.Net Foundation Design Reviews, 16 Apr 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: GitHub Quick Reviews

Description

-15:-41:-49 - Approved: Json serializer support for a property name policy https://github.com/dotnet/corefx/issues/36351#issuecomment-483773918
00:38:51 - Approved: Evolving EventCounter API https://github.com/dotnet/corefx/issues/36129#issuecomment-483778953
00:48:31 - Approved: Add missing async methods in System.Data.Common and implement IAsyncDisposable https://github.com/dotnet/corefx/issues/35012#issuecomment-483779404

A

Already then, we'll be alive. Let's do this thing where go to this I. Think the first of all we look at this Steve. That's the white one right.

B

C

Okay, I uh I guess I can start here now right now, Eric's not here, but we did kind of work together on this and James also was involved with some discussions here, and this is kind of what we came up with um it's. Basically, we need support for changing the way a property named.

C

Our local object is handled um so we're proposing four things here, an attribute that you can specify explicit name just say: hey: here's, the name of the property, no other post processing occurs on that the Nestea attribute and then two properties on the options class, the serializer and then, along with. That, then, is one of the properties returns of class and we provide a base class for that um with uh with a property, aesthetic property to get the default key location policy, we're not providing any other policies like snake case in etc, out-of-the-box.

C

But the idea is that someone else can bring their own.

D

That makes sense it does.

E

What's the camel casing policy, just whether it's camel cased or Pasco case yeah,.

C

But by default property names on you know, on a polka object or typically Pascal case and in a lot of cases the underlined Jason is um camel case. You know starts to the lowercase letter so that so basically it's very common that the Jays son. Then we have to be able to convert the Jason hommes property names. We have to be able to match those up with the Polka object, which is Pascal cased, and that policy allows you to do that and then case.

C

Insensitivity um is another thing related to the property name and then that's kind of independent of the name, because you need camel case or whatever or not, and then in still want your property name to be case. Insensitive um when your DC Rises.

C

I'll, maybe just scroll down to the example section.

F

A

Just quickly get my mouth: it's not that easy with two three yeah it just sort of a few questions that I had.

C

I'll, just let you read this I mean this pretty self-explanatory.

B

The camel casing policy is it the case that this would seem to apply to everything like every property would end up with this same policy? Is that okay, that's right.

C

Okay, great yeah, that was um this is what we started with. Originally, we had a way to specify it on a per type and even up for property level, but um for preview five. This is what.

B

We came up with I mean we do have a way to set out on the property level by yes,.

C

B

Using jason name right, you just do it yourself, yep.

C

Yep, so um yes, and then on with it with the jason, are with the jason name attribute, you can have any policy you want right. I mean you just have to put it under appropriate.

B

And I assume that overrides whatever policy you set like the policy doesn't apply, whatever string you put in there. Yes,.

C

Exactly so, if you put jason name, that's exactly what its gonna be um irregardless of any kind of global study.

A

After that, specific questions, adjacent name is not subjected to the policy right, so you never get called for this guy. As soon as he put Jays name, you basically override whatever policies.

F

E

Is case-insensitive preferred over ignore case looks like we have a mix of both across the framework, depending on whether you're working with like string or a collection or reg ex cetera, so ask again like it like its case-insensitive versus ignore case. Preferred, oh, you mean the yeah so like should we name it property, name, case-insensitive or a property name ignore case we seem to have a mix of both usages across the framework, depending on what you're dealing with.

C

We actually have kick insensitive used anywhere, yeah.

E

About 20 to 30 uses usages of each okay, depending on where you looking.

C

Indicate the default to be false, um I'm open to naming.

E

Like for ordinal, we of course use ignore case, and then we've got stand-alone ignore case in a few places like link and XML and reg X. But then we also have peace and sensitive in some places. Like collections looks to be mostly in collections, yeah.

A

It looks like also it looks like it's mostly in the legacy stuff right, so it's system collections its magics, which is also non generic. So every time you talk about ignore case, it seems to come from the either the string ones or the was the other one. Well, it's just in strings there.

