.Net Foundation Design Reviews, 26 May 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review

Description

00:00:00 - Needs Work: RequireMethodImplToRemainInEffectAttribute https://github.com/dotnet/runtime/issues/35315#issuecomment-634163634
00:21:10 - Approved: Add activity Ids and Context to all logs https://github.com/dotnet/runtime/issues/34305#issuecomment-634181323
00:54:58 - Approved: Add a GetEncodings method to System.Text.EncodingProvider to support enumerating available character encodings https://github.com/dotnet/runtime/issues/25819#issuecomment-634199380
01:26:58 - Approved: StringSplitOptions.TrimEntries https://github.com/dotnet/runtime/issues/31038#issuecomment-634204949
01:37:59 - Approved: Obsolete RuntimeHelpers.OffsetToStringData https://github.com/dotnet/runtime/issues/31406#issuecomment-634212546
01:52:23 - Approved: expose System.Text.Encoding.Latin1 getter publicly https://github.com/dotnet/runtime/issues/31549#issuecomment-634217668

A

A

All right, good morning, friends, this is our I think I can post build stream, so maybe a few issues we're going to go off go over like it's, mostly things that are blocking over progress in other areas. So we start with the acquire method in pole to remain in effect, and you want to talk about it.

B

Yeah I guess like so basically.

B

Have this you called covariant return.

C

B

And have a Pass where you have earning some type and have a derived past. It is right for that method, but returns a type. That's derived from you know, type. Those returning Oh.

B

One example is fa, boss a bear if over a punk one survey of cars, we have override of the learning in terms of B for exam, and it it allows you asked in the when you knew. No, you are using the touch pass, and this attribute is a something that the compiler to mark such the Assads, that that override the base class methods, colonel and something that the users would I'll ER.

A

Maybe I miss the nuance, but like isn't that already legal in Nile I qui do we need an attribute.

B

This attribute allows contact.

B

Understand it, the do pride to remain.

D

B

E

B

B

And scroll down a little I have this: it's almost at the end, where you have this example, a little.

F

Another I mean.

B

C

B

You have the class a passkey Anthony, no pasty that overrides I thought in the virtual function. This reviewed each photo for the runtime problem and all Oh.

B

The virtual overriding, so the the virtual result is in master dispersion, would put the correct I thought.

A

A

They're not regular overnight.

B

G

B

Right, basically, there are not writes the a our new odds in the virtual table at all right methods, inter like in the base classes in and.

B

So that you can, you can do this if you PIL? If you apply this that override thing, then you can. You can override up I thought from the base class, so you can say that this methods, the method overrides a method, is class.

B

It has to be used for the core entry, because regular overrides allow just basic internal hiding in I same city, the same return type and same name, while this override.

B

Inge the return, return type, basically work for injured, right.

A

But and that's different from the normal like new and overwrites and medicines, we shall be, as you say, override or give me a new thought. But basically this thing I am occupying the same veto with slot as a guy over.

B

This is basically saying it's kind of / today, in the effect will place a new slot right from the ice class. So, basically, if you then invoke something using today's class and a method to start a bit, then be the override will be all even it's not.

E

Or why is the attribute.

B

A

The thing I'm trying to figure out as well yeah I just.

B

I have no good, in the run time actually yeah. That might be better for David to be here, because you give you the correct answer to buy. It was that about so. Basically, this.

E

Is even if it's making slides so that it can have a different return type seems like that's a public slide on a nun. They.

B

E

B

It some time and we did a virtual function and we could end up whole day somehow I think the resolving I thought from the. If you call.

B

Using the be using the b-class, then I mean, but if you call the virtual I look to the beat pass and without it at run time with all functional pass because right of the of the a and the distributors, the would apply the the other the be castle as well. So basically, a virtual function called you right in this. If you I mean your option.

A

Yes, so basically, what it says is like like, if you apply this attribute on a method that has an override specification, then it means applied to this one and any of the base pipes in the hierarchy that have an override on em, well, yeah,.

A

Interesting and a normal override is literally just as one thing.

B

I know this is implies them.

A

I mean it makes sense, I mean my question: I mean I think this is the generic one that we always asked like. Does it have to be an attribute or I? Guess the the method in pull flags or there's only so many of them don't want to put it there, but it seems like.

B

I, wasn't there and I the agreement between the charity and and the runtime dogs that were originally discussing that I try to be if there is any, but, like reasoning recorded anywhere I mean.

A

If I had to guess I, don't think this is charging cares one way or the other they just care about what they were supposed to emit so long. The semantics are the one thing: I think if we say that's a flag somewhere on an existing one, they also find that against it's more question for the runtime /tp format, it's self weather method, Impa Flags would work for that as well, but I guess burgeoning. That is always in the SS.

B

Let me see if I have I was they are defined by email just way. I can.

A

What is the semantics to you or the semantics of this thing, for if I now create a new type called D that extends B and also overrides virtual function, as the compiler then also apply? The attribute the override in Dior is, is the fact that C has this attribute applied that it's the only thing in the hierarchy you, it should be I I do.

A

It has to be applied by the compiler right, like there's.

C

No implicit inheritance.

B

I'ma 100-year about the house.

B

Trying to if the the discussion mentions that.

B

And I was joined the light so I I can't see the learning of all.

A

That yeah I mean like we can just collect the questions and then give it back. I may like to me the desired semantics, make sense: orell works I'm, not sure the name makes no sense to me, like it seems a bit abstract for my taste.

A

As say, is kind of like Reserve overrides of bathe, it close it based on Highlanders.

G

A

Other than that I think, like the namespace, makes sense, that's usually very put stuff that only the compiler is supposed to use.

A

We don't expect humans to actually reason about this name anyway.

A

Of course, people will see it, and you know this assembly view. We should at least make sure that the name on what sense to people.

B

Yeah I think the Amos I said I.

B

There is bad that I, don't know by was picked whether they were like some mothers, considerate or.

B

Because it ones when a party is left I did this thing like three weeks ago or something like that I I wasn't trying to. Why is.

A

It still effects, then that's just.

A

H

B

Little bit of in the access.

B

B

Yeah that this is to ensure that any virtual call to the battery base or derive signature of the method placing it was derived Oh. What we've! Basically as before,.

B

B

All right, so he wouldn't be done next step. Dad should I try to I, mean I, try to find a bad pun or so.

A

The way it normally works, as he just holds a bunch of questions you flip it back to API, needs work and then basically, what would be good if you would have somebody who's, basically picking up father's work and is able to you know, speak to the thing in more detail and, like they say, shepherded through the process. I think most of the stuff seems fine. I mean like again like it seems, like the name says, is right. Everything is an attribute seems likely the right choice.

A

I will still ask, but I think I know the answer to that already and yeah. I think again, the sous-chef compiler team poly won't care whatever the name is. If then, if this is the name that will encode that one, if we tell them it's a different name, then code that one so I think that's that's. Fine um that'd have to be an entry bid.

A

A

A

Anybody has any additional feedback, Jeremy.

E

I'm, basically, just still confused by it did sounds like how did it actually what I think the runtime ending fun time, behavior didn't, have it essentially, what's the it'll help? What's the? What is it changing about the run.

B

Well, yeah I mean fiction in this issue. I think that it tries to explain that step-by-step.

B

By a but I I don't want to like leave I guess you have more immersed outside I'll just buy you like.

B

You down by it by trying to fully understand the engine, but, but what's here I don't know, would it make sense? A mother I think that you have to review and I just like to read again and I can better some like more details.

A

I mean it would probably help to have that I think I think I have a hunch of how this is working, but it would probably still help to have more out there.

A

Jeremy said like the gentleman problem with these features is often that, depending on how they actually work on shape might work better than the other, and it's a bit hard to judge that without know exactly what it does describing.

E

That it's bearing returns, but everything there is a way to function, takes no parameters. I can't map concepts from one to the other.

C

Kid and then calling function, virtual function and using dot, override and I think it would be great if it was virtual and if it used pride like normal c-sharp syntax and if there's a case of where ICI virtual function, hiding a dot version. Currently press in today.

G

They would know it.

C

Would help explain like maybe use it interface, implementations X to indicate you're overriding a dot here.

C

Because I think my understanding is that it's a dot virtual and then the new virtual function, e virtual or override a dollar which means you've, got e or writing a and b's a new topping.

B

The thing is: it's not really new pasta, I, if I or correctly, if you call.

B

Once of a and you end up calling direction from a only if you use it to be and derived classes you up going to functioning, if you use.

C

The new syntax.

B

Well, this this mean if the B overrides a it basically replaces the body by either like a function and basically, if you call a virtual function, you end up calling this e.

C

I wish we just entered yeah.

B

In this, this is this case. It is that's true that that the friend vibes will be better here by different return. Types to explain. Behavior I think that the the those months rather to I embody all this T it rather than I do so so maybe body when he was writing this on to like mix in more or detail the.

B

It's a covariant winter, but I'm.

E

Just does that mean for me the where it's like, oh and see now has B entries in a V table in it far as I can tell there should only all the parts where it's like, oh the vtable got fatter I I Oh supposed to be, is returning a back to return a returning e like understand when the new lots are popping up and what at the attribute at the run time with you see.

