.Net Foundation Design Reviews, 21 May 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: Nullable Annotation Attributes

Description

-10:-23:-04 - Approved: Attributes for nullable annotations https://github.com/dotnet/corefx/issues/37826#issuecomment-494505569

A

You all right so I call this issue based on the email that Matt found out so I've not updated since I guess that the afternoon or whatever. So, if there's any late modification, then we can talk about this, so I, basically just pasted all the you know the things we had.

A

What I haven't done yet is I have not applied the attributes, targets kind of thing, because I don't quite understand where these go in all cases like I bet some of the descriptions that I have some questions myself, so I I did not do all of that. Yet, but if.

B

You scroll scroll down to the bottom.

B

Right sprite there yeah.

A

C

B

A copy and paste of what I have right now that in my drafts, PR I don't know that I got them all right, but I can tell you why I chose what I chose dad.

A

Told it works, yeah I just wanted to make sure that we actually look at api's and not just some very rough.

A

Or something right, how do you wanna walk? There's like I, think math group them by certain things like preconditions and postconditions.

D

Families, I guess.

D

Yeah, these are generally in the same ordering so I'm the first category. First two are about regulating input beyond what the nullable annotations are right. So it's about saying and all it's allowed in here, even though, even if then, you know sometimes use a generic contact right, even even if the type would normally allow it, so it expresses especially Nancy or the opposite null is not allowed here. Even if the type of house, those are fruit, antigens I, was.

B

Gonna say, and it's relevant for both generics and non generics. Generics is obvious because the define you know the person defining the class if it's unconstrained, doesn't get to specify question marks on things and non generics in a variety of situations. If a property is defined to be say, string, but the getter and the setter disagree on what can be returned, you know null versus non null and what can be set null versus non null. These allow you to override what you set us, but the property level or.

D

There might be an argument as to could we make do with fewer like why? Why is allow null, and maybe now why? Why are they different? One describe one describes preconditions: restrictions on certain, the other describes what can be expected of output and, in certain cases like ref parameters, for instance, they really have to it really do mean different things on the same entity. So you can't like elapsed. I mean to wanna know, is it the same thing? Is.

C

The interesting output I, don't.

E

Know if allowing all and disallow null would be reasonable, longer turns if we have them yeah.

F

Under turns I.

E

Don't know as though they go on returns.

F

Yeah, these are.

D

F

D

Just be a copy of it right, these brain food only so they should not I should probably not go on. Why.

G

Are they allowed on methods in order to.

B

Allow them on get it getting set. Accessors I had to allow method.

G

And that's why we don't have delegate, because it doesn't make sense there right.

D

Like- and that goes to one of your open questions, I guess, which is when these go on property, should they go on the accessors, or should they go on the property itself, but Steve Derrick for now? But it's done something we it's something we could close on here is to a lot of the money accesses, which is why you need the meta thing right, but we, if we had them instead go directly on the attribute just like they would go directly I'm sorry, they recommend the probably just like they would go directly on the field.

D

That will look a little different.

A

F

G

From the IL perspective, you wouldn't want it on the property. If you were going on the accessors, because in IL you can have more than just kids.

G

You could also have like properties with parameters like you can be right.

G

um Kids, you can do indexers, but with union property yeah.

D

G

So you wouldn't wander off the property.

D

It would then apply to the value of, and probably I guess think, both reasonable, truces.

A

When the question so many things you want to support right, I mean what happens when you them on the property and then also on the parameter like that. At some point, you just have to decide what what burden you have right.

A

I! Guess you can't with them, because this is what I guess only Nixon.

E

I mean if you, if you put them on both I guess like in your source, and they were disagreeing right. You know but disallowed null on the on the property, and then you put in a lot of mill on the getter.

H

E

Whatever then or the setter I mean, then we probably just error right I imagine will habitat.

D

I wouldn't be desirable to decide one place or the other here and then just say: that's the only way you can do it yeah.

A

I'll give you that, because I mean that seems more logical, he also don't want I. Think you know. Inconsistency in usages were all right. You want people to have one way of doing it.

G

And I was saying: I think if you do the property, then there will be some non-si sort of languages that wouldn't be able to properly express things. They've got more than just said, or different variation.

A

And we can always add to the attribute usage right. If you find another thing, we need to put it on there just edit there and how does that? Work with other properties, Nance.

E

You can you annotate the getter and setter for another problem. I.

D

Mean can you technically put attributes on yeah I believe so yeah.

G

Because we do that with.

E

I mean we do that in sorte sure I meant syntactically.

A

But, given that you say get and said, I mean you should be able to put stuff in front of it. I think.

D

I mean it's, it's just I think it's mostly just a design decision. That's on the one side, if you have a field for sure you have to put it on the field itself, you're not get to put it on the input or the output side of the field visit. There's no context in distinction, so we can either use the field analogy and say: property should be annotated at the same way. The fields are, let's change the field to a property.

D

You know the attribute states in the same place or you can say well, attributes properties have more specific places. They actually have a place where input and a place for output, and it should go there either principle makes sense to me and I'm I'm vaguely in favor of the analogy with with the fields, but no I, don't think I. Think.

A

That some people think about it I mean like the only thing is like do we ever want to have different different representation for input and output. I would hope not well.

D

They will, they will actually often differ. Input and output will differ in even in the core library yeah, but for properties. Yes, women.

I

Have cases when annotating like the simple mobile annotations, we could add question mark to the return type, but we didn't actually wanted to return that nor tell you or we didn't actually want to accept an old value. But we usually angle.

B

To return a null value.

I

But we didn't except no like.

B

A property that lazily initializes, if it's null and then a setter that allows null- and you know basically resets it yeah.

D

Yeah they actually go both different both ways, but there are also fields, but there's at least one field: a public mutable field somewhere I can't remember which one it is it that's a couple in strongbox, for instance, yeah.

B

Strongbox is the main one, but.

D

You have a field that differs on input and output as well, but this will come up various places.

G

What if we said you had both and be for the property form? It was only a lot for auto promise, in which case we would put it off the private field in the back infield that was emitted, I see no way.

G

J

G

For reflection or other.

D

I, don't know why you would only or I mean I get. You raise a question of what your auto properties do with their back. That's something we can.

D

We could do the right thing, I, don't think all the properties, it's sort of user facing level they don't differ from other property.

D

Hey I'm a little sick, so you say you have to put the method thing on in order to be able to use the accessors as target okay.

B

If you want to use yeah at least that's my experience, if you want it that the the accessors are considered methods as far as attribute usage goes it's unfortunate because.

D

Then the it will be overly lenient another that you would actually be allowed to put these on methods, even though makes no sense.

E

We could make that a compiler error.

G

All right, let's see sharp side, but that doesn't stop anyone from believing you could also just not have methods in compiler could special case. These attributes would put on X to know they go either on the return type of them together or on the value input.

G

B

B

Iii know I'm, the one who you know default is putting them on the accessories, but I would actually sort of lean with Mads and say we get rid of the message thing on the top four: it prohibits you from using it in places that don't make sense and the first to apply to the indirection. They always apply to the indirection there's really only regardless of language. There's really only one thing on properties that has an indirection and there's really only one thing in properties that has announced direction.