E

Was also one hit in reg X, the rank X looks to use both oh great.

A

Yeah I honestly.

A

You know one works for me, I mean insensitive kind of I know, it's insensitive, makes more sense to me in speaking or case, but it's just me.

E

Does it make a difference from the Unicode perspective.

F

C

Here we are with that um normalizing, the the Jason name coming in to Indian culture.

E

C

E

Since we'll be using it in the context of string comparisons, most.

G

Often, can you repeat the lesson and she said Steve I didn't catch, that.

C

The way that is actually done internally is that.

C

Is that we take the EJ's the name coming in on the you know, on a deserialize and we normalize it meaning in this case you know we converted to uppercase, but what they could do, lowercase using the invariant culture, and then we compared against our internal cache of the property names word trip, which were also normalized. The same way.

A

What is Jason gonna do well away.

C

On the they have a strategy kind of classic or your.

C

A class you basically derived from and manage this yourself ICC.

A

Doesn't have pre-canned one for those that know.

C

It does it's just not it's just not like this here arm upfront, I kind of go through more hoops to do this.

C

And okay, so unless I hear otherwise I'm just gonna keep the name yeah.

A

G

People will know what I mean yeah yeah I'm, just a little.

C

And then this accessibility model here the idea is that you know we're gonna provide the camel case one because that's by far the most common people want their own snake case. You know, whatever policy they might have, um they can write their own and derive from this class.

A

The home training.

C

A

Let's say I like the camel casing, policy and I want to just tweak some of my names because they are not crappy right, like XML or I, know some other abbreviations that I want to tweak and I extend the camel case in order to have to implement this again myself.

C

Currently, you have to you know, copy the code, I guess, because that actual implementation is internal yeah.

A

And the only thing you could do is just use the static one yourself, but then the first time you say you know come the case. Name, you just say: I, don't know where the aesthetic API is and then you just converted again yourself right are.

B

You actually sorry, would you actually have to I mean? Could you just call Jason name policy ignore case dot, convert name.

H

Right before the call, if you want yeah.

A

So what you do in ten people to be able to inherit from this right see what you would need to protect a constructor on Jason, a policy I mean if you write the code like this in c-sharp. That means the compiler will omit the the protected one, but an API reviews. If this is the metadata, that would mean that you don't have a constructor, so nobody couldn't have it from this. You want! Oh okay,.

F

So that yeah this you know you.

A

Probably intend to have a quadratic I'm, sorry I, look in my ear and then the my other question is: should be half one that doesn't return a string, but basically it does it inside a span. But do we know there.

B

Okay: what's the, how does the implementation handle the coach and around this? Is this called like once per type and cached.

C

Yeah, it's um kind of during the warm-up phase, look through all of the properties and detect this global setting and then cache it and.

C

And then the ji-sung incoming and so there's no perfect, honestly, there's no perfect once it gets warmed up, because the incoming Jason is expected to match exactly yeah.

C

Yeah serialization. We take, of course, the camelcase name and and I'll put that right.

G

I might change my mind on the name because you're not performing a comparison, you're performing a normalization before posting.

F

Which actually does matter the rich quality? We're talking about? The vision am I talking about so the.

G

You'll want to convey the idea that you're not doing a case insensitive comparison of the data coming across a wire against a property name on the type. The idea that you want to convey is you normalize both names, and then you do an ordinal comparison at both names against that bit. Actually does matter good reason that you acknowledge the.

C

Turkey Berrien culture, sorry.

I

What was that? Even.

C

Though I'm using than Berrien culture, why would it matter so.

G

If you have a lowercase, dauntless Turkish line and compared against a regular capital I that will pass as true for ordinal ignore case comparison. But if you have a lowercase, dauntless Turkish high and you call to upper invariant on it, it will return back the original lowercase dauntless Turkish I, which no longer compares as equal to a capital. I like there. There actually is a difference between normalization and ordinal ignore case comparison.

C

Okay, I do mentioned something along normalization performance comment: I have but yeah I will change. The case-insensitive then say something along yeah.

G

C

At all these things.

G

You have to like normalized.