A

I think the way I understand it is because, because of the way il works, Coburn returns another thing in IL. So if you want to have a different return type, you have to have a new slot with ordinate for.

C

A

New signature, then you need to patch up the other ones to point to the new one as well, so that you have the right semantics and the way understand is attribute is what this attribute does. Is it basically tells is the runtime to say well, I haven't really patched up all my base cards, so please do that. For me.

A

Sorry I haven't patched up all the overrides in the base potentially so you have to do this for me right, but.

E

If, if is the particular ease of right of AIDS birth function as all this function, that has the same name, but it turn type I'll earn all slot. One I don't understand where any of you catching is. He has now overridden a's thing to say, call slop and then see. Does whatever and callings up hero that says, earned all slot one. But let us see.

B

Is the for it over I, just love I been it overrides zero.

B

The base class function would inherit see in this in this world. He virtual function.

E

Right iron, Minton Eilert write down a, but it's overriding osvárt and and points to slot the e version and then B's version and points to the ski version. Like sounds like a compiler problem.

A

I think I know I think the.

E

A

E

Express that.

A

I think it comes time the.

E

Il you call it the call to the old method, signature turn type, I understand, dial.

A

E

The reason oh and.

A

So I think the problem is that so, if you look at C right so see the extends from B the thing you're overriding is basically like the like. The original definition is effectively a virtual function.

A

And I think there is no way you can express this in IL to say: I want this thing to override a and B's.

A

I think that might be the problem and so.

B

That's why, basically, by this override both yes.

A

Effectively and I think that's.

B

A

You can't express, or the attribute.

C

A

Ai doesn't necessarily.

C

State that, and it's not clear how that word if you bolts up despairing over but if it has to be or is the default behavior, it's all yeah.

A

D

The idea of the.

A

Attribute is that you preserve the entire hierarchy right, but but yeah, that's the other thing where I think having more detail would help, because it's the concern I have as well that I understand how this works, for this case I'm entirely sure how that works for all cases, and if you eventually want to attribute that can base overwrite part of it. I, don't know whether the name makes.

B

Little sense you function brides functioning a then this attributed means to be applied to Michael all the like overrides so that the future. So in case it's like both a and B. If it's more I mean there is no up to that is just like buying the override to because the a was an open, bioody. Oh I, don't and carry this listener.

E

Right but it sounds more like its purpose of the attribute override all compatible limiters like a require method. Impulse remain enough. I, don't like a link, Demming feature like it. It sounds to me like it's don't, but it's not don't extra. The name doesn't.

H

E

I understand sounds like copy a piece of data from things Tyler, maybe didn't about the as it really exists, but that doesn't sound anything it'll. Maybe it makes sense. David humming doesn't make sense. He from outside.

B

And I said I think there's one. It was not here. He would end up riding slot in of the a virtual function who basically, basically, who called using and himself of a then you would get the right thing. If you put e, then it would call the B version out by a wild wheel. I wanted to all the exceed function as that's the difference that would be. It would mean in this specific example. Currently.

E

Sounds like it's into computation.

E

What happens? Ii did variant in a later of the assembly. Where does the attribute go, etc? It sound.

D

E

Very, very very calm attribute, isn't holding that patient. It sounds like it's like sort of patching, never.

B

Happen I think.

F

That might be the.

B

Details on the type system, because that's out of my like, if.

E

It was every covariant and it was like splat meet. Others and I could understand it, but the fact that it's only appeared also I don't know what it should.

B

Be a issues they should have been applied thing all right. This is a beauty on all the yeah and you are right. It's not clear whether the valerian I, don't know would be eight or clear.

A

I headed a few more questions at the bottom, so, but until issue about this attribute it does, they should talks about covalent returns for all samples use or it's one of the methods used virtual functions or the behavior differences without the attribute. If this can't be explained in CA explained may be the wrong word here, but this can be expressed in C chef that Allah doesn't cool, would help to the name seems a bit abstract. Can we make this clearer, like prisoner of overrides attribute?

A

Okay, your base overrides attribute doesn't have to be a new attribute, or could it be a new blank? How does a tribute play with default interface methods, and how does this play with versioning will override side edit it'll of the hierarchical data did I miss anything or is that it.

E

I think if I had answers it.

I

E

A

My gut feel is once we have answers to all these questions. We might be even more convinced. I was like it's a complicated feature. Perhaps.

B

And they David answers those questions. There will be another round, yeah.

A

So like depends on what the answers are, but my gut feel is probably his one-line comment will probably not help us, so we probably have to get him on next time.

A

It would probably be helpful, so the next meeting we have is Thursday, so that would work.

A

As well, if you want.

B

I think that Thursday I have training and about the David or I other. That's good.

A

All righty, then the next this.

A

Remember anything you say can and will be used against you.

G

G

G

This interesting context, but these two things is not related to each other at all, but we got some scenarios that somebody will need to relate the Jason context with the luminance coming.

G

So basically this this all of this implementation should be determined, but I mean to enabled relating the facing context to the to the law in discovering it to require, like I'm, getting some external memories and like splitting more tickets to the lobby and so on. So we prefer to have like pop in behavior, and you know like who ever need to relate the recent context with with knowing is going so they showed to us they should opt in. To that I mean it will not be sinful to give you this.

G

This actually request came from is because net and and also app inside as your loving manage this stuff. So basically, all proposed API here is just like how to opt-in to get this I'm increasing context inside the donators coming and also to control. What level of the of the tracing context you want to add to the to salumi basic basic scoping. Terrific tracing context contains something is called trace, ID, an ideal parent ID and some state and some flags.

G

So if you look to the example here, I'm not like you know, so the idea is how to enable is that triskin text inside the logging is why you create plane, creates a link builder. You we are introducing a new extension missile on the Builder itself is called configure.

G

This is the same way I mean like the other logging Softworks. So we introducing the configure extension missile on the Builder and inside this configure method, it will take a parameter, which is we call it a factory options. This is what we introduced in this proposed new yards yeah. So basically, I mean like why you create why why we create a factory and getting the Builder? You configure the Builder with factory options, and you enable your neighbor getting the activity tracking options. All active I mean the tracing economics.

G

So if you look to that what is expecting to get in the logging, this is the next thing which is start with info. I mean like he would get something like span ID and you have a participant ID. You have this ID or gets a three side, the parent, ID and so on. So this is the whole idea about what we want to do here. So if you go down a little bit, we can start looking at the proposed device itself.

A

So in yourself, like maybe, let's just roll down all the way to the this one, here's the extension method. First I guess: logger a factory options already exists. No, this is what we will introduce.

G

A

Mean I mean the I'll Augie build already exist, yeah.

G

Everything gives exist, I means that only new things we introducing here is local local factory options and extension method and some UNAM class. That's all that's all sentencing.

E

So is the configure method, does it mutate the input or does it return a distinct value as the return.

G

A configur method doesn't return anything we're just like it's just enable some options on the factory itself, so I'm from this options we can detect about, like you know also, we will turn it on to get that that risk context in the moment, so we will start using it to.

J

Your question Jeremy: it's a it's a fluent pattern, so it's intended to return the disclaimer yeah. You can call talking figure back and figure yeah.

J

Okay, this Headroom is already well established within ASP nuts, so I don't think anyone is that's.

E

Fair, just in general, in the BC L, we don't do fluent so yeah. So the question is: is this already being done in other dotnet logging AP is and if it's not being done in other net logging api's, then we get to ask the big question of. Should we be doing it here, we're in general? The answer is we don't like introducing new concepts, yeah.

A

I think the ship has kind.

E

Of sailed with.

A

My cousin extensions api's because they basically behave like that already across the board right.

E

So this is consistent with in this area. Then.

G

E

Mean everything that I know about this feature is on the screen right now,.

J

Part of the stuff that asp net originally worked on that got kind of handed over the wall to the libraries team. Everything in microsoft, thought extensions, follows asp.net design patterns rather than standard ECL design patterns. Am I mean.

A

That's why it ended up being called my code extensions, because neither side could agree with the other side. So that's kind of the the compromise we landed so.

E

My my easy first thing is: you need to get rid of activity tracking options that default, because we had this with SSL protocols we had default and then, when we needed to evolve it later, we were unable to do so because you can't change the value of an enum field. Instead, this should be a something like a static read-only on logger factory options. So you have the flexibility of modifying the value in future releases. Well,.

A

I mean I would even go a step further and say, like just initialize the default in logger factory options constructor like. Why do you sure, if.

E

You, if you need to expose what the default is then.

A

It should be a static read-only.

E

Somewhere, instead of a an enum, remember sure.

A

I mean like I would just say like if it's the default, like just I, have it to be the default right. I mean well yeah. My other concern is like default on. An enum is always weird when you also have none, because now the keyword default and the member default don't mean the same thing. Oh that's.

G

We rename it well.

E

The problem is it will, if you, if it's an enum member, it will always for the end of time, be span, ID, trace, ID and parent ID, so the only good name for it is span. Id, trace, idea. Everything, because.