B

You know, there's there are other potential accessor, but neither of them we can just say these apply to getting set. If there's fear in a language, tough luck and then we make them as restrictive as possible and are forced to consistency and usage and.

E

K

Cannot we can always loosen this later? You can always add to it. Yeah the words name, I.

A

Said it means we move method from I, don't know.

D

And if you want to use them on properties, you have to use in the write down, probably right, I'll.

B

While you're at an email, please remove return value from the top to.

G

And so the compiler would know if on a property you put allowed on, it would go on together and if you put maybe no, it would go on the setter.

B

I

What why you solo multiple sentences.

A

I

B

I wasn't sure what to do there there somewhere, it was clear to me like it didn't make any sense and for the top four I wasn't sure so. I left it as basically true and figured we talked about it.

E

D

You don't have any parameters, so they can't be different. It.

B

Right it was, it was more a question of: do we I wasn't sure what our philosophy was if we explicitly prohibit such things or if we just treat it as okay? Well, there's multiple, but so what that's at least one. So it's.

G

The compiler going to error, if you try to put both, allow and disallow it all on the same.

E

I imagine we will I, don't think. We know how to interpret that. So.

A

They would have said what do you do when you implemented it or with that you're just saying: well, it's both it's a it's a quantum, though.

D

There's a language position there: we could make it always like we can make it like. We can deal with it, so we could apply the most restrictive or.

G

I would think the least restrictive in that that.

B

That does raise the question if they are usually.

F

B

Should they be the same attribute with a boolean, you know no allowed true, no allowed false yeah.

E

We did start there right, we started, we started, I have one with a with a boolean flag, and then we came to this yeah.

I

I think having one like a now allowing all attribute grew, allow no false yeah. You represent the intent hi I'm.

D

A little I get what you say, but there's say the problem here is that it sort of gives the illusion that there are only those two possibilities, but there's a third option which is leaving it off, which means neither of those two sometimes right, especially in a generic setting. That's allowing all true and I Arnold false and there's allowing all whatever it said. Whatever it, you know, it resolves to from the type.

D

So if you so, if you have the true and false it just sort of gives the illusion that it's just those are the only two options and you could almost make one of them, they are beautiful. We.

B

Could have allowed no that takes a nullable of bool and then you could have allowed no, no.

A

I would say just from a synthetic example and I would avoid having parameters, because if you actually put them on parameters, it will get very busy quickly, right, I! Think if they're just simply attributes just one name in brackets, I think that's much easier to read.

G

Yeah I do think it reads better when they've broken out, but I.

D

G

It's also smaller metadata because you don't have to oh the mall yeah code. The rule. That's wrong, bites every powder, so.

A

Okay, so we agree that alone, multiple make some loose ends in the first two days right: first, four and the first four, even okay, I think.

G

A

Think that's the default right. I.

G

Think a lot actually does make sense of the baby, though wins, because you might have both cool cases where and you can Howard I make them. You could have multiple conditions where it might be dull. Well, the.

B

Lien about the return value.

E

Okay, there's only two possibilities and if both of them are the case, then just used, maybe no yeah.

D

Maybe nellwyn false and maybe not one true. Okay, it's just baby sing, we're better off forcing you to go.

A

Though he wasn't asking if the chant should be I, think apparently in the salad notation of the native side, they have. You know it's all underscores all over the plate, so, but they all start with in or out to make it clear where, where they go, did we think about that at all? Let me not when everything for the first time also I had to read the comments to understand which one goes where wasn't super clear, just by the naming, which one is the opener, which one is in summer.

B

D

Are they yeah when you get to the conditional? There are actually.

E

D

I seen so, we would lose the name inconsistency if we did that, but.

A

We could have in order I didn't really care, but I guess at that point it will be very way to say no because you would have to well. It's been faithful to me um and my only concern is like I mean I wrote this in the in this issue and I said like I want to group them somehow, because compiler service already is pretty large in terms of API surface I. Think a prefix would just make the API it's really submitter long again and then same problem. It gets very busy when you have actually signatures.

A

I'm, not sure when well I mean in is at least short done. Still it's like additionals to be. You know two to three here: cuz you have to put everywhere.

A

Maybe it's just the first time he'd roll.

E

What would that be like in allow? No in or out, maybe not something.

A

Like that yeah, it also beats funny. You know, I could.

E

See like maybe now out or allow no end those redefines. It's.

C

A suffix yeah, that's fair, follow in or nephilim input gets longer.

D

C

D

And you have maybe null.

H

D

And then that be maybe Nolan or out when Brendan.

A

Hiep sets obviously don't include the underscores.

A

E

It's just: firstly, there are still only outputs.

D

Right there, the effort, not the advances, but maybe Nollan not null themselves- are only how good and.

E

No, not no one make no one or both on the other. No.

D

They also an encode because they can be applied. They they are post-conditions, so they talk about what is what is known after the method has been called right, but but they can either be. They can either be mutating and now producing an output and saying something about that output, or they can be testing an input and saying something about that input. So for.

B

Example, string dot is null or empty. Its argument is not null one false.

D

Yes, oh that one too, let's go on an in code, but it's still a post condition. It doesn't restrict the call. It just tells us something about what what's the case after they're, not about what true.

B

You know it's not no one fault. If.

E

Spring, that is another empty returns false! Oh, yes, no yep! Now.

D

Yeah I wish bullying only had one value so I.

A

Called this is Julian uh-huh, all right, so, first more than we're covered with that right.

D

Yes, so the so the third and fourth retain the air return value. Our game.

A

I think somebody I think Steven brought up there should. Maybe the cabbage will be me as a non-native speaker. I honestly did not get the differentiation. Little Steven wrote an email, I mean it's just I. Don't care like that. It's.

G

Two words versus won't work.

A

And I'm always confused wishes all right. I.

D

I would argue vaguely for keeping what's there because of the symmetry with not null, so you say the the the value is may be null or the value is not no. Those sentences both make sense. If the maybe is one word in.

G

D

Why is it? One of them has a verb and one doesn't if you may becomes a word in itself and that's a verb, but the element does not I would.

G

Also say, regardless of the semantics of the language that wasn't better, it.

A

Certainly looks nicer that.

D

One exira it might, then the other one should be called. May not be no don't thing. Yes,.

A

B

I mean it's, but the phrase they use the value is may be null news sort of a it's, not really English I mean it's, it is, but it's like more of a compiler speaker, whereas the value may be null is sort of the more. What someone you know would normally say if they were speaking English I understand.

D

That it's more the analogy, they'd be plugged into the same sentence in the same gap.

D

Faster we're speaking.

G

Sioux language.

A

A

Yeah, like I, mean I, don't anybody feel strongly like I? For me, it's either way. I have a slight bias towards this one. What's currently written, because it just looks nice well when we were designing.

E

The compiler API I had originally had it, as maybe like our public enums for the compiler, and we ended up going with this spelling because it reads better and it wasn't in the same place. I had.

B

No problem with this until I spelled it out in my head and then once I saw it with a capital. B I can't unsee it, but you can keep seeing it in your.

