.Net Foundation Design Reviews, 14 May 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: System.Data

Description

00:00:00 - Approved: Add missing async methods in System.Data.Common and implement IAsyncDisposable https://github.com/dotnet/corefx/issues/35012#issuecomment-492336595
00:31:15 - Approved: Add missing feature detection properties to DbProviderFactory https://github.com/dotnet/corefx/issues/35564
00:36:11 - Approved: New System.Data.Common batching API https://github.com/dotnet/corefx/issues/35135#issuecomment-492347155

A

All right we are live, but apparently my screen does not quite work. Let me first fix that because the pillow, my second one it or does not show up this one does, but not the here we go. No. This is one there and apparently also have audio. So that's not too bad alright. So let me zoom a little bit summary of its answer. What's.

B

Going on the order you want to go through or I mean, there's one, that's quite big, which is the patching one. It might be good to get this out of the way first, possibly because the other one will take up some time. That's right, so this issue is basically about filling gaps that we have in a do that with regards to async api's.

B

So there are FECA some KPIs and a lot of the important places, but there are also some important places where it's missing and it also adds a async disposable as well to all classes. All the abstractions are currently implemented. Opposable mm-hmm.

C

B

So there's there's already been some discussion, including with you on the issue, so this is already undergone some changes. Yep.

A

Yeah, we actually do something asynchronous when you do a sip, this, both sorry, which one we.

B

A

Something in synchronously and you actually decide.

B

So it depends. This is an abstraction, so it's gonna depend on on whatever you know, whatever provider it is, but, for example, a connection could be an unpeeled connection, in which case when you dispose it would be closing a connection, possibly sending a server message over the network and.

D

Even if we were even if provider has a pool, there's a reasonable chance that it's full and we go to return the connection we end up having to dispose it rather than return it to the poor. It didn't.

B

So it depends which kind of pooling strategy you have I mean I'm, not aware of a provider that has like a full pool and then which closes it because the pool is full. Usually you would you always have a place to return it into the pool. So, but pooling is, you know, is open and can be implemented in any way possible. It would grow the pool better than a human connection. Usually you don't allocate more than than the pool, so you have a very fixed. We had a conversation about this one. So.

D

That someone asking to make a connection with my.

B

Weight I would wait exactly so there's a very hard operon limit, there's a max pool size and you never want to surpass.

B

Usually so that means that you never end up in a situation where you have too many connections in you can't return other.

D

Things that would cause a connection to be poisoned, so it couldn't be returned to the pool it.

B

Could be broken for a variety of ways there could be. There could be an I/o error, obviously, and that kind of stuff.

E

No, it was the one there are situations in which, for instance, about time detects that the the connection is completely broken and needs to imaginate the whole pool. All of them are going to be broken, hmm and presumably there are non in-memory.

D

Providers that don't cool yes.

B

So there's, for example, sequel light which is not necessary. So it's not an in-memory one later, there's an embedded one, but a pool is not necessarily something. That's gonna enhance performance. Very much because it's you know it's a local.

D

Thing so what I mean? What's the situation, we're disposing that the connection would actually do basic coursework so.

B

One case is: if the user has disabled cooling, for example, one one good cases you may be using an out of casa, spooning solution, that's not inside the driver, so you may be connecting to another. This happens a lot with first person. You connect to an external component that actually does believe for you, but which means that you are doing a sink and a sink operations. When opening and closing and you're connecting to that external component over socket CCD.

D

B

Yes, that's a that's a good and that's a strategy for when you have a lot of client service and you want to do pooling at a centralized level. Kherson embossed, the driver pooling, is going to be in process on your specific server, whereas you may have other servers. You want to balance. Of course.

B

It's an example. One.

E

Aspect I want to point out, because this is on the obstruction. Then we need to fall back implementations for all the guys that on automatic, oh, if a client is not going to Clemente a bunch of pieces right there, but we need to implement there.

D

Sure I mean the fact there to virtual by net New. York necessarily means that, and we have that on stream. We have that on text writer, where the base implementation, you can't add absolutely.

C

D

Are we went back and forth on what to do in those cases whether it should retrieval work calls disposed whether it should just calls I suppose that we ended up just saying just a synchronously call disposed in the.

B

Rest of the idea done that a guy already does the same thing with ACP guys, which were introduced at some point. So it's consistent.

A

So, what's the difference between begin transaction.

B

And beginning the B transaction, so that's a general pattern in. Do you see well.

A

I'm just looking at it yeah it's an ad on general.

B

Pattern which will also see what the batching, where you have one non-virtual public API like create command, which delegates to a protected virtual one, which the providers in it implement and that allows and provider implementations to you both override the protected virtual one. And you do one new begin. So you do a new create command which hides the the original one.

E

You know so the one that arise returns to provide a specific transaction object instead of returning the new transaction.

D

Pattern instead of I guess so so begin transaction ends up calling begin deep, Insurance exactly in.

B

Updates non-virtual, so that's the none'll rideable one and it allows you to code directly against the provider and half the provider specific classes rather than having to cast down and missile one. So.

D

You're, adding a bunch of new things as public virtual is that inconsistent with the rest of your model.

F

Is it again more about having a protective.

G

Virtual public error.

B

Checking so if I remember correctly, that doesn't apply for those api's pattern, but maybe we should look at it. I.

E

Think would apply because it returns a transaction and the toy you want to return as equal transaction or anything Civitas. Basically,.

A

Yes, it were my kids like what I'm wondering is. Do you need to begin DB transaction.

B

A single role, so the beginning of the transaction, is that's actually one of the good cases where beginning a transaction is certainly gonna, be nice and clean you're, asking in general about whether.

A

Or not because you basically add so, it seems like there is. It said the pattern is begin transaction with. Overall desolation level is a non-virtual on DB connection and then the one that is virtually is begin, DB transaction and it's protect. So it seems like yeah exactly it seems like no you're adding public methods that are virtual yeah. It's breaking the family right, but it seems just odd with the way that the existing one is fracturing. You're.