D

E

Yeah thing in the next version that gets 0 X to 0. You can't add that to default, even if you really really really wanted to so you should not have the default member in the enum.

E

Something to express like if it's expressing a preference like a default as opposed to it, it is an actual semantic combination of multiple flags. Then a static read-only on some other type is better because that means that it will use the value as defined in the current, build, as opposed to the value defined in the reference assembly that you compiled against.

G

Okay, but in the future, this is static, can be changing to different values as a.

E

Static read-only account as a brick cannot and as a constant cannot.

A

Okay, I mean: do you need it ever.

G

This is I, actually do it. Follow is the one whom requested yeah.

K

I mean maybe not it just seems nice to expose what the actual default is gives you my said it or reset it to something else.

A

It's just that, even if you do it, what people discover it right, because you can't put it on the actual enum.

K

Yeah I mean I I, wouldn't I, wouldn't down the hill for yourself, I'm.

L

K

L

Like wherever we're like the place, where you're most likely wanting to know what the default is, is.

A

In a place where you can access laga factory options already, in which case just ask activity tracking options, what its value currently is and then do whatever you want to do with it like, for example, let's say you want to enable say when it is a was trace. Id right, you would just say you know, activity tracking options, equals activity, tracking options and then and ill. The price ID right, yeah.

K

A

That seems like the the the most sensible thing.

K

J

That's big that also fits with the flu and pattern more because the idea is you're, never you're. Never over. Writing someone in a fluent pattern, you're, always building on top of what they do right. Well,.

K

You can over any right. You.

J

Can but it's not you don't think it's that common, its more common to chain right goes there yeah and in those example of you, know, value and equals tilde. Something is an example: chaining.

K

Alright I seems fine.

A

Let's do that and then um the rest makes sense to me.

G

Yeah so for the local factory class I mean this is already existing class, but we are adding some extra constructors that takes these options, and this is mainly to make dependency injection work nicely was was a feature.

K

This is a sucky part of the eye. Well, it's.

H

Not really it's not really.

K

The eyes it's just a fact of having to evolve api's that have constructors, you add new things. You have to add more well new required things. It's kind of the problem right so so.

A

K

A

News of laga factory is that constructed by the eye yeah.

C

A

C

Well, when you.

A

Add these new constructors, how? How does a spirit decide which one to use.

K

The Container by default, picks up the longest, so that's kind of how it works. They can take ambien, it can be ambiguous either. It has to be kind of the longest thing that has the most yeah I provided arguments in.

A

Other words, the one that is the most arguments, is only overloaded by arity, then.

K

A

Means if you have five arguments, that's the longest. You cannot have another thing that also has five arguments. Otherwise the Iowa Fair records, if so by us, editing, basically I suppose the last one is the one you will not call red providers photo option.

K

Normally, what we try to do is tap a just, not go crazy, but first of all you can't add any required thanks. Is that that's the across the issue right new new required things? Make this make this a problem?

K

We tend to deprecated older constructors ones that are less less used, but that is also a problem. If we want you people to support giving up the factory. The reason it was this way is because people in the past, like before 3.0 with new at the factory by hand.

K

Now we have the the create helper that does for you, but all right that yeah, you could do new factory and then call a provider on the fly. So I don't have a good answer for this. This is just like I said about the point in to construct three inject: the pen sees that come from us like a shared location. That's.

J

Pertinent question: David: um does a too young system support a property introduction or just constructor injection Jess, certain injection in the properties? If you, if you have property injection and then study everything that softly becomes a property instead of a constructor parameter? Would that solve this problem? It.

K

Would help yeah.

A

Not their problem, but.

K

J

K

J

There are other issues around it.

K

But if this is going.

J

To keep coming up over and over and over and over again does it make sense to not try to solve this problem until that other a more underlying problem, so first I'll.

K

Tell you I'll tell you so one strategy that we have used is we have the implementations, be internal and the attractions be public that we could change around the conductor's like as required, but then that has it that has different issues, but.

E

It worked with the current tech if, if your longest constructor as default options for default parameters, it does work, but that one that does work today.

K

Seems like a combination.

E

Of having an existing default constructor, which I assume the type has us making only this longest one and giving everything as a default parameter and then just following default parameter evolution rules would solve this going forward of you'll end up with a bunch of constructors that are marked as editor browsable, never just for legacy reasons, but essentially you end up with two default: visible public constructors ones, the empty one and one has everything- is a optional premium.

A

But I don't think that will work right because, let's say in v1 you have a B and C as options you pass through the constructor. Now you want to add an optional D. It's a no! If you edit in another constructor that no takes ABCD, you mark D is optional. Well, then, now the Kolbe of ABC is ambiguous.

A

E

If only one of them has default parameters, you'll you'll still have an exact binding to the ABC one. You can't remove it and it has to forward correctly, but it has to do the the behaviors correctly, but it's it's just the evolution of default parameters. I see.

A

You're saying because the compiler will prefer, the one that had takes ABC will not consider the one that takes D right, I, guess that makes sense. Intellisense.

E

Can't intellisense won't see it because it see B never, but.

A

E

Why I should one that the compiler will pick I.

A

Mean it's a good example here, a little bit at the Constructors you have here. It does not look like that's what you do today right, like otherwise I'd understand about logger filter options in the middle is different. Yes,.

K

Early-Round, actually, this whole coulis, because.

E

You have yeah auger factories.

K

E

Be like hi options on the top. It's not wrapped in another thing in the middle and his options monitor in the bottom Jesus.

E

So I don't know that was some copy, pasta or.

K

The providers I option is factory options, so I think they were too are. You were empty.

G

So yeah I mean like we have exactly like or other constructors today, which is exactly like this one. Without the options.

G

So basically about like you know, if you look to the first one, which is taking options, only we already already have one which is doesn't take any options. It doesn't.

A

Take any parameters I.

G

A

You today, I have one that has no arguments. You have one that only takes providers. You've won the text.

K

Providers yeah and that's.

A

A

But then I don't send from here to here you have log and photo options and you have AI options, monitor absorb photo options.

G

Right because I think in example, imitation it's really creating one from Gaza right.

A

Yeah, that's the thing that makes it hard, because now you have.

A

D

Understand that that's.

A

Why I'm stead, but but to to Jeremy's point right if we like in order to keep the I brand, that simple, you want simple chaining right. You basically want to have all previous ones to be busy prefixes of the current one. As soon as you have discrepancies like this, it now means when you add new things, you probably need to add them to two to two disjoint groups effectively and that that can result in ambiguities. If you, if you're, not super careful.

E

Yeah yeah I mean you know if, if the one that takes logger filter options is internally just deferring to something to do the options monitor version of it, then nothing says you have to allow somebody who went from that particular two parameter constructor to then still be able to call the the one with all the defaults like you can make them convert to using a monitor instead of a fixed value.

G

A

I, don't have a suggestion. He.

G

A

Damage has already been done, kind of.

K

Almost dead, so if.

D

K

To avoid breaking changes right, at least in five, maybe six we can.

C

Remove more things, what.

K

Is the minimum that we could add and we add one overload like the very last one that we know is yeah activated, passing the talk to your options and then the rest? You get. No, you can't change the options unless you used the DEA activated when I was little the longest so.

E

My yeah, my recommendation, would be to delete the first three proposed new constructors and the last one with every parameter: defaulted, yep.

D

E

Only thing that it cuts out is passing filter options without the I options, monitor wrapper yeah, and that just means that the direction that the API is taking in the future is that that should not have been a constructor. That was public.

K

Okay, oh yeah, that seems reasonable. This is.

J

Yeah the whole thing.

K

J

Think separately, we're going to have to have guidance on di instantiated types anyway, because they don't yeah at the patterns that we tend to use for PCL types. So I.

K

J

Think the typical things that we tend to apply apply here. It's.

K

Kind of like um patterns there's like if you, if you need to add new required things, this will come up every time and there's a bunch of different strategies like have an options: object that can be mutated. Pass into your cutter. Add more constructors, make it internal, so it doesn't matter, I mean kind of have. We have a mix of all three of those in XP dotnet and there's. No really good, like amazing solution is kind of like trade offs right or you can stop adding me required things if.

J

You don't use a service provider for anything.

K

Things Liz Lee, it's kind of it yeah. Those are the things that you emerge from using using today, beautiful I, know and.

J

It's beautiful I mean it's one of the interesting things. There was a it's.

J

You know you, you tell people you're not ever really supposed to do. These things up by hand always use a DI system to do it up for you, yeah, even when your unit testing like get a helper method to do it up, for you don't new it up within your test. Yeah.

K

And, and and to be fair in logging at first was at was not titled to DL. Didn't have any of these characters. It had like a super empty one, but you could call out the waiter on and then, as we added more stuff to it, it got hard as a career. Behind me, I got. The passing like for I was just passing one.

A

And that's the problem with.

K

A

That it's fundamentally viral right, like UV. If you really want to get the benefits of the eye you kind of have to use it all the way down.