A

Video early for direct and so.

I

They allow multiple. We agreed that it should be false right. We shouldn't know: I want to pull on the first of all.

E

Five o'clock, somebody man so.

A

We should allow multiple with.

B

Shipping ship, no, it should be false. Yeah.

A

So but that's the default correct.

A

I just write it here supposed to be clear and then.

A

I mean whatever the default pauses the default all right, I'll leave it up.

A

B

I left off inherited here but inherited defaults to true I'm, assuming we won't that's hope to walk in general.

E

I would think so we'd bonnet.

A

And that's a default for annotations tool right if you want them differently, you override them or do that so.

D

That is that, for what does inherited mean actually well, if you.

B

Means whether if you look at a derived override or so, if you look at the override of the method, somebody that's applied to whether it's meant to be applied to that as well. So.

D

This is not just for classes, it's.

A

Actually, good point I, don't know it might actually only apply to the types I'm not sure to place the message, then you all write them right.

B

I thought it to be wrong: I mean.

A

Either way, let's say: I mean if it doesn't grow, then too bad there's nothing. We can do about it and if it does then I would say we probably want to be here anyway right. The dogs.

I

A

I

Become being very tight, the right classes and overriding members, okay,.

A

Okay, I think we want that.

D

Yeah I think it.

G

Will help you, we can also just write a test about they. Do we want that and then are.

D

There cases where the combination would you you've? Certainly you might get redundant ones. You know we have generic instantiations override I mean.

B

That the downside to inherit true, allow multiple false is I believe. If it arrived over, I tried to specify the attribute it wouldn't be able to again I think.

E

The other thing is: if it was inherited, we might run into scenarios where someone's trying to loosen an assertion. For example, someone has disallowed null on that on the top level one, but you would have an implementation and you want to allow no for that one, because you Andale it correctly. We might, we might run into a scenario where we have both don't know which one is the real one. I would actually probably.

D

Then suggest that we don't have a married and then people will have to specified manually we we sort of intended to have given the compiler that would actually regulate this, compare overrides and make sure that they were in accordance with what they all were overriding but not necessarily the same, but we probably won't get to implementing that. They.

A

Could be wrong but I thought and you if the attribute you can ask effectively what type they were applied.

A

Recently, actually, the data I think that both present present many foggy custom attitudes but I'm not sure all the competitors ever the symbol. Api has been I would say if you couldn't tell which time they were applied to. You would basically just say if any other attributes are applied on the on the inherited one. You basically ignore all the ones from the base. If none of them are one on the inherited wanted to take all the ones from the base. All.

D

Right, the problem is that absence of an attribute means something that you can't express with a nephew. There's.

A

None of them are present at all, says me, so.

D

You finally looked it.

B

Looks like I was, it looks like I was wrong. The even with allow multiple false looks like an overwrite or can specify the attribute again.

G

Yeah, so as long as you could actually detect that it was inherited or not, you would be able to act present.

D

But then, what's the point of inheriting it, if we want to detect, if we wanted to take the cases where it's not repeating and say we behave as if it wasn't inherited thrones ank bored over it, then at that point it's pointless.

G

To narrative, well, you could you could have the compiler detected another way such as like. If the entire type is annotated, then you can avoid omitting the attributes of the method again provided that they match of what their inherited. So, if you could use it, you could still detected the compiler while reducing metadata, which other is also hurting startup time.

A

But think but but if no novel attributes are present, if I basically put you back to, you have no annotations that everything would be but you're calling button people managers they oblivious right, but would you ever want that I, don't think you were doing okay I can see. We want to loosen it from not now to now or the other way around but like. Why would you ever go back to I, don't know well.

G

It's an existing libraries in here they a type so there's an oblivious library, that's referencing in all the where library, yes and they're overriding a tight consider normal where library, but they're still oblivious. If so, you don't want, you wouldn't have any attributes neck, but like I was saying the compiler could have an attribute on the type that says the type stole a little air and if it's aware, but there's no afternoon on the method, it could say.

D

Go look at the inherited state. I just think we're getting into something very complicated here.

G

D

I if we want the compiler to disregard canned heritance of the attributes this alone, then we shouldn't do the in-home well.

G

I think we do I think inherited. The attributes is worthwhile for the perspective of producing in hurting start-up time for nullable and your life. If that's what we want and we.

D

D

Then we need the attributes to be different things. We need to be able to express all the options for that review, not just with the absence of that reviews, and we need to the ability for when you do supply an attribute to contradicts a previous one. We need to compiler to be smart enough to replace the meaning and sort of.

G

So why do you want the attributes that annotate not the case?

G

hmm So so, given a type in a null oblivious library, there will be no nullable. Actually what so I released it, so the compiler can detect it. This.

D

Is not about null obliviousness. This is about I want to make it more permissive, overwrite I'm totally, even in total null aware, land I want to make it more permissive override right, so I want to no longer have the disallow null attribute on my generic brand I'm, a parameter that has a generic type. I can't could allow null, because that's truly I and my only option with the current set we're proposing areas believe it all. But if the attributes inherit, then that won't work either because I mean still inheriting the disallow.

D

That's why I'm saying then I don't be led to type the site attribute and then compiler needs to be smart about that. This seems like a lot of machinery. It also seems, like you.

B

Know it the developer, specifying the override already has to specify the annotations again. So why wouldn't they have to specify the attributes again and.

H

B

If all that wasn't true we're talking about the compiler here, and it doesn't care about the inherited property regardless, they can do whatever it wants right.

D

So I get your concern about the metadata. Bloat I think that these attributes are probably not what's going to be because the for metadata bloat, the the attributes generated by the compiler and much much more numerous and those are the ones those are where we need to be smart, be.

A

Hopefully those are well memorize. It would hopefully use them sparingly in your library. Hopefully then, on everywhere cuz then The Syndicate also saw quite a bit for, though.

B

Yeah I mean if you look at the draft PR I put up I'm gonna, wait my hands or something like 400 of these attributes applied, approximately 300 of which were does not return. So there aren't that many compared to the literally there's a nullable attribute being generated by the compiler right now on every single method. Right. Every.

L

B

Metric core loop, regardless of whether it was annotated with that with annotations or not yes,.

D

All right so conclusion not inherited yes, I'm northerner hood yep.

C

We wanna, we have plenty lot. Yes, we should finish the other times before all the time we've been here.

H

Neither question and them may be null well just the wind, the winds ones like our wheeze or special casing. Boolean return values. Yes,.

D

Yeah, the reason being that this this clicks in with the compilers flow analysis or the compiler both when it does a definite assignment analysis, but also the null analysis. It's able to have a maybe null when true and maybe know when false state, where, if you then branch on the result of the method, it can keep track of the different null States in the two branches is.

G

That also going to be usable by F sharp that doesn't use the iPad.

M

G

They return a couple basically with.

E

The pool of the result instead well, they return a tuple right and it's an option. So the option will be annotated. Breakfast they'll have the valid case in the valid case, and they won't have this problem at all. Okay,.

