing fancy about it. Now it conforms to the interfaces, so it can be linked. I guess.

E

Object, enumerator returns it enumerates over this new structure, which is Jason property property, has two properties. One is the element that is the value, and the other is a string. That is the name. The structure itself is read-only as an implementation detail for those who care name is delay. The the name property just calls into the document and says, give me what the name is, so it can use a a delay.

E

A son escaped but still cached value for compound, because originally I had it as a a memorized property, but that meant I couldn't have a read-only struct and then I that I kicked it to be a really strong ru overriding to string on all.

I

These x element.

E

Overrides to string, which is it which ones do the correct thing.

G

What does look if my element represents an object? Does it work? Does it return like open curly? It returns.

E

From the open curly to the closed curly without supplying a Satan okay, and that, following my work, design guidelines, you should never depend on the value of to string. It's only been pretty buggy yeah.

K

Yeah so like, for example, takes on property, doesn't know dragon, but there's good, useful, correct. It.

E

Does not override it because I don't know, I mean if someone wants to define what we think of good to string is for a property we can at it. But what does a name column fun to dream and.

F

G

Except for the Albert curly, basically.

K

G

An object like you have open, brace.

K

And the reason I'm asking for it is again I'm using now reason to be there in there, when you send lots of feedback to ask some for things, not overriding for string and not being able to. Even but.

G

This is just for debugging all right, yeah, okay, so.

E

Again, we could say that it was similar to what object does of it. It finds. Where did the quote started this start out to? Where is the thing or where is the value end, and it feels like you're gonna bring them party on and that's fine yeah, so there's Jason value type, which is as I said, Jason.

K

Token type of at the ends- and the word start to removed so I notice that you have Jason right there good rights element.

K

Would it make sense to it won that and your rights property.

E

We could it's really just an accelerator for right element: property that main kind of property that value that property dot; name.

G

Transcodes, perhaps back and forth, so it would in theory, be somewhat of an optimization. Yes,.

E

It depends as soon as are as soon as our API representation is you say that we will always write the backslash U version of a quote. That means that were always green escaping, which means that we have to do the work anyway. So no.

K

Doesn't Jason properties charcoal to.

I

The actual payload it does, but if we guarantee.

E

That we're writing at escape strings using a particular escapement format. Then we can't write. We can't blip. We have to know what's going on check whether it's already like.

G

I would be willing to bet on that, but I just think you're, the one who cares about the way. Yes,.

H

I

Yeah I mean that's, that's fine, dad. So another thing.

E

Which I guess would be right elements Jason property, because it contains the name and the value. They are me weird, because it would be the only thing in that type of departments have value that doesn't take a property.

K

Name so element can be very very large, because I, like element can be an object with another inside and stuff like this yeah. What happens if it doesn't think when you write.

K

Opposed to small, you mean yeah, so it writes to the JSON right there in JSON. The writer currently takes is a buffer writer, but the right doesn't have you can get new batteries? Yes, what happened? See I tried.

H

To write a gig.

G

So I uh sort of Elvis he does have a limitation on how large any individual incoming string can be, and the reason for that is because he pessimistically computes the maximum length that that string will expand into after escaping and then does a bunch of work under the covers to make sure that the I buffer writer can actually give him a properly sized snap. So it should work. Unless you have some. Unless you're trying to write a gigabyte, long string, it should work well.

K

It's not going to come back to long string, so I think this is almost a problem that we should head of discussing a larger group, okay, which is, if I'm using JSON writers directly and I'm. Writing. You know property by property from time to time. I would flash that the actual stream well flash I think is not on my path right. It's on pipe right!

K

So now we are introducing a bunch of ID. Is that potential? You can write massive amounts of data if I have no way to flush, I think periodically between you know every hundred megabytes whenever I think one gigabyte, tracer I mean you should be flashing so Yuri quickly, then once every hundred megabyte, but.

G

This this assumes that you have already built up in memory and element a single element. The corresponds to 100 megabytes adjacent correct now. Is that realistic? Well,.

C

I mean what about the case where you're parsing the file, your and then you're just going to rewrite it to normalize all the text to be either minified or pretty print, or something like that. That to me sounds like there features missing.

I

K

I kind of agree: that's what I said it's it's! This whole thing that pipe writer has classic same exactly. For this reason, it's basically periodically also call it after writing some amount of data, and then you can get more buffers right to them, advance them and then flashes yeah. Well now we are introducing ap. Is that that kind of high level ideas that potentially can write large amount of data? They should be flashing a sync from time to time, but it's not me Dean, there's no way to do it.