B

Right so following the pattern, we would actually have to add two methods here again written like a a non-virtual begin transaction, a sink and a virtual protected begin. Eb transaction music, yeah I'm, not really sure.

A

Like all the pattern is used, I guess it's on the derived clubs that they can call them about. Basically, maybe doing a bunch of casting right.

D

A

Clothes again, no reason.

D

For the pattern is sort of error, checking is not allowed to be like this, or it doesn't have to be able or more particularly, doesn't fail to get me implemented by yeah. They doing I mean. Is there error? Checking like these don't take isolation.

B

D

Don't remember there, it makes sense I, don't.

B

Remember, they're actually being on your checking in the base nice implementation of anything here. These are completely empty. Delegating methods, if I remember correctly, but it does again allow. So, if you're coding against the specific provider classes, you can be calling the so not be overridden, but the new methods and you're interacting with those types, that's I, think that's the the value of, and it's true that this API is not following down looking.

D

At your existing and presumably your begin transaction async that doesn't take the isolation level, wouldn't be non virtual, it would add, and then begin transaction async that fix the isolation level would also be non virtual. But then you would have began debe tener that can async. That does take asset, which.

B

Would be protected function and then.

D

I guess the question is at this point: would you make? Would you collapse the to non-virtual publics into one public with two default parameters, yeah by selection level being default and the cancellation because it looks like here.

G

Previous was just the I want the default isolation.

G

Hectic virtual would be non-default practice. The public virtual would have default parameters, decided public, nonverbal or.

E

Demeanor is it compelling to add a default isolation level parameter if the same version of API does metallic and oh.

D

You know somebody could go from the sink to the a sink with the same number of commas and parentheses. If you do default default, the isolation level, cancellation.

E

Yeah I'm fine with that I'm, just wondering whether it MIT I mean I, see reasons to just leave it as the sink API looks like and just use it all. Instead of different.

A

Little methods but I mean I, would Polly generally like stick to the existing pattern because it makes it less confusing. Otherwise, you look at this and, like me, why does is have two methods? Is it all.

E

Like in the dogs.

A

Places where there's much knowledge you don't need to care much for.

D

The begin transaction dates thanks for turning past me, we had a long discussion about task forces value tasks. How did you decide on tasks of DB, transactional value tasks.

B

So for that particular method, the idea was that, as a general rule, this is always going to invite the Phoenicians do a round trip and natural database round trip, and it's very unlikely to benefit from an I value task source, at least as advanced things stand today. That was the general reason, so it's pretty much always going to complete asynchronously. Yes,.

D

And it's that true or.

D

For an in-memory provider that would seem to not be the case. Yeah.

B

I mean I, don't know of an in-memory provider. That's actually gonna. Are you talking about something like sequel right sure I mean it's very hard to have transaction? Well, no very in memory right, yeah, but sequel light. Doesn't Amanda I mean super light, actually has an in memory mode as well and also does transactionality, so you actually do have that it, so it doesn't guarantee when you crash that it makes insane I think it does, but I'm not so I'm in memory. If you crash, then it's gone.

B

What if this is wrong, but it does I think it does guarantee on this. Don't because.

A

I mean if you but.

B

A

Definition then, if you actually rash, you have to have it written to disco mall that you get something. That's the like that kind of implies. They.

B

Can be cool a memory I mean they could be doing some sort of right ahead. Log which does it and.

D

Can also you can imagine, there are transaction systems that don't provide all aspects of acid, but you know if, when the process and the data goes away anyway, so anyway, they could use your spree in merit, so it just provides for consistency within memory access and it.

A

Would be worth just thinking.

D

About whether there are with.

A

Virtual listener.

C

Who comes even.

D

More so if we belief or like sequel or something I totally understand the assessment, but if there are places where it would be possible to optimize, this further might be worth re-evaluating up to you. Yeah.

E

Secret life, it's interesting because it doesn't cover using API so because.

B

It doesn't give maps to not nobody, analysis, yeah, but I, wonder so to be completely honest, I mean I'm a bit surprised, but my own driver actually does an optimization where it defers the the actual setting of the begin transaction to the server uttanasana come on until you send the actual next command, so it batches them together.

B

That's a bit embarrassing that you know.

E

B

Means you'll need a simple intersection at all: I, don't I don't need it at all. I never need it, but it.

D

Would be an example.

B

D

You, but it is an example.

B

Of an optimization, which means that for nuisance.

E

If you think that when I hear you saying basically we're running that this planet does.

D

C

D

Separate value tasks of value task of tea- yes, yes, Elsie, and for value task of tea for virtuals when in doubt, if there's any chance yeah, if you don't completely understand your entire derive Manmatha l--, then value task of tea is probably a better option, because someone may actually be faster than you expect. Okay, even applying aspect of the usability problems of value test well,.

G

D

Then it gets into the this is why it's hard to come up with a match with the qualta, why I say in general deferred tasks, because performance aside task is better.

D

When performance comes into play, value tasks can be better or value test achieve a better variety situations, but its worst from a usability perspective, and so then you have to factor in how does? How is the API actually going to be used if it's an API that it's likely that someone, if it's an API, where the only thing people are ever going to do, is a wait? Just wait, X no composing. Yet then, then it makes value tasks of T on a virtual ism, pretty much better assistant. If somebody's gonna compose over a versus well.

B

So for something.

D

Like begin trans transaction async, like it's pretty likely that someone's doing you know a weight, begin transaction dou, X, dou, Y, dou Z, stop it's less likely that they're gonna store it into a they're gonna. Do lazy, initialization of some data, storing the resulting task into a data structure that then lots of people didn't go access? It doesn't really apply to that. Such.

C

D

G

Kick off three transactions in the same time and wait I mean like.

D

It's just less likely and if those situations occurred you could say as task yeah, but for there are other api's where you it is likely that you're gonna kick off three of them, like downloading data, promote HP, client or something yeah.

C

You're very likely.

D

To download multiple things at the same time, yeah you're very likely to durian.

C

Say yeah you want.

