.Net Foundation Design Reviews, 24 May 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: .NET Design Review: Nullable Annotations

Description

We're reviewing nullable annotations for System.Runtime and System.Runtime.Extensions.

A

All righty, so how do you want to do this? Do you want to just go top to bottom I? Guess, that's probably easiest way for us to do right, yep, yeah, I think that's so tough, well an opportunity to review everything, because everything that is currently not marked at the question mark it's implicitly not now it just realize it would have been really helpful if my tool were to find out the reference typing. I think these are all the types of they're fine.

A

Greek civilization influence trimming context of both.

B

Yeah, are you sure.

A

In context, this is right, yeah, realization,.

A

That's correct: it's a glass answer, class yeah, but both here to say that you've any about this one just curve to the right.

C

Anything how you're still looking at the first one yeah.

A

I don't know, there's no different.

D

The question about show business.

E

Oh, it is showing the question marking green.

F

But it should be like the boxers notice, these guys.

A

A

Maybe who's already was marked like that. I, don't think.

A

Yeah, none of them inaudible, because we can't.

A

And so this guy.

A

Yeah totally furtive say that everything that is arks.

A

For example, but we are also going to be similar right equation, my fishing line.

A

And I think both Byner and culture info are basically effectively Oh.

E

Great instance is a weak one that it can return all right. You can post new as the inner exception.

A

And it's just walk this on precedent to the return mages, and so what does it mean affiliates and returns now it.

E

Was for a millivolt I bribed see like if you pass type of tune a.

G

Lot of these things are these like corner cases that force us to put question marks like if you pass a nullable in to create instance, you know what you're know yeah it's this is like. I was disagreeing on this one for a little thing,.

G

But I mean like if you're you know, if you're in a generic method- and you have no a generic T, it might be a nullable of the intonation.

C

G

C

Is a special case because the struck in generally cannot be knew about digital struck. This kind of to me is apparently.

G

If it's sure, like this feature in general, is about helping you catch, no reference exceptions, and so it's the corner cases that you're gonna miss yeah.

C

But the int question Marcus doesn't feel like you ever gonna end up with new reference exception, but I mean.

G

If you try to dereference the result of great instance, you.

A

Know or I guess, yeah I guess it depends on how you call it right. If you know the types in all, it's a reference type, you just suppress it right. Yeah I know better. Okay. If the type is an argument to your method, then probably you shouldn't it.

G

Really depends what we're trying to achieve with this whole feature. I mean if it's like you know, hand, wavy, we're gonna, we're gonna, tell you what you offer the 80% common case of inputs. What the 80% common case of outputs will be so.

C

Like honestly, like I, I think it would be better to actually specialize this specific thing in the compiler instead of nautical, but I bring in special cases like basically like.

H

Remarkable but.

C

Like basically tell the compiler somehow actually activate or something that this specific thing is special and like, but what you want to compile to do you want in vitamin C, they're trained men, you never know know so, like generally like, like when someone checks in the intellisense. That would show us no Morrible but like with this one case where someone is not movable of something, then I would actually make the compiler to treat it as something this can actually do. So what do you do if.

G

You were in a generic man that took a tee yeah and you.

A

Did type of tea yeah no I, think honestly it so the way I've seen this right. So let's say I call create instance right: mind you and mining Eve understanding is there's never returns, not because it will just blow up with the type dozen exits right, so I know use this thing like a lot of gives me squiggles I'm, like what and then I read the documentation and says, give me another scope of another reference type. Then I'm like well in my case I know for a fact: it's never the case. That is another bow.

A

Sorry. Another billable.

H

Know is that and also I can just bang it.

A

Away or debugger sorted or whatever it is right, but I think the problem is, if you do more special handling, then I think the question is just what you want to do right. I mean the case is that the compiler can do to mystical II, look at your code and decide with certainty that this is not the case. It's probably not being very high for some of the methods and the other problem is that just makes the whole feature extremely complicated.

A

Right and now the question is now: how much special occasion came to compiler realistically do when we have? You know how many you guys did. We have any I, think twelve thousand or something like so like that's just justin runtime, let's assume on time. Extensions is open. Their object, control, isn't the struct.

F

Right she was number checking our become the otherwise have been show up like that now object humbly is not a script yeah! Fine! Looking you can thank me like founded, but I got.

C

A

The problem is these all be more in type, so it's basically logic leading remoting handling, its I think that's.

D

A

G

Just saving graces pretty much; no one uses this, so it doesn't like. You know that someone uses activator that create instance of T. Then this doesn't matter because there are no question marks there.

A

Yeah I think it depends on the context right. I mean, like I, think the all the ones that take tie but probably super general code. Will you literally have no idea what the T is right, in which case yeah you probably want inaudible I, don't.

E

I think we need to represent the intent. It sucks that we're gonna add a lot of warnings for the common usage. But you know basic.

C

Like almost all the use of use, action I have to put the bug in there like they do not see. Oh yeah aquarium too.

G

And I totally understand that perspective, but from my perspective, if we're gonna start waving our hands to say well, just this could be null, but in the common case it won't be so we're not gonna, put a question mark right that entirely deep values, the feature it may as well just take it off. Yes,.

A

I think also that it depends on the API, I think to string I felt more strongly towards not having a question, my purely because you call it over the place right, but these are some of the API. Is you call all over the place right so like, even if, even if you have to bang it away? Well, let's pull one of your more complicated components that do things like this, and so you probably want to think about that anyways. So I think! That's that's! Why we'll be okay with that I need to strain.

A

We convinced ourselves that, while to string as a topic is used to lower like the particular method, is probably not used that much all in the context where you would have to bring it away right so, but yeah we will have customer paper for those kind of things, but I think we largely have to document reality right.

G

You might have a question on the tool with for things that aren't highlighted in yellow green or red. What does that mean? Yeah.

A

I have no idea. You basically talked about the aggregate exception case right where the question mark here shows up is not yellow or perhaps context.

F

And North entirely sure what a lot of things about the lead on the exception case would actually remove them other than the same.

A

Variant, so basically, what this tool does it aligns? Api is based on logically the dollar fee, but will align them regardless of question mark or not question mark, because I don't change the idea of the API and then it will basically generate the tokens for left and right and then just dip them so that, logically, the question mark should have showed up as it did. I suspect. What handed up happening is that whatever module I have locally already must have had the question mark in it and non tell you why that is.

G

I mean we're looking at everything anyways, we kind of ignore the colors yeah. Well,.

A

It does help right because he's like oh there's, something that little linear that would have been useful but yeah yeah, but in.

E

Aggregate exceptions most.

A

Of the exception.

E

Array and he wasn't marketable.

A

I also like some of the stuff here, like I'm, very concerned right now. Likewise, this showing up as.

B

C

Anyway, there is something fishy going on, maybe compiler I thought they did it like indirectly, or something no.

A

The dish is literally over the tokens, I mean unless I mean I looked at this, like you know a lot for three years ago, but maybe Santi spear broke some invariant and the way and the way the different works I don't know, but we'll find out.

A

Yes, so the everything exception case I mean clearly, you would alone all right or the inter exception. Yeah. We don't actually.

G

A

For the whole array we all finished up now, none of the inner exceptions can be known, though I'm in the array itself. It.

I

Cannot be known, it's.

A

A pair of, though only to accept no but I guess yeah I mean.

G

If you, if you don't specify anything programs, you get an empty right.

A

Well, there must be a reason compiler to that originally I think they actually.

G

A

If you don't specify no.