C

Okay: okay, wait: no I'm you're, referring just to the camel case and then okay well.

G

I'm not going to anything that conveys the idea of comparison. Okay,.

A

Selecting complementation standpoint like is it then, okay, that we actually normalize them, rather than just comparing them on the fly? Actually.

G

A

Don't think so, but.

G

If, if the I, so here's the thing right you we need to generate something right right like when we're sending stuff out on the wire. We actually got to generate something like that point. That's normalization, because you don't have anything else to compare your data to, but.

A

You mean normalizing from the properties as the original, a that we have to apply some sort of yeah. So.

G

I'm, distinguishing between these two operations, when you have a comparison, you have two strings as inputs and you return a boolean measure output right when you have normalization, it's straight string out right when you're serializing you have to normalize or I, might experience string out when you're deserializing, it's a bit wonky, but.

A

I think Steve correct me if I'm wrong, my understanding is that this property named case-insensitive is not for describing this it's for, describing when you destabilize how you match the thing in the JSON payload to the proper conditions on the top, and so.

C

A

Normalization, then that would be just watching.

C

But but because I'm, a performance I think that it does make sense to normalize.

D

C

Because, basically, the the way that the properties are cached in a special type of dictionary.

C

Iii think that otherwise we'd have to loop, have it have a real dictionary or some other kind of mechanism? But I search is the only case and like you.

G

Have a concept of a lump over property Valkyr up when your deserializing.

I

Currently, we don't have we don't, but yes, we plan on having them. What would the string compare beyond that property? Does.

G

Whatever you said here in variant culture in your case, hopefully I want to know if it's ordinal in your case, then this behaviors interval, because now you have now you now, someone can send you data which won't bind to a property, but where you look up that exact property name inside the leftover data dictionary. What we're trying.

A

So, okay, so that basically means you should we should be consistent in the way we describe this right, yeah.

G

A

Still don't understand why normalization is more efficient, though, because to me like you, basically, you basically read the payload. You take the string, that's in the payload and you look up a property somehow in your in your cache things meters, the dictionary lookup in you have to agree on what the comparator using so no matter like you either we say if you set this to true we're using ordinal. If you set this to false sorry, if you set this to true, we are saying or don't ignore, case and set it to false. It means ordinal.

A

Then that would make sense to me. Yeah.

C

Because of the way the hash table is structured, it does byte wise comparisons at once. After the normalization occurs. That's why I mean I. If we don't want to do byte wise comparisons for the key, um you know, that's just a perfect and I have to say: okay, if property name is insensitive and take this other slower path, otherwise um you know: do the best map just go.

I

Get slower no matter.

G

What? Because we have to normalize it.

C

G

That might be the best time to do right now, honestly, just so that we don't paint ourselves into a corner when you say this would you're affected. If this were we, we don't have a backing store of normalized names where we compute stuff on the fly. Yeah I'm, but I- don't really want to sidetrack this. On implementation, details I was trying to figure out if this actually affects the API.

A

It's behavior, but I think that the API has to reflect that behavior and I think we need to agree on what the behaviors that we think is the right one. So that's not the done thing at sidetracking.

C

I know early on or something goals, of course were high performance for the serializer older functionality yeah. But you don't want to subscribe all, though that's the thing we don't want to do the wrong thing. um So if this is, you know a very rare thing to do, and it's gonna be a surprise. People then find we should. You know, probably change it um and, like I said it already has a performance concern.

C

But now you know, if you um a lot of properties on your vocal class, basically have we have to fall back to a different algorithm, um much slower over in your dictionary. Basically.

I

G

At your place, I think we can actually I think we can actually solve that under the covers. Without the Navy I guess, I have high confidence. We can do that because property names are almost always going to be asking. So we can special case out.

F

Yeah we we with them yeah we can. We can do.

A

With the name then change, but with the name same thing.

G

Let's leave that alone: okay, I'll.

C

Sink with you, then Levi afterward, yes, I, know.

G

It we are your friends doing it just means. Convert name can really only be used for a radiant a once. It can't ever be used for deserve, because it will return potential, which.