K

E

Could I mean so we can also drop this entirely, because one of the things we talked about was how it was going to be so great because we passed this in and it still has a functor back to what the raw text was and I can just copy it. The problem is: if we care about the way that were escaping if we care about the representation of the state values, we don't want to do that and at one point this just becomes that's.

G

Why I said I'm willing to bend on that because I was I was talking about the initial escaping of a value? If you have an incoming payload with data already escaped by because the verifier checks out, so it can just go right out to the output. Yes,.

E

But we don't know it could have backslash quote instead of.

I

K

Okay, let me I would not want my comment to be all the removal of these things. I think they are super useful, yeah I do think that we should spend some time I I, it's um it's it's a thing that we hand way.

G

So maybe Jason, why they're? Just not all right, maybe Jason wider, just like Jason readers is very very low level and Jason document deserves something higher level. As far as writing, those to be its compliment would.

E

Be all the api's off of Jason writer, but with the word a thing could be about them. No I, don't.

G

Really know what, if it looked like to be honest,.

E

K

Maybe it's for a summation not for this yeah, yeah and and way to budget, probably.

E

Responding to the 300 comments on the day, yeah yeah.

G

Buried under an avalanche awarding.

E

Yeah I'm fine, with whatever I think having seen the brighter today. These are also missing the skip escaping or whatever the name of the optical brand or the view.

D

E

The end of the video of it okay.

E

Yeah, so the only thing that I can think of it was in the last review that I said no one which I think I have edit text about this. You think that I would have been saying it actually open. My own.

E

Is James had suggested that he wanted an API that returns nullable and the problem is in nowhere in the framework? Do we have in a method that has the word nullable in it, and we've already have used get in 32 to be a nominal in 32 and I? Don't think that nullable would be the default that we would want so I said no, unless Christoph right now, once the end wants to decide that we can have given 32 and get knowable with.

K

The behavior that James was proposing is you index into a property and if the property doesn't exist, it returns. No, he.

I

E

To the notion that a property is either a number or null and say that an element is either a number or no and then because what he does is he has conversion operators with no make alternatives, so he has a cast to end and a guest to nullable and you just use whichever one you wanted and you get the thing the following the guideline that any conversion operators should also have a method. That is the same thing as the conversion operator. Well, what do we call that method?

E

Do we actually write the word nullable as in UML a ble, and we.

I

K

Done them so today, when, when you most people scroll to the method that we are talking about, if you call that in 32 and pass property 9, you throw.

E

So get in 32 you're already added. You already have an element. Yes in the element type, if it is no get in 32 throws I'm, not a number. Yes,.

K

Well, you can't round API, so you should fix your program, there's! No, no! No right! So James.

E

Is indicating he he thinks that it that this should have had a notion that this is either in it or a nullable, and in this position in your Dom right and I think that if he wants that as an extension method, you can write it. I can.

K

Imagine something like try gravity. What are you fascinate I can imagine those know.

A

His argument is I. Remember what what James said it said like in JSON null is often used in the context of the videos absent because you're civilizing, for example, database structures or other things, and the the polymer in practice is. If we, if we give you these two, a few I get property in whatever. That makes that useful, slightly more cumbersome to use any editor yeah.

K

This one yes I, need to know I need to know what the type of the body was. Yes, yeah.

A

He's just saying: there's another overload that I get another billion thirty-two because you expect it to be, but.

K

Whether or not cold, if Jason value time, who is no and do whatever you wanted to.

I

Do because because the consumer basically eventually wants to get an alibi in right, so it's James.

E

The scenario is you have in your in your class in your c-sharp class, you have a nullable sent ya. In order to read this, you have to write the if the type is known yourself, I'm, fine,.

K

With that, okay, yes,.

E

K

Me too, because I thought, if it was very convenient for us to add the color directly, so you should do it, but I don't know how it would, even though we would have to basically not have get that for return in 64, it would have to return no. No, it would.

I

Be different or we have be pasting, it's hidden, 64 and.

E

Get doable in six but.

K

By the way, my.

I

Argument is we've never added the word.

E

G

I, don't remember if we talked about this last time, but operators like explicit cast operators from Jason I went to the private parts. What's a rule on when we should have explicit operators like a cast jason element to enter, for instance, and it would be the equivalent of given 32 right. So.