B

Eric's question there are just using Korolev as an example, we talked about this yesterday or the day before offline. There are a few cases in Cortland, for example, where we returned any new instead of a pool, and so the you know ends up being if foo equals um success or something like that, and so we were talked about well, you know, could we add instead of taking a pool here?

B

Could this be an object and have it be any constant value that you can specify an attribute, but it was the omit that president that's just not feasible for the compiler and I think if it becomes feasible, we could always add another constructor overloading that just takes an object in a new parameter or.

H

A new property and a new property I mean that the property.

B

Is be like we're.

H

Almost building that could have to add a whole new attribute. You.

B

Can do that or I mean or it could just be- maybe no one attribute I take some object in a return value object or something like that. Also I mean I added these. Let me run that note. I added these properties to the attributes, because it's sort of you know good style, hygiene bunion. It's not clear to me that they're actually useful.

A

Yeah, it's the compiler that garden says you should do that, because the assumption is that some people still call get custom attributes and that kind of actually instances of the attributed right. But anybody who reads metadata wouldn't not need them.

A

It's just that cold.

F

H

Just wanted to make sure I was understanding that correctly.

E

I mean since their attributes, if we wanted to add a return, valuable object, we wouldn't even like we wouldn't even really bloat the metadata. Like return value, ask returned by the object to a boolean yeah.

D

But they we probably have an option to expand in the future.

D

B

Fred I afraid I missed what you were saying. What would we change that wouldn't be a breaking change? You.

E

Could make return value that you can make together for that? Just cast the return value object to a boolean, and anybody who reads metadata would be would be broken because they are looking at. You know, they're, looking at the echo metadata, but.

B

No, but just but this property, as it currently exists would be. We would need a second property, no.

E

No, yes, yes, okay, right I'm, just saying it wouldn't be an auto problem. Yes right! Yes,.

A

E

We want to, we want to move on to the other.

D

Attributes so one thing to just to note about these two attributes, maybe Nolan and not know when is that they are actually.

D

Only two of the four possible combinations that are found in or left today, I was.

B

Actually wrong about that. Oh there's, three of the four combinations that are kind of cordless Wow.

D

Okay: okay: forget that we look at that. All four should be expressible, because symmetry elevators.

D

Oh never gets more complicated.

D

The knowledge nominal attribute, that's probably the most complicated one of them all. Yes and buddy. It turned out to be very useful, so this one says that the.

D

A given output is not null if a given input is not know, so it's it's describing the correlation between an input and an output. I think.

B

Just to know I think I got the attribute usage wrong in this one as well. I mean.

H

I

B

Method and property.

I

Yes, it's also relevant to parameter.

B

What's that it is so, for example, interlocked exchange or volatile right, they would go on. You would put not null if not null on the first. The first parameter specifying the second parameter, because it writes the second parameter to the first, and so, if the second is non null than the first wall, so become non null upon the other post condition. Yeah.

I

Make sense I see.

E

Okay, so if it's not on property, how does that work with indexers? What do where do you specify something for an indexer? Yes, I could have an indexer with parameter that I could say that the getter returns not null if the parameter to the indexer is not known.

D

If you can stand it, that's our go ahead. Focus.

B

I hadn't considered properties with or maybe I had, and that's why I have it on here. I'll give myself the benefit of the doubt, because I now play more what I think reality.

F

Was a copy-paste.

B

I hadn't considered indexers I, don't.

D

Know that you can specify a return target on that index. I don't see.

B

How you do I, don't believe you can just like you can't specify return on a regular property right.

A

um But then you would put it on the Robert here suppose yeah and just say, but.

D

Probably it's not properties, don't have parameters, so this one wouldn't apply their two names. Well, indexers do have why already Nexus do and the reason we go.

B

Now why would it apply to delegate.

N

You don't have that we'll get up there right.

G

Because it doesn't apply to a delegate itself, it applies to the return type or a parameter epidemic.

B

Right, the mr. method here is erroneous method shouldn't be there, that's what you were looking at. Basically,.

A

Its parameter property and return me. You then.

B

And property only if we care about it, which.

D

Yeah the index should place, it did probably punch, it seems like structurally possible but very unlikely and I think most people will use.

G

Index Oh mass's yeah.

E

But having an making a dictionary.

G

D

That returned you, one of the keys yes um and then.

N

Shouldn't it also so everywhere we have parameter shouldn't, have generic parameter.

N

B

I, don't know what that would mean. Can you give an example: I, don't.

N

Know what it means I'm, just looking at after the targets and generic parameter, is one of a.

D

Type parameter.

N

D

That mean type parameter. These can't go on type friend, yes,.

K

A

We even have a you, can't even express it into sure. Okay, yeah yeah.

J

You can put after your son, yeah I promise.

B

J

B

Put them on arguments, but you can put them on parameters like.

F

Improvement between the brick.

B

Okay, you can put them in the like in the specification of the type but like where I really wanted to use it with inside of generic arguments like action, open, brace or how open bracket you know attribute and if.

D

We start allowing that at some point, then we could get the next level of mileage out of these right.

C

A

This is a lot multiple true, because you can have different yeah.

B

This one we actually have cases where there are multiple, because.

A

You can specify different parameters right.

B

For example, delegate combined that takes a two delegates. This has to return not null, if not null, attributes on it, one for each parameter, because the return of delegate a combined will be non null if either the inputs was not known in.

A

The parameter name B now it seems like looking at your metal body, can be. How are you talking with exception to my money? I? Did it already all right which, by the way, you should never different attributes.

B

A

Because, if you're, otherwise, if you say get custom attributes you just basically just blows up by getting in the data.

B

What are you supposed to do? I just.

A

B

So that means parameter name here should be nullable straight yeah.

E

C

E

Be nullable string, but it should Chris fully not die. What.

B

Does that mean.

E

D

Means you will die later, we could the reference for there, the compiler when it sees that attribute it. It will blow up. A test. Pilot should complain that this thing is no. You.

A

Should probably convert notes to empty string formally or something yeah.

D

Then all the parameters that have empty name is for I mean it will very soon die. Thank you. This.

A

Particular thing I think Stephens asking in the general case. What do you do with them? I haven't.

B

Actually trusted if I write an attribute like this and I specified null and I'm in a novel of aware context, will the compiler complaint.

E

A

More to go alright, the never just school here, cuz I'm, looking at a different thing on my right, editing.

I

So do not know if not no attribute. If you have multiple on the same parameter it will. It is exclusive right like if one, if, if one value is not know, then you know that that guy is, you know if one of them is not around here. Yeah.

B

I'm gonna my can or.

I

Basically, okay, good, it makes sense to instead accept a parent string that you can specify multiple strings in the same attribute rather than having multiple attributes.

I

In the parameter, usability.

D

Wise that might be an advantage. I would worry about here. I would worry a little bit about the bloat, given that in a so rare now we would have to have a string array or something inside of the attribute instead of Christmas string, but that's the only it.

F

Seems rewrite it.

B

Is pretty where I think I I think that's the only case there? There are several cases where there are multiple, not null of not in allatra buttes, on various aspects of a method like the two I mentioned before interlock exchange and volatile I'm. Sorry interlocked exchange has not knoweth not null as on both the return and on the first argument, the first parameter but I think delegate combined is the only one in correlate that has to not know if not nulls, on the same, like both for unn return, for example on this.