G

It's always been an empty, but but what changed a couple releases ago is we use our raid on empty instead of new T sub 0 Oh Susan yeah.

C

The prettiest miss.

G

Raises an interesting question is: if we ever wanted to loosen a great exception, to allow inner exceptions to be now, would we consider that a breaking change to change the you know to add question marks? Presumably adding question marks to parameters would not be considered breaking change.

A

I guess it depends, but if it's a virtual in my favorite.

J

Yeah, it was virtually and wouldn't be yeah.

A

But I think we generally said like, depending on the case, you might actually accepted it. You know base and customer feedback right, so I think eventually, I would get to the point where we have high confidence in our annotations and don't turn them anymore, but until we have someone one might achieve over have to tweak them.

H

Do you still need to check for null? Even if you don't have the question mark on there, because if somebody else is calling you that's not using the latest compiler yep.

G

Yeah yep, so we didn't, we added these annotations or didn't. Add annotations. We didn't change any of the exception handling code, you, you know it was throwing before it still throwing, but.

H

I mean: did anybody writing a new library gun well.

G

It's up to them.

E

You kind of still doesn't bow no with using a bank to shut up the compiler I. Given it the consuming. Api says it is not nullable. You can call that API quit no bank and the compiler. We will just ignore that I mean.

H

I mean in that case, like virtuals, it really wouldn't be a breaking change right, because somebody still could have passed it null well right. The question was changing an input parameter from it.

G

Was just worse breaking change the if someone had written source that overrode a virtual with a plan and their parameter was not knowable, and then we made the base prouder nullable, the compiler when compiling the drive source would warn it.

H

Wouldn't be there wouldn't be a binary pretty no.

A

I mean none of these are fighting breaking changes in practice, but, like the the thing is, though, you know beef, you know before this feature. We couldn't express malleability right, so I'm 100% certain we someone broken invariant right. We just said: well, it's not represented in the type. So it's fine right, like it's more like this gray area, behavior changes that may or may not affect you, and so the way I see it is that their notations just expressed that a little bit better. But in the end of the day, is that the same discussion.

A

A

And it seems all fine to me so far, not seeing anything that is controversial.

A

What does it mean if an app contents, the base director is now.

G

So there was one case: I'm gonna have to go. Look at it now.

G

I think it was when it's hosted or something it's possible.

K

G

Appears reported.

K

By the host so there's almost report announcement.

J

Sure it comes from a deaf country, but.

A

They are practically speaking. You have the base directory somewhere right, it's just like, what's the whole sedated to compute it or there's an actual scenario, but not make sense, but.

H

It's not necessarily the dotnet Exe host right, any other. Anybody else could have hosted core CLR for when they spun it up. They passed in doll for the base directory yeah.

A

But the question.

L

Is meant one incinerator.

A

G

C

Is the problem.

G

Is like we, so this calls app domain based directory, which ends up saying. What is my entry assembly get its location and use that and that's tall? Then it returns null and tree assembly, but might be null if you were hosted.

A

Oh so you're not calling actually the host you're, basically just saying well, if it's hosted, then I just don't have an entry assembly. So there's nothing. I can do that's the current implementation, at least so. In order for us to fix that, you would have to have some host extension point where the host came to turn the base directory.

H

But it's like it's it's any C++ code, that's loading! Of course you are and insert, and then they can start calling any method right. They don't have to call a main method.

A

Right right, which is why you would have to say, like the host, would have to detective the C++ code, activating that would have to pass in the data somehow if they wanted to have that thing or return. Now there.

G

Is a way house, so it looks like there's a some app context based directory that can be said least. According to this comment and that'll be used. If a host sets it if a host doesn't set it, then we hit this get entry assembly thing. So, basically, if the host doesn't set it and you're hosted you get no.

A

No target favorite name same logic right because we've inferred from the attribute that's applied to the entry assembly. So if you neither have the attribute or you don't have an entry sembly, we just don't know and.

G

This is the thing like right now, these this kind of gets to Christoph's point, which is the 99% use case for this. It's not going to be no, but it could be known so we attributed as such.

A

I think this kind of goes like if you'd see this first pass more as an you know, we actually represent. What's going on and I think the other question is like: should this ever be done right or should we just fix the code that this corner case just can't occur anymore, because I mean like how much code have we written? That assumes based directly to actually be there? Oh, it.

H

Just returned empty string; instead, no I'm, saying in this case you could just return empty string. Knowledge.

J

Well, that would.

A

J

Thing with your own yeah.

H

But what should it return in the scenario we just said well,.

A

That's what I'm saying I think I think like if you look at the API right now we're just document what actually happens right but but I think. The other question we should ask ourselves is like: should we just fix the corner case, so you always have a valid based a rectory, because practically speaking, you have it. You just can't compute it which is more like if you think of it as a lacking feature, or you know a bargain or extension I.

H

Don't care what it in that scenario, what what should be the base directory so uninterested actual value? Basically,.

A

We would say the contract is the host has to settle on the host has to decide whether it's what it's based or vector is so that apps have the same way to deal with that. What, if you like some fun where you like, so it's hosting the browser right.

D

A

Which case you would say, the browser I mean that's more like the expression for the file system in the browser, but if they have one you could assume you could basically say well hole use base directory, usually by for loading other stuff, but you would say well either you can't load any other stuff, in which case the thing they were matters where you say well, it's whatever. The logical route is for your file system inside of a browser which might be somewhat arbitrary. You know way to live. It I think.

G

What I hear you saying is this is right for now, but in the future we might choose to remove the question mark and then say: if the host does not set almost much says set this app context switch yeah. If it doesn't and there's no engine assembly, we will throw yes, it's.

A

Basically more like so, basically no, we document the actual behavior and then I think. The other question is: do we like the behavior, but that's that's a very different discussion. Yes,.

A

I mean getdata seems reasonable if the key doesn't exist, don't have data for them. Okay,.

H

Target framework claim B know the.

A

Same thing, like you, don't have an entry in assembly, in which case we don't know what. What what to look for, but even if you have, while you may not have the attribute applied.

C

The citizens and they inspire go so what.

G

There's no return null otherwise get its target framework attribute yeah. If it's null return, though otherwise return its brain were thing. So, there's like all these steps.

A

Yeah I think for for hosted apps billion all the tired fabric name because the holes is activated in the target family, but for for exes yeah you kind of screwed, because that's you know or not exits, but logical axes. Where the entry point is one run one time you're on.

G

There were a bunch of times when we were going through all these adding these annotations, where you know I see a PR like I think Miriam sent one and I was like. How could this possibly be? No, why are you making the snow and then yon chimed in well? It says in the docs right here this could be cobble yeah I've, never protected myself against that possibility. Yeah.

A

That's the thing right: I mean most people, don't with dogs, you only read them and you hit them towards the first. My raft right I know you read them the first time you needed one. So in a sense, Petry accomplished right. The operation in the conflict.

K

K

Yes, we're dining for Oregon Tech session.

A

Choose, except not everywhere, like a pair of name, is Kim Yoon dal. You have whatever.

E

We're injecting technology- if the message for example, is known, then we use I think we just returned the type of exception so.

K

Why they are calling the constructors that take message and passing month in the first place? Maybe they wanted to pass down the inner exception.

H

I mean if you have a default constructor right like it, should be logically, the same thing like the default constructor should just call the one brand name with null all right. That's logically, what you're doing it.