K

The I think it's you tend to design types take in more things than you would think about, which makes it kind of hard to create and that's kind of the main thing right like. If things are talking to you, then you can't have the fuss that kind of grab things from the ambient nature of the universe. Yes,.

H

K

That exploding things audibly here and there, which means you can test it, but it also means you have to pass in all the contacts everywhere. That's kind of like double edged sword, yeah.

I

um How many times has the safety I've been broken in the past and yet factor itself? None: okay, okay,.

K

Yeah I know it's.

F

K

Bar is really high. The bar is late.

D

K

Be sealed pretty much is it. This thing runs on forever kind. Net core nice standard um super very we're breaking the api, sir, but I still think. Maybe you want to deprecated things that we think are are bad and then remove it in a future version.

J

Demo sounds like we need to fast-track some of the better off solution, stuff, yeah.

K

Isn't that more thing them I mean.

A

The implementations they might be half of all the things we need to know. You just apply the attribute yeah.

F

J

We've been putting off for quite some time, actually applying them to AP ice, because we we keep saying oh we'll get around to help me to meet him for it. It sounds like we should actually have that meeting. Yeah.

A

We should get this list. Like I mean it was a list earlier right. We just I I just have to update this thing again. Actually, yeah.

J

A

What we actually meant there right.

J

Well it especially because it's getting to be so late in the preview cycle like we need to start acting on some of those oscillations like I know, you.

A

So are there any things that should be defaulted, the constructor of logger factory, or should we just have the options be required.

A

Should options be.

G

A

To null or what what's, the no.

G

Options is at legs, I mean so options. It has to be defaulted to, none which is no I'm in.

E

A little bit not taking the options you're taking and I options of options.

G

E

It can be default to no and no means use the default or use whatever internal version of defaults. You feel like using don't have to be public, so.

K

Helper called options not create, but it's.

E

Not a huge deal yet if there's.

K

E

Scenarios for somebody calling new on this directly and I think that the new constructor should be all defaulted parameters, and if there aren't, then it doesn't matter yeah.

E

So just saying they're all defaulted seems the most flexible because it doesn't have a major cost and it doesn't boxing out.

E

Yeah and then it says essentially so it would behave whatever the same as the default. Constructor does right. So if you don't provide a an ienumerable of logging provider, that does something right: okay, unless providers is actually a required parameter, but I don't think so, given that you were it's.

C

E

An options only constructor, so the calling the longest constructor with essentially calling it empty would be the same as calling the default constructor. So just everything that was Oh is then acceptable. Okay, so if null abilities turned on for this one, then all of them also get marked as null accepting and.

A

So basically should we know.

I

Make providers.

A

And filter options both be equal to No.

K

So, okay, so one one didn't come with the IDI. The thing about equals, no I think it works everywhere, but we need to make sure that it works on all the providers that we support. So that's that's the other things we we do support it and then we do support it. I think it is a recurrent feature of our of our container traction but I'm a little wary of making it no everywhere, just maybe there so maybe that that doesn't work on other container providers and just for context like our data container, isn't abstraction.

K

Well, it's both an implementation, and so, if you plug in a different DI container implementation, it should still work with these constructors, but.

E

Even if the DI system itself didn't understand that the API had a suggested default like it's not going to like, if it just looked at it and said, I need a thing that produces an I options and logger factory options, but can't find one. That's failure, which would be the same as whether it is.

K

E

As default here, well,.

K

It could call constructor just works right. Ilia is fine, picking a different, a different overlays. Worse, that's what you don't want.

E

Okay, so if it finds a signature, a constructor, it can't call it excludes it from the set instead of failing yeah a might yeah.

K

That's the danger yeah, let's just verify that that works, and then we can make it one dollop. If it does work, it's not that we should just add a new required parameter in Olano constructor.

A

So help me understand why you need to have one can't you export a default instance, or does it that that's not supported.

K

Export a default, since what do you mean you mean like math, yeah,.

A

Do you have the notion of the user can specify options of log of factory options and when they don't, you have a default one? Oh, it doesn't even.

K

A

Have to be omitted from the graph.

K

Before has to be so, there's no defaults, you have to have something in the container that you can produce a blah blah I haven't asked for options. We happen to have one added by someone else by default, so the others work. There's always there's always an options of blood in the container right, if you use it in a spirit, cooler or in hosting. If you were to crea container like from scratch and not call the things that add the options implementation, it would just blow up. Cuz I couldn't find one and.

A

That seems like something you should probably think about like how do you add new new things that are effectively required.

K

A

K

Require things all right: yeah, because.

A

You can either do one or two things you can either say. There's this first-class feature of like there's a default, the user can override the default, or you say you know the the thing that actually constructs things and injects things is is aware of default parameters right, like that's yep really, the fan I had I think.

K

It works, we just confirm it before we like.

A

K

Know the use of third-party.

A

Like containers, is it literally every di container ever written or is it like three or four.

K

We have a list on the actual, okay might have gotten lost in the move.

D

But in the inner readme.

K

Of the of the extensions repo that maybe, if you look at a branch there's a list of container but um that work.

A

K

A

We work with the top ones and make sure that they.

K

Actually, work.

A

K

Alright, worse, is we ship a set of specs? There are texts, and then you run those tests on your implementation and if they pass, you can have works.

A

K

Put the sticker.

A

On your project, yeah.

K

I passed ten tests is that gonna test I believe we used to actually run the third prize since I'm, not sure. If I got removed in the move, we actually run those containers as part of the test, but I'm pretty sure. It's good I know.

G

So I come for almost I am to see I'm.

K

Like 86% sure they're gone, but her in the chat, I pasted, the the actual teams- oh look there. It is good job Sarah. Here's the actually link to the the test for you.

K

And there's more than that, but but this is kind of the main ones that are I believe are.

H

You talking about the specification test.

K

There's a as I said, a test called the you know, the external di taps I think it kind of has all these.

H

They might be an extension sweep. Oh yeah.

K

They're well past.

K

It's definitely in a branch.

K

G

Okay, what kind of take this line.

G

We can't forget more about how to run this list.

A

F

A

Looks good to me: oh, it doesn't look good for Jeremy, but he's just not enlightened. Yet oh dude.

K

Just wait until we bring it, get a ps4 extension to get love everywhere.

A

Breaking changes, breaking changes everywhere, you get one, you get one, you get one all right.

A

G

So this considered yeah.

A

I already marked the proof that I did not.

G

The thing thank you.

A

So Oh party go break the world.

G

A

Comes natural, I'm sure all right, so no idea who this is not Terry. You also on this one. Yes,.

G

This should be simple or hopefully so. Basically, this one is approved before so.

G

To summarize, the issue here is: we have including not getting Cooling's he eyes that can enumerate all, including we support and in the.net cool we started to have the model of the including provider which is which is like you can plug in your your provide, your including if you want- and we have almost most of the our including support to the provider, is called the page, including provider, and we proposed about like in order to expose a new it I from the provider which is kinda liberate.

G

They, including the tech support, and this will get reflected to the main main vessel in Zion coding, which is getting coatings. So this VI you get gotta prove it with some tweak, but but there's a catch here is getting coatings is returning array of of including info right and we discovered, after this item, get a blue to. We discover that including info class doesn't have public constructor, so the whole idea additions that I'm proposing here is just to add extra. A new constructors are getting coding and make it public on the, including info.

G

If you go down, we can I can show you and.

A

Was just reading my notes from like two years ago and I have literally zero memories of writing those notes. It's a yeah.

A

I, guess is your comment somewhere right, yeah.

G

A

The one this is the one I just read like the right, okay and then I think you probably.

F

Have some comments.

A

F

It's literally just taking the existing internal structure, that's there that takes code, page name and display name and making that public. Yes,.

E

What does it mean if you don't know what a code page means is: zero the I'm, not a code page based in coding.

G

No, we don't have all, including, as we support it, has to have a good page. Okay, but today could be 0 means the default.

G

E

Really I mean the question: is: is really if it if there should be an overload that doesn't take a code page or or any other forms of overloading that need to be on here? But if, if all not net encodings.

G

The encoding class itself, it has a properties called cool page, so we have to support it anyway.

H

F

Tag that this the get encoding method on encoding info, this turns around and calls encoding dot get encoding with that code page. So that suggests that there's a one-to-one mapping between code, page and encoding info, you could never have two different encodings with the same College. That's.

G

Right, but the catch here is we have the provider model. So if somebody created his home provider and tried to to provide exactly the same encoding, we are trying to get the first one and returned to the user. I mean so the result would not have any duplicates, but it couldn't be easier. Dublicate between providers.

F

Right, but when even it is what I'm pointing out is if there is a duplicate between providers, the way encoding info is currently implemented. When you call dot get encoding on it, it's not necessarily going to return that providers encoding.

G

So we go to the provider first and try to get the including from the provider, and they penny right.

F

But there could be multiple, multiple providers with the same code, page right. Sorry, it's yes and they're all they're all going to return me and coding from the first one.

E

Yes, which suggests that maybe you want encoding info, that takes code, page name display name and which provider is producing it, so that getting coding will ask that provider first or only so.