D

D

All right, the last two have nothing to do with nobility per se. I mean.

C

F

Won't go fluorite.

N

F

N

Flow but that's not returned that should at least have attribute targets.

D

But method covers that right, don't know it's different together. What.

B

Would happen on the constructor? Do.

N

Kinds of people can.

B

Oh sorry, dot for it does not return. You said: yeah okay, I mean.

N

It sleep I mean we're here to support people's terrible ideas.

N

B

We need it on delicate thin as well. That's.

N

It that's a question. I was going to ask yes I! Think we do. What does the billikin crybaby here Seamus.

H

N

H

Only valid if delegated.

N

H

Alone instances.

N

Of this delegate can only bind to methods which our markets do not does not return. Well,.

J

Looks like we wouldn't check that, but we just win good.

N

Okay, that's I'm, saying it's like that. Now, that's that's a sure see sharpen core FX should agree on whether not that's a good idea or not, because one thing is, they cannot put it now and that's easy to choose to do it later in the min conjunction with the C sharp feature. It add that to that with it, I mean.

A

It seems a bit contrived, but.

K

I thought at least one person walking in there well.

N

I'd say that if we're not gonna I would say: if we're not gonna support it, the language then I would say, don't put it in that's what use targets. So it's clear, we don't support it and then, if that one person makes a compelling case, yeah.

A

I mean normally I would I said it sucks to version caller, but giving them. You know version everything at the same time like it seems very much so.

G

Do we consider throwing as part of does not return? Okay,.

A

There's anything that construct exceptions. Log exceptions terminates the process, anything.

G

Well, throwing is the exception because it goes back to the parent context. If there's a catch statement, no.

N

Return, this means that this would be the invocation of the method where you called it, and you will not reach that subscription. Yeah that is does not return, does not return means that you do not meet next instruction after some coffee, it.

K

Doesn't mean they'll continue.

N

Next on, they are continue next, the coaches only he would reach next instruction. The.

D

Language has a notion of the end point of a expression of statement. The end point of the call is not reachable well.

G

So what I was saying for on-air continue? Next is, if you have a method that smart does not return, but in BB you have on air continue next, it would still return from the method. No.

N

Of the the language makes it appear as if the next statement in reality you do not have BB, doesn't knock you over their nose lesson. huh Okay, doesn't have permanent control.

J

A

But but he's running again the normal exception case in VB. You would be able to reach the next one because you guys have a catch and then you go to the next day and.

J

Even basically, this attribute in English for honorary, be.

A

Right so, basically, you can intervene. Ship between fell fats and both an exemption right yeah, because telephone resume next is nothing that the core points will definitely terminate yeah.

D

So the only so VB will have if we do that start respecting every reason. Viva media will have to sort out its own rules. They may not be able to forget the benefit for those actually, because.

G

I was also wondering if this would be confusing that someone who sees does not return, and they think that throwing an exception does return in a way.

E

Sure I mean does, if people think of exceptions, not in any way that has ever been described, I got that well.

G

They're, putting people that use exceptions for control flow, but.

A

That's the right music for that by the way, but.

D

It still means that they do not execute a return statement or the implicit return at the end of it. I I mean, if there's a better, it's like. If you have a better name in mind in my, we can consider it I, don't.

G

Think I've got a better name, I was just it might be clearer if it was two separate ones for always throws verses, does not return. Anything I.

E

Don't I don't think I think there is an equal number of people who would be confused by those two. Those two as there would be people were confused by this I mean.

A

This is something where I would honestly say: the documentation or system describe the behavior, because no matter what name you pick there's no way we can describe all possible control flow exits in the name.

A

I think the only quit so then we okay with beat both of them I, suppose that they are right.

B

Does not return, if should be allowed multiple faults, I think.

A

The Richmond, the doesn't return, if you know multiple phones, yeah.

E

D

E

D

The same reason, and only allows you'll have two options: yeah, yes, and if you want both of them, there's the unconditional one and.

G

In C++, this is no return for always through.

E

If we had to, we just be explaining in 10 years why we have both of these things in music.

A

The answer is because there was a five minute review where we messed it up a.

B

K

Of returns, sorry.

B

For does not return, we made a constructor and Method. We've decided no on delegate.

E

Yes, I think we decided no for now with the potential that we would the compiler and for FX could coordinated supporting that way. Yes,.

A

So then, the only question that, having are many two minutes, it should be have a base type for all of them. The reason I suggest the basis of all of them is that when you go to MST and you can click on the base time- and we show you all of them neatly combined. If you don't hide it, then you basically have to have a documentation paper. Somebody created, you know something by hand. Basically, if.

B

We did I, don't think it should be for does not return. I.

A

Think that's very fair.

D

Yeah I agree with that. They, even though they're very useful for mobility they're, not about malleability in the same way as I did this area all.

E

Right I am also in support of this we're gonna fight. Yet for two minutes, one with the name should be thank.

A

You fine, do you have I, just I just hacked up this thing here. Don't like this name. Let me know.

G

And we maybe have a compiler, annotations and knowable annotations. That way you can have the does not return inherit from compiler animate. You have the room to extend in the future. Thank you.

A

Section one tube editing, yeah.

B

So it doesn't mean: aren't there already a bunch of attributes and system time, compiler services that are similar in nature? It does not return in that regard. Yeah.

A

Let me can I debase that that's under breaking chain.

D

We get, we would have to medic find the name Lansing.

B

Might yeah because I think my point is we should, if we're going to add, based I'm sure we should review it in the context of everything that would inherit it. Yeah.

D

Yeah, that seems like a battle for another day where we have more than a million well.

G

Then maybe maybe key he does not return in particular as inheriting from add for you, like the other ones, and put the rest of these is pretty notable yeah I.

D

Don't mind the novel annotation one, some are cheap, maybe it's a little. Maybe it's a little confusing, because the question marks are also called nullable annotations, but it's sort of like an unknowable I.

C

Would expect another will actually that you guys.

K

Have food also extent.

A

E

Only thing I'd be concerned about there is then, when you click on nullable, annotation, well, I, guess a lot of a lot of the annotations in insisting on time to time, services already or document to visit. You shouldn't ever take this because you shouldn't ever type the normal attribute right.

A

And if you're trying to comply.

D

On do it, but that would be a would having I think this based on, because you are deliberately looking for documentation, Thunder duping them for people to look for what they can right right. But the.

G

Null of the route knowable annotation is also good for that, like you would have knowable after view inaudible annotations at to do, and if you, if you said, give me all custom attributes that inherited from notable annotation and get knowable and all of these post three additions and therefore you would have all the context you eat for breakfast.

E

And can choose. The point is that you can't write nullable or taught, and that's that's the theory is that you can, when you're looking for this stuff, if you're looking to look through metadata, you can you know we can, we can put a link in the documentation, but if we're talking about a page of these are the things I can write, that's I think what we were talking about here. Well,.

A

I think that was inaudible like I think attribute is not a signatory think it is it's in the menu in the source one, but if you do reflection or if you look at things and I'll spy until the actually supported that gets visible to people read and that sends I, don't think it's bad if it extends that it just means. Yet you cannot turn in source code. If you try, you get a copepod I. Also don't like it. So.