G

Does raise an interesting question, which is let's say we would say we have argument exception. You know, I, do kind of I think it was Tareq's point like well. If you don't have a message in an ER exception, you should be calling the shortest possible constructor. So would we want to push people towards that by even though these technically can take no pretending they can't? The flip side is, if you do that, people that are passing know are gonna get four names or people that pass things when the compiler can't tell are gonna get warnings.

J

The parlament, when you do that, is that basically.

A

You enter in overloads that I look exactly like this one. We don't have to change any more. You basically have you know incompatible shapes if you've, all right, where eventually you run into problems where adding an overload gets really complicated, because the question is now, what will all existing code compile or I'll be accidentally binding against something? So basic here when you say argument null, except you an exception, you pass not common are already not great right. I mean for me for.

C

Me, the point would be like 20 or direct for this class and you don't uh kamon it personal or not, and then you just run out whatever.

A

Yeah I mean elbow: writing is also a problem but or inheritance in general, but I think like like I've seen over those in the fabric that did what Stephen suggested and is usually a pain.

G

Now to clear, I wasn't suggesting or just highlighting the old yeah you.

A

I think those are all like if you think about it, they actually make sense this first time. It's like really me, except now there are names, but I guess parenting can always been done so don't but we'll be able to bother.

G

Some of them are a little interesting on, except like serialization in phone.

G

You know the I, don't even remember if we check whether it's an all new implementation or not, but we never expect it because it's always provided by infrastructure. So we just say this can't be no, even if technically something might allow it to be.

E

I, don't think we ever checked, we never say yes, we we could always no ref is.

M

F

For most reform.

K

I'm just wondering about the goal here: what like you know basically are: are we trying to reduce the warnings that we can come on the in every this nullable or just? We want to make sure that we're providing a strong feature that, like you know, people can spend some more time to enhance it. So.

G

I think the the end goal that everyone for this whole novel feature the end goal. That I think everyone would agree on is trying to reduce a particular class of exceptions and failures that people have in their applications and doing it via static analysis at compile time, and that means the compiler helps you to understand how nulls flow, through your code and, if you, and that means that any library code that might give you back something that's null- has to tell you that and anything that allows you to accept.

G

That can accept null has to tell you that the compiler factors, all that information in where we end up disagreeing and sometimes I, think I'm sort of at one end of the spectrum and mads is probably at the other end of the spectrum, is how quote quote accurate? Do our annotations need to be and I've erred on the side and I know, I've had disagreements with Santi and Christoph and others as well.

G

I've erred on the side of we should be as accurate as possible if it's possible that something could in any situation, hand you back null. We should say so. The flip side of that is, it means you're, probably going to get more warnings in your consuming codes than you. Otherwise would so. My goal hasn't been to reduce the warnings that people get in their code. It's if someone enables this feature, it's tell them every possible place where it's they might even you know, have a null issue, and that may be the right approach.

G

That may be the wrong approach. But to me it's about like it's been what we should be doing from our library perspective and what.

A

I like about it, though originally I was more on the mats, Kempe or saying you know, let's make it so that we don't basically drown the consumer in warning messages, because it kind of results in the FX code for now and people just turn it off entirely because they get frustrated with it. But what is nice about? It is that if you have earring on the a let's just document reality we will find API. Is that arguably have a you know a question of a contract? Right, like you know, or should does actually be now?

A

Should we make sure we never returned out I like which is kind of the base to back to the discussion right and if we, we basically limey, say yeah, you can pretend it's not. Now then you're effectively sweeping the problem under the rug, when in fact, you probably won't either either you want to say no, you have to handle all you want to say. Okay, let's make sure we actually don't produce enough.

G

Yeah and then yeah this there's an effort currently underway sort of a DCR to add a bunch of attributes that allow us to more finally Express the sort of know behavior of these methods and that's an effort to alright well, we've attributed we've annotated everything sort of a very conservative way. Can we then use these attributes to be more specific and eliminate warnings in cases where we can actually help the customer understand?

G

Well, even though we say this that you know this returns no you're going to be fine, given the inputs that you provide to the function right and so that sort of sets our expression. That's what it's about all right, we'll let let's help the customers safely drive down the number of warnings they received right so.

K

What I'm hearing here is like, then, we should do the right thing in our code, regardless of the warning, and we assume that these attributes is going well at some point produces warnings, but we had some cases the way I see- and this is my problem. What I mean like going to every case like this and then like? It's not there I say like.

G

We should be alright, I say in general. That is the approach that we've taken. There are a couple gray areas where we had to make some judgment calls so, for example, one of the biggest issues with this. This feature in general is strux, and you know if you say that if you have a reference field on a struct- and you say it's non nullable you're lying because someone can always walk up to any struct and say default of T, and then every reference field on there is going to be known.

G

And so, but is that really helpful or not so like with the race segment? Well, there are cases where you get default, created array segments, and so we do have a question mark on the array, an array segment.

G

But there's like this interface map struct in reflection, where no one ever creates one of those things using default interface, but they only ever get them unless they're doing something weird out of one of our reflection, api's, which will not return to fault initialize things, and so there there are a bunch of reference fields on that and we didn't make the we made those non nullable, and so you know there are these gray areas. 99% of stuff is very clear and there's this, like 1% gray area we've had to make some judgment, calls.

A

But but I wanna add one thing, sir I think the other thing to say is like if you look at existing code, that is already in production, I, don't think turning off. This turn on this feature will find a torn-up box. You probably find bugs no matter how you know robust Rico phases, but there will always be like some corner cases, but but there's a lot of like maybe somewhat false positives, because the compiler cannot actually understand your code at the level where it actually executes but I.

A

Think to me the big savings is imagine somebody writes a net new code. You don't have to waste a lot of time on reading the documentation, because most of this is not happening at compile time.

A

The compiler already tells you all you're, just dotting into something that could be now, but you'll probably find this bug immediately when you actually run it for the first time, but you don't have to do that anymore, but, like I, think there is some accumulative effect of productivity savings that just come from when you actually evolve in code or you're writing that new code. This is where this feature is I, think really shining but I think for existing code. I! Don't expect that this is going to make. You know your life substantially better right.

A

It will probably make your life somewhat worse in the beginning, because you have to accommodate the future, but then eventually, as you evolve in your co-pays, you actually have payoffs. So in other words, if I have an existing app I'd, barely touch I would never activate the feature. There's like very little gate, valuable to be to be gained by that. So.

K

This feature has to be enabled for the whole project: you're combining oh, no, you can do it by source file, which is basically what we did. So maybe this could be your guidelines about like telling like if you're writing you could enable this, but don't enable it for me. Well.

A

Today, so when we ship free, oh the idea is that it's it's it's never on. You have to always.

H

A

For five, oh there's a discussion on whether we should in the project template enable it by default, but it would still be if you have existing code, it would not be on. But if you do find your project, you get a part. It is already anything I expect.

K

At that time, I mean like Yui the new source of file. You wanna put it for don't enable form whatever existed before.

A

We could do that, but it's like you know, you have a you know: pound nullable, enable and every single source file that you create I'm, not sure. That's what you want.

K

Because, like you know, because.

M

It mean if you look at.

K

M

We are worried.

K

About like a lot of forums that people can get I mean. So if we say like yes, I mean enable just for the new code, and this would be fair about tonight. This warning in the new code, so you have the chance to yes, but.

C

Other a new fire to a project you already have a bunch of ideas, would like in your own project, which are not annotated, which I.