G

My my point here is: like you know: we are going to filter that before returning the result to the user, so we will not. We don't rural. We I mean the VI encoding to get encodings.

E

Right, but if, if I have an encoding provider that says that I'm using the.

E

The Linux c1 that we're adding support for on Windows that, if I have that encoding provider, object and I call get encodings and it says it has an encoding for I, don't know, 451 I think is Ian us.

I

E

It says that there's there's an encoding for this code page and then on that encoding info I called get encoding. It's not necessarily giving me the one off the provider that I asked for, because if the system one is registered, first I'll get the system encoding and not the the provider specific encoding.

G

Yeah I mean so this is the behavior we need to define here. I mean, like you know, so: Louie returns the system one more, so we return zero by the pump. So my my tick here is to return the provider one because.

E

Right, you can't accomplish that with the instructor you're exposing because it doesn't know which provider it came from yeah.

J

You're suggesting just adding the encoding provider as a parameter to the constructor right. That's.

E

My suggestion yeah you.

J

E

This would happen.

G

In life, including dot, dot, get encodings, EDI itself is the one which is cooling the provider. So it knows what provider it's cooling right.

F

But actual character encoding empowers what's handed back from getting coatings right along the provider. Yes,.

F

C

F

Least, the only consumer of encoding info is encoding, dot get encodings. Yes,.

G

F

J

Hey I'm a little confused because I'm looking at the colors of encoding info dog get encoding and it looks like there are no colors internally of that API. This is a yes, but because there are no internal colors, it sounds like we can actually change the behavior of it to match what our customers expect, rather than what the runtime itself expects.

G

Right I mean we are free to define whatever behavior.

F

We want to you, but, like I, said I guess my question that he is there if this method is never used, no one can currently construct one of these things, and we know we don't ever hand these out of the runtime um behavior of getting Coney kinda doesn't matter right well, we do hand these out encoding.

J

And encoding info rate is exposed to be encoding dog, getting codons right.

F

Oh, it is yeah yeah.

E

Oh okay, but can I I remember the name now it's ICU, so can I directly instantiate an ICU encoding provider.

G

Yeah you can create you wonder why Dan and Brian like implemented a API is that written so.

E

My concern would be if I said you know: new ICU and coding provider friend friend get encodings dot; first dot get encoding, I'm, not actually necessarily getting the one from ICU right, and that seems like a semantic behavioral problem to me because I won I said: I wanted ICU providers and coatings.

G

E

Just suggesting that in coding in from shatrak which provider it came from so that, if you get this, if you call to get encodings on a provider that, when you call getting coding on that encoding info, assuming that's a member that it gives you the one from the provider that created the encoding info as opposed to the one from the system.

G

Today, I mean so, if you need to just write it because I'm not seeing what we need to support it well,.

E

The problem is you're, making get encodings public on the provider if get encodings is protected and is only called by plumbing. Then I don't care anymore, but if it's public, then the user, then someone can instantiate, a provider call get encodings and try to from that encoding info instantiate that encoding and then they might not get the one from that provider, and that is a problem.

G

Without that's.

F

We're talking about Jack, it's got, it's kind of encoding info. Has a method called get encoding, yes, which just turns around and calls encoding get encoding with the coded page right yeah. So it's not going back to the provider. It's not necessarily going back to the provider that this encoding in cocaine, bro, you're.

E

Losing information along the way yeah.

G

We kind of changed the behavior to to ensure that it will return cycling from the provider to Newton. How do I do that?

G

We have all information signs of provider itself, we can know which- including they are requirements from.

F

Zoo, if I have, if I, have three different providers, but all of which construct an encoding input with the same code, page name and display me. There is no information internally. That would allow you to distinguish between those three providers and which ones to go back to.

G

You, okay, you have talked about what you live, an onion from that public API things, including itself or from the provider level. Oh.

J

Well, I, if you called encoding provider, don't get encodings and it gave you back in encoding info right, like presumably you know, which instance of the encoding provider you call it on so I. Don't think that's a huge concern, but if you're calling these static encoding don't get encodings method and it's aggregating everything then! Yes, now it's a problem! Well,.

F

Assuming it works with the singleton encoding provider? Well, you know if you're calling it from me encoding provider because you're still getting back in encoding info, that's not tied to that provider and.

J

F

J

At that point, you wouldn't called get encoding because it might return the wrong thing at that point. What you would do is you would get the encoding name and then go back to the encoding provider directly and say: hey. Please give me the encoding with this with this tonight. Okay, that's exactly the issue is exposing.

F

J

F

All because it's going to do the wrong thing. Oh.

J

Yeah yeah yeah I'm, saying right right now, there's a way around it, but it's very unclean yeah.

E

Yeah like if the get encodings on the provider is only supposed to be used by plumbing than it needs to not be public it, it should be protected and then there's some stuff inside, where encoding, with all that static methods can call into it using an internal method that wraps the protected method. I mean.

A

My question is, generally speaking like do we support that different encodings from different providers of the same Co page? It seems like the API. It makes the assumption that it's unique, but otherwise what does encoding dot get encoding with a code page mean no.

G

There's some scenarios, of course I mean I, didn't see it in real life, but some scenario is like: if somebody needs to overwrite the current encoding I mean they don't like I mean like we have a bug that we never fix and they need to fix it themselves, so they can provide their own provider with this encoding, which.

A

Is alright, yeah and then I agree with what Stephen said like? If, if that's a support in scenario then, but then I'm not sure that they fix that Jeremy proposed is the right one per se, because if I can still call encoding dot and getting coding 42, what does it mean if I'm not replaced encoding 42 like that seems not very fast.

E

That's gonna be a problem, no matter what right.

A

But it's writing.

E

A

If we, if we, if we say it's a problem regardless, then why is this encoding thing and the including provider a new problem, because if I can safely say that an encoding Maps, a particular called page, then this encoding provider thing is just yet another version of that, because now completely depends on how you got your encoding.

A

But if you walk the encoding provider and got the including info and then constructed the encoding from there will give you different results from saying: oh I'm, going straight to encoding or get encoding right, but, like you already are in the world where you lose information, depending on which entry point you chose I'm.

E

Looking at it, as so my sort of semi scenario here is you, you have a specific provider, you call the get encodings and you have a sorting function. That's going to say of all the ones that this provider supported. This is the one that is the best match, and then you call get encoding off the one that was the best sort if you got it from a particular provider.

E

So you expect that this is being a factory method from the provider, but instead it goes back to the system default order list and doesn't necessarily come from the provider. You need to yourself have tracked the provider to call not getting coding with that code. Page.

G

And slinging, we can't have a constructor that can take the including itself. So, who are we in written from the provider? We create encoding, classic self and determine.

G

So you can yet including info and from the constructor, which is the exact Kooning, including class.

J

That might be a bit overkill. Honestly, um you can imagine caching the encoding info instances, but not necessarily caching. The encoding instances themselves, the.

E

Encoding instances always stateless, there's, nobody, there sometimes they're mutable.

G

I

G

I'm just trying to think about today, if there would be any problem to Korea, including itself from inside the provider without needs a public.

J

Api IBM, but but at that point, why return in encoding info? Why not just return the encoding directly like you? Can.

G

E

But if encoding info took code, page name display name and a knowable provider where null means just use the system, the system, the meta provider and that allows the get encoding method to know which provider it came from and to just do the callback- it's yes, it's one pointer bigger in the structure, but it who cares.

E

Even if the system top-level static encoding that get encodings does deduplication and and it has its own order and help somebody already called code page zero, so no one else is allowed to return an encoding info for it. If you went to the provider directly, you'll still get a sensible result, and that's that's my concern here. So.

A

Can you can you accidentally describe what the ill-defined behavior is that you would observe like who do you basically call.

E

With the current API yeah, so you have a specific provider, so you have the ICU encoding provider. You call getting codings. You run that through a sorting function that returns the best encoding info that you have. So you get that back and now you say great. This was the best encoding info for my scenario: encoding info dot get encoding.

E

If that does not return in ICU encoding or something from the ICU encoding provider. That feels like a very bad API to me. So.

A

Okay, I buy that, but like what I understand, so you call ICU and coding provider get encodings which gives you back in encoding info right yeah. So why would that not give you only encoding infos? It came.

F

Down listen. The problem is that, on that instance, encoding info has a method get' encoding, which just calls the static encoding dot get encoding Coppage. So it's going to go back to the global lookup.

E

And if the, if a higher binding provider also provides the same code page and you get back to that providers, version of that code page instead of I, see use okay.

A

So how do I construct an encoding then normally like I'm not supposed to call get in for encoding and forget encoding, I, assume.

E

There's encoding provider that getting coding of int and that's exactly.

D

E

Encoding info needs to cool. There is yeah.

A

I see, you know, I see, so the problem is really encoding and forget. Encoding. Then right.

E

So the solution that I suggest is at a fourth parameter to the public constructor, which is a nullable encoding provider. Null means use the system default meta provider, and otherwise it's only used this one provider that it was bound to yeah.