E

So so here's the evolution of how we got to here, yeah jason net, has conversion operators with no named methods yeah, so Absalom had conversion operators with no name methods in the prototype yeah when I was moving it into core FX I followed the guideline that says. If you have a conversion operator, you need to have a named alternative to the conversion operator. Steve asked why I have both so I deleted the conversion on.

K

This is that don't conversion operators are very cute, but they are not visible in intellisense and the indian people don't know how to use the ideas. Okay, I get that element and they like color, corrector, I, get now in the end yeah, and they don't see in methods. They see. You can't value that returns a string, they call it and then they parse the end and then it's everything yes super. So, okay. So if.

E

We get a bunch of feedback later that people really really want them. We can consider them, but I think right now, smaller is better. That's.

G

Totally fun yeah.

K

I will be totally fine with adding the operators, in addition to the get methods- oh yeah, you know when, like good code in the future of people demanded.

E

And and if I did all that they would do, is they would just been following what's written in the book, they would just call them, which means that they are tea or syntactic sugar. That's totally fine! Okay,.

E

K

In you know, value is object and basically the nut box. Whatever is there.

A

K

Would be the scenario for that, for example, if I want to write a dynamic, I.

E

Mean yeah, we can add it later. It's an easy extension method. It's.

G

Why doesn't this implement like a dynamic object or whatever it is actually does that.

K

Okay, the only problem I had with it is.

G

Because or dynamic API is one that, for calm implication. Yes, are.

A

You saying all dynamic feature is very heavy weighted, not worried, isn't.

K

Perform isn't wasn't visiting all of the DLR to basically make it not just for calm.

A

Basically is but like you implemented in an attentional static type system, so it's not going to be amazingly faster, but.

K

It looks fine, so.

I

And then the only other thing that I wrote down that was.

E

Based on class feedback is email can ask for the property index to return a nullable, but the operator didn't lift and then that's when we had the discussion of the property actually wasn't cheap. So now it's a method with federal property.

E

H

I'm, looking at the words property, oh so say a property okay,.

B

Jeremy, when Venus got the reader, maybe was writer around big integer if I want, if I had a an element that contained you know, data that was too large to fit in along, for example, but I still wanted to use Jason document. How would I create my own being at the jurors they're like a Jets fan or something that I would use to then pass to being integer parse or something? How does that work? We.

E

Cut get spam during the last year because it was the only thing that cared what the encoding of the input data was. Okay,.

B

So like get it like a integer out longer than long max value well,.

E

If you're, if you really want to and you're willing to ignore the advice and framework design guidelines, you can call to string which will give you the number text, and then you can just parse that.

C

As a big reason, why we don't have a inner text property like we, you on XML.

E

Element, first of all, it wouldn't be a property because it's gonna be hellaciously, expensive and because it's hellaciously expensive yeah. What? What does it think string.

G

Yeah yeah, no, it literally just have to have a method on a property, have a method called get in or text it transpose to shrink. If you want the raw text there, you go like you're, taking the perfect for it. Okay, well,.

K

E

But the question it like so is that it is that transcode, no one escaped is that transcode none escape like there's. We have to decide Wow raw text, so it will be on escaped.

G

Friends, so sorry.

E

It would be it.

G

Would be escaped s? Okay, no just transcoded yeah,.

E

So if I, the FI, the encoding transform from yeah from whatever it was in the own TM, because it's always separate to.

E

We can, it makes sense on element. I, don't know that make sense on property property two strings actually going to be, but it doesn't really make sense for, like property, give.

G

Me your tax: if you want people not to take a dependency on property doctrine, you could always just add a debugger display attribute.

G

I

Gonna go sleep under the table now, so when is.

G

I

Feature available for people who want political API is to know and I'll, say, John.

E

Because you know fortunately.

K

My usage of Jason reader because nobody can grow what I wrote.

K

I couldn't delete to pass a small structure that felt like STEM fields. It's like hundreds of lines of code, so.

E

I think this is going to be in the same situation of the reader so for sharing the reader resourcefulness.

I

E

I

When we like of it, I.

E

Need things from a song which he says that he will get to this week and then I will open the PM or at in fact it's I might open the PR blocked on and those two bugs awesome. So that's this is everything is implemented. Everything is ready accepted. It gives you illegally form strings because we don't have on escaping and it depends on beheading half fixed, a bug in the reader that I want ass on to Pittsburgh and.