A

Think you're, a new source files I think just enable the newest. So as far as the video will not be very high, like the basically, the problem is, is Christoph said, but you have to annotate your existing code base to actually flow information as a meaningful way right. This interaction between the noose, waffle and the other and in the other system, I, would generally say to me when I say code, existing code I really refer to the project as a whole, and you should make this on a project-by-project decision me.

A

The pound nullable is only an intermediary step, as you are upgrading your project towards malleable enable, but the end result is always you turn on for the whole project. I, don't think many people will have this half stage where, but this is just selecting.

K

With what you said before about, like you, know, yeah, this is very useful for the new code and at the same time, we're saying no. We should enable it for the older code for be music or any project, so.

A

And what I'm saying is that if you have existing code like isn't it right for your project, yeah.

K

A

To make a decision, how often do you touch this? If the answer is very rarely I would not bother at all. If this is an if it is that your production app that you're actively evolving, then you know take a point in time where you migrate, this to be inaudible, do which file by file, so we actually drive down on the ones for your project and once you're done and able it for the project and remove all the palpable file across all your sauce.

A

But it's just a in order for you to actually make progress without immediately drowning in 10,000 warnings. That's what you do! It fall by file, but that is just the the process itself, like eventually your when you're done. You should be at the point where it's enable for the whole project. You should not have this as medicines.

K

In the middle, it.

A

K

Be a clear guidelines from yeah, but.

A

I think the other thing is like we're: the first project that is not hello world, but he actually tried this right. I think there wasn't folks, I've tried it, but they didn't have enough time to actually drive the annotations to all their co-pays I. Think, as we are finding more code bases where we do this, where everything a little bit more about. Okay, whatever works, what doesn't work but I think we already have a pretty good understanding of what worked for us all right. Let's keep doing, because we have only like 12,000.

C

Mp at first did they not take their only I see nothing. So what did you say so.

A

Rosen did so Jared had this goal that base today the Rosen team. They generally are very aggressively doctoring their own features, but the problem is that this is not a feature that you can just you know surgically add some where you actually have to go. The whole file and some of the balls and files are ginormous, so it's not easy to enable and they have a lot of stuff where the type itself is ginormous.

A

It's spread over like 25 files, and so you, like anything like half a tribe, is usually not very useful right, so you kind of have to you know, take some time off to actually do it and they just didn't have enough time doing that.

M

Just one thing regarding a civilization input looked into the sources we always create and now a will always personal knowledge if it's right to wear since the civilization info yeah. But you will see.

G

M

All the place, no.

G

One that's always saying like we didn't really focus on that at all I'm, not even sure we look at the implementations. At least I was notice. It's called by our own code. I mean.

A

Quite frankly, I don't even care what be positive if the implement, if all the implementations today will not regress when you pass in Allah, it should not be now. The thing goes that the buck is off the chain right like that, but given that we don't out reduce your lives, the way that works.

C

C

The partner said the question mark alright.

L

uh You mean tea question mark colon reason, we're.

F

Gonna do that? Oh! This is because it is a generic yeah. You can constrain. Generics can't have question.

G

Marks alone.

F

A

Logically, yes,.

K

Those are combined up, doesn't, except if it is America.

G

K

G

Have let's say you have a list of T, it's pretty bad as a list of string, then add it takes a string if I created as a list of string question mark then add takes a string question mark.

G

But this gets into weirdness in the implementations like VM. Implementing in generic method is weird because the compiler forces you to assume that every T that comes in might be null because it doesn't know whether it was going to be constructed with a nullable or non knowable type. But then, when, when you return ax T, you can't return null. Because again it doesn't know whether it might have another done, though so it has to assume not know for results and know for inputs.

A

Which is not confusing at all.

G

It makes sense this feature. It's a really interesting one, because it's taken a ton of work across, especially the language came on our team, but on us as well- and there are it's like most of the features- are like a hundred percent buttoned up, you know any feature. The languishing does end to end. It's like really polished this one. There are some pretty big gaps like you know. If I say new string already with the novelty journal, that is an array of non null strings right, except what I just said, new string array.

G

Everything in that array is known, yep.

A

I think one string eraser should just be populated with some chords from Shakespeare. That seems a bit.

G

G

Ideally, somewhere in the GC, just bury it in it.

A

Well, the good news is you're kind of viewing it up. Oh good.

A

It looks like everything here is effectively not knowledge.

G

Some of these are going to be impacted by new entrants. Yeah so example the find method that you just scrolled past that that returns a T yeah. This is so. This is where the I view scroll your cursor, like five lines up or yeah like right there. This is one of those interesting cases where we at the moment. We lack the ability to express the fact that, regardless of whether you use it nullable or non malleable, t if find, doesn't find what it's looking for.

G

It will return null right, it'll return to fault of T rather right, and we have no way at the moment of expressing that with the new attributes, there will be a maybe null attribute on the return type here. So basically is saying, even if you, even if you pass in a non malleable T, this still might return them. Do.

A

You know all this will actually show up in tooling, so when I actually instantiate this guy with string, will it then say yep it's find of string, but it returns. The string question mark no, probably normally were just doing. Control flow, probably know that it could be no array and I.

G

Think that's the intent it needs. You know in these. In the extent of time it would say no matter what you pass in, because it was maybe not on the T it'll put a question mark on the result. Oh really, she will show up is there and that will that's the intent like that the attributes would sort of manifest. You know reason whatever in.

H

G

Possible way of intellisense, yeah I think that's I, think in be string where to mark is more useful than maybe null string, but I don't know if that's actually going to happen for c-sharp a /v s 16.2 or whatever. This might come here.

A

Yeah, this will all say, take a reflection, api's interesting. If you have one exposed, not ability.

A

I mean ya, find all funny.

G

G

And generally with I, don't know how all the attributes are gonna be reflect like some of them. There isn't a good way with question marks. You know of expressing what they mean so I, don't entirely know when it's gonna look like an intelligence, anything that's codependent, it's probably exact all the the meat you know not know when or not know. If not know, all those attributes are gonna be weird.

E

E

Hello so set value. All of those take a question mark, but the one that takes three index doesn't ality.

G

Oh, it's like a bug.

G

A

By Asante eagle eyes, this is what we have young people here, because the old people can see the quarry.

F

What is that face all right, good venue.

A

Those odds are the woman's not defined.

G

This also highlights that know that, like you know, there are well 60,000 methods or something in.net are hidden. You know, are 20,000 in Corinth are know what it is. There's a really good chance, we're gonna ship all this stuff and have a few mistakes like this. No.

A

By the way, how does it work with roughing out? They can never be now correct, because in the ref in the out pointer themselves cannot be known right. They always point to something very foundation: I mean.

G

Technically they could be if used unsafe code, but from a c-sharp perspective. No, they can't be no but good. So the annotations end up referring to the null Nisour. The thing that's referenced. Yes, when you say ref string, question.

A

Mark it means there's a reference to something that came contain this confession right or.

G

If you, if you have a ref string and whatever and you put the may, be null or not null or something attribute on it, that's referring to again the location of the references.

A

Yeah I was just wanting money, ref question mark on out question mark or something sanity like this. Tomorrow there are kids, but I, don't think they ever cared.

G

In a weird way that the generic types were the easier ones because it was just way less stuff to have to pay attention to because couldn't put question marks where you want to do yeah.

L

Or just put a bunch of banks, so why is the compare inaudible, but the comparison is not.

G

Because that's I believe that's actually correct, although a second.

J

But it desirable what seemed very like rule.

C

No I think the compiler by default will use some default. Contrary web versus different version, so.

A

The comparisons are the delegate version of the interface right, so like they're killing is a default for that as well. Right.