A

It was about to say, like it seems like. The easiest thing would be that encoding info would take an encoding provider as one of the arguments in the constructor yeah right.

E

And then maybe it should expose that back as a property but I don't it seems less necessary.

A

I mean it seems to me the most logical thing.

G

Yep yeah, so so I guarantee about play any including info object will always create encoding from the providers that it created this, including game flow yeah.

C

J

Yeah, this also means that we would need to just fabricate a dummy provider for built-in things like asking UTSA so on, but there can be an internal implementation, so.

E

J

D

So it's so it's.

J

A billable parameter, not oh, that's what I said. Yes, sorry, I misheard that then yeah, okay, no Liz.

E

The meta provider.

G

H

I

A

That means encoding in, for the first argument would be encoding provider that will be nullable.

E

Or the last one I, don't care.

G

E

Last one would.

G

Be better I mean.

A

Considering that we only make it public so for people that implement encoding providers to call it seems, themes first seems the most simple to me, but.

E

I mean the question, it really is: is it knowable and then.

A

E

Or is it just knowable and if it's knowable and default and it has to go last.

A

Fair enough, but.

I

I, don't have a strong opinion, I mean.

A

It seems like if I were to derive from encoding provider. You want me to pass it in right.

G

A

Given that the only customers of this constructor right now will be the ones doing that, it seems almost a before I parameter would be better.

E

That's fair and then the system one can and cheat and called the existing internal constructor and has an old provider and knows how to deal with that. It doesn't have to be the meta provider, doesn't have to be part of the contract. Yeah.

A

It seems almost the most logical thing then so we'll be then even exposed to product I guess we would select spores the property around.

J

A

J

Property, no, you need to know you need to have a dummy internal provider like you can't make that knowable. No.

E

It would be weird that it was a required constructor parameter from the only constructor you can see, but it says that it can sometimes be null when you ask, for it, I mean it's.

A

J

A

I guess I buy that too, so that I mean it seems like the value is all too low. So I'm, okay, with just a million right, opportune time.

E

It also just make an internal type of system encoding provider, which is not public, and it returns the singleton instance of that which just wraps all the static methods, yeah.

A

But then one seems more like what Eric said earlier like nobody's asking for that so like maybe we should just not have the.

J

Well, the provider type itself, the abstract type is defined in core, but you're right, the implementations are all out of and yeah the base class has to be in corridor wise core, wouldn't be able to call into the outer band implementations, yeah I, think you're.

G

Right this okay, yeah I, mean.

E

If, if the, if it is true that encoding provider is a higher layer, then the solution would be.

D

E

Of into encoding encoding, factory-like.

J

And encoding provider.

G

Exists, we don't we don't need to work yeah, but anyway, I mean like I. Don't think you need. Talks was property from now at least, and we can have just the constructor. We take the extra parameter, no yeah to the property leader. Yes,.

E

Okay, so none knowable provider in the public, constructor implementation detail we deal with it. However, we feel like it and no property to expose the input parameter seems fine.

J

Yeah and we need to add the get encodings method to encoding provider. Yeah.

E

Which is part of the proposal yeah.

H

A

Basically, you are supposed to cool, so basically would call pudding.

A

A

A

One that takes a string or the one who takes an INT.

J

You eat as an implementation detail.

A

If you have to override both of them and to have consistent names anyway, right yeah, but we should probably.

J

A

Spec, what we look, what we used to look up things right.

J

Right now it uses codepage so may as well. Just use the same on the provider and.

E

Yeah, unless, at the same time, we wanted to add.

E

Getting coding of encoding info, but then that seems too complicated.

E

You know, because I would let you do something like use, because I don't remember: if code pages are different between Ian us and E and CA. If they are, then the code page solves it. Otherwise it would give you the or the sub locale of in United or English United States, whereas a place that I've not had a big fun of southern Indiana, having a different set of sorting that that excludes y'all's or something I.

J

Believe y'all's is your own possessive.

E

Well, I was using it as the plural of the singular y'all okay, which is a mass now.

J

C

J

Default implementation of get encodings is returns of nominal, empty array, I guess.

A

That already we already said this last time he said a little bit of where is it or passed? My comment.

A

A

Yeah we said in encoding she'll be version between an empty list. Instead of throwing.

C

A

This isn't correct: it means that one provided that is an update. It doesn't spoil the enumeration of everyone else right.

G

We shouldn't throw from yeah commented from the previous design review is getting coatings in the provided, shot written I instead of array, I'm, not sure I mean.

J

I, don't care enough about that. Yeah.

G

But this I just mentioned because I mean.

E

It honestly, it seems nicer, especially it lets a provider, use, yield, return and any other iterator pattern where they don't have to pre count it, and if all that, if the target scenario of this is you're going to just or each prefer, if you're doing essentially select many provider, not get encodings dot unique of code, page yeah I can I can do link that ienumerable seems better because then you get to you're, not allocating the fixed size, arrays and then throwing them away. Okay sounds.

G

E

Recommend ienumerable of encoding info and then the default implementation just returns, enumerable empty of encoding info.

J

A

A

Because it's getting coding in force- oh, do we know turning not.

G

A

G

Already, using it and including class, get encodings, which is system, including game for so yeah.

E

I think this is be consistent in the space instead of be it have the correct name right.

A

And should to be public or should we make.

J

It protected hazal make it public, there's no good reason to make it protected right. Yeah.

A

For me, alright, so copied and pasted the thing down or fighting over that empty behavior.

E

The but you currently have on the screen- maybe you just didn't hit refresh, still has returning an array instead of an enumerable.

A

Yeah I said it in the comments that didn't air.

M

All right, I think that's good enough. Yeah.

A

Even your amp hi.

J

So we actually talked about this back in January and it was partially approved at that point. Basically, a string, split option, string, dot split turns out. There are lot of places in the code where we split, we get all of the elements of the resulting array and then we call trim on every single one of them because we don't want whitespace. A very typical use case, for this is parsing like comma separated values. Stuff like that so yeah. The proposal is, we add an entry to string, split options right now.

J

The two entries are none and remove empty entries. I had proposed also adding trim entries and the behavior is pretty much what it says. It trims members of the array before it returns the array back to you. If you specify both trim and remove empty entries, it trims the entries and then sees if anything is empty and then at that point it Chuck's them from the array.

A

Is it already a flexi room I believe it.

J

Is? Yes? Yes, but.

A

You only have two options right now: yep.

J

And that's why we do this right, so we could add a so. We could ask up in the future. I guess.

A

That make sense, yeah I mean.

E

Yeah, it's plural, so it should have been flags good to see there. We were consistent back in the framework 1o days, yeah.

J

When we, when we last reviewed this in January, it got punted back to me saying we should see if we need, trim or trim start trim end broken out when I looked through. All of the call sites. I didn't see anyone that would actually benefit from trim. Start incrementing split out like that, the only people who would the only people who were doing this for people who just called written basically so I figure may as well not complicate things to.

E

Devil's advocate the scenario in a CSV, particularly if you do Excel export to CSV. If the value itself contains another comma, then it gets quoted. I assume that there's no intention of making string dot split support, complicated CSVs with a essentially with quoting and escaping and correct. Okay, just you brought up CSVs is your trim target and that making sure that we know that it doesn't solve all CSVs, only trivial ones,.

J

In fact I, while I was actually implementing this, I found two examples of that in opened up bugs I mean: can we.

A

At least say on record that you should not use string, dot split for CSV files, I mean as soon as you do, CSV files you need to kind of handle it quotes and you need to kind of handle separators in quotes and so strings that will stop working for you quickly. Yeah.

J

There it turns out, there are actually quite a few cases where people are doing kind of like fake CSV, instead of real CSV, where they know for a fact that everything is supposed to be a number yeah.

A

J

Need Xcode and stuff like that yeah, but.

A

Even numbers like that, you know you good old germany or I'd rather use commas as decimal separator.

A

Yeah I can tell you the number of cases where I worked for companies or with companies that used those souls called cheating, CSV parsers, and always regretted that yeah.

E

I mean nothing is stopping us from in the future, adding another flag that says both.

A

E

Essentially, you know, support quotes or something which understands quotes and or removes that, like I mean we could make this API be smarter in the future. We're not limited by the lack of that flag right now, right, yeah.

A

I mean Nick trim, start tremendous seems more likely honestly, like that, and that seems already fairly Ramone. If.

J

You, if you look at the may be fixed by a link at the top of the issue that that github lovingly put there. You can see all the different places in the code that I already fixed up to use this API, and none of them were doing CSV specific things. They were just doing. Csp des like things.

A

May be fixed by him what why maybe like dude they're, basically telling you all why we don't know whether he actually fixes that I caught my eye so.

J

Yeah, something like I.

A

J

Person says, resolves.

A

Then I think come on.

J

Or babe baby maple, yes,.

A

Area so that yeah, that seems reasonable them.

J

There is one other thing that I wanted to call out here, which is: there is some weird edge case, behavior of string das flight, where you can pass in the maximum number of items that you want to receive in the backing array. So you can say things like string, dot, split on comma return, max 10 items, and even if it sees 20 comments in there, for instance, you will get back at an element array which makes sense, except for the behavior of string.