A

Is what I would generally expect anyways, because otherwise to create a fuckton of strings in your decoding? No.

G

What I mean is um you can't call convert, name, cash them and then say we expect only these exact things to be inbound from where you say that would be incorrect.

G

It just means we would need for the convert name proper. For that convert a method we would want to say this method is only intended to be used during the serialization. You shouldn't call this and catch the results in order to speed up this orientation, behavioral good you, you will potentially deserialize along Dale, where you'll potentially perform incorrect decisions, meaning.

A

You're you're not mentioning something that you should have yes, okay,.

C

Well, I mean um we can talk about the implementation later, but I mean my assumption is that we can catch the result from convert name um regeneration.

I

G

Don't think I can do that I think well, I think there are salts. It tells that, but again, okay, that doesn't respect the API. So I don't talk back to her. We should move on.

A

Well, I understand like if you say you say it's insensitive true, you can't do an exact match, anyways right. So how do you compare them? Then you store both the lowercase in all uppercase. It's.

G

You you convert them with dusto.

G

There are ways to do it. It's just non-trivial.

A

C

Values false and we're gonna, you know look identically as the basically coming in at the default.

A

Your mother's new consumption is you just have a comparison there and that's what it used to index into your dictionary. Yeah.

G

Tim Harris can't be used to generate data. Comparison will be used to retail well,.

A

There's I'm, saying I think that I would assume that it is earlier. The convert name is only done once on startup to effectively pre compute. What the output name needs to be and.

G

A

You store this in some sort of property, descriptor that you cache based on property income whatever, and when you look up data from the payload using this property you either choose or annoyed water or ordinal, ignore case to look up your data and then that's how you match it.

F

Would be my name.

A

You, the world.

F

F

G

G

G

A

Just said he thinks that's the correct behavior I.

G

A

To think about my own fundamentally.

G

Different operations like they are truly fundamentally different operations, yeah.

J

I think it's difficult to.

G

Reconcile this in my mind, and we yeah doing paper or.

C

Apparent pacing you mean are truly fundamentally different or.

G

Generally, generating an output string and comparing two strings for equality like normalizing, the string and comparing two stories for equality are just fundamentally different like they. They are completely different things and we do tend to paper over that in the framework yeah.

A

Basically, logically, is if you say, compare these two strings and before and in comparison, that is case. Insensitive is different from saying. Let me convert both strings for a casing to upper or two lower and then compare the results. Provide equality, correct, they're, not result in the same output.

G

A

All right so then I think my only other question was earlier like should this be spam based, but given the V cache name, it seems, like you know, the payload sizes. You know the other dude, sorry that the types are finite. So no matter how many requests you get per. Second, you can pre catch them. If you really want to do the.

G

Only the only way that that might be a problem is, if you do, if we do in the future, have a concept of a just, a regular property bag, and that we want to be able to send out over the wire what all of those have to go through this as well.

G

Does that mean we need a stand? Overload.

A

Yeah I would think.

I

Go to the north.

I

A

I think the poverty bag would only be the residue from what you got I, don't think we would support a property bag for output Oh. What.

G

About if you're serialized in the dictionary.

C

No, we do plan on us on heaven all mad and you can put on a property that contains a essentially a dictionary. What the overflow in there underneath serialization and then serialize that.

G

Does this policy also apply to keys of a dictionary personalizing? It.

C

There is a dictionary key policy property here on this, the last one, so we.

G

Just said, Rick I'm.

C

A property name on a focal versus a property name. You know, I mean a key game in a dictionary because you might, you might not want to have both enabled I, don't think you'll do snake gates into one camera on the other, so that you might want your dictionary to be on.

D

Regarding naming, should Jason name policy convert a message to all be adjacent property name policy; convert property, name policy.

D

Similarly, with the attribute strobing Jason name be Jason property name, my.

C

Census use a dictionary keys as well, it's not really properly, it could just be called Jason.

C

Well, I mean string, converter, Paul's here, I, don't know, I mean, but but.

A

Jason, it's always the property, though right, no matter how you got there, but that's about my property or got net key, but it's still a property in Jason.