D

To get at startup, you want to download some data store it somewhere for everyone to access, so you're gonna store the task of string in a collection and, in fact, so, for those value tasks, probably isn't as good and so from a usability perspective, because there are, there are all these limitations on what you can do with a value task, and it doesn't always tell you if you're doing a phone in the best situation best case. Is you get an error? Your worst case? Is you get corruption? Yes,.

C

D

Yes, so from a usability perspective, you always want to err on the side of task or performance perspective. Value task of T is generally, but not always a better answer, and so our guidance is basically. This performance is always sort of secondary to usability and correctness or fir tasks, but but.

B

It's true that we're also dealing with a pretty low level thing, that's more likely to be used by layers rather than end-users. I.

D

Wish I wish my answer could just be always do yeah, and if I have to choose an always to claw, it's gonna be tasks. How many times have you had this feature area, I'm kind of curious twice this week, I've been wracking, my head about whether I could come up with like a always low, jerks and I can definitely come up with a flow chart. Maybe I should write that down it.

D

Just it bothers me that I I have all these if else's and maybe I'll be kind of beaten into submission and I'll just come up with a okay. So, firstly, third task of T always use value task of T in there and done since.

E

I think that a big eyes close in general, they are not thread safe. So when, when you try to do three things on the connection on the same connection, exactly you're doing it wrong. So maybe what it does is how pretty a good thing, because and if you don't try, it's.

B

E

B

Even that I mean so when you say at the same time, do you mean concurrently yeah? So you start.

E

Them without their waiting and then you try to a little and all of the time or something like imposing and even with something like more.

D

E

Okay, yeah I saw you could.

D

Go in the game, I.

E

D

About like Mars, with sequel.

E

Well, Mars allows a interleaving operations, but not really concurrency yeah, so you can start execute, execute the query and then exhibit inequity and then start calling read on. You should know.

D

That I could say, execute execute, execute await all right. No.

E

You execute and your weight on the execute you execute under weight and execute, but then you get these faults and then you start reading and that's what Mars made that's it and you can interleave those weeds. But you know the things that you executed, but that for.

B

Performance reasons: there's a bit of a shift towards a different database government model which does multiplexing, which allows you to do exactly that, which would allow me to use the same pipe to send a lot of queries down into the pipe and pipeline them together. I mean.

D

With with value tasks, you're never locked into it, you can always call that as tasks and get back a task, it's just it's a usability.

D

And ironically, if someone did the super-advanced thing and returned to value tasks that wrapped in I value test source calling as tasks will actually be more expensive.

B

D

As the allocated separate tasks anyhow I digress.

B

Sure, no, but anyway, that makes sense to me actually to have a value, tester, I'm, not sure or anything.

D

But the rest of them significant, you scroll down once anymore I'm, going to take the whole.

B

We slide by the section they are quiet, but then there is the one thing that emo we were discussing a little bit, so they D be beta. Reader has a close, async next to dispose async and they both return value to us, because they're largely identical expected to be. This is mirroring the sync API for consistency, and we don't know if some provider may have implemented some nuance between disposing close or the ones are going to be B data reader.

B

You can't reopen it or we get anything with it, you get it and then you dispose it, and it's done we chose to have valued asked their just dispose have to be virtual. In that case, like does it just delegate.

D

To close I think.

B

Right now, it's I mean in the sink API both have to be implemented, if I'm not mistaken, but maybe we should.

A

Confirm this, when you don't have an open right, so there's no open, there's no way, I mean in principle, I, don't even think you should make them both work for.

B

Should be somewhat one-way, delegate, the other the others like yeah Justin, it is so that's not how to sync. Api looks like that at this point. No, the question.

C

Is English? You want.

B

To stick to the sink due to having the same thing happening on the same side of things and.

A

It is a general pollen in the VCL. That slows very often is the same as this pulse, but not always, if you look at the guidelines, this basis said use the verb that is appropriate for the stuff because they couldn't agree that panels are being disposed, not close, but basically there's the other distinction, which is well things you can close. Sometimes you can reopen, which means brush the.

C

Knee because yeah.

A

I mean you disposed you're done right doc! That's yes, but that see.

D

Data reader has virtual close and personal disposable.

E

You know so and I think a derived type code and I think it's. The implementation on that one usually throws can throw from close that it eats all the exceptions from the voice. I, don't know why, as.

G

E

G

D

Poses a lot of exceptions finalizes, you know.

G

B

That's a question like I've seen, yeah I've seen that in sample.

E

Why they feel like they? So it's an unwritten thing.

B

That's not standardized in any way, but I know we had this discussion, but whether we should be throwing from dispose or not. Okay, we can take. You know the.

D

Guidelines say it's fine. You.

A

Only see, maybe in the gunman's to is double this pose is fine and some people just lazy, and they do it, but I just need all except.

B

Okay, NDB connection by the way is the other case where you can open and close at many many times and dispose it once at the end, which is why closed async in that case has returns a task versus. Of course, this poses a sink which obviously returns a value offense.

E

That's the same, the same reasoning that we just did on the apply to one of those on.

B

D

Data reader, do you expect.

B

E

D

Accumulate the pattern.

B

With the beauty.

D

But it has been.

B

Set up fast because well in this case, there's no one team, there's.

E

B

80%, do it, but.

E

It means what just yeah it's.

D

Do you expect anyone on DB data reader to to call clothes, as opposed to just using.

D

If you expect that someone who's using your async API would just use to dispose async pattern, then then maybe it would be reasonable to allow for your divergence to say: do you have clothes as in so it returns tasks and it returns to suppose they sink that as tasks there.

B

Is one slight small difference which is? There are two properties on DB daito-ryu according to the documentation which are valid after closing, but not after disposing, which is Records affected by the command that you are having the reader on? So in theory? Yes, there's there's a difference there yeah, but what's the benefit.

D

B

D

People are only gonna call call this pose async than having to close their safe return, dispose async, I'd, ask ask, and you never get any colors would mean you've mirror API in now. You don't have to and you're like well this one is this pose or value task, because the interface says so and this one is passed because the guidelines say so and guidances don't call close call. This pose I'm.

A

Gonna do find surprising, is more people Co close than their bows, which means very confident so.

B

You would you would advocate returning fast from there as long.

C

B

Long as we have to wait guys.

C

Why not use that just orange.

A

Gate like I can't Devine Tibbets been using and with the combing. It look good wouldn't with this show up and using it.

B

Because the second eyelets is you just detect the fits both call right, yeah in theory, it's slightly more efficient. If you have.

C

A losing better.

B

Transit passenger, so users can use that. Then they don't go through the overhead with a value bathroom.

D

You like to compile everything like generator cope. What a classical it's a disposable I! Don't see that I mean I, don't.

B

Make that as actually I.

A

Wouldn't want increase.

B

The ref count of clothes I mean I, wouldn't.

A

Sorry I mean providing implementation. The ILR ability to implement.

D

Officials, by having value fast, niceness, similar.

B

D

If they soldier DB reader or is a calling, you disposed I'm, not disposable. It's the right answer. I.

A

Think that I can suggest from this chair I think it's.

D

G

A

The method as public I think it would just fall through that's the thing if they explicitly implemented and then his war for the interface I mean otherwise. It would basically have a probably a zero uses very good point, and if you look at I, don't know we just look at the derived tribes for a second. Oh I can see.

G

Doneita see the base.

F

Okay, I need to call suppress finalize this.

A

Whatever we decided, we say that is an alias. We just call to make it non virtual. What do we say? We have them both virtual and yeah. He really wanted I.

B

Think we should be keeping them virtual. Even we don't know what providers may have done in the sink side of things and if M distinction has been introduced on the same setting should allow them to happen on the ice inside and what is that? You cannot limitation, go all this pole dancing or go food I think if there's nothing but I would have to be different.

E

When touching our, when we are suppose a single person right, yeah.

B

Oh so sorry, yes, I was talking animals.

A

Of all clothes not to spa, they say so yes, so that if somebody has different logic, it's you know consistent.

G

A

Yes, all right that just might.

D

Be worth checking here the logic we'll just step aside, the tracking.

A

Yeah I'm uncle issue: it depends on how CCI represents that, to which there's sometimes different, yeah yeah. It seems interesting that I don't see that.

F

Actually, to the point that I just mentioned, what would the correct you know see us if they did overall hey what way so, like I had mentioned that the visa stage, however long the tight as it currently, is this? um What what does it look like correctly if someone tried to do this opposing.

D

Basically, from the pattern perspective to expose a stink should functionally do the same thing as disposed true.

D

So a check all suppress finalize assuming there was a potential for finalizar.

C

But it's also not technically, above my all, struck by lies, because the contract states that multiple calls to dispose should be nothing so finalize, we'll just call it it's a performance, kids, yeah.

F

But I think it's worth clarify yeah different bug, bug yeah yeah.

B

Yeah yeah, it's a.

F

It's the ultimate. It's.

B

Not correctly, that's what is there any plans to kind of extend the disposable pattern to take to take that into account the async it's.

D

Logically, the same logically, the same as just it does factor into the pattern. It's just. You were expected to call this post true from you're expected to have functionality equivalent to supposed true okay.

D

If you go to the same state, yeah.

G

Possibly be a synthetic yes.

D

A

Do we say that the type can only implement, and you can't let both in fact I would expect motion.

D

A

Can you only improve the a single version, not the same version no I mean and then ever finalizer, so that seems kind of I wouldn't be broken right with.

D

You, our guidance, is, if you have a finalizer a, is a safe handle instead and b be disposable when.

G

I dispose women I using delusional if.

D

You have a finalized er, that's going to always be a synchronous thing. So now you should be synchronous disposable to solvent if you can benefit from AC dispose and also delays. Engine I mean yes and know that a lot of times you'll find finalized errs that do like they're, often to things that need to be done in disposed.

D

Water, which is only done in a finalized earth-like dispose, will both flush and close the handle, whereas the finalizer will simply close the handle and I I, don't think, there's anything under mentally wrong with having a sink disposable and a finalizar and no I disposable, but I think it's going to be super.

C

D

I would I would definitely raise it as a question of any new eight guys that showed up I. Think.

A

That seems fundamentally something good yeah I'd like to see a kids where they will be seen. One.

D

Of these days, but I think I did spend I suppose makes it work. You're gonna get the racing. The.

A

Only thing we should avoid is like having like this, both boolean and dispose a singular. Then yeah.

D

There's no reason to have there's.

A

D

To have disposed async pool the only reason that exists for dispose is so that the finalizar can share the same implementation and call just call it with false. Instead of arrow, all right. There's.

B

D

You should never have a finalized or call it dispose a sink. You.

C

Really shouldn't have fun lies in the first place, because well.

D

That's your yes, yes, guidance! Number one is: if you're writing a finalizer stop and do something else. You know if you think you still need it. Ci disposable.

A

Right so then, any concerns of the API service performance.

A

Honey only 40 minutes in yeah.

B

And that's the small issue well.

F

There's just the you know the.

D

General for each item: they consider you want tasks or value paths and they're sneaks happy to answer mail, but it doesn't really change.

F

Well, what was the decision earlier on whether they should be long virtuals delegating to virtual it.

B

Will change the ignited about yes to align with the synchronicity, so I know that that I know that changing begin transaction to return value tasks as well and possibly close a sink on TV data reader to return tasks but I'm, not sure about.

A

A

And let me just hope you pass the the feedback items in minute has to feel like the bestest. Mostly asking questions could be two key things. You said it said the first one is the one that you might just put up in the second one. Is they consider whether you one way to task with tierracast 15 but I? Think that was pretty much.

B

So just one question I mean the work we're approving now for that matter or three, and is it still possible in any way to get stuff into it on its Android or if so I mean every I? Will.

A

Go of everything that was API improving candidate for Stan, and we just submit you know for it. Okay, there's not that much! It's like, thankfully, she's I.

B

Don't feel like that, all right, so there's another yeah, there's another ratio we already approved and merchant to done it for which we were you know, may want to put in yeah I wanted to do this like sometimes this we could just do the okay.

A

And it was funny they think so that I will check yeah, we'll check with that half that I give them this ring over. That's right, we're talking. Should we focus on this one first, because it's the short one, that's super short.

B

B

Basically, there's a small feature: detection, API or rather pattern that allows you also to film. Not all providers implement everything if you call a method that is not supported and you will get either an exception or no. The idea here is to simply add to feature detection properties, the DB provider factor a factory that tell you whether that provider can create a data adapter or not.

B

We have at least one adapter at one provider that doesn't support the adapters and command builders and we'd like to expose the ability to know that it's more of a completeness.

A

Did you look over the other ones? Is this like their sense to add the other ones as well right away? They are there if there's.

B

Only one right now, there's only can create it, so the numerator right now yeah, because the other ones are mandatory and they do that. It doesn't make sense to have the provider with no connection. For example, he started.

E

When we need a running car of 1.0 and for the first time here was one provider that he has of all data said because because it wasn't there and now utilize still exists and it still doesn't relate. Sir, yes, cuz.

A

Which a David Trent fault, the true then, who was supposed to write when it's supported I mean the lethal imitation returned, true I, suppose right I think it should return for. So we.

E

Wind cuts this but I, don't know what is it kind of struggle.

D

So, let's you call this instead of calling the method and saying the throws.

F

Exactly bullet weight, that's something the problem is, that would be a breaking, because these methods find if the default behavior would have to be the dissolve return false, which means that if someone has extended the class and.

G

F

Default has to be I.

A

Think, because if the current implementation is your Kern expectation is you can always create data adapter and always return something understand you.

E

Probably want to return the.

A

Default to be true, and then say, only new idea of things are.

E

A

E

We had this question I think that we came to the same conclusion, but also the different implementation of the create adapted method. I think throws it actually returns null. If it's not okay, do.

D

We are we going to consume these methods in area like at these purely for consumers,.

E

Of our libraries you could be actually using so is Ronis. Well, these providers, the providers that don't support them. They are going to categorize them.

D

Do we have any? Are we adding any call sites for these properties not as far as I know, they.

B

A

No, no, no! You.

B

Are world right, but they're, mostly in the warm brother and or am whatever so that, therefore layers that are using a deodorant generically, which is already a pretty rare occurrence, only or arms and various other layers to check whether it exists.

A

I mean it kind of makes sense. You want to check them about having to refer to the moment where you actually need that right. You may want to check this earlier on hide the stack and just all right confuse the provider, because what pop-up or if so, there's you're like no I, don't support that I need something that actually can create. I, don't know data deficit, I.

D

Guess you know, suggest you're going to just try calling creating thing if it returns a non-null non-throw. Yes,.

B

Based on the fact, is you overrode this they? Actually, this is the default implementation that we discussed not is to actually call it and see if it returns now what you're working yeah, no right, yeah, yeah, ouch! Sorry, that's not expensive at all. I.

A

D

Assuming you catch it, yeah.

A

You better catch it: yes, yes, better I'm attached it costed, yes, I think that makes an ID for that, and that means I'm. Okay with anybody else.

D

All the other kin methods that are already there or abstract only one and it's watch those water show.

B

Motech National results in the implementation.

A

All right so then we have whole 15 minutes for the big people.

B

Yeah I don't quite know how to go about this kind of issue. There's a lot of API surface here, the general problem here. If you want to really go through it, is that right now I do that it doesn't have a first-class matching API. So you can't execute multiple sequel, safe statements in one round round trip what has evolved. It has has been what you see there: a concatenation based approach where you just stick the multiple statements into a command. There are various problems with that.

B

Some providers, just don't work that way on the wire protocol level, so Daniel split protocol messages forcing the driver to do string, parsing of sequel on the client side, which is bad for perf and also for correctness.

B

It's hard to well to do this yeah, even the sequel, server implementation, which does support this on the protocol level is inefficient because it sends to that as a single statement and does plan cache pollution yeah, whereas if you broke it down, which is also something that's supposed to be possible, you would have a more efficient implementation of batching, but why.

A

Would you vacuum in this particular example? You have here, isn't the general pattern, you just do a parameter and do a loop like. Why would you if your person well, because you don't want to send them and it's individuals and.

B

Basically, single batch of the server you want around it. Basically, the idea here is just to do. One mantra of core is a very intensive consumer of batching at the moment. So when you do a lot of updates and you to Save Changes at the end and get batches, everything and I know it uses this. This API.

A

I see, can you outline your general strategy for batching? That is the idea that you basic read about you. Yeah basically add commands to it or whatever exactly and then how do you have the data and of command like parameterize, don't parameterize so.

B

The troll, the way the API has currently proposed, is that you have something called a DB batch, which is a container for an I list of DB batch commands, and each batch command has a sequel and a parameter collection. What's basically a container of a lot of sequel and parameter sauce. This is the the current proposal and then the execution, the execute methods are on the batch itself and not on individual commands. Invasive.

D

Are you going to have any providers in your initial implementation that do not that return something other than list of T for the property you have called? It is probably not aah like I mean one collections, guidances users will cast it. If you don't have anything in the initial implementation that uses the.

G

Site showing you type use.

D

The concrete type because you're stuck with it anyway.

G

D

As well be virtualized recalls, why.

B

Do you need a new bench come on? Why can't you do in this big list in DB that that's the way the proposal actually started out one of the problems, so one problem is that EB command has some properties which would be dead when used within a batch like timeout, which which doesn't make sense on a batch command within a batch.

B

It's something that applies to the batch as a whole, if also some other properties which are dead if I remember correctly, another problem is the DB command is disposable and that creates various lifetime lifecycle issues. So if you dispose a DB command as part of a batch, how does that work with the batch? Is the batch itself disposable? So all these questions we kind of went back and forth on them and find and kind of decided that we prefer to avoid them. Does it make sense for the thing inside the batch to be disposable?

B

There's there's a lot of reasoning about exactly this question below, so the command is.

E

Heavy we wanted to have something yeah.

D

The default implementation, other sodapoppin rotation- do we need.

D

B

A can batch, yes, okay, exactly that's that! There's a feature detection again, one once again for this on connection on the V connection and DB provider factory, that's further down summer.

B

Is a good day we also you talked about a default implementation. We also went back and forth a lot about having a shim or not implementing this API over. Let's say concatenation based matching so that you can use this API or sequel services. We decided against it for various reasons, and yet we don't want us, so we don't want to simulate batching without there actually being batching yeah and there's also a lot of edge cases which make it kind of impractical to wrap concatenation batching.

B

With this API, which providers will support batching I mean I, would that we own that we own yeah, so mpg and then PG sequel is not owned, but I would be implementing that we can.

C

B

It's owned by me: Microsoft, then sequel, a sequel, client we'd, really love to see it, but unfortunately, at this point I'm not sure when that's going to happen, and it might imply a pretty big rewrite of the internals, we're not sure it was hoping to do maybe a sort of an exploratory prototype. This.

D

I mean you can.

B

Rephrase my question, but one.

D

B

D

That we've always faced when adding new abstractions is having very few implementers of those abstractions and, as a result, getting it wrong. I see yeah. So how is it that were validating we're getting it right? We.

E

We want to show you this API and to get your feedback and developer. We are actually going to decide later whether we actually try to put it into SEO based on the same catalog. We want to cut more than one advantage, I think the.

A

One thing we should make your choice, which is every time we originally when we had the whole DB common conversation with the secret team. Might they felt very strongly that this should be involved with the design of it, because gonna make sure they can actually implement that right? So, even if they don't have resources to implement it, we should at least make sure they have seen it in they're saying all that will work for us.

A

I take the one thing that should be: what is that the approval, and we added and then later on secret, says: oh I, can't oh I don't want to implement it like this right and then we're just a group. So.

B

We have two other provider, writers, commenting and interacting on this, so the Mexico provider writer is kind of involved in this, and one of the guys was actually submitting all the sequel time for requests. Rave to write. Also kind of these are people who are involved. We at least get some sort of triangulation on it, but I completely agree that I mean the idea. Is that there's not a sender pull request for this tomorrow? In any case, yeah, there is also a.

E

Secret I, don't really does implement batching for data set and data table it doesn't it went all the functionality. Essentially, you can execute a command at resonance, just rows affected, but you cannot execute the command returns results right, like a data reader using that API. It's only that that API is completely internal. Is you only get.

B

To it, if you pass it into table- and we know that an hibernate, for example, user reflection, time, susceptor cuz, it's efficient and that's exactly that's one of the points behind us as proposal is a surfaced and I think we have been.

E

Talking also with people that have like the original try out knowledge of still applying their opinion is that it can be implemented. Then we have all I'm saying is like here. You know we don't need an official like email from them, but I get these. You know cube or should have seen.

A

E

Kind of support the design we are on the same page, but we wanted to let the API. Do you see.

A

What it's so, the one thing I do by with the DB. That recommend is that you want something that it's not a DB man, because this is more specialized to do the one thing we should make sure those that we don't think we might change our mind later, because if we basically end up in the water we add more and more stuff to DB batch command, then you just have duplication and then it's like whoa. No, if somebody wants to add normal Messer, we have to remember to add.

E

It in two places and then we're pretty confident that that we're not going back we've had that in the original design and even one more batch command crafts records affected, the energy c'mon doesn't have it yeah and we need people so.

A

What happens in you know like women, the curious service, not manipulation for you, but a select statement the basically get effectively. The multiple result sets on the on the badge.

B

Itself in general, you get the results on the match itself. We have single DB data reader. So if you have, you can battle five selects in The Dispatch and you get back at a single DB data reader with multiple result sets, which is how concatenation matching or any words today or sometimes you can call a short procedure. That's going to return five result sets. So it's consistent with the way gig that currently works. So basically.

A

Logically, everything that is on EB command right now. That would basically give you a reader would effectively.

B

Be on on that sheet, so exactly so all these methods, you see mirroring DB command. All the executors execute non query. Executes scalar are copy-paste s--, almost right.

B

It's it's intentionally supposed to merrily, become and, and there's even a note that we could. We could even have some sort of interface to capture the similarities. The 80s are very similar, they're, both executable. You know things.

A

C

A

Me very convincing that you will never wish to turn a batch command into a program, because you will always would like to add them on the bank side of it.

B

Yeah, we ran into a lot of trickiness when we try to use DB command inside a batch, and then the same type can be used inside of action outside of a tree, ran into difficulties with the API, with clarity with life cycle and dispose also. So what this also proposes now that you are executing a batch, an exception can be raised and we want to allow users to know which of the batch commands generated it. We're proposing adding a property referencing.

B

The DB batch commands from DB exception and again that raise a lot of lifetime lifecycle issues. If you disposed the command but reference it from the exception, then that's a world of you know complicated 310,.

D

B

All four commands each encourage an error.

B

We, so that's not something that would be supported on the generic API level, just like DB exception nowadays doesn't have a concept of that, but sequel exception does have a concept of that. So that would be something I think that would be extended in this in the specific provider. Did you consider making the badge itself derived from DB coming derive from it mean to be bad commander? The TV batch TV badge yeah.

D

Upstairs TV, oh I, see right, you think all the things we can do on a command you I do.

A

All of that aside, it looks like it it's kind of almost on a person over that way.

B

Naturally, I'm not sure I found out.

A

The neck, is you just get everything.

B

For free right, and then everybody just extends that guy, so some okay, so it's one one example is that executes reader does not accept a command behavior, which is something you'll find on VB command, because we allow a different command behavior to be specified for each batch command. Inside of that right, so that's an API like you know, that's a discrepancy.

A

I mean this is something that you can. You cannot have to walk through that in your head. I guess, like I, don't have a strong opinion. It's just that I would generally say if there's only one or two exceptions. Maybe it's worthwhile doing that, because then anybody can treated batches of command, which might actually be beneficial. I was it was not the case, then don't but.

B

I get if it snows.

A

And I 100 percent.

B

I would just I mean I felt. The same way and kind of proposed may be having an interface to capture this, but then, if it's, if it's practical to simply have a superclass instead of me, I mean it's even better. Maybe if there's a reason to keep this place.

D

And you're not gonna, go with the interface then consider making time out the time span instead, event.

G

B

C

Always are we doing the dykes, pen, yeah.

B

I mean, but then so that would be really like encouraging the lady be commended to be back even further, so people are used to be become and with that with an int I mean I, don't know, I might be a bit of worse. Didn't use all the time machine that we need to.

D

I mean it's the problem, if you make when you're, making a new thing and you make it, how much do you make it look like the old thing and how much do you make it look like.

G

Yes, dude, if Lincoln really do like into.

D

Timeout just jumped out at me cuz it's in the middle of this game, I.

A

Would say within a particular area like we try to stay within the area as much as possible to not have multiple patterns at once, because then you over time like it just gets really complicated to reason about the API but yeah in a brand new area. We should definitely, you know, choose the more appropriate patterns, but here I don't know, timeout is everywhere, and eighty over an engine is always it in. So we should.

E

A

Because it's inspired by a bee, oh yeah, so Jeremy. What are you thoughts on a list of the magical man? The right thing at the top of the screen right so.

D

Yeah Christophe pulling up before when I brought in a thing that had a property that was returning a and I some form of collection is whatever gets returned in the first version. Someone is going to hard cast it and then, as a breaking change, we will never be willing to change what it is. So if you're always gonna, have it be a list of T, get rid of the eye and be honest, but.

B

This is an abstraction over several providers. There is that aspect of it yeah.

D

If none of them in the Box return anything other than list of T users will hard cast it because users love our tastic things really.

A

D

Time I mean usually.

A

It's not necessarily at the direct call, so they're just passed. The on is minus 15, some cool passages pass things and then you never sort.

D

Of something to have a list, someone wrote an extension method, then they wrote an over list of TN.

B

D

List of T, and then they cast this thing, and it did some stuff like I mean their code has worked now for five years and then you great your provider in their code. If we want this to be as I list of the easy answer is from the implementations in this that we own don't return the most yeah. Well, I meant the.

C

Other ones just return.

A

D

Thing that might be cast honesty.

A

In your case, it might actually be rough about. To just add the you know: it's like five lines of code to do the. What does the DB batch collection or whatever similar to the DB primal.

B

Collection like.

A

A natural because.

B

A

Because if you ever, if you ever want to add a message to this guy right, convenience, add overload or something right that just directly takes the string, for example, then just be attached for you, it would I need you screw of an interface, so the parameter list as it follows that thing. So, there's a proper collection type, hey fzd was a DB parameter collection or something when you would have your DB patch selection, I. Think what is the TV video.

B

C

Our collection, I think.

A

It extends um nothing actually, it's virtually just implementing everything from scratch. I would just probably extend collection of whatever my collection of DB batch. Then you can just add your own methods on top, but these collection types are designed to give you interceptors right. If somebody adds something you want to add something: okay, I can override and there's a.

F

Future proofing yeah my nobody's collection of.

E

Yeah, we're gonna call it now that just takes it.

E

Can you see I got that come on Texas these and.

B

Oh yeah, like a nicer and necessary gonna: okay.

B

D

General, its and I think as a brand or when you're going to read from it and if it's a property limits it online you've picked a concrete type in your implementation. Just be honest about it or have like this. If you instantiate at this guy it'll be less than if you instantiate at this guy it'll, be you know some others planted and then at least there's the book. You have two options in the Box one of them. You could park past you and couldn't, and we told you it as an interface I mean.

A

If you really care about versioning, one thing you can also do is just have an internal type. The people can don't cause yeah. Just do your own Eliza statistics and.

C

With certainty- and they just can't organize those two together, they can sit passively you, but.

A

You basically have to provide your own reference, your, but probably better off just having a public type, yeah yeah.

D

You could have yeah internal, you know something list extends list of T like yeah.

B

I guess if it extends list of to you know, but that's not.

G

You also allows you to add.

B

Stuff in the future as well- yes, it's not it's almost like both problems and not just or maybe.

A

Yet this of array right very basic, trying to be smart with the innumerable dog get empty and we basically said well, we always allocate the enumerate or so how about? We give you a specialized handwritten version of a new Google that doesn't do that wrong. Everybody, because me, when we returned an empty array and people inadvertently took a dependency on the fact it's an array because they see all and output and the custom still does it for the customer.

A

I knew more voluminous realizable, and so it's like well like this happens all the time with things like it's, it's not so much that the perv that the person calls the API nomini there's a downtown sir. You are usually not that dumb, but it happens, but.

C

It does have a larger system.

A

Because you pass stuff around and then somebody else's anyway, so they yeah modulo the feedback that we had on the you know, reviewed of the other people make sure you have more fun invitation. That seems reasonably. Nothing stands out as.

B

As an issue, there's a whole lot of text below on how this behaves but and how this is going to be used. But if you want to see how it's going to be used scroll a bit more down, because that's there's the horrible example and a good example, that's the very thing of using it directly rather than through the generic again.

A

The one thing I never was clear an idea of my luck. Is there any benefit of me calling prepare or is it something you do automatically it's a very provider-specific.

B

And then PG sequel, it's super important clipper on sequel server, it's an Noah, almost automotive anymore, but that's a very, very provider-specific, but but it is important in databases in general. Yes,.

A

And then I assume you can't preparing the bench and they're just basically psychologically prepare on all of the exams. They.

D

Both of these were selects in this example, rather than one updated, one select mm-hmm.

B

How do I read the results of differentiators, so DB data reader, which has returned from executed reader below actually already as an API, supports multiple results it. It has a next result, method missing, which has a different shape and they'll be in the same order. They.

A

See the other thing you say well, move next and then you fall to just say: you have an autumn that says once the.

D

First, one was call or exactly which other one see guys with maybe I'm close and the stored procedure itself returns. Seven results that sweetness without hate result sense. Yes,.

B

D

A thing that and threw a bunch of stuff at the streams and then did another day.

B

So, there's a bag from the community. It's important to mention because of this thing, because there are some variable number of results of cases wanted us to return a reader per batch command so that it could be easier to distinguish in boundaries between the commands which I thought was so the thing is that there are other solutions to do that. You can place icon marker. You can do a select one to know what it's the you can use the shape of there. If you know that the shape of the you know the result.

B

So it looks like this and you can use that. So no, there are other worker ants to to do this and it complicates the API for everybody else right and it's a pretty exotic use case in general, not.

D

Opaque as well, you just pushed all this stuff, but like sequel server, and it just gives you back like what gave me a thing that had seven commands. So there's a thing with in results rings. Yes, because some of them could be like the update is there's no.

G

Result that comes back but.

A

Even like, as you kind of know, know how close to a proc I know, what's.

B

A

With her to 22 or two dots and.

B

That always, is it possible either.

A

B

Server has a thing that they're in the know, an expert voice, but with a variable number I can.

E

Write a stalker sphere and then inside this parameter was passed and do a select and decide parameter was passed, select and then you don't know how many true.

A

But I mean I'm, saying the caller really needs to know what there's yes to do.

C

Regardless off.

E

You mean like, if I don't.

A

Know what I get for words: five like how am I going to write that goes whenever.

E

Some yeah, you didn't equate that put it forest, where it's the same personalized consuming them, but.

B

There could be something dish analogy: it's true on the client-side or you don't know, you're making proceed four or five, and then how are you going to distinguish between it?.

A

It seems like a column of the different results, edges or thought onto the matching, and if I just call this to our program, the same problem right I mean kind of idea. Yeah.

B

That's very easy.

D

So, in that sense, it's really yeah. It's not. We have the problem today as well yeah. What are the command streams like update mixed with the result? Streams like select sorry, say again, so when you do this and you.

B

Still find out the number of rows that were updated by the update ah good question so that the whole point there's another big section here. One of the specific goals here is to allow non aggregated records affected right now, so we want to know for each if you look at the effect of rows below so this is why we added on TV command batch Batman. Sorry, we added the records affected on the command itself, which.

C

Is going to be.

B

Populated by the driver and allows you to get command I command, another force effective, all right so.

D

You listen to execute the batch, then you look back at the Commandments.

B

D

Asked what was the rows affected and that gives.

B

Me a bit of matching.

D

In the results which.

B

Is not a perfect API, because the command is usually supposed to be an input you know object rather than an output. I think that it is now possible, but it is possible or.

C

Before it was not exactly.

B

That's another explicit goal of this: it goes like I mean if you really wanted to. You could just have a method on the batch that just gives you an array and.

A

We went through this a little.

B

Said I don't want to okay, you.

A

Know, hey I also think it's weird.

B

Because it's probably easier to just oh, let them on the command. We have this array of commands. Why.

A

Don't you made.

B

As well, just right put the property on them yeah. That was the logic no.

A

It seems to be too little all right anything else. Otherwise, it's just who that is, and then yeah yeah.

D

So I guess the biggest feedback is, if, if you think it makes sense in retrospect to make a big batch extend DB command, that would be a good consolidation of your API said.

F

Oh and also validate.

D

Against multiple ionizations yeah.

F

D

F

Very the question so I was looking for the uses. I mean it looks like you're, introducing the concept of Rose effective here that wasn't originally, and even it's.

B

Not on DB command, but it is in two other places in the API. It is the result set of execute non query and it's also on DB data reader. So right now, if you, if you do execute, let's say whatever and you give anything you get back a reader execute reader on it. You have a records affected property which is aggregated. So you have no idea. If you executed multiple things, you can't know which one that's exactly the problem and.

F

You expect that that Cal whole never exceed you're.

B

Asking about the type I mean.

F

It is a database, you know, I can I can imagine the database where, especially if you're operating on matches.

C

F

Tables just like I could imagine, maybe you just dropped everything. We're.

G

B

You think just go ahead and.

F

Make it long and I don't know I, don't know, probably why not, there's no constellation by one and I wonder if.

B

It was a question because you poaching the expertise, right I, don't think I've ever heard or Seema scenario like this, but yeah. It's.

D

Not to be you know the moment we make it in someone's going to write it and said this is not big enough for me to go back to the stored procedure. If you call the stored procedure in it and it does a thousand different update statements that I feel like in the command window, it could you know, like sequel, Explorer, you see all the individual numbers. Is there any way that you get them back and how are you aggregating the result stream on that one.

B

A very provider-specific question so I how the rows affected is gonna, be surfaced by wrong and I have no idea what the answer. Okay, simple server! You know with automatically activates everything that happened inside this or procedure. You know that or the left no way.

D

B

It's the because.

D

Theoretically, you get back, you said two statements and you get back. 11 update your values, they're like how do you know which ones were which so.

B

One Postgres you just don't get that information at all. For example, I, don't know what the one on sequels so does that look an eyesore model I mean it's yeah, I know which.

E

I think you're going to pretend that number of perfect and huge replication is going to yeah, never finish anyway. So I don't know for any of nothing me not big data right. It's in.

B

The relational you.

E

Are bringing stuff to the client, this IP I start bringing stuff to apply for changing something I thought.

B

E

The ticket now would.

B

Be like a little update, update Sagamore.

G

F

B

Look at lead everything from a huge table that.

F

Was it forever drop it baby? You need to know the number of rows you're, just wrong begin transaction to count drop.

D

C

B

Is not always allowed in transaction. So what valid point.

E

You seem to think because they is the arrogance, then the ones that returned the early truckers affected, that we should changed and we cannot change them. Yeah.

B

It's yet they are more critical here. I got iffy, that's actually very true.

G

A

Bye guys online.