J

Dot split today is as soon as it sees that night, comma, and it says oh I'm only supposed to return 10 elements. It literally returns the entire remainder of the string without doing any type of further post-processing I preserved that behavior here, even though I believe it to be incorrect. No.

E

I mean that makes sense, I think I've used that from my own lame CSV of I know that the first n values can't have any nested commas in them, because their enum names or you know, integers or whatever and then and then it's like oh yeah and the rest of the string. I have to process separately. Yeah.

J

But what that means, though, is um like, if you say, remove empty, ignore, ignored, trim entries, say that you pass and remove empty entries, and the very last thing in the string is a comma. Then that means that the last element in the array that you get back will literally be an empty string, which is presumably not what you wanted, because you passed in remove empty entries, and that is an existing behavior. That I did not change and presumably.

E

That means you're also not trimming it correct yeah. That seems consistent.

J

Utf-8 string is different by the way, because this is also this is already implemented on utf-8 string and the utf-8 string. Behavior actually looks at everything, so if you see nine commas and you said, remove or give me back ten things. Sorry, if you see 20 commas and you said, give me back ten things. It will check to see. How is everything else emptied? Let me not actually yield in this part. The return value I would.

E

Recommend you match the existing behavior yeah.

E

Yeah, it's string! That split has been around for 20 years in utf-8 string that split isn't public yet.

J

Yeah well, we also have a read-only span of char version of string, dot splint, which is in a which is in PR review right now, I haven't I, haven't checked to see what the behavior on that nothing is.

A

All right, so no one's done now. This one you're still up yeah.

J

We should obsolete this.

A

I mean I have no concerns, I mean just for context. This guy is basically used by the compiler when you use fixed statements over strings yeah.

C

A

You said, because we added get printable reference. The compiler will prefer that one now over there offset to string yeah the.

J

Sea-Shore compiler will yeah there's the f-sharp. Compiler is already updating their stuff. C++ CLI apparently still calls this, but I just need someone to point me to where to open up that bug, but even.

F

If, even if it didn't fix that for five Oh mark eNOS obsolete, wouldn't break anything right because presumably the compilers aren't saying I wanted to use that API, but it was obsolete. So I'm not going to yeah.

J

The compilers, the compilers, aren't going to be broken by working us off, so wait really what this is ism, if you are, if you yourself, are calling this API for some reason, we wanted to stop right.

F

Right I just read in C++ CLI and F sharp, but they did nothing. They shouldn't be broken by putting obsolete on this yeah.

J

The people who might be calling this API are people who are emitting dynamic methods, for instance, and we want. We want to tell those people like don't call this instead call get a reference, but.

A

I mean this point, that's the whole point of obviously doing this is so we can change the behavior so that and that behavior change will break anybody who called the API. But it's a compiler or a human being, read correct. So we.

J

I mean where we are obviously going to have to we're obviously going to have to allow the existing ecosystem of binaries to continue to work, people who might still be calling this API. So it would require that existing dll's, which are already compiled with us like when we're ingesting them into the into the chip importer like we would still need to somehow handle them correctly. But if we can start reducing the number of dll's that we have to special case, because we can start getting people off of this API.

J

That makes it less likely that we will. You know miss, recognize, a pattern when we eventually do make this change. What is the change that we would make this? One of the things that we were trying to do was experiment with doing narrow strings and the framework strings that are backed with ASCII instead of utf-16, and that would have that be offset to the data, because we.

C

J

In front it would it would mean that the data for the string is actually not in line with the string object. It's actually in a backing array is a completely separate object.

J

So when you there is no offset yeah, like literally, if you call, if you call this property like the property, is going to have to throw platform, not supported exception, because it's, if you try to use it you're, going to get an access violation which means that, when we're analyzing, when we're code chatting aisle that calls it, we actually have to pretend that you had written a completely different code pattern. But we experimented with doing that a while back, and it turns out that logic is very error-prone and we have gotten it wrong.

J

So if we can get as few people as possible continuing to use that pattern because we're updating the compilers and everyone else to call the new API, that increases our odds of success, because it means that we're messing with your ILS during the import phase.

A

I didn't even know that we would have fix a pile on loading. We would have to yeah.

J

Because you're, you want your nuts tender to OD, yellow to continue to work on dotnet $5, $6 7 right no.

A

Maybe I don't agree with that as I'm just saying it's cool that we do that, like I, wasn't aware that we even had this ability to you know I'm I'm, thinking of more creative ways to use this, which you probably don't want, but we.

E

Already have a couple scenarios that whether you've done you know ready to run or engine or anything else that it the runtime, can understand. This was not built in the in the mode that it should have been, and it throws everything back at the jet so and then the jet does magic.

A

J

That's so that's what this is. Basically, don't use this old API, because it's it's possibly going to return the wrong thing. If we ever make this change use a new API instead, which is always guaranteed to work.

G

So if somebody uses the compiler version against the new runtime.

D

G

Will not generate any warning or anything during the.

J

Compile time right, correct the compiler actually already special cases, the new API and calls it if you're targeting a compliant runtime. Oh, but.

E

If they used the last pre Roselyn version of the compiler yeah, but that was against Eve's comment of it would be very surprising that the compiler, which is the one producing the call to this, would bother looking up this thing that I'm about to blind call: hey, wait a minute that was marked obsolete, yeah.

J

G

G

E

Would get the you you built something or you you uh like? Do a reverse, compile.

E

And then so you build like it just builds dumb c-sharp to call what the isle said. Instead of trying to do pattern matching to get back to what it might have looked like and then now, if you recompiled that it would and say, hey wait a minute you're calling this get offset number and that's deprecated yeah.

J

Conveniently you, actually, you can't represent in c-sharp what the compiler would have generated a Nile there. We go.

A

Is that a kid complaining me way, yeah.

J

There's there's a spider above me and the cats warning me that there's a spider. Well then, it's very cute I.

A

E

A

E

Wanted more expressive, c-sharp semantics! That's.

M

What I thought too, but you should.

J

Listen to it later, I should it's. It's actually a few feet over to the side right now, so I'm not freaking out quite yeah.

A

But if you let the cat lose, you know what happens right like every time Mike had ate his spider, it will basically eat the whole spider and then spit out the individual legs, which this is very creepy.

A

So let's do this then on this it was like an obvious one that we can just approve, as is, but my only question is: should we put a diagnostic nd on this guy.

J

So that that I actually don't know and I'll defer to this group, um we obsoleted drink coffee in a previous release without doing any type of diagnostic ID, because that's we didn't have diagnostic IDs at that point. Right um and I see this as kind of in the same vein, like don't misuse strings and in this particular case you should literally never ignore this warning, but you should always address it in your code always and to.

A

Me, the only benefit here would be I mean Jenny speaking I. Think absolutely members is generally much less of a problem of seeing types right, because types will.

F

Usually, result.

A

In potentially hundreds of violations, but like the member seems less problematic, I mean the only benefit here would be, as you said like if there's other things that are like it like, for example, standard copy having the same diagnostic ID might make it easier to point people to why that's a bad idea. Yeah I, don't think of.

C

A

Should suppress it, it's more like okay, here's, the diagnostic ID! You should just freakin fix, yeah.

J

And that's fine I'll I'll defer to the wisdom of this group here, but it's I know that when we had talked about better off solution in the past, we were we kind of bucketed things into a few different categories. One of the categories was like I know what I'm doing I kind of accept the fact that this is obsolete and the other category is like here's something you really do need to address in your code. This falls under the second bucket yeah.

E

I mean the the benefit of a diagnostic ID would be that there's a Doc's page explaining the problem. Yeah.

A

I mean in fact like yeah, the only boob would see this uh people who are literally emitting code by hand, which is a very small, said right. Everybody else is just as fixed with just by recompiling in the compiler would just pick get, pin up a reference now right, correct.

J

And in fact our I mean we have unit us which use dynamic method to call this API, but I mean, of course, where the runtime, of course, with a rights, Atlantis, yeah.

A

Yeah I mean I I mean unfortunately can't tell who causes API, because you can't differentiate between competitors emitting a call to distress.

C

A

C

A

I had to guess the only public do. This are on the YouTube right now, because people like Kevin and then that's pretty much it right so probably yeah doesn't seem worthwhile overthinking. This I mean I would also generally say. If we later on decide, we want to have a diagnostic ID. It is technically a breaking change to add one, but I would not necessarily object to that because to me that's in the same vicinity of analyzers right, it's you just improve the diagnostic or I, which seems fine.

J

A

This what's breaking change because you change basically what Diagnostics you produce right so like if you recompile you, you may have suppressed the generic obsolete one and if you change the diagnostic ID well, now you get a diagnostic ID right, but oh.

F

We were suppressing the generic one, it doesn't suppress individual ones. No, that.

A

Was the feeder bead? We discussed this with the compiler team and we added the ability to add one we concluded like if we would have to do extra work to do that and we said, like the cons outweigh the pros or the pros outweigh the cons. I guess it's like a it seems makes more sense you to turn to not do that and I think in general I'm. Okay with that, as this was breaking change because kind of the only cases where I can think of they would like to do that.