C

A

C

A

I mean honestly I'm a little slow, I. Think Jason name as an attribute makes a little bit more sense when you put it on the property because it reads nicer, but yeah either one works with me. You.

A

Awesome depend some more like hold on. You feel here.

D

I think generally, the small name is not descriptive enough and so I think it should be properly named.

D

I rarely because I don't want things to get confused with Jason strain, or anything else relate to that, because there is a concept of name generally M Jason raised properly name. So what is name.

C

Yeah I know it originally came from because of the attribute was called Jason name so I mean we could definitely call. The policy gave some property name policy that works.

D

And maybe it should be Jason naming policy, it show Jason name.

A

Works for me as well.

C

If properly naming policy works,.

C

Mmm-Hmm then we should probably look at property. Juice are a property name, case-insensitive has but um casein. You know well I'm, not gonna, revisit that I guess I'm, but I think that's fine.

C

Okay, if there's not any more comments on it, I did have a couple quick questions for feedback on behavior other than what I haven't need to discuss with Levi afterwards about normalizing um right now, I mean I, know this. Isn't this really design, but what's the general thinking like if one of these policies would return a null or an empty string? Should we just ignore that or does it make sense to throw them I.

A

Mean a convert name returns now. Yes,.

C

If somebody writes a bad policy, basically, we.

G

Like within the framework, we generally don't care about our dependencies, we're turning bad data to us generally, we just know rough or something like that really now, generally speaking like, if, if, if we call a virtual method on our own type, for instance, because maybe someone some class test like we generally don't defend against the virtual method, returning all like, we generally just keep our people also. If this is hooked up with bulla Ville and well, it doesn't return a no little string.

A

Empty string will be pulled mining because he wouldn't lie revenues presume one correct payload. You.

G

Produce some halo, they told you to produce.

A

G

A perfectly valid key value.

G

F

A

Yeah this looks like missions, video, alright,.

B

My intuition would be we we check for null and throw it for no other reason. Then it would help someone diagnose where they have a bug and doesn't seem like it's on a per sensitive path: ya,.

I

B

Like one check and a throwing of a, we must have some Jason serialization exception, or something and just use a string that says: hey, hey idiot, don't return, though yeah.

A

We should probably also do the same thing for empty string and let's really feel strongly that we should support them. But normalizing property name to an empty string seems like a bad idea.

B

And then to Levi's point: when we add the knowability annotations it'll be a non nullable string return types of someone will came to my other warnings if they try and return null most of the time right.

A

B

A

The argument of leg should be guard against this.

A

So if you think about the most common use case here, which is all I want to I mean I'm generally happy with pasture case or camel case, but I don't know my SDK that I use for Ezra has some weird wacky names that need to normalize so for the most part I just forward to the existing policy, then a few things that do my own coding and then that up so I didn't do that no means normal testing will probably find that your payloads are mostly correct, except for the one case where you had the special case, where your code now returns invalid data, which might actually be hard to find.

A

Is it somewhere deep in the bowels of your? You know, I work effectively, so I don't like if it's not particular I. Think honestly guarding is probably better anything you and we should probably throw a method operation or something like this or I.

A

Know what else you would grow, but an argument speaking.

C

For the empty string case.

A

Yeah I'm: do you know in both cases it's like it's not an argument. Right, you're kicking an output video, but but.

C

But back to yeah you can't really have a empty property name and a poco object, but I mean it would be Villa Jason. So you could put it in a property there or something like that. And then you know we, the converter. We could special case empty string out before we call the converter, assume there's no conversion, but someone might want to say: hey I want to convert an empty string into something else. Who knows so I'm inclined just to leave empty string alone and then.

J

Throw our null web for Jason empty key, finds a lot of people complaining about frameworks that blocked it. So we shouldn't bother yeah yeah the.

A

Only thing I would say is like I think the way you would do this is, you would just put a Jace name attribute on the property. You want to be the empty string like I. Don't think it's a very likely scenario that you want to be able to take a non empty property converted to empty or take an empty one and it'd. Make it non empty right.