K

Also, another thing is I asked some this question in email yesterday, but he's not replying he's.

F

K

So lots of data comes as streams, but your new ad ice take either span. So we read only sequences. Are you guys gonna an adapter, hey Steve,.

I

E

If anyone knows the plan for sequencing stream, it would be you. Are we having a generic transformation of one to the other, so I I.

B

Pushed for us to have just easy stream support, but I was told no, so the next best thing is one of the issues that showed up on emos list that we skipped over was from David about creating an adapter between streams and pipelines and there's pipelines to sequence. You'd go stream to pipeline to sequence.

B

We have obviously some boilerplate in there to do the whole get the next sequence from the pipeline.

B

That's the only thing that I'm aware of I pushed Josh and acid on this, but I've been told it's not a priority.

E

So is your question first off there is your assertion that we should add JSON document that far subscreen, which treats the on the string did.

A

The one you have about feeling.

K

Like you know, there's one question: what I want I need and what yeah a majority of our developers I, don't care whether it's a you know helper, that takes a stream or it's an adapter. Well, but I need one of those otherwise again, I'm. Writing lots of PCL libraries to be able to parse JSON, but if you're. So, if we don't think that we're going to.

E

Get an easy form of this, then. Presumably your question would be, or your suggestion would be, that we had JSON document parse, that they stream yeah I.

B

Think that would actually make sense, although it would probably need to be an async method. Yeah.

B

The reason I heard you know we're not gonna, do it for Jason reader/writer is this is low-level and as such, we're using refs trucks and all that kind of stuff, and so I could kind of get okay. We need something higher level, but Jason document is already higher level and I. Don't believe, there's any reasons. It's already pulling everything into memory anyway, and it would just be a single method that goes from stream to JSON document. I, don't see anything stopping us from and as.

D

I

So parse async that takes stream and then the unconstitution don't good and.

E

Then do we also want the synchronous method, because I think that in the asp.net suggested updates to our async guidances? If you have an async, also have a same set of. Do you think everything yeah.

K

So we need to work on the acing guidance because I, don't know I mean we have there. There are lots of people who push forward, always getting synchronous version for asynchronous API, but I'm not sure about like how much sense it plays or like who are your operations can.

G

Can we go back to the first method, real fast, the person that has overloads to take away the most and bite right, yeah? Okay, that would probably the reason I'm asking is. Historically, it's been very common for people to take like a byte array, wrap it inside a memory stream and then pass it to an API that takes a string as an input.

G

That would be an example of where you would want a sync equivalent, but here just passed the binary underarmor, so we would have to get people used to having this overload instead of wrapping stuff and then calling it. The overload.

B

Sorry one thing that popped in my head and we can go back to what I was just saying, but we've I think made me think of it. If we take a stream, we probably also need to take an encoding.

E

Yeah or two instead of taking the stream, we take a text reader.

E

Stream, encoding and I mean I can do a quick test of is utf-8 encoding great for the byte array. Yeah.

K

Why not just stream without encoding and basically say use custom utf-8 data I mean we don't take encoding for read-only memory and you could do the same thing and.

I

We can call it stream, utf-8 stream and.

E

Where we, because in the ones that take bikes, we've said in the parameter name on any key I, think, like I, told.

I

K

Famous yeah, my point is basically it's equivalent to this method that takes read on memory and we don't state encoding and ring code and trans code on the fly. How can basically tell you this is the expectation if you have something else.

E

But once where I can.

K

H

E

Adding string should we also add text reader which would solve the I, have text and an encoding and.

K

Text reader, that's then, okay, taxi that doesn't let you do what this thing needs to do, which is read by its utf-8 bytes, so we can pass it to Jackson reader. That's for me, I'm just.

E

Just asking what scenarios we might yeah because, like if you wrote, we wanted to connect it to that console back in right, like Miami I'm, going to input Jason stuff, that's that's the text. Reader I'm gonna say the utf-16.

K

So maybe it would be useful, I'm saying I, don't know how you would actually implement it. We would drain into the chart 100 it would drink it. We don't all text yeah.

G

It would have it if it would have I think surprising performance characteristics, which is why I'm nervous about just adding it as an overload.

E

Because I mean that's what the stream one's gonna do, yeah, at least until we have reason that we need to do something differences. Even if it's a sink. It's just gonna keep basic training into progressively larger printed buffers until it decides what size where that should meet. It gives you the.