G

For reflection and expression generation you would want to be able to find all of them.

D

Towards whether we optimize for the simple common case or for the advanced case, because the nullable attribute is you know, roughly two orders of magnitude more complex and these put together is how to understand it.

D

So that's so I'm just I'm inclined to keep that a little on the side. Sure.

G

If you feel strongly, we could do that start in the college. Revisit it later.

A

Okay, then there was the other question ahead. So.

N

The white services namespace, like that's a that's a name say so we generally tell customers, don't go there like we tell customers. It's like this is where we put things that the compiler needs as a function of its hosted. Job right here is different. This is one more saying. No, we want you to oh I, think that I, don't I, don't doing the classic like this is bad and I. Don't have a better suggestion.

A

G

Occur to me, annotations III system, annotations.

A

About system diagnosis, doc, contract I,.

A

Take that as a citizen.

K

A also headed the recording I.

A

Mean all seriousness it single Diagnostics dot. Something seems like most don't replace. Then I mean it's to be fair. There's enough, I think there's like what twelve classes or something I think they would be okay with like having their own thing. But it's a.

H

Way of grouping.

A

It is, it is another way of grouping in which I would be in favor of, and then you base it against your new role.

A

Then the inaudible one would not be there, but another one would go to compiler services because that's submitted by the compiler. You should never use it and then the only ones it would go into the namespace or the ones that we actually expect people to write in source, yeah, yeah, okay, so.

B

All of the caller attributes are in runtime compiler services. We think that was a mistake.

F

M

Yeah, that was.

A

But to be fair everything that brown, that the compiler has knowledge of we put in compiler services, I, don't think me necessarily always considered that there's not.

E

Many certain things like string.

F

J

That was I realize we've done that, for which ones, but.

A

We feel they are also very rare right. I mean they're, not use that much. These ones will be used a lot more right. So comparatively speaking well,.

G

Safe and other stuff in yeah.

A

But that one is still true, we don't want it to use it.

A

N

You know actually.

J

N

G

A pole for aggressive optimisation.

N

Pepa dental is in right, definitely yeah see those. So if I wanted to rationalize them being at a different name, space I would say it's because they took the the home and though the attributes that are there affect how the compiler he observably affects how the compiler mix code, whereas this just changes the Diagnostics I, guess.

E

One thing to consider with having them in this namespace is: how is the effect on say, intellisense, because now everything from compiler services to he, and so they more often.

D

It definitely feels nicer to say using system dot null ability, annotation.

A

And don't use notability because then you get the same discussion with the remaining two attributes again.

G

I think just something down here: you should I mean.

A

We could electrical systems, diagnosis, good, annotations, right.

E

That a name says we already have no again.

A

It will be appear to contract, so the contracts are like don't use dumb go here. The stuff is deprecated, go to the other side over there.

D

Yeah, okay, actually I mean.

D

If that's the case, you.

A

Look at the chat window well nobody's barking in the chat window, so the indicator and.

D

You know what it says: if the names is nothing about null ability, then the flow attributes might also belong that they also. What do they drive? Do? They belong in the Diagnostics namespace as well. Doesn't return yeah.

J

Cuz, that's again, it only affects Diagnostics. It does infect holtsville that.

D

Case they they could actually go in the same place. Yeah. What about coding notices is.

B

That way is that true chart doesn't want the compiler do dead code, elimination based on reachability I.

N

The will do we won't do that based on the attribute, because we can't stop that Sharpe from doing.

N

We want, if you want the compiler to have if the intent of that is to change how the compiler, in its code, I, think that we would have to do something stronger than an attribute or we'd have to like that did not meet what my understanding the attribute so because we would have like there's. That is it's unavoidable.

B

Are going to Co gen around a call to it does not return? That said it like what happens if you, if you have something after is that you know it doesn't return out of the method or something the.

N

Way the only thing that I my understanding of that attribute what I envisioned it was that that affects on gullibility analysis, meaning that for the purpose of flow analysis, we pretend things like throw helpers. Bah blah does not actually return, and so we consider everything knowable beyond that. We would not change like region right for reach ability. So.

B

Okay, that's different in my mind what I thought the surgery was. I thought this was solving the general issue of wanting I mean I can be wrong. That's what it! What were you.

N

Thinking, how did you do it as a an attribute is tricky to do that? What you want like we needs like it's something like we can't. My fear of using an action before you're talking about Stephen is like at like F sharp, can add that attribute anywhere it it's like now. What are we doing? Well,.

G

I mean even if you did something else, you're still going to be and may they can attribute under the covers right, but when but I'm sure could still go and do whatever at one. So in El.

N

Reno you're presuming an implementation, don't do that I can mob wrap the out of it and after it doesn't respect water. No, they do because we mod wreck so much crap and switch everything to do. They respect my rights now, okay, specifically because we started telling them you have to, because this is dangerous as I'm saying like we can, we can do what we, if we, if we actually want.

N

We have options if we want to do the thing that Steve said, but I would think just a simple attributes, not enough, because we like the way we do read like dead code and stuff like we, the compiler alternates between, for instance, like if you do, while true like. If you ever return, you can have return value in c-sharp and do-while. True, there's no RET instruction and I'll like we.

N

We know because the CLR we there's an overlap between like CLR does not return an hour analysis, and we know, for instance, that while true never returns, therefore, you don't need a dot read, whereas if we called another method, that head does not return like what do we do? Do we just say well we'll put like a dot throw no after that, like? How do we react in the case where it does return, because.

A

It's illegal I'll fix it, it would've failed. We.

N

Ever do something.

A

Like if you have.

N

A method by that has a return value and something which just does not return. What do we do? We have to admit some code there to have a legal eye out and that's why we've never like that's the trick of this feature is that you do that and that's why I thought that the attribute was just affecting all ability, analysis, I'm, not saying you can't do the full feature, but that was a different feature than I was imagining when I read that dog.

N

D

The the full teacher is not one that we can be doing in c-sharp anytime thing right.

G

Is it something that you could determine how to omit the absolute in c-sharp, a type Raven, but not a compiler to do it yet that way, if we decide to support it in the future, we don't have to create yet another attribute in order.

E

To make it work well, the problem with system there's multiple problems there with that, but that the main thing is that we would have to create some new opt-in system for this thing, because upgrading to c-sharp 9 shouldn't suddenly start warning you about unreachable code, because we updated our unreachable code, analysis and suddenly half your method bodies aren't.

E

A

Let's, let's make a table, it's discussion a little bit because I mean the thing that we just started was just the namespace choice right so like, regardless of what Sichuan does I think the namespace choice should be somewhat orthogonal to it, because well.

B

Evident but I think there are two things are one that if it's Diagnostics and the attribute is affecting reach ability. That's no longer diagnostics and two I made a comment earlier about the attribute not deriving from nullable annotation. If it's only affecting null ability than I think it has to derive from nullable annotation.