H

Delegate can't we.

A

Throw when the comparison is moving, no I I get that alone and see, if kids very odd, that one accepts not the other one. Doesn't it's.

G

Questioning the current behavior, not the annotation, yeah yeah I liked it.

A

A

This is an example where I'm just me. It seems odd something we would have never caught in a previous API review, because I don't know what the code does.

C

There's a high structure comparable compared to and take the icomparer and I think that should be.

C

They wonder, which is great down to the second from the bottom.

G

Oh I mean that you're saying that the compare should be novel, yeah I think so I think most implementations as soon. What's not.

F

Yeah, the great dope and.

G

It's possible, we could say those are bugs, but like.

A

Isn't an example where the Walker kind of like how would the Walker you can work if you don't have that.

G

I could like all the tuple implementations of I structural, comparable they assumed the compare is non novel, no rep. If it is.

E

Or they don't even bother checking we so, for example, in the one that in the I structural equal equals explicit implementation from array we don't check. If the compare is no, we could just no rep default.

F

You as something is really really small, like I, don't know, Oh.

A

Every reference type show the dooblydoo.

C

Knows that? Because it's like my mama used to be mega normal think.

H

What did we say about the difference between I, compare and comparison.

G

The annotations are correct: according to the implementation yeah, it's just a question of. Do we like the babe and that's been there since I think thought that too oh yeah.

A

We compare a comparison was never great, because it's always this, maybe I- did I think it method in four or five. So you can actually construct the comparison from the compare about dividing before yourself, but.

H

A

H

Comparison is a delegate right and all the other places we take delegates. We don't accept no right like predicate. When we take predicates, we don't we don't take no there, so we kind of feels like that should be consistent with other delegate usages so that, if you want to make them consistent, the only way to make them consistent would be. You can't take a null I compare.

G

H

Think we're good with different yeah.

G

I think I mean I, don't think it's gonna change, yeah.

A

Yeah just write them up right now we can, we can review them separately right. It's just. It just seems law that we can't. You see the question max. It just backs me as odd. None of the super, informative yeah so in the structure is now and.

E

Also, it doesn't make sense because, like it doesn't matter much because there are explicit implementations but the compared to equals and get hash code or a yeah. No.

A

Rep, if the proper.

E

Names of the call.

A

Is also not passing in they, you hear where they're, just everything in line.

A

Or in a race segment, so Steven just said we assume awaken me now, even though we don't accept one, you can be constructed with one.

G

This one's this one's a little painful but I it because, especially because like when you call a method that, like a try, get method that has an array segment as an out if it returns false. That array segment is generally a default array segment. If you tried to access its array, it's gonna be null.

G

So this one felt like we had to had to make the array or turn inaudible, even though most of the time you try and actually access, the array won't be.

G

A

What are we doing enough ready equal to be a? We don't blow up right right.

G

In general, all of our equals methods do all the necessary checks to I need to do null comparison or to factor in no comparison as part of the operation. In.

A

All the methods we have an array segment does any of them lie Rev when the array is a queen now I.

G

Think that used to be the case, I feel like we've had some bugs and they just I believe at this point. Nothing will no ref if you give it a default or a second.

D

With their progeny box already, and this.

A

Yeah, what do we do to care in the field? Probably undress, not a woman.

G

Psychic, whatever you ask me on the field in a race segment, is a nerd yeah. It's a Claire doesn't know.

M

What's the reason that you turned in the implementation, the underline your way in every segment can be now because I picked was the question down the line, the backing away filled in every segment? Can its current the Marxist can be now. What's the reason for that default of our our segment.

G

Yeah, so you mean it's a teep, it's a TT or a question mark. It's all you're, saying yeah, it's it's again because you can say to feel people to raise segments of tea, I, just trust to always have a both of those customers. That's right.

I

G

It's that that's one of those fish on the new stuff. It guys that will pass and the Ray segment isn't out like now try and get a memory stream dot, try get array or did.

H

G

And it returns false that array segment was initialized to default. Where is it that makes the whole implementation well do it.

M

You mean anyone using.

G

M

No I mean I mean implementation in the VCL. For because now you have to apply a lot of thing, a lot of things yeah. There are a bunch of places. Okay,.

G

It's one of those cases where we were talking about earlier, where, where we're exposing, what's we're exposing it accurately it can be null, but the majority it's I'm, the volbeat, okay and one of the things we talked about as a potential attribute was some way to express for, like you know, so we're adding a not null when attribute where you can put it on one of these out parameters and basically say if this function returns.

G

True, this will not be known which works great for reference parameters, but for something like a race segment, where it's a struct that wraps a reference type. You don't have a good way of expressing that, and so for this release that you're going to get warnings.

G

D

This one's people.

K

Oh good custom attributes almost right and not.

F

Return, you can see empty.

C

Is this for to be singular canary, can move yeah I think the two are cannot because it's very empty a thing. I remember this.

C

Oh Justin and Jenny I think I've checked this particular case. I think it was incorrect, but I think it was a while ago. So.

D

A

Generally, say: that's also what we should do. We should never return in our collection. We should always first get returning everyone, one, because otherwise it is very easy to make the mistake of saying just for you to know that, and nobody ever expects a collection to me now.

K

Yeah I mean like get custom, attribute, send another one and return the not I mean, so nobody must.

C

Say a specific attribute and you don't have it for giving some present spot version. The array you want a list of attributes which was certain criteria and that in that case there is this entity.

A

In fact, I think we have explicitly I love it. It says down which one our collection, the native code, has for us to be different behaviors and that gets very.

F

A

E

A

They all kind of measure what everything.

M

You know ask one more general question: joy, I'm, regarding binary civilization here, um have you thought about the cases where you see you love something and something can be now, but you wouldn't expect it to be now because it because you could modify the payload or on doesn't framework, it's different yeah.

C

So the generic base is the same as you put the bank and you just basically make it grows, but at the same well you could call something by reflection and it completely perfectly okay, the.

G

You know anything I think we you know for those kinds of things. We assume that either the implementation of the DC realization is gonna balk at a bad payload that over as Christoph was saying, we basically say you know what that's that's just like using unsafe code to modify something. If the implementations do serialization isn't validating that it was correct and you modified it using effectively. Unsafe code is you're modifying the payload. Well, that's on you static analysis.

C

So not that you have, because if you know someone, someone is tampering, the payload and you get new reference exception, and you know your code should not be throwing on a fan section. You know something is wrong and you should probably investigate either way. Oh this.

M

Right, oh yeah, okay,.

F

The polymers everything is binary, logical, right, sorry,.

A

Not everything well, virtually everything 250 well, they buy marked, but you can think you can store one every two any anytime. You want you just don't get very good. Behavior.

K

A bad image Matt exception. The message follows a terminal string.

G

So this is another one of those special cases where we had a sort of gray area where, in theory, someone could override message on exceptions in general, like it's a virtual method for virtuals, you could override it and in return whatever you want, including null, but we look through all of our exceptions and none of them return null from message and there's no point in returning null. You can always just returned based on message whatever. What, if you created but.

K

Image exception, object was none thing and you asked for the message of concern. You'll get the base.

G

G

It's you will always get a non-null message, even if you pass it a null message.

A

Let's base message on exceptions: this is just an error occurred.

G

Yeah, it includes the name of the exception type.

F

Oh yeah, there were five black or yeah.