K

You know the now resides there. Are you gonna try to get it into a single button? Oh these gets constraints. This rains days we got to read on sequences. Well, the current.

E

Implementation does the API shape says that we can expand the type to have better support for sequence in the future, but right now we don't have a news page justifying complicated utility.

G

But because, because we're doing API review right and part of API reviews, it is, you know future proofing. This is it possible that we can add such such an API overloads in the future without destroying this type. So the.

E

I mean the absolute easiest answer is because Jason element opaquely hides what it's document is. We can I can internally virtualize the entire partial structure of things. If you gave me a super once I'd give me back a different object that understands dynamic aging between things like I can I.

G

Can internally virtualize the.

E

Whole thing so.

G

Yeah one disease: yeah: okay, so you want a private constructor on JSON document on which isn't reflected here, but here I.

E

Have a private, yes, apparently I, you know, I wrote this one by hand instead of copying. You know the bottom double to her. Yes, the the type because you can't change it once it's built. It makes no sense anyway. He or she did activate your create instance. One mood elements just gonna throw up I, I, think I'm disposed perfect section, because.

G

That way, we can also, if we need to in the future on the steel, this just make it abstract and still give it a uh an internal constructor or something yeah.

K

Anyway, I would start simple by just adding and over love the thesis P. We call it. You.

K

B

Do have all the other complications that come with taking streams like do. We also have a pool, leave open and sort of thing, I.

K

Mean there is no, it doesn't make sense for it to to have this lip open because it basically consumes the whole stream stores the data, so you pass the stream you make them whatever you want with it. The.

E

Only reference should be if it's a memory stream I would drain it to the end of the current stream in, and you could write more into it later, but.

G

If it's a, if it's a memory stream like.

B

With the existing overloads of like parse, that takes a read-only memory. What happens if the utf-8 Jason contains two documents, one after the other.

E

K

B

That's what I was asking it's like: on the one hand, you get that reaction, though I have pre parse at first and then, if we switch to saying no, you don't we're. Just gonna represent the first one. Well then, with a stream, you kind of do want to leave open because you might have stream that contains multiple documents. Well,.

E

But you can't agree the stream, so if we found in one middle of a read segment that you, what do you mean yet we have done? We industry, you can see back to zero, assuming.

G

That the stream supports secret yeah, but some do some networks turn doesn't.

K

Yeah but like all.

C

The file ones do so, if you already have a JSON file on disk and you want to create both a pretty animated by version of it, why we did twice when you can read it once see back to zero and then do it again. Well,.

K

I mean oh yeah, so, but by the way the reading, I guess.

E

The easy answer is I'm. Not there is I, none, nothing that ever disposes. The stream ever should have disposed the stream, especially this shouldn't, because it's a method and amount of constructor, which means our convention, is, it would not put it in using a statement. So it does not leave leave open cuz, it won't hold it yeah.

K

Now the thing that I would change now that you said that I think the methods should basically bars till they find the closing brace and then returned how many bytes they passed from memory, otherwise I again, I kind.

C

Of I'm not sure if a.

K

Person does exactly because otherwise what happens if there's some like trailing whitespace is legal. What.

E

K

What, if there's something else in there in the memory yeah? Well, then it means how you always have to kind of parse it before I pass it because.

E

To believe that what you have sent me is a JSON document yeah, so we can do that. We can add types consumed where.

I

It takes is charged on also why.

E

K

I

K

Why would it be yeah? Because what if it's not there, you know not possible if.

G

It's not awesome right. Why are you calling this API if you have reason to suspect it's, not a JSON document. I, don't know, I got.

C

It from you a doctor again file from disk, and it wasn't validation.

G

C

They forgot a something somewhere sure.

E

That it's an Bill so.

C

You throw yeah.

E

What's your implementation of false, it's probably throw well.

K

We could so, where do we have tryparse on end and not on this one.

G

Why do we have tried in 232 tryparse? My understanding is because for performance reasons we wanted to be able to tell if a value coming in and such such as you know, coming over the network or something was a. It was something that we could very easily throw an error for or report an error for. So.

K

Here that argument would be this is so expensive that we don't care about the cost of the exception. Yeah.

E

And if you absolutely need to have your exception right, you can or say, zero.

E

G

You need to have your reception counter, say zero. You should.

H

A

Say like if you get an exception and says you know, string is one another. It's probably easy to tell why it's when I run it in. If you have an error in the entire document and says the document valid Jason you're like where's the problem, it's a 1099, wait I! Think when Adam is my semicolon right, so Falls is not going to be super useful. There. That's.