A

So think of the context of converted seems like the wrong thing to do. It's.

J

A they're sort of orthogonal, because the attribute implies that you own the type and can change it, and the name applies to anything that you walk through the serializer or sorry. The policy is anything that you can walk through the serializer. So if you wanted to replace everything name to do it with empty, then there's no way to do that without letting us let him to go through I.

A

Guess that's fair! Let's take that.

J

The runtime versus design time and we shouldn't have different filters, but null doesn't make anything.

C

Yeah thrown on then um okay. My other question rate someone related, is in the sense of the egg, where I guess is, if you have case insensitive or you have camel casing, and it's not possible that you have the same property on this. Oh sorry, the same Jason property coming in that can map to more than one focal. You know property um right now, I mean in the implementation. It's just not defined because you know, like whatever reflection orders. You happen to find whatever property. First I don't know. If I should special case.

C

That could be an exception case as well. To say: hey, you have you know, poco class I, don't know what property to assign this value to.

C

Are we just um is: um is none undefined, okay,.

G

If the developer makes two different Koko properties with the same name effectively just for different case yeah,.

C

So I have a property called foo and uppercase and a property called foo and lowercase. And then my jason has blue in a mixed case, say for example, and I turn on case insensitive, um which one do I said well.

G

I would think that as you're in the D serializer like I'm I would imagine that you're building up the list of properties to be bound before you start deserializing I would imagine that at that point you could probably just detect and fail that hey something's, something's ambiguous here.

C

All right yeah, we.

H

Die because we want to give you a lot, we.

C

Do lock that property down on the since you can't change it? You know what sterilization started.

C

I'm just yeah, like I, said it's just a question Joey. Do we want to try to detect that and throw some exception or are not worried about it? I.

G

Would point you I wouldn't detect that on the serialization I would probably detect that on the instantiate in the DC realizer itself be honest.

G

You like, before you even have a chance to house a table into it.

C

Yeah normally, then, on the first time, you deserialize is the time that we would build up that cash.

I

At the same time,.

C

I

C

I mean but yeah you're right that it could be done once I had a time arm during one of my plays and not you know, have to do something every single time.

C

Okay, design, I'll say you request, but okay before I and I also don't have to apply to camel casing as well, and so I have to loop through every single property, vertical power policy on it and see. If there's a duplicate.

G

Which is fast well, it's actually not that bad, because it's only done once and the way that you can do. It is literally just new up a half Sun using whatever compare, is appropriate and throw all the names in there and see if there's enough.

G

But it's work you do once and it's good pretty fast. So.

D

What happens if somebody sets the property amusing, Jason name attribute that matches another property name, I mean.

G

That would be what you do right like you, you would when you're, looking at all of the properties that are eligible for or binding on, the type you would run all of the property names we're the appropriate jason name policy get what the converted name is and then take all of those and throw those into a hash set and see if there's any collision there and that half set would use either an ordinal compare if case-insensitive is false or an old, ordinary lowercase compare and place insensitive differently and if there's a match, then you say, like I I, don't know what you're asking me to do.

G

And this would happen well for both for serialization and deserialization. It would only happen producer negotiation right now, like serialization, tend not to be worried about.

G

I mean if you, if you wanted to garden for these serialization or for serialization you code, it's just I, guess I, don't care all that much I, mainly care about data coming in going out. Okay,.

C

So I mean the feedback I'm getting is yeah, we should. We should verify all attributes and all properties on the cocoa um against. You know, kennel K or whatever policy you might have and then case insensitive and then zero.

C

Yep, just everybody agree with them: I mean are not agreed. All right has an opinion.

C

Even though we say it's cached on, there still is a cost associated with it, especially if people misuse the options class.

C

But yeah it is cached um and only done once. Forgive us hope.

C

Okay, I have not nothing else. Then I'll make the changes. Some of the naming changes in the protected on constructor. We talked about and announced a validation logic to the design.

A

All right after the day, I posted the notes and I updated the mail all righty. Then let's people.

A

Let's talk about.

F

This one here Sonia this isn't suitable. It's.