A

So let me just point out that nobody showed up yet so I'm mildly, confident that nobody will kick us out in the next twenty minutes either. So maybe just close on that- let's just try to close on this, then so like so. First of all, what do people think of the system? Diagnostics? Our code knows his name space because it already exists and has two things in it: it's not like a super used name space.

E

Feels too tied to the name that we chose for our brothel namespace? Why.

A

Would that matter.

E

Because that's a detail of the implementation: I guess it doesn't we toda now the name code analysis is a compiler that happens to compile C sharp because.

A

This namespace, it existed like sins v1 on the can work right so like doing.

B

There's there's two attributes there in core: there's exclude from code coverage and suppress message yet.

E

I heard that innocent it was to do.

A

um I mean they are I, mean the grand scheme of things, that's what they are at their base. It is support for Diagnostics at compile-time right, so that seems largely in line with what their two kids are used for.

A

And yeah there's only two things in it, so if you are using to that one, nobody will say: oh, my god, she just backed in my intelligence and hundreds of types that I should never use. So that seems like a good choice.

B

It's fine to me, yeah.

A

And then so, if we have that nice face to be still thinking, we need the attribute the the the base top because quite thing at that point, I'm fine, with not having the base up at all. Like you.

E

Can just click on the namespace and then you see it.

C

E

That will also get to the advantage, because we don't have to worry about whether or not does not return to have the same base type. This is just in this namespace right, yeah,.

B

Although I do wonder if we need to make a different names or does not return, I mean it's I had a very different picture in my mind of what it is and I get that we can't make that work procedure our base, but I feel like that's. Taking the good name.

G

I think there's gonna be a lot of I, don't want it extended if they're not going to stop filing issues and if language ever wants to do it, I don't think we want to have two conflicting names for what is effectively does not return and does not return, but does billable and analysis.

A

But well the only other thing what majority' did sounded like whatever the feature would be would not necessarily use just an attribute. Is that fair, yeah.

D

I think that I mean one of the proposals for the getting the actual behavior around non returning methods is to have a new return type, never type, essentially never time. It's like the void is a change.

F

Like boy exhale.

D

But it also doesn't give you a result back, but it even doesn't even return to you.

D

A

D

Then we can officially, then we can encode that, whichever way we like but there's no use of facing attributed right, you pick up if you think a some syntax solution.

N

When he said we did this Midori, it was great and the future was fantastic, but we use that abuse because, but all we cared about was.

D

Which also better judge with the philosophy to not have too much semantics like magic impact with attribute? Yes, one thing is to affect the Diagnostics, but now there is to change culture, yep.

A

Marek, you were about to say something: yeah.

L

I'm sure it was still something month about the naming, because I can definitely imagine, for example, all the cases we have semi. We generate code that as soon as the guys, you see the attributes, they were kind of user attribute and just generated everywhere. Simply because the name is so nice, it might not be exactly what the sous-chef will use it for what kind of Philby just abused, because it says in the framework.

L

A

I mean you guys will never use it outside of Nunavut all right. There there's no other co-pastor checks for the attribute of today. That is not related to not.

C

A

Because I could see people just trying to like get the control flow. All of that right, like I, mean that's, for example, like was this thing that always wanted to this I forgot like.

A

I can think.

F

Of it well I'm in.

G

Like the Roscommon cases, you have an attack that throws and you might have a helper method. And today, if you throw in line the compiler uses that for reach ability and initialization of variables below right, but you might have to do an explicit else statement or something else if you use a helper method. So the thing compiler understand that today, yeah.

A

And it was back to Jarrod's point I'd like it's, not it wouldn't be. Legal in I are compatible to decide to emedco differently to make that correct code.

D

If I get what you're saying with a name, he says steal a little, it does feel absurd. To be I mean we.

N

Kind of wanted, worse names. Actually we are the absurdity of we're in a name debate and we can't agree because we all agree. We have the right name and.

F

The maybe to find the wrong man right. It just feels.

D

Absurd but I do get. The point. Does.

A

Not return, but it's not the way to work it should be. You seems like a good name for me. I.

E

G

E

A

What actually that really.

N

Actually, another thing we could do is the compiler doesn't care. If stuff is public or private, not in L, don't grunt, hmm sorry, I'm, making internal I have put it on there. We can do the Mel ability analysis and do the real feature later. No.

A

That seems like word for that in starts of age I'm, not saying it meant, then we all thinking it I.

B

Think the word is helpful.

G

And if it does that using a keyword like never uh-huh and couldn't you differentiate between does not return these for no ability and does not return using the keyword syntax, whether he's not the mod record, so we could just use the name and extend it later. No, it's just all mob brats.

N

Do if mod rec say give me a tight name, you can put this tight name in the namespace and if we use this, if we, if we use a syntax solution, we use my breath on top of that, so we could use the name. I mean honestly I find it hard to believe that we would ever do the feature that steamin wants without a syntax solution, because it is so like it's like change the way to generate.

E

Io, but you know very.

N

Well, Neil.

E

Would I don't have an aneurysm suggested changing the way we generate al based on like didn't code flow based on an app should be running now we know how to get him.

D

So then the remaining, if that's true, I, think that's true- that we wouldn't do the feature you you think this is with an attribute yeah, but then the only remaining concern is people might think we did and they see.

H

D

They might think it does more expect it to do more than it does, and it.

G

Will make some tooling difficult because they'll have to know to look for regular, attribute versus attribute with BA, but I think that's no different than other attributes really.

E

And I think I think they'll learn it I think we'll learn that we didn't and she be now. He had didn't add the feature that we thought. I wanted pretty quickly right like right, I'll say: oh they'll have.

G

To return something, maybe would we doesn't blog post detailing these after the we can explicitly list like hey, this doesn't work where.

N

You think it does actually no it's the best to know you Stella what it works and when people complain and then c-sharp not even fix a customer.

D

That's true: that's how we as that's, how we see the desire for the next feature, you're.

B

Like that, the Dilbert cartoon about running yourself a minivan this afternoon, yes.

C

You know they're.

B

Like the boss offers to pay like 20 bucks for every bug fixed and then the one of the engineers as though I'm gonna write myself a minivan this afternoon.

E

A

Stephen, what what.

E

A

Did we annotate with these attributes it's basically a certain belfast and what else wit.

B

Does not return yeah, so the 300 something methods in corliss that throw like regardless throwing and other throw methods they all have does not return. Fail-Fast has does not return. Debug assert contract assert have does not return if on the UH Napoleon.

B

Tryna, remember if there were others is.

A

There against Americans those are whole well, do they actually worked in the completa still doesn't quite know emit, so it doesn't actually execute past. This point I mean.

B

It's no worse than it is today, where it doesn't. The compiler doesn't know that environment failed fast, never returned, so you have to have a throw null or something after it if it is necessary, for you know to get the compiler to actually compile and.

N

Also, you do realize, like the feature you're asking for, would mean that we would introduce new warnings for non volkov, because exactly that, if we said that we know environment fail-fast doesn't return, then we've saved the throw after it is unreachable.

N

It's like I guess.

A

My question is like, like these to achieve is highlighting how many one is to death fix, because it sounds like we're kind of in this half pregnant state. Every basically say: oh.