G

At the exception that we throw I believe does contain information about where the error was right: I just throw whatever exception. Okay, then, you know it'll contain many vendors from there it'll.

E

Tell you where it was when it yeah alright. So so, if you, if we want it to parse the first one and the consumer, does that mean that you want to add the out so that we want to overload IBM I.

G

I'm still not sold that we wanted to parse the first one and then leave everything else there, because our other, like Dom representations, don't do that yeah.

E

They all assume it's the entirety and they they read to the end. They.

A

Also, don't think they just read and rely, including the possible well.

E

Nothing clear, disposals, anything no.

A

I meant like, if you, if you would accept the stream and I, think I, don't think it actually feeds the stream, but saying that call us, which is good for doing that. Right.

E

So yes for the problem, the tureen with read first 7 or higher is we now have a either we throw because we unconditionally call seek back 5, bytes and it says I'm, sorry, I'm not see Keable or we say. If you give me an unsuitable stream. I have corrupted your data party on and so I think that if we're adding the stream overload that is always bringing in and that what I wanted anyone, which means we shouldn't have.

G

Mm-Hmm I am not there. I'm I'm checking XML document just for poor precedence.

H

E

Yeah, if we're doing read first and tell you how much we process.

H

G

If we call it lowered rather than cars,.

E

Parses, aside from in the XML document, API parses the words that we will use to mean turn-turn, something.

D

That looks like text.

E

Into something that looks like an object.

G

Because I'm just looking to see like what XML document does, for instance, when we loaded well.

A

The difference in XML document is that, if you say load with with a string, you know we expect it to be your eyes. We download the contents in parts no with.

G

A stream a system iostream, so there's no parse of the streamer believe there's an XML document, upload of Sriman yeah.

A

There's load, but it's no part right, so parse is basically given text. You get a document. Everything else is load given text, you get a document. Yes, how is that different than load? Because if you pause the string to load yeah, it usually is a path.

G

So that's one: yes, I'm string load of string as a final email. So yes, yes, so that would be why we don't rename Arsenal okay yeah because it takes abstract. But every other overload expects the actual data.

H

A

Which is by the way, it not great I, think that's a yes.

E

E

Yeah so again, like I mean we do have some conflicting guidance here of things that we've things that were written in framework design, guidelines and things that we did our design guideline.

I

Says don't use out well, our new.

K

And since spam, our safety is the arts as much as they can, and they tell you how many yes.

G

But those are mainly meant for our developers, who are trying to just who are trying to read chunks of data from effectively an endless stream. This is this has meant for, like a higher level, consumer right.

K

But the thing is so: what is my because, let's say yeah, somebody did start in a text file. They don't approve JSON documents. What is my wrinkles now? What what code can I write? Do people actually do.

G

This, like it seems like we're trying to it, seems like we're trying to say here to behavior that we should have without evidence that people have a dog Jason file that just has invalidation data, which.

K

G

Effectively, what you're saying so.

K

So what would we do like when we discover that it's an important scenario.

G

Then we can change the apos at that point we can maybe we can maybe have a parse first valid JSON object or something if we took what we have right now. So, let's ignore string for what.

E

We have right now and we copy and paste all of these with a.

E

Trailing as much as we can give it a has a default parameter out in consumed, so bytes considered for the bikes ones in chars consume for the charts in strength.

E

That one would be you call the out one and that one will read the first one and tell you how much it read and the one that doesn't give you the out is you expect this is built-in, and so it builds in the if the length of what I read is not equal to the length of the input.

E

We could do that and then that would also be that if we don't do that, that would be the extension me. The stream one again complicates things were freed first, because not all stream support, crap I went too much you know and, and that one feels to me like we would not have the consumed one. It would.

I

Be this one drains to the end? Well, it could be you.

K

Pass me a strain, you stream, that it's not suitable. It's your problem, I.

E

Mean that's fair.

K

I mean that would be the.

E

Way like seek throw.

B

Considering the implementation right now is just gonna basically offer everything into a giant array. Until we have a better option, a better implementation. We could just externalize this and have you know stream, read or read to end or something that returns the binary from the stream? And then you just pass it into one of these existing overloads and leave it entirely outside of jason document. And if you do that, then this whole discussion about seeking and whatnot entirely goes away.

K

We do add such a.