A

A 61, so they have 40 minutes left because I'd love to.

A

Anything else you want to talk about. Alright, let's just go over this one here, so.

G

It's just a decent profit.

G

I guess we just posted.

A

Think that the exam yes.

A

Yeah, basically, they are proposing a bunch of new methods, but none of them exist today.

G

You can transaction you soon. What.

A

Yeah, it's not, unfortunately, begin is used in the ASIC model. That's all that's in there. No.

G

I know because we're getting a transaction is this open network Ventura.

A

Yeah and the API will exist today and the non synchronous version. So it's just.

B

Do you know what the difference between clothes async and dispose async is.

A

So, in a do the net on the connection you can close it and open it and closing it and open it. I can only disclose it once got.

B

It that make sense.

B

Ironic, that you can cancel, cancel.

B

That feels weird to me: what is that? How does not mean I.

G

Would assume that.

F

Terminates the I/o request so.

A

When you run a command, but the command might be, you know, delete the entire database right and halfway through you say, cancel, but then they say it rolls back. The transaction I see.

B

E

Could also say like if it takes too long to send this cancellation request, then don't hang around waiting for that to go through extreme I. Guess.

B

What I'm wondering is what generally triggers a call to cancel I? Think if it's a cancellation token, is it the same cancellation token that's being passed in, because then that would be a no op kind of you're.

A

Saying if I call begin transaction and then, like you know, I, don't know, pull whatever the thing is called execute a single DB command and then later on just kind of the cancellation token how's that different from quality, cams amazing.

B

Yeah I guess I'm asking two things. One is exactly what you just said, and the other is who's generally responsible for calling cancel a sync. My intuition- and this could be entirely wrong since I'm, not particularly I, haven't used. Db command in like ten years is generally with a cancel method. You would do something like cancellation, token, dot register and the delegate you pass call something like cancel right so that when the cancellation token is triggered, it calls cancel.

B

But in this case, if you did that, presumably you'd be passing in that cancellation token and by definition, would have already been cancelled when this was called so I'm trying us I'm just wondering what the what is a typical usage pattern for cancel a single look like.

B

And to your point: if it's actually just cancelling something on operations, you previously started, they all take cancelation token. So, but I may just not understand its usage. The.

J

Only places I see internally to do it or on cancellation token, that register.

G

Hold this way, no.

J

Ones who call the current cancel.

G

Is the only thing I'm just trying to imagine scenarios where you would want to pass an account solution, token, which is like if the network operations taking too long, you want to be able to cancel to cancel operation, but if you cared enough to even send a cancel command, you probably wanted it to finish yeah without interrupted up right.

B

Which is the same generally, the same logic that we used to say to suppose a sink doesn't take a cancellation token, because.

G

B

Is about tearing down your resources? It's often happening when you've already cancelled something you cancel something, and then you call dispose and clean up and a finally block and so I'm wondering if cancel I think is a similar thing.

G

B

Suppose they think is special, so the the reason we chose value task is its. While most implementations will probably just end up returning a value test that wraps a task or that's just default, because it's already, you know complete synchronously. There are going to be a fair number where it actually can be sort of the more sophisticated implementation that we use as an existing object and in.

I

B

All of the async generators the compiler generates, do that the same object. That's used for the enumerator, that's used for the services for the innumerable is used for the enumerator is used for every value task returned from move next async and it's also used as the backing for dispose a syncs result. Yeah.

G

I, just I really wanted to make sure that wasn't the type of you know. That's.

B

G

B

Be I think disposable interface says that the supposed async returns a value task.

G

When, when people were implementing async disposables, do you generally recommend that they expose a public disposed async method as done here, or that it is an explicit interface implementation.

B

Let's go question: we haven't really discussed it much.

G

Which means that if you really wanted to override them, like you said, make them humble yeah.

B

I mean given that it's virtual I think it makes sense for to be public if it wasn't virtual, I still probably I'm still not sure, there's any good reason to hide it.

A

We don't have to dispose better and effectively than 4ac.

B

No because the dispose pattern is entirely predicated on just notice.