G

So if for bad image, format exceptions specifically, if you pass in a null message, then it has its own internal message that it uses that may or may not incorporate the file name with the invention. Question anyhow it. But this is one of those gray areas where for virtuals in general, we have this problem. In theory, every single virtual on every type- that's not sealed, should return nullable, but that wouldn't be particularly useful and it's not really accurate because from an intent perspective message should never return null and across all the code we looked at.

G

It never returns null. So we're sort of putting a stake in the ground saying all right message returns, not null if you're returning null that's on, you fix it and yeah. So again, one of these judgment calls.

G

K

K

K

I think the question: what could be the most used character in all so I'm.

G

Really I really do want to run and analyze her over core live and see. How many question marks we added yeah and thanks like it's, a high number.

A

Yeah I'm curious what the ratio between Vegas in question works is I'm. Pretty sure banks are like wise.

G

We have a lot of things. Pher will have many fewer once we get these attributes in particular. That does not throw attribute, because right now pretty much every public API we have has at least one bang per parameter.

A

The big thing for the button put a bunch.

B

It's a big bang.

B

A

Did you have a cross-check like across like, for example, this is all tryparse accepting the downstream.

G

I'm not sure we've done that analysis, but in general. Yes,.

A

Then pretty sure I ran into issues with some foot, throwing some woman.

G

If I remember correctly in general on all the primitive types, parse does not accept know the try parse does.

K

We were 10 just all some of the time, yeah.

K

G

About Santi and Christophe and booyah, and ten or others who were to stuff, but when I was going through and annotating, so I came to a type like daytime. It was such a relief that there were no reference types anywhere. You count. Oh yeah, Joe.

E

Sugg vanilla will enable of.

G

Yeah, there's that big, please, like ah yeah, I, think this is why you are not duplicating.

G

A

With a glad we replace string brackets.

G

A

G

G

B

I think question mark or a question mark I hate hell does the loot, which is good because you should never do it. My.

G

Least favorite, though, is bang boom; bang, which means not food suppressed warnings related in it. Oh yeah.

A

Actually so wrong about bang.

A

E

I was doing the work to teach an IDIQ know of what I came with a swath of you such weird test cases. It looked really really ugly, like a dictionary that could take tuples that had tuples that could be knowable that could have mobile types and then they des terreaux could be no yeah. I mean.

A

Look at the Roseland none of the reference types test: here's a hundred thousand lines of something.

A

And they also gave up the units as methods adventure, just of numbers like yeah. Well, my test case 1 3, 3, 8,.

A

A

A

I'm just trying to write down some general conventions like, for example, tripods should always you know the boat was always except malleable. I have phone provider should call they always be inaudible, because those are things you can easily swap check magic style.

A

G

Well, all this reflection stuff is another area where people are going to have a ton of warnings, because we you know here, we have no idea what you're going to invoke and whether it can return null or not. So we basically always have to say it can return all the cranes not.

C

A good if the target is smooshing, isn't it doesn't that mean like a stopping by, though so I when you were a dog ate when the target is stoned? Doesn't that mean static, oh yeah, so why is this.

I

G

That's a good question. No one actually calls that code. If that's, no one constructs delegates explicitly. So it probably shouldn't be question mark. But it's not going to impact anyone and nothing. You.

A

Can even survive from the show everything they block that.

G

But we should probably be accurate.

G

Oh I think you're right that that first line should probably be object. Question mark target, but yeah I, don't think it unpacked anyone.

A

Well, I guess it's the second one you're supposed to call right.

G

Oh, sorry, you don't use any of you won't use any of those protected constructors ever so.

A

I'm saying what were the intent was, this is for an instance method. This is for aesthetic method. Oh.

G

I, don't I have no idea, I've never used.

D

E

They create delegates, one that returned knowable is because of the throw online radio parameter right. Yeah.

G

Because we don't have the ability to say this will be no. This could be null if and only if thrown by and failure is true. We can know right right, yeah I mean unless you make it a condition like this. So if you look at the first overload of create delegate compared to the second one, the first one just calls the second one with throw and bind failure.

G

True, and that's why this one doesn't return an unknowable delegate, the second one which takes the option of having thrown by a failure, be false needs to be able to return, though, can be expressive of our bank today. Holding on right, no, we would be an attribute about allows you to basically say kind of we're, adding a not null when attribute yeah. We would need like the results we have. We haven't not know when that refers to the return value.

G

We would need to not know when that refers to a parameter, so you would say, return not bail when parameter name throw unbind. Failure is true.

A

Are you also couldn't express the fact that it will throw like it never returned when, when this thing is true and in our course.

G

Right, that's what I was yeah as an undecided. You did it. You need more expressiveness than we currently had, but.

A

It seems on supplying, like I, mean it sucks for the person implementing this overload here. But it also is nothing a good example where you just have to say a lot better than the types.

G

The first overload of combine above is another one of those cases where we're gonna have an attribute that will help us if the compiler can see that either input delegate is not enough and it'll know that the results it's not know as well, but we don't have the ability to do the same for the second combined overload, because we don't have the ability to say if any of the params coming in is not know and the result will be none known.

G

So if you, if you pass in and non null, is the first argument and a null is the second argument the compiler will know. The result is not known. If you then add one more null to your argument list and won't know whether the result is not or not.

B

You always going.

C

F

16 very good.

A

Taking 17 we're like like you, should do something differently: I, don't know what it is, but what are we doing right now.

D

D

F

You're gonna be grappling from the eye from a provider right yeah. All the things that are here we'll just do a spot check.

F

F

G

You remind me of what situation exceptions. Stacktrace can be known.

A

Although stick.

A

I think it's hoping that I might, after you constructed one right.

A

But do you see by the way.

G

Two lines about my reading: great, oh yeah, well,.

A

I think, once you I think you know, if you say new exception and someone to be now right, you have to actually throw to get affected. All right, it'll be okay, but there goes also back to Janet.

M

Yeah I had to show in binary format to get a well. It stacked us to see you like this, yet our game, yeah.

A

Yeah, this is awesome.

D

G

Yeah you're right I just tried a new exception about stacktrace equals null.

A

This seems something where maybe empty string is actually the better answer. Yeah, that's what I thought in.

I

J

It's quite gently like: why should everybody who.

A

Doesn't know, thank you first, the default is a valid answer of.

D

D

A

Into streak, or something fair enough still, those is good.

A

G

I mean in truth, a lot of these Streng returning methods that are diagnostic. You just pass them to something else that prints it out. Debug got it debunked, discerned or console dot, write line and all of them except string, question mark and ignore nulls. So in most of these cases it doesn't really matter, but you know it could.

M

So if those annotations is it possible to do markets so that it would not cause an error? If you are in a certain scope log in the catch statement, we don't have the ability to do that. I mean it could be special case by.

I

The compiler, because I said something you know cool and then after that part.

C

A

Polymers, it will break very quickly why all you have to do is whoever factory or the creation and access is in different methods in the Cottonwood. No assume it's not now because well didn't create it right.

G

I mean it's possible, we'll start to find I. Don't think that was me particularly problematic, it's possible to after we ship and people start using it in earnest that we'll start to find things where either we decide.

G

We need new attributes to express something or the compiler special case, something there are any couple places the compiler is likely go in a special case like interlock compare exchange, equality, things would be, are difficult to express with attributes and very kind of complicated rules, but the compiler might a special case, so we might find something like that would need to get into the list. So we.

A

Didn't add any funds or actions that are marked with other words right, because it would be they're all generic component or a little law. I'm saying we didn't add, like a you, noticed, an action kind of thing where we have another rule anyway, right.