B

Conveniently as part of it, because it's necessary to complete the scenario, but it would actually be useful for other things.

E

We have a screen, reader read and it's.

A

Playable, though, because in the sense that it's some there yeah green hope and paste the code, if you want screen.

K

Reader, we're not giving you like it's a whole discussion started with me saying. Basically that is gonna have the doctors, because I kept streams and now he's saying yes, we're gonna we're not gonna have a permit on the Texas stream. We come at it an IP I, but you will not be able to well.

A

But it is going.

K

To be bought like ten lines.

A

Of code, no, it's not.

K

Copy the entire screen to allocated by the way. Why would they be a lot of work because well, I, don't know if it's to an array or many race I mean there. Are it's not if you won't have to resize, you have to come like it's, it's not the.

E

Happy stream that tells you how long it is a priority.

E

G

Think I think crystal sure, we're coming down to is like its it's easy for us to do the naive implementation of take this dream and turn it into a single byte array. That is that, actually what people want to practice well,.

A

Mean I think I think the collection in general is that if you, if you do this like what what are the sizes, you expect to be passed in right. If you expect this to be like gigabytes of data, then maybe that's not quite but I, don't they will be gigabytes of data in practice, but it will be in the megabyte. Well, no.

K

It was nothing to do their megabytes or not if the stream doesn't give you life and even if it gives you.

D

K

Not trivial because it can basically depending what the stream is. It could.

D

K

A tiny chunks and like it's basically it has. There are many modes stream, is an abstraction for so many different behaviors and like to write this drain to a single byte. Arrays needs to account for all the behaviors.

A

Maybe I don't understand how this thing would differ from you allocated buffer that you guess for K. You can copying and then well you keep calling like. You know, lead into that thing. Once you want out of thing, you relocate you by the way of growing it by whatever, whatever number you come up with, you keep doing this and eventually every byte array. They whose length you know and then you just feed one more one sizing.

K

Byte arrays and doing it with you, know, constellations this and that, like it's, it's not trigger I mean it's not like rocket science. The point is that you know like people shouldn't you write in DC, oh right,.

G

I mean it's list of T I mean this is what list of T does already covers, or are you out story it shuts it resizes, but, like Rafael said there is now more complication because you're doing it in a loop, you have a cancellation. Okay, well.

A

You only have ad range all right, so that's I'm, saying if you look at the we look at the code that does that in list of T it just depend on to advent. I, don't think it's that much code, like you, have a policy how you grow and like that's it I.

G

Think I think we could do it without too much hassle. Yeah, it's just as crystal said. We don't want. We don't want to the average developer to have to write this.

A

G

But I think that's a.

A

Difference Amy needs a pub label, yeah I'm, saying I'm all agree with you. We should have the API I think in practice for the people that need a portable API, that it's not that many people and those can copy deploy. They really care that much like I think that's. It doesn't strike me as a necessarily bad idea. I think it would be way of friends to expose a single method and Apollo professionally, just because we cannot have streamed or you know which end of something which is where the API or duty belong. I.

K

Don't know I'm.

D

K

Sure I'm, not even convinced that stream lead to end is long term the best solution, because, for example, if this implementation is smarter and actually was doing something we print on the sequence, you don't want to eat, least remarried to end.

K

A

Saying, like you, don't even want stream anything, you don't even want a bite out of that. You want something that literally takes water scream, yeah.

K

I think that would be actually like I, don't know why we started to discuss it. Then we kind of wrote back and so.

E

The question is: what's what's the policy on so if we, if we do parse as we throw the moment that there's something passed, the final semantic token then, but then that one is fine, I'm easy to do as a stream. The problem is: if it's, if it's the one that does only the first Jason payload, then either we throw in non suitable streams or we have some notion of if it's not secret, but we corrupt or or whatever. So it's it's about. The policy of this we add parts.

E

First kind of that being is an overload that outs the number consumed or we make a profit where we make a new method called parts first, which outs the number of things consumed is.

A

As far as first always well define do we know when the thing is at the end, we.

K

Know when we are at the end, but it may be too late, we might might have already read too much data. It is gonna strain that step as.

G

We can only read from a streaming chunks, so.

K

We have the following options: we could, as Jeremy said in this case, we throw we only throw if it's Nancy Kabul, which is we unconditionally third option. We mean we, no, no, we believe the stream as it is, and we say we don't care. You should have passed that see Kabul stream and for a function. Is we give you the data back, that we over read.