B

A

B

For null ability, they fix. Like half of the warning half of the exclamation points we have in core liver because of does not return.

A

B

So because, because the common pattern, it's it's a little, it's really surprising but, like let's say you have a method that has a non knowable string as an argument. Well, we can't be sure that someone's using a null ability feature it isn't passing. No, so we have to say if string equals, no and we've been say, throw help or throw the compiler helpfully says. Oh this does not know, but then you checked it for no, so it might actually be no and then you're not provably. Doing anything.

B

That's going to exit so I have to assume that all subsequent code might see a null here and therefore it ends up warning anytime. We use the thing that we just checked for now that it might be now I see.

A

So, okay, the feature is actually oh: okay, yeah.

B

L

But Steve you you, you have this to help to work around the Steve sharp kind of co-chair limitation. So if you could undo that and rely on the C sharp to generate better so I guess you would love me that sure.

B

But the the code isn't just like when the code you're talking about is all the code associated with constructing you know: loading the resources, loading, the string loading. The argument Spock, like that's a huge amount of code, often to generate these exceptions, that we throw that all has to be separated out so.

D

B

The c-sharp compiler could, you know, do the opposite of in lining with with linux every time it sees throw new and take everything and put it into a separate method, I'm sure, but we're not at that point.

A

All right so I think like as far as I'm concerned, I I see the problem with the name. People will probably try it, they will get disappointed and life will go on. I, think that's I think the I don't know an alternative. Obviously, to that we can try to name these guys, better, more scope towards knowledge, but, like I've seen.

A

D

I'm with you, yeah I,.

E

Like the name city, all.

A

Right so then, let's just say this is it I will paste this guy in as far as I have it right now, and then you can all double check that what I have here is actually correct. Oh that's, beautiful.

A

C

A

Was actually nice and my editor good.

J

Does not return Oh.

E

No, it's just not on a new one like everyone now, no.

F

It's not alphabetical correct.

L

And actually it just occurred to me still did you consider extending the method improve actually we'll? Basically, the one which has the inlining towards I have like no return. It's basically enough right.

A

Nothing, the compiler sees them necessarily right.

A

The question with method imple is: we usually be stripped method imple entirely from reference, and it's because they are not meant to be for consumers when the name suggests it's an implementation. Detail like, as is the runtime call, it was not a runtime called. They want to inland this or know.

E

Where, as cortex is doing exactly the opposite, with these right here, yeah putting them on the reference assemblies only and not putting them in in the final assemblies actually.

A

We put those who'll in there I'm not going.

E

C

E

Assembly, that's.

C

What we actually wrote.

E

You're saying you're working a foot novel attribute only there.

G

Was talk about stripping them, but that might hurt certain scenarios and David Wright has been working on the kooky hash filter to try and help with startup yeah. So I you had mentioned at one point: I thought that was I mean.

B

Basically may I be concerned with putting this and Method info options if, for no other reason and everything in method info options is about runtime behavior and everything we're talking about here is compiler.

G

Well, I mean if the.

N

B

Features and what we're talking about here right, yeah,.

A

A

A little feelings of the matter and pull options. We should be careful because you're running on a bit there, eventually, oh it seems everyone up there. We.

N

Can make it work? Look, we didn't make the see chef reach I'm, not worried about that. Then we can make it work by using a mod rack on that afternoon.

A

I think Marik was asking that, instead of having the attributes, we would just have them as nothing yeah.

D

But it sounds like there's a mismatch between what the intention of method in follows: amount, whatever, oh and this one here, this one is yeah specifically. Therefore, the purpose of being in may today that so others can pick it up and get end unlikely.

E

Entirely compile-time that's supposed to entirely one time.

A

All right, then, I will comment on this. One I can find my mouse again.

A

And we consider this guy approve of this awesome. You guys needed in is already in this. Is the I'm progress right now? It doesn't.

G

Seem it already have a PR up, hey.

B

For ya, so I'm not entirely sure. We want to do, though, because I'm pretty sure that there's going to be analysis, that the compiler does that creates warnings that I haven't either thought about or flagged or fixed. So if I were to merge this, and then we took the compiler update, I am expecting there's gonna, be a reasonable amount of fix-up work to be done. Yeah so I'm wondering if I should I've got the draft PR I'm wondering if I should just sit on it until we take the compiler update, I, think.

E

The compiler doesn't need we all we need is the shape of this API to test it right. Yeah, hey Stu, welcome.

N

I think, if you guys have the shape the API, then we can tighten place that into our code and start going and then I sent you an email, Steve you're on the issue where I have a schedule. Wait out. Okay, I'll take a look yeah, but it's what the goal is to get a compiler done by 610, as that should hopefully give three weeks or core effects that come adopt a nice work please, before.

M

We can know that we can update the compiler now we're.

I

Kind of like 30, something that's always changing,.

M

I texted, you of learn just the one fixed cost Oh tests not to run anymore because he put users bed I'll wrap, but we fix that. But that's.

N

Not a breaking cake, that's just a bug. Yeah, okay, breaking changes. You guys did something intentionally.

M

N

The thing right, but did you guys because we hit the bug we thought that's the same, but we just fixing course you'll are okay.

M

I, don't know about that I'm. Sorry I thought you were on livid O's. You.

G

Mean Neos bug, yeah I thought his that bit burnt. Yet it wasn't worse this morning he's working on a fixed for right.

N

G

Is up and stop? It comes to three or four days, and you.

N

Send me to live to them when they took something note that here is right now yeah, that's like we need to say that I was so thinking, that's the same one. If not, then we need to look at it hit the exact same time: I'm creature, okay,.

B

Regardless, though, went on 6/10 the last features get merged for the compiler. We should be able to within a day, I assume get a version of that into core FXR intercourse CLR rather.

G

Right we could get as soon as the bill for Rossum gets up. I could fast-track the change into build tools. Oh.

I

Yeah, of course, this.

G

Course you'll: are you yeah.

B

Because I'm assuming there's gonna, be a little bit of back and forth on this to kind of nail things down the way we all expect them to be so the sooner we can get that the better right and then I'll give a slight thing. As I was saying, you know two weeks or something to work through probably do another. You know one or two rounds of find issues fix the compiler find issues fixed to compiler.

B

And then call it quits.

F

Garrett won't be around for those bz2 didn't yeah.

G

That's what hot Victor yeah Bucky were. You guys were getting a good core FX stack Alec for spam. Oh no! It's something good for invalid.

M

Broken exception: bed I'll on my mapping on system to have it open system, io enumeration, filesystem.

N

You can send us the bad I, always.

A

It's not better l, it's just not great I owe everyone mindset. Man there.

B

G

Yeah, you think so well, because the the one method- Lily that's interesting- has a stack, alec local bands. Oh, is.

E

It just span to fight equal or something equals.

G

M

Why I thought of.

G

Sanity tenet equals.

E

G

E

That's the bug yeah, who.

M

Will have a building is the motion of no.

G

G

I'd had the equivalent yeah, but it hasn't been touched in four days. It's got three side officer.

J

J

He's working on the stock.

D

The game during the.

A

All right, then, I think we are done right, anything else to add otherwise. Does it.