J

G

All of our actions and funks, except an output generic, so there's nothing to there's, there's nothing to annotate.

A

Everybody because they actually feel well, we could constraint it. But what happens if you can't read et to the object whose to look on right.

G

We wouldn't want to do that because we won't be able to to be able to use novels as their arguments.

J

Yeah I'm not saying changing.

A

Those what I'm saying is, like you know in span action. We added a new delegate, try to just be able to express that you're saying.

G

An action in funk well, it would be difficult to do because you can't overload unknowability, but we would, you know, need to come up with some other signature yeah, but liquid these and in order to be able to say these will never be. No. We didn't do that. My guess. We can look at it in the future. Yeah.

A

I mean I also think it's fun in bed. You could fissile no come on you Taurus gunpoint, as you can I have like Ian's and reference types like intertwine right.

G

No good but but I didn't make you if you have a method, foo a generic method like foo of T and you take a func of T and then in your foo of T. You add a constraint where T : class, for example, you're effectively saying I, want I only accept a func of T that will never return.

G

But as the constraint we'll end up applying to the donkey as well.

K

You see keepalive can tick none. It.

G

Can it doesn't make much sense, so this is a place where we could. We could choose to say well, the intent. Is you don't pass null here if you're passing no you're probably doing something wrong, but GC keepalive doesn't actually do anything other than serve as a marker to say, like basically keep this thing moving by the stack.

K

And they cannot accept not for suppress finalize and sometime.

G

Suppress our lives actually throws if Mel's taking, because it actually has to modify us to kind of touch the object itself well. Keep a life has nothing to do with anything on the object itself. Keep alive just says pretend this thing is still rooted. Suppress finalize actually has to go into the object in the mutated right. What did you see? You wouldn't mark something inside the object. It's a bit on my run after keep alive, I think.

A

They do is that, because you call the method on the stack low. Logically, the thing is still in use. I think the method does nothing like the only thing I know is. Is that it's it's never get in line or anything like this is just basically it's.

L

Just different seeing.

A

K

So, just through reference.

A

To things that said,.

G

I mean I, said I'll call to GC ducting polite null is useless, so I mean we could choose to say this doesn't make any sense accepting and all, even though it's perfectly happy accepting null and that's it I mean that's. One of those judgment calls because I'm totally fine with that. If people want to do that, I think.

E

We're trying to represent the intent right so I think they tend to make sense to make you numb off these games to me. So the intent of using keep alive is to keep an object alive right and you cannot keep no alive. If you think about is a little book box like more.

C

Oh and yeah, if you have a notable in it box it if.

G

Yeah, if you, if you had a an int question, mark a equals null and then you did GC keep alive a that would end up calling AC keep alive with no.

G

But I don't know that that's I mean I'll.

I

K

People can you call suppress finalized on object like this, a good throw? No well, maybe we'll throw argument. No internet, 5 OH.

K

A

A

Turns out enzyme undead expect.

A

D

D

A

See try past is not acceptable.

D

A

D

F

What won it? Police will be. My guess, interesting. You try to polish exercise. I want.

G

Actually should be nullable yeah.

E

Yeah, we don't throw, we just say the result is default and return polls. Okay, try.

C

Parsing my perch is nice, for this level is when you're adding move the empty string.

C

Fixing reality or like no strength, was gnostics, but something about this room with a bag.

C

A

A

There's a lot of API.

E

We looked at a system, one time and open extensions, probably happy what we need to relieve.

A

Which is good, which also means it's not that bad don't go through everything tonight that today not on the flats review it. Yes, if you later too will be a bad father party.

D

Although these from the we're gonna be a completely accurate.

A

As long as there's a funny game of shots, that involves every time somebody finds about me all doing that could be a fun party Mike a progressively harder to drink, though, which might actually be good too I.

G

Just tried changing the GC keepalive thing and it ends up producing about 10 warnings across quarterly, not because we're explicitly passing null, but because we're passing something the compiler can't prove is not null yeah.

G

So that con raises the question whether it's a good idea to manage it or not, because there's no harm no foul in leaving it not leaving it knowable right.

A

Does one of those areas where I think intent is also less important right? Basically, you intended I want to keep this alive. It's not null, and if it's not well, then I guess you doesn't matter what I like or.

J

A

I mean if you call it if knowledge just odd, but I, think anybody in the code review would say exactly you're trying to cheat me in 64.

A

I'm very happy that we do inaudible pointers that would be really ugly good-looking.

A

K

B

The mushroom RF object is get lifetime service, always known now,.

A

Watch the back. Okay, like we don't support your morning, we actually return anything meaningful. You Danny Boy return. Now.

A

Lovely runtime call, or something like that. Wait currently.

G

We throw in platform not supported exception, so.

A

Telling cubature nothing.

J

That's awesome.

A

Nevermind them I guess aim is to initialize lifetime service, then and memorize claw.

A

A

A

A

The heck is the signature by the way, I remember my name, EC member itself.

A

Is that the messenger.

D

A

B

Is that basically, the metadata representation assisted signature? This.

A

B

A

D

K

K

Moving London into handy.

F

A

Oh everybody is a funny offending number.

B

C

A

It's appalling these both areas.

F

C

As the object name, our property on the object disposed exception level in the constructor.

G

Thank you. It returns string that empty. If it's not.

K

Why object to string.

K

Another we don't return.

A

What we had there like I wanted, can our discussion I.

E

A

When I go over and.

G

The short answer is, we have a lot of disagreements about it and, while the object, the base object not to straighten them turned all, and while in fury no one should ever return now, people do.

A

Usually by accident, but there's too many of them.

A

K

But this will not create a lot of warnings.

E

It was only created if you call it directly an object.

G

Yeah, so the theory that we're going operating under- and we could be wrong- is that there actually won't be all that many warnings because in general either you're calling to string on a specific. You have a variable of a specific type and when you over I to string, you can constrain it further. So most overrides of two string returns string, not string question mark, and so, if you had a type- and you wanted to say parse the result of two string, there's a really good chance that it's not actually returning a string.

G

Question mark is returning string. If you have, if you have object and you're calling to string you're, probably not parsing, it you're, probably passing it to something else, to work with you're passing it to string. Can cap you're passing it to console.writeline you're, passing it to whatever all of which except string question mark so you're, probably not getting not getting it warnings they are either. So the places where you get warnings- and there are some- are limited.

C

The class above the ultimate disposal exception has the two string please very shortly that subject.

B

That is object on.

C

B

Has to be version somewhere an exception exception.

D

A

Look at this beautiful couple here: doesn't it Vita nice yeah.

K

B

The capital, L capital, L, really selfie,.

A

And I think there's not help at all.

E

I think I know why yeah you think if the attributes right yeah and as I change them the way that we that, when API, creates a types now.

A

I thought like, if you do, if you basically remove attributes that should not return you to fear, because I floated them all right, otherwise they would rather.

D

Oh, which might be doing.

M

F

Was the question ring.

E

To string yeah.

D

E

Read always bad I think we return the string empty.

M

M

Is it the generic one? Yes,.

F

F

The mouse fries always been an original reason, our best, that is there's.

D

No reason that we have to.

A

Then, what's the reason that one time attract handle eyes and non no I was training. For my that's what my.

I

Cruiser wise left, non-market unknowable.

I

Because you are except you're just one having faults wrong.

G

It should accept a customer yeah.

F

Which one the runtime.

G

Candle lanterns have another two comparisons: all.

L

G

Arguments and all the operator equals and not equals. Okay, all of them to my beard. It's.

A

Also subtract so technically speaking, I mean this mean accepts an object. That's because the compiler can prove that it's not now did not give you a warning. I mean worst case returns false at one time, so.

K

A

Added the rule that we saw, we should say anything that takes any contact operator, equals non-disclosure except malleable,.

G

You you I, get nervous now, any time you scroll up to things where tryparse is about to be on the screen.

A

I made a promise dear.

A

K

K

And, like my not analyzing, goes through training right for time, make sure yeah.

A

We should add some rules to this idea would be.

A

Actually, any API rules are super easy to write. Two doesn't have to do. Look at the code user to look at the.

A

Blinkin charismatic 7m annul removal that I'd only eat or nouns right so.

K

Come pair can take knowledge string.

K

G

K

Why not close it fix anything? Nothing I think it looks.

G

Much like it I think if throws yeah, but the index is nonzero and the string is serum is not, but I can look.

A

A

A

Once we don't accept it knowledge, anything.

A

Back atcha, it can usually be done to sorry where you looking at the culture input. One isn't cutting for normally also considered honorable Oh for.

I

A

For to whatever the behavior is usually heard, sometimes it might.

G

Be that you know I want actually that actually does throw if it's no.

C

Probably because there is an overall pretty doesn't think.

G

You you it's usually a I format provider where we allow it to be known. It's rare that we actually take culture explicitly. Oh that's one.

K

Thing we attract in the know, it doesn't make sense to pass I formatting here, because we need to get a compare info from. So I from what doesn't I mean if you best like dick time format. The end point, it doesn't mean oh I,.

G

Know I'm saying that I think we may was thinking. I was I for methods take an eye for map provider. We almost always allowed to be. No we're always allowed to be known right, but this is different. This is taking a culture.

A

The are we done with this one. We are in agreement that this looks right, yeah.

E

I just looked at the code and if the, if any of the strings are null, we consider its length to be zero and then we we do an index comparison against. That knife make sure it is right, and so if the index is greater than zero, then we would throw.

D

A

Probably by accident accident and.

E

You asked about ienumerable values yeah like a con card. We throw for that if it is No.

G

Okay, maybe we throw if the values is no yeah you'll laugh, but it contains to be done. I'm sorry attempt to be known, yeah I get if the enumerables numerators concurrent returns and all that's fine.

K

And the concat watches take two strings and mean, except.

C

Man, yes, if you add a low strength, less impossible string between the return table, which will be stranger than tea, does it make sense, but it's.

A

Police doesn't return on a number.

G

It's it's pretty convenient that the cat allows null to be translated into string, not empty. It's weird to do two nulls, but it kind of logically falls out of it.

E

And that will help also with the object to string warnings.

G

I think when Santi is its string, duck and cat object are zero that you recently fixed I object.

H

I was an issue with with.

G

Feel like it like, didn't you change that to return string.empty rather than Noah or something yes, yeah I changed.

E

One of those I mean.

G

I think I was like the one of maybe two changes that we made to make the annotations what we wanted them to be. Let me find it right.

G

Yawns Iran's Bon Secours.

G

J

And something is oddly Oh.

G

K

What is internal.

F

Yeah, that sounds like a rule to me. It's like it's in trying to fall.

B

G

I think if it's not interned, it'll return No if it is it'll return, the intern version.

A

I'm saying if it is more internal returns,.

A

Never listen, try which one returns, the internal representation.

G

I believe that's what it does.

G

And then sometime you come away deep, yeah yeah the church, the dog say, returns a reference to string if it is in the common language run time in turn. Otherwise, no.

A

It sounds like this is the conditional version of this one, so this one basically says if it is in turn, give me the internal representation, but don't add it to the in current thing like this. One would either give you the internal representation or add it to the Internet.

A

So the scenario makes sense: if you guys just bad.

A

I'm glad that is not empty takes.

I

Another bit strange.

G

Pretty useless bitten.

A

The join does not allow the elements to be now. It seems odd, considering that for object, we've accepted, you know which join and these guys, except neither both.

I

Sorry look, but these guys this guy does not of the innumerable one might be a mistake.

G

And not join of innumerable of string.

G

C

This gate that.

G

Should be spin question mark yep, yes,.

E

I didn't know my PR.

D

I think we're to join.

A

But it's even do you know how covariance and contravariance works with nullable? Is that how it's supposed to work once the buns are fixed? Okay, so, okay? So it will be that.

G

Between questions, it should work the way you expect it to work. If you have a base a parameter on a base, virtual that accepts string, you should be able to override it to accept string question mark. If you have return value that returns string, question mark, you should be able to override it to return string, but that also goes to casting of generics right.

G

A

Won't way so, if I have a have a method that accepts in ienumerable of non level of sort of object, now level of the question mark sorry and I pass in an ienumerable of string. No question mark yeah.

G

You should be able to yeah it's possible that doesn't work for any.

G

A

What does it mean to split on a non separator.

G

I believe it uses either a space or all white space.

G

uh No separator or knitting.

G

Yeah it treats all white space separators.

A

But why is it none about let's, he was odd, no.

G

Uh-Huh, it's basically how you say, like passing an array of every known, whitespace character,.

A

You can unfold, it keep forgetting the HRA and the string when I'm, not using you.

K

Replace different culture in phonology.

D

C

D

C

C

So I'm not saying what's inconsistent.

A

This one here doesn't accept inaudible device. The.

C

Longest terminal.

C

If a joke about the culture and they want to decipher them, yeah.

K

Yeah another place, yeah culture was not malleable, but here they.

E

Remember the other API was.

K

F

Can conquer yeah also spring can be ready. You.

G

Yeah this one used this current culture. If cultures know yeah and there's a post point earlier, it kind of highlights how having this in the API makes it easier to spot differences like that. Yep, perhaps true change the other one through something.

F

K

A

Breaking chain, while accepting more than one breaking change, if you have not accepted knowledge safe to accept now, okay, you have code that relies on an argument on existing, should change something hikers.

G

K

It's so people yep.

A

But McCoy the answer is, it would be holding it wrong, so I should wait and see. You know why not yeah the guy would already listed. Oh no I know actually I didn't because the iPhone for what a comment. Let me actually have this so we said spring. Compare.

F

Doesn't accept culturing.

K

I know what you think.

E

Yeah to lower those myself.

C

To lower or doable because there is an overall presentation.

K

D

E

Too low, where empty uses current culture.

E

To offer also those and I said one.

A

Thing that I, don't like this room is I, have no way of checking. What's next in this room, and you think that that was an area and we got the problem outside, because if there's no meeting twelve.

A

We should at least try to oh yeah, that's great under the effects of the tour from the left end of not like high-dpi I can say that finish the string now, let's finish stringing this and then we can.

F

That's a good ending, there's nothing at all now, but it seems like enough. People have to leave. Are there any breakfast? Now you can pop it? Did you annotate to offer hey William order, don't cover to Chari.

A

C

Anything I'm saying Houston discovery, yeah.

A

Looks fine all right, so no Meah done with string, which is at least you think of the system, is based features the big hitter and then from there honestly, but he's except for reflection, big and.

J

E

That what we you, where we should start on the next name, long.

F

Thank you. There yeah.

A

I need I need to book. I need to ask her one where the Jemez, who has something short, that they need to get unblocked by and then either we either. We spend the second hour, but there's also another API review that the WinForms guys asked me to do so. I probably have to I'm. Can you get away with just accused.

A

All righty things online, this.

L

E

Also, you could probably help so I. Have these.

E