E

So the string one without without the without over, read there out by bracket bracket over red instead of Elvis, and this is just complicating.

K

A

Honestly, the one that I think, if the most logical one would be the you basically check in parts in the stream, see cable and if not you English Lords, that you do. You can't do that. But if we do that, then by definition, that stream has a length which no means creating a by the rate of the right size and copying all the data in this video. Well.

E

It is trivia, it is trivia that people often get wrong because they assume that if they call read that it will fill the buffer that they give them instead of the return out a number smaller than that. So it's there is a correct thing to do, and.

K

Also, the big is like so this its own for many low level like the eyes we kind of like said those abilities. Let's not. What is the reason that we had this dream.

A

K

I'm saying is: do we do you do what I just said? No I, first I, don't it's still not it's easier, but it's not true. So there's no reason not to have this method. I.

A

Think, by the way, I'm trying to be together are you, okay, with the stream method that throws when it's not suitable, like I, our data.

K

After the end, I wouldn't throw if it's not signal stream, yeah I would maybe I'm okay with it throws, if it's if it reaches at the end, there's more data and it's not single.

K

Because otherwise, you cannot pass a network stream, even if it's completely valid Jason, because it's not seqable, you don't want that became. You want to be able to pass an unsinkable streets I, don't.

G

Think you would want to throw a road network stream into this API, to be honest generally, if you're, if you're getting Jason from the network. Generally, it's going to be like within a mile envelope as part of an HTTP request which will have its own inner stream, which represents just so request payload, which means it right upstream terminates non-sequel wrapping the underlying stream. That's not suitable.

K

Content stream, that's it! You know.

G

That's fine because you expect the stream only to ever have a JSON document and absolutely nothing else ever so shouldn't be, throwing if parse is called the non sequel stream know that that actually goes back to what we were saying earlier. Oh, if there's anything at the end of the document fail just straight up for.

K

Not suitable, no just.

G

Rate a fail so.

K

Now this is this is one of the options, which is so we could always throw if it's more yes or we could only throw.

E

If it's not single right, so I think if we any method that doesn't help you how much it read throws. If there's right like to me, that seems obvious again. The parse method, as it currently as the signature currently is, should.

I

Throw if there is an extraneous data at the end, because it doesn't return you how much that true, yes,.

E

I

Now, if we make overloads.

E

That returning the value of how much was read, then the stream equivalent I think would be it Out's a byte array of what it over read and if the stream was C Keable. This gives you back a raid on empty because it simply put the stream in the right position. Okay,.

K

It's a very strange idea, as we yeah I said it's an option, but it's missing.

I

But it actually seems to make sense of.

E

Its the these are the bytes. These bytes are bike that I had that I I can't put back and like you need to take them, and this is your problem buddy. If you're composing your.

I

If that ever returned so.

K

I think it's an awesome, API.

I

For a low-level.

K

Type, this one is too high level right so.

E

Alternatively, we don't we up the stream one. We only have the read to end option and if you have a stream that has more than one thing you have to buffer it yourself. So.

G

With the with the utf-8 Jason Reader API, you should be able to tell yourself at the end of a document is right. Yes, but I.

K

Have to then demean twice: I have to first read to determine. Where is the end? Yeah then wrap this in a bounded stream that I have to develop because we don't have it in DCL and then parse it to you. Well, you know, then.

E

You ran from the array for the number of bytes that you care about, and then you fill that and then you give me my.

G

Operating on what we should do is offline, like we should just sit down with you and figure out what the scenario is, where you can't tell ahead of time where the end of the JSON document goes as.

K

I said I was not even proposing that we do it in d1. I was just saying: what would we do if we run into a scenario butters, you know like we. Should men API something like soon? We now have a scenario. Okay, now it's butchers so.

E

Now this is just a future question, then my answer would be. We either overlook those guys we'll make a new method called parse first, which then helps whatever piece of data you need. If it's all these guys, it's out to the length of it.

E

I

I suppose genius, and what about the memory same thing we would make parse. We would make parse first memory, our flights consumed okay.

K

So I'll give you.

I

A concrete scenario for we do have to start shopping this down by the way yeah.

K

The ones with memory would I worry about. This I have I rented the bytearray from a buffer pool because they tie in it. I write JSON to it to take some ends, but there is some data because it was not cleared from the pool all.

A

K

A

Guys online and.

I

Why did you not slice the memory to what you were because I need to know where it's the end, so that requires.