G

B

From either this disposer takes a pool being called from either a public dispose or the finalizer, and you don't call this from the finalizar. So there's only one so.

A

What does it mean if you, if you are, if you're, a sink disposable and synchronous disposable, that basically means that the synchronous this blows? It's.

B

Up to the caller decide which to use very and.

I

B

Functionally they should be equivalent just one doing sync I own one doing Asian guy, oh yeah,.

G

Jeremy, can we kill the dispose pattern in the new frame with design clapping, because.

J

I know, but we don't migrants, bowl yup,.

G

Sorry, man sorry go ahead.

J

No because we don't make and we don't change rules unless they're broken and that rule is just outdated, but there are still times that you want to add a finalizer. So it still makes sense.

G

And we don't generally.

J

Don't really hide dispose as making it in an explicit cast, because a lot of people find out if something is disposable by using intellisense and typing dot. Be.

J

Hiding should be done only when needed.

J

The synchronous dispose will keep the pattern. The way it is and we'll say, async dispose, doesn't need it because there's no final under.

B

These are all these objects already employed on disposal right, correct.

A

The other question I have for you is like the clothes acing, also be translated tasks. Your thoughts in there, everything else, those tests except those days, yeah.

B

I know shy thoughts through these and was sort of taking into account my guidance that unless it was perfect, it should return tasks and perf critical on the async completion path and radicular, I'm masu I, don't know why I'm having trouble understanding why clothes async would be valued. Tasks could be worth asking him and maybe there's some scenario: I'm, not sure what the scenario would be. They.

G

Might have this knows async all fours a sink, and it might just be for the meaning.

B

While disability, because Bo's they think, could call closed I, think that would be fine, because you can wrap a value task around a task. You couldn't do the opposite, you couldn't have well, you could have I mean close I. Think could shall wait to suppose they think it would just be. Oh, be an async implementation yeah. It's.

G

Just they, but if I were if I were designing this like if I have, is this those method calling a closed method? I would probably just or ease of understanding that will be just propagate the return type because it'll mean yeah. It just wouldn't occur to me that one is implicitly convertible to the other right.

B

Yeah I think my two questions are basically around close, a sinks, return type and cancel they sink in general.

A

Yeah that recorded both of them I will ask them that and I think we should.

B

And what do you think about the comment at the end, yeah.

A

That was the other question that I just love like it seems a bit weird to add a method just for consistency reasons, but.

G

I mean I guess there was a lot more discussion there. We most well.

B

C

J

B

May be an explanation for clothes.

B

B

B

B

B

Yeah and it seems like- maybe it should just be returned tasks, but it doesn't I know whatever is fine.

A

F

I mean, if you do your other performance.

B

Just closed actually disclose return, the connection to the connection pool or is it actually close, the connection and just get rid of it.

A

I, don't remember: I thought that at least sequel the spooling is a stevie connection, waffles, yep or doing so well, but the bonus, that's an abstraction that you would have to go to the actual I think pooling is not done in the framework. I think it's provided specific.

A

Like at some point, I'm pretty sure secret the connection pooling already under the covers, but.

A

hmm But do you assume I.

B

Was trying to understand how expensive clothes, actually it I get clothes implies. You have to open a brand new connection and do all those round trips. Then it's definitely expensive and- and it's not just a close call that matters it's what you do to get another connection. But if clothes async is just putting the connection into a collection and then the next time you go to say open it just grabs it from the collection.

B

Then in theory I could imagine, maybe there's a tight loop where you're calling open-close open-close open-close and it's just reusing the same connection over and over and over in order. Yeah.

G

It looks like they actually do, have some kind of cooling mechanism. Opener I mean I'm just getting the code, but I see references to your connection but and that connection pool is at least touched from within the dispose method or from within the close method. Sorry.

A

About to leave the room, so let me just write the last thing.

A

You all right, we bounce back, I needs work.

A

I mean they can come back to us all right. Let me enter scream them and then, if anything urgent comes up this week week, I can set up a one-off I was hoping for two hours, but apparently the room got lost. Yeah.

F

A