A

It's probably partially because you want to get it into its own bucket and to me it doesn't seem bad because the way I see it is the same as analyzer. So it would be very, very okay if I take an analyzer and that would also have its own diagnostic ID. So it seems fine, maybe.

F

This is a separate conversation which is caught me a little off guard because you know something like analyzers. I do have a big hammer which is I can turn off the whole analyzer. How do I turn off all op solution?.

M

Mornings- you don't in that world, but like how do you turn of all analyzers.

F

And my analyzer can pay I just go and delete it.

A

So what do you delete because I mean you would you would have to add one more, a diagnostic ID to the things you suppress right?

A

Remove the class I.

F

A

E

Or the assembly yeah.

J

E

Sniff it out of the list of stuff to do.

J

Yeah, you don't suppress it, you just remove the reference to it entirely. Yeah.

A

But like I mean that's the general thing right I mean like like as soon as you have to modify your project like I. Don't think this is I mean to me it's just the way. What you do is the same thing effectively you. Basically, we configure your project removing a particular animal I.

F

Think to me it's a service, it's how much work I need to do. Maybe it's not a deal big deal, because we're not gonna have a lot of these. But if we end up in a world where I have 300 different diagnostic IDs for different obsoleted, things feel like you would wanna be able to turn off can't.

C

You just know warn nothing.

F

Maybe that's the thing. What would you know warn ya if you.

M

Don't have it done work.

C

To get eat it rustic item, the compiler gives a general warning for all obsolete to.

F

Stop we were just talking about that Emo's saying that doing that doesn't actually suppress the ones that are at their own diagnostic ID right.

A

Yeah, because there was the feature right so the compiler raises the warning under that ID, right and so I know, if it's a present for that, it won't and won't do that I mean that was called it in the spec as the behavior. Nobody liked the compiler team agreed with that on or said nobody had concerns with this, so that lets OB ended up doing I mean.

D

Yeah and shifted.

A

Yet, like so I think, if, if you feel strongly, we can reopen that I mean it has pros and cons right like the con is that a lot of people already have suppressants for obsolete one. So none of them would benefit from the new ones and that's what we the biggest con right. There's.

J

This kind of dovetails with what I had hinted at earlier, where, because it's getting to be kind of late in the release cycle, we probably need to start getting some of these observations in sooner rather than later, so that we can get feedback on them.

J

Not justice but other observations, yeah.

A

A

Let me remove the blocking one as well.

A

A

J

So did we decide.

A

A diagnostic ID, or not for this, so what I'm hearing is that we don't feel strongly for adding one. The means also adding one right now would be not super useful, because if you were to add one I would probably change the one on copy as well.

A

The caveat I just called out whether it is technically a breaking change, because you change the diagnostic ID. But to me that seems ok, especially because the old day we just shipped is like six months ago, right or 12 months ago, or something like that. So there a recent observation, but given that the amount of people who call this API I can probably get into one room, doesn't seem worth it.

A

All right- and this was this one.

A

One more so price.

I

Four minutes.

A

Also Levi's.

J

This there was actually a customer request for this. Basically, the Latin wedding code, the Latin one encoding, is pretty common in web scenarios, because it's the encoding that you use for things like HTTP, headers, they're, not actually ASCII they're, not actually utf-8 they're, actually, iso 8859-1, which is colloquially known as latin one right now, you can use encoding, not get encoding of the literal string, iso 8859-1, but like no one knows to do that.

J

It's just kind of difficult to remember that that's what you have to do, so this is exposing it as a first-class property, alongside utf-8, alongside unicode and so on.

J

The only public api being added here is the latin one: property getter we're not making the Latin one encoding class of South public.

J

F

There was a discussion about the name: I, don't know it up, theorists reference. No! You re yeah.

G

Yeah, it's I mean for me, I'm, not sure about how many people outside Microsoft knows what latin-1 means in general. Actually people knows what.

G

It's used for anyway,.

C

If you have a Latin one, then the wiki page on iso-8859 s, the first thing to come up. Yes,.

J

Yeah yeah I don't mind, calling it iso 8859-1, but we need to somehow have punctuation between the 8859 and the one, and I don't really know how to do that. Other than use an underscore and that's.

E

What we did in the TLS cipher suites because of the same there's a number here in that it needs to be broken. Yeah I mean I've, always known this as Latin one or I guess Latin, one I can't say if I knew of it before working at Microsoft. Since its you know a decade and a half ago, but to me Latin one seems obvious and yeah doing a search for Latin Latin, one with no or Latin one, either way the ISO 8859 one Wikipedia page is the top hit.

C

So it's worth noting that at that top of that wiki page, it gives a link to the Latin one supplement, Unicode block, which is the official name of the unicode block for the upper half of iso-8859 Walton.

G

So well, after creating this latin-1, including a cuckoo clock named Annette, what is returns.

J

That's a good question: it's going to be whatever encoding getting coding. Iso 8859-1 talking aver terms, because it's literally the equivalent of calling that api.

G

G

If the property will return at their company, I mean, however, does really mean. Okay,.

J

Yeah I don't remember offhand what it was yeah.

G

A

It seems like a reason compromise to say, probably like encoding dot. Property names should probably match encoding, get encoding dot, name.

G

Say about like, if somebody knows about like eyes: oh it's five 9-1, so they will call get including with this name. They will not put the prophesy believe, but but I mean like we are providing this property. Latin-1 property is just affordable, familiar those Latin, one yeah, it's an accelerator, that's all guys yeah! So.

C

Encoding can encoding Latin one is valid and it returns iso-8859 one because.

I

We special case that name.

A

Right, so if we were to use I mean looking at search engine results, it looks like iso-8859.

A

One is more like has more hits than Latin one, so I'm fine with using either one I just don't know how we would name the the eyes or one, because you would have to put IV can't just omit the right and then underscore would look extremely ugly. It's dead and.

E

Writing writing it as iso-8859 da Sh. One would be a very strange.

A

J

A

I think letting one seems more ansible.

J

Name, I also have an error in my description here. The narrowing conversion does best-fit mapping right now. It says that it, uh it throws out everything. That's above the asking range.

J

E

They are smart.

J

Books and apostrophe or double quote as oh, no sorry I do call that out correctly or sorry, I I can't even read what I wrote to know. The description here is correct. Ignore what I said it's.

A

Always fine to find out that you are smarter than you think it would be.

A

That to me is almost irrelevant because it I think whatever the behavior of encoding got getting--getting coding with that name, it should match whatever encoding dot. Lettin one gives you, it would seem very weird and it would differ so whatever the behavior is. The idea that we have so.

E

You want it to return a special latin-1 encoding object that wraps the real latin-1 encoding and just says the same as latin one. No.

A

No I'm saying he was just talking about what the behavior is for certain askew ranges and I'm, saying I: don't even care because, whatever the behavior is, we currently have when you're saying cordoned or getting coding and coding or letting one should match that behavior right. Okay,.

I

A

Should yeah, which is why it to me, it's a red herring, whether we expose the property or not like this.

A

E

I think it seems reasonable. I know that you know in the ASN code we would have used this, but we can't cuz we're at standard, and presumably this is that court. So.

A

Is there any reason I mean I know that we expose ASCII, encoding and UTF I, say Unicode, encoding I think is dedicated types like did. We just do this because we didn't know better in v1, or is they actually said ah so you want to use them directly.

J

You can, when you instantiate things like utf-8 encoding, you can pass in parameters that control things like whether the byte order mark is emitted. Latin-1 encoding doesn't really have such a concept, um so there is no need to configure the encoding in that fashion so because there's no need to pass in a constructor parameter. There's no need for a constructor which follows there is no need for a type yeah.

J

You I'm drawing lots like I'm, jumping quite a bit here, but no one's asking for the type right now people are just asking for how do I do this, and because there's no accelerator, they don't know how to do it. Yeah.

G

We should use get including it at that time. Yeah yeah, yeah.

J

There are, there are options that you can pass into encoding dot, get encoding that control things like what happens when you encounter bad data and those calls will continue to be honored, even even when this goes public.

A

That makes sense.

A

J

Gone through all of my issues in 35 minutes.

A

Well, we just walk all this to newest. You just were lucky, oh yeah, so it's not that.

D

E

A

E

What it's worth in in my sequel character sets. They call it Latin one, not ISO 8859.

A

Yeah I was about to say that, like that's how I know what lettin one is, because that's with sequel server is using yeah.

E

So I think in sequel and Latin one is the is the name and the ISO name is not one that people know yeah.

J

And I think in our documentation we would. We would explicitly clarify what we mean by Latin one just to remove all ambiguity. Yeah.

A

Alright I think that's it for today and then we will continue on Thursday I think we have how many more we have. Let me just feel refresh here. We have.

A

One two three, four, five, more issues that are marked looking and then maybe we can even took a look, take a look at the backlog. Maybe I know we keep saying that, but.

F

We'll get there one day, one.

A

Day, all right, thanks, friends, see you later